RECENT ADVANCES

IN WIRELESS
COMMUNICATIONS
AND NETWORKS
Edited by Jia-Chin Lin
Recent Advances in Wireless Communications and Networks
Edited by Jia-Chin Lin

Published by InTech
Janeza Trdine 9, 51000 Rijeka, Croatia

Copyright © 2011 InTech

All chapters are Open Access articles distributed under the Creative Commons
Non Commercial Share Alike Attribution 3.0 license, which permits to copy,
distribute, transmit, and adapt the work in any medium, so long as the original
work is properly cited. After this work has been published by InTech, authors
have the right to republish it, in whole or part, in any publication of which they
are the author, and to make other personal use of the work. Any republication,
referencing or personal use of the work must explicitly identify the original source.

Statements and opinions expressed in the chapters are these of the individual contributors
and not necessarily those of the editors or publisher. No responsibility is accepted
for the accuracy of information contained in the published articles. The publisher
assumes no responsibility for any damage or injury to persons or property arising out
of the use of any materials, instructions, methods or ideas contained in the book.

Publishing Process Manager Niksa Mandic

Technical Editor Teodora Smiljanic
Cover Designer Jan Hyrat
Image Copyright Edyta Pawlowska, 2010. Used under license from Shutterstock.com

First published July, 2011

Printed in Croatia

A free online edition of this book is available at www.intechopen.com

Additional hard copies can be obtained from orders@intechweb.org

Recent Advances in Wireless Communications and Networks, Edited by Jia-Chin Lin

p. cm.
ISBN 978-953-307-274-6
Contents

Preface IX

Part 1 Physcial and MAC Layers 1

Chapter 1 A Study of Cramér-Rao-Like Bounds and

Their Applications to Wireless Communications 3
Kao-Peng Chou and Jia-Chin Lin

Chapter 2 Synchronization for OFDM-Based Systems 23

Yu-Ting Sun and Jia-Chin Lin

Chapter 3 ICI Reduction Methods in OFDM Systems 41

Nadieh M. Moghaddam and Mohammad Mohebbi

Chapter 4 Multiple Antenna Techniques 59

Han-Kui Chang, Meng-Lin Ku, Li-Wen Huang, and Jia-Chin Lin

Chapter 5 Diversity Management in MIMO-OFDM Systems 95

Felip Riera-Palou and Guillem Femenias

Chapter 6 Optimal Resource Allocation in OFDMA

Broadcast Channels Using Dynamic Programming 117
Jesús Pérez, Javier Vía and Alfredo Nazábal

Chapter 7 Primary User Detection in

Multi-Antenna Cognitive Radio 139
Oscar Filio, Serguei Primak and Valeri Kontorovich

Chapter 8 Multi-Cell Cooperation for Future Wireless Systems 165

A. Silva, R. Holakouei and A. Gameiro

Part 2 Upper Layers 187

Chapter 9 Joint Call Admission Control in

Integrated Wireless LAN and 3G Cellular Networks 189
Chunming Liu, Chi Zhou, Niki Pissinou and S. Kami Makki
VI Contents

Chapter 10 Near-Optimal Nonlinear

Forwarding Strategy for Two-Hop MIMO Relaying 211
Majid Nasiri Khormuji and Mikael Skoglund

Chapter 11 Connectivity Support in

Heterogeneous Wireless Networks 221
Anna Maria Vegni and Roberto Cusani

Chapter 12 On the Use of SCTP in Wireless Networks 245

Maria-Dolores Cano

Chapter 13 Traffic Control for Composite

Wireless Access Route of IEEE802.11/16 Links 267
Yasuhisa Takizawa

Part 3 Applications and Realizations 297

Chapter 14 Wireless Sensor Network: At a Glance 299

A.K. Dwivedi and O.P. Vyas

Chapter 15 Software Defined Radio Platform for

Cognitive Radio: Design and Hierarchical Management 327
Amor Nafkha, Christophe Moy,
Pierre Leray, Renaud Seguier and Jacques Palicot

Chapter 16 Dealing with VoIP Calls During “Busy Hour” in LTE 345
Angelos Antonopoulos, Elli Kartsakli,
Luis Alonso and Christos Verikoukis

Chapter 17 A Semantics-Based Mobile

Web Content Transcoding Framework 361
Chichang Jou

Chapter 18 Power Supply Architectures for

Wireless Systems with Discontinuous Consumption 379
Jose Ignacio Garate and Jose Miguel de Diego

Chapter 19 Wireless Sensor Networks in

Smart Structural Technologies 405
Yang Wang and Kincho H. Law

Chapter 20 Extending Applications of Dielectric Elastomer

Artificial Muscles to Wireless Communication Systems 435
Seiki Chiba and Mikio Waki
Preface

Many exciting impacts on our daily life are shortly anticipated due to recent advances
in wireless communication networks that enable real-time multimedia services to be
provided via mobile broadband Internet on a wide variety of terminal devices. The
trend is mainly driven by the evolution of wireless networks and advanced wireless
information and communication technology (ICT). The progression to fourth-
generation (4G) or IMT-advanced systems is expected to significantly change usage
habits and introduce new services, such as the services supported by higher spectral-
efficiency communication technology and self-configurable, high-feasibility networks.
The Third Generation Partnership Project (3GPP) Long-Term Evolution (LTE)
continues to be enhanced as this book is being written.

Although there have been many journal and conference publications regarding
wireless communication, they are often in the context of academic research or
theoretical derivations and sometimes omit practical considerations. Although the
literature has many conference papers, technical reports, standard contributions and
magazine articles, they are often fragmental engineering works and thus are not easy
to follow up. The objective of this book is to accelerate research and development by
serving as a forum in which both academia and industry can share experiences and
report original studies and works regarding all aspects of wireless communications. In
addition, this book has great educational value because it aims to serve as a virtual,
but nonetheless effective bridge between academic research in theory and engineering
development in practice, and as a messenger between the technical pioneers and the
researchers who followed in their footstep.

This book, titled Recent Advances in Wireless Communications and Networks, focuses
on the current hottest issues from the lowest layers to the upper layers of wireless
communication networks and provides “real-time” research progress on these issues.
In my endeavor to edit this book, I have made every effort to ask the authors to
systematically organize the information on these topics to make it easily accessible to
readers of any level. The editor also maintains the balance between current research
results and their theoretical support. In this book, a variety of novel techniques in
wireless communications and networks are investigated. The authors attempt to
present these topics in detail. Insightful and reader-friendly descriptions are presented
to nourish readers of any level, from practicing and knowledgeable communication
X Preface

engineers to beginning or professional researchers. All interested readers can easily

find noteworthy materials in much greater detail than in previous publications and in
the references cited in these chapters.

This book is composed of twenty chapters that were authored by the most
knowledgeable and successful researchers in the world. Each chapter was written in
an introductory style beginning with the fundamentals, describing approaches to the
hottest issues and concluding with a comprehensive discussion. The content in each
chapter is taken from many publications in prestigious journals and followed by
fruitful insights. The chapters in this book also provide many references for relevant
topics, and interested readers will find these references helpful when they explore
these topics further. These twenty chapters are arranged in order from the lowest layer
to the upper layers of wireless communication. This book was naturally partitioned
into 3 main parts. Part A consists of eight chapters that are devoted to physical layer
(PHY) and medium access control (MAC) layer research. Part B consists of five
chapters that are devoted to upper layer research. Finally, Part C consists of seven
chapters that are devoted to applications and realizations.

Chapter 1 is an introduction to topics at an inner receiver in wireless communications,

including a historical perspective and a description of Cramér-Rao-like bounds and
relevant applications to estimation techniques in wireless communications. Chapter 2
conducts a thorough review of the initial synchronization techniques applied to
wireless orthogonal-frequency-division-multiplexing (OFDM) communications.
Chapter 3 is devoted to deeply investigating novel techniques of inter-carrier
interference (ICI) reduction in practical OFDM communications. Chapter 4 is focused
on multiple-antenna techniques from diversity, spatial multiplexing to beamforming
techniques. Chapter 5 deeply investigates diversity management techniques in MIMO-
OFDM communication systems. Chapter 6 thoroughly investigates resource allocation
methods in OFDMA broadcast channels. This is the downlink scenario in either LTE-A
or IEEE 802.16m wireless communications. Chapter 7 is dedicated to primary user
detection in multi-antenna environments. Spectral sensing techniques are considered
as the most important issue in recent research regarding cognitive radios (CRs) or
cognitive networks. Chapter 8 focuses on multi-cell cooperation methodology. In third
generation mobile communications, macro-diversity was investigated. In similar
environments, multi-cell cooperation may be expected in opportunistic communication
networks.

Chapter 9 covers a novel technique on joint call admission control in integrated

wireless local area networks (WLAN) and cellular networks. Chapter 10 is devoted to
studying near-optimal nonlinear forwarding strategies in two-hop MIMO relaying
scenarios. Chapter 11 discusses a research on connectivity support in heterogeneous
wireless networks for next-generation multimedia communication networks. Chapter
12 studies the use of a stream control transmission protocol (SCTP) in wireless
communication networks. Chapter 13 investigates traffic control for composite
wireless access route of IEEE802.11/16 links.
 Preface XI

Chapter 14 comprehensively covers wireless sensor networks. Chapter 15 is a

complete study of CRs built by a software-defined radio (SDR) platform. Chapter 16
deals with VoIP calls during rush hours in LTE communications. Chapter 17
demonstrates a semantics-based mobile web content transcoding framework. Chapter
18 proposes novel and effective power supply architectures for wireless
communication systems. Chapter 19 applies wireless communication technology to
smart structural technology. Chapter 20 provides extending applications of dielectric
elastomer artificial muscles to wireless communication systems.

In summary, this book covers broad areas of communications and networks. The
introductions, derivations, discussions and references in this book significantly
improve the readers’ understanding of communications and networks and encourage
them to actively explore these broad, exciting and rapidly-evolving research areas.

Jia-Chin Lin
Distinguished Professor
Dept. of Commun. Engr.
National Central University,
Taiwan, R.O.C.
 Part 1

Physcial and MAC Layers

 1

A Study of Cramér-Rao-Like Bounds and

Their Applications to Wireless Communications
Kao-Peng Chou and Jia-Chin Lin
Department of Communication Engineering, National Central University
Jhongli, Taoyuan,
Taiwan

1. Introduction
Estimation theory has been developed over centuries. There are several approaches to
utilizing this theory; in this chapter, these approaches are classified into three types. Type I
includes the oldest two methods, the least squares (LS) and moment methods; both of these
methods are non-optimal estimators. The least squares method was introduced by Carl
Friedrich Gauss. Least squares problems fall into linear and non-linear categories. The linear
least squares problem is also known as regression analysis in statistics, which have a closed
form solution. An important feature of the least squares method is that no probabilistic
assumptions of the data are made. Therefore, the linear least squares approach is used for
parameter estimation, especially for low complexity design (Lin, 2008; 2009). The design
goal of the least squares estimator is to find a linear function of observations whose
expectation is a linear function of the unknown parameter with minimum variance. In
addition, the least squares method corresponds to the maximum likelihood (ML) criterion if
the experimental errors are normally distributed and can also be derived from the moment
estimation. As an alternative to the LS method, the moment method is another simple
parameter estimation method with probabilistic assumptions of the data. The general
moment method was introduced by K. Pearson. The main procedure in the moment method
involves equating the unknown parameter to a moment of distribution, then replacing the
moment with a sample moment to obtain the moment estimator. Although the moment
estimator has no optimal properties, the accuracy can be validated through lengthy data
measurements. This is mainly because the estimator based on moment can be maintained to
be consistent. Type II includes the methods of minimum variance unbiased estimator
(MVUE) and the Bayesian approach, which are both optimal in terms of possible minimum
estimation error, i.e., statistical efficiency. MVUE is the best guess of an unknown
parameter. The standard MVUE procedure includes two steps. In the first step, the Cramer-
Rao lower bound is determined, and the ability of some estimator to approach the bound. In
the second step, the Rao-Blackwell-Lehmann-Scheffe (RBLS) theorem is applied. The MVUE
can be produced by these two steps. Moreover, a linear MVUE might be found under more
restricted conditions.
In the Bayesian method, the Bayesian philosophy begins with the cost function, and the
expected cost with respect to the parameter is the risk. The design goal of Bayesian
philosophy is to find an estimator that minimizes the average risk (Bayes risk). The most
4 Recent Advances in Wireless Communications and Networks

common cost function is a quadratic function because it measures the performance of the
estimator in terms of the square of the estimation error. In this case, the Bayes risk is the
mean square error (MSE), and thus, the Bayes estimate is a minimum mean square error
(MMSE) estimator. Another common cost function is the absolute function, which regards
the absolute estimate error as the Bayes risk. In this case, the Bayes estimate is a minimum
mean absolute error (MMAE) estimator. Another estimation, which is not a proper Bayes
estimation but fits within Bayes philosophy, is the maximum a posteriori (MAP) estimation.
The MAP criterion considers the uniform cost function, and the parameter is discretely,
randomly distributed under this assumption. Although this estimate usually only
approximates the Bayes estimate for uniform cost, the MAP criterion is widely used for
estimator design. Type III includes the maximum likelihood (ML) estimate, which is the
most important estimation theory in the 20th century. The ML estimate can be referred to as
an alternative MAP without knowledge of apriori probability of the parameters. The ML
estimator is the most popular approach for obtaining a practical estimator, which was
previously used by Gauss. The general method of estimation was first introduced by R. A.
Fisher with the concepts of consistency, efficiency and sufficiency of the estimation function.
The ML estimator is required when MVUE does not exist or cannot be found. An advantage
of the ML estimator is that a practical estimation is easy to obtain through the prescribed
procedures. Another advantage of this approach is that MVUE can be approximated due to
its efficiency. Thus, from the theoretical and practical perspectives, the ML approach is the
most important and widely used estimation method of this century (Lin, 2003).
Because the ML estimator is essential in estimation theory, the analysis of its performance is
a benchmark of estimator design. This benchmark is commonly known as the Cramer-Rao
lower bound (CRLB), which is named after Harald Cramer and Calyampudi Radhakrishna
Rao. In section 2, the definition of the CRLB is introduced with several examples. A general
case of CRLB under two common communication channels is then introduced in section 3.
To establish basic knowledge of hybrid parameter estimation, random parameter estimation
is presented in section 4. In section 5, Cramer-Rao-like bounds for hybrid parameter
estimation are introduced and compared with each other. Lastly, we summarize some
practical cases and compare these cases with modified CRB which is most common used
Cramer-Rao-like bounds.

2. Cramer-Rao lower bound (CRLB)

The Cramer-Rao lower bound (CRLB) is a lower bound on the variance of any unbiased
estimator. Many other variance bounds exist, but the CRLB is the easiest one to derive and is
thus widely used in many estimation studies. This theory provides a benchmark for
examining the performance of novel estimation algorithms and also highlights the
impossibility of finding an unbiased estimator with a variance less than this lower bound.
Before introducing the definition of CRLB, there is a simple estimation example that may
could help promote understanding of the basic CRLB concept.
Example 2.1
There is a simple signal transmission model with a transmitted signal s , a received signal
r[n] and an additive white Gaussian noise w[ n] .

r[n] = s + w[n] (1)

A Study of Cramér-Rao-Like Bounds and Their Applications to Wireless Communications 5

Here, the index n refers to the n ’th observation. In this problem, the transmitted signal s is
assumed to be an unknown parameter that is deterministic during n observations. The first
idea estimate s takes one observation as our estimation, e.g., the n ’th observation,
namely sˆ = r[n] . To analyze the estimation accuracy, we check the likelihood function of
r[n] as shown.

1 ⎡ 1 2⎤
p( r[ n]; s ) = exp ⎢ − 2 ( r[ n] − s ) ⎥ (2)
2πσ 2⎣ 2σ ⎦

Substituting the estimator we chose in this likelihood function yields

1 ⎡ 1 2⎤
p( sˆ ; s ) = exp ⎢ − 2 ( sˆ − s ) ⎥ (3)
2πσ 2 ⎣ 2σ ⎦

Now, the mean value is the target parameter s , and the estimation variance is σ 2 . The
estimation accuracy can then be determined as

−1
⎛ ∂ 2 ln p( r[ n];s ) ⎞
var(sˆ ) = σ 2 = −E ⎜⎜ ⎟⎟ . (4)
⎝ ∂s 2 ⎠

Furthermore, we are interested in finding a more accurate estimator by lowering the

variance σ 2 . This can be achieved by exploiting multiple observations. Assuming the
observation samples are identical independently distributed, the likelihood function for
multiple observations is

1 ⎡ 1 N −1 2⎤
p( r[n]; s ) = N
exp ⎢ − 2 ∑ ( r[n] − s ) ⎥ . (5)
⎢⎣ 2σ n = 1
( 2πσ 2 2 ) ⎥⎦

A ML estimator can be derived in the same way as for a single observation to yield

N −1
∑ r[ n]
n=1
sˆ = , (6)
N
which is an unbiased estimator, namely E{sˆ} = s . We can also find the estimation variance
using equation (4); the result is similar to the single observation MLs with a factor N in the
denominator:

σ2
var(sˆ ) = . (7)
N
An extreme case occurs when N approaches ∞ , and the process reduces the estimation
variance to 0. From this simple example, we can summarize that the ultimate goal of
estimator design is to find the minimum variance unbiased estimator (MVUE), and if we
wish to illustrate the performance of our estimator, then estimation variance can be found
through the likelihood function. Now, we are ready to define the CRLB (Kay, 1998).
6 Recent Advances in Wireless Communications and Networks

<Theorem>
Assume the pdf, p( r ;θ ) ,satisfies the regularity condition

⎡ ∂ ln p(r ;θ ) ⎤
Er ;θ ⎢
∂θ ⎥ = 0 for all θ . (8)
⎣ ⎦

Then, the variance of any unbiased estimator θˆ has a lower limitation

1
var(θˆ ) ≥ . (9)
⎡ ∂ 2 ln p(r ;θ ) ⎤
−Er ;θ ⎢ ⎥
⎣⎢ ∂θ 2 ⎦⎥
An unbiased estimator may be found that attains the bound for all θ if and only if

∂ ln p(r ;θ )
= I (θ )( g( r ) − θ ) (10)
∂θ
for some function I (θ ) and g(r ) . This estimator can be stated as θˆ = g( r ) , which is a MVUE with
variance 1 / I (θ ) . To attain the variance lower bound, Fisher’s information is defined as

⎡ ∂ 2 ln p(r ;θ ) ⎤
I (θ ) = −Er ;θ ⎢ ⎥, (11)
⎣⎢ ∂θ 2 ⎦⎥
which is used to calculate the covariance matrices associated with maximum-likelihood estimates.
An unbiased estimator that achieves the variance lower bound is referred to as “efficient”.
In other words, an unbiased estimator that achieves the CRLB is an efficient estimator and
must be MVUE. Figures 1 and 2 are illustrations of the relationship between a MVU
estimator and the CRLB.

Fig. 1. θˆ1 MVU and efficient

A Study of Cramér-Rao-Like Bounds and Their Applications to Wireless Communications 7

Fig. 2. θˆ1 MVU and not efficient

Although there are some theories capable of finding MVUE by sufficient statistics and the
Rao-Blackwell-Lehmann-Scheffe theorem, we will not introduce the details in this chapter.
However, we encourage readers to fully inform themselves concering MVUE from the
references in this chapter (Kay, 1998).
A question may be raised concerning why the minimum variance estimator should be an
unbiased one. Although the unbiased estimator seems to sucessfully find an perfect
estimator ϕ because the expectation value approaches the true parameter i.e., E[θˆ ] = θ 0 , but
a biased estimator may outperform than an unbiased one. For example, in some situations,
the relationship between a MVUE and a Bayesian MSE estimator may be illustrated in figure
3.

Fig. 3. MVUE vs. Bayesian estimator

8 Recent Advances in Wireless Communications and Networks

In this example, the Bayesian MSE estimator is an unbiased estimator. The performance
comparison in figure 3 shows that within a certain parameter interval, the biased Bayesian
estimator may have lower estimation variance than MVUE’s. However, this comparison also
shows that the biased estimator performs terribly outside this interval. Thus, the unbiased
estimator has an advantage in terms of consistent performance.

2.1 Asymptotic CRLB

For some cases in which the closed form of the CRLB may not be derived, the asymptotic
CRLB can be used instead; this form can be attained by assuming that infinite observation
samples are available. Under this assumption, we have an observation sample with an
infinite signal-to-noise ratio (SNR).

3. General case CRLB

3.1 Gaussian noise
The AWGN channel is the most common channel model in wireless communication, which
was also used in the example in the last section. In example 2.1, we only consider the
estimate of symbol s . Now, a general form of any parameter θ is derived.
Example 3.1
Assuming symbol s is transmitted with a general unknown parameter θ and added with
an AWGN wn (t ) . The signal model is describe as

rn (t ) = s(t ;θ ) + wn (t ) , (12)

where n indicate the n th observation. Following the general CRLB derivation steps, the
likelihood function is found first and differentiation with respect to θ is then performed
twice.

1 ⎡ 1 N −1 2⎤
p( rn (t ); s(t ),θ ) = N
exp ⎢ − 2 ∑ [ rn (t )−s(t ;θ )] ⎥ (13)
⎣ 2σ n = 1
( 2πσ 2 2 ) ⎦

∂ ln p(rn (t );s(t ),θ ) 1 N −1 ∂s(t ;θ )

= 2 ∑ [ rn (t )−s(t ;θ )] (14)
∂θ σ n=0 ∂θ
2
∂ 2 ln p(rn (t );s(t ),θ ) 1 N − 1 ⎛ ∂s(t ;θ ) ⎞ ∂ 2 s(t ;θ )
= ∑ ⎜ ⎟ +[ rn ( t )− s( t ;θ )] (15)
∂θ 2 σ 2 n = 0 ⎝ ∂θ ⎠ ∂θ 2

∂ 2 ln p( rn (t );s(t ),θ )
Taking the expectation of with respect to p(r ; s ,θ ) into Fisher’s information
∂θ 2
yields
2
⎪⎧ ∂ ln p(rn (t );s(t ),θ ) ⎪⎫ 1 N − 1 ⎛ ∂s(t ;θ ) ⎞
2
I (θ ) = −Er ;s ,θ ⎨ ⎬= 2 ∑ ⎜ ⎟ (16)
⎭⎪ σ n = 0 ⎝ ∂θ ⎠
2
⎩⎪ ∂θ
Finally, the inverse recipocal of the Fisher‘s information produced by the CRLB in the
AWGN channel.
A Study of Cramér-Rao-Like Bounds and Their Applications to Wireless Communications 9

1 σ2
var(θˆ ) ≥ = (17)
I (θ ) N − 1 ⎛ ∂s(t ;θ ) ⎞
2
∑ ⎜ ⎟
n=0 ⎝ ∂θ ⎠

3.2 Complex Gaussian channel

Another commonly seen channel is complex Gaussian channel. The mobile communication
and wireless communication usually introduce the Rayleigh fading due to multipath delay
spread and Doppler shift. In numerical simulation we may use the Jake’s (Clarke) model,
but in theoretical analysis, complex Gaussian channel is more popular, because it has
Rayleigh distributed amplitude with an uniformly distributed phase, which is convenient to
use and without loss of generality.
Example 3.2
The signal model can be extended from the general AWGN channel model. We multiply the
− jφ
Rayleigh distributed channel gain α 0 and the uniformly distributed channel phase e 0
with the symbol s(t ;θ u ) .

− jφ0
rn (t ) = α 0 e s(t ;θu ) + wn (t ) (18)

Alternatively, using complex coordinates, i.e., the Gaussian distributed α I and αQ with
mean η A and variance σ A2 yields

rn (t ) = (α I + jαQ )s(t ;θ ) + wn (t ) (19)

Because the α I , αQ and wn (t ) terms are Gaussian distributed, the received signal rn (t ) is
also Gaussian distributed. To find the joint likelihood function, the mean mr and variance
σ r2 of the received signal should be derived.

mr = η A (1 + j )s(t ;θ ) (20)

σ r2 = 2σ A2 Ps (t ;θ ) + 2σ N2 (21)

Here, Ps (t ;θ ) = s(t ;θ )s(t ;θ )* is the power of the transmitted signal. The joint likelihood
function turns out is then described by

1 ( r (t )−mr )2
pr (r (t ); s(t ;θ )) = exp( − ) (22)
2πσ r2 2σ r2

4. Random parameter estimation

In previous sections, some basic knowledge of estimation bounds were introduced based on
unknown parameters with random interference. These kinds of estimation problems are
categorized in the classical estimation approach. Some properties of estimation methods are
listed in Table 1.
10 Recent Advances in Wireless Communications and Networks

Parameter Sample Parameter

types distribution distribution
LS Unknown Unknown Non
Moment Unknown Known Non
MVUE Unknown Known Non
Bayesian Random Known Known
MAP Random Known Known
ML Both Known Uniform
Table 1. Some estimation properties
Another research area focuses on random parameters estimation, and several approaches,
including the Bayesian theorem, MAP and ML, are widely used already. One of the most
popular and well-known Bayesian approache is the MMSE estimator. Below, the MMSE will
be briefly introduced with an example.
Example 4.1
Assuming that we received signal r (t ) that is composed of a random symbol s and white
Gaussian noise w(t ) , the following relationship can be described.

r (t ) = s + w(t ) (23)

The conditional pdf of r (t ) with a priori information can be stated as

1 ⎛ 1 N −1 ⎞
p( r (t ); s ) = exp ⎜ − ∑ ( rn (t )−s )2 ⎟ (24)
2πσ 2⎝ 2 n = 0 ⎠

Using Bayes’ rule,

p(s ;r (t ))p(r (t ))
p( r(t ); s) = (25)
p(s )

After certain computations, the conditional pdf with a posteriori information is obtained as

1 ⎛ 1 ⎞
p(s ; r (t )) = exp ⎜ − 2 (s − μs ;r )2 ⎟ , (26)
2πσ s2; r ⎜ 2σ ⎟
⎝ s ;r ⎠
where
1
σ s2;r = ; (27)
N 1
2
+
σ σ s2

⎛N μ ⎞
μs ;r = ⎜⎜ 2 x + s2 ⎟⎟σ s2;r . (28)
⎝ σ σ s ⎠

The MMSE estimator is then determined as

sˆ = E{ s|r (t )} = μs ;r = α x + (1 − α )μs (29)

A Study of Cramér-Rao-Like Bounds and Their Applications to Wireless Communications 11

where

σ s2
α= 2
(30)
σ 2 +σ
s
N
The Bayesian mean square error is defined as

⎛ ⎞
σ2 ⎜ σ s2 ⎟ σ2
Bmse(sˆ ) = E[( s −sˆ ) ] =
2
⎜
N ⎜ σ 2 +σ 2 ⎟≤ N (31)
s ⎟
⎝ N ⎠

As σ s2 → ∞ i.e., without any information from a prior knowledge, the bound would be the
same with the sample mean estimator. This result can be compared with that of the first
example in this chapter, and an important concept of Bayesian estimator is revealed: any
prior knowledge will result in higher accuracy of the Bayesian estimator.

5. Hybrid parameter estimation

In addition to classical estimation and random parameter estimation, there is a more
complicated scenario called hybrid parameter estimation. In hybrid parameter estimation,
the desired parameter is a vector that is composed of several unknown paramters and
random parameters. The parameter vector can be constructed as

T
θ = ⎡⎣θTr θTu ⎤⎦ , (32)

where θr is a random parameter vector and θu is an unknown parameter vector. Because

we are considering the random parameters, we assume that we have some prior knowledge
of these parameters, such as the probability distribution function. Several techniques for
calculating hybrid parameter Cramer-Rao like bounds are described below.

5.1 CRLB with nuisance parameter

In our first case, θr is treated as a nuisance parameter, which means that these random
parameter are undesired.
Example 5.1
Reformulating the signal model and likelihood function yields

rn (t ) = s(t ; θ) + wn (t ) (33)

p( rn (t ), θr ; s(t ), θu ) =
1
exp( −
[rn (t )−s(t ;θ)]2 ) . (34)
2πσ 2 2σ 2

Because we assumed that the pdf is well-known and these denoted parameters are
unimportant, the marginal likelihood function is derived first, and the nuisance parameters
are integrated out of the equation.
12 Recent Advances in Wireless Communications and Networks

p( rn (t ); s(t ), θu ) = ∫ p(rn (t ),θr ;s(t ),θu )p(θr )dθr (35)

θr

Now, the resultant problem becomes a classical estimation problem, and the CRLB can be
derived step by step.

∂ ln p(rn (t );s(t ,θu )) 1 N − 1 ∂s(t ,θu )

1. = 2 ∑ [ rn (t )−s(t ,θu )] (36)
∂θu σ n=0 ∂θu

2
∂ 2 ln p(rn (t );s(t ,θu )) 1 N − 1 ⎛ ∂s(t ,θu ) ⎞ ∂ 2 s(t ,θu )
2.
2
= 2 ∑ ⎜ ⎟ +[ rn (t )−s(t ,θu )] (37)
∂θu σ n = 0 ⎝ ∂θu ⎠ ∂θu 2

⎧ ⎫
3. I ( θ ) = Er ⎪⎨ ∂ ln p(rn (t );s(t ,θu )) ∂ ln p( rn (t );s(t ,θu )) ⎪⎬ (38)
i, j
⎪⎩ ∂θ i ∂θ j ⎪⎭

⎡ ⎤
4. CRLB(θˆi ) = ⎢ 1 ⎥ ≤ var(θˆi ) (39)
⎣⎢ I ( θ ) ⎦⎥ i ,i

5.2 Hybrid CRLB

In some scenarios, the effect of these ramdom parameters cannot be ignored. Another
method that considers the joint pdf called joint estimation. The CRLB for this kind of joint
estimation is called hybrid Cramer-Rao bound (HCRB). The derivation process is nearly
identical to that of ordinary CRLB; the likelihood function is determined, and partial
differentiation with respect to the desired parameter is performed twice.

rn (t ) = s(t ; θ) + wn (t ) (40)

p(rn (t ), θr ; s(t ), θu ) =
1 [r (t )−s(t ;θ)] )
exp( − n
2
(41)
2πσ 2 2σ 2

∂ ln p( rn (t ),θr ;s(t ),θu ) 1 N −1 ∂s(t ;θ )

= ∑ [ rn (t )−s(t ;θ )] (42)
∂θ 2
σ n =0 ∂θ

2
∂ 2 ln p( rn (t ),θr ;s(t ),θu ) 1 N − 1 ⎛ ∂s(t ;θ) ⎞ ∂ 2 s(t ;θ )
2
= 2 ∑ ⎜ ⎟ +[ rn (t )−s(t ;θ)] (43)
∂θ σ n = 0 ⎝ ∂θ ⎠ ∂θ 2
Because the joint pdf is considered, the expection of Fisher’s information should be taken
with respect to p( r (t ),θr )

⎧⎪ ∂ ln p(rn (t ),θr ;s(t ),θu ) ∂ ln p(rn (t ),θr ;s(t ),θu ) ⎫⎪

I ( θ )i , j = Er , θr ⎨ ⎬ (44)
⎪⎩ ∂θi ∂θ j ⎭⎪
The joint pdf p( r (t ),θr ) is not easy to determine, and an alternative approach using double
layer expectation which computes the expectation with respect to the conditional pdf first.
We define the information matrix with respect to the conditional pdf p(r (t );θr ) as
A Study of Cramér-Rao-Like Bounds and Their Applications to Wireless Communications 13

⎧⎪ ∂ ln p(rn (t ),θr ;s(t ),θu ) ∂ ln p( rn (t ),θr ;s(t ),θu ) ⎫⎪

I ( θ0 )i , j = Er ;θ ⎨ |θr ⎬ . (45)
r
⎪⎩ ∂θ i ∂θ j ⎭⎪
Then expectation is computed with respect to p(θr ) , and all of the random pararameters are
eliminated.

⎧⎪ ⎧⎪ ∂ ln p(rn (t ),θr ;s(t ),θu ) ∂ ln p( rn (t ),θr ;s(t ),θu ) ⎪⎪ ⎫⎫

I ( θ )i , j = Eθ ⎨Er ; θ ⎨ |θr ⎬⎬
r r
⎩⎪ ∂θ i ∂θ j ⎭⎪ ⎭⎪
⎩⎪ (46)
I ( θ )i , j = Eθ
r {I ( θ ) }
0 i, j

Finally, the HCRB is derived as

⎡ 1 ⎤
HCRB(θˆi ) = ⎢ ⎥ ≤ var(θˆi ) . (47)
⎣⎢ I ( θ ) ⎦⎥ i ,i

5.3 Modified CRLB

During the process of deriving the HCRB, an important step involves taking the inverse of
the Fisher’s information matrix. In some cases, the inverse of the Fisher‘s information matrix
may not exist or cannot be derived into a closed form lower bound. We can then try the
modified or simplified bound, such as the MCRB. Instead of taking the inverse of the matrix
first, we select the desired estimation element from the information matrix first and then
execute the inverse step. After choosing the desired estimation element, the Fisher’s
information is no longer in a matrix form, and derivation is easier.

⎡ 1 ⎤
MCRB(θˆi ) = ⎢ ⎥ ≤ var(θˆi ) (48)
⎣⎢ ( )i ,i ⎦⎥
I θ

An previously reported example can help distinguish the difference between these CR-like
bounds (F. Gini, 2000).
Example 5.2
When considering a data-aided joint frequency offset estimation case, the signal model can
be described as

rn (t ) = Ae − j 2π f Dt s(t ) + wn (t ) (49)
− jφ
Here, A is the complex channel, which can be rewritten as A = α 0 e 0 = α I + jα Q , and
e − j 2π f Dt represents the frequency offset. The estimation parameter matrix
θ = [ f D α I α Q ]T can be defined. Because this is a data-aided case, s(t ) can be a pilot or
preamble, and we can assume that s(t )s(t )* = 1 without loss of generality. Then the signal
after pilot removal is

xn (t ) = rn (t )s(t )*
(50)
xn = (α I + jαQ )e j 2π f Dt + vn (t )
14 Recent Advances in Wireless Communications and Networks

xn (t ) is also Gaussian distributed. Following the derivation of S. M. Kay (1998) and F. Gini
(2000), we can find the conditional Fisher’s information matrix.

⎡ 2π 2 N ( N −1)(2 N −1) 2 2 π N ( N −1) π N ( N −1) ⎤

⎢ (α I +αQ ) − αQ αI ⎥
⎢ 3σ N2 σ N2 σN 2
⎥
⎢ ⎥
π N ( N −1) N (α I −η A )2
I (θ 0 ) = ⎢ − αQ + 0 ⎥ (51)
⎢ σN 2
σ N2 σ A2 ⎥
⎢ 2
⎥
⎢ π N ( N −1) N (αQ −η A ) ⎥
⎢ αI 0 + ⎥
⎣ σ N2 σ N2 σ A2 ⎦
By computing the expectation of α , the Fisher’s information for the frequency offset is

I ( f D ) = Eα {I (θ 0 )} (52)

Then the MCRB is derived as

1 3
MCRB( f D ) = = (53)
⎡⎣ I ( f D ) ⎤⎦ 4π 2
N ( N − 1)(2 N −1)ρ
11

where ρ = (η A2 + σ A2 ) / σ N2 is the SNR. Now, the difference between the MCRB and the HCRB
can be checked. As mentioned previously, the HCRB is

⎡ 1 ⎤
HCRB( f D ) = ⎢ ⎥
⎣⎢ I ( f D ) ⎦⎥ 11 , (54)
3(K R + 1)(K R + 1+ N ρ ) 1
= 2
2π N ( N −1) 2(2 N −1)(K R + 1)(K R + 1+ N ρ )−3N ( N −1)ρ K R

where K R = η A2 / σ A2 is the Rice factor, which is the power ratio between direct path signal
and other scatter path signals. A comparison of the HCRB and MCRB can be evaluted as.

HCRB( f D ) 2(2 N −1)(K R + 1)( K R + 1+ N ρ )

= . (55)
MCRB( f D ) 2(2 N −1)(K R + 1)(K R +1+ N ρ )−3N ( N −1)ρ K R

Based on the equation above, in the general case, the ratio is always larger than 1, which
means that the HCRB is generally a tighter bound than the MCRB. Conversely, when
K R → 0 or K R → ∞ , the ratio of HCRB to MCRB approaches 1. It is interesting that these
two bounds only meet for two extreme scenarios, namely the Rayleigh channel and direct
path.

5.4 Miller Chang bound

The Miller Chang bound (MCB) is proposed by R. W. Miller and C. B. Chang (1978). They
state that the MCB can apply to a more restricted class of estimator that is unbiased for each
value of the nuisance parameter, which is referred to as locally unbiased, whereas the
standard Cramer-Rao bound (CRB) can applies to any estimators that are unbiased over the
ensemble. The Miller Chang bound is defined as
A Study of Cramér-Rao-Like Bounds and Their Applications to Wireless Communications 15

⎧⎪ 1 ⎫⎪
MCB(θˆi ) = Eθ ⎨ ⎬. (56)
r I(θ )
⎩⎪ 0 i ,i ⎭⎪
The MCB has a similar form to the MCRB, but the MCB is always tighter than the MCRB.
More directly, the MCB applies to more restricted estimators than the CRLB, which implies
that the MCB is tighter than CRB, and the MCRB is looser than the CRB, which was derived
by A. N. D’Andrea (1994). Therefore, the MCB is tighter than the MCRB. Alternatively, we
can also explain this relationship using Jensen’s inequality for any convex function ϕ and
random variable x

ϕ ( E[ x ]) ≤ E ⎡⎣ϕ ( x ) ⎦⎤ . (57)

In our case, the inverse function for a positive defined matrix is a convex function, so

1 ⎪⎧ 1 ⎪⎫
MCRB(θˆi )= ≤ Eθ ⎨ ⎬ =MCB(θˆi ) . (58)
{
Eθ I ( θ0 )i , i
r
} r I(θ )
⎪⎩ 0 i , i ⎪⎭

Now, from example 5.2 in the MCRB subsection, the MCB of the joint estimated frequency
offset is

3σ N2 ⎧⎪ 1 ⎪⎫
MCB( f D ) = Eα ⎨ ⎬
2π N ( N −1)(2 N −1) ⎩⎪α I2 +α Q2 ⎭⎪
2
(59)
⎪⎧ 2(η 2 +σ 2 ) ⎪⎫
MCB( f D ) = MCRB( f D )Eα ⎨ 2A 2A ⎬
⎩⎪ α I +α Q ⎭⎪
The final result still remains the expectation term, so it cannot be derived into a closed form.
Although the MCB is a tighter bound than the MCRB, the MCRB is more likely to derive
into a closed form. In addition, the MCB requires a locally unbiased estimator, which is also
a harsh restriction for estimator design, so the MCRB is more popular for theoretical
analysis.

5.5 Summary of the relationship between Cramer-Rao-like bounds

Some of the relationship between Cramer-Rao-like bounds has been derived previously
(Reuven, 1997). In this work, they consider the signal model with Gaussian distributed
channel gain and an unknown timing delay. We can also derive this relationship from our
examples in subsection 5. Following from example 5.1, if we carry through the calculation
to the end, then we will obtain the marginal CRB of the frequency offset f D .

3( K R +1+ N ρ ) (60)
CRB( f D ) = 2
2π N ( N −1)ρ[ N ( N + 1)ρ + 2(2 N −1)K R ]
Then, this result is compared with that for the HCRB, which was derived in equation (55).

2(2 N −1)(K R + 1)( K R +1+ N ρ )− 3N ( N −1)ρ K R (61)

CRB( f D ) = HCRB( f D )
(K R + 1)[ N ( N + 1)ρ + 2(2 N −1)K R ]
16 Recent Advances in Wireless Communications and Networks

After calculations, the CRB can be summarized into the HCRB multiplied by a function. We
simplified the fraction in equation (62) and found that it is larger than 1 only if N < 1 . This
result implies that CRB( f D ) ≥ HCRB( f D ) , and the relationship HCRB( f D ) ≥ MCRB( f D ) has
been proven by equation (56). Another way to prove this is to use a corollary.
“For any positive defined matrix M , ⎡ M −1 ⎤ ≥ [ M11 ]−1 , an equal occur if M is diagonal”.
⎣⎢ ⎦⎥ 11
Finally, we summarize the relationship between CRB, HCRB and MCRB as

CRB( f D ) ≥ HCRB( f D ) ≥ MCRB( f D ) (62)

However, the relationship between the MCB and MCRB was also derived in equations
(58-59) using Jensen’s inequality. Because the MCRB seems to be a looser bound in the
Cramer-Rao-like bounds family, we normalized all other bounds to the MCRB, as shown
in figure 4.

Fig. 4. Normalized bounds versus the Rice factor

From the figure above, the MCB exhibits drastic variation near K R = 1 , which indicates that
the locally unbiased estimatior of f D is difficult to find when the power of scatter signal is
larger than line-of-sight (LOS) signal. Moreover, when K R = 0 (Rayleigh channel), the ratio
of the normalized MCB approaches infinity, which means that no locally unbiased estimator
exists. The Rayleigh fading channel is the most frequently used channel model in a wireless
A Study of Cramér-Rao-Like Bounds and Their Applications to Wireless Communications 17

communication environment, which is another reason that the MCRB is more popular than
the MCB. In multiple parameters estimation, joint estimation techniques have been a
popular topics recently. In terms of hybrid parameter joint estimation, the benchmark for
comparison with is the HCRB. Based on equation (56) and figure 3, the HCRB has a feature
that approaches the MCRB when K → 0 or K → ∞ . As mentioned previously, the
R R
scenario K R → 0 implies the Rayleigh channel. The analysis shows that the MCRB is quite
sufficient as a benchmark to design an estimator in the Rayleigh channel environment.
Some prior research has been reported on the relationship among the joint estimate initial
phase, timing delay and frequency offset (D'Andrea , 1994). The author summarized and
derived some cases in which the CRB is equal to the MCRB.
i. Estimation of φ when f D , τ and data are known
ii. Estimation of τ when f D , φ , and data are known
iii. Estimation of f D with M-PSK modulation, when τ and differential data are available
but φ is unknown.
Here, φ , f D and τ are the initial phase , frequency offset and timing delay. Other cases
may exist in which the CRB is equal to the MCRB, but these cases are difficult to analyze. An
important conclusion here is that if an estimator approaches the MCRB, then the MCRB
must be closed to the CRB.

6. Advanced topics
6.1 Carrier phase and clock recovery
As summarized by A. N. D’Andrea (1994), there are several synchronization techniques that
can attain or approach the MCRB for a carrier phase θ and timing τ estimation. Under the
assumption that the frequency offset and timing are known, MCRB(θ ) can be attained
using two algorithms.
i. Maximum likelihood decision-directed (ML-DD), proposed in H. Kobayashi (1971)
ii. Ad hoc non-data-aided (ad hoc NDA) method, proposed by A. J. Viterbi (1983).
The MCRB(τ ) can also be attained using the ML-DD algorithm with derivative-matched
filters (DMFs); however, the use of DMFs also makes the estimator impractical to
implement. Several alternative algorithma have been found that can approach MCRB(τ )
without using DMFs.
i. DD early-late scheme with T / 2 sample space, proposed by T. Jesupret (1991).
ii. DD scheme, proposed by K. H. Mueller (1976).
iii. NDA scheme, proposed by F. M. Gardner (1986).
Although these alternative algorithms can approach MCRB(τ ) without using DMFs, they
are subject to some restrictions that require θ to be known and a roll-off factor α that
should be small.

6.2 Frequency offset estimation

In this subsection, three practical carrier frequency estimation techniques are overviewed
and compared with the popular MCRB.
A. NDA loop algorithm
The first algorithm is a non-data-aided carrier frequency estimation; a block diagram
representing this algorithm is shown in figure 5. The received signal r (t ) first passes
18 Recent Advances in Wireless Communications and Networks

through the matched filter G * ( f ) and the so-called “frequency-matched filter” dG * ( f ) / df .

Assuming that the timing is perfectly synchronized, the frequency error is described as

ek = Re{ x k y k* } . (63)

Then, the frequency error passes through a loop filter and triggers the voltage-control
oscillator (VCO) to compensate for the frequency offset. If the loop filter is implemented by
a simple digital integrator, then the VCO output can be written as

fˆD ( k + 1) = fˆD ( k ) + γ ek (64)

Fig. 5. NDA loop algorithm

The next step is to evaluate the estimation noise performance. There are three assumptions
i. The frequency errors are small as compared to the symbol rate.
ii. The pulse shaping filter G * ( f ) is a root-raised cosine function with a roll-off factor α .
iii. There is perfect timing delay synchronization
Under these assumptions, the frequency jitter is minimized, and the estimation variance of
f D is derived to be

4α BLT 1 ⎛ 1 ⎞,
σ 2fD = ⎜ 1+ ⎟ (65)
π 2T 2 Es /N 0 ⎝ Es /N 0 ⎠
where BL is the loop noise bandwidth and T is the symbol duration.
B. Differential decision-directed algorithm
The second algorithm is a differential decision-directed (DDD) algorithm that is used on
PSK signals; the block diagram for this algorithm is shown in figure 5. This algorithm is
similar to the NDA algorothm except for the frequency error generator. The assumptions for
this algorithm include the following:
i. The frequency errors are small compared to the symbol rate.
ii. G * ( f ) is the same as was defined previously.
iii. Timing is perfectly synchronized.
Because we are discussing the M-PSK signal, we can denote our symbol by
A Study of Cramér-Rao-Like Bounds and Their Applications to Wireless Communications 19

c k = exp( jϕ k ) , (66)

where

ϕ k = 2π n / M , n = 1, 2...M . (67)

Fig. 6. Differential decision-directed (DDD) algorithm

Then the phase difference between x k and x k − 1 will be

Δφk = Δϕ k + ( f D − fˆD )T + δ k , (68)

where Δϕ k is due to modulation, ( f D − fˆD ) is caused by estimation error, and δ k is the

phase noise with other interferences, which can be modeled as an uniformly distributed
random variable from −π to π . When the difference between x k and x k − 1 is correct,
perfect Δϕˆ k = Δϕ k is obtained. The most important component of this block diagram is the
frequency error that is defined as

{
ek = Im xk x k* − 1 exp( − jΔϕˆ k )} (69)

The performance of the estimator is then

BLT 1 ⎛ 1 ⎞
σ 2f = ⎜ 2 BLT + ⎟ (70)
D π 2T 2 Es /N 0 ⎝ Es /N 0 ⎠

Prior to the simulation, we assume that BLT = 5 × 10 −3 and the QPSK signals have a roll-off
factor α = 0.5 . The result is compared with the MCRB in figure 7.
As shown in figure 7, these two algorithms yield much greater variance than the MCRB,
which indicates that there is still room for improvement.
C. Feed-forward NDA
The third algorithm is the feed-forward NDA for M-PSK signal modulation; the block
diagram for this algorithm is shown in figure 8. The F function in the middle of the block
diagram is a 4th-powered non-linear function. Similar to the previous analyses , the received
phase can be separated into three parts:
i. A step-wise increasing quantity 2π Mf D kT due to the frequency error f D .
ii. A constant initial phase.
20 Recent Advances in Wireless Communications and Networks

iii. A phase noise caused by thermal noise and inter-symbol interference that is uniformly
distributed from −π to π .

Fig. 7. Comparison of the variance of the two algorithms with that of the MCRB

Fig. 8. Feed-forward NDA

The estimation variance has been derived (Bellini, 1990) in a scenario with a very high SNR,
the estimation variance can be approached as
A Study of Cramér-Rao-Like Bounds and Their Applications to Wireless Communications 21

3 1
σ2 ≈ (71)
fD 2π 2T 2 L(L2 -1)m Es /N 0

The MCRB in this case is

3T 1
MCRB( f D ) = (72)
2π 3
( LT ) 3
Es /N 0

Thus, when L  1 and m = 1 ,the algorithm performance will attain the MCRB. However,
this result is obtained under very high SNR. Further research is needed to design estimators
that can approach or attain the estimation bounds with less restriction.

7. References
Bellimi, S., Molinari, C. and Tartara, G. (1990). Digital Frequency Estimation in Burst Mode
QPSK Transmission, IEEE Trans. Commun., Vol.38, No.7 , (July 1990), pp. 959-961,
ISSN: 0090-6778
Cramer, H. (1946). Mathematical Method of Statistics, Princeton University Press, ISBN-13:
978-0691005478, Uppsala, Sweden.
D’Andrea, A. N., Mengali, U. and Reggiannini, R. (1994). The Modified Cramer-Rao Bound
and Its Application to Synchronization Problems, IEEE Trans. Commun., Vol.42,
No.2/3/4, (Febuary 1994), pp. 1391-1399, ISSN: 0090-6778
Gini, F. and Reggiannini, R. (2000). On the Use of Cramer-Rao-Like Bounds in the Presence
of Random Nuisance Parameters, IEEE Trans. Commun., Vol.48, No.12, (December
2000), pp. 2120-2126, ISSN 0090-6778.
Gardner, F. M. (1986). A BPSK/QPSK Timing Error Detecor for Samples Receivers, IEEE
Trans. Commun., Vol.34, No.5, (May 1986), pp. 423-429, ISSN: 0090-6778
Jesupret, T., Moeneclaey, M. and Ascheid, G. (1991). Digital Demodulator Synchronization,
ESA Draft Final Report, ESTEC No. 8437-89-NL-RE., (Febuary 1991)
Kay, S. M. (1998). Fundamentals of Statistical Signal Processing, Prentice Hall, ISBN 0-13-
345711-7, Upper Saddle River, New Jersey
Kobayashi, H. (1971). Simultaneous Adaptive Estimation and Decision Algorithm for
Carrier Modulated Data Transmission Systems, IEEE Trans. Commun., Vol.19, No.3,
(June 1971), pp. 268-280, ISSN: 0018-9332
Kotz, S. and Johnson, N. L. (1993). Breakthroughs in Statistics: Volume 1: Foundations and Basic
Theory, Springer-Verlag, ISBN: 0387940375, New York.
Lin, J. C. (2003). Maximum-Likelihood Frame Timing Instant and Frequency Offset
Estimation for OFDM Communication Over A Fast Rayleigh Fading Channel, IEEE
Trans. Vehic. Technol., Vol.52, No.4, (July 2003), pp. 1049-1062.
Lin, J. C. (2008). Least-Squares Channel Estimation for Mobile OFDM Communication on
Time-Varying Frequency-Selective Fading Channels, IEEE Trans. Vehic. Technol.,
Vol.57, No.6, (November 2008), pp. 3538-3550.
Lin, J. C. (2009). Least-Squares Channel Estimation Assisted by Self-Interference
Cancellation for Mobile PRP-OFDM Applications, IET Commun., Vol.3, Iss.12,
(December 2009), pp. 1907-1918.
22 Recent Advances in Wireless Communications and Networks

Mueller, K. H. and Muller, M. (1976). Timing Recovery in Digital Synchronous Data

Receivers, IEEE Trans. Commun., Vol.24, No.5, (May 1976), pp. 516-530, ISSN: 0090-
6778.
Miller, R. W. and Chang, C. B. (1978). A Modified Cramer-Rao Bound and its Applications,
IEEE Trans. On Inform. Throey, Vol.IT-24, No.3, (May 1978), pp-389-400, ISSN : 0018-
9448
Poor, H. V. (1994). An Introduction to Signal Detection and Estimation, Springer-Verlag, ISBN:
0-387-94173-8, New York.
Viterbi, A. J. and Viterbi, A. M. (1983). Nonlinear Estimation of PSK-Modulated Carrier
Phase with Application to Burst Digital Transmission, IEEE Trans. Inform. Throey,
Vol.IT-29, No.3, (July 1983), pp. 543-551, ISSN : 0018-9448.
 2

Synchronization for OFDM-Based Systems

Yu-Ting Sun and Jia-Chin Lin
National Central University, Taiwan,
R.O.C

1. Introduction
Recently, orthogonal frequency division multiplexing (OFDM) techniques have received
great interest in wireless communications for their high speed data transmission. OFDM
improves robustness against narrowband interference or severely frequency-selective
channel fades caused by long multipath delay spreads and impulsive noise. A single fade or
interferer can cause the whole link to fail in a single carrier system. However, only a small
portion of the subcarriers are damaged in a multicarrier system. In a classical frequency
division multiplexing and parallel data systems, the signal frequency band is split into N
nonoverlapping frequency subchannels that are each modulated with a corresponding
individual symbol to eliminate interchannel interference. Nevertheless, available bandwidth
utilization is too low to waste precious resources on conventional frequency division
multiplexing systems. The OFDM technique with overlapping and orthogonal subchannels
was proposed to increase spectrum efficiency. A high-rate serial signal stream is divided
into many low-rate parallel streams; each parallel stream modulates a mutually orthogonal
subchannel individually. Therefore, OFDM technologies have recently been chosen as
candidates for fourth-generation (4G) mobile communications in a variety of standards,
such as 802.16m and LTE/LTE-A.

2. OFDM fundamentals
2.1 System descriptions
The block diagram of an OFDM transceiver is shown in Fig. 1. Information bits are grouped
and mapped using M-phase shift keying (MPSK) or quadrature amplitude modulation
(QAM). Because an OFDM symbol consists of a sum of subcarriers, the n − th N × 1 mapped
signal symbol Xn is fed into the modulator using the inverse fast Fourier transform (IFFT).
Then, the modulated signal xn can be written as

1 N −1
xn = ∑ X k e j 2π kn
N k =0
N
, n = 0,1, ,N - 1 (1)

where N is the number of subcarriers or the IFFT size, k is the subcarrier index, n is the
time index, and 1 N is the normalized frequency separation of the subcarriers. Note that xn
and X k form an N − point discrete Fourier transform (DFT) pair. The relationship can be
expressed as
24 Recent Advances in Wireless Communications and Networks

1 N − 1 − j 2π kn /N
Xn = DFTN {xn } = ∑ xk e
N k =0
, n = 0,1, ,N - 1 (2)

Fig. 1. The block diagram of the OFDM transceiver

The data symbol X k can be recovered approximately by using a DFT operation at the
receiver if the orthogonality of the OFDM symbol is not destroyed by intersymbol
interference (ISI) and intercarrier interference (ICI). A cyclic prefix (CP) is used in an OFDM
system to prevent ISI and ICI. The CP usually repeats the last L samples of an OFDM block
and then is arranged in front of the block. The resulting symbol sn can be represented as

⎧x , n = −L , −L + 1, , −1
sn = ⎨ N + n (3)
x
⎩ n , n = 0,1, ,N − 1

The transmitted signal may pass through a channel h depending on the environments. The
receiver signal rn can be written as

rn = sn ⊗ h + w (4)

where w denotes the additive white Gaussian noise (AWGN). The data symbol Yn can be
recovered by using a DFT operation and is determined as

1 N −1
Yn = ∑ y k e j 2π kn
N k =0
N
, n = 0,1, ,N - 1 (5)

Fig. 2 (a) shows the spectrum of an OFDM subchannel, and (b) shows an entire OFDM
signal. At the maximum value of each subcarrier frequency, all other subcarrier spectra are
null. The relationship between the OFDM block and CP is depicted clearly in Fig. 3.
The OFDM technique offers reliable effective transmission; however, it is far more
vulnerable to symbol timing error and carrier frequency offset. Sensitivity to symbol timing
offset is much higher in multicarrier communications than in single carrier communications
because of intersymbol interference. The mismatch or instability of the local oscillator
inevitably causes an offset in the carrier frequency that can cause a high bit error rate and
performance degradation because of intercarrier interference. Therefore, the unknown
Synchronization for OFDM-Based Systems 25

OFDM symbol arrival times and mismatch/instability of the oscillators in the transmitter
and the receiver are two significant synchronization problems in the design of OFDM
communications. A detailed description of symbol timing error and carrier frequency offset
is given in the following sections.

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2

-0.4 -0.4
-6 -4 -2 0 2 4 6 -6 -4 -2 0 2 4 6
Frequency Frequency

(a) (b)
Fig. 2. Spectra of (a) an OFDM subchannel and (b) an OFDM signal

Fig. 3. An OFDM symbol with a cyclic prefix

2.2 Synchronization issues

2.2.1 Timing offset
OFDM systems exploit their unique features by using a guard interval with a cyclic prefix to
eliminate intersymbol interference and intercarrier interference. In general, the symbol
timing offset may vary in an interval that is equal to the guard time and does not cause
intersymbol interference or intercarrier interference. OFDM systems have more robustness
to compare with carrier frequency offset. However, a problem arises when the sampling
26 Recent Advances in Wireless Communications and Networks

frequency does not sample an accurate position; the sensitivity to symbol timing offset
increases in OFDM systems. Receivers have to be tracked time-varying symbol timing offset,
which results in time-varying phase changes. Intercarrier interference comes into being
another attached problem. Because an error in the sampling frequency means an error in the
FFT interval duration, the sampled subcarriers are no longer mutually orthogonal. The
deviation is more severe as the delay spread in multipath fading increases; then, the tolerance
for the delay spread is less than the expected value. As a result, timing synchronization in
OFDM systems is an important design issue to minimize the loss of robustness.

2.2.2 Carrier frequency offset

In section 2.1, it is evident that at all OFDM subcarriers are orthogonal to each other when
they have a different integer number of cycles in the FFT interval. The number of cycles is
not an integer in FFT interval when a frequency offset exists. This phenomenon leads to
intercarrier interference after the FFT. The output of FFT for each subcarrier contains an
interfering term with interference power that is inversely proportional to the frequency
spacing from all other subcarriers (Nee & Prasad, 2000). The amount of intercarrier
interference for subcarriers in the middle of the OFDM spectrum is roughly twice as larger
as that at the OFDM band edges because there are more interferers from interfering
subcarriers on both sides. In practice, frequency-selective fading from the Doppler effect
and/or mismatch and instability of the local oscillators in the transmitter and receiver cause
carrier frequency offset. This effect invariably results in severe performance degradation in
OFDM communications and leads to a high bit error rate. OFDM systems are more sensitive
to carrier frequency offset; therefore, compensating frequency errors are very important.

3. Application scenarios
The major objectives for OFDM synchronization include identifying the beginning of
individual OFDM symbol timing and ensuring the orthogonality of each subcarrier. Various
algorithms have been proposed to estimate symbol timing and carrier frequency offset.
These methods can be classified into two categories: data-aided algorithms and non-data-
aided (also called blind) algorithms. By using known training sequences or pilot symbols, a
data-aided algorithm can achieve high estimation accuracy and construct the structure simply.
Data-aided algorithms require additional data blocks to transmit known synchronization
information. Nevertheless, this method diminishes the efficiency of transmission to offer the
possibility for synchronization. Non-data-aided (blind) algorithms were proposed to solve
the inefficiency problem of the data-aided algorithm. Alternative techniques are based on
the cyclic extension that is provided in OFDM communication systems. These techniques
can achieve high spectrum efficiency but are more complicated.
In the data-aided technique, several synchronization symbols are directly inserted between
the transmitted OFDM blocks; then, these pilot symbols are collected at the receiving end to
extract frame timing information. However, the use of pilot symbols inevitably decreases
the capacity and/or throughput of the overall system, thus making them suitable only in a
startup/training mode. The data- aided technique can provide effectively synchronization
with very high accuracy. Thus, it can be used to find coarse timing and frequency offset in
the initial communication link. Several data-aided techniques have been proposed (Classen
& Meyr, 1994, Daffara & Chouly, 1993, Kapoor et al., 1998, Luise & Reggiannini, 1996, Moose,
1994, Warner & Leung, 1993). Moreover, the SNR at the front end in the receiver is often too
Synchronization for OFDM-Based Systems 27

low to ineffectively detect pilot symbols; thus, a blind approach is usually much more
desirable. A non-data-aided technique can adjust the fine timing and frequency after the
preamble signal. Some non-data-aided techniques have been proposed (Bolcskei, 2001, Daffara
& Adami, 1995, Lv et al., 2005, Okada et al., 1996, Park et al., 2004, Van de Beek et al., 1997).

3.1 Non-data-aided method

The cyclic extension has good correlation properties because the initial TCP seconds of each
symbol are the same as the final seconds in OFDM communications. The cyclic prefix is
used to evaluate the autocorrelation with a lag of T . When a peak is found in the correlator
output, the common estimates of the symbol timing and the frequency offset can be
evaluated jointly. The correlation output can be expressed as
TCP
x(t ) = ∫ r (t − τ )r * (t − τ − T )dτ (6)
0

where r ( t ) is the received OFDM signal, x(t ) is the correlator output, τ denotes the timing
offset. The correlator output can be utilized to estimate the carrier frequency offset when the
symbol timing is found. The phase drift between T seconds is equivalent to the phase of the
correlator output. Therefore, the carrier frequency offset can be estimated easily by dividing
the correlator phase by 2π T . The carrier frequency offset denotes the frequency offset
normalized by the subcarrier spacing. Fig. 4 shows the block diagram of the correlator.

Fig. 4. Correlator using the cyclic prefix

3.2 Data-aided method

Although data-aided algorithms are not efficient for transmission, they have high estimation
accuracy and a simple architecture which are especially important for packet transmission.
The synchronization time needs to be as short as possible, and the accuracy must be as high
as possible for high rate packet transmission (Nee & Prasad, 2000). Special OFDM training
sequences in which the data is known to the receiver were developed to satisfy the
requirement for packet transmission. The absolute received training signal can be exploited
for synchronization, whereas non-data-aided algorithms that take advantage of cyclic
extension only use a fraction signal of each symbol. In training sequence methods, the
matched filter is used to estimate the symbol timing and carrier frequency offset. Fig. 5
shows a block diagram of a matched filter. The input signal is the known OFDM training
sequence. The sampling interval is denoted as T . The elements of {c 0 c1 c N − 1 } are
the matched filter coefficients which are the complex signals of the known training
sequence. The symbol timing and carrier offset can be achieved by searching for the
correlation peak accumulated from matched filter outputs.
28 Recent Advances in Wireless Communications and Networks

Fig. 5. Matched filter for the OFDM training sequence

4. Examples
4.1 Example 1: Non-data-aided, CP-based, fractional/fine frequency offset
According to previous researches, very high computational complexity is required for joint
estimation for timing and frequency synchronization. Moreover, one estimate suffers from
performance degradation caused by estimation error of the other. Thus, an effective
technique is proposed (Lin, 2003).

Fig. 6. The OFDM transceiver (Lin, 2003)

Synchronization for OFDM-Based Systems 29

The proposed technique which employs a two-step method that estimates the frame timing
instant and frequency offset by the maximum-likelihood (ML) estimation criterion. First, it
estimates a frame timing instant such that the estimate is completely independent of the
frequency offset estimation with no prior knowledge of the frequency offset; thus, a much
lower estimation error of the frame timing instant is achieved by avoiding any power loss or
phase ambiguity caused by frequency offset. The main reason for this arrangement is that
frame timing instant estimation has to take place completely before frequency offset
estimation because the latter actually requires frame timing information.
The block diagram of the OFDM system investigated here is depicted in Fig. 6. The received
signal can be expressed as

rk = α k sk −θ e j 2πε k /N + nk (7)

where θ is the unknown delay time; α k denotes a channel fade, which has a Rayleigh-
distributed envelope and a uniformly distributed phase; ε denotes the carrier frequency
offset in a subcarrier spacing; and 1 N is the normalized frequency. In accordance with
Jake’s model of a fading channel (Jakes, 1974), α k can be expressed as a complex Gaussian
random process with the autocorrelation function given as

{ }
⎛ T ⎞
E α k1 α k∗2 = J 0 ⎜ 2π f D k1 − k2 u ⎟
⎝ N⎠
(8)

where E{} ⋅ denotes the statistical expectation operation; ∗ denotes taking complex
conjugation; J 0 ( ⋅) is the zeroth-order Bessel function of the first kind; f D is the maximum
Doppler frequency caused directly by relative motion; and Tu is the OFDM block duration,
which actually corresponds to the time interval of an N -sample OFDM block. In a previous
work (Van de Beek et al., 1997), the log-likelihood function for θ and ε can be written as

Λ (θ , ε ) = log f ( r θ , ε )
⎛ ⎞
= log ⎜ ∏ f ( rk , rk + N ) ∏ f ( rk ) ⎟ (9)
⎜ ⎟
⎝ k∈I k∉I ∪ I ′ ⎠
⎛ f ( rk , rk + N ) ⎞
= log ⎜ ∏ ∏ f ( rk ) ⎟
⎜ ⎟
⎝ k∈I f ( rk ) f ( rk + N ) k ⎠
where f ( ⋅) denotes the probability density function; r = [ r1 r2 r2 N + L ] is the
T

observation vector; I = [θ ,θ + 1, ,θ + L − 1] ; and I ′ = [θ + N ,θ + N + 1, ,θ + N + L − 1] . It

must be noted that the correlations among the samples in the observation vector are
exploited to estimate the unknown parameters θ and ε , and they can be written as

⎪ { }
⎧E r 2 = σ 2 + σ 2 ,
k s n m=0
⎪
∀k ∈ I : E {
rk , rk∗+ m } { }
= ⎨E rk , rk∗+ m = σ s2 J 0 ( 2π f DTu ) e − j 2πε , m = N (10)
⎪
⎪0, otherwise
⎩

where σ s2 = E ⎡ sk ⎤ is the average signal power and σ n2 = E ⎡ nk ⎤ is the average noise power.
2 2
⎣ ⎦ ⎣ ⎦
30 Recent Advances in Wireless Communications and Networks

Because the product ∏ k f ( rk ) in (9) is independent of θ and ε , it can be dropped when

maximizing Λ (θ , ε ) . Under the assumption that r is a jointly Gaussian vector and after
some manipulations reported in the reference Appendix (Lin, 2003), (9) can be rewritten as

⎡θ + L − 1
{
Λ (θ , ε ) = c1 + c 2 ⎢ ∑ Re rk rk∗+ m e − j 2πε −
⎣ k =θ
ρ θ + L −1
2
}
∑ rk + rk + N ⎥
k =θ
2 2 ⎤

⎦
( ) (11)
= c1 + c 2 ⎣Re {λ1 (θ )} cos ( 2πε ) − Im {λ1 (θ )} sin ( 2πε ) − ρλ2 (θ ) ⎤⎦
⎡

where

ρ=
{
E rk rk∗+ N } =
σ s2 J 0 ( 2π f DTu )

{ } E{ r }
E rk
2
k+N
2 σ s2 + σ n2

θ + L −1
c1 = − ∑
k =θ
(
log 1 − ρ 2 )
2ρ
c2 =
( )(
1 − ρ 2 σ s2 + σ n2 )
θ + L −1
λ1 (θ ) = ∑ rk rk∗+ N
k =θ

λ2 (θ ) =
1 θ +L −1
∑
2 k =θ
2
{
rk + rk + N
2
}
In the above equation, it is assumed that the random frequency modulation caused by a
time-varying channel fade and the phase noise of the local oscillator are negligible; thus,
{ }
rk rk∗+ N has almost the same phase within the range k ∈ [θ ,θ + L − 1] ; therefore, rk rk∗+ N { }
can be coherently summed up in the term λ1 (θ ) . If the partial derivative of Λ (θ , ε ) is taken
with respect to ε , one can obtain the following equation:

∂
Λ (θ , ε ) = −2π c 2 ⎡⎣Re {λ1 (θ )} sin ( 2πε ) + Im {λ1 (θ )} cos ( 2πε ) ⎤⎦ (12)
∂ε
To obtain the value of εˆ that maximizes Λ (θ , ε ) , the above partial derivative is set to zero
and equality stands only when

Re {λ1 (θ )} Im {λ1 (θ )} 1
= = (13)
cos ( 2πε ) − sin ( 2πε ) c3

where c 3 is set as a constant 1 L for simplicity. As a result, the carrier frequency offset
estimate can be expressed as

1 ⎛ Im {λ1 (θ )} ⎞
εˆ = − tan −1 ⎜ ⎟ (14)
2π ⎜ Re {λ1 (θ )} ⎟
⎝ ⎠
Synchronization for OFDM-Based Systems 31

The carrier frequency offset estimator derived above actually requires accurate frame timing
information to effectively resolve the carrier frequency offset by taking advantage of a
complete cyclic prefix. As a result, accurate frame timing estimation has to be performed
before a carrier frequency offset is estimated.
To develop a frame timing estimation scheme without prior knowledge of frequency offset,
the log-likelihood function in (11) can be approximated as follows:

Λ (θ , ε ) ≈ c1 + c 2 ⎡⎣c 3 Re {λ1 (θ )} ⋅ Re {λ1 (θ )} + c 3 Im {λ1 (θ )} ⋅ Im {λ1 (θ )} − ρλ2 (θ ) ⎤⎦

⎣ ( )
= c1 + c 2 ⎡c 3 Re 2 {λ1 (θ )} + Im 2 {λ1 (θ )} − ρλ2 (θ ) ⎤
⎦
(15)

⎡ 2
= c1 + c 2 c 3 λ1 (θ ) − ρλ2 (θ ) ⎤
⎢⎣ ⎥⎦
Thus, one can obtain a frame timing estimator independent of frequency offset estimation.
The proposed technique provides a more practical estimate of the frame timing instant
because frame timing estimation is very often performed before frequency offset is
estimated or dealt with. As a result, the proposed estimator of the frame timing instant and
frequency offset can be expressed as

p
θ
3 1{
⎧Step 1: θˆ = arg max c λ (θ ) 2 − ρλ (θ )
⎪ 2 }
⎪
⎨
⎪Step 2: εˆp = −
1
⎛ Im λ θˆ ⎞
tan −1 ⎜
1 p ⎟ { ( )} (16)

⎪
⎩
2π ⎜⎜ Re λ θˆ ⎟⎟
⎝ 1 p
⎠ { ( )}
Its structure is depicted in detail in Fig. 7. The proposed frame timing estimator inherently
exploits the highest signal level by disregarding any phase ambiguity caused by residual
error in frequency offset estimation. Therefore, the proposed technique performs frame
timing estimation in a manner independent of frequency offset estimation; then, frequency
offset estimation can be properly achieved in the next step by effectively taking advantage
of accurate timing information.

Fig. 7. The estimator (Lin, 2003)

Because the effect of fast channel fading is considered here, the proposed technique has to
account for a maximum Doppler frequency fD on the same order of 1/Tu. Therefore, the
proposed estimator of the frame timing instant is often dominated by its first term because
the correlation coefficient term ρ in (16) approaches zero in such an environment. As a
result, estimating of the frame timing instant can be simplified as follows to reduce the
hardware complexity:
32 Recent Advances in Wireless Communications and Networks

{
θˆp′ = arg max λ1 (θ )
θ
2
} (17)

In addition, several techniques for combining multiple frames have also been investigated
(Lin, 2003) to increase the robustness of the proposed technique under low SNR conditions.
Other simulation experiments show that the proposed techniques can effectively achieve
lower estimation errors in frame timing and frequency offset estimation.

4.2 Example 2: Data-aided, preamble, integral/coarse frequency offset

Previous works often employ signal-estimation techniques on a time-indexed basis in the
time direction. However, very few previous works have dealt with frequency-offset
problems by applying a detection technique on a subcarrier-indexed basis in the frequency
direction. An effective technique for frequency acquisition based on maximum-likelihood
detection for mobile OFDM is proposed. The proposed technique employs a frequency-
acquisition stage and a tracking stage. We mainly focus on frequency acquisition because
tracking has been investigated (Lin, 2004, 2006b, 2007). By exploiting differential coherent
detection of a single synchronization sequence, where a pseudonoise (PN) sequence is used
as a synchronization sequence, we can prove that data-aided frequency acquisition with
frequency-directional PN matched filters (MFs) reduces the probabilities of false alarm and
miss on a channel with a sufficiently wide coherence bandwidth. Strict statistical analyses have
been performed to verify the improvements achieved. Furthermore, the proposed technique
can operate well over a channel with severe frequency-selective fading by exploiting
subcarrier-level differential operation and subsequent coherent PN cross-correlation.

Fig. 8. The OFDM transceiver (Lin, 2006a)

In the investigated OFDM system, a PN sequence with a period N p (say, N p < K ) is
successively arranged to form an OFDM preamble block. The complex representation of the
received baseband-equivalent signal can, thus, be written as
Synchronization for OFDM-Based Systems 33

K
1 ⎛ kl ⎞ ⎛ l ⎞
rl =
N
∑ ck Np
exp ⎜ j 2π ⎟ exp ⎜ j 2π ( d + ε ) ⎟ + n′′′l ,
⎝ N⎠ ⎝ N⎠
l = 0,1,… , N − 1 (18)
k =− K

where l denotes the time index, the term exp ( j 2π ( d + ε ) ( 1 N ) ) represents the effect of
the CFO that is mainly caused by instability or mismatch that occurs with the local
oscillator at the front-end down-conversion process, d and ε are the integral and fractional
parts of the CFO, respectively, which are normalized by the subcarrier spacing (i.e., frequency
separation between any two adjacent subcarriers), c k is the k N p th chip value of the PN code
Np

transmitted via the kth subchannel, whose normalized subcarrier frequency is ( k N ) , k N p

denotes the k modulus N p , and n′′′l is complex white Gaussian noise. With the FFT
demodulation, the pth subchannel output can be expressed as
N −1
1 ⎛ pl ⎞
Yp =
N
∑ exp ⎜⎝ − j 2π N ⎟⎠ ⋅ rl
l =0 (19)
K
⎛ N −1
= ∑ ck g( k + d − p + ε ) ⋅ exp ⎜ jπ ( k + d − p + ε ) ⎟⎞ + n′′p , p = − N 2 ,… , N 2 − 1
k =− K
Np
⎝ N ⎠
where
sin (πυ )
g(υ ) =
N sin (πυ N )

and n′′p has a noise term. If the demodulation outputs {Yp , p = 0, 1,… , N p − 1; N p < K} are
cross-correlated with a locally generated PN sequence with a phase delay d̂ using PN MF,
then the output of the PN MF can be obtained.

Np
Z0 =
σ0
( ) ⎛ N −1
g d − dˆ + ε exp ⎜ jπ
⎝ N
( ⎞
d − dˆ + ε ⎟ + n0
⎠
) (20)

The detailed derivation has been shown elsewhere (Lin, 2006a). As a result, coarse frequency
offset can be detected through subcarrier acquisition. The detection procedure is equivalent
to testing the following two hypotheses:

⎧ f 2 (η A1 , H 1 ) ∼ χ 2
⎪ Zo
⎪ sin (πε )
⎪ A1 = g1 ( ε ) = = g ( ε ) , H 1 : d = dˆ ,
⎪⎪ N sin (πε N )
⎨ (21)
⎪ f Zo 2 (η A0 , H 0 ) ∼ χ
2

⎪
⎪ sin (π ( d′ + ε ) )
⎪ A0 = g0 ( ε ) = g ( d′ + ε ) = , H 0 : d ≠ dˆ
N sin (π ( d′ + ε ) N ) d ′= d − d ≠ 0
ˆ
⎪⎩

where H 1 and H 0 denote the two hypothesis that the local PN sequence has been
aligned (i.e., d = dˆ ) and has not been aligned in-phase (i.e., d ≠ dˆ ), respectively, with respect
to the post-FFT-demodulation PN sequence.
The previous derivations show that the major difficulty with the ordinary likelihood
functions results from the very complicated probability density functions of the derived
34 Recent Advances in Wireless Communications and Networks

random variable, A0 = g0 ( ε ) and A1 = g1 ( ε ) . Therefore, the two derived random variables

A0 and A1 are first set to be constant for the worst cases, and thus, the (fixed) noncentrality
parameters can be exploited in the likelihood functions to simplify the detection procedure.
The probabilities of false alarm and miss for noncoherent detection can be written as

Pfanc ( tnc ) = P ( S > tnc H 0 )

f S ⎛⎜ s H 0 ,max g0 ( ε ) ⎞⎟ ds
∞
≤∫ (22)
tnc ⎝ ε ⎠
= Q1 ( λnc ,0 , tnc )
nc
Pms ( tnc ) = P ( S ≤ tnc H 1 )
f ⎛⎜ s H 1 ,min g1 ( ε ) ⎞⎟ ds
∞
≤1−∫ (23)
tnc S ⎝ ε ⎠
= 1 − Q1 ( λnc ,1 , tnc )
where

(( ))
2
sin π d − dˆ + ε Np
λnc ,0 = max 2 g 2 ( 1.5 ) N p ⋅ SNR
N sin (π ( d − dˆ + ε ) N )
ε ≤ 0.5 σ0
d ≠ dˆ
2
Np
λnc ,1
ε ≤ 0.5
(
= max g d − dˆ + ε ) σ0
= 2 g 2 ( 0.5 ) N p ⋅ SNR
d = dˆ

and
( μ −2)
( )
4
∞ 1⎛ x ⎞ ⎛ x + a2 ⎞
Qμ ( a, b ) = ∫b ⎜ ⎟ ⋅ exp ⎜⎜ − ⎟Iμ a2 x dx
2 ⎟⎠ 2 − 1
2
2 ⎝ a2 ⎠ ⎝
is the generalized Marcum Q-function, which is defined as the complementary cumulative
density function of a noncentral χ 2 random variable with μ degrees of freedom and
noncentrality parameter a 2 , and where tnc is a design parameter representing the decision
threshold of the derived noncoherent detection.
The above noncoherent detector can be further improved by a differentially coherent
detection technique that consists of coherent accumulation of cross-correlations subchannel-
by-subchannel by means of PN MFs. The detailed derivation has been provided elsewhere
(Lin, 2006a). As a result, the probability of false alarm and miss for the proposed differentially
coherent subcarrier-acquisition technique is given by

1 −s 2
( )
∞ ∞ ∞
Pfadc = P ( γ a − γ b > tdc H 0 ) = ∫ fγ b ( s H 0 ) ∫ f (η H 0 )dη ds = ∫0 e Q1 λdc ,o , s + tdc ds (24)
0 s + tdc γ a 2

1 −s 2
( )
∞
Pfadc = P ( γ a − γ b ≤ tdc H 1 ) = 1 − ∫ e Q1 λdc ,1 , s + tdc ds (25)
0 2
where
Synchronization for OFDM-Based Systems 35

λdc ,0 = Λ 2 H0 = 4 g 2 ( 1.5 ) N p SNR

d ≠ dˆ

λdc ,1 = Λ 2 H1 = 4 g 2 ( 0.5 ) N p SNR

d = dˆ

and tdc is a design parameter denoting the decision threshold when the above differentially
coherent detection is used.
It can be easily seen from simulation results (Lin, 2006a) that no matter what values of the
decision threshold are chosen, the proposed techniques can achieve sufficiently low
probabilities of false alarm and miss and that differentially coherent detection can achieve
lower probabilities than its noncoherent counterpart. The main reason for this difference is
that differentially coherent detection primarily tests two more distantly separated
distributions than does the noncoherent detection.
Although the previous derivations were conducted only on an AWGN channel, similar
results and conclusions hold for a flat-fading channel or in an environment whose coherence
bandwidth is wide enough to accommodate several subchannels. The relative contexts are
shown completely in the reference paper (Lin, 2006a).

5. Synchronization in LTE/LTE-A systems

5.1 Introduction
Requirement of transmission data rate grows up rapidly as time flies. The Long Term
Evolution (LTE) specification proposed by 3rd Generation Partnership Project (3GPP) has a
significant influence on recent wireless communications. LTE communication systems are
expected to achieve a data rate of 100 Mb/s on downlink and 50 Mb/s on uplink
transmissions; it can also provide flexible bandwidths of 1.4, 2.5, 5, 10, 15 and 20 MHz. An
LTE communication is based on the OFDM techniques and adopts single-carrier frequency-
division multiple access (SC-FDMA) on uplink transmission and OFDMA on downlink
transmission. It is clear that LTE can provide a high data rate, robust performance over
multipath fading channels and high spectral efficiency. However, an LTE system has a
major drawback: it is sensitive to frequency error as OFDM systems. Timing and frequency
synchronization is a key component for initial synchronization of an LTE system. For a link
initiative, a mobile station has to search for a base-station by means of synchronization
sequences, which are broadcasted in all directions in which the station provides coverage.
This search is called cell search in cellular systems. In the cell search, a sector search must be
performed at first. The following tasks comprise the sector search: coarse timing alignment,
fine timing synchronization, fine frequency offset compensation, coarse frequency offset
detection, and sector identification.

5.2 LTE frame structure

An LTE supports 504 different cell identifications. To avoid the high complexity of a cell
search procedure, these cell identifications are categorized into 168 cell-identification groups,
N (1) (2)
ID ; additionally, each cell-identification group contains three identities, N ID . Therefore,
cell (1) (2)
cell identification (ID) can be stated as N ID = 3N ID + N ID . Initially, the sector of the received
signal has to be identified. Then, the cell that can provide service must be identified. After
the above procedure is completed, communication can begin. An LTE supports two
36 Recent Advances in Wireless Communications and Networks

synchronization signals for the cell search procedure. One is the primary synchronization
signal (P-SCH), and the other is the secondary synchronization signal (S-SCH). P-SCH and
S-SCH are inserted into the last two OFDM symbols in the first slot of the sub-frame zero
and sub-frame five, where the frame structure is shown in Fig. 9. The P-SCH signal is
transmitted twice in each 10-ms frame. It can provide frame timing synchronization with a
tolerance of 5 ms. The main goal of the P-SCH is to conduct timing synchronization, coarse
frequency-offset detection and sector identification. Each frame has a pair of S-SCH signals
that can be chosen from the 168 different cell identifications. Therefore, the S-SCH signal is
used to determine the cell ID.

Fig. 9. LTE frame structure

The frame structure of the LTE system is depicted in Fig. 9, and the length of each frame is
10 ms. Each frame is divided into ten 1-ms sub-frames. Each sub-frame contains two slots
with lengths 0.5 ms. Additionally, each slot consists of seven symbols, and each symbol
contains 2048 samples. The zeroth and fifth sub-frame convey P-SCH and S-SCH signals.
According to the LTE specification, the CP length is 160 samples in the first symbol of a slot
and 144 samples in the other 6 symbols of the slot. When the occupied bandwidth is 20
MHz, the system parameters are as follows: the sampling rate is 30.72 MHz, the FFT size is
2048, and the subcarrier spacing is 15 KHz. The synchronization signals occupy only the
central 72 sub-carriers of the 2048 sub-carriers. Both boundaries of the band conveying the
synchronization signals accommodate 5 null subcarriers. Therefore, the synchronization
signals only adopt 62 subcarriers.

5.3 P-SCH signal

The number of physical-layer cell identifications is very large. To avoid high complexity in
the cell search, the cell identifications are partitioned into three sets, physical-layer cell-
(2)
identification group N ID or sector. The P-SCH signal is composed of three Zadoff-Chu (ZC)
Synchronization for OFDM-Based Systems 37

sequences with lengths of 62 in the frequency domain. Each sequence represents a sector
identification. The ZC sequences employed in the LTE (3GPP LTE, 2005) are written as

⎧ π un(n + 1)
⎪ −j 63
⎪e , n = 0,1,...., 30
du (n) = ⎨ (26)
⎪ −j π u( n + 1)( n + 2)
⎪e 63 , n = 31, 32,....,61.
⎩

N (2)
ID Root index u
0 25
1 29
2 34
Table 1. Root index u of sector identification (3GPP LTE, 2005)
where u is the root index for which values are set to 25, 29, and 34, which correspond to
(2)
N ID =0,1 or 2, respectively. A ZC sequence is a chirp-like sequence and is symmetric both in
the time domain and frequency domain. The sequence has good correlation properties.
Therefore, the P-SCH signal employing the ZC sequence is utilized to help coarse timing
synchronization and frequency-offset detection.

5.4 Cell search method

Research regarding sector search in LTE systems has been studied extensively (Chen et al.,
2009, Manolakis et al., 2009, Tsai et al., 2007). Three methods were studied previously (Tsai
et al., 2007). They mainly take advantage of auto-correlation, cross-correlation and hybrid
detection. The first method adopts auto-correlation to search for P-SCH location by taking
advantage of the repetitions of P-SCH sequences. Coarse frequency-offset acquisition depends
on the output of the auto-correlator. Its main advantage is low complexity, but the timing
detection is inevitably distorted on signals with low SNR. The second method employs
cross-correlation between the received signal and the locally-generated P-SCH to detect timing
and frequency offset. Additionally, the cross-correlation can be divided into several segments
to counter any effect caused by a high frequency offset. The method has a trustworthy
timing detection, but its complexity is higher than auto-correlation detection. Hybrid
detection combines advantages of auto-correlation and cross-correlation. Its complexity is
less than that employing cross-correlation detection. The auto-correlation detection obtains
coarse timing and frequency offset, and compensates for the frequency error. Then, the
accurate timing can be obtained by exploiting cross-correlation.A previous study (Manolakis
et al., 2009) used maximum likelihood (ML) criterion to estimate the fractional frequency
offset and the OFDM symbol timing; its performance is improves by averaging 8 OFDM
symbols. Next, cross-correlation between the three P-SCH sequences and the received signal
is obtained; and the sector ID can be determined by selecting the highest cross-correlation
according to the orthogonality among the used Zadoff-Chu sequences.

5.5 Carrier aggregation

Carrier aggregation is one of the most important technologies in the new LTE-Advanced
standards. This technique will also play a significant role for 4G communication systems. By
38 Recent Advances in Wireless Communications and Networks

using carrier aggregation, a peak data rate up to 1 Gb/s is possible in future 4G mobile
communications. Because of the flexibility of effective transmission, the user can exploit
numerous carriers at the same time. In addition, these carriers may lie in the same or
different band and may have different bandwidths. Carrier aggregation provides diverse
combinations and flexible spectrum usability and has attracted attention. Carrier aggregation
techniques can be classified into two categories: continuous and discontinuous as shown in
Fig. 10. These two categories can be subdivided into three types: intraband contiguous,
intraband discontinuous and interband. A diagram describes their difference in Fig. 11.

Fig. 10. Carrier aggregation types: (a) intraband contiguous; (b) intraband discontinuous;
(c) interband (Iwamura et al., 2010)

Fig. 11. Carrier aggregation categories: (a) continuous; (b) discontinuous (Yuan et al., 2010)

6. Summary
In this chapter, the authors intend to introduce the OFDM communication systems and take
care of the main issue, frequency offset, can lead to severe performance degradation. Two
classifications of synchronization techniques are introduced. Several novel techniques have
been thoroughly discussed in great detail in this chapter. LTE/LTE-A systems have been
chosen as candidates for 4G mobile communication. The concept of LTE-LTE-A systems is
mentioned in the end of this chapter.
Synchronization for OFDM-Based Systems 39

7. References
3GPP LTE (2005). TS 36.211 V8.3.0: Technical Specification Group Radio Access Network;
Evolved Universal Terrestrial Radio Access (E-UTRA); Physical Channels and
Modulation (Release 8)
Bolcskei, H. (2001). Blind estimation of symbol timing and carrier frequency offset in
wireless OFDM systems, IEEE Transactions on Communications, Vol.49, No.6, (June
2001), pp.988-999
Chen, Y., Wen, X., Zheng, W. & Lin, X. (2009). Symbol timing estimation and sector detection
algorithm based on LTE TDD system, Proceedings of IEEE Network Infrastructure and
Digital Content Conference, 2009 (IC-NIDC 2009), Beijing, China, pp.828-832.
Classen, F. & Meyr, H. (1994). Frequency synchronization algorithms for OFDM systems suitable
for communication over frequency selective fading channels, Proceedings of IEEE
Vehicular Technology Conference, 1994 (VTC’94), Stockholm, Sweden, pp. 1655-1659
Daffara, F. & Chouly, A. (1993). Maximum likelihood frequency detectors for orthogonal
multicarrier systems, Proceedings of IEEE Communications Conference, 1993 (ICC’93),
Geneva, Switzerland, pp. 766-771
Daffara, F. & Adami, O. (1995). A new frequency detector for orthogonal multicarrier
transmission techniques, Proceedings of IEEE Vehicular Technology Conference, 1995
(VTC’95), Chicago, USA, pp. 804-809
Dahlman, E., Parkvall, S., Skold, J. & Beming, P. (2007) 3G Evolution HSPA and LTE for Mobile
Broadband, Academic Press
Iwamura, M., Etemad, K., Fong M.-H., Nory, R. & Love, R. (2010) Carrier aggregation
framework in 3GPP LTE-advanced [WiMAX/LTE update], IEEE Communications
Magazine, Vol.48, No.8, (August 2010), pp.66-67
Jakes, W. C. & Cox, D. C. (1994). Microwave Mobile Communications. Wiley-IEEE Press
Kapoor, S., Marchok, D. J. & Huang, Y.-F. (1998). Pilot assisted synchronization for wireless
OFDM systems over fast time varying fading channels, Proceedings of IEEE Vehicular
Technology Conference, 1998 (VTC’98), Ottawa, Canada, pp. 2077-2080
Lin, J.-C. (2002a). Noncoherent sequential PN code acquisition using sliding correlation for
chip-asynchronous DS/SS, IEEE Transactions on Communications, Vol.50, No.4,
(April 2002), pp.664-676
Lin, J.-C. (2002b). Differentially coherent PN code acquisition with full-period correlation in
chip-asynchronous DS/SS receivers, IEEE Transactions on Communications, Vol.50,
No.5, (May 2002), pp.698-702
Lin, J.-C. (2002c). Differentially coherent PN code acquisition based on a matched filter for
chip-asynchronous DS/SS communications, IEEE Transactions on Vehicular
Technology, Vol,51, No.6, (November 2002), pp.1596-1599
Lin, J.-C. (2003). Maximum-likelihood frame timing instant and frequency offset estimation
for OFDM communication over a fast Rayleigh-fading channel, IEEE Transactions
on Vehicular Technology, Vol,52, No.4, (July 2003), pp.1049-1062
Lin, J.-C. (2004). Frequency offset estimation by differentially coherent frequency error
characterization for OFDM wireless communications, Proceedings of IEEE
Communications Conference, 2004 (ICC’04), Paris, France, pp. 2387-2391.
Lin, J.-C. (2005). Frequency offset acquisition based on subcarrier differential detection for
OFDM communications on doubly-selective fading channel, Proceedings of IEEE
Communications Conference, 2005 (ICC’05), Seoul, Korea, pp. 1952-1956.
40 Recent Advances in Wireless Communications and Networks

Lin, J.-C. (2006a). Coarse frequency-offset acquisition via subcarrier differential detection for
OFDM communications, IEEE Transactions on Communications, Vol.54, No.8,
(August 2006), pp.1415-1426
Lin, J.-C. (2006b). Frequency offset estimation technique based on error characterization for
OFDM communications on time-varying multipath fading channels, Proceedings of
IEEE Communications Conference, 2006 (ICC’06), Istanbul, Turkey, pp.2911-2916.
Lin, J.-C. (2007). A frequency offset estimation technique based on frequency error
characterization for OFDM communications on multipath fading channels, IEEE
Transactions on Vehicular Technology, Vol,56, No.3, (May 2007), pp.1209-1222
Luise, M. & Reggiannini, R. (1996). Carrier frequency acquisition and tracking for OFDM
systems, IEEE Transactions on Communications, Vol.44, No.11, (November 1996),
pp.1590-1598
Lv, T., Li, H. & Chen, J. (2005) Joint estimation of symbol timing and carrier frequency offset
of OFDM signals over fast time-varying multipath channels, IEEE Transaction on
Signal Processing, Vol.53, No.12, (December 2005), pp.4526-4535
Manolakis, K., Gutierrez Estevez, D. M., Jungnickel, J., Xu, W. & Drewes, C. (2009) A closed
concept for synchronization and cell Search in 3GPP LTE systems, Proceedings of
IEEE Wireless Communications and Networking Conference, 2009 (WCNC 2009),
Budapest, Hungary, pp. 1-6.
Moose, P. H. (1994). A technique for orthogonal frequency division multiplexing frequency
offset correction, IEEE Transactions on Communications, Vol.42, No.10, (October
1994), pp.2908-2914
Nee, V. R. & Prasad, R., (2000). OFDM for Wireless Multimedia Communications, Artench House
Okada, M., Hara, S., Komaki, S. & Morinaga, N. (1996). Optimum synchronization of
orthogonal multi-carrier modulated signals, Proceedings of IEEE Personal, Indoor and
Mobile Radio Communications Conference, 1996 (PIMRC’96), Taipei, Taiwan, pp. 863-867
Park, B., Ko, E., Cheon, H., Kang, C. & Hong, D. (2001). A Blind OFDM synchronization
algorithm based on cyclic correlation, IEEE Signal Processing Letters, Vol.11, No.2,
(February 2004), pp.83-85.
Popovic, B. M. (1992). Generalized chirp-like polyphase sequences with optimum
correlation properties, IEEE Transactions on Information Theory, Vol.38, No.4, (July
1992), pp.1406-1409
Van de Beek, J.-J., Sandell, M. & Borjesson, P. O. (1997). ML estimation of time and
frequency offset in OFDM systems, IEEE Transaction on Signal Processing, Vol.45,
No.7, (July 1997), pp. 1800-1805
Tsai, Y., Zhang, G., Grieco, D. & Ozluturk, F. (2007). Cell searrch in 3GPP Long Term Evolution
systems, IEEE Vehicular Technology Magazine, Vol.2, No.2, (June 2007), pp.23-29
Warner, W. D. & Leung, C. (1993). OFDM/FM frame synchronization for mobile radio data
communications, IEEE Transactions on Vehicular Technology, Vol,42, No.3, (August
1993), pp.302-313
Yuan, G., Zhang, X., Wang, W. & Yang, Y. (2010). Carrier Aggregation for LTE-advanced
mobile communication systems, IEEE Transaction on Communication Magazine,
Vol.48, No.2, (February 2010), pp.88-93
 3

ICI Reduction Methods in OFDM Systems

Nadieh M. Moghaddam and Mohammad Mohebbi
Iran University of Science and Technology
Iran

1. Introduction
The principles of multicarrier modulation have been in existence for several decades.
However, in recent years these techniques have quickly moved out of textbooks and into
practice in modern communications systems in the form of orthogonal frequency division
multiplexing (OFDM). OFDM is a special form of multicarrier modulation technique which
is used to generate waveforms that are mutually orthogonal and then distributes the data
over a large number of carriers that are spaced apart at precise frequencies. This spacing
provides the "orthogonality" in this technique which prevents the demodulators from seeing
frequencies other than their own. In an OFDM scheme, a large number of orthogonal,
overlapping, narrow band subcarriers are transmitted in parallel. These carriers divide the
available transmission bandwidth. The separation of the subcarriers is such that there is a
very compact spectral utilization. With OFDM, it is possible to have overlapping sub
channels in the frequency domain (Figure 1), thus increasing the transmission rate.

Fig. 1. Power spectrum of the transmitted signal

In order to avoid a large number of modulators and filters at the transmitter and
complementary filters and demodulators at the receiver, it is desirable to be able to use
modern digital signal processing techniques, such as fast Fourier transform (FFT).
OFDM is a promising candidate for achieving high data rates in mobile environment
because of its multicarrier modulation technique and ability to convert a frequency selective
fading channel into several nearly flat fading channels.
This technology has been chosen as the transmission method of many standards, such as
Digital Subscribe Line (DSL), European Digital Audio and Video Broadcasting terrestrial
(DAB/DVB-T), European HIPERLAN/2 and IEEE 802.11 a/g for wireless local area
networks (WLAN), Worldwide Interoperability for Microwave Access (WiMAX), etc.
However, OFDM systems exhibit a sensitivity to phase noise higher than single carrier
modulations due to its long symbol period. Because carriers are kept very close to each
other, OFDM is very sensitive to distortion that may remove the orthogonality between
carriers. The crystal oscillator used in a mixer generates phase noise. It can also be caused by
42 Recent Advances in Wireless Communications and Networks

AWGN present at the input of a Phase Locked Loop (PLL) in a coherent receiver. Phase
noise can cause several types of signal degradation that are usually very difficult to quantify
analytically. When the modulation experiences phase noise, it encounters two problems: 1) a
common phase rotation over all the carrier frequencies which rotate the entire signal space
for a given OFDM symbol and 2) inter-carrier interference due to the loss of orthogonality
between subcarriers. Especially, the ICI seriously degrades system predominance because it
may break down the orthogonality between subcarriers.
There have been many previous works on the phase noise, frequency offset and reduction of
ICI. Among them the following methods are discussed and compared in this chapter. In the
next section the OFDM system is introduced and its benefits along with its drawbacks are
analyzed. ICI reduction methods such as pulse shaping and self-cancellation are given in
section 3 and the last section concludes the chapter.

2. OFDM system
Figure 2 shows the block diagram of a typical OFDM system. The transmitter section
converts digital data to be transmitted, into a mapping of subcarrier amplitude and phase. It
then transforms this spectral representation of the data into the time domain using an
Inverse Discrete Fourier Transform (IDFT). The Inverse Fast Fourier Transform (IFFT)
performs the 20 same operations as an IDFT, except that it is much more computationally
efficient, and so is used in all practical systems. In order to transmit the OFDM signal the
calculated time domain signal is then mixed up to the required frequency. The receiver
performs the reverse operation of the transmitter, mixing the RF signal to base band for
processing, then using a Fast Fourier Transform (FFT) to analyze the signal in the frequency
domain. The amplitude and phase of the subcarriers is then picked out and converted back
to digital data. The IFFT and the FFT are complementary function and the most appropriate
term depends on whether the signal is being received or generated. In cases where the
signal is independent of this distinction then the term FFT and IFFT is used interchangeably.
The high data rate serial input bit stream is fed into serial to parallel converter to get low
data rate output parallel bit stream. Input bit stream is taken as binary data. The low data
rate parallel bit stream is modulated in Signal Mapper. Modulation can be BPSK, QPSK,
QAM, etc. The modulated data are served as input to inverse fast Fourier transform so that
each subcarrier is assigned with a specific frequency. The frequencies selected are
orthogonal frequencies. In this block, orthogonality in subcarriers is introduced. In IFFT, the
frequency domain OFDM symbols are converted into time domain OFDM symbols. Guard
interval is introduced in each OFDM symbol to eliminate inter symbol interference (ISI). All
the OFDM symbols are taken as input to parallel to serial data. These OFDM symbols
constitute a frame. A number of frames can be regarded as one OFDM signal. This OFDM
signal is allowed to pass through digital to analog converter (DAC). In DAC the OFDM
signal is fed to RF power amplifier for transmission. Then the signal is allowed to pass
through additive white Gaussian noise channel (AWGN channel). At the receiver part, the
received OFDM signal is fed to analog to digital converter (ADC) and is taken as input to
serial to parallel converter. In these parallel OFDM symbols, Guard interval is removed and
it is allowed to pass through Fast Fourier transform. Here the time domain OFDM symbols
are converted into frequency domain. After this, it is fed into Signal Demapper for
demodulation purpose. And finally the low data rate parallel bit stream is converted into
high data rate serial bit stream which is in form of binary.
ICI Reduction Methods in OFDM Systems 43

Fig. 2. OFDM system implementation

By the insertion of an extra guard interval between successive OFDM symbols the Inter
Symbol Interference (ISI) can be avoided. The guard interval could be a section of all zero
samples transmitted in front of each OFDM symbol and its duration should be more than
the channel delay spread (Lc). It should be considered that in practical systems the guard
interval is not used. Instead, Cyclic Prefix (CP) is inserted to combat the multipath-channel
by making the channel estimation simple. The cyclic prefix is a replica of the last Lp samples
of the OFDM symbol where Lp > Lc. Because of the way in which the cyclic prefix was
formed, the cyclically-extended OFDM symbol now appears periodic when convolved with
the channel. An important result is that the effect of the channel becomes multiplicative.
For the better understanding of this issue assume that the impulse response of the channel is
, ,…, and the i-th transmitted signal block in the output of IFFT block is
, , , , … , , . In this condition the cyclic prefix would be , , , ,…, , . The
symbols of the received baseband signal after the transmission through the channel are
equal to:

, , , ,
, , , ,

, , , , (1)
, , , ,

, , , ,

At the receiver the first Lc+1 symbols are discarded and the N remained symbols are
demodulated using an N-point FFT. So the data on the k-th subcarrier is as follows:
44 Recent Advances in Wireless Communications and Networks

, , , , (2)

where is the channel impulse response in the frequency domain. It can also be considered
as the channel gain on the k-th subcarrier. , ∑ , is the input symbols of the
IFFT block at the transmitter as:

, _ , (3)
It can be seen that the main symbol stream , could be detected with the estimation of
the channel coefficient at the receiver, while there is no Inter-Symbol Interference (ISI) or
Inter-Carrier Interference (ICI).
Thus, a multipath channel is converted into scalar parallel sub-channels in frequency
domain, thereby simplifying the receiver design considerably. The task of channel
estimation is simplified, as we just need to estimate the scalar coefficients for each sub-
channel and once the values of are estimated, for the duration in which the channel
does not vary significantly, merely multiplying the received demodulated symbols by the
inverse of yields the estimates of , .
The benefits of OFDM are high spectral efficiency, resiliency to RF interference, and lower
multi-path distortion. This is useful because in a typical terrestrial broadcasting scenario
there are multipath-channels (i.e. the transmitted signal arrives at the receiver using various
paths of different length). Since multiple versions of the signal interfere with each other
(inter symbol interference (ISI)) it becomes very hard to extract the original information.
With the rapid growth of digital communication in recent years, the need for high speed
data transmission is increased. Moreover, future wireless systems are expected to support a
wide range of services which includes video, data and voice. OFDM is a promising
candidate for achieving high data rates in mobile environment because of its multicarrier
modulation technique and ability to convert a frequency selective fading channel into
several nearly flat fading channels.
However, there are some non-idealities which can affect the performance of an OFDM
system. These non-idealities are as follows:
Noise: Like other communication systems, the performance of an OFDM system is affected
by different kind of noise such as uniform noise (AWGN), non-uniform noise (colored
noise), and impulse noise.
LO phase offset: This condition occurs when there is a difference between the phase of the
output LO and the phase of the received signal.
FFT window location offset: In practice, a correlation is often used with a known preamble
sequence located at the beginning of the transmission. This correlation operation aids the
receiver in synchronizing itself with the received OFDM symbol boundaries. However,
inaccuracies still remain, and they manifest themselves as an offset in the FFT window
location. The result is that the N symbols sent to the FFT will not line up exactly with the
corresponding OFDM symbol.
Sampling frequency offset: A sampling frequency offset occurs when the A/D converter
output is sampled either too fast or too slow.
Non-linearity in the transmitter and receiver circuits: All transmitters and receivers in
communication systems contain devices such as amplifiers which are often designed to be
non-linear in order to minimize power consumption. On the other hand, an OFDM signal is
ICI Reduction Methods in OFDM Systems 45

made up of multiple simultaneous signals that, for a given average power, have a higher
peak signal level. Thus, OFDM signals result in an increase in the peak-to-average ratio
(PAR) of the signal. Because of the non-linear transfer functions of amplifiers, these higher
peak amplitude levels will create more severe distortion than a single carrier case even if the
average power levels of each are the same.
Phase noise: OFDM systems are very sensitive to phase noise caused by oscillator
instabilities in both the transmitter and the receiver. Without loss of generality, in this study
the local oscillator in the receiver will be considered as the phase noise source. As
mentioned before, the modulated subcarriers overlap spectrally, but since they are
orthogonal over symbol duration, they can be easily recovered as long as the channel and
other non-idealities do not destroy the orthogonality. An unwindowed OFDM system has
rectangular symbol shapes. Therefore, in the frequency domain the individual sub-channels
will have the form of sinc functions where the first sidelobe is only some 13 dB below the
main lobe of the subcarrier (Figure 3). A practical oscillator has spectral components around
the centre frequency. These components cause the loss of orthogonality of the OFDM
carriers. In the frequency domain it can be viewed as interference caused by the high
sidelobes of the adjoining carriers on a particular subcarrier.

Fig. 3. OFDM spectrum with 5 subcarriers

In this study the phase noise is resolved into two components, namely the Common Phase
Error (CPE), which affects all the sub-channels equally and the Inter-Carrier Interference
(ICI), which is caused by the loss of orthogonality of the subcarriers.
As described above, the IFFT of D is given by d . The k-th sample of can be represented
by:

1
, , 0,1, … , 1 (4)
46 Recent Advances in Wireless Communications and Networks

The phase noise is modeled as a phasor e , where the phase noise process is zero-
mean and wide-sense stationary with a finite variance . An approximation for the PSD of
a free-running oscillator can be found in [Robertson & Kaiser, 1995] is as follows:

10 | |
10 | |
(5)
10 | |

Fig. 4. Phase noise PSD of a typical oscillator

Parameter c determines the noise floor of the oscillator and a determines the noise level in
the frequency ranges from the center frequency to . Parameter b gives the noise fall off rate
from the noise floor at to the noise level at . (See Figure 4).
The demodulated data symbol of the k-th subcarrier of the i-th OFDM symbol , with the
consideration of phase noise is given by:

/ (6)
, , 0,1, … , 1

Using (2), (6) and considering AWGN,

1 / /
, , (7)

where is the contribution due to AWGN on the k-th subcarrier. We can further
simplify (7) as:

1 1 /
, , , (8)

The first term on the right-hand side of (8) rotates the useful component , of each
subcarrier by an equal amount and is independent of the particular subchannel concerned, k.
ICI Reduction Methods in OFDM Systems 47

This is commonly known as the Common Phase Error (CPE). The second term is the Inter-
Carrier Interference (ICI) caused by contributions from all subcarriers on k due to the
loss of orthogonality. Unlike the CPE, ICI is not easy to estimate.
Local oscillator frequency offset: It drives from the difference between LO frequency at the
transmitter and the receiver. In addition to phase noise, the frequency offset which is caused
by the high Doppler spread and the mismatch in the oscillator frequency, produces inter-
carrier interference (ICI).
For a block of data, the modulated signal at the n-th instant of time can be written as:

1 /
, 0,1, … , 1 (9)

The signal at the receiver, after passing through a frequency selective fading channel is
expressed as:

1 / (10)

At the receiver, the signal is mixed with a local oscillator signal which is ∆ above the
correct carrier frequency. The signal received at the k-th subcarrier after performing the
FFT is expressed as:

1 1

(11)
0 , 0,1, … , 1

Where ∆ is the frequency offset normalized to the OFDM symbol rate 1/ . The
denotes the OFDM symbol duration excluding the guard interval. The sequence is
defined as the ICI coefficient between l-th and k-th subcarriers, which can be expressed as:

1
1 (12)

The first term in the right-hand side of (11) represents the desired signal. Without frequency
error 0 , 0 takes its maximum value. The second term is the ICI components, which as
ε becomes larger, the desired part | 0 | decreases and the undesired part | | increases.
The undesired ICI degrades the performance of the system. It is not possible to make
reliable data decisions unless the ICI powers of OFDM system are minimized. Thus, an
accurate and efficient Inter-Carrier Interference (ICI) reduction procedure is essential to
demodulate the received data. Several methods have been presented to reduce ICI,
including windowing at the receiver [Muschallik, 1996; Müller-Weinfurtner, 2001; Song &
Leung, 2005], the use of pulse shaping [Tan & Beaulieu, 2004; Mourad, 2006; Maham &
Hjørungnes, 2007], self-cancellation schemes [Zhao & Haggman, 2001], and frequency
domain equalization. The next part of this chapter introduces these techniques.
48 Recent Advances in Wireless Communications and Networks

3. ICI reduction techniques

In the OFDM systems, N subcarriers are used for data transmission of N symbols
, , , ,… , . By using the IFFT operation for the data modulation, rectangular pulse
shaping filter is implicitly applied. Thus, the spectrum of each individual subcarrier equals a
sinc-function defined as / and is given by:

, , 0,1, … , 1 (13)
where represents the frequency f shifted to the carrier frequency of the OFDM system
and normalized to the sampling frequency 1/ . The normalized frequency is given by:

(14)
Accordingly, is defined as the normalized center frequency of the k-th
subcarrier with representing the center frequency of the k-th subcarrier. The spectrum of
the transmitted OFDM symbol is the superposition of the spectra of all individual
subcarriers:

(15)

The sidelobe power of this sum signal and also the sidelobe power of each subcarrier
spectrum only decays with 1/z resulting in a high interference caused by the high sidelobes
of the adjoining carriers on a particular subcarrier.
Here, some techniques introduced to reduce the power of the interfering components.

3.1 Pulse shaping

As we have seen in the OFDM spectrum each carrier consists of a main lobe followed by a
number of sidelobes with reducing amplitudes. As long as orthogonality is maintained,
there is no interference among the carriers because at the peak of the every carrier, there
exists a spectral null. That is at that point that the component of all other carriers is zero.
Hence the individual carrier is easily separated.
In the presence of the frequency offset the orthogonality is lost because the spectral null
does not coincide to the peak of the individual carriers. So some power of the sidelobes
exists at the centre of the individual carriers which is called ICI power. The ICI power will
go on increasing as the frequency offset increases. The purpose of pulse shaping is to reduce
the sidelobes which leads to the significant decrease in the ICI power.
In a simple OFDM system, symbols are performed using an N-FFT function. This implies
that the received signal r(k) is shaped in the time domain by a rectangular pulse function.
One possible countermeasure to overcome the interference is making the PDS of the OFDM
modulated subcarriers ( ) go down more rapidly by shaping the transmit signal of the
OFDM subcarriers. This makes the amplitude go smoothly to zero at the symbol boundaries.
The N-subcarrier OFDM block with pulse-shaping is expressed as:

(16)
ICI Reduction Methods in OFDM Systems 49

where p(t) is the pulse shaping function. The transmitted symbol is assumed to have zero
mean and normalized average symbol energy. Also we assume that all data symbols are
uncorrelated, i.e.:

1, , , 0,1, … , 1
(17)
0, , , 0,1, … , 1
where is the complex conjugate of . To ensure the subcarrier orthogonality, which is
very important for OFDM systems the equation below has to be satisfied:

, , 0,1, … , 1 (18)

In the receiver block, the received signal can be expressed as:

(19)
where denotes convolution and h(t) is the channel impulse response. In (19), w(t) is the
additive white Gaussian noise process with zero mean and variance N /2 per dimension.
For this work we assume that the channel is ideal, i.e., h(t) = δ(t) in order to investigate the
effect of the frequency offset only on the ICI performance. At the receiver, the received
signal r ′ t becomes:

∆ ∆ (20)
′

Where θ is the phase error and ∆ is the carrier frequency offset between transmitter and
receiver oscillators. For the transmitted symbol , the decision variable is given as
∞
′
(21)
∞

By using (18) and (21), the decision variable can be expressed as

∆ ∑ ∆ , 0, . . . , 1 (22)

where P(f) is the Fourier transform of p(t) and is the independent white Gaussian noise
component. In (22), the first term contains the desired signal component and the second
term represents the ICI component. With respect to (18), P(f) should have spectral nulls at
the frequencies 1/ , 2/ , . .. to ensure subcarrier orthogonality. Then, there exists no
ICI term if ∆ and θ are zero.
The power of the desired signal can be calculated as [Tan & Beaulieu, 2004; Mourad, 2006;
Kumbasar & Kucur, 2007]:

∆ ∆ | ∆ | | ∆ | (23)
The power of the ICI can be stated as:

∆ ∆ (24)
50 Recent Advances in Wireless Communications and Networks

The average ICI power across different sequences can be calculated as:

∆ (25)

As seen in (25) the average ICI power depends on the number of the subcarriers and P(f) at
frequencies: ∆ , , 0,1, … , 1
The system ICI power level can be evaluated by using the CIR (Carrier-to-Interference
power Ratio). While deriving the theoretical CIR expression, the additive noise is omitted.
By using (23) and (25), the CIR can be derived as [Tan & Beaulieu, 2004; Mourad, 2006;
Kumbasar & Kucur, 2007]:

| ∆ |
(26)
∑ ∆

Therefore, the CIR of the OFDM systems only depends approximately on the normalized
frequency offset. A commonly used pulse shaping function is the raised cosine function that
is defined by:

1 1
, 0
2 2
1, (27)
1 1
, 1
2 2
where α denotes the rolloff factor and the symbol interval is shorter than the total symbol
duration (1 + ) because adjacent symbols are allowed to partially overlap in the rolloff
region. Simulation shows that the benefit of the raised cosine function with respect to the ICI
reduction is fairly low.
A number of pulse shaping functions such as Rectangular pulse (REC), Raised Cosine pulse
(RC), Better Than Raised Cosine pulse (BTRC), Sinc Power pulse (SP) and Improved Sinc
Power pulse (ISP) have been introduced for ICI power reduction. Their Fourier transforms
are given, respectively as [Kumbasar & Kucur, 2007]:

, (28)

, (29)

, (30)

, (31)

, (32)
where (0 1) is the rolloff factor, / 2, a is a design parameter to adjust the
amplitude and n is the degree of the sinc function.
ICI Reduction Methods in OFDM Systems 51

Fig. 5. Comparison of REC, RC, BTRC, SP, and ISP spectrums

Fig. 6. CIR performance for different pulse shapes

The purpose of pulse shaping is to increase the width of the main lobe and/or reduce the
amplitude of sidelobes, as the sidelobe contains the ICI power.
52 Recent Advances in Wireless Communications and Networks

REC, RC, BTRC, SP, and ISP pulse shapes are depicted in Figure 5 for a=1, n=2, and 0.5.
SP pulse shape has the highest amplitude in the main lobe, but at the sidelobes it has lower
amplitude than BTRC. This property provides better CIR performance than that of BTRC as
shown in [Mourad, 2006]. As seen in this figure the amplitude of ISP pulse shape is the
lowest at all frequencies. This property of ISP pulse shape will provide better CIR performance
than those of the other pulse shapes as shown in Figure 6 [Kumbasar & Kucur, 2007].
Figure 5 shows that the sidelobe is maximum for rectangular pulse and minimum for ISP
pulse shapes. This property of ISP pulse shape will provide better performance in terms of
ICI reduction than those of the other pulse shapes. Figure 7 compares the amount of ICI for
different pulse shapes.

Fig. 7. ICI comparison for different pulse shapes

3.2 ICI self-cancellation methods

In single carrier communication system, phase noise basically produces the rotation of
signal constellation. However, in multi-carrier OFDM system, OFDM system is very
vulnerable to the phase noise or frequency offset. The serious inter-carrier interference (ICI)
component results from the phase noise. The orthogonal characteristics between subcarriers
are easily broken down by this ICI so that system performance may be considerably
degraded.
There have been many previous works in the field of ICI self-cancellation methods [Ryu et
al., 2005; Moghaddam & Falahati, 2007]. Among them convolution coding method, data-
conversion method and data-conjugate method stand out.

3.2.1 ICI self-cancelling basis

As it can be seen in eq. 12 the difference between the ICI coefficients of the two consecutive
subcarriers are very small. This makes the basis of ICI self cancellation. Here one data
ICI Reduction Methods in OFDM Systems 53

symbol is not modulated into one subcarrier, rather at least into two consecutive subcarriers.
This is the ICI cancellation idea in this method.
As shown in figure 7 for the majority of l-k values, the difference between ) and
1 is very small. Therefore, if a data pair (a,-a) is modulated onto two adjacent
subcarriers , 1 , then the ICI signals generated by the subcarrier will be cancelled out
significantly by the ICI generated by subcarrier l+1 [Zhao & Haggman, 1996, 2001].
Assume that the transmitted symbols are constrained so that
, … , then the received signal on subcarrier k considering
that the channel coefficients are the same in two adjacent subcarriers becomes:

1 (33)

In such a case, the ICI coefficient is denoted as:

1 (34)
For most of the values, it is found that | ΄ | | |.

Fig. 7. ICI coefficient versus subcarrier index; N=64

For further reduction of ICI, ICI cancelling demodulation is done. The demodulation is
suggested to work in such a way that each signal at the k+1-th subcarrier (now k denotes
even number) is multiplied by -1 and then summed with the one at the k-th subcarrier. Then
the resultant data sequence is used for making symbol decision. It can be represented as:

"
1 2 1 (35)

The corresponding ICI coefficient then becomes:

54 Recent Advances in Wireless Communications and Networks

" 1 2 1 (36)
Figure 8 shows the amplitude comparison of | |,|΄ | and | " | for N=64
and 0.3. For the majority of l-k values, | ΄ | is much smaller than | |, and the
|" | is even smaller than | ΄ |. Thus, the ICI signals become smaller when
applying ICI cancelling modulation. On the other hand, the ICI cancelling demodulation can
further reduce the residual ICI in the received signals. This combined ICI cancelling
modulation and demodulation method is called the ICI self-cancellation scheme.
Due to the repetition coding, the bandwidth efficiency of the ICI self-cancellation scheme is
reduced by half. To fulfill the demanded bandwidth efficiency, it is natural to use a larger
signal alphabet size. For example, using 4PSK modulation together with the ICI self-
cancellation scheme can provide the same bandwidth efficiency as standard OFDM systems
(1 bit/Hz/s).

Fig. 8. Amplitude comparison of | |,| ΄ | and | " |

3.2.1.1 Data-conjugate method
In an OFDM system using data-conjugate method, the information data pass through the
serial to parallel converter and become parallel data streams of N/2 branch. Then, they are
converted into N branch parallel data by the data-conjugate method. The conversion process
is as follows. After serial to parallel converter, the parallel data streams are remapped as the
form of D'2k = Dk , D'2k+1 = -D*k , (k = 0, … , N/2-1). Here, Dk is the information data to the k-th
branch before data-conjugate conversion, and D'2k is the information data to the 2k-th
branch after ICI cancellation mapping. Likewise, every information data is mapped into a
pair of adjacent sub-carriers by data-conjugate method, so the N/2 branch data are extended
to map onto the N sub-carries.
The original data can be recovered from the simple relation of Z'k = (Y2k – Y*2k+1)/2. Here, Y2k
is the 2k-th sub-carrier data, Z'k is the k-th branch information data after de-mapping.
Finally, the information data can be found through the detection process. The complex base-
band OFDM signal after data conjugate mapping is as follows.
ICI Reduction Methods in OFDM Systems 55

. . . , 0 (37)

where, N is the total number of sub-carriers, Dk is data symbol for the k-th parallel branch
and is the i–th sub-carrier data symbol after data-conjugate mapping. d(n) is corrupted by
the phase noise in the transmitter (TX) local oscillator. Furthermore, the received signal is
influenced by the phase noise of receiver (RX) local oscillator. So, it is expressed as:

. . (38)
where s(t) is the transmitted signal, w(t) is the white Gaussian noise and h(t) is the channel
impulse response. and are the time varying phase noise processes generated in
the transceiver oscillators. Here, it is assumed that, and
for simple analysis. In the original OFDM system without ICI self-
cancellation method, the k-th sub-carrier signal after FFT can be written as:

1
. (39)

In the data-conjugate method, the sub-carrier data is mapped in the form of ′ , ′

. Therefore, the 2k-th sub-carrier data after FFT in the receiver is arranged as:

(40)

1
(41)

w2k is a sampled FFT version of the complex AWGN multiplied by the phase noise of RX
local oscillator, and random phase noise process is equal to .
Similarly, the 2k+1-th sub-carrier signal is expressed as:

(42)

In the (40) and (42), corresponds to the original signal with CPE, and corresponds
to the ICI component. In the receiver, the decision variable of the k-th symbol is found
from the difference of adjacent sub-carrier signals affected by phase noise. That is,
′
1 1
2 2 2
(43)
∑

where 1⁄2 is the AWGN of the k–th parallel branch data in the
receiver. When channel is flat, frequency response of channel equals 1. Z'k is as follows.
56 Recent Advances in Wireless Communications and Networks

1
(44)
2

3.3 CPE, ICI and CIR analysis

A. Original OFDM
In the original OFDM, the k-th sub-carrier signal after FFT is as follows:

. (45)

The received desired signal power on the k-th sub-carrier is:

| | | | (46)
ICI power is:

| | (47)

Transmitted signal is supposed to have zero mean and statistically independence. So, the
CIR of the original OFDM transmission method is as follows:

| | | |
∑ | | ∑ | | (48)

B. Data-conversion method
In the data-conversion ICI self-cancellation method, the data are remapped in the form of
′
, ′ .
So, the desired signal is recovered in the receiver as follows:

1
2 1
2 2

1 (49)
2

CPE is as follows:

2
(50)

ICI component of the k-th sub-carrier is as follows:

ICI Reduction Methods in OFDM Systems 57

2 4
. (51)

So
| 2 | | 2 |
(52)
∑ | 2 | ∑ | 2 |

C. Data-conjugate method
In the data conjugate method, the decision variable can be written as follows:

1
(53)
2

Through the same calculation, CPE, ICI and CIR of the data conjugate method are found.

0 (54)
The fact CPE is zero is completely different from the data conversion method whose CPE is
not zero like (14).
Then, ICI of data conjugate method is:

1 4
. . . (55)

The above term is the summation of the signal of the other sub-carriers multiplied by some
complex number resulted from an average of phase noise with spectral shift. This
component is added into the k-th branch data Z ′ . It may break down the orthogonalities
between sub-carriers. So, CIR is:
4
(56)
∑

4. Conclusion
OFDM has been widely used in communication systems to meet the demand for increasing
data rates. It is robust over multipath fading channels and results in significant reduction of
the transceiver complexity. However, one of its disadvantages is sensitivity to carrier
frequency offset which causes attenuation, rotation of subcarriers, and inter-carrier
interference (ICI). The ICI is due to frequency offset or may be caused by phase noise.
The undesired ICI degrades the signal heavily and hence degrades the performance of the
system. So, ICI mitigation techniques are essential to improve the performance of an OFDM
system in an environment which induces frequency offset error in the transmitted signal. In
this chapter, the performance of OFDM system in the presence of frequency offset is
58 Recent Advances in Wireless Communications and Networks

analyzed. This chapter investigates different ICI reduction schemes for combating the
impact of ICI on OFDM systems. A number of pulse shaping functions are considered for
ICI power reduction and the performance of these functions is evaluated and compared
using the parameters such as ICI power and CIR. Simulation results show that ISP pulse
shapes provides better performance in terms of CIR and ICI reduction as compared to the
conventional pulse shapes.
Another ICI reduction method which is described in this chapter is the ICI self cancellation
method which does not require very complex hardware or software for implementation.
However, it is not bandwidth efficient as there is a redundancy of 2 for each carrier. Among
different ICI self cancellation methods, the data-conjugate method shows the best
performances compared with the original OFDM, and the data-conversion method since it
makes CPE to be zero along with its role in significant reduction of ICI.

5. References
Robertson, P. & Kaiser, S. (1995). Analysis of the effects of phase-noise in orthogonal
frequency division multiplex (OFDM) systems, Proceedings of the IEEE International
Conference on Communications, vol. 3, (Seattle, USA), pp. 1652–1657, June 1995.
Zhao, Y. & Haggman, S.G. (2001). Intercarrier interference self-cancellation scheme for OFDM
mobile communication systems, IEEE Transaction on Communication. pp. 1185–1191.
Muschallik, C. (1996). Improving an OFDM reception using an adaptive Nyquist
windowing, IEEE Transaction Consum. Electron. 42 (3) (1996) 259–269.
Müller-Weinfurtner, S.H. (2001). Optimum Nyquist windowing in OFDM receivers, IEEE
Trans. Commun. 49 (3) (2001) 417–420.
Song, R. & Leung, S.-H. (2005). A novel OFDM receiver with second order polynomial
Nyquist window function, IEEE Communication Letter. 9 (5) (2005) 391–393.
Tan, P. & Beaulieu, N.C. (2004). Reduced ICI in OFDM systems using the better than raised-
cosine pulse, IEEE Communication Letter 8 (3) (2004) 135–137.
Mourad, H.M. (2006). Reducing ICI in OFDM systems using a proposed pulse shape,
Wireless Person. Commun. 40 (2006) 41–48.
Kumbasar, V. & Kucur, O. (2007). ICI reduction in OFDM systems by using improved sinc
power pulse, ELSEVIER Digital Signal Processing 17 (2007) 997-1006
Zhao, Y. & Häggman, S.-G. (1996). Sensitivity to Doppler shift and carrier frequency errors
in OFDM systems—The consequences and solutions, Proceeding of IEEE 46th
Vehicular Technology Conference, Atlanta, GA, Apr. 28–May 1, 1996, pp. 1564–1568.
Ryu, H. G.; Li, Y. & Park, J. S. (2005). An Improved ICI Reduction Method in OFDM
Communication System, IEEE Transaction on Broadcasting, Vol. 51, No. 3, September
2005.
Mohapatra, S. & Das, S. (2009). Performance Enhancement of OFDM System with ICI
Reduction Technique, Proceeding of the World Congress on Engineering 2009, Vol. 1,
WCE 2009, London, U.K.
Moghaddam, N. & Falahati, A. (2007). An Improved ICI Reduction Method in OFDM
Communication System in Presence of Phase Noise, the 18th Annual IEEE
International Symposium on Personal, Indoor and Mobile Radio Communications
(PIMRC'07)
Kumar, R. & Malarvizhi, S. (2006). Reduction of Intercarrier Interference in OFDM Systems.
Maham, B. & Hjørungnes, A. (2007). ICI Reduction in OFDM by Using Maximally Flat
Windowing, IEEE International Conference on Signal Processing and Communications
(ICSPC 2007), Dubai, United Arab Emirates (UAE).
 4

Multiple Antenna Techniques

Han-Kui Chang, Meng-Lin Ku, Li-Wen Huang and Jia-Chin Lin
Department of Communication Engineering, National Central University, Taiwan,
R.O.C.

1. Introduction
Recent developed information theory results have demonstrated the enormous potential to
increase system capacity by exploiting multiple antennas. Combining multiple antennas
with orthogonal frequency division multiplexing (OFDM) is regarded as a very attractive
solution for the next-generation wireless communications to effectively enhance service
quality over multipath fading channels at affordable transceiver complexity. In this regard,
multiple antennas, or called multiple-input multiple-output (MIMO) systems, have emerged
as an essential technique for the next-generation wireless communications. In general, an
MIMO system has capability to offer three types of antenna gains: diversity gains,
multiplexing gains and beamforming gains. A wide variety of multiple antennas schemes
have been investigated to achieve these gains, while some combo schemes can make trade-
offs among these three types of gains. In this chapter, an overview of multiple antenna
techniques developed in the past decade, as well as their transceiver architecture designs, is
introduced. The first part of this chapter covers three kinds of diversity schemes: maximum
ratio combining (MRC), space-time coding (STC), and maximum ratio transmission (MRT),
which are commonly used to combat channel fading and to improve signal quality with or
without channel knowledge at the transmitter or receiver. The second part concentrates on
spatial multiplexing to increase data rate by simultaneously transmitting multiple data
streams without additional bandwidth or power expenditure. Several basic receiver
architectures for handling inter-antenna interference, including zero-forcing (ZF), minimum
mean square error (MMSE), interference cancellation, etc., are then introduced. The third
part of this chapter introduces antenna beamforming techniques to increase signal-to-
interference plus noise ratio (SINR) by coherently combining signals with different phase
and amplitude at the transmitter or receiver, also known as transmit beamforming or
receive beamforming. Another benefit of adopting beamforming is to facilitate multiuser
accesses in spatial domain and effectively control multiuser interference. The optimal
designs of these beamforming schemes are also presented in this chapter.

2. Diversity techniques
Diversity techniques have been widely adopted in modern communications to overcome
multipath fading, which allows for enhancing the reliability of signal reception without
sacrificing additional transmission power and bandwidth (Rappaport, 2002; Simon &
Alouini, 1999). The basic idea of diversity is that multiple replicas of transmitted signals
which carry the same information, but experience independent or small correlated fading,
60 Recent Advances in Wireless Communications and Networks

are available at the receiver. In fading channels, some samples are severely faded, while
others are less attenuated; hence, in statistics, the probability of the signal strength of all
samples being simultaneously below a given level becomes small, as compared with the
case without applying diversity techniques. Consequently, we can overwhelm the channel
fading by imposing an appropriate selection or combination of various samples, so as to
dramatically improve the signal quality. Based on signal processing domains to obtain
diversity gains, diversity techniques can be classified into time, frequency and space
diversity. Here, we focus on space diversity techniques where multiple antennas are
deployed at the transmitter or receiver sides. One category of space diversity schemes is to
combine multiple signal replicas at the receiver, which is termed as receive diversity. The
other category is to use multiple antennas at the transmitter, and this kind of diversity
schemes is called transmit diversity (Giannakis et al., 2006).
In this section, we first present various receive diversity schemes, including selection
combining, switch combining, equal-gain combining (EGC), and MRC. The well-known
Alamouti’s transmit diversity scheme using two transmit antennas and one receive antenna
is then introduced. The generalized case using two transmit antennas and multiple receive
antennas is shown as well. Subsequently, space-time block codes (STBCs) with the number
of transmit antennas larger than two (Tarokh et al., 1999) are presented. Finally, a maximum
ratio transmission (MRT) scheme is discussed to simultaneously achieve both transmit and
receive diversity gains and maximize the output signal-to-noise ratio (SNR) (Lo, 1999).

2.1 Receive diversity techniques

In cellular systems, receive diversity techniques have been widely applied at base stations
for uplink transmission to improve the signal reception quality. This is mainly because base
stations can endure larger implementation size, power consumption, and cost. In general,
the performance of the receive diversity not only depends on the number of antennas but
also the combining methods utilized at the receiver side. According to the implementation
complexity and the extent of channel state information required at the receiver, we will
introduce four types of combining schemes, including selection combining, switch
combining, EGC, and MRC, in the following.

2.1.1 Selection combining

Selection combining is a simple receive diversity combining scheme. Consider a receiver
equipped with nR receive antennas. Fig. 1 depicts the block diagram of the selection
combining scheme. The antenna branch with the largest instantaneous SNR is selected to
receive signals at every symbol period. In practical, since it is difficult to measure the SNR,
one can implement the selection combining scheme by accumulating and averaging the
received signal power, consisting of both signal and noise power, for all antenna branches,
and selecting one branch with the highest output signal power.

2.1.2 Switch combining

Fig. 2 shows the switch combining diversity scheme. As its name suggested, the receiver
scans all the antenna branches and selects a certain branches with the SNR values higher
than a preset threshold to receive signals. When the SNR of the selected antenna is dropped
down the given threshold due to channel fading, the receiver starts scanning all branches
again and switches to other antenna branches. As compared with the selection diversity
Multiple Antenna Techniques 61

scheme, the switch diversity scheme exhibits lower performance gain since it does not pick
up the branch with the highest instantaneous SNR or received signal power. In spite of this
performance loss, it is still very attractive for practical implementation as it does not require
to periodically and simultaneously monitor all the antenna branches. Another advantage is
that since both the selection and switch diversity schemes do not require any knowledge of
channel state information, they are not limited to coherent modulation schemes, but can also
be applied for noncoherent modulation schemes.

Fig. 1. Block diagram of selection combining scheme

Fig. 2. Block diagram of switch combining scheme

62 Recent Advances in Wireless Communications and Networks

2.1.3 Maximum ratio combining

Fig. 3 shows the block diagram of the MRC scheme. MRC is a linear combining scheme, in
which multiple received replicas at the all antenna branches are individually weighted and
summed up as an output signal. Since the multiple replicas experience different channel
fading gains, the combining scheme can provide diversity gains. In general, there are several
ways to determine the weighting factors. Consider a receiver having nR receive antennas,
and the received signals can be expressed as a matrix-vector form as follows:

⎡ r1 ⎤ ⎡ h1 ⎤ ⎡ n1 ⎤
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
r = ⎢ ⎥ = ⎢ ⎥s + ⎢ ⎥ = sh + n (11)
⎢r ⎥ ⎢ h ⎥ ⎢n ⎥
⎣ nR ⎦ ⎣ nR ⎦ ⎣ nR ⎦

where ri , hi , and ni are the received signal, channel fading gain, and spatially white noise
at the ith receive antenna branch, respectively. After linearly combining the received
signals, the output signal is given by

y = w † r = w † ( sh + n ) = sw † h + w † n (2)

where w represents the weighting factors for all antenna branches, and ( i ) is the
†

Hermitian operation. Subsequently, from (2), for a given h , the output SNR is calculated by

2
Es w † h
SNRo = 2
(3)
σ n2 w

where Es and σ n2 are the signal power and the noise power, respectively. According to the
Cauchy-Schwarz inequality, we have

2 2 2
w†h ≤ w h (4)

Hence, the upper bound for the output SNR is given by

2
Es w † h Es h
2
2
SNRo = 2
≤ = h SNRi (5)
σ n2 w σ n2

where SNRi = Es σ n2 is defined as the input SNR. We can further observe that the equality
in (5) holds if and only if w = h , and therefore, the maximum output SNR can be written as

2
SNRo = h SNRi (6)

The method adopting weighting factors w = h is called MRC, as it is capable of maximizing

2
the output SNR with a combining gain of h . However, the main drawback of the MRC
scheme is that it requires the complete knowledge of channel state information, including
both amplitude and phase of hi , to coherently combine all the received signals. Hence, it is
not suitable for noncoherent modulation schemes.
Multiple Antenna Techniques 63

Fig. 3. Block diagram of MRC scheme

2.1.4 Equal gain combining

Equal gain combing is a suboptimal combining scheme, as compared with the MRC scheme.
Instead of requiring both the amplitude and phase knowledge of channel state information,
it simply needs phase information for each individual channels, and set the amplitude of the
weighting factor on each individual antenna branch to be unity. Thus, all multiple received
signals are combined in a co-phase manner with an equal gain. The performance of the
equal gain combining scheme is only slightly worse than that of the MRC scheme, while its
implementation cost is significantly less than that of the MRC scheme.

2.2 Transmit diversity techniques

Although the receive diversity can provide great benefits for uplink transmission, it is
difficult to utilize the receive diversity techniques at mobile terminals for downlink
transmission. First, it is hard to place more than two antenna elements in a small-size
portable mobile device. Second, multiple chains of radio frequency components will
increase power consumption and implementation cost. Since mobile devices are usually
battery-limited and cost-oriented, it is impractical and uneconomical for using multiple
antennas at the mobile terminals to gain diversity gains at forward links. For these reasons,
transmit diversity techniques are deemed as a very attractive alternative. Wittneben
(Wittneben, 1993) proposed a delay diversity scheme, where replicas of the same symbol are
transmitted through multiple antennas at different time slots to impose an artificial
multipath. A maximum likelihood sequence estimator (MLSE) or a MMSE equalizer is
subsequently used to obtain spatial diversity gains. Another interesting approach is STC,
which can be divided into two categories: space-time trellis codes (STTCs) (Tarokh et al.,
1998) and STBCs. In the STTC scheme, encoded symbols are simultaneously transmitted
through different antennas and decoded using a maximum likelihood (ML) decoder. This
scheme combines the benefits of coding gain and diversity gain, while its complexity grows
exponentially with the bandwidth efficiency and achievable diversity order. Therefore, it
64 Recent Advances in Wireless Communications and Networks

may be not practical or cost-effective for some applications. Alamouti‘s STC was historically
the first STBC to provide two- branch transmit diversity gains for a communication system
equipped with two transmit antennas. It has been recognized as a remarkable, but simple,
diversity technique, and adopted in a number of next-generation wireless standards, e.g.,
3GPP long-term evolution and IEEE 802.16e standards.
In this section, we overview Alamouti’s transmit diversity technique. We focus on both
encoding and decoding algorithms, along with its performance results. Then, we introduce
the generalized STBCs with an arbitrary number of transmit antennas to achieve full
diversity gains, which are proposed by Vahid Tarokh (Tarokh et al., 1999) based on
orthogonal design theory.

2.2.1 Alamouti’s space-time encoding

The encoding procedure of Alamouti‘s Space-time codes for a two-transmit antenna system
is depicted in Fig. 4. Assume that data symbols, each of which is mapped from a group of
m information bits through an M -ary modulation scheme, are going to be transmitted,
where m = log 2 M . Let C denote the set of constellation points. For each encoding round,
the encoder successively takes a pair of two modulated data symbols x1 ∈ C and x2 ∈ C to
generate two transmit signal sequences of length two, according to the following space-time
encoding matrix:

⎡x − x2* ⎤
X=⎢ 1 ⎥ (7)
⎢⎣ x2 x1* ⎥⎦

Fig. 4. Block diagram of Alamouti’s space-time encoder

The Alamouti’s STC is a two-dimensional code, in which the encoder outputs are
transmitted within two consecutive time slots over two transmit antennas. During the first
time slot, two signals x1 and x2 are transmitted simultaneously from antenna one and
antenna two, respectively. Similarly, in the second time slot, the signal − x2 * is transmitted
∗
from antenna one and the signal x1* is from antenna two, where ( i ) denotes the complex
conjugate operation. It is clear that the encoding process is accomplished in both spatial and
temporal domains. Let us first denote the transmit sequence from antenna one and two by
x 1 and x 2 , respectively, as

x 1 = ⎡⎣ x1 − x2* ⎤⎦
(8)
x 2 = ⎡⎣ x2 x1* ⎤⎦
Multiple Antenna Techniques 65

We can observe that these two signal sequences possess the orthogonal property with each
other. That is, we have

( )
†
x1 x2 = x1x2* − x2* x1 = 0 (9)

Where ( i ) denotes the Hermitian operation.

†

In other words, the code matrix, X , satisfies the orthogonal matrix property as follows:

⎡x 2 + x 2
0 ⎤
1 2
XX † = ⎢ ⎥
⎢ 0 x1
2
+ x2 ⎥⎦
2
(10)
⎣
(
= x1 + x 2
2 2
)I 2

where I 2 is a 2 × 2 identity matrix.

Fig. 5. Block diagram of Alamouti’s space-time encoder

Let us assume that there is only one receive antenna deployed at the receiver side. The
receiver block diagram for the Alamouti‘s scheme is shown in Fig. 5. Assume that flat fading
channel gains from transmit antenna one and two to the receive antenna at the time slot t
are denoted by h1 (t ) and h2 (t ) , respectively. Under the assumption of quasi-static channels,
the channel gains across two consecutive symbol periods remain unchanged, and they can
be expressed as follows:

h1 (t ) = h1 (t + T ) = h1 (11)

and

h2 ( t ) = h2 ( t + T ) = h2 (12)

where hi , for i = 1 and 2 , is a complex constant value corresponding to the channel gain
from the transmit antenna i to the receive antenna, and T denotes the symbol period. At
66 Recent Advances in Wireless Communications and Networks

the receive antenna, the received signals across two consecutive symbol periods, which are
denoted by r1 and r2 for time t and t + T , are respectively given by

r1 = h1x1 + h2 x2 + n1 (13)

and

r2 = − h1x *2 + h2 x *1 + n2 (14)

where n1 and n2 are independent additive white Gaussian noise with zero mean and
variance σ 2 . It is noticed here that although we present Alamouti’s space-time codes under
flat fading channels without concerning the multipath effect, it is straightforward to extend
the Alamouti’s scheme to the case of multipath channels by using an OFDM technique to
transform a frequency selective fading channel into a number of parallel flat fading channels
(Ku & Huang, 2006).

2.2.2 Maximum likelihood decoding for Alamouti’s scheme

The successful decoding for Alamouti’s space-time codes requires the knowledge on
channel state information h1 and h2 at the receiver side. In general, channel estimation can
be performed through the use of some pilot signals which are frequently transmitted from
the transmit side (Ku & Huang, 2008; Lin, 2009a, 2009b). Here, we focus on the decoding
scheme and assume that channel state information is perfectly estimated and known to the
receiver. From the viewpoint of minimum error probability, the decoder intends to choose
( )
an optimal pair of constellation points, x 1, x 2 , to maximize the a posteriori probability
given by the received signals r1 and r2 . Mathematically, we can express the decoding
problem as

(x 1, )
x 2 = arg max 2 Pr ( x1 , x2 r1,r2 )
( x1 , x2 )∈C
(15)

where C 2 is the set of all possible candidate symbol pairs ( x1, x2 ) , and Pr ( i ) is a
probability notation. According to the Bayes’ theorem, we can further expand (15) as

Pr ( r1 , r2 x1, x2 ) P ( x1 , x2 )
(x 1 )
, x 2 = arg max
( x1 , x2 )∈C 2
P ( r1 , r2 )
(16)

By assuming that all the constellation points in C 2 occur with equal prior probabilities and
the two symbols of each pair are generated independently, all symbol pairs ( x1 , x2 ) are
equiprobable. As the decision of the symbol pairs ( x1, x2 ) is irrelevant to the probability of
received signals r1 and r2 , we can rewrite (16) as

(x 1, )
x 2 = arg max 2 Pr( r1 , r2 x1, x2 )
( x1 , x2 )∈C
(17)

Furthermore, since the noise n1 and n2 at time t and time t + T , respectively, are assumed
to be mutually independent, we can alternatively express (17) as

(x 1, )
x 2 = arg max 2 Pr( r1 x1, x2 )Pr( r2 x1, x2 )
( x1 , x2 )∈C
(18)
Multiple Antenna Techniques 67

Recall from (13) and (14) that r1 and r2 are two independent Gaussian random variables
( )
with distributions r1 ∼ N h1x1 + h2 x2 , σ 2 and r2 ∼ N − h1x2* + h2 x1* , σ 2 . Substituting this ( )
into (18), we then obtain a ML decoding criterion:

( x x ) = arg ( 2 2
1, 2 min r1 − h1x1 − h2 x2 + r2 + h1x2* − h2 x1*
x1 , x2 )∈C 2
(19)
= arg min 2 d 2 ( r1 , h1x1 + h2 x2 ) + d 2 r2 , −h1x2* + h2 x1*
( x1 , x2 )∈C
( )
where d ( s1 , s2 ) denotes the Euclidean distance between s1 and s2 . The ML decoder is,
2

therefore, equivalent to choosing a pair of data symbols x 1, x 2 to minimize the distance ( )

metric, as indicated in (19). By replacing (13) and (14) into (19), the ML decoding criterion
can be further rewritten as a meaningful expression as follows:

( x x ) = arg (
1, 2 min
x1 , x2 )∈C2
(h 1
2
+ h2 − 1
2
)( x 1
2
+ x2
2
) + d (x ,x ) + d (x ,x )
2
1 1
2
2 2
(20)

where x 1 and x 2 are two decision statistics obtained by combining the received signals r1
and r2 with channel state information h1 and h2 , given by

x 1 = h1*r1 + h2 r2*
(21)
x 2 = h2* r1 + h1r2*

By taking r1 and r2 from equation (13) and (14), respectively, into (21), the decision
statistics is given by

(
x 1 = h1 + h2
2 2
)x + h n + h n
1
*
1 1
*
2 2
(22)
=(h )x − h n + h n
2 2 * *
x2 1 + h2 2 1 2 2 1

It is observed that for a given channel realization h1 and h2 , the decision statistics x i in
(22) is only a function of xi , for i = 1, 2 . Consequently, the ML decoding criterion in (20)
can be divided into two independent decoding criteria for x1 and x2 ; that is, we have

( 2
x 1 = arg min h1 + h2 − 1 x1 + d 2 x 1 , x1
x1 ∈C
2
) 2
( ) (23)

and

( 2
x 2 = arg min h1 + h2 − 1 x2 + d 2 x 2 , x2
x2 ∈C
2
) 2
( ) (24)

Particularly, if a constant envelope modulation scheme such as M -phase-shift-keying ( M -

( )
PSK) is adopted, the term h1 2 + h2 2 − 1 xi 2 , for i = 1, 2 , remains unchanged for all
possible signal points with a fixed channel fading coefficients h1 and h2 . Under this
circumstance, the decision rules of (23) and (24) can be further simplified as

x 1 = arg min d 2 x 1 , x1 ;
x1 ∈C
( ) x 2 = arg min d 2 x 2 , x2
x2 ∈C
( ) (25)
68 Recent Advances in Wireless Communications and Networks

From (25), for the case of constant envelope modulation, the decoding algorithm is just a
linear decoder with extremely low complexity to achieve diversity gains. On the other hand,
when non-constant envelope modulation, e.g., quadrature-amplitude-modulation (QAM) is
adoped, the term (h1
2 2
) 2
+ h2 − 1 xi , for i = 1, 2 , may become different for various
constellation points and cannot be excluded from the decoding metric. Therefore, we should
follow the decoding rules as shown in (23) and (24) to achieve the ML decoding.

2.2.3 Alamouti’s scheme with multiple receive antennas

We now extend the Alamouti’s scheme to an MIMO communication system with nR
multiple receive antennas. Let us denote r1j and r2j as the received signals at the jth receive
antenna at the time slot t and t + T , respectively. According to (13) and (14), it follows

r1j = h j ,1x1 + h j ,2 x2 + n1j

(26)
r2j = −h j ,1x2* + h j ,2 x1* + n2j

where h j , i , for i = 1, 2 and j = 1, , nR , is the channel fading gain from the transmit
antenna i to the receive antenna j , and n1j and n2j are assumed to be spatially and
temporally white Gaussian noises for the receive antenna j at time t and t + T ,
respectively. Similar to the derivation in the case of single receive antenna, the ML decoding
criterion with multiple receive antennas now can be formulated as below:

nR
( x x ) = arg (
2 2
1, 2 min ∑ r1j − h j ,1x1 − h j ,2 x2
x1 , x2 )∈C j = 1
2
+ r2j + h j ,1x2* − h j ,2 x1*
(27)
nR
= arg min
( x1 , x2 )∈C2
∑d
j =1
2
( r1j , h j ,1x1 + h j ,2 x2 + d) 2
( r2j , − h j ,1x2* + h j ,2 x1* )
We then define two decision statistics by combining the received signals at each receive
antenna with the corresponding channel link gains, as follows:

( )
j *
x 1 = h *j ,1r1j + h j ,2 r2j
(28)
j
x 22 = h *j ,2 r1j − h j ,1 (r )
2
j *

Note that by replacing r1j and r2j , given in (26), into (28), the decision statistics can be
explicitly written as

( )
*
x 1 = ⎛⎜ h j ,1 + h j ,2 ⎞⎟ x1 + h *j ,1n1j + h j ,2 n2j
j 2 2

⎝ ⎠ (29)
( )
*
j ⎛ 2 2⎞
x 2 = ⎜ h j ,1 + h j ,2 ⎟ x2 − h j ,1 n2j + h *j ,2 n1j
⎝ ⎠
2 2
where Geff = h j ,1 + h j ,2 is the effective channel fading gain, and it is shown that the
Alamouti’s STC scheme can therefore extract a diversity order of two at each receiving
branch, even in the absence of channel state information at the transmitter side. Following
Multiple Antenna Techniques 69

the derivation in (19) and (20), the ML decoding rules, under the case of nR receive
antennas, for the two data symbols x1 and x2 can be represented by
nR
x 1 = arg min ∑ ⎛⎜ h j ,1 + h j ,2 − 1 ⎞⎟ x1 + d 2 ( x 1 , x1 )
2 2 2 j
(30)
j =1 ⎝ ⎠
x1 ∈C

and
nR
x 2 = arg min ∑ ⎛⎜ h j ,1 + h j ,2 − 1 ⎞⎟ x2 + d 2 ( x 2 , x2 )
2 2 2 j
(31)
x2 ∈C
j =1 ⎝ ⎠

In particular, for constant envelope modulation schemes whose constellation points possess
equal energy, the ML decoding can be reduced to finding a data symbol x i , for i = 1, 2 , to
j
minimize the summation of Euclidean distance d 2 ( x 1 , x1 ) over all receive antennas, in the
following:

( )
nR
j
x 1 = arg min ∑ d 2 x 1 , x1 (32)
x1 ∈C
j =1

and

( )
nR
j
x 2 = arg min ∑ d 2 x 2 , x2 (33)
x2 ∈C
j =1

2.2.4 BER performance of Alamouti’s scheme

The bit error rate (BER) performance of the Alamouti’s transmit diversity scheme is
simulated and compared with the MRC receive diversity scheme in the following. At the
beginning, it is assumed that a flat fading channel is used, and the fading gains from each
transmit antenna to each receive antenna are mutually independent. Furthermore, we
assume that the total transmission power from the two transmit antennas for the Alamouti’s
scheme is the same as that for the MRC receive diversity scheme.
Fig. 6 compares the BER performance between the Alamouti’s with one or two receive
antennas and the MRC receive diversity with two or four receive antennas. The Alamouti’s
scheme with two transmit antennas and one receive antenna has the same diversity order as
the MRC receive diversity scheme with two branches. In other words, the slopes of these
two BER performance curves are identical. However, the Alamouti’s scheme has 3dB loss in
terms of E b N 0 . This is due to the fact that for fair comparisons, the total transmission
power is fixed and the energy radiated from each transmit antenna in the Alamouti’s
scheme is a half of that from a single transmit antenna in the MRC receive diversity scheme.
Similarly, the Alamouti’s scheme with two receive antennas can introduce the same
diversity order as the MRC receiver diversity scheme with four branches, while there is still
3dB loss in BER performance. In general, the Alamouti’s scheme with two transmit antennas
and nR receive antennas can provide a diversity order of 2 × nR , which is the same as the
case that the MRC scheme uses 2 nR receive antennas.
In Fig. 7, it is shown that the BER performance of the Alamouti’s scheme with quadrature
phase-shift keying (QPSK) modulation over flat fading channels. It is obvious that the more
number of receive antennas it uses, the higher diversity order it can achieves.
70 Recent Advances in Wireless Communications and Networks

0
10
SISO
Alamouti (NR=1)
-1 MRC (NR=2)
10
Alamouti (NR=2)
MRC (NR=4)
-2
10
BER

-3
10

-4
10

-5
10
0 5 10 15 20 25 30 35 40 45
Eb/N0 [dB]

Fig. 6. Comparison of BER performance between Alamouti’s and MRC schemes with binary
phase-shift keying (BPSK) modulation

0
10
Alamouti QPSK (NR=1)
Alamouti QPSK (NR=2)
-1
10

-2
10
BER

-3
10

-4
10

-5
10
0 5 10 15 20 25 30
Eb/N0 [dB]

Fig. 7. BER performance of Alamouti’s scheme using QPSK modulation

Multiple Antenna Techniques 71

2.2.5 Generalized space-time block codes

As we discussed, the Alamouti’s scheme shows a very elegant way to achieve full diversity
gains, i.e., a diversity order of two, with a low-complexity linear decoding algorithm by
utilizing two transmit antennas. The key feature of the Alamouti’s scheme is the orthogonal
property of the encoding matrix in (1), i.e., the sequences generated by the two transmit
antennas are independent of each other. In (Tarokh et al., 1999), Tarokh generalizes this idea
to any arbitrary number of transmit antennas by applying the orthogonal design theory, and
proposes a series of STBCs which can fulfill transmit diversity specified by the number of
transmit antennas nT. Meanwhile, these STBCs also enable a very simple maximum-
likelihood decoding algorithm, based only on a linear processing of the received signals at
different time slots.

Fig. 8. Encoder structure of STBCs

The encoder structure for generalized STBCs is presented in Fig. 8. In general, a STBC can be
defined via an nT × p transmission matrix X , where nT represents the number of transmit
antennas and p is the time duration for transmitting each block of space-time coded
symbols. Consider a M -ary modulation scheme, where we define m = log 2 M as the
number of information bits required for each constellation point mapping. At each encoding
operation, a block of km information bits are mapped onto k modulated data symbols xi ,
for i = 1,… , k . Subsequently, these k modulated symbols are encoded by the nT × p space-
time encoder X to generate nT parallel signal sequences of length p which are to be
transmitted over nT transmit antennas simultaneously within p time slots. The code rate of
a STBC is defined as the ratio of the number of symbols taken by the space-time encoder as
its input to the number of space-time coded symbols transmitted from each antenna. Since
p time slots are required for transmitting k information-bearing data symbols, the code
rate is given by

k (34)
Rc
p
Therefore, the spectral efficiency for the STBC is calculated by

rb ( rsm ) Rc km
η = = ( bits sec Hz ) (35)
B rs p
where rb and rs are the bit and symbol rate of a space-time coded symbol, respectively, and
B represents the total bandwidth. For simplicity of notations, we usually denote a STBC
with nT transmit antennas as X nT . Based on the orthogonal designs in (Tarokh et al., 1999),
to obtain full diversity gains, i.e., diversity order is equal to nT , the space-time encoding
matrix should preserve the orthogonal structure; that is, we have
72 Recent Advances in Wireless Communications and Networks

(
X nT ⋅ X n† = c x1 + x2 +
T
2 2
+ xk
2
)I nT (36)

where c is constant, ( i ) takes the Hermitian operation, and InT is an nT × nT identity

†

matrix, the entries in X nT take the values of modulated symbols xi , their conjugate xi∗ , or
their combination. The orthogonal structure allows the receiver to decouple the signals
transmitted from different antennas by using a simple linear decoder derived based on the
ML decoding metric. Tarokh et al. (Tarokh et al., 1999) discovered that the code rate of a
STBC with full diversity must be less than or equal to one, i.e., Rc ≤ 1 . In other words, the
STBCs cannot be used to increase bandwidth efficiency, but provide diversity gains. It is
noted that the full code rate, Rc = 1 , requires no additional bandwidth expansion, while the
code rate Rc ≤ 1 requires a bandwidth expansion by a factor of 1 Rc .
Based on modulation types, STBCs can be classified into two categories: real signaling or
complex signaling. For a special case of p = nT , it is evident from (Tarokh et al., 1999) that
for an arbitrary real constellation signaling, e.g., M -amplitude shift keying ( M -ASK),
STBCs with an nT × nT square encoding matrices X nT exist if and only if the number of
transmit antennas nT is equal to two, four, or eight. Moreover, these code matrices can not
only achieve the full code rate Rc = 1 but also provide the full diversity gains with a
diversity order of nT . However, it is desirable to have code matrices with the full diversity
gains and the full code rate for any number of transmit antennas. It has been proved that for
nT transmit antennas, the minimum required value for the transmission periods p to achieve
the full diversity nT and the full code rate Rc = 1 must satisfy the following condition:

min 2 4c + d (37)
{(c , d ) 0≤c , 0 ≤ d ≤ 4, and 8c + 2 d
≥ nT }

nT p X nT
⎡x − x2 ⎤
2 2 X2 = ⎢ 1
⎣ x2 x1 ⎥⎦
⎡ x1 − x2 − x3 −x4 ⎤
⎢x x1 x4 − x3 ⎥⎥
X4 = ⎢
2
4 4
⎢ x3 −x4 x1 x2 ⎥
⎢ ⎥
⎣ x4 x3 x2 x1 ⎦
⎡ x1 − x2 − x3 −x4 − x5 − x6 − x7 − x8 ⎤
⎢x x1 −x4 x3 − x6 x5 x8 − x7 ⎥⎥
⎢ 2
⎢ x3 x4 x1 − x2 − x7 − x8 x5 x6 ⎥
⎢ ⎥
x4 − x3 x2 x1 − x8 x7 − x6 x5 ⎥
8 8 X 8 = ⎢⎢
x x6 x7 x8 x1 − x2 − x3 −x4 ⎥
⎢ 5 ⎥
⎢ x6 − x5 x8 − x7 x2 x1 x4 − x3 ⎥
⎢x − x8 − x5 x6 x3 −x4 x1 x2 ⎥
⎢ 7 ⎥
⎢⎣ x8 x7 − x6 − x5 x4 x3 − x2 x1 ⎥⎦

Table 1. Square code matrices with full diversity gains and full code rate for nT = 2, 4, 8
Multiple Antenna Techniques 73

Accordingly, the minimum values of p for a specific value of nT ≤ 8 , and the associated
STBC matrices X nT for real signaling are provided as follows, where the square transmission
matrices X 2 , X 4 , and X 8 are listed in Table 1, and the non-square transmission matrices
X 3 , X 5 , X 6 and X 7 are listed in Table 2.

nT p X nT
⎡ x1 − x2 − x3 −x4 ⎤
3 4 X 3 = ⎢⎢ x2 x1 x4 − x3 ⎥⎥
⎢⎣ x3 −x4 x1 x2 ⎥⎦
⎡ x1 − x2 − x3 −x4 − x5 − x6 −x7 − x8 ⎤
⎢x x1 −x4 x3 − x6 x5 x8 −x7 ⎥⎥
⎢ 2
5 8 X 5 = ⎢ x3 x4 x1 − x2 −x7 − x8 x5 x6 ⎥
⎢ ⎥
⎢ x4 − x3 x2 x1 − x8 x7 − x6 x5 ⎥
⎢x x6 x7 x8 x1 − x2 − x3 −x4 ⎥⎦
⎣ 5
⎡ x1 − x2 − x3 −x4 − x5 − x6 − x7 − x8 ⎤
⎢x x1 −x4 x3 − x6 x5 x8 − x7 ⎥⎥
⎢ 2
⎢x x4 x1 − x2 − x7 − x8 x5 x6 ⎥
6 8 X6 = ⎢ 3 ⎥
⎢ x4 − x3 x2 x1 − x8 x7 − x6 x5 ⎥
⎢x x6 x7 x8 x1 − x2 − x3 −x4 ⎥
⎢ 5 ⎥
⎢⎣ x6 − x5 x8 − x7 x2 x1 x4 − x3 ⎥⎦
⎡ x1 − x2 − x3 −x4 − x5 − x6 −x7 − x8 ⎤
⎢x x1 −x4 −x3 − x6 x5 x8 −x7 ⎥⎥
⎢ 2
⎢ x3 x4 x1 − x2 −x7 − x8 x5 x6 ⎥
⎢ ⎥
7 8 X 7 = ⎢ x4 − x3 x2 x1 − x8 x7 − x6 x5 ⎥
⎢x x6 x7 x8 x1 − x2 − x3 −x4 ⎥
⎢ 5 ⎥
⎢ x6 − x5 x8 −x7 x2 x1 x4 − x3 ⎥
⎢x − x8 − x5 x6 x3 −x4 x1 x2 ⎥⎦
⎣ 7
Table 2. Non-square code matrices with full diversity gains and full code rate for nT = 3, 5, 6, 7
The other type of STBCs belongs to complex constellation signaling, and just as the case for
the real constellation signaling, these complex STBCs also abide by the orthogonal design
constraint in (36). In particular, Alamouti’s scheme can be regarded as a complex STBC for
two transmit antennas; that is, the code matrix can be expressed as

⎡x −x2* ⎤
X C2 = ⎢ 1 ⎥ (38)
⎢⎣ x2 x1* ⎥⎦

where we use XCnT to denote a complex STBC for nT transmit antennas in order to
discriminate between real and complex matrices. It is noted that the Alamouti’s scheme can
provide a diversity order of two and the full code rate. As compared with those real STBCs,
it is much more desirable to invent complex STBCs since complex constellation schemes
74 Recent Advances in Wireless Communications and Networks

usually exhibit higher bandwidth efficiency. In addition, one might wonder if there exist
other complex STBC matrices for nT > 2 . Unfortunately, it has been shown that Alamouti’s
scheme is the only complex STBC with an nT × nT square code matrix to simultaneously
achieve the full diversity gains and the full code rate. For the case of nT > 2 , we might
intend to construct complex orthogonal matrices XCnT that can achieve full diversity gains,
but with high code rate Rc and minimum decoding latency p . We summarize below that
for any number of transmit antennas, there exist complex STBCs with code rate of Rc = 1 2 .
For example, for the cases of three or four transmit antennas, the code matrices are given by

⎡ x1 − x2 − x3 −x4 x1* - x2* - x3* - x 4* ⎤

⎢ ⎥
XC3 = ⎢ x2 x1 x4 − x3 x2* x1* x 4* - x3* ⎥ , Rc = 1 2 (39)
⎢ ⎥
⎣⎢ x3 −x4 x1 x2 x3* - x 4* x1* x2* ⎦⎥

⎡ x1 − x2 − x3 −x4 x1* - x2* - x3* - x 4* ⎤

⎢ ⎥
⎢x x1 x4 − x3 x2* x1* x4* - x3* ⎥
XC4 = ⎢ 2 ⎥ , Rc = 1 2 (40)
⎢ x3 -x4 x1 x2 x3* - x4* x1* x2* ⎥
⎢ ⎥
⎣ x4 x3 - x2 x1 x4* x3* - x2* x1* ⎦

Still, for the case of nT > 2 , we can acquire other higher code-rate complex code matrices at
the expense of complicated linear encoding and decoding processing. For example, the
following two matrices XC3 and XC4 are STBCs with a code rate of Rc = 3 4 :

⎡ x3* x3* ⎤
⎢ x1 − x2* ⎥
⎢ 2 2 ⎥
⎢ ⎥
x3* - x3*
XC3 ,3 4 = ⎢ x2 x1* ⎥ , Rc = 3 4 (41)
⎢ 2 2 ⎥
⎢ ⎥
⎢ x3 x3 ( −x 1 − x1* + x2 − x2* ) ( * *
)
x 2 + x 2 + x1 − x1 ⎥
⎢ ⎥
⎣ 2 2 2 2 ⎦

⎡ x3* x3* ⎤
⎢ x1 −x2 ⎥
⎢ 2 2 ⎥
⎢ x3* − x3* ⎥
⎢ x2 x1 ⎥
⎢ 2 2 ⎥
XC4 ,3 4 =⎢ * ⎥ , Rc = 3 4
(42)
⎢ x3 x3 ( −x1 − x1* + x2 − x2* ) ( *
x 2 + x 2 + x1 − x1 ⎥ )
⎢ 2 2 ⎥
⎢ 2 2 ⎥
⎢
⎢ x3 − x3 ( −x2 − x2* + x1 − x1* ) ( − x1 + x1* + x2 − x2* ⎥
⎥
)
⎢⎣ 2 2 2 2 ⎥⎦

2.3 Maximum ratio transmission techniques

The multipath fading is considered as a detrimental effect to degrade the performance of
wireless communications. The most common and simplest way to mitigate the multipath
Multiple Antenna Techniques 75

fading is to adopt antenna diversity techniques. As we introduced in the previous section,

with multiple receive antennas, the MRC scheme can attain receive diversity gains and
maximize the output SNR, while with multiple transmit antennas, STC schemes are
proposed to obtain the MRC-like transmit diversity gains. For example, Alamouti’s STC can
obtain a diversity order of two by encoding a pair of two symbols to transmit over two
transmit antennas and two contiguous time slots. However, these transmit diversity
techniques are designed with the object of providing diversity gains, other than to maximize
the post-output SNR at the receiver side, which is usually taken as one of the most
important performance figure to minimize the BER performance. Furthermore, more and
more wireless communication systems are now equipped with multiple transmit and
receive antennas, and therefore it is desired to simultaneously obtain transmit and receive
diversity gains so as to combat the severe fading effects.
In this section, we introduce a MRT scheme to fulfill the above two challenges, namely
achieving transmit and receive diversity gains and maximizing the post-output SNR (Lo,
1999). Finally, numerical results are presented.

2.3.1 MRT systems and schemes

The developed framework of MRT schemes can be regarded as the generalization of MRC
for the systems with multiple transmit-and-receive antenna pairs. Figure 9 presents the MRT
system model for which nT transmit antennas and nR receive antennas are equipped.
Accordingly, there are nT × nR channel links between the transmitter and the receiver. We
assume channel coefficients are statistically independent, as follows:

⎡ h11 h1nT ⎤ ⎡ h1 ⎤
⎢ ⎥ ⎢ ⎥
H=⎢ ⎥=⎢ ⎥ (43)
⎢h hnR nT ⎥⎦ ⎢⎣ hnR ⎥⎦
⎣ nR 1
where the entry hnR nT denotes the flat fading channel gain from the nT th transmit antenna
to the nRth receive antenna. In general, it is essential to acquire the channel state
information for successful implementation of a multiple-input multiple-output
communication system. The channel estimation issue is beyond the scope of this chapter,
and it is assumed that channel state information of (43) is perfectly known at the both
transmitter and receiver sides.

Fig. 9. System model of MRT

The MRT system is a diversity-achieving technique. For each time slot, only one data
symbol, denoted by s , is transmitted by multiplying with antenna-specific weighting
76 Recent Advances in Wireless Communications and Networks

coefficients vm , m = 1,… , nT . The average symbol power is assumed to be E ⎡ s 2 ⎤ = Es .

⎣ ⎦
T
Denoting the weighting coefficients vector as v = ⎡ v1 ,… , vn ⎤ , the signals transmitted by
⎣ T ⎦

antennas can be expressed by

T
x = ⎣⎡ x1 ,… , xnT ⎦⎤ = sv (44)

From (43) and (44), the received signals is given by

r = Hx + n (45)
T
where n = ⎡n1 ,… , nn ⎤ is the additive white Gaussian noise vector. Subsequently, a
⎣ R ⎦
T
weighting vector w = ⎡ w1 ,… , wn ⎤ is applied for combining the received signals at the
⎣ R ⎦

receive antennas, followed by symbol decision. In the MRT system, we construct the
weighting factor v from the channel matrix H through a linear transformation as follows:

1
v= ( gH )† (46)
a

where g = ⎡⎣ g1 gnR ⎤⎦ is a parameter to be determined. The transmitted signal can be

alternatively written as

1
x= ( gH )† s (47)
a
where a is a normalization factor such that the transmit power is equal to one, which can be
represented as

1
⎛ nR nR nT ⎞ 2
a = gH = ⎜ ∑ ∑ g p gq* ∑ hpk h gk
*
⎟ (48)
⎜ p=1 q =1 ⎟
⎝ k =1 ⎠
Therefore, from (47), the received signal vector is given by

1
r = H ( gH ) s + n
†
(49)
a
In order to make symbol decision, the weighting vector w is applied to combine the
received signal vector x for signal reception. If wT is equal to g , the output is given by

1
s = wT r = g H ( gH ) s + gn = as + gn
†
(50)
a
The post-output SNR after combining is thus calculated as

a2 a 2γ 0
γ= γ0= (51)
gg † 2
∑ p =1
nR
gp
Multiple Antenna Techniques 77

where γ 0 = Es σ n2 denotes the average SNR for a single transmit antenna. It is observed
from (51) that the post-output SNR only depends on the value of g . Thus, to maximize the
post-output SNR is equivalent to choosing an appropriate value of g . By assuming all
channel links between the transmitter and receiver are statistically identical, each of the
transmitter antenna has to be allocated with the same weighting power; that is, we have
g1 = g2 = = gnR . Without loss of generality, we can set gi = 1 , for i = 1, , nR , for
simplicity. Accordingly, the post-output SNR in (51) is simplified as

a2
γ= γ0 (52)
N
By applying the Cauchy-Schwarz inequality, we can further get that a 2 is maximized, i.e.,
maximizing the post-output SNR, if and only if we set

nT
∑ hpk hgk*
(g g ) =
* *
p q
k =1
nT
(53)
∑ hpk hgk*
k =1

It is noted that the denominator term in (53) is due to the fact of gi = 1 and g p gq* = 1 . In
accordance with (52) and (53), the maximum post-output SNR is given by

nR nR nT
γ0
γ max =
N
∑ ∑ ∑ hpk hgk*
p =1 q =1 k =1
(54)
nR nR
γ0
=
N
∑∑ h p hq†
p =1 q =1

On the one hand, if h p and hq are mutually orthogonal, i.e., h p hqH = 0 , γ max takes the
minimum value, given by
nR nT
γ0 2
γ max =
N
∑ ∑ hpk (55)
p =1 k =1
2
On the other hand, if h p and hq are fully correlated, i.e., h p hqH = hq , γ max takes the
maximum value, given by
nR nR nT
γ0 2
γ max =
N
∑ ∑ ∑ hpk (56)
p =1 q =1 k =1

Form (55) and (56), the average post-output SNR is therefore bounded by

Mγ 0 ≤ γ max ≤ NMγ 0 (57)

where γ 0 = γ 0E ⎡⎢ hpk ⎤⎥ is defined as the average SNR at each branch. Also, for an MRT system
2

⎣ ⎦
with nT × nR antennas, it is expected from (55) that the diversity order is equal to nT × nR .
78 Recent Advances in Wireless Communications and Networks

0
10
NR=1
NR=2
-1
10

-2
10
BER

-3
10

-4
10

-5
10
0 5 10 15 20 25 30
Eb/N0 [dB]

Fig. 10. Comparison of MRT schemes with NT = 2 and QPSK modulation

0
10
NR=1
NR=2
-1
10

-2
10
BER

-3
10

-4
10

-5
10
0 5 10 15 20 25 30
Eb/N0 [dB]

Fig. 11. Comparison of MRT schemes with NT = 4 and QPSK modulation

Multiple Antenna Techniques 79

2.3.2 BER performance of MRT scheme

The BER performance of the MRT scheme is simulated as follows. At the beginning, we
assume that the complex Gaussian fading gains among transmit-receive pairs are mutually
independent. The total channel power from all the transmit antennas to a receive antenna is
normalized to one. The QPSK modulation is adopted in the simulation. Moreover, it is
assumed that channel fading gains are perfectly available to both transmitter and receiver
sides. Fig. 10 and Fig. 11 demonstrate the BER performance curves for an MRT system with
two transmit antennas and two and four receive antennas, respectively. It is observed from
these two figures that the BER performance can be improved as the number of transmit and
receive antennas increases, and we can gain much diversity gains by increasing the number
of antennas.

3. Spatial multiplexing techniques

Recent research resutls from information theory aspects have disclosed that rich scattering
wireless channels can provide enormous capacity which increases proportional to the
minimum number of transmit and receive antennas (Foschini, 1996). Spatial multiplexing
based MIMO (SM-MIMO) systems, in which multiple data streams are simultaneously
transmitted through multiple antennas, is a new attractive technique to realize high data
rate transmission, while signal detection at the receiver is a difficult task for SM-MIMO
systems. Although the well-known ML detector can be applied for the SM-MIMO systems
achieve the best performance, its complexity increases exponentially as the number of data
stream and the number of transmit antennas increase.
Foschini proposed a layered space-time (LST) or Bell laboratories layed space-time (BLAST)
architecture (Foschini & Gans, 1998), allowing for processing multi-dimensional received
signals in spatial domain. At the transmitter, multiple data streams are first encoded and
then distributed over multiple antennas. At the reciever, the received signals are separated
and subsequently decoded by integrating interference supression or cancellation tehcniques
with decoding algorithms, thus leading to much lower computation complexity, as
compared with the ML decoding. Various BLAST architectures, depending on whether error
control coding is adopted or not, have been investigated, e.g., horizontal BLAST (H-BLAST),
diagonal BLAST (D-BLAST), and vertical BLAST (V-BLAST). The error control coding is
applied for both the H-BLAST and D-BLAST architectures, while the V-BLAST does not
consider the use of error control coding. Although the H-BLAST and D-BLAST architectures
can gain better performance due to the coding gains in spatial domain, they suffer from a
spectral efficiency loss and higher implementation complexity (Foschini & Gans, 1998).
Therefore, the uncoded and simplified V-BLAST architecture, proposed in (Wolniansky et
al., 1998), is considered as an effective solution to fullfill the potential spatial multiplexing
gains. In this section, we focus on the V-BLAST architecture, followed by data detection
techniques used to separate and detect the V-BLAST signals. Two essential V-BLAST
receivers based on ZF and MMSE are introduced. Finally, we discuss the BER performance
of the V-BLAST architecture with different data detection algorithms.

3.1 V-BLAST systems

The block diagram of the V-BLAST system is presented in Fig. 12, where there are nT
transmit and nR receive antennas and nT ≤ nR. A single data stream is divided into nT
substreams, each of which is then individually modulated into symbols and transmitted
80 Recent Advances in Wireless Communications and Networks

from its corresponding transmit antenna. Without loss of generality, it is assumed that the
same constellation is applied for all substreams, and the transmission duration is a burst
consisting of L symbols. We assume that the multiple-input multiple-output channel is flat
fading and quasi-static over the duration of L symbols, and the channel matrix is denoted
by HnR ×nT , whose ( i , j ) th entry hij represents the complex channel gain from the jth
transmit antenna to the ith receive antenna. We denote the data vector to be transmitted as

Fig. 12. Block diagram of V-BLAST systems

T
s = ⎡⎣s1 ,… , snT ⎤⎦ (58)

where si is the modulated symbol at the ith transmit antenna. The received signals at the
nR receive antennas can thus be represented as a vector, as follows:

r = Hs + n (59)
where n is an independent and identical distributied (i.i.d.) white Gaussian noise vector
with zero mean and a covariance matrix σ n2 I .

3.2 V-BLAST data detection

The idea to perform data detection for this system is to incorporate conventional linear
equalizers with nonlinear interference cancellation methods. Each substream is in trun
detected, and the reminders are considered as interference signals. The equalizer is designed
according to some specific criteria, e.g., ZF and MMSE, to null out the interference signals by
linearly weighting the received signals with equalizer coefficients. In the following, we
discuss the ZF-based V-BLAST receiver and the MMSE-based V-BLAST receiver.

3.2.1 ZF-based V-BLAST algorithm

First of all, the ZF nulling is performed by strictly choosing nulling vectors w i , for
i = 1,… , M , such that

w †i ( H ) j =δ ij , for i , j = 1, 2, nT (60)
Multiple Antenna Techniques 81

where ( H ) j denotes the jth column of H , the notation ( i ) takes the Hermitian, and δ is
†

the Kronecker delta function. Then, the ith substream after equalization is given by

y i = w †i r (61)

The ZF-based detection procedures to extract substreams for any arbitrary detection order
are elaborated in the following. Let us set j = 1 , r1 = r and the ordering set for data
{ }
detection as ς = k1 , k2 , , knT , where k j ∈ {1,… , nT } .
Step 1. Use the nulling vector w k j to obtain the decision statistic for the k jth substream

y k j = wTk j rj (62)

Step 2. Slice y k j to obtain the hard decision sˆk j

( )
sˆk j = Q y k j (63)

where Q ( i ) denotes the hard decision operation.

Step 3. Reconstruct and cancel out the currently detected substream from the received
signal rj , resulting in a modified received signal vector rj+ 1

rj + 1 = rj − s k j ( H ) k (64)
j

Update j to j+1 for the next iteration, and repeat the Step1~Step3, where the k j th ZF nulling
vector is given by

⎧0, for i ≥ j
w †k ( H ) k = ⎨ (65)
⎩1, for i = j
j i

Thus, if the inter-antenna interference is perfectly reconstructed and cancelled, the weighting
vector w †k is orthogonal to the subspace spanned by the interference-reduced vector rj + 1 .
j +
Accordingly, the solution to w †k in (65) is the k jth row of H j − 1 , where the notation
j
H j − 1 denotes the matrix acquired by deleting columns k1 , k2 , , k j −1 of H and ( i )
+

denotes the Moore-Penrose pseudoinverse (Abadir & Magnus, 2006). The post-detection
SNR for the k j th substream of s is therefore given by

Es
SNRk j = 2
(66)
σ n2 wkj

⎡ 2⎤
where Es = E ⎢ sk j ⎥ .
⎣ ⎦

3.2.2 MMSE-based V-BLAST algorithm

The MMSE is another well-known criterion for designing V-BLAST data detection. For the
MMSE criterion, we intend to find the equalizer coefficients to minimize the mean squared
error between the transmitted vector s and the equalized output W † r , as follows:
82 Recent Advances in Wireless Communications and Networks

WMMSE = arg min E (s − W † r)2

W
{ } (67)

The optimal MMSE equalizer WMMSE is expressed as

( )
-1
WMMSE = HH† + σ n2 InR H (68)

where σ n2 denotes the noise power and InR represents an nR × nR identity matrix. Thus, the
decision statistic for the ith substream is given by

yi = (WMMSE )†i r (69)

where (WMMSE )i denotes the ith column of the matrix WMMSE . Subsequently, the hard
decision for the ith substream is given by

sˆi = Q ( yi ) (70)

To further improve the performance, one can incorporate the interference cancellation
methodology, similar to the idea in subsection 3.2.1, into the MMSE equalizer. Concerning
an arbitrary detection order, the interference suppression and cancellation procedure is the
same as the Step1~Step3 in subsection 3.2.1, but the MMSE equalizer is used instead of the
j
nulling vector as follows. Define the MMSE equalizer at the jth iteration as WMMSE :

-1
⎛
( ) ⎞
†
j
WMMSE = ⎜ Hdj − 1 Hdj − 1 + σ n2 I nR ⎟ Hdj − 1 (71)
⎝ ⎠

where Hdj − 1 represents the truncated channel matrix by deleting the columns
k1 , k 2 , , k j − 1 of H . At the jth iteration, the weighting vector for detecting the k jth
j
substream, WMMSE ( ) kj
j
, is thus obtained from the k j th column of WMMSE .

3.2.3 Ordering V-BLAST algorithm

Since the ZF-based and MMSE-based V-BLAST data detection algorithms iterate between
equalization and interference cancellation, the order for detecting the substreams of s
becomes an important role to determine the overall performance (Foschini et al., 1999). In
this subsection, we discuss an ordering scheme for the two V-BLAST detectoin algorithms.
Although the ZF or MMSE equalizer can null out or supress the residual inter-antenna
interference, it will introudce the noise enhancement problem, leading to incorrect data
decision, and the incorrect interference reconstruction will cause the error propagation
problem. Assuming that all the substreams adopt the same constellation scheme, among all
the remaining entries of s (not yet detected), the entry with the largest SNR, i.e., from (66),
2
having the minimum norm power w k j , is choosen at each iteration in the detection
process. The iterative procedures for the ordering ZF-based or MMSE-based V-BLAST
detection algorithms are described in the following.
Multiple Antenna Techniques 83

Initialization Step:

Set j = 1
( )
-1
Calculate G1 = H+ (if ZF) or G1 = HH† + σ n2 InR H (if MMSE)
2
Choose k1 = arg min ( G1 )i
i

Recursion Step:

While j ≤ nT
{
Interference supression & cancellation Part:
( )k ( )k
†
Calculate the weighting vector: w †k j = G j (if ZF) or w †k j = G j (if MMSE)
j j

Equalization: y k j = w †k j rj

( )
Slice: sˆk j =Q y k j

Interference cancellation: rj + 1 = rj − s k j ( H ) k
j

Ordering Part:
+
Calculate the equalizer matrix of the updated channel matrix: G j + 1 = H j (if ZF) or
-1
⎛
( ) ⎞
†
G j + 1 = ⎜ H dj H dj + σ n2 I nR ⎟ H dj (if MMSE)
⎝ ⎠
2
Decide the symbol entry for detection: k j + 1 = arg
{
min
i∉ k1 , k2 , ,k }
j
( G j + 1 )i
Update: j = j + 1
}

In the above iterative procedure, for the ZF case, the vector (G j )i denotes the ith row of the
matrix G j , computed from the pseudoinverse of H j − 1 , where the columns k1 , , k j − 1 are
set to zero. However, for the MMSE case, the vector (G j )i denotes the the ith column of
the matrix G j , computed from the MMSE equalizer of Hdj − 1 , where the columns
k1 , , k j − 1 are set to zero. This is because these columns only related to the entries of
sk1 , , sk j which have already been estimated and cancelled. Thus, the system can be
regarded as a degenerated V-BLAST system of Figure 12 where the transmitters k1 , , k j are
removed.

3.2.3 BER Performance of various V-BLAST detection algorithms

The BER performance of the V-BLAST with ML, ZF, and MMSE algorithms is presented in
the following. Both the transmitter and the receiver are equipped with four antennas, and a
flat fading channel is used for simulation. Fig. 13 compares the BER performance of the ML
detector with those of the ZF-based or MMSE-based V-BLAST algorithms without ordering,
in which the detecting order is in sequence from the last transmit antenna to the first
84 Recent Advances in Wireless Communications and Networks

transmit antenna. As compared with the ZF-based and MMSE-based V-BLAST algorithms,
the ML detector has better BER performance. However, the computation complexity of the
ML detector exponentially increases as the modulation order or the number of transmit
antennas increase. Instead, the ZF-based and MMSE-based V-BLAST algorithms, which are
the linear detection methods combined with the interference cancellation methods, require
much lower complexity than the ML detector, but their BER performance is significantly
inferior to that of the ML detector.

0
10
ZF-based V-BLAST
MMSE-based V-BLAST
-1 ML
10

-2
10
BER

-3
10

-4
10

-5
10
0 5 10 15 20 25 30
Eb/N0 [dB]

Fig. 13. BER performance of 4 × 4 V-BLAST systems without ordering

Fig. 14 and Fig. 15 demonstrate the BER performance of the ZF-based and the MMSE-based
V-BLAST algorithm, respectively, with or without ordering. We can observe from these two
figures that the V-BLAST algorithms with ordering can achieve better performance than that
of the algorithms without ordering. An ordered successive interference suppression and
cancellation method can effectively combat the error propagation problem to improve the
BER performance with less complexity, although the ML detector still outperforms the
ordering V-BLAST algorithms.

4. Beamforming
Beamforming is a promising signal processing technique used to control the directions for
transmitting or receiving signals in spatial-angular domain (Godara, 1997). By adjusting
beamforming weights, it can effectively concentrate its transmission or reception of desired
signals at a particular spatial angle or suppress unwanted interference signals from other
spatial angles. In this section, we will introduce the general concepts of beamforming
techniques as well as some famous beamforming methods.
Multiple Antenna Techniques 85

0
10
ZF without ordering
ZF with ordering
-1 ML
10

-2
10
BER

-3
10

-4
10

-5
10
0 5 10 15 20 25 30
E b/N0 [dB]

Fig. 14. BER performance of 4 × 4 V-BLAST systems using ZF-based V-BLAST algorithm
with or without ordering

0
10
MMSE without ordering
MMSE with ordering
-1 ML
10

-2
10
BER

-3
10

-4
10

-5
10
0 5 10 15 20 25 30
Eb/N0 [dB]

Fig. 15. BER performance of 4 × 4 V-BLAST systems using MMSE-based V-BLAST algorithm
with or without ordering
86 Recent Advances in Wireless Communications and Networks

4.1 Linear array and signal model

Fig. 16 depicts a linear array with L omni-directional and equi-spaced antennas, and the
antenna spacing is set as d .

Fig. 16. Linear array

Let the first antenna element of the linear array be the reference point at the origin of the x-y
coordinate system. Consider M uncorrelated far-field sources with the same central carrier
frequency f 0 , for i = 1,… , M , and due to the far-field assumption, each source can be
approximated as a plane wave when arriving at the linear array. Accordingly, the time
arrival of the plane wave from the ith source in the direction of θ i to the lth antenna is
given by

τ l (θi ) = d ( l − 1) sin θi (72)

c
where c is the speed of light. Denote mi ( t ) as the baseband signal of the ith source. By
assuming that the bandwidth of the source is narrow enough, i.e., mi ( t + τ l (θ i ) ) ≅ mi ( t ) , the
bandpass signal on the lth antenna contributed by the ith source is expressed as

{
bi ( t ) = ℜ e mi ( t + τ l (θ i ) ) e
j 2π f 0 ( t +τ l (θ i ) )
} (73)
≅ ℜ e {m ( t ) e }
j 2π f 0 ( t +τ l (θ i ) )
i

Thus, the corresponding baseband equivalent signal on the lth antenna is given by

j 2π f 0τ l (θ i )
mi ( t ) e (74)

Let xl ( t ) denote the baseband representation of the received signal on the lth antenna,
including both M sources and noise on the lth antenna, and from (74), it is given by
M
xl ( t ) = ∑ mi ( t ) e j 2π f 0τ l (θi ) + nl ( t ). (75)
i =1
Multiple Antenna Techniques 87

where nl ( t ) is the spatially additive white Gaussian noise term on the lth antenna with
zero mean and variance σ n2 .

Fig. 17. Antenna beamforming

Fig. 17 shows the spatial signal processing of a beamforming array, in which complex
beamforming weights wl are applied to produce an output of linear combination of the
received signals xl ( t ) , expressed as
L
y ( t ) = ∑ wl* xl ( t ) (76)
l =1
T
Define w = [ w1 , w2 ,… , wL ] and x ( t ) = ⎡⎣ x1 ( t ) , x2 ( t ) ,… , xL ( t ) ⎤⎦ . From (75), we can express
T

x ( t ) in a matrix-vector form as follows

M
x ( t ) = ∑ mi ( t ) a i (θi ) + n ( t ) (77)
i =1

T
T j 2π f 0τ 1 (θ i ) j 2π f 0τ L (θ i ) ⎤
where n ( t ) = ⎣⎡n1 ( t ) , n2 ( t ) ,… , nL ( t ) ⎦⎤ , and a i (θ i ) = ⎡⎢ e ,… , e is known as
⎣ ⎦⎥
the steering vector for the ith source. Then, we can rewrite (77) into a matrix notation,
leading to a compact representation:

y ( t ) = w †x ( t ) (78)

From (77) and (78), for a given beamforming vector, the mean output power is calculated as

P ( w ) = E ⎡⎣ y ( t ) y ∗ ( t ) ⎤⎦ = w †Rw (79)

where E [ i] denotes the expectation operator and R is the correlation matrix of the signal
x ( t ) , which is defined and given by
M
R = E ⎡⎣ x ( t ) x † ( t ) ⎤⎦ = ∑ pi a i a †i + σ n2 I (80)
i =1

where pi is the power of the ith source, I is an identity matrix of size L × L . In the
following, we introduce three famous beamforming schemes to determine the complex
beamforming weights.
88 Recent Advances in Wireless Communications and Networks

4.2 Conventional beamforming

A conventional beamforming scheme is to form a directional beam by merely considering a
single source. Since all its beamforming weights are set with an equal magnitude, this
scheme is also named as delay-and-sum beamforming. Without loss of generality, assume
that the targeted source for reception is in the direction of θ0 , and the beamforming weight
vector is simply given by

1
w= a (θ 0 ) (81)
L
Now we consider a communication environment consisting of only one signal source, and
with this delay-and-sum beamforming scheme, the output signal after beamforming is given
by

y ( t ) = w † x ( t ) = w †a (θ 0 ) m0 ( t ) + w † n ( t ) (82)

By assuming that the power of the source signal is equal to p0 , the post-output SNR after
beamforming is then calculated as

2
w †a (θ 0 ) m0 ( t ) Lp0
SNR = = (83)
E ⎡⎢ w † n ( t ) ⎤⎥
2
σ n2
⎣ ⎦
In fact, the conventional beamforming can be regarded as an MRC-like scheme, as
introduced in (Godara, 1997). It can be easily proved that for the case of a single source, the
conventional beamforming scheme can provide the maximum output SNR. We can also
observe that the output SNR in (83) is proportional to the number of antenna arrays, and as
the number of antennas increases, it can facilitate to reduce the noise effect. However, when
there are a number of signal sources which can interfere with each other, this conventional
beamforming scheme does not have ability to suppressing interference effectively.

4.3 Null-steering beamforming

Consider a communication environment consisting of one desired signal source with an
angle of arrival θ0 and multiple interfering signal sources with angles of arrival θ i , for
i = 1,… , M . To effectively mitigate the mutual interference, a null-steering beamforming
scheme can be designed to null out unwanted signals from some interfering sources with
known directions. The null-steering beamforming scheme is also named as ZF
beamforming. As its name suggested, the design idea is to form beamforming weights with
unity response in the desired source direction θ 0 , while create multiple nulls in the
interfering source directions θ i , for i = 1,… , M . Now assume that the steering vector for the
desired and interfering signal sources are respectively denoted by a 0 and a i , for
i = 1,… , M . Then, the beamforming weights can be designed by solving the following
equations (D’Assumpcao & Mountford, 1984):

w †a 0 = 1
(84)
w †a i = 0, for i = 1, ,M
Multiple Antenna Techniques 89

Using the matrix notation, we can rewrite (84) as follows

w †A = eT1 (85)

where A is a matrix containing M + 1 steering vectors, i.e., A = [ a 0 , a 1 ,… , a M ] , and we

define e1 = [ 1,0, ,0 ] . If the total number of desired and interfering sources is equal to the
T

number of antennas, A is an invertible square matrix as long as θ i ≠ θ j , for all i , j . As a

result, the solution for the beamforming weights is given by

w † = eT1 A −1 (86)

Otherwise, the solution is obtained via taking the pseudo inverse of A . The solution for the
beamforming weights becomes

( )
−1
w † = eT1 A † AA † (87)

It is noted that although this beamforming scheme has nulls in the directions of interfering
sources, it is not designed to maximum the output SNR. Hence, it is not an optimal one from
the viewpoint of maximizing the output SNR.

4.4 Optimal beamforming

Although the null-steering beamforming scheme can deal with the interference problem by
nulling out unwanted signals, it suffers from two critical problems. One is that the null-
beamforming scheme requires the knowledge of interfering source directions, which is not
easily acquired in practice. The other is that the output SNR is not the maximum, even
though this scheme is quite simple. To overcome these two problems in the environment
comprising of one desired signal source with the steering vector a 0 and M interfering
sources with the steering vectors a i , for i = 1,… , M , from (79), we can derive the optimal
beamforming weights by minimizing the total output power, under the constraint that the
output gain in the direction of desired signal source is equal to one; that is

min w †Rw
w (88)
subject to w †a 0 = 1

where R = ∑ i = 0 pi a i a †i + R n , pi is the signal power of the corresponding source, and R n is

M

the covariance matrix of the spatial noise. Here, we consider a more general case where the
noise is not necessary spatial white. This optimal beamforming scheme is also known as
minimum variance distortionless response (MVDR) beamforming. The Lagrange multiplier
can be applied to solve this optimization problem, and the Lagrange function associated
with (88) is given by

( )
L ( w , v ) = w †Rw + v w †a 0 − 1 = w † ( Rw + va 0 ) − v (89)

where v is the Lagrange multiplier. By applying the Karush-Kuhn-Tucker (K.K.T.)

conditions, the sufficient conditions to reach the optimal solution are given by
90 Recent Advances in Wireless Communications and Networks

∇ wL ( w , v ) = 2Rw + va 0 = 0 (90)

w †a 0 = 1 (91)

We can further rewrite (90) as

v∗ −1
( )
†
w† = − R a0 (92)
2
Since R is a Hermitian matrix, i.e., R = R † , by taking (91) into (92), we have

v* 1 (93)
− =
2 a †0R −1a 0

From (92) and (93), the optimal beamforming weights (Vural, 1975; Applebaum, 1976) are
therefore given by

R −1a 0 (94)
w=
a †0R −1a 0

In general, the corelation matrix R = E ⎡⎣ y ( t ) y ∗ ( t ) ⎤⎦ can be approximated by computing the

time average of the received signal power over a period of time duration T, i.e.,
T −1 2
R ≅ ( 1 T ) ∑ t = 0 y ( t ) . It is worthwhile to mention here that for this scheme, the output SNR
can be maximized without requiring the knowledge on the directions of the interference
sources. Now consider a particular case where only one desired signal source exists and
noise is with the covariance matrix R n , i.e., R = p0a 0a †0 + R n . Use the inversion lemma, we
have

p0R n−1a 0a †0R n-1

R −1 = R n−1 − (95)
1 + p0a †0R n−1a 0
After some straightforward manipulation, the optimal beamforming scheme is then
degenerated to

R -1a 0 R n-1a 0 (96)

w= =
a †0R -1a 0 a 0†R n-1a 0
In practice, it is hard to know the noise correlation matrix R n ; however, it is evident from
(96) that one can use R , which could be approximately obtained through the time average
of the received signal power, instead of R n for calculating the optimal beamforming
weights in this single-user case. For another special case where noise is spatially white, the
optimal beamforming scheme is then further degenerated to the conventional beamforming
scheme by substituting Rn = σ n2 I into (96):

R n-1a 0 1 (97)
w= = a0
a 0†R n-1a 0 L
Multiple Antenna Techniques 91

4.5 Performance of various beamforming techniques

We show some examples to evaluate the performance of these three beamforming schemes.
The number of antenna and the carrier frequency of the sources are set as L = 20 and
f0 = 2.3 GHz, respectively. The antenna spacing is set to be half the wavelength. The number
of sources is given by M = 3, and the directions of the desired and interfering signals are
given by 20°, 5°, and 45°, respectively. Fig. 18 compares the angle responses of the
conventional and optimal beamforming schemes. It is shown that the optimal beamforming
scheme can efectively filter out the two interfering sources at the directions of 5° and 45°, as
compared with the conventional beamforming scheme. Fig. 19 compares the angle response
between the null-steering and the optimal beamforming schemes. Although the null-
steering beamforming scheme can form deep nulls in the directions of the interfering
sources, the performance of filtering out the interference sources degrades dramatically
when there is small discrepancy between the true and esimated directions. In other words,
the null-steering scheme requires extremely accurate estimation on the directions of
interfering sources. Since the optimal beamforming does not require the knowledge of the
directions of interfering sources; in fact, it can be performed by using the received signals to
calcuate the correlation matrix in (90) instead, the optimal beamforming has more robust
performance than the null-steering one.

10
Optimal beam former, 20 antennas
Conventional beam former, 20 antennas
0
Amplitude of angle response (dB)

-10

-20

-30

-40

-50

-60

-70
-100 -80 -60 -40 -20 0 20 40 60 80 100
Angle
Fig. 18. Angle responses of optimal and conventional beamforming schemes
 92 Recent Advances in Wireless Communications and Networks

-50
Amplitude of angle response (dB)

-100

-150

-200

-250

-300 Optimal beam former

Null-steering beam former

-350
-100 -80 -60 -40 -20 0 20 40 60 80 100
Angle
Fig. 19. Angle responses of optimal and null-steering schemes

5. References
Abadir, K. M. & Magnus, J. R. (2006). Matrix Algebra, Cambridge University Press, ISBN 978-
0-521-53746-9, New York, USA.
Applebaum, S. P. (1976). Adaptive Arrays, IEEE Transactions on Antennas and Propagation,
Vol.24, No.5, (Sep 1976), pp.585–598, ISSN: 0018-926X.
D’Assumpcao, H. A. & Mountford, G. E. (1984). An Overview of Signal Processing for
Arrays of Receivers, Journal of the Institution of Engineers (Australia), Vol.4, pp.6–19,
ISSN 0020-3319.
Foschini, G. J. (1996). Layered Space-Time Architecture for Wireless Communication in a
Fading Environment When Using Multi-Element Antennas. Bell Labs Technical
Journal, Vol.1, No.2, (Summer 1999), pp.41-59, ISSN 1538-7305.
Foschini, G. J. & Gans, M. J. (1998). On Limits of Wireless Communications in a Fading
Environment when Using Multiple Antennas. Wireless Personal Communications,
Vol.6, No.3, (March 1998), pp.311-335, ISSN 0929-6212.
Foschini, G. J., Golden, G. D., Valenzuela, R. A. & Wolniansky, P. W. (1999). Simplified
Processing for High Spectral Efficiency Wireless Communication Employing Multi-
Multiple Antenna Techniques 93

Element Arrays. IEEE Journal on Selected Areas in Communications, Vol.17, No.11,

(Nov 1999), pp.1841-1852, ISSN 0733-8716.
Giannakis, G. B., Liu, Z., Zhuo, S. & Ma, X. (2006). Space-Time Coding for Broadband Wireless
Communications, John Wiley, ISBN 978-0-471-21479-3, New Jersey, USA.
Godara, L. C. (1997). Application of Antenna Arrays to Mobile Communications. II. Beam-
Forming and Direction-of-Arrival Considerations. Proceedings of IEEE, Vol.85, No.8,
(Aug 1997), pp.1195-1245, ISSN 0018-9219.
Ku, M.-L. & Huang, C.-C. (2006). A Complementary Code Pilot-Based Transmitter Diversity
Technique for OFDM Systems. IEEE Transactions on Wireless Communications, Vol. 5,
No. 2, (March 2006), pp.504-508, ISSN 1536-1276.
Ku, M.-L. & Huang, C.-C. (2008). A Refined Channel Estimation Method for STBC/OFDM
Systems in High-Mobility Wireless Channels. IEEE Transactions on Wireless
Communications, Vol. 7, No. 11, (Nov 2008), pp.4312-4320, ISSN 1536-1276.
Lin, J.-C. (2009). Channel Estimation Assisted by Postfixed Pseudo-Noise Sequences Padded
with Zero Samples for Mobile Orthogonal-Frequency-Division-Multiplexing
Communications. IET Communications, Vol.3, No.4, (April 2009), pp.561-570, ISSN
1751-8628.
Lin, J.-C. (2009). Least-Squares Channel Estimation Assisted by Self-Interference Cancellation
for Mobile Pseudo-Random-Postfix Orthogonal-Frequency-Division Multiplexing
Applications. IET Communications, Vol.3, No.12, (Dec 2009), pp.1907-1918, ISSN
1751-8628.
Lo, T. K. Y. (1999). Maximum Ratio Transmission. IEEE Transactions on Communications, Vol.
47, No. 10, (Oct 1999), pp.1458-1461, ISSN 0090-6778.
Rappaport, T. S. (2002). Wireless Communications: Principles and Practice (2nd Edition),
Prentice Hall, ISBN 0-13-042232-0, New Jersey, USA.
Simon, M. K. & Alouini, M. S. (1999). A Unified Performance Analysis of Digital
Communication with Dual Selective Combining Diversity over Correlated Rayleigh
and Nakagami-m Fading Channels. IEEE Transactions on Communications, Vol.47,
No.1, (Jan 1999), pp.33-43, ISSN 0090-6778.
Tarokh, V., Seshadri, N. & Calderbank, A. R. (1998). Space-Time Codes for High Data Rate
Wireless Communication: Performance Criterion and Code Construction.
Information Theory. IEEE Transactions on Information Theory, Vol.44, No.2, (Mar
1998), pp.744-765, ISSN 0018-9448.
Tarokh, V., Jafarkhani, H. & Calderbank, A. R. (1999). Space-Time Block Codes from
Orthogonal Designs. IEEE Transactions on Information Theory, Vol.45, No.5, (Jul
1999), pp.1456-1467, ISSN 0018-9448.
Vural, A. M. (1975). An Overview of Adaptive Array Processing for Sonar Application.
Proceedings of Electronics and Aerospace Systems Conference, 1975 (EASCON’75),
pp.34A–34M.
Wittneben, A. (1993). A New Bandwidth Efficient Transmit Antenna Modulation Diversity
Scheme for Linear Digital Modulation. Proceedings of IEEE Communications
Conference, pp.1630-1634, ISSN: 0-7803-0950-2, Geneva, Switzerland, May 23-26,
1993.
94 Recent Advances in Wireless Communications and Networks

Wolniansky, P. W., Foschini, G. J., Golden, G. D. & Valenzuela, R. A. (1998). V-BLAST: An

Architecture for Realizing Very High Data Rates over The Rich-Scattering Wireless
Channel. Proceedings of URSI International Symposium on Signals, Systems, and
Electronics, pp.295-300, ISSN 0-7803-4900-8, Pisa, Italy, Sep 29-Oct 2, 1998.
 0
5

Diversity Management in MIMO-OFDM Systems

Felip Riera-Palou and Guillem Femenias
Mobile Communications Group
University of the Balearic Islands
Spain

1. Introduction
Over the last decade, a large degree of consensus has been reached within the research
community regarding the physical layer design that should underpin state-of-the-art and
future wireless systems (e.g., IEEE 802.11a/g/n, IEEE 802.16e/m, 3GPP-LTE, LTE-Advanced).
In particular, it has been found that the combination of multicarrier transmission and
multiple-input multiple-output (MIMO) antenna technology leads to systems with high
spectral efficiency while remaining very robust against the hostile wireless channel
environment.
The vast majority of contemporary wireless systems combat the severe frequency selectivity
of the radio channel using orthogonal frequency diversity multiplexing (OFDM) or some of its
variants. The theoretical principles of OFDM can be traced back to (Weinstein & Ebert, 1971),
however, implementation difficulties delayed the widespread use of this technique well until
the late 80s (Cimini Jr., 1985). It is well-known that the combination of OFDM transmission
with channel coding and interleaving results in significant improvements from an error rate
point of view thanks to the exploitation of the channel frequency diversity (Haykin, 2001, Ch.
6). Further combination with spatial processing using one of the available MIMO techniques
gives rise to a powerful architecture, MIMO-OFDM, able to exploit the various diversity
degrees of freedom the wireless channel has to offer (Stuber et al., 2004).

1.1 Advanced multicarrier techniques

A significant improvement over conventional OFDM was the introduction of multicarrier
code division multiplex (MC-CDM) by Kaiser (2002). In MC-CDM, rather than transmitting
a single symbol on each subcarrier, as in conventional OFDM, symbols are code-division
multiplexed by means of orthogonal spreading codes and simultaneously transmitted onto
the available subcarriers. Since each symbol travels on more than one subcarrier, thus
exploiting frequency diversity, MC-CDM offers improved resilience against subcarrier fading.
This technique resembles very much the principle behind multicarrier code-division multiple
access (MC-CDMA) where each user is assigned a specific spreading code to share a group of
subcarriers with other users (Yee et al., 1993).
A more flexible approach to exploit the frequency diversity of the channel is achieved by
means of group-orthogonal code-division multiplex (GO-CDM) (Riera-Palou et al., 2008). The
idea behind GO-CDM, rooted in a multiple user access scheme proposed in (Cai et al., 2004),
is to split suitably interleaved symbols from a given user into orthogonal groups, apply a
spreading matrix on a per-subgroup basis and finally map each group to an orthogonal set of
96
2 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

subcarriers. The subcarriers assigned to a group of symbols are typically chosen as separate
as possible within the available bandwidth in order to maximise the frequency diversity gain.
Note that a GO-CDM setup can be seen as many independent MC-CDM systems of lower
dimension operating in parallel. This reduced dimension allows the use of optimum receivers
for each group based on maximum likelihood (ML) detection at a reasonable computational
cost. In (Riera-Palou et al., 2008), results are given for group dimensioning and spreading code
selection. In particular, it is shown that the choice of the group size should take into account
the operating channel environment because an exceedingly large group size surely leads to a
waste of computational resources, and even to a performance degradation if the channel is not
frequency-selective enough. Given the large variation of possible scenarios and equipment
configurations in a modern wireless setup, a conservative approach of designing the system
to perform satisfactorily in the most demanding type of scenario may lead to a significant
waste of computational power, an specially scarce resource in battery operated devices. In
fact, large constellation sizes (e.g., 16-QAM, 64-QAM) may difficult the application of GO
techniques as the complexity of ML detection can become very high even when using efficient
implementations such as the sphere decoder (Fincke & Pohst, 1985). In order to minimise the
effects of a mismatch between the operating channel and the GO-CDM architecture, group
size adaptation in the context of GO-CDM has been proposed in (Riera-Palou & Femenias,
2009), where it is shown that important complexity reductions can be achieved by dynamically
adapting the group size in connection with the sensed frequency diversity of the environment.

1.2 Multiple antennae schemes

Multiple-antenna technology (i.e., MIMO) is the other main enabler towards high speed
robust wireless networks. Whereas the use of multiple antennae at the receiver has been
long applied as an effective measure to combat fading (see, e.g. (Simon & Alouini, 2005)
and references therein), it is the application of multiple antenna at the transmitter side what
revolutionised the wireless community. In particular, the linear increase in capacity achieved
when jointly increasing the number of antennas at transmission and reception, theoretically
forecasted in (Telatar, 1999), has spurred research efforts to effectively realize it in practical
schemes. Among these practical schemes, three of them have achieved notable importance
in the standardisation of modern wireless communications systems, namely, spatial division
multiplexing (SDM), space-time block coding (STBC) and cyclic delay diversity (CDD). While
in SDM (Foschini, 1996), independent data streams are sent from the different antennas
in order to increase the transmission rate, in STBC (Alamouti, 1998; Tarokh et al., 1999)
the multiple transmission elements are used to implement a space-time code targeting the
improvement of the error rate performance with respect to that achieved with single-antenna
transmission. In CDD (Wittneben, 1993) a single data stream is sent from all transmitter
antennae with a different cyclic delay applied to each replica, effectively resulting as if the
original stream was transmitted over a channel with increased frequency diversity.

1.3 Chapter objectives

The combination of GO-CDM and MIMO processing, termed MIMO-GO-CDM, results in a
powerful and versatile physical layer able to exploit the channel variability in space and
frequency. Nevertheless, the different MIMO processing schemes coupled with different
degrees of frequency multiplexing (i.e., different group sizes) gives rise to a vast amount
of combinations each offering a different operating point in the performance/complexity
plane. Choosing an adequate number of Tx/Rx antennas, a specific MIMO scheme and the
Diversity Management
Diversity Management in MIMO-OFDMin MIMO-OFDM Systems
Systems 973

subcarrier grouping dimensions can be a daunting task further complicated when Tx and/or
Rx antennas are correlated. To this end, it is desirable to have at hand closed-form analytical
expressions predicting the performance of the different MIMO-GO-CDM configurations in
order to avoid the need of (costly) numerical simulations.
This chapter has two main goals. The first goal is to present a unified BER analysis of the
MIMO-GO-CDM architecture. In order to get an insight of the best possible performance
this system can offer, attention is restricted to the case when ML detection is employed
at the receiver. The analysis is general enough to incorporate the effects of channel
frequency selectivity, Tx/Rx antenna correlations and the three most common different forms
of spatial processing (SDM, STBC and CDD) in combination with GO-CDM frequential
diversity. The analytical results are then used to explore the benefits of GO-CDM under
different spatial configurations identifying the most attractive group dimensioning from
a performance/complexity perspective. Based on the previous analysis, the second goal
of this chapter is to devise effective reconfiguration strategies that can automatically and
dynamically fix some of the parameters of the system, more in particular the group size of
the GO-CDM component, in response to the instantaneous channel environment with the
objective of optimising some pre-defined performance criteria (e.g., error rate, complexity,
delay).
The rest of this chapter is organized as follows. Section 2 introduces the system model of a
generic MIMO-GO-CDM system, paying special attention to the steps required to implement
the frequency spreading and the MIMO processing. In Section 3 a unified BER analysis
is presented for the case of ML detection. In light of this analysis, Section 4 explores
reconfiguration strategies aiming at the optimisation of several critical parameters of the
MIMO-GO-CDM architecture. Numerical results are presented in Section 6 to validate the
introduced analytical and reconfiguration procedures. Finally, the main conclusions of this
work are recapped in Section 7.
Notational remark: Vectors and matrices are denoted by bold lower and upper case letters,
respectively. The superscripts ∗ , T and H are used to denote conjugate, transpose and complex
transpose (Hermitian), respectively, of the corresponding variable. The operation vec(A)
lines up the columns forming matrix A into a column vector. The symbols ⊗ and  denote
the Kronecker and element-by-element products of two matrices, respectively. Symbols Ik
and 1k×l denote the k-dimensional identity matrix and an all-ones k × l matrix, respectively.
The symbol D(x) is used to represent a (block) diagonal matrix having x at its main (block)
diagonal. The determinant of a square matrix A is represented by |A| whereas x2 = xx H .
Expression  a is used to denote the nearest upper integer of a. Finally, the Alamouti
 
transform of a K × 2 matrix X = [x1 x2 ] is defined as A (X )  −x2∗ x1∗ .

2. MIMO GO-CDM system model

We consider a MIMO multicarrier system with Nc data subcarriers, equipped with NT and NR
transmit and receive antennas, respectively, and configured to transmit Ns (≤ NT ) spatial data
streams. Following the group-orthogonal design principles, the available subcarriers are split
into Ng = Nc /Q groups of Q subcarriers each. In the following subsections the transmitter,
channel model and reception equation are described in detail.

2.1 Transmitter
As depicted in Fig. 1, incoming bits are split into Ns spatial streams, which are then processed
separately. Bits on the zth stream are mapped onto a sequence sz of symbols drawn from an
98
4 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

GO-CDM

Symbol Segment Grouping CDD IFFT CP

Spreading
mapping S/P
Spatial stream parser

Antenna mapping
Input bits

STBC
GO-CDM

Symbol Segment Grouping Spreading

mapping S/P CDD IFFT CP

Fig. 1. Transmitter architecture for MIMO GO-CDM.

M-ary complex constellation (e.g. BPSK, M-QAM) with average normalized unit energy. The
resulting Ns streams of modulated symbols {sz }zN=s 1 are then fed to the GO-CDM stage, which
comprises three steps:
1. Segmentation of the incoming symbol stream in blocks of length Nc (i.e., eventual OFDM
symbols), and serial to parallel conversion (S/P) resulting, over the kth OFDM symbol
period, in sz (k).
  Ng
2. Arrangement of the symbols in the block into groups szg (k) , where szg (k) =
g =1
 T
szg,1 (k) . . . szg,Q (k ) represents an individual group.
3. Group spreading through a linear combination

1
s̃zg (k) = √ Cszg (k), (1)
NT

where C is a Q × Q orthonormal matrix, typically chosen to be a rotated Walsh-Hadamard

matrix (Riera-Palou et al., 2008).
Before the usual OFDM modulation steps on each antenna (IFFT, guard interval appending
and up-conversion), the grouped and spread symbols are processed in accordance with the
MIMO transmission scheme in use as follows:
SDM (Ns = NT ) : In this case the blocks labeled in Fig. 1 as STBC and CDD are not used,
and the spread symbols are directly supplied to the antenna mapping stage, which simply
connects the incoming zth data stream to the ith transmit branch (1 ≤ i ≤ NT ), that is,

s̆ig (k) = ŝig (k) = s̃zg (k). (2)

STBC (Ns = 1, NT = 2) : Two consecutive blocks of spread symbols, s̃1g (k) and s̃1g (k + 1), are
Alamouti-encoded on a per-subcarrier basis over two OFDM symbol periods,
 ∗
ŝ1g (k) = s̃1g (k), ŝ1g (k + 1) = − s̃1g (k + 1) ,
 ∗ (3)
ŝ2g (k) = s̃1g (k + 1), ŝ2g (k + 1) = s̃1g (k) .
Diversity Management
Diversity Management in MIMO-OFDMin MIMO-OFDM Systems
Systems 995

In the antenna mapping stage, STBC-encoded streams are connected to two transmit
branches, one for each symbol of the STBC code, that is,

s̆ig (k ) = ŝig (k). (4)

CDD (Ns = 1) : In a pure CDD scheme, the same data stream is sent through NT antennas
with each replica being subject to a different cyclic delay Δi , typically chosen as Δi =
Δi−1 + Nc /NT with Δ1 = 0 (Bauch & Malik, 2006), resulting in transmitted symbols


s̆ig,q (k) = s̃1g,q (k) exp − j2πdq Δi /Nc , (5)

where dq denotes the subcarrier index.

Hybrid schemes The analytical framework developed in this chapter can also be applied
to hybrid systems combining SDM, STBC and/or CDD. Nevertheless, for brevity of
presentation, the analysis to be developed next focuses on scenarios where only one of
the mechanisms is used.

2.2 Channel model

The channel linking an arbitrary pair of Tx and Rx antennas is assumed to be time-varying
and frequency-selective with an scenario-dependent power delay profile
P −1
S(τ ) = ∑ φl δ(τ − τl ), (6)
l =0

where P denotes the number of independent paths of the channel and φl and τl denote the
power and delay of the l-th path. It is assumed that the power delay profile is the same for
all pairs of Tx and Rx antennas and that it has been normalized to unity (i.e., ∑lP=−01 φl = 1). A
single realization of the channel impulse response between Tx antenna i and receive antenna
j at time instant t will then have the form
P −1
ij
hij (t; τ ) = ∑ hl (t)δ(τ − τl ), (7)
l =0
 
ij
where it will hold that E | hl (t) |2 = φl . The corresponding frequency response can be
expressed as
P −1
ij
h̄ij (t; f ) = ∑ hl (t) exp(− j2π f τl ), (8)
l =0
which when evaluated at the Nc OFDM subcarriers yields
 T
h̄ij (t) = h̄ij (t; f 0 ) . . . h̄ij (t; f Nc −1 ) . (9)

In order to simplify the notation, assuming that the channel is static over the duration of a
block (i.e., an OFDM symbol), the frequency response between Tx-antenna i and Rx-antenna
j over the Nc subcarriers during the kth OFDM symbol can be expressed as
 T
ij ij
h̄ij (k) = h̄0 (k ) . . . h̄ Nc −1 (k) . (10)
100
6 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

CP FFT ML group Segment Symbol

Spatial stream deparser

detection P/S De-map

STBC pre-processing

Subcarrier grouping

Estimated bits
ML group Segment Symbol
CP FFT detection P/S De-map

Fig. 2. Receiver architecture for MIMO-GO-CDM.

Since the subsequent analysis is mostly conducted on per-group basis, the channel frequency
response for the gth group is denoted by
 T
ij ij ij
h̄g (k ) = h̄ g,1 (k) . . . h̄ g,Q (k) , (11)

with correlation matrix given by

    H
ij ij ij
Rhg = E h̄g (k)2 = E h̄g (k) h̄g (k) , (12)

which is assumed to be constant over time, common for all pairs of Tx and Rx antennas
and, provided that group subcarriers are chosen equispaced across the available bandwidth,
common to all groups.
Now, considering the spatial correlation introduced by the transmit and receive antenna
arrays, the spatially correlated channel frequency response for an arbitrary subcarrier q in
group g can be expressed as (van Zelst & Hammerschmidt, 2002)
 T
H g,q (k) = R1/2 1/2
RX H g,q ( k ) R TX , (13)

where R RX and R TX are, respectively, NR × NR and NT × NT matrices denoting the receive

and transmit correlation, and
⎛ 1NT
⎞
g,q ( k ) . . . h̄ g,q ( k )
h̄11
⎜ .. .. ⎟
Hg,q (k) = ⎜
⎝ . .
⎟.
⎠ (14)
NR 1 NR NT
h̄ g,q (k) . . . h̄ g,q (k)

2.3 Receiver
As shown in Fig. 2, the reception process begins by removing the cyclic prefix and performing
an FFT to recover the symbols in the frequency domain. After S/P conversion, and assuming
ideal synchronization at the receiver side, the received samples for group g at the output of
the FFT processing stage can be expressed in accordance with the MIMO transmission scheme
in use as follows:
Diversity Management
Diversity Management in MIMO-OFDMin MIMO-OFDM Systems
Systems 1017

SDM and CDD: In these cases,

 
rg (k) = vec rg,1 (k) . . . rg,Q (k) = H g (k)s̆g (k) + υg (k), (15)

where the NR Q × NT Q matrix

 
H g (k) = D H g,1 (k) . . . H g,Q (k) , (16)

represents the spatially and frequency correlated channel matrix affecting all symbols
transmitted in group g, the Ns Q-long vector of transmitted (spread) symbols is formed
as    T
s̆g (k) = vec s̆1g (k) . . . s̆gNT (k) , (17)

and finally, υg (k) is an NR Q × 1 vector representing the receiver noise, with each
component being drawn from a circularly symmetric zero-mean white Gaussian
distribution with variance συ2 .
STBC: As stated in (3), STBC encoding period η = k/2, with k = 0, 2, 4, . . ., spawns
two consecutive OFDM symbol periods, namely, the kth and (k + 1)th symbol periods.
Assuming that the channel coherence time is large enough to safely consider that H g (k +
1) = H g (k), then,

r̃g (k) = H g (k)s̆g (k) + υg (k),

(18)
r̃g (k + 1) = H g (k)s̆g (k + 1) + υg (k + 1),
and, therefore, we can define an equivalent received vector in STBC encoding period η as
     
r̃g (k) H g (k) υg (k)
rg (η )  ∗ = A s̃g (η ) + ∗  H̃ g (η )s̃g (η ) + υ̃g (η ), (19)
r̃g (k + 1) H g (k) υ g ( k + 1)

where 
  
HA
g (k)  D A H g,1 (k) . . . A H g,Q (k) (20)
and  T 
s̃g (η )  vec s̃1g (k) s̃1g (k + 1) . (21)

In order to facilitate the unified performance analysis of the different MIMO strategies, it is
more convenient to express the reception equation in terms of the original symbols rather than
the spread ones. Thus, defining
 T 
Ns
sg (k ) = N vec sg (k) . . . sg (k)
√ 1 1 SDM
T
 T 
sg (η ) = √1 vec s1g (k) s1g (k + 1) STBC (22)
2
sg (k) = √1N s1g (k) CDD
T

it is straightforward to check that the symbols to be supplied to the IFFT processing step are
given by,
s̆g (k) = (C ⊗ I Ns ) sg (k) SDM
s̆g (k) = s̃g (η ) = (C ⊗ I2 ) sg (η ) STBC
s̆g (k) = EgΔ (C ⊗ 1 NT ×1 ) sg (k) CDD
102
8 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks
   
ΔQ Δq
with EgΔ  D EgΔ1 . . . Eg , where Eg = D e− j2πdq Δ1 /Nc . . . e− j2πdq Δ NT /Nc (Bauch &
Malik, 2006). Furthermore, since processing takes place either on an OFDM symbol basis for
SDM and CDD systems or on an STBC encoding period basis for STBC schemes, the indexes
k and/or η can be dropped from this point onwards, allowing the reception equation to be
expressed in general form as
rg = Ag sg + νg
where ⎧
⎪
⎪ H g (C ⊗ I Ns ) SDM
⎨
Ag = H̃ g (C ⊗ I2 ) STBC
⎪
⎪
⎩
H g EgΔ (C ⊗ 1 NT ×1 ) CDD
and 
υg for SDM/CDD
νg = . (23)
υ̃g for STBC
It should be noted that, regardless of the MIMO scheme and group dimension in use, the
system matrix Ag has been normalised such that the SNR can be defined as Es /N0 = 1/(2συ2 ).
Upon reception, all symbols in a group (for all streams in SDM and for both encoded OFDM
symbols in STBC) are jointly estimated using an ML detection process. That is, the vector of
estimated symbols in a group can be expressed as

s̄g = arg min Ag sg − rg 2 . (24)

sg

This procedure amounts to evaluate all the possible transmitted vectors and choosing the
closest one (in a least-squares sense) to the received vector. Nevertheless, sphere detection
(Fincke & Pohst, 1985) can be used for efficiently performing the exhaustive search required
to implement the ML estimation.

3. Unified bit error rate analysis

3.1 BER analysis based on pairwise error probability
Using the well-known union bound (Simon et al., 1995), which is very tight for high
signal-to-noise ratios, the bit error probability can be upper bounded as

Ng M NQ M NQ
1

Pb ≤
Ng NQ M NQ
∑ ∑ ∑
log2 M g=1 u=1 w=1
P sg,u → sg,w Nb (sg,u , sg,w ), (25)
w
=u

where,
⎧
⎪
⎪ Q Ns for SDM
⎨
NQ = 2Q for STBC . (26)
⎪
⎪
⎩
Q for CDD


The expression P sg,u → sg,w , usually called the pairwise error probability (PEP), represents
the probability of erroneously detecting the vector sg,w when sg,u was transmitted and
Diversity Management
Diversity Management in MIMO-OFDMin MIMO-OFDM Systems
Systems 1039

Nb (sg,u , sg,w ) is equal to the number of differing bits between vectors sg,u and sg,w . To
proceed further, the PEP conditioned on Ag can be shown to be (Craig, 1991)
⎛ ⎞

 1 Ag (sg,u − sg,w )2
P sg,u → sg,w |Ag = erfc ⎝ ⎠
2 4συ2
 (27)

1 π/2 Ag (sg,u − sg,w )2
= exp − dφ.
π 0 4συ2 sin2 φ

Now, defining the random variable d2g,uw  Ag (sg,u − sg,w )2 , the unconditional PEP can be
expressed as

  
1 π/2 +∞ − x/4σv2 sin2 φ
P sg,u → sg,w = e pd2g,uw ( x ) dx dφ
π 0 −∞
 π/2   (28)
1 1
= Md2g,uw − 2 2 dφ,
π 0 4συ sin φ

where p x (·) and M x (·) denote the probability density function (pdf) and moment generating
function (MGF) of a random variable x, respectively.
Let us now define the error vector eg,uw = sg,u − sg,w . Using this definition, it can be shown
that
d2g,uw  Ag eg,uw 2 = H gH Tg,uw
H
Tg,uw H g , (29)
where 
  
H g  vec vec H g,1 . . . vec H g,Q , (30)
and Tg,uw can be expressed as

 
1Q×1 ⊗ Sg,uw  IQ,NT ⊗ I NR SDM/CDD
Tg,uw = (31)
11×Q ⊗ Sg,uw
T  IQ,2
T
⊗ I2NR STBC

with 

eTg,uw C T ⊗ I NT SDM/STBC
Sg,uw =
 (32)
eTg,uw C T ⊗ 11× NT EΔT CDD

and In,m  In ⊗ 11×m . The expression of d2g,uw reveals that it is a quadratic form in complex
variables H g , with MGF given by
! ! −1
Md2g,uw (s) = !I N − sGg,uw ! , (33)

where N is equal to QNR for the SDM and CDD schemes, and equal to 4QNR for the STBC
strategy. Furthermore,
Gg,uw = Tg,uw Rg Tg,uw
H
, (34)
with
Rg = Rhg ⊗ R TX ⊗ R RX . (35)
104
10 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Now, let λg,uw = {λ g,uw,1 , . . . , λ g,uw,Dg,uw } denote the set of Dg,uw distinct positive eigenvalues
 
of Gg,uw with corresponding multiplicities αg,uw = α g,uw,1 , . . . , α g,uw,Dg,uw . Using the
results in (Femenias, 2004), the MGF of d2g,uw can also be expressed as
Dg,uw Dg,uw α g,uw,d
1 κ g,uw,d,p
Md2g,uw (s) = ∏ (1 − sλ g,uw,d ) α g,uw,d = ∑ ∑ (1 − sλ g,uw,d ) p
(36)
d =1 d =1 p =1

where, using (Amari & Misra, 1997, Theorem 1), it can be shown that
⎡ ⎤!
!
p−α
λ g,uw,d
g,uw,d
α g,uw,d − p Dg,uw !
∂ ⎢ 1 ⎥!
⎢ ⎥!
(α g,uw,d − p)! ∂sαg,uw,d − p ⎣ d∏
κ g,uw,d,p = α g,uw,d ⎦!
 =1 (1 − sλ g,uw,d ) !
d
=d !
s= λ 1
g,uw,d (37)
nd α g,uw,d +nd −1
λ g,uw,d ( )
Dg,uw
p−α nd
= λ g,uw,d ∑ ∏
g,uw,d
 λ 
α g,uw,d  + n d
Φ d  =1 1− g,uw,d
λ g,uw,d
d
=d
 
with Φ being the set of nonnegative integers n1 , . . . , nd−1 , nd+1 , . . . , n Dg,uw such that
∑d
=d nd = α g,uw,d − p, which allows (28) to be written as
⎛ ⎞p

 D α  π/2 2
1 g,uw g,uw,d sin φ
P sg,u → sg,w =
π d∑ ∑ κ g,uw,d,p 0 ⎝ 2 λg,uw,d ⎠ dφ
=1 p =1 sin φ + 4σ2
v
⎛  ⎞p ⎛  ⎞g (38)
λ g,uw,d   1 + Ω λg,uw,d
Dg,uw α g,uw,d 1 − Ω 4σ 2
p − 1
p − 1 + g 4σ 2
= ∑ ∑ κ g,uw,d,p ⎝ v ⎠ ∑ ⎝ v ⎠ ,
d =1 p =1
2 g =0
g 2
(
with Ω(c) = c/(1 + c). By substituting (38) into (25), a closed-form BER upper bound for an
arbitrary power delay profile is obtained. It is later shown that this bound is tight, accurately
matching the simulation results.

3.2 BER analysis based on PEP classes

Since there are many pairs (sg,u , sg,w ) giving exactly the same PEP, it is possible to define
a pairwise error class C( Dg,c , λg,c , αg,c ) as the set of all pairs (sg,u , sg,w ) characterized by
a common matrix Gg,uw = Gg,c with Dg,c distinct eigenvalues λg,c = {λ g,c,1 , . . . , λ g,c,Dg,c }
with corresponding multiplicities αg,c = {α g,c,1 , . . . , α g,c,Dg,c } and therefore, a common PEP
denoted by P ( Dg,c , λg,c , αg,c ). A more insightful BER expression can then be obtained by
using the PEP class notation, avoiding in this way the exhaustive computation of all the
PEPs. Instead, the BER upper-bound can be found by computing the PEP for each class and
weighing it using the number of elements in the class and the number of erroneous bits this
class may induce. The BER upper bound can then be rewritten as
1
Pb ≤
Ng NQ M NQ log2 M
Ng NQ log2 M (39)
× ∑ ∑ ∑ N W ( Dg,c , λg,c , αg,c , N )P ( Dg,c , λg,c , αg,c ),
g=1 ∀C( Dg,c ,λg,c ,αg,c ) N =1
Diversity Management
Diversity Management in MIMO-OFDMin MIMO-OFDM Systems
Systems 105
11

where W ( Dg,c , λg,c , αg,c , N ) corresponds to the number of elements in the class
C( Dg,c , λg,c , αg,c ) inducing N erroneous bits.

3.3 Asymptotic performance

Now, in order to gain further insight on the parameters affecting the BER performance, let us
focus on the asymptotic case of large SNR. When Es /N0 → ∞, the argument of the MGF in
(28) also tends to infinity, and it can easily be shown that when s → ∞ the MGF in (36) can be
approximated by
1
Md2g,uw (s)   , (40)
Dg,uw α g,uw,d Dd,uw
∏d=1 λ g,uw,d (−s)∑d=1 αg,uw,d
allowing the asymptotic PEP of the different classes to be expressed as


 1 π/2 (4συ2 sin2 φ) D̃g,c (2 D̃g,c )! ( Es /N0 )− D̃g,c
Pasym Dg,c , λg,c , αg,c = dφ = , (41)
π 0 D α
∏d=1 λ g,c,d
g,c g,c,d 2 D̃g,c !2 ∏ D̃min λαg,c,d
d=1 g,c,d

D
where D̃g,c = ∑d=g,c1 α g,c,d is the rank of the matrix-defining class Gg,c . From (41) it is clear that
the probability of error will be mainly determined by the groups and classes whose matrices
 H
Gg,c = Gmin min min
g,c  T g,c R g
min
Tg,c (42)

have the smallest common rank, denoted by

  H 
D̃min = rank(Gmin
g,c ) = rank T min min
g,c R g T min
g,c , (43)

allowing the asymptotic BER to be written as

Ng NQ log2 M − D̃
(2 D̃min )! W ( D̃min , λg,c , αg,c , N ) ( Es /N0 ) min
Pb ≤ ∑ ∑ ∑ N
2( D̃min !)2 α g,c,d . (44)
g=1 ∀C( D̃min ,λg,c ,αg,c ) N =1 Ng NQ M NQ log2 M ∏dD̃=min 1 λ g,c,d

In light of (44), the asymptotic BER minimisation is achieved by maximising the minimum
group/class rank D̃min and the eigenvalue product of all the groups/classes with rank D̃min .
In the following, only the maximization of D̃min (i.e., maximisation of the diversity order)
is pursued since the maximization of the product of eigenvalues is far more difficult as it
involves the simultaneous optimization of all the eigenvalue products in the groups/classes
with rank D̃min .
On the rank of Tg,c
min : As stated in (Cai et al., 2004; Riera-Palou et al., 2008), choosing the

subcarriers for a group equispaced across the whole bandwidth minimizes subcarrier
correlation allowing the optimization of the system performance if an adequate family of
spreading codes is properly selected. To this end, rotated spreading transforms have been
proposed for multicarrier systems in (Bury et al., 2003) where it is shown that the often
used Walsh-Hadamard codes lead to poor diversity gains when employed to perform the
frequency spreading. This can be explained by the fact that for certain symbol blocks the
energy is concentrated on one single subcarrier and, thus,
 N SDM
min R
rank Tg,c = (45)
NT NR STBC/CDD.
106
12 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

A deep fade on this subcarrier dramatically raises the probability of error in the detection
process, regardless of the state of all other subcarriers, limiting in this way the achievable
diversity order (asymptotic BER slope). A similar effect can be observed when using
other spreading sequences such as those based on the discrete Fourier transform (DFT).
As pointed out in (Bury et al., 2003), a spreading that has the potential to maximize the
diversity order can be found by applying a rotation to the columns of the conventional
spreading matrix Cconv as C = Cconv D(θ ), where θ = [θ1 . . . θQ ] with each θq denoting
the chip-specific rotation, which in the proposed scheme is given by
 
j2π (q − 1)
θq = exp ,
QΘ

with Θ being constellation dependent and selected so as to make 2π/Θ the minimum
angle producing a rotation of the transmit symbol alphabet onto itself (e.g., Θ = 2 for
BPSK, Θ = 4 for MQAM). This indicates that while using conventional Walsh-Hadamard
spreading no frequency diversity gain will be achieved, the rotated spreading has the
potential (depending on the channel correlation matrix Rg ) to attain a frequency diversity
gain proportional to the number of subcarriers per group, common to all groups and
classes. That is, when using optimally rotated spreading codes,
 QN SDM
min R
rank Tg,c = (46)
Q NT NR STBC/CDD.

g : The correlation matrix R g

On the rank of Rmin min can be expressed in general form as

Rmin
g = Rmin
h g ⊗ R TX ⊗ R RX , (47)

and consequently (Petersen & Pedersen, 2008),

 
rank Rmin
g = rank Rminhg rank (R TX ) rank (R RX ) . (48)

Except for pathological setups exhibiting full spatial correlation between pairs of transmit
or receive antennas (scenario not considered in this analysis), R TX and R RX are full rank
matrices with rank (R TX ) = NT and rank (R RX ) = NR , and therefore,
 
rank Rmin
g = NT NR rank Rmin hg . (49)

Therefore, the maximum attainable frequency diversity order can be directly related to
Rmin
h g and is given by the number of independent paths in the channel delay profile. If
error performance isto be
optimized, enough subcarriers per group need to be allocated
to ensure that rank Rminhg = P. In fact, defining the sampled channel order L as the
channel delay spread in terms of chip (sampling) periods, it is shown in Cai et al. (2004)
that the maximum rank of Rminh g is attained by setting the number of subcarriers per group
to Q = L + 1. While this is a valuable design rule in channels with short delay spread, in
most practical scenarios where L can be in the order of tens or even hundreds of samples,
the theoretical number of subcarriers required to achieve full diversity would make the use
of ML detection difficult even when using efficient search strategies (i.e., sphere decoding).
Diversity Management
Diversity Management in MIMO-OFDMin MIMO-OFDM Systems
Systems 107
13

Moreover, very often maximum diversity would only be attained at unreasonably large
Es /N0 levels.
In order to determine the number of subcarriers worth using in a given environment (i.e.,
a particular channel power delay profile), it is useful to use as reference the characteristics
of the ideal case where all subcarriers in the group are fully uncorrelated (frequency
domain iid channel). It is straightforward to see
 that, in this case, the frequency correlation
matrix is given by Rmin hg = IQ , with rank Rmin hg = Q, and furthermore, it has only
one non-zero eigenvalue λhg ,1 = 1 with multiplicity αhg ,1 = Q. Therefore, for any given
MIMO configuration and a fixed number of subcarriers, the frequency domain iid channel
results in the maximum frequency diversity order (Q) and will also lead to the minimum
probability of error.
Since, for most realistic scenarios, setting the group size to guarantee full diversity (Q =
L + 1) is unfeasible, we need to be able to measure what each additional subcarrier is
contributing in terms of frequency diversity gain. Ideally, each additional  subcarrier

should bring along an extra diversity order, that is, an increase in rank Rmin hg by one
as it is indeed the case for uncorrelated channels. For correlated channels, however, this
is often not the case and therefore to choose the group size it is useful to have some form
of measure. A widely used tool in principal component analysis (Johnson & Wichern,
2002) to assess the practical dimensionality of a correlation matrix is the cumulative sum
 Q
of eigenvalues (CSE) that, for the correlation matrix Rmin
h g with eigenvalues λ h g ,q , is
q =1
defined as
∑nq=1 λhg ,q
Ψ(n) = Q
. (50)
∑q=1 λhg ,q
For the frequency domain iid channel, Ψ(n) is always a discrete linearly increasing function
of n, and it can serve as a reference against which to measure the contribution of each
subcarrier in arbitrary realistic channels.
As an example, suppose we are trying to determine the appropriate group size for models
B and E from the propagation studies conducted in the definition of IEEE 802.11n (Erceg,
2003). Both models have been measured across a total bandwidth of 20 MHz with a channel
sampling chip period of 10 ns. On one hand Model B is made of 11 paths and it has an
rms-delay spread of 15 ns and very low frequency selectivity. On the other hand Model
E corresponds to a channel with 38 paths (split in 4 clusters) with an rms-delay spread of
100 ns, resulting in large frequency selectivity. While Model B is representative of typical
office indoor environments, Model E corresponds to large outdoor spaces such as airports
or sport halls.
Figure 3 depicts the CSE for channel profiles B, E and the iid model, for different number
of subcarriers (Q = 2, 4, 8 or 16) chosen equispaced across a bandwidth of 20 MHz. It
can be inferred from the top left plot that when only two subcarriers are used per group
(Q = 2), Models B and E behave qualitatively in a similar manner to the iid model and
each of the subcarriers contributes in a significant way towards the achievement of the
maximum diversity. When increasing the number of subcarriers (e.g., Q = 4, 8, 16), this no
longer holds, notice how the CSE values for Model B quickly saturate and get farther apart
from those of the iid channel, indicating that the additional subcarriers do not contribute
substantially in increasing the frequency diversity order. For the case of Model E, a similar
108 Recent Advances in Wireless Communications and Networks

Fig. 3. Cumulative eigenvalue spread for Models B and E from (Erceg, 2003) and iid channel
for different group sizes.
effect can be appreciated but to a much lesser extent, with the departure from the iid model
being more evident for Q = 16 subcarriers. These results seem to indicate that, for Model B,
Q = 2 would provide a good compromise between performance and detection complexity.
In contrast, for channel E, Q = 8 would seem a more appropriate choice to fully exploit
the channel characteristics. Notice that, according to the number of paths of each profile,
Models B and E should attain diversity orders of 11 and 38, respectively. From the results
in Fig. 3 it is obvious that far more moderate group sizes should be chosen in each case to
operate in an optimal fashion from a diversity point of view at a reasonable (ML) detection
complexity.
In conclusion, provided that scenarios with full spatial correlation are avoided, setting the
number of subcarriers per group Q using the proposed CSE-based approach yields

rank Rmin
g = Q NT NR . (51)

g,c : Given an m × n matrix A and an n × p matrix B, it holds that (Meyer,

On the rank of Gmin
2000)
rank(A) + rank(B ) − n ≤ rank(AB ) ≤ min {rank(A), rank(B )} . (52)
Thus, using optimally rotated spreading codes and setting the number of subcarriers per
group Q using the proposed CSE-based approach, provided that pathological scenarios
with full spatial correlation are avoided, we can use (46) and (51) in (52) to show that the
global diversity order for the analysed MIMO strategies is given by
 QN SDM
D̃min = rank Gming,c = R
(53)
Q NT NR STBC/CDD.
Diversity Management in MIMO-OFDM Systems 109

Wireless channel
Fig. 4. Communication architecture for a MIMO-GO-CDM with group-size adaptation.

4. Reconfiguration strategies
It is clear from (44) and (53) that the (instantaneous) rank of the group frequency channel
correlation matrix Rminh g determines the asymptotic diversity of a MIMO-GO-CDM system,
and therefore, it can form the basis for a group size adaptation mechanism. Strictly speaking,
the maximum possible rank of Rmin h g is given by the number of independent paths in the
channel profile. However, as shown in Subsection 3.3, very often the practical rank is far
below this number as maximum diversity is only achieved at unrealistically low error rates.
The adaptive group dimensioning scheme proposed next exploits this rank dependence to
dynamically set the group size as a function of the channel response between all pairs
of transmit and receive antennas. Figure 4 illustrates the architecture of the adaptive
MIMO-GO-CDM system, where it can be appreciated that, in light of the acquired channel
state information (CSI) and system constraints (complexity, QoS), the receiver determines the
most appropriate group size to use and communicates this decision to the transmitter using
a feedback channel. Note, as shown in Fig. 4, that CSI nd SNR information can also be used
to determine the most appropriate modulation and coding scheme in conjunction with the
GO-CDM dimensioning and MIMO mode selection. However this topic is beyond the scope
of this chapter and in this work only fixed modulation and uncoded transmission modes are
considered.
In order to perform the adaptive dimensioning of the GO-CDM component, the receiver
min
requires an estimate R̃hg of the group frequency channel correlation matrix. An accurate
estimate of the full correlation matrix Rmin
h g could be computed by means of time averaging
over the frequency domain, however, in indoor/WLANs scenarios where channels tend to
vary very slowly, this approach would require of many OFDM symbols to get an adequate
estimate. Fortunately, only the group channel correlation matrix is required, thus simplifying
the correlation estimation. Exploiting the grouping structure of GO-CDM-MIMO-OFDM
and assuming the channel frequency response is a wide-sense stationary (WSS) process, it
min
is possible to derive an accurate estimate R̃hg from the instantaneous CSI, provided the
subcarriers in a given group have been chosen equispaced across the available bandwidth.
110
16 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

It is assumed) that the group

* size to be determined is chosen from a finite set of possible
values Q = Q1 , . . . , Qmax whose maximum, Qmax , is limited by the maximum detection
complexity the receiver can support. Suppose that at block symbol k the receiver acquires
knowledge of the channel to form the frequency response h̄ij (k) over all Nc subcarriers.
Now, using the maximum group size available, Qmax , it is possible
to form the frequency
ij ij
responses for all Ngmin = Nc /Qmax groups, h̄1 (k ), . . . , h̄ N min (k) . Taking into account the
g

WSS property it should hold that

    
ij ij ij i  j
E h̄ g,q (k)h̄ g,v (k) = E h̄m,q (k)h̄m,v (k) , (54)

for all pairs of transmit and receive antennas (i, j) and (i , j ) and any q, v ∈ {1, . . . , Qmax }, as
the correlation among any two subcarriers should only depend on their separation, not their
absolute position or the transmit/receive antenna pair. A group channel correlation matrix
estimate from a single frequency response can now be formed averaging across transmit and
receive antennas, and groups,
min
NT NR Ng
min 1 ij ij
R̃hg = ∑∑ ∑
NT NR Ngmin i=1 j=1 g=1
h̄g (k)(h̄g (k)) H . (55)


min
Using basic properties regarding the rank of a matrix, it is easy to prove that rank R̃hg ≤

min Ngmin , Qmax , therefore, Ngmin = Qmax maximises the range of possible group sizes using
min
a single CSI shot. Let us denote the non-increasingly ordered positive eigenvalues of R̃hg
 Q̃
min
by Λ̃hg = λ̃hg ,q where, owing to the deterministic character of R̃hg , they can all be
q =1
assumed to be different and with order one, and consequently, Q̃ represents the true rank of
min
R̃hg . For the purpose of adaptation, and based on the CSE criterion, a more flexible definition
of rank is given as ⎧ ⎫
⎨ ∑nq=1 λ̃hg ,q ⎬
Q̃ = min n : Ψ(n) = ≥ 1− , (56)
⎩ Q̃
∑q=1 λ̃hg ,q ⎭

where n ∈ {1, . . . , Q̃} and  is a small non-negative value used to set a threshold on the
normalised CSE. Notice that Q̃ → Q̃ as  → 0.
Since the group size Q represents the dimensions of an orthonormal spreading matrix C,
restrictions apply on the range of values it can take. For instance, in the case of (rotated)
Walsh-Hadamard matrices, Q is constrained to be a power of two. The mapping of Q̃ to an
allowed group dimension, jointly with the setting of , permits the implementation of different
reconfiguration strategies, e.g.,

Maximise performance : Q = arg min { Q̂ ≥ Q̃ } (57a)

Q̂∈Q

Minimise complexity : Q = arg min {| Q̂ − Q̃ |}. (57b)

Q̂∈Q

It is difficult to assess the feedback involved in this adaptive diversity mechanism as it

depends on the dynamics of the underlying channel. The suggested strategy to implement
Diversity Management
Diversity Management in MIMO-OFDMin MIMO-OFDM Systems
Systems 111
17

6
10

5
10

Expected number of operations (Ω , Ω )

T
g
4
10

3
10

2
10 N =1, group
s
N =1, total
s
N =2, group
s
1
10 N =2, total
s
N =4, group
s
N =4, total
s
0
10
1 2 3 4 5 6 7 8
Group size (Q)

Fig. 5. Complexity as a function of group size (Q) for different number of transmitted
streams.

this procedure is that the receiver regularly estimates the group channel rank and whenever a
variation occurs, it determines and feeds back the new group dimension to the transmitter. In
any case, the feedback information can be deemed insignificant as every update just requires
of log2 Q feedback bits with Q denoting the cardinality of set Q. Differential encoding of Q
would bring this figure further down.

5. Computational complexity considerations

The main advantage of the group size adaptation technique introduced in the previous section
is a reduction of computational complexity without any significant performance degradation.
To gain some further insight, it is useful to consider the complexity of the detection process
taking into account the group size in the GO-CDM component while assuming that an efficient
ML implementation, such as the one introduced in (Fincke & Pohst, 1985), is in use. To this
end, Vikalo & Hassibi (2005) demonstrated that the number of expected (complex) operations
in an efficient ML detector operating at reasonable SNR levels is roughly cubic with the
number of symbols jointly detected. That is, to detect one single group in a MIMO-GO-CDM
system, Ω g = O( NQ 3 ) operations are required.

Obviously, to detect all groups in the system, the expected number of required operations is
given by Ω T = NQc Ω g . Figure 5 depicts the expected per-group and total complexity for a
system using Nc = 64 subcarriers, a set of possible group sizes given by {1, 2, 4, 8} and
different number of transmitted streams. Note that, in the context of this chapter, Ns > 1
necessarily implies the use of SDM. Importantly, increasing the group size from Q = 1 to
Q = 8 implies an increase in the number of expected operations of more than two orders of
magnitude, thus reinforcing the importance of rightly selecting the group size to avoid a huge
waste in computational/power resources. Finally, it should be mentioned that for the STBC
setup, efficient detection strategies exist that decouple the Alamouti decoding and GO-CDM
112
18 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Spatial Division Multiplex Cyclic Delay Diversity Space−Time Block Coding

0 0 0
10 10 10
Q=1
Q=2
Q=4
−2 −2 −2 Q=8
10 10 10
BER

BER

BER
−4 −4 −4
10 10 10

−6 −6 −6
10 10 10

0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
Eb/N0 (dB) Eb/N0 (dB) Eb/N0 (dB)

Fig. 6. Analytical (lines) and simulated (markers) BER for GO-CDM configured to operate in
SDM (left), CDD (centre) and STBC (right) for different group sizes in Channel Profile E.

detection resulting in a simplified receiver architecture that is still optimum (Riera-Palou &
Femenias, 2008).

6. Numerical results
In this section, numerical results are presented with the objective of validating the analytical
derivations introduced in previous sections and also to highlight the benefits of the adaptive
MIMO-GO-CDM architecture. The system considered employs Nc = 64 subcarriers within
a B = 20 MHz bandwidth. These parameters are representative of modern WLAN systems
such as IEEE 802.11n (IEEE, 2009). The GO-CDM technique has been applied by spreading
the symbols forming a group with a rotated Walsh-Hadamard matrix of appropriate size. The
set of considered group sizes is given by Q = {1, 2, 4, 8}. This set covers the whole range
of practical diversity orders for WLAN scenarios while remaining computationally feasible at
reception. Note that a system with Q = 1 effectively disables the GO-CDM component. For
most of the results shown next, Channel Profile E from (Erceg, 2003) has been used. Perfect
channel knowledge is assumed at the receiver. Regarding the MIMO aspects, the system is
configured with two transmit and two receive antennas (NT = NR = 2). As in (van Zelst &
Hammerschmidt, 2002), the correlation coefficient between Tx (Rx) antennas is defined by a
single coefficient ρ Tx (ρ Rx ). Note that in order to make a fair comparison among the different
spatial configurations, different modulation alphabets are used. For SDM, two streams are
transmitted using BPSK whereas for STBC and CDD, a single stream is sent using QPSK
modulation, ensuring that the three configurations achieve the same spectral efficiency.
Figure 6 presents results for SDM, CDD and STBC when transmit and receive correlation
are set to ρ Tx = 0.25 and ρ Rx = 0.75, respectively. The first point to highlight from the
three subfigures is the excellent agreement between simulated and analytical results for the
usually relevant range of BERs (10−3 − 10−7 ). It can also be observed the various degrees of
influence exerted by the GO-CDM component depending on the particular spatial processing
mechanism in use. For example, at a Pb = 10−4 , it can be observed that in SDM and CDD,
the maximum group size considered (Q = 8) brings along SNR reductions greater than 10
dB when compared to the setup without GO-CDM (Q = 1). In contrast, in combination
with STBC, the maximum gain offered by GO-CDM is just above 5 dB. The overall superior
performance of STBC can be explained by the fact that it exploits transmit and receive
Diversity Management
Diversity Management in MIMO-OFDMin MIMO-OFDM Systems
Systems 113
19

0
Spatial division multiplexing 0
Cyclic delay diversity 0
Space−time block coding
10 10 10

Analytical, ρrx=0
Analytical, ρtx=0
−1 −1 −1
10 10 10
Simulation, ρrx=0
Simulation, ρtx=0
BER

BER

BER
−2 −2 −2
10 10 10

−3 −3 −3
10 10 10

−4 −4 −4
10 10 10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
ρ or ρ ρ or ρ ρ or ρ
rx tx rx tx rx tx

Fig. 7. Analytical (lines) and simulated (markers) BER for GO-CDM configured to operate in
SDM (left), CDD (centre) and STBC (right) for different transmit/antenna correlation values.

diversity whereas in SDM there is no transmit diversity and in CDD, this is only exploited
when combined with GO-CDM and/or channel coding.
Next, the effects of antenna correlation at either side of the communication link have been
assessed for each of the MIMO processing schemes. To this end, the MIMO-GO-CDM system
has been configured with Q = 2 and the SNR fixed to Es /N0 = 10 dB. The antenna correlation
at one side was set to 0 when varying the antenna correlation at the other end between 0 and
0.99. As seen in Fig. 7, a good agreement between analytical and numerical results can be
appreciated. The small discrepancy between theory and simulation is mainly due to the use
of the union bound, which always overestimates the true error rate. In any case, the theoretical
expressions are able to predict the performance degradation due to an increased antenna
correlation. Note that, in CDD and SDM, for low to moderate values (0.0 − 0.7), correlation at
either end results in a similar BER degradation, however, for large values (> 0.7), correlation
at the transmitter is significantly more deleterious than at the receiver. For the STBC scenario,
analysis and simulation demonstrate that it does not matter which communication end suffers
from antenna correlation as it leads to exactly the same results. This is because all symbols are
transmitted and received through all antennas (Tx and Rx) and therefore equally affected by
the correlation at both ends.
Finally, the performance of the proposed group adaptive mechanism has been assessed by
simulation. The SNR has been fixed to Es /N0 = 12 dB and a time varying channel profile
has been generated. This profile is composed of epochs of 10,000 OFDM symbols each. Within
an epoch, an independent channel realisation for each OFDM symbol is drawn (quasi-static
block fading) from the same channel profile. For visualisation clarity, the generating channel
profile is kept constant for three consecutive epochs and then it changes to a different one. All
channel profiles (A-F) from IEEE 802.11n (Erceg, 2003) have been considered. Results shown
correspond to an SDM configuration.
The left plot in Fig. 8 shows the BER evolution for fixed and adaptive group size systems as the
environment switches among the different channel profiles. The upper-case letter on the top
of each plot identifies the particular channel profile for a given epoch. Each marker represents
the averaged BER of 10,000 OFDM symbols. Focusing on the fixed group configurations it is
easy to observe that a large group size does not always bring along a reduction in BER. For
example, for Profile A (frequency-flat channel) there is no benefit in pursuing extra frequency
114
20 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

5
10 10
A B F D E C A B F D E C A B F D E C

9 Q=8
−2
10 8

7 4 Q=4

ML detection complexity
10
6
−3

Rank/Q
10 VarQ
BER

5 Q=2

4
3
10
−4 Q=1 3 Q=1
10
Q=2
Q=4 2
Q=8
varQ Q(k)
1
rank
−5 2
10 0 10
0 3 6 9 12 15 18 0 3 6 9 12 15 18 0 3 6 9 12 15 18
4 4
x104 (OFDM symbols) x10 (OFDM symbols) x10 (OFDM symbols)

Fig. 8. Behaviour of fixed and adaptive MIMO GO-CDM-OFDM over varying channel profile
using QPSK modulation at Es /N0 =12 dB. NT = NR = Ns = 2 (SDM mode). Left:
epoch-averaged BER performance. Middle: epoch-averaged rank/group size. Right:
epoch-averaged detection complexity.

diversity at all. Similarly, for Profiles B and C there is no advantage in setting the group
size to values larger than 4. This is in fact the motivation of the proposed MIMO adaptive
group size algorithm denoted in the figure by varQ. It is clear from the middle plot in Fig. 8
that the proposed algorithm is able to adjust the group size taking into account the operating
environment so that when the channel is not very frequency selective low Q values are used
and, in contrast, when large frequency selectivity is sensed the group size dimension grows.
Complementing the BER behaviour, it is important to consider the computational cost of the
configurations under study. To this end the right plot in Fig. 8 shows the expected number
of complex operations (see Section 5). In this plot it can be noticed the huge computational
waste incurred, since there is no BER reduction, in the fixed group size systems with large Q
when operating in channels with a modest amount of frequency-selectivity (A, B and C).

7. Conclusions
This chapter has introduced the combination of GO-CDM and multiple transmit antenna
technology as a means to simultaneously exploit frequency, time and space diversity. In
particular, the three most common MIMO mechanisms, namely, SDM, STBC and CDD, have
been considered. An analytical framework to derive the BER performance of MIMO-GO-CDM
has been presented that is general enough to incorporate transmit and receive antenna
correlations as well as arbitrary channel power delay profiles. Asymptotic results have
highlighted which are the important parameters that influence the practical diversity order
the system can achieve when exploiting the three diversity dimensions. In particular, the
channel correlation matrix and its effective rank, defined as the number of significant positive
eigenvalues, have been shown to be the key elements on which to rely when dimensioning
MIMO-GO-CDM systems. Based on this effective rank, a dynamic group size strategy has
been introduced able to adjust the frequency diversity component (GO-CDM) in light of the
sensed environment. This adaptive MIMO-GO-CDM has been shown to lead to important
power/complexity reductions without compromising performance and it has the potential
to incorporate other QoS requirements (delay, BER objective) that may result in further
energy savings. Simulation results using IEEE 802.11n parameters have served to verify three
Diversity Management
Diversity Management in MIMO-OFDMin MIMO-OFDM Systems
Systems 115
21

facts. Firstly, MIMO-GO-CDM is a versatile architecture to exploit the different degrees of

freedom the environment has to offer. Secondly, the presented analytical framework is able to
accurately model the BER behaviour of the various MIMO-GO-CDM configurations. Lastly,
the adaptive group size strategy is able to recognize the operating environment and adapt the
system appropriately.

8. Acknowledgments
This work has been supported in part by MEC and FEDER under projects MARIMBA
(TEC2005-00997/TCM) and COSMOS (TEC2008-02422), and a Ramón y Cajal fellowship
(co-financed by the European Social Fund), and by Govern de les Illes Balears through project
XISPES (PROGECIB-23A).

9. References
Alamouti, A. (1998). A simple transmit diversity technique for wireless communications, IEEE
JSAC 16: 1451–1458.
Amari, S. & Misra, R. (1997). Closed-form expressions for distribution of sum of exponential
random variables, IEEE Trans. Reliability 46(4): 519–522.
Bauch, G. & Malik, J. (2006). Cyclic delay diversity with bit-interleaved coded modulation
in orthogonal frequency division multiple access, IEEE Trans. Wireless Commun.
8: 2092–2100.
Bury, A., Egle, J. & Lindner, J. (2003). Diversity comparison of spreading transforms for
multicarrier spread spectrum transmission, IEEE Trans. Commun. 51(5): 774–781.
Cai, X., Zhou, S. & Giannakis, G. (2004). Group-orthogonal multicarrier CDMA, IEEE Trans.
Commun. 52(1): 90–99.
Cimini Jr., L. (1985). Analysis and simulation of a digital mobile channel using orthogonal
frequency division multiplexing, IEEE Transactions on Communications 33(7): 665–675.
Craig, J. W. (1991). A new, simple and exact result for calculating the probability of error
for two-dimensional signal constellations, IEEE MILCOM’91 Conf. Rec., Boston, MA,
pp. 25.5.1–25.5.5.
Erceg, V. (2003). Indoor MIMO WLAN Channel Models. doc.: IEEE 802.11-03/871r0, Draft
proposal.
Femenias, G. (2004). BER performance of linear STBC from orthogonal designs over MIMO
correlated Nakagami-m fading channels, IEEE Trans. Veh. Technol. 53(2): 307–317.
Fincke, U. & Pohst, M. (1985). Improved methods for calculating vectors of short length in a
lattice, including a complexity analysis, Math. Comput. 44: 463–471.
Foschini, G. (1996). Layered space-time architecture for wireless communication in a
fading environment when using multi-element antennas, Bell Labs Technical Journal
1(2): 41–59.
Haykin, S. (2001). Communication Systems, 4th edn, Wiley.
IEEE (2009). Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer
(PHY) Specifications Amendment 5: Enhancements for Higher Throughput, IEEE
Std 802.11n-2009 .
Johnson, R. & Wichern, D. (2002). Applied Multivariate Statistical Analysis, fifth edn, Prentice
Hall.
Kaiser, S. (2002). OFDM code-division multiplexing in fading channels, IEEE Trans. Commun.
50: 1266–1273.
116
22 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Meyer, C. (2000). Matrix analysis and applied linear algebra, Society for Industrial and Applied
Mathematics (SIAM).
Petersen, K. B. & Pedersen, M. S. (2008). The matrix cookbook. Version 20081110.
URL: http://www2.imm.dtu.dk/pubdb/p.php?3274
Riera-Palou, F. & Femenias, G. (2008). Improving STBC performance in IEEE 802.11n
using group-orthogonal frequency diversity, Proc. IEEE Wireless Communications and
Networking Conference, Las Vegas (US), pp. 1–6.
Riera-Palou, F. & Femenias, G. (2009). OFDM with adaptive frequency diversity, IEEE Signal
Processing Letters 16(10): 837 – 840.
Riera-Palou, F., Femenias, G. & Ramis, J. (2008). On the design of uplink and
downlink group-orthogonal multicarrier wireless systems, IEEE Trans. Commun.
56(10): 1656–1665.
Simon, M. & Alouini, M. (2005). Digital communication over fading channels, Wiley-IEEE Press.
Simon, M., Hinedi, S. & Lindsey, W. (1995). Digital communication techniques: signal design and
detection, Prentice Hall PTR.
Stuber, G., Barry, J., Mclaughlin, S., Li, Y., Ingram, M. & Pratt, T. (2004). Broadband
MIMO-OFDM wireless communications, Proceedings of the IEEE 92(2): 271–294.
Tarokh, V., Jafarkhani, H. & Calderbank, A. (1999). Space-time block codes from orthogonal
designs, IEEE Transactions on Information Theory 45(5): 1456–1467.
Telatar, E. (1999). Capacity of Multi-antenna Gaussian Channels, European Transactions on
Telecommunications 10(6): 585–595.
van Zelst, A. & Hammerschmidt, J. (2002). A single coefficient spatial correlation model
for multiple-input multiple-output (mimo) radio channels, Proc. Proc. URSI XXVIIth
General Assembly, Maastricht (the Netherlands), pp. 1–4.
Vikalo, H. & Hassibi, B. (2005). On the sphere-decoding algorithm ii. generalizations,
second-order statistics, and applications to communications, Signal Processing, IEEE
Transactions on 53(8): 2819 – 2834.
Weinstein, S. & Ebert, P. (1971). Data transmission by frequency-division multiplexing using
the discrete Fourier transform, IEEE Trans. Commun. Tech. 19: 628–634.
Wittneben, S. (1993). A new bandwidth efficient transmit antenna modulation diversity
scheme for linear digital modulation, Proc. IEEE Int. Conf. on Commun., Geneva
(Switzerland), pp. 1630–1634.
Yee, N., Linnartz, J.-P. & Fettweis, G. (1993). Multi-carrier CDMA in indoor wireless radio
networks, Proc. IEEE Int. Symp. on Pers., Indoor and Mob. Rad. Comm., Yokohama
(Japan), pp. 109–113.
 0
6

Optimal Resource Allocation in OFDMA

Broadcast Channels Using Dynamic
Programming
Jesús Pérez, Javier Vía and Alfredo Nazábal
University of Cantabria
Spain

1. Introduction
OFDM (Orthogonal Frequency Division Multiplexing) is a well-known multicarrier
modulation technique that allows high-rate data transmissions over multipath broadband
wireless channels. By using OFDM, a high-rate data stream is split into a number of lower-rate
streams that are simultaneously transmitted on different orthogonal subcarriers. Thus, the
broadband channel is decomposed into a set of parallel frequency-flat subchannels; each one
corresponding to an OFDM subcarrier. In a single user scenario, if the channel state is known
at the transmitter, the system performance can be enhanced by adapting the power and data
rates over each subcarrier. For example, the transmitter can allocate more transmit power and
higher data rates to the subcarriers with better channels. By doing this, the total throughput
can be significantly increased.
In a multiuser scenario, different subcarriers can be allocated to different users,
which constitutes an orthogonal multiple access method known as OFDMA
(Orthogonal Frequency Division Multiple Access). OFDMA is one of the principal
multiple access schemes for broadband wireless multiuser systems. It has being
proposed for use in several broadband multiuser wireless standards like IEEE
802.20 (MBWA: http://grouper.ieee.org/groups/802/20/), IEEE 802.16 (WiMAX:
http://www.ieee802.org/16/, 2011) and 3GPP-LTE (http://www.3gpp.org/). This chapter
focuses on the OFDMA broadcast channel (also known as downlink channel), since this is
typically where high data rates and reliability is needed in broadband wireless multiuser
systems. In OFDMA downlink transmission, each subchannel is assigned to one user at
most, allowing simultaneous orthogonal transmission to several users. Once a subchannel
is assigned to a user, the transmitter allocates a fraction of the total available power as well
as a modulation and coding (data rate). If the channel state is known at the transmitter,
the system performance can be significantly enhanced by allocating the available resources
(subchannels, transmit power and data rates) intelligently according to the users’ channels.
The allocation of these resources determines the quality of service (QoS) provided by the
system to each user. Since different users experience different channels, this scheme does not
only exploit the frequency diversity of the channel, but also the inherent multiuser diversity
of the system.
In multiuser transmission schemes, like OFDMA, the information-theoretic system
performance is usually characterized by the capacity region. It is defined as the set of rates
118
2 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

that can be simultaneously achieved for all users (Cover & Thomas, 1991). OFDMA is a
suboptimal scheme in terms of capacity, but near capacity performance can be achieved when
the system resources are optimally allocated. This fact, in addition to its orthogonality and
feasibility, makes OFDMA one of the preferred schemes for practical systems. It is well known
that coding across the subcarriers does not improve the capacity (Tse & Viswanath, 2005),
so maximum performance is achieved by using separate codes for each subchannel. Then,
the data rate received by each user is the sum of the data rates received from the assigned
subchannels. The set of data rates received by all users for a given resource allocation gives
rise to a point in the rate region. The points of the segment connecting two points associated
with two different resource allocation strategies can always be achieved by time sharing
between them. Therefore, the OFDMA rate region is the convex hull of the points achieved
under all possible resource allocation strategies.
To numerically characterize the boundary of the rate region, a weight coefficient is assigned
to each user. Then, since the rate region is convex, the boundary points are obtained by
maximizing the weighted sum-rate for different weight values. In general, this leads to
non-linear mixed constrained optimization problems quite difficult to solve. The constraint is
given by the total available power, so it is always a continuous constraint. The optimization
or decision variables are the user and the rate assigned to each subcarrier. The first is a
discrete variable in the sense that it takes values from a finite set. At this point is important to
distinguish between continuous or discrete rate adaptation. In the first case the optimization
variable is assumed continuous whereas in the second case it is discrete and takes values from
a finite set. The later is the case of practical systems where there is always a finite codebook,
so only discrete rates can be transmitted through each subchannel. Unfortunately, regardless
the nature of the decision variables, the resulting optimization problems are quite difficult to
solve for realistic numbers of users and subcarriers.
This chapter analyzes the maximum performance attainable in broadcast OFDMA channels
from the information-theoretic point of view. To do that, we use a novel approach to
the resource allocation problems in OFDMA systems by viewing them as optimal control
problems. In this framework the control variables are the resources to be assigned to each
OFDM subchannel (power, rate and user). Once they are posed as optimal control problems,
dynamic programming (DP) (Bertsekas, 2005) is used to obtain the optimal resource allocation.
The application of DP leads to iterative algorithms for the computation of the optimal resource
allocation. Both continuous and discrete rate allocation problems are addressed and several
numerical examples are presented showing the maximum achievable performance of OFDMA
in broadcast channels as function of different channel and system parameters.

1.1 Review of related works

Resource allocation in OFDMA systems has been an active area of research during the last
years and a wide variety of techniques and algorithms have been proposed. The capacity
region of general broadband channels was characterized in (Goldsmith & Effros, 2001), where
the authors also derived the optimal power allocation achieving the boundary points of the
capacity region. In this seminal work, the channel is decomposed into a set of N parallel
independent narrowband subchannels. Each parallel subchannel is assigned to various
users, to a single user, or even not assigned to any user. In the first case, the transmitter
uses superposition coding (SC) and the corresponding receivers use successive interference
cancelation (SIC). If a subchannel is assigned to a single user, an AWGN capacity-achieving
code is used. Moreover, a fraction of the total available power is assigned to each user in
Optimal Resource
Optimal Resource Allocation
Allocation in OFDMA inChannels
Broadcast OFDMA Broadcast
Using Dynamic Channels Using Dynamic Programming
Programming 1193

each subchannel. Then, taking the limit as N goes to infinite (continuous frequency variable),
the problem can be solved using multilevel water-filling. Similarly, in (Hoo et al., 2004) the
authors characterize the asymptotic (when N goes to infinite) FDMA multiuser capacity
region and propose optimal and suboptimal resource allocation algorithms to achieve the
points in such region. Here, unlike (Goldsmith & Effros, 2001), each subchannel is assigned to
one user at most and a separate AWGN capacity-achieving code is used in each subchannel.
In OFDMA systems the number of subchannels is finite. Each subchannel is assigned to one
user at most, and a power value is allocated to each subcarrier. OFDMA is a suboptimal
scheme in terms of capacity but, due to its orthogonality and feasibility, it is an adequate
multiple access scheme for practical systems. Moreover, OFDMA can achieve near capacity
performance when the system resources are optimally allocated.
In (Seong et al., 2006) and (Wong & Evans, 2008) efficient resource allocation algorithms are
derived to characterize the capacity region of OFDMA downlink channels. The proposed
algorithms are based on the dual decomposition method (Yu & Lui, 2006). In (Wong & Evans,
2008) the resource allocation problem is considered for both continuous and discrete rates,
as well as for the case of partial channel knowledge at the transmitter. By using the dual
decomposition method, the algorithms are asymptotically optimal when the number of
subcarriers goes to infinite and is close to optimal for practical numbers of OFDM subcarriers.
Some specific points in the rate region are particularly interesting. For example the
maximum sum-rate point where the sum of the users’ rates is maximum, or the maximum
symmetric-rates point where all users have maximum identical rate. Many times, in practical
systems one is interested in the maximum achievable performance subject to various QoS
(Quality of Service) users’ requirements. For example, what is the maximum sum rate
maintaining given proportional rates among users, or what is the maximum sum-rate
guarantying minimum rate values to a subset of users. All these are specific points in the
capacity region that can be achieved with specific resource allocation among the users. A
crucial problem here is to determine the optimal resource allocation to achieve such points.
Mathematically, these problems are also formulated as optimization problems constrained
by the available system resources. In (Jang & Lee, 2003) the authors show the resource
allocation strategy to maximize the sum rate of multiuser transmission in broadcast OFDM
channels. They show that the maximum sum-rate is achieved when each subcarrier is
assigned to the user with the best channel gain for that subcarrier. Then, the transmit power
is distributed over the subcarriers by the water-filling policy. In asymmetric channels, the
maximum sum-rate point is usually unfair because the resource allocation strategy favors
users with good channel, producing quite different users’ rates. Looking for fairness among
users, (Ree & Cioffi, 2000) derive a resource allocation scheme to maximize the minimum
of the users’ rates. In (Shen et al., 2005) the objective is to maximize the rates maintaining
proportional rates among users. In (Song & Li, 2005) an optimization framework based on
utility-function is proposed to trade off fairness and efficiency. In (Tao et al., 2008), the authors
maximize the sum rate guarantying fixed rates for a subset of users.

2. Channel and system model

Fig. 1 shows a block diagram of a single-user OFDM system with N subcarriers employing
power and rate adaptation. It comprises three main elements: the transmitter, the receiver
and the resource allocator. The channel is assumed to remain fixed during a block of OFDM
symbols. At the beginning of each block the receiver estimates the channel state and sends
this information (CSI: Channel state information) to the resource allocator, usually via a
120
4 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Fig. 1. Single-user OFDM system with power and rate adaptation.

feedback channel. The resource allocator can be physically embedded with the transmitter
or the receiver. From the CSI, the resource allocation algorithm computes the data rate and
transmit power to be transmitted through each subcarrier. Let vectors r = [r1 r2 · · · r N ] T
and p = [ p1 p2 · · · p N ] T denote the data rates and transmit powers allocated to the OFDM
subchannels, respectively. This information is sent to the transmit encoder/modulator
block, which encodes the input data according to r and p, and produces the streams of
encoded symbols to be transmitted through the different subchannels. It is well known
that coding across the subcarriers does not improve the capacity (Tse & Viswanath, 2005)
so, from a information-theoretic point of view, the maximum performance is achieved by
using independent coding strategies for each OFDM subchannel. To generate an OFDM
symbol, the transmitter picks one symbol from each subcarrier stream to form the symbols
vector X = [ X [1], X [2], . . . , X [ N ]] T . Then, it performs an inverse fast Fourier transform (IFFT)
operation on X yielding the vector x. Finally the OFDM symbol x’ is obtained by appending
a cyclic prefix (CP) of length L cp to x. The receiver sees a vector of symbols y’ that comprises
the OFDM symbol convolved with the base-band equivalent discrete channel response h of
length L, plus noise samples

y’ = h ∗ x’ + n. (1)
It is assumed that the noise samples at the receiver (n) are realizations of a ZMCSCG
(zero-mean circularly-symmetric complex Gaussian) random variables with variance σ2 : n ∼
CN (0, σ2 I). The receiver strips off the CP and performs a fast Fourier transform (FFT) on the
sequence y to yield Y. If L cp ≥ L, it can be shown that

Yk = Hk Xk + Nk , k = 1, . . . , N, (2)
where H = [ H1 , H2 , . . . HN is the FFT of h, i.e. the channel frequency response for each
]T
OFDM subcarrier, and the Nk ’s are samples of independent ZMCSCG variables with variance
σ2 . Therefore, OFDM decomposes the broadband channel into N parallel subchannels with
channel responses given by H = [ H1 , H2 , . . . HN ] T . In general the Hk ’s at different subcarriers
are different.
Note that the energy of the symbol Xk is determined by the k-th entry of the power allocation
vector pk . It is assumed that the transmitter has a total available transmit power PT to be
distributed among the subcarriers, so ∑ kN=1 pk ≤ PT . The coding/modulation employed for
the k-th subchannel is determined by the corresponding entry (rk ) of the rate allocation vector
r.
Optimal Resource
Optimal Resource Allocation
Allocation in OFDMA inChannels
Broadcast OFDMA Broadcast
Using Dynamic Channels Using Dynamic Programming
Programming 1215

Fig. 2. Multi-user OFDM system with adaptive resources allocation.

Fig. 3. M-users broadcast broadband channel.

Fig. 2 shows a block diagram of a downlink OFDMA system. It comprises the transmitter,
the resource allocator unit and M users’ receivers (Fig. 2 only shows the m-th receiver).
The resource allocator is physically embedded with the transmitter. It is assumed that the
transmitter sends independent information to each user. The base-band equivalent discrete
channel response of the m-th user is denoted by hm = [ hm,1 hm,2 · · · hm,Lm ] T , where now L m is
the number of channel taps and nm ∼ CN (0, σm 2 I) are the noise samples at the m-th receiver.

Noise and channels at different receivers are assumed to be independent. A scheme of a

M-user OFDM broadcast channel is depicted in Fig. 3.
Let Hm = [ Hm,1 Hm,2 · · · Hm,N ] T denote the complex-valued frequency-domain channel
response of the OFDM channel, as seen by the m-th user, for the N subchannels. As it was
mentioned, Hm is the N-points discrete-time Fourier transform (DFT) of hm .
It is assumed that the multi-user channel remains constant during the transmission of a
block of OFDM symbols. At the beginning of each block each receiver estimates its channel
response for each subcarrier, and informs the resource allocator by means of a feedback
channel. Then, it computes the resource allocation vectors r, p and u = [ u1 u2 . . . u N ] T , where
u k denotes the user assigned to the k-th subcarrier. Each subcarrier is assigned to a single
user, so it is assumed that subcarriers are not shared by different users. Note that, since
u k ∈ Su = {1, 2, . . . M }, there are M N possible values of u, so M N different ways to assign
the subcarriers to the users. Once these vectors have been computed, the resource allocator
122
6 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

informs the transmitter and receivers through control channels. Then, the transmitter encode
the input data according to the resource allocation vectors and stores the stream of encoded
symbols to be transmitted through the OFDM subchannels. The OFDM symbols are created
and transmitted as in the single-user case. Each user receives and decodes its data from the
assigned subchannels (given by u).
Let γ be a M × N matrix whose entries are the channel power gains for the different users and
subcarriers normalized to the corresponding noise variance

| Hm,k |2
γm,k = . (3)
σm
2

Assuming a continuous codebook available at the transmitter, rk can take any value subject
to the available power and the channel condition. The maximum attainable rate through the
k-th subchannel is given by

rk = log2 (1 + pk γuk ,k ) bits/OFDM symbol, (4)

where pk is the power assigned to the k-th subchannel. The minimum needed power to
support a given data rate rk through the k-subcarrier will be

2r k − 1
pk = . (5)
γuk ,k

We assume that the system always uses the minimum needed power to support a given rate
so, for a fixed subcarriers-to-users allocation u, the rk ’s and the pk ’s are interchangeable in the
sense that a given rate determines the needed transmit power and viceversa.
In practical systems there is always a finite codebook, so the data rate at each subchannel is
constrained to take values from a discrete set rk ∈ Sr = {r (1) , r (2) , . . . , r ( Nr ) } where each value
corresponds to a specific modulation and code from the available codebook. The transmit
rates and powers are related by

rk = log2 (1 + β(rk ) pk γuk ,k ) bits/OFDM symbol, (6)

where the so-called SNR-gap approximation is adopted Cioffi et al. (1995), being 0 < β(r ) ≤ 1
the SNR gap for the corresponding code (with rate r). For a given code β(r ) depends on
a pre-fixed targeted maximum bit-error rate. Then, the SNR-gap can be interpreted as the
penalty in terms of SNR due to the use of a realistic modulation/coding scheme. There will
be a SNR gap β(r ( i) ), i = 1, . . . Nr associated with each code of the codebook for a given
targeted bit-error rate. The minimum needed power to support rk will be

2r k − 1
pk = . (7)
β(rk )γuk ,k

Since there is a finite number of available data rates, there will be a finite number of possible
rate allocation vectors r. Note that there are ( Nr ) N possible values of r, but, in general, for a
given u only some of them will fulfil the power constraint.
Optimal Resource
Optimal Resource Allocation
Allocation in OFDMA inChannels
Broadcast OFDMA Broadcast
Using Dynamic Channels Using Dynamic Programming
Programming 1237

3. The rate region of OFDMA

For a given subcarriers-to-users and rates-to-subcarriers allocation vectors u and r, the total
rate received by the m-th user will be given by
N
Rm (r, u) = ∑ δm,u rk k
bits/OFDM symbol, (8)
k =1
where δi,j is the Kronecker delta. The users’ rates are grouped in the corresponding rate vector

R(r, u) = [ R1 (r, u), R2 (r, u), · · · , R M (r, u)] T , (9)

which is the point in the rate region associated with the resource allocation vectors r and u.
Let R0 denote the points achieved for all possible combinations of u and r

R0 = R(r, u), (10)
r∈Sr ,u∈Su
where Sr and Su are the set of all possible rates-to-subcarriers and subcarriers-to-users
allocation vectors, respectively. Therefore, R0 comprises the rate vectors associated with
single resource allocation strategies given by u and r. Later, it will be shown that, in general,
R0 is not a convex region. Let (r1 , u1 ) and (r2 , u2 ) be two possible resource allocations that
achieves the points R1 = R(r1 , u1 ) and R2 = R(r2 , u2 ) in R0 . By time-sharing between the two
resource allocation strategies, all points in the segment R1 -R2 can be achieved. Therefore, the
rate region of OFDMA will be the convex hull of R0 : R = H (R0 ). Note that the achievement
of any point of R not included in R0 requires time-sharing among different resource allocation
schemes.
The next two subsections analyze the OFDMA rate region for the cases of continuous and
discrete rates. Mathematical optimization problems for the computation of the rate region are
posed, and their solution by means of the DP algorithm is presented.

3.1 Continuous rates

Let us first consider the achievable rate region R0 (u) for a fixed subcarriers-to-users allocation
vector u. It will be the union of the points achieved for all possible rates-to-subcarriers
allocation vectors r

R 0 (u ) = R(r, u). (11)
r∈Sr
It can be shown that R0 (u) is a convex region (Cover & Thomas, 1991), being its boundary
points the solution of the following convex optimization problems

maximize λ T R(u) = ΣkN=1 λuk rk

r
subject to ΣkN=1 (2r k − 1)/γuk ,k ≤ PT (12)
r k ∈ Sr , k = 1, . . . , N,
for different values of vector λ = [ λ1 λ2 · · · λ M ]T ,
where λm ≥ 0. λ can be geometrically
interpreted as the orthogonal vector to the hyperplane tangent to the achievable rate region
at a point in the boundary. The components of λ are usually denoted as users’ priorities.
Note the constraint regarding the total available power. This is a well-known convex problem
(Boyd & Vandenberghe, 2004) whose solution can be expressed in closed-form as follows
124
8 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Fig. 4. Rate regions of all vectors u. The total transmit power is PT = 1.

 λ  
u k γ u k ,k
log2 if μ ≤ λuk γuk ,k
rk∗ = μ (13)
0 if μ ≥ λuk γuk ,k
where μ is a Lagrangian parameter which can be implicitly obtained from
N  +
λuk 1
∑ μ
−
γuk ,k
= PT , (14)
k =1
where ( a)+ = max{ a, 0}.
Fig. 4 shows the rate regions R(u) for all possible subcarriers-to-users allocation vectors (u) in
a toy example with M = 2 users, N = 2 subcarriers and PT = 1. (Although it is not a realistic
channel, it is used here to illustrate the resource allocation problem in OFDMA channels).
Here and in the following results, the rates are given in bits/OFDM symbol.
When the system allocates all subcarriers to a single user (u k = u, ∀k), the broadcast channel
turns into a single-user channel and the solution of (12) does not depend on λ. Therefore, in
these cases the rate region degenerates in a single point on the corresponding axis. The rate at
this point is the capacity of the corresponding single-user OFDM channel. Once the optimal
rate vector r∗ is obtained, the power to be allocated to each subcarrier is given by (5).
The achievable points for all possible values of u and r will be

R0 = R 0 ( u) . (15)
u∈Su
In general, R0 is not convex. This fact can be observed in the example of Fig. 4. The rate
region of the OFDMA broadcast channel is the convex hull of R0 : R = H (R0 ). The rate
region for the example of Fig. 4 is depicted in Fig. 5 as the convex hull of the region achieved
by all vectors u. In general, there are subcarriers-to-users allocation vectors u that are never
optimal. This is the case of u = [1, 2] T in the example.
As it was mentioned, there are M N possible subcarriers-to-users allocation vectors u.
Therefore, an exhaustive search among all the possible vectors requires the computation of
Optimal Resource
Optimal Resource Allocation
Allocation in OFDMA inChannels
Broadcast OFDMA Broadcast
Using Dynamic Channels Using Dynamic Programming
Programming 1259

Fig. 5. OFDMA rate region. It is the convex hull of the rate regions achieved for all vectors u
(see Fig. 4).

M N waterfilling solutions for each vector λ, which is not feasible for practical values of N and
M. An alternative is to jointly optimize over u and p simultaneously, so the problem becomes

maximize λ T R(r, u) = ΣkN=1 λuk rk

r,u

subject to ΣkN=1 (2r k − 1)/γuk ,k ≤ PT (16)

u k ∈ Su , k = 1, . . . , N
r k ∈ Sr , k = 1, . . . , N
This is a mixed non-linear constrained optimization problem. In general these kind of
problems are difficult to solve. However, it has the structure of a DP problem with the
following elements (see appendix: Dynamic Programming):
• The process stages are the subchannels, so the number of stages is N,
• Control vector: ck = [ u k , rk ] T ,
• State variable: xk = Σki=−11 (2r i − 1)/γui ,i ,
• Initial state x1 = 0,
• Subsets of possible states: 0 ≤ xk ≤ PT ,
• Subsets of admissible controls:

Ck ( xk ) = [ u k , rk ] T | u k ∈ Su , rk ∈ Sr , rk ≤ log2 (1 + ( PT − xk )γuk ,k ) ,

• System equation: xk+1 = f k ( xk , ck ) = xk + (2r k − 1)/γuk ,k ,

• Cost functions: gk (ck ) = λuk rk .
126
10 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Fig. 6. Rate regions for the two-users channel of Fig. 7 considering different values of average
transmit power per subchannel (P).

The entries of the control vector ck are the user and the rate allocated to the k-th subchannel,
that take values from the sets Su and Sr , respectively. The state variable xk is the accumulated
power transmitted in the previous subchannels. Therefore 0 ≤ xk ≤ PT , and the initial state
is x1 = 0. The control component rk is constrained by the available power at the k-th stage:
PT − xk . Note that the solutions of (16) for different λ’s are the points of R0 located in the
boundary of the rate region, and the convex hull of these points are the boundary of the rate
region.
By using the DP algorithm, rate regions for a more realistic two user channel have been
computed. They are depicted in Fig. 6. In this example the number of subcarriers is N = 128
and the users’ subchannel responses are shown in Fig. 7. They are normalized so the average
gain (averaging over the subchannels and users) equals 1. These channel realizations have
been obtained from a broadband Rayleigh channel model with L = 16 taps and an exponential
power delay profile with decay factor ρ = 0.4. This channel model will be described in section
5. The noise variances are assumed to be σm 2 = 1, identical for all users. Fig. 6 shows

the rate region for two different values of average power per subchannel: P = PT /N = 1
and P = PT /N = 10. Note that if the OFDM subchannels were identical (frequency-flat
broadband channel), the average SNR at the OFDM subchannels would be 0dB and 10dB,
respectively.
To obtain the rate regions of Fig.6, (16) has to be solved for each λ by using the DP algorithm.
In each case, the solution is a pair of optimal resource allocation vectors u∗ and r∗ . Then, the
power to be transmitted through the subchannels p∗ is given by (5). The corresponding users’
rate vector are obtained from (8) and shown in Fig. 6 as a marker point in the boundary of the
rate region. Therefore, the marker points are the points of R0 located in the boundary of the
rate region and associated with pairs of resource allocation vectors (u, r) which are solutions
of (16) for different users’ priority vectors. For example, for λ = [0.4, 0.6] T and PT = 10 the
optimal resource allocation vectors are shown in Fig. 8, as well as the transmit power through
the OFDM subchannels (p∗ ). The resulting users’ rates are R(u∗ , p∗ ) = [82.7, 350.7] T .
Note that different vectors λ can lead to identical solution of (16), and hence to identical
points/markers in the boundary of the rate region. The convex hull of the marker points
Optimal Resource
Optimal Resource Allocation
Allocation in OFDMA inChannels
Broadcast OFDMA Broadcast
Using Dynamic Channels Using Dynamic Programming
Programming 127
11

Fig. 7. Normalized subchannel gains at the OFDM subchannels.

Fig. 8. Optimal resource allocation vectors u and r for λ = [0.4, 0.6] T and PT = 10N. The
figure also shows the resulting power allocation vector p.

constitutes the boundary of the rate region. Any point in the segments between two markers
is achieved by time sharing between the corresponding optimal resource allocations.
The application of the DP algorithm requires the control and state spaces to be discrete.
Therefore, if they are continuous, they must be discretized by replacing the continuous spaces
by discrete ones. Once the discretization is done, the DP algorithm is executed to yield
the optimal control sequence for the discrete approximating problem. Hence, it becomes
necessary to study the effect of discretization on the optimality of the solution. In the problem
(16), the state variable xk and the second component of the control vetctor (rk ) are continuous.
To obtain the rate regions of Fig. 6, Sr was uniformly discretized considering Nd = 2000
possible rate values between 0 and a maximum rate which is achieved when the total power
PT is assigned to the best subchannel of all users. The state variable xk was discretized in Nd
possible values accordingly. To study the effect of discretization of xk and rk the rate regions
for different values of Nd are shown in Fig. 9. The channels and simulation parameters were
as in Fig. 6. It shows that the required Nd is less than 500.
128
12 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Fig. 9. Rate regions for different number of rate discretization values Nd

Fig. 10. Example of OFDMA rate region with discrete codebook Sr .

3.2 Discrete rates

Now, Sr is a finite set and therefore the set of achievable points R0 is finite. It comprises all
points resulting from the combinations of r and u that fulfill the power constraint. Therefore,
the cardinality of R0 depends on γ, Sr and PT .
As an example, let us consider again the channel example of Fig. 4 and 5 with PT = 1
and a codebook with the following available rates Sr = {0, 1/4, 1/2, 2/3, 3/4, 1}. Note that
by including zero rate in Sr we consider the possibility of no transmission trough some
subchannels. All achievable rate vectors (R0 ), and their convex hull, are shown in Fig. 10,
where β(r ) = 1, ∀r ∈ Sr is assumed.
The set R0 can be viewed as the union of the points achieved by different vectors u ∈ Su :
R0 (u). For example, in Fig. 11 the points achieved by u = [1, 2] T are highlighted, as well
as their convex hull. It can be observed that, for this particular value of u the attainable rate
vectors are always under the convex hull of R0 . Therefore, u = [1, 2] is not the optimal
subcarriers-to-users allocation vector in any case.
Optimal Resource
Optimal Resource Allocation
Allocation in OFDMA inChannels
Broadcast OFDMA Broadcast
Using Dynamic Channels Using Dynamic Programming
Programming 129
13

Fig. 11. Achievable points for u = [1, 2], and their convex hull.

In general the vertex points in the boundary of R0 (u) will be the solutions of (12), but now Sr is
a finite set. Therefore, both the state and control variables are discrete so (12) is a fully integer
optimization problem. Unlike the continuous rates case, there is not closed-form solution
when the allowed rates belong to a discrete set. The problem can be formulated as a DP
problem where
• The process stages are the subchannels, so the number of stages is N,
• Control variable: ck = rk ,
• State variable: xk = Σki=−11 (2r i − 1)/γui ,i , 0 ≤ xk ≤ PT , x1 = 0, x N +1 = PT
• Initial state x1 = 0
• Subsets of possible states: 0 ≤ xk ≤ PT
 
• Subsets of admissible controls: Ck ( xk ) = rk |rk ∈ Sr , rk ≤ log2 (1 + ( PT − xk )γuk ,k ) ,
• System equation: xk+1 = f k ( xk , ck ) = xk + (2r k − 1)/γuk ,k ,
• Cost functions: gk (ck ) = λuk rk .
The control variables ck are the rates allocated to the subchannels, that take values from the set
Sr . The state variable xk is the accumulated power transmitted in the previous subchannels
(up to k − 1-th subchannel). Therefore 0 ≤ xk ≤ PT , and the initial state is x1 = 0. The control
component rk is constrained by the available power at the k-th stage: PT − xk .
Since there are M N possible values of u, an exhaustive search among all possible vectors u
requires to solve the above DP problem M N times for each vector λ, which is not feasible
for practical values of M and N. In this case one has to jointly optimize over u and
r simultaneously, as in (16). Now, unlike the continuous rates case, (16) is an integer
programming problem because the control variable ck is fully discrete taking values from a
finite set Su × Sr .
Fig. 12 shows the rate regions for the two user channel of Fig. 7 considering continuous
and discrete rate allocation. As it is expected, continuous rate adaptation always outperforms
the discrete rate case. In all cases the rate region has been obtained from the DP algorithm.
The figure depicts the rate region for two different values of average power per subcarrier:
P = 1 and P = 10. It is also assumed that the noise at the OFDM subchannels are
130
14 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Fig. 12. Rate regions for the two user channel of Fig. 7 considering continuous and discrete
rate allocation. P denotes the average transmit power per subchannel

i.i.d. with variance σm 2 = 1, identical for all users. Then, if the subchannels were
identical and frequency-flat, the average SNR at the OFDM subchannels would be 0dB
and 10dB, respectively. Now, the following set of available rates have been considered:
Sr = {0, 1/2, 3/4, 1, 3/2, 2, 3, 4, 9/2}. These are the data rates of a set of rate-compatible
punctured convolutional (RCPC) codes, combined with M-QAM modulations, that are
used in the 802.11a (http://grouper.ieee.org/groups/802/11/). The advantage of RCPC
codes is to have a single encoder and decoder whose error correction capabilities can
be modified by not transmitting certain coded bits (puncturing). Therefore, the same
encoder and decoder are used for all codes of the RCPC codebook. This makes the RCPC
codes, combined with adaptive modulation, a feasible rate adaptation scheme in wireless
communications. Apart from the 802.11, punctured codes are used in other standards like
WIMAX (http://www.ieee802.org/16/, 2011).
To obtain the rate regions of Fig. 12, the maximization problem (16) has been solved, using
the DP algorithm, for each λ. In each case, the solution is a pair of optimal resource allocation
vectors u∗ and r∗ . Now, the corresponding users’ rate vector is obtained from (8) and shown
in 12 as a marker point in the boundary of the rate region. For example, for λ = [0.4, 0.6] T
and PT = 10 the optimal resource allocation vectors are shown in Fig. 13, as well as the
transmit power assigned to each OFDM subchannel (p). These vectors produces the users’
rate vector R(u∗ , r∗ ) = [89.5, 340.0] T . Note that this is quite similar to the users’ rate vector
attained for this λ in the continuous rate case. Comparing 8 and 13 one can observe that
the resource allocation vectors are similar in the cases of continuous and discrete rates. In
fact, for this particular case, the subcarriers-to-users allocation vectors u are identical and the
rates-to-subcarriers allocation vectors r are quite similar. Similar behavior is observed for any
other vector λ.

4. Maximum sum-rate
In the case of continuous rate adaptation, the maximum sum rate will be the solution of (16)
for λ = 1 M . But, it was shown in (Jang & Lee, 2003) that the sum-rate is maximized when
each subcarrier is assigned to the user with the best channel
Optimal Resource
Optimal Resource Allocation
Allocation in OFDMA inChannels
Broadcast OFDMA Broadcast
Using Dynamic Channels Using Dynamic Programming
Programming 131
15

Fig. 13. Optimal resource allocation vectors u and r for λ = [0.4, 0.6] T and PT = 10N. The
figure also shows the power transmitted through the OFDM subchannels p

Fig. 14. Optimal resource allocation vectors u and r to achieve the maximum sum-rate. The
total transmit power is PT = 10N. The figure also shows the power allocation vector p

u ∗k = arg max{γm,k }, k = 1, . . . , N, (17)

m
Then, the optimal rates can be calculated from (13) and (14) with λ = 1 M , and the power
allocated to each subcarrier is given by (5).
The maximum sum-rate point is always in the boundary of the rate region. In the channel
of Figs. 4 and 5, the maximum sum rate is 1.43, which is achieved by u∗ = [2, 2] T and r∗ =
[0.53, 0.90] T . The power allocation is p∗ = [0.40, 0.60] T . In this particular case the maximum
sum-rate is achieved by allocating all the system resources to the user 2, so the rate for user 1
is zero. In the two-users channel of Fig. 7, the maximum sum rate, for P = 10, is 445.64. This
is achieved by the resource allocation vectors depicted in Fig. 14.
In the case of discrete rate adaptation, the maximum sum rate is also achieved by allocating
each subcarrier to the user with the best channel on it. But now, the rates allocated to the
subchannels will be the solution of (12) with u given by (17). So, the optimal rate allocation
132
16 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Fig. 15. Optimal resource allocation vectors u and r to achieve the maximum sum-rate. The
total transmit power is PT = 10N. The figure also shows the power allocation vector p

can also be obtained from the DP algorithm. In the simple case of Fig. 10, the maximum
sum rate is 1.417, which is achieved by u∗ = [2, 2] T , r∗ = [2/3, 3/4] T and p∗ = [0.52, 0.47] T .
Like in continuous rate adaptation, the maximum sum-rate point is always in the boundary
of the rate region. In the two-user channel of Fig. 7, the maximum sum rate is 441.0 assuming
that the average transmit power per subcarrier is P = 10 and the set of available rates is
Sr = {0, 1/2, 3/4, 1, 3/2, 2, 3, 4, 9/2}. It is achieved by the resource allocation vectors depicted
in Fig. 15. Again, the maximum sum-rate and the corresponding resource allocation vectors
are quite similar to the continuous rate case.

5. Outage rate region

The previous results show the achievable performance (rate vectors) for specific channel
realizations. However, due to the intrinsic randomness of the wireless channel, the channel
realizations can be quite different. To study the performance for all channel conditions we
resort to the outage rate region concept Lee & Goldsmith (2001). The outage rate region for a
given outage probability Pout consists of all rate vectors R = [ R1 , R2 , . . . , R M ] T which can be
maintained with an outage probability no larger than Pout . Therefore, the outage rate region
will depend on the statistical parameters of the broadband channel.
In the following results the so-called broadband Rayleigh channel model is considered. This a
widely accepted model for propagation environments where there is not line of sight between
the transmitter and receiver. According to this model the time-domain channel response for
the m-th user hm is modeled as an independent zero-mean complex Gaussian random vector
hm ∼ CN (0, diag(Γ m )), where Γm = [ Γ m,1 Γ m,2 . . . Γ m,L ] T is the channel power delay profile
(PDP), which is assumed to decay exponentially

Γ m,l = E {hm,l h∗m,l } = Am ρlm , l = 1, . . . , L m , (18)

where L m is the length of hm , ρm is the exponential decay factor and Am is a normalization
factor given by

1 − ρm
A m = Em Lm
, (19)
ρ m (1 − ρ m )
Optimal Resource
Optimal Resource Allocation
Allocation in OFDMA inChannels
Broadcast OFDMA Broadcast
Using Dynamic Channels Using Dynamic Programming
Programming 133
17

Fig. 16. Outage rate regions for different values of probability of outage Pout .

being Em the average energy of the m-th user channel. Note that the frequency selectivity of
the channel is determined by ρm , so the higher the ρm the higher is the frequency selectivity
of the m-th user channel. The exponential decay PDP model is a widely used and it will be
assumed in the following results. Any other PDP model could be used. Unless otherwise
indicated, the parameters of the following simulations are
• Number of OFDM subcarriers: N=128
• i.i.d. Rayleigh fading channel model with ρ = 0.4, L = 16 and E = 1, for all users
• Available transmit power: PT = 10N
• Same probability of outage for all users: Pout = 0.1
• In the case of discrete rates, Sr = {0, 1/2, 3/4, 1, 3/2, 2, 3, 4, 9/2}
To obtain the outage rate region, 5000 channel realizations have been considered in each case.
The rate region of each channel realization has been obtained by solving (16) with the DP
algorithm.
Fig. 16 shows the outage rate regions for different values of outage probability (Pout ). One can
observe the performance gap between continuous and discrete rates, which is nearly constant
for different values of Pout . Since the channel is identically distributed for both users, the rate
regions are symmetric.
Fig. 17 shows the outage rate regions when the users’ channels have different average energy
(Em ). The sum of the average energy of the channels equals the number of users (2). In this
case, only continuous rate adaptation is considered. As it is expected, the user with the best
channel gets higher rates.
Fig. 18 shows the two-user outage rate regions for different values of average transmit power
per subcarrier P = PT /N. Note that the performance gap between continuous and discrete
rates increases with P.
Finally, Fig. 19 compares the outage rate regions for different values of channel frequency
selectivity. The figure clearly shows that the higher the frequency selectivity the more useful
is the resource adaptation. The gap between continuous and discrete rates does not depend
on the frequency selectivity of the channel.
134
18 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Fig. 17. Outage rate regions when the users’ channels have different average energy (Em ).

Fig. 18. Outage rate regions for different values of average power per subcarrier P.

6. Conclusions
This chapter analyzes the attainable performance of OFDMA in broadband broadcast
channels from an information-theoretic point of view. Assuming channel knowledge at
the transmitter, the system performance is maximized by optimally allocating the available
system resource among the users. The transmitter has to assign a user, a fraction of the
available power and a data rate (modulation and channel coding) to each subchannel. The
optimal allocation of these resources leads to non-linear constrained optimization problems
which, in general, are quite difficult to solve. These problems are solved by means of a
novel approach to the resource allocation problems in OFDMA systems by viewing them
as optimal control problems, where the control variables are the resources to be allocated
to the OFDM subchannels (power, rate and user). Once the problems are posed as optimal
control problems, dynamic programming is used to obtain the optimal resource allocation that
maximizes the system performance. This constitutes a new methodology for the computation
of optimal resource allocation in OFDMA systems. The achievable performance for given
Optimal Resource
Optimal Resource Allocation
Allocation in OFDMA inChannels
Broadcast OFDMA Broadcast
Using Dynamic Channels Using Dynamic Programming
Programming 135
19

Fig. 19. Outage rate regions for different values of PDP exponential decay factor (ρ).

channel realization is characterized by the rate region, whereas the overall performance of
OFDMA in random channels is characterized by means of the outage rate region. The cases
of continuous and discrete rate allocation are addressed under a general framework. The first
case leads to mixed optimization problems, whereas in the second case, the optimal resource
allocation is the solution of integer optimization problems.
By using the dynamic programming algorithm, the achievable rate region of OFDMA for
different channels and system parameters has been computed. The simulation results shows
that the performance gap between continuous and discrete rate adaptation is quite narrow
when the average transmit power per subcarrier is low, and it increases for higher values
of transmit power. The frequency selectivity of the broadband channel has a important
influence in the system performance. The higher the frequency selectivity, the more useful is
the adequate resource adaptation. The gap between continuous and discrete rate adaptation
does not depend on the frequency selectivity of the channel. The gap between continuous and
discrete rate adaptation remains nearly constant for different values of outage probability.

7. Appendix: Dynamic programming

Dynamic Programming (DP) is a well-known mathematical technique for solving sequential
control optimization problems with accumulative cost functions Bertsekas (2005). In a DP
problem there is always an underlying dynamical discrete process governed by a set of system
functions with the form

xk + 1 = f k (x k , c k ), 1≤k≤N (20)
where k is an index denoting a stage in the system evolution, N is the number of stages, xk
represents the state of the system at stage k and ck denotes the control action to be selected
at stage k (see Fig. 20). In general xk ∈ Sk ⊂ Rn , so the subset of possible states (Sk ) can
be different at different stages. In fact there is a fixed initial state x1 = x(0) , so the set of all
possible states at the first stage has an unique value S1 = {x(0) }. The control vector ck at
each stage are constrained to take values in a subset Ck (xk ), which, in general, depends on the
current state xk and on the stage (k).
136
20 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Fig. 20. Scheme of a DP problem showing the system states (xk ), controls (ck ) and costs (gk ) at
different stages

At each state the system incurs in an additive cost gk (xk , ck ), which, in general, depends on
the state and on the applied control. The objective is to minimize the total cost of the system
along its evolution by selecting the optimal controls at each stage
N
min
N
∑ gk (xk , ck ). (21)
{ck ∈Ck (xk )}k =1 k =1

Therefore, the elements of a DP problem are:

• The number of stages N,
• The control vector ck ,
• The state vector: xk
• The initial state: x1
• The subset of possible states at each stage Sk
• The subset of possible controls at each stage Ck ( xk ),
• System equation: xk+1 = f k ( xk , ck ),
• Cost functions: gk (ck ).
In general, these problems are difficult to solve. A key aspect is that controls cannot be viewed
in isolation since the controller must balance the cost at the current stage with the costs at
future stages. The DP algorithm captures this trade-off. The DP algorithm simplifies (21) by
breaking it down into simpler subproblems in a backwardly recursive manner Bellman (1957),
which is described in the following lines.
The optimal cost from state xk at stage k can be expressed as follows

 
Jk∗ (xk ) = min gk (xk , ck ) + Jk∗+1 ( f k (xk , ck )) , k = N − 1, . . . , 1 (22)
ck ∈Ck (xk )
∗
JN (x N ) = min gk (x N , c N ),
c N ∈ C k ( xN )

and the optimal control policy at stage k for state xk is

 
μ ∗k (xk ) = arg min gk (xk , ck ) + Jk∗+1 ( f k (xk , ck )) , k = N, . . . , 1
ck ∈Ck (xk )

Finally, the optimal control sequence c∗ = [c1∗ , c2∗ , . . . , c∗N ] and the corresponding system
evolution x∗ = [x1∗ , x2∗ , . . . , x∗N +1 ] are easily obtained by iteratively applying the optimal
control policies from the initial state as follows
Optimal Resource
Optimal Resource Allocation
Allocation in OFDMA inChannels
Broadcast OFDMA Broadcast
Using Dynamic Channels Using Dynamic Programming
Programming 137
21

x1∗ = x(0) , (23)

c∗k = μ∗k (xk ), x∗k+1 = f k (x∗k , c∗k ), k = 1, . . . N

8. Acknowledgements
This work was supported by the Spanish Government, Ministerio de Ciencia e
Innovacit’on (MICINN), under project COSIMA (TEC2010-19545-C04-03) and COMONSENS
(CSD2008-00010, CONSOLIDER-INGENIO 2010 Program).

9. References
Bellman, R. E. (1957). Dynamic Programming, Princeton University Press, Princeton, NJ, USA.
Bertsekas, D. P. (2005). Dynamic programming and optimal control, volume I, Athena Scientific,
Belmont, Massachusetts, USA.
Boyd, S. & Vandenberghe, L. (2004). Convex Optimization, Cambridge University Press,
Cambridge, UK.
Cioffi, J., Dudevoir, G., Eyuboglu, M. & Forney Jr, G. (1995). Mmse decision-feedback
equalizers and coding ii. coding results, IEEE Transactions on Communications
43(10): 2595–2604.
Cover, T. M. & Thomas, J. A. (1991). Elements of information theory, Wiley.
Goldsmith, A. & Effros, M. (2001). The capacity region of broadcast channels with intersymbol
interference and colored gaussian noise, IEEE Transactions on information Theory
47(1): 219–240.
Hoo, L., Halder, B., Tellado, J. & Cioffi, J. (2004). Multiuser transmit optimization for
multicarrier broadcast channels: Asymptotic fdma capacity region and algorithms,
IEEE Transactions on Communications 52(6): 922–930.
Jang, J. & Lee, K. (2003). Transmit power adaptation for multiuser ofdm systems, IEEE Journal
of Selected Areas in Communications 21(2): 171–178.
Lee, L. & Goldsmith, A. (2001). Capacity and optimal resource allocation for fading
broadcast channels - part ii: Outage capacity, IEEE Transactions on information Theory
pp. 1103–1127.
Ree, W. & Cioffi, J. (2000). Increase in capacity of multiuser ofdm system using dynamic
subchannel allocation, Proceedings of IEEE Vehicular Technology Conference (VTC’00),
Tokyo, Japan, pp. 1085–1089.
Seong, K., Mohseni, M. & Cioffi, J. M. (2006). Optimal resource allocation for ofdma
downlink systems, Proceedings of 2006 IEEE International Symposium on Information
Theory (ISIT’06), Seattle, Washington, USA.
Shen, Z., Andrews, J. G. & Evans, B. L. (2005). Adaptive resource allocation in multiuser ofdm
systems with proportional fairness, IEEE Transactions on Wireless Communications
4(6): 2726–2737.
Song, G. & Li, Y. (2005). Cross-layer optimization of ofdm wireless networks - part ii:
Algorithm development, IEEE Transactions on Wireless Communications 4(2): 625–634.
Tao, M., Liang, Y. C. & Zhang, F. (2008). Resource allocation for delay differentiated
traffic in multiuser ofdm systems, IEEE Transactions on Wireless Communications
7(6): 2190–2201.
Tse, D. & Viswanath, P. (2005). Fundamentals of Wireless Communications, Cambridge University
Press, Cambridge, UK.
138
22 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Wong, I. & Evans, B. (2008). Resource Allocation in Multiuser Multicarrier Wireless Systems,
Springer, New York, USA.
Yu, W. & Lui, R. (2006). Dual methods for nonconvex spectrum optimization of multicarrier
systems, IEEE Transactions on Communications 54(7): 1310–1322.
 0
7

Primary User Detection in

Multi-Antenna Cognitive Radio
Oscar Filio1 , Serguei Primak1 and Valeri Kontorovich2
1 The
University of Western Ontario
2 Centre of Research and Advanced Studies (CINVESTAV-IPN)
1 Canada
2 Mexico

1. Introduction
It is well known in the wireless telecommunications field that the most valuable resource
available is the electromagnetic radio spectrum. Being a natural resource, it is obviously finite
and has to be utilized in a rational fashion. Nevertheless the demand increase on wireless
devices and services such as voice, short messages, Web, high-speed multimedia, as well as
high quality of service (QoS) applications has led to a saturation of the currently available
spectrum. On the other hand, it has be found that some of the major licensed bands like
the ones used for television broadcasting are severely underutilized Federal Communications
Commission (November) which at the end of the day results in a significant spectrum
wastage. For this means, it is important to come up with a new paradigm that allows us to take
advantage of the unused spectrum. Cognitive radio has risen as a solution to overcome the
spectrum underutilization problem Mitola & Maguire (1999),Haykin (2005). The main idea
under cognitive radio systems is to allow unlicensed users or cognitive users (those who have
not paid for utilizing the electromagnetic spectrum), under certain circumstances, to transmit
within a licensed band. In order to perform this task, cognitive users need to continuously
monitor the spectrum activities and find a suitable spectrum band that allows them to:
• Transmit without or with the minimum amount of interference to the licensed or primary
users.
• Achieve some minimum QoS required for their specific application.
• Share the spectrum with other cognitive users.
Therefore, it is easy to observe that spectrum sensing is the very task upon which the entire
operation of cognitive radio rests Haykin et al. (2009). It is of extreme importance for the
system to be able to detect the so-called spectrum holes (underutilized subbands of the radio
spectrum). This is why in this chapter we focus all our attention to analyze some important
aspects of spectrum sensing in cognitive radio, and particularly the case when it is performed
using multiple antennas.
In order to take advantage of the cognitive radio features it is important to find which parts
of the electromagnetic spectrum are unused at certain moment. These portions are also
140
2 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

called spectrum holes or white spaces. If these bands are further used by a licensed user the
cognitive radio device has the alternative of either moving to another spectrum hoe or staying
in the same band but altering its transmission power lever or modulation scheme in order to
avoid the interference. Hence it is clear that an important requirement of any cognitive radio
network is the ability to sense such spectrum holes. As the most recent literature suggests
right now Akyildiz et al. (2008),Haykin et al. (2009), the most efficient way to detect spectrum
holes is to detect the primary users that are receiving data within the communication range
of a cognitive radio user. This approach is called transmitter detection which is based on the
detection of the weak signal from a primary transmitter through the local observations of
cognitive users. The hypotheses cab be defined as

H0 : n(t)
x (t) = , (1)
H1 : hs(t) + n(t)

where x (t) is the signal received by the cognitive user, s(t) is the transmitted signal of the
primary user, n(t) is the AWGN and h is the amplitude gain of the channel. H0 is a null
hypothesis, which states that there is no licensed user signal in a certain spectrum band. On
the other hand, H1 is an alternative hypothesis, which states that there exist some licensed
user signal. Three very famous models exist in order to implement transmitter detection
according to the hypotheses model Poor & Hadjiliadis (2008). These are the matched filter
detection, the energy detection and the cyclostationary feature detection.

1.1 Matched filter detection

When the information about the primary user signal is known to the cognitive user, the
optimal detector in stationary Gaussian noise is the matched filter since it maximizes the
received signal to noise ratio (SNR). While the main advantage of the matched filter is that
it requires less time to achieve high processing gain due to coherency, it requires a priori
knowledge of the primary user signal such as the modulation type and order, the pulse shape
and the packet format. So that, if this information is not accurate, then the matched filter
performs poorly. However, since most wireless networks systems have pilot, preambles,
synchronization word or spreading codes, these can be used for coherent detection,

1.2 Energy detection

If the receiver cannot gather sufficient information about the primary user signal, for example,
if the power of the random Gaussian noise is only known to the receiver, the optimal detector
is an energy detector. In order to measure the energy of the received signal, the output signal
of bandpass filter with bandwidth W is squared and integrated over the observation interval
T. Finally, the output of the integrator Y, is compared with some threshold λ to decide
whether a licensed user is present or not. Nevertheless, the performance of the energy detector
is very susceptible to uncertainty in noise power. Hence, in order to solve this problem, a pilot
tone from the primary transmitter can be used to help improve the accuracy of the energy
detector. Another shortcoming is that the energy detector cannot differentiate signal types
but can only determine the presence of the signal. Thus the energy detectors is prone to the
false detection triggered by the unintended signals.
Primary
Multi-Antenna User
CognitiveDetection in Multi-Antenna Cognitive Radio 1413
Primary User Detection in
Radio

1.3 Cyclostationary feature detection

An alternative detection method is the cyclostationary feature detection. Modulated
signals are in general couple with sine wave carriers, pulse trains, repeating spreading,
hopping sequences or cyclic prefixes, which result in built-in periodicity Kontorovich et al.
(2010). These modulated signals are characterized as cyclostationary since their mean and
autocorrelation exhibit periodicity. These features are detected by analyzing a spectral
correlation function. The main advantage of the spectral correlation function is that it
differentiates the noise energy from modulated signal energy, which is a result of the fact
that the noise is a wide-sense stationary signal with no correlation, while modulated signals
are cyclostationary with spectral correlation due to the embedded redundancy of signal
periodicity. Therefore, a cyclostationary feature detector can perform better than the energy
detector in discriminating against noise due to its robustness to the uncertainty in noise power.
Nonetheless, it is computationally complex and requires significantly long observation time.
Most of the previously mentioned techniques are investigated for a single sensor albeit some
use of multiple sensors is suggested in (Zhang et al., 2010). In the latter, the authors consider
a single sensor scenario equipped with multiple antennas and derived its performance in
assumption of correlated antennas and constant channel. Also, most of these studies are
focused on investigating the performance of particular schemes in ideal environments such
as independent antennas in cooperative scenario or in uniform scattering. However, such
consideration eliminate impact of real environment and its variation, while it is shown in
many publications and realistic measurements that such environments change frequently,
especially in highly build areas. Understanding how particular radio environment affects
performance of cognitive radio sensing abilities is, therefore, and important issue to consider.
Furthermore, it is well known (Haghighi et al., 2010) that the distribution of angle of arrival
(AoA), itself defined by scattering environment (Haghighi et al., 2010), affects both temporal
and spatial correlation of signals in antenna arrays. For these reasons in the first part of
the chapter we utilize a simple but generic model of AoA distribution, suggested in (Abdi
& Kaveh, 2002), to describe impact of scattering on statistical properties of received signals.
Later the concept of Stochastic Degrees of Freedom (SDoF) is incorporated in order to obtain
approximate expressions for the probability of miss detection in terms of number of antennas,
scattering parameters and number of observations. Following, the trade-off between the
number of antennas and required observation duration in correlated fading environments
is investigated. It is shown that at low SNR it is more convenient having just a single antenna
and many time samples so the noise suppression performs better. On the contrary, at high
SNR, since the noise is suppressed relatively quickly is better to have more antennas in order
to mitigate fading. Now, most of the existing spectrum schemes are based on fixed sample size
detectors, which means that their sensing time is preset and fixed. Hence, in the second part of
the chapter, we present some results based on the work of A. Wald (Wald, 2004) which showed
that a detector based on sequential detection requires less average sensing time than a fixed
size detector. We show that in general, it is possible to achieve the same performance that other
fixed sample based techniques offer but using as low as half of the samples in average in the
low signal to noise ratio regime. Afterwards, the impact of non-coherent detection is assessed
when detecting signals using sequential analysis. We finished using sequential analysis as a
new approach of cooperative approach for sensing. We call this an optimal fusion rule for
distributed Wald detectors and a evaluate its performance. The last section of the chapter is
devoted to conclusion remarks.
142
4 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Fig. 1. System Model

Fig. 2. Filtered Observations

2. Impact of scattering environment in spectrum sensing in multi-antenna
cognitive radio systems
2.1 Signal model
Let us consider a primary transmitter which transmits some pilot signal s over L symbols
in order to sound the primary channel. CR can sense the same signal using NR receiving
antennas. The received signal matrix X of size NR × L can be written in terms of the NR × L
complex channel matrix H = {hrl } and the noise matrix W of the same size as
X = Hs + W, (2)
Here W is a zero mean Gaussian matrix of covariance σn I and H is a zero mean Gaussian
matrix with covariance matrix RH respectively. Element hrl is the channel transfer coefficient
from the transmitter to r-th antenna measured at l-th pilot symbol. Using vectorization
operation ((van Trees, 2001)) , one can rewrite (2) as
x = hs + w, (3)
where x = vec X, h = vec H and w = vec W1 . Therefore, the detection problem is to
distinguish between the hypotheses
H0 : x [ n ] = w[n] n = 0, 1, . . . , NR L − 1
. (4)
H1 : x [ n ] = h [ n ] s + w [ n ] n = 0, 1, . . . , NR L − 1
1 The vec(·) operator is defined as the NR L × 1 vector formed by stacking the columns of the NR × L

matrix i.e. vec H = [h1 h2 . . . hL ]
Primary
Multi-Antenna User
CognitiveDetection in Multi-Antenna Cognitive Radio 1435
Primary User Detection in
Radio

The sufficient statistic in this case is given by (van Trees, 2001), (Kay, 1998)
  −1
T = x H Qx = |s|2 x H Rh |s|2 Rh + σn2 I x, (5)
 
where Rh = E hh H is the correlation matrix of the channel vector h = vec H.
This correlation matrix reflects both spatial correlation between different antennas and the
time-varying nature of the channel. Let Rh = UΛU H be eigendecomposition of the correlation
matrix Rh . In this case, the statistic T could be recast in terms of the elements of the
eigenvalues λi of the matrix Λ and filtered observations y = UH x:
  −1 NR L λ2k
T = y H Λ Λ + σn2 I y= ∑ + σn2
| y k |2 , (6)
k =1 λ k
2

which is analogous to equation (5.9) in (van Trees, 2001). Elements yk of the vector y could
be considered as filtered version of the received signal x with a set of orthogonal filters uk
(columns of the matrix U), i.e. could be considered as multitaper analysis (Thomson, 1982).
Linear filtering preserve Gaussian nature of the received signals, therefore, the distribution of
T could be described by generalized χ2 distribution2 (Andronov & Fink, 1971):
NR L
p( x ) = ∑ αk exp(− x/2λk ), (7)
k =1
and  
NR L
λl
αk−1 = 2λk ∏ 1− . (8)
l =1,l =k
λk
Theoretically, equation (7) could be used to set up the detection threshold γ. However, it is
difficult to use it for analytical investigation. Therefore, we would consider a few particular
cases of the channel when the structure of the correlation matrix could be greatly simplified
to reveal its effect on the detection performance.

2.2 Performance of estimator-correlator for PU detection

2.2.1 Constant independent channels
In this case the full covariance matrix Rh = σh2 O L ⊗ I NR is a Kronecker product of NR × NR
identity correlation matrix I NR and O L = 11 H is a L × L matrix consisting of ones. Therefore,
there are NR eigenvalues λk , k = 1, · · · NR equal to L. The k-th orthogonal filter uk is the
averaging operator applied to the data collected from the k-th antenna. Thus, the decision
statistic is just
2
NR L NR

TCI = ∑ ∑ xkl = ∑ Pk , (9)

k =1 l =1
k =1
where 2
L

Pk = ∑ xil . (10)
l =1

2 Assuming that all eigenvalues λk of Rh are different.

144
6 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

In absence of the signal, samples xkl are drawn from an i.i.d. complex Gaussian random
variable with zero mean and variance σn2 . Therefore, the distribution of Pk is exponential, with
the mean value Lσn2  
1 P
p( P) = exp − , (11)
Lσn2 Lσn2
and the distribution of T is just central χ2 distribution with NR degrees of freedom
 
1 T NR −1 T
pCI (T |H0 ) = exp − . (12)
Γ( NR ) ( Lσn2 ) NR Lσn2

If γ is a detection threshold for the statistic T then the probability PFA of the false alarm is

∞  
Γ NR , γ/Lσn2
PFA = pCI (T |H0 )dT = , (13)
γ Γ( NR )
or
γCI = Lσn2 Γ−1 [ NR , PFA Γ( NR )] , (14)
where Γ−1 [ NR , Γ( NR , x )]
= x. If the signal is present, i.e. if the hypothesis H1 is valid, then the
signal yi is a zero mean with the variance σ2 = L2 |s|2 σh2 + Lσn2 . As the result, the distribution
of the test statistic T under the hypothesis H1 is given by the central χ2 distribution with NR
degree of freedom and the probability of the detection is just

∞  
1 γ 1 1
PD = p(T |H1 ) = Γ NR , 2 = Γ NR , Γ−1 [ NR , PFA Γ( NR )] ,
γ Γ( NR ) σ Γ( NR ) 1 + Lμ̄
(15)
where
σ2
μ̄ = |s|2 h2 , (16)
σn
is the average SNR per symbol. Performance curves for this case could be found in (van Trees,
2001).
It could be seen from both (9) and (15) that under the stated channel model, the improvement
in performance PD comes either through reduction of noise through accumulation of signal
in each of the antennas (i.e. increase in the effective SNR) or through exploitation of diversity
provided by NR antennas Kang et al. (2010). Thus, increasing number of antennas leads to a
faster detection.

2.2.2 Spatially correlated block fading (constant spatially correlated channel)

Now let us assume that the values of the channel remain constant over L symbols but the
values of the channel coefficients for different antennas are correlated. In other words we will
assume that Rh = σh2 O L ⊗ Rs where Rs is the spatial correlation matrix between antennas. Let
Rs = Us Λs UsH be spectral decomposition of Rs . Then the test statistic T could be expressed,
according to equation (6), as
NR |s|2 σh2 λk NR
μ̄λk
TCC = ∑ | y k | 2
= ∑ | y k |2 , (17)
| | μ̄λ k +1
s 2σ λ + σ
2 2
k =1 h k n k =1
where σh2 is the variance of the channel per antenna. The eigenvalues λk of Rs reflect time
accumulation of SNR in each “virtual branch” of the equivalent filtered value yk . In general,
Primary
Multi-Antenna User
CognitiveDetection in Multi-Antenna Cognitive Radio 1457
Primary User Detection in
Radio

all the eigenvalues are different so one should utilize equation (7). While these calculations
are relatively easy to implement numerically, it gives little insight into the effect of correlation
on the performance of the detector.
Under certain scattering conditions (Haghighi et al., 2010), the eigenvalues of the matrix Rs
are either all close to some constant λ > 1 or close to zero. If there is Neq < NR non-zero
eigenvalues, their values are equal to λk = NR /Neq to preserve trace, and the rest NR − Neq
are equal to zero. In this case, the test statistic TCC could be further simplified to
Neq
TCC ( Neq ) = ∑ | y k |2 , (18)
k =1

where the index k corresponds to non-zero eigenvalues. Thus, the problem is equivalent to
one considered in Section 2.2.1 with Neq independent antennas and the expression for the
threshold γCC and the probability of detection are given by
 
αγCC = σn2 Γ−1 Neq , PFA Γ( Neq ) , (19)

where 0 < α < 1 performs as a corrector variable.

The effect of correlation between branches has dual effect on performance of the system. On
one side, the number Neq of equivalent independent branches is reduced, comparing to the
number of antennas NR , therefore reducing diversity. However, increased correlation results
into additional accumulation of SNR (or, equivalently, additional noise reduction through
averaging) by factor of NR /Neq ≥ 1. Therefore

∞  
1 αLγCC
PD = p(T |H1 )dT = Γ Neq , =
γ Γ( Neq ) σ2
 
1 1  
Γ Neq , Γ−1 Neq , PFA Γ( Neq ) . (20)
Γ( Neq ) 1 + LNR μ̄/Neq

2.2.3 Independent channel with temporal correlation

In the case of independent antennas but temporally correlated fading, the full correlation
matrix can be represented as Rh = R T ⊗ I L where R T = U TH Λ T U is the eigen decomposition
temporal correlation matrix of an individual channel Paulraj et al. (2003). The decision statistic
can now be represented as
NR   −1 NR
1
T ICC = ∑ xkH R T RT +
μ̄
IL xk = ∑ Tk , (21)
k =1 k =1

where xk is 1 × L time sample received by the k-th antenna. Therefore, each antenna signal is
processes separately and the results are added afterwards.
Taking advantage of eigendecomposition of the correlation matrix R T calculation of decision
statistic Tk can be recast as a multitaper analysis
  −1 L
1 λl
Tk = yk Λk Λk + I L yk = ∑ | y |2 . (22)
μ̄ λ
l =1 l
+ 1/μ̄ kl
146
8 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Once again, we can utilize approximation of the correlation matrix by one with constant or
zero eigenvalues as in Section 2.2.2. In this case there will be

(tr R T )2
Leq = (23)
tr R T R TH

eigenvalues of size L/Leq and the rest are zeros. Therefore, there is NR Le q terms in the sum
(21) each contributing
L/Leq μ̄L + Leq
= , (24)
L/Leq + 1/μ̄ μ̄L
into the variance of T ICC . Corresponding equations for choosing the threshold become
 
γCC = Lσn2 Γ−1 NR Leq , PFA Γ( NR Leq ) , (25)

∞
1 γ
PD = p(T |H1 )dT = Γ Neq , 2 =
γ Γ( Neq ) σ
 
1 1  
Γ Neq , Γ−1 Neq , PFA Γ( Neq ) . (26)
Γ( Neq ) 1 + LNR μ̄/Neq

2.2.4 Channel with separable spatial and temporal correlation

The correlation matrix of the channel with separable temporal and spatial correlation has the
correlation matrix of the form Rh = R T ⊗ Rs . Correlation in both coordinates reduces total
number of degrees of freedom from NR L to Neq Leq ≤ NR L The loss of degrees of freedom is
offset by accumulation of SNR due to averaging over the correlated samples. The equivalent
increase in the average SNR is NR L/Neq Leq . Thus, the problem is equivalent to detection
using
(trRs )2 (trR T )2
Keq = Neq Leq = , (27)
||Rs ||2F ||R T ||2F
independent samples in the noise with the average SNR
NR L
μ̄eq = μ̄. (28)
Neq Leq

The sufficient statistics in the case of the channel with separable spatial and temporal
correlation could be easily obtained from the general expression (5) and (6). In fact, using
Kronecker structure of Rh one obtains
Keq
T = ∑ | z k |2 . (29)
k =1

2.3 Examples and simulation

2.3.1 Correlation models
While the Jakes correlation function J0 (2π f D τ ) is almost universally used in standards on
wireless channels (Editors, 2006), realistic environment is much more complicated. A few
other models could be found in the literature, some chosen for their simplicity, some are based
on the measurements. In most cases we are able to calculate Neq analytically, as shown below.
Primary
Multi-Antenna User
CognitiveDetection in Multi-Antenna Cognitive Radio 1479
Primary User Detection in
Radio

1. Sinc type correlation If scattering environment is formed by a single remote cluster (as it is
shown in (Haghighi et al., 2010)), then the spatial covariance function Rs (d) as a function
of electric distance between antennas d is given by

Rs (d) = exp ( j2πd sin φ0 ) sinc (Δφd cos φ0 ) , (30)

where φ0 is the central angle of arrival, Δφ is the angular spread. This correlation matrix
has approximately 2Δφ cos φ0 N + 1 eigenvalues approximately equal eigenvalues with
the rest close to zero (Slepian, 1978).
2. Nearest neighbour correlation Neglecting correlation between any two antennas which
are not neighbours one obtains the following form of the correlation matrix Rs
⎧
  ⎨ ⎪ 1 if i = j
⎪
ρ if i = j + 1
Rs = rij = ∗ if i = j − 1 , (31)
⎪
⎪ ρ
⎩
0 if |i − j| > 1

where ρ is the correlation coefficient. The eigenvalues of (31) are well know (Kotz &
Adams, 1964)
kπ
λk = 1 − 2|ρ| cos , 1 ≤ k ≤ N. (32)
N+1
The equivalent number of independent virtual antennas is given by

N2 N
Neq = = . (33)
N + 2( N − 1)|ρ|2 1 + 2|ρ|2 (1 − 1/N )

3. Exponential correlation    
Rs = rij = |ρ|i− j . (34)
Eigenvalues of this matrix are well known (34) (Kotz & Adams, 1964)

1 − | ρ |2
λk = , (35)
1 + 2|ρ| cos ψk + |ρ|2

where ψk are roots of the following equation

sin( N + 1)ψ − 2|ρ|ψ sin N + |ρ|2 sin( N − 1)ψ

= 0. (36)
sin ψ

4. Temporal correlation model for nonisotropic scattering Considering the extended case of
the Clarke’s temporal correlation model for the case of nonisotropic scattering around the
user, we have the temporal correlation function as (Abdi & Kaveh, 2002):
 
I0 κ 2 − 4π 2 f D
2 τ 2 + j4πκ cos( θ ) f τ
D
Rs (τ ) = , (37)
I0 (κ )
where κ ≥ 0 controls the width of angle of arrival (AoA), f d is the Doppler shift, and
θ ∈ [−π, π ) is the mean direction of AoA seen by the user; I0 (·) stands for the zeroth-order
modified Bessel function.
148
10 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Theory (solid) vs. Simulation (x)

1
P = 0.1
FA
0.9 P = 0.01
FA
PFA = 0.001
0.8

0.7

PD 0.6

0.5

0.4

0.3

0.2

0.1

0
−25 −20 −15 −10 −5 0 5
SNR, dB

Fig. 3. ROC approximation vs. simulation results (α = 0.8) . Solid lines - theory, x-lines -
simulation.

7
κ=0
κ=1
6 κ=2
κ=3
κ=4
5
κ=5

4
ΛRs

0
1 2 3 4 5 6 7 8 9 10
No. eigenvalue

Fig. 4. Eigenvalues behavior of Rs temporal correlation matrix for nonisotropic scattering

(N = 10, μ = 0 and f d = 50Hz)

Figure 4 shows the eigenvalues behaviour for different values of the κ factor. Notice that
for κ = 0 (isotropic scattering) the values of the eigenvalues are spread in an almost equally
and proportional fashion among all of them. As κ tends to infinity (extremely nonisotropic
scattering), we obtain N − 1 zero eigenvalues and one eigenvalue with value N. In other
words, as κ increases, the number of “significant” eigenvalues decreases and hence so the
value of Neq as shown in Figure 6.

2.4 Simulation procedure

In order to perform the simulations which verified these results, the hypothesis in eq. (4) was
formed by giving the channel matrix H the desired correlation characteristics as shown in
Primary
Multi-Antenna User
CognitiveDetection in Multi-Antenna Cognitive Radio 149
Primary User Detection in
Radio 11

0.9

0.8

0.7

P
D 0.6

0.5

0.4

Nr = 1 L = 100
0.3 Nr = 4 L = 25 (ρ = 0)
Nr = 4 L = 25 (ρ = 0.3)
Nr = 4 L = 25 (ρ = 0.6)
0.2
Nr = 4 L = 25 (ρ = 0.95)

0.1
−25 −20 −15 −10 −5
SNR, dB

Fig. 5. Effect of correlation between antennas in the probability of detection.

0.9 κ = 0 (Theoretical)
κ = 0 (Simulation)
0.8
κ = 10 (Theoretical)
κ = 10 (Simulation)
0.7

0.6
PD

0.5

0.4

0.3

0.2

0.1

0
−25 −20 −15 −10 −5 0 5
SNR, dB

Fig. 6. ROC approximation for the estimator correlator considering the temporal correlation
for isotropic and nonisotropic scattering (κ = 0 and κ = 10 respectively).

section 2.3.1. Therefore, the vectorization operations are performed and after evaluating the
respective statistical tests, Monte Carlo method is utilized.

2.5 Space-time processing trade-off

It is common to assume that increasing number of antennas improves performance of
detection algorithms due to increased degree of diversity. Such proposition is correct when
the number of time samples remains the same. However, in cognitive networks it is desired to
reduce decision time as much as possible, sometimes by introducing some added complexity
in the form of additional number of antennas. The goal of this section is to show how one can
trade speed of making decision with a number of antennas available for signal reception.
It can be seen from equation (18) that the processing of the signal consists of two separated
procedures: averaging in time and accounting for diversity and suppressing noise in spatial
150
12 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

diversity brunches. Depending on amount of noise (SNR) and fading (fading figure (Simon
& Alouini, 2000)) one of these two technique brings more benefit to the net procedure.
For relatively low levels of SNR noise suppression is a dominant task, therefore it is more
advantageous to have a single antenna and as many time samples as possible. On a contrary,
if SNR is somewhat higher, noise is sufficiently suppressed even by short time averaging and
suppressing fading through diversity combining is more beneficial. This can be seen from
Fig. 5. This graph shows performance different configurations of the receiver in such a way
that product NR L remains constant. The same Figure shows effect correlation, and thus, the
scattering environment, plays on quality of reception. For very strong correlations ρ ≈ 1 and
Neq ≈ 1. Therefore all samples collected are used to reduce noise. Such scheme performs
the best at low SNR. However, when ρ = 0 and Neq = NR the gain from the diversity is
highest and the scheme performs best for higher SNR. Intermediate case allows for smooth
transition between these two regions. It also can be seen, that only in the case of ρ = 1
there is an equivalent trade off between number of antennas and time samples: i.e. the
performance depends only on Q = NR L regardless how the product is divided between NR
and L. However, lesser correlation result in unequal trade-off with gain or loss defined by
SNR and amount of correlation.

3. Sequential analysis for multi-antenna cognitive radio

3.1 Sequential analysis overview
The sequential analysis and the sequential probability ratio test (SPRT) introduced by A. Wald
in 1943 (Wald, 2004) have proved to be highly effective in taking decisions between two known
hypothesis (H0 , H1 ). Moreover, as it is shown in (Wald, 2004), the sequential probability ratio
test frequently results in a saving of about 50 percent in the average number of observations
in comparison to other well known detection techniques such as the Neyman-Pearson (NP)
decision test which is based on fixed number of observations. Unlike NP test, which utilizes
the logarithm of the Likelihood Ratio (log-LR) and compares it with a single predefined
threshold γ, the sequential probability ratio test compares the log-LR with two thresholds A
and B which are obtained in terms of the target probability of false alarm (PFA ) and probability
of misdetection (PMD ) (or the complementary probability of detection PD = 1 − PMD ) (Wald,
2004), (Middleton, 1960). Furthermore, in contrast to fixed decision time of NP test, the
duration of testing of sequential analysis is a random variable.
The thresholds A and B are approximated as (Wald, 2004):
1 − PMD PMD
A = ln , B = ln (38)
PFA 1 − PFA
The test procedure consists of sequential accumulating of m samples and calculating the
cumulative sum of the m-th log-LR as
m
B< ∑ Λ i < A, (39)
i =1

where Λi is a single log-likelihood ratio sample.

If Eq. (39) is satisfied, the experiment is continued by taking an additional sample increasing
m by 1. However, if
m
∑ Λi ≥ A, (40)
i =1
Primary
Multi-Antenna User
CognitiveDetection in Multi-Antenna Cognitive Radio 151
Primary User Detection in
Radio 13

Neyman−Pearson
Wald Test

2
10

No. of Samples

1
10

0
10
−20 −15 −10 −5 0
SNRdB

Fig. 7. Comparison of Neyman-Pearson Test and Sequential Probability Ratio Test

(PFA = 0.1,PD = 0.9).

the process is terminated with the acceptance of H1 . Similarly

m
∑ Λi ≤ B , (41)
i =1

leads to termination with the acceptance of H0 .

3.2 Wald test for complex random variable

Let us consider testing zero mean hypothesis in complex AWGN

H0 : zi = xi + jyi = wi
. (42)
H1 : z i = m + wi

Here m = m I + jmQ = μ exp( jφm ) = 0 is complex non zero mean and wi is i.i.d. complex zero
mean Gaussian process of variance σ2 .
A single sample log-likelihood ratio Λi is given by

p1 ( z i ; H1 ) 2μ ( xi cos φm + yi sin φm ) − μ2
Λi = ln = . (43)
p0 ( z i ; H0 ) σ2

After N steps of the sequential test the cumulative log-likelihood Λ ratio becomes
N
2μ Nμ2
Λ= ∑ Λn = σ2
TN − 2 ,
σ
(44)
n =1

where
N N
T N = cos φm ∑ xn + sin φm ∑ yn . (45)
n =1 n =1
The rest of the test follows procedure of Section 3.1.
152
14 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Figure 7 shows the performance comparison in number of samples needed between Wald Test
and Neyman-Pearson Test. Notice that the latter needs in general almost twice the number of
samples in order to detect the presence of the signal.
It follows from (45) that the sufficient statistic in the case of complex observation is
N
T = ∑ { x exp(− jφm )}. (46)
n =1

In other words, processing of the received signal is implemented in two stages: first, the data
is unitary rotated by the angle φm to align the mean along real axis; then the real part of data
is analyzed using the same procedure as purely real data.

3.2.1 Average number of samples

We have defined the sequential test procedure in eq.(39). Using this we can call
m
p(z , . . . , zm |H )
Λ= ∑ Λi = ln p(z11 , . . . , zm |H10 ) , (47)
i =1

where the random variable m stands for the required number of samples needed to terminate
the test. As stated in Wald (2004) it is possible to neglect the excess on threshold A and B ,
hence the the random variable can have the four possible combinations of terminations and
hypotheses such as ⎧
⎪
⎪ PFA A if H0 is true
⎨
PD A if H1 is true
Λ= . (48)
⎪ (1 − PFA )B if H0 is true
⎪
⎩
PM B if H1 is true
Following same reasoning we can get the conditional expectation for the random variable Λ
as 
PFA A + (1 − PFA )B if H0 is true
Λ̄ = . (49)
PD A + PM B if H1 is true
It is possible now to obtain the average number of samples (decision time) for accepting one
of the two hypothesis (H0 , H1 ) as:

PFA A + (1 − PFA B)
n̄(H0 ) = , (50)
Λ̄(H0 )
P A + (1 − PD B)
n̄(H1 ) = D , (51)
Λ̄(H1 )

where the term Λ̄(H0 ) can be calculated as

∑nN=1 Λn
Λ̄(H0 ) = , (52)
N
if no signal is present. The term Λ̄(H1 ) can be calculated analogously assuming there is a
signal present as follows

∑nN=1 Λn
Λ̄(H1 ) = . (53)
N
Primary
Multi-Antenna User
CognitiveDetection in Multi-Antenna Cognitive Radio 153
Primary User Detection in
Radio 15

3
10

Simulation
Theoretical
2
10

1
10

Samples (n)
0
10

−1
10

−2
10
−20 −15 −10 −5 0 5 10 15 20
SNR(dB)

Fig. 8. Average Number of Samples for Detection in Sequential Analysis

0.025

Simulation
Wald Distribution

0.02

0.015

0.01

0.005

0
0 50 100 150 200 250

Fig. 9. Approximation of decision time using Wald Distribution.

Fig. 8 shows the average number of samples needed to achieve a PD = 0.9 for different SNR.
The deviation at high SNR’s is due to the same effect already explained before. In practice for
very high SNR’s only one sample is enough to detect the presence of primary users.

3.2.2 Decision time distribution

as it was described earlier, decision time when using sequential analysis for detection is a
random variable. Hence it can be described by its PDF. Although an exact shape for this PDF
is not known in general, a very good approximation is available (specially for the low SNR
regimen) called Wald distribution or inverse Gaussian distribution defined as

λ − λ ( x − μ )2
f (x) = exp x > 0, (54)
2πx3 2μ2 x
where μ is the mean and λ > 0 is the shape parameter. In figure 9 it is shown Wald’s
distribution in order to approximate the decision time for a PD = 0.9.
154
16 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Though application of sequential analysis with high reliability of the hypothesis testing
(PFA , PM → 0) can provide an effective censoring of the information sent to other SU or FC
together with reduction of the sampling size at SU.

3.3 Sequential probability ratio test for partially coherent channel

Let us consider detection of a signal in a channel a with partially known phase. The received
signal can be modeled as
zi = m exp( jΔ) + wi , (55)
where m = m I + jm Q = μ exp( jφm ) is a deterministic and known complex constant, wi
is complex zero mean Gaussian noise with variance σ2 . Random variable Δ represents
uncertainty in measuring the phase of the carrier. Its distribution can be described by PDF
pΔ (Δ). In the following we assume that the phase uncertainty is described by the Von Mises
(or Tikhonov) PDF (von Mises, 1964)

exp [κ cos(Δ − Δ0 )]
pΔ (δ) = . (56)
2πI0 (κ )

Parameter Δ0 represents bias in the determination of the carrier’s phase, while κ represents
quality of measurements. A few particular cases could be obtained from (56) by proper choice
of parameters
1. Perfect phase recovery (coherent detection): κ = ∞, Δ0 = 0, and, thus, pΔ (Δ) = δ(Δ)
2. No phase recovery (non-coherent detection): κ = 0 and, pΔ (Δ) = 1/2π
3. Constant bias: κ = ∞, Δ0 = 0, pΔ (Δ) = δ(Δ − Δ0 )
We will derive the general expression first and then investigate particular cases to isolate
effects of the parameters on performance of SPRT.

3.3.1 Average likelihood ratio

For a single observation zi probability densities p1 (zi ) and p0 (zi ) corresponding to each of the
hypothesis H1 and H0 are given by
   
( xi − μ cos(φm + Δ))2 (yi − μ sin(φm + Δ))2
p1 (zi ) = C exp − exp − , (57)
σ2 σ2

and  
xi2 + y2i
p0 (zi ) = C exp − , (58)
σ2
respectively Filio, Primak & Kontorovich (2011). For a given Δ, the likelihood ratio Li could
be easily calculated to be
 
p (z ) 2μ ( xi cos(φm + Δ) + yi sin(φm + Δ)) − μ2
Li = 1 i = exp . (59)
p0 ( z i ) σ2
The conditional (on Δ) likelihood ratio L( N |Δ), considered over N observation is then just a
product of likelihoods of individual observations, therefore
Primary
Multi-Antenna User
CognitiveDetection in Multi-Antenna Cognitive Radio 155
Primary User Detection in
Radio 17

 
p1 ( z n )
N 2μ ∑nN=1 ( xn cos(φm + Δ) + yn sin(φm + Δ)) − Nμ2
L( N |Δ) = ∏ = exp
p (z )
n =1 0 n σ2
   
2μT ( N, Δ) Nμ2
= exp exp − , (60)
σ2 σ2
where
N N
T ( N, Δ) = cos(φm + Δ) ∑ xn + sin(φm + Δ) ∑ yn . (61)
n =1 n =1
Let us introduce a new variables, X ( N ), Y ( N ), Z ( N ) and Ψ( N ), defined by
N
X ( N ) = Z ( N ) cos Ψ( N ) = ∑ xn (62)
n =1
N
Y ( N ) = Z ( N ) sin Ψ( N ) = ∑ yn (63)
n =1

Using this notation equation (61) can now be rewritten as

T ( N, Δ) = Z ( N ) cos [φm + Δ − Ψ( N )] . (64)

The average likelihood (Middleton, 1960) L( N ) could now be obtained by averaging (60) over
the distribution of pΔ (Δ) to produce
 
π  
Nμ2 2μZ ( N ) cos [φm + Δ − Ψ( N )]
L̄( N ) = exp − 2 exp pΔ (Δ)dΔ. (65)
σ −π σ2
In turn, this expression could be further specialized if pΔ (Δ) is given by equation (56)
⎡ ⎤
  2 Z2 ( N )
Nμ2 1 4μ 4μZ ( N ) κ
L̄( N ) = exp − 2 I ⎣ + cos [φm − Ψ( N ) − Δ0 ] + κ 2 ⎦ . (66)
σ I0 (κ ) 0 σ4 σ2

It can be seen from (66) that it is reduced to (45) if Δ0 = 0 and κ = ∞. Furthermore the
deterministic phase bias Δ0 could be easily eliminated from consideration by considering z̃i =
zi exp[− j(φm + Δ0 )] instead of zi . Therefore, equation (66) could be simplified to
    
Nμ2 1 1 2 Y 2 ( N ) + [2μX ( N ) + κσ2 ]2 .
L̄( N ) = exp − 2 I0 4μ (67)
σ I0 (κ ) σ2
In the case of non-coherent detection, i.e. in the case κ = 0, expression (67) assumes a well
known form
   
Nμ2 2μZ ( N )
L̄( N ) = exp − 2 I0 . (68)
σ σ2
Formation of the likelihood ratio could be considered as a two step process. As the first
step, inphase and quadrature components independently accumulated to lessen the effect
of AWGN. At the second step, values of X ( N ) and Y ( N ) must be combined in a fashion
depending on available information. In the case of coherent reception it is know a priori that
156
18 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

3
10
κ = 0 (Non coherent)
κ = ∞ (Coherent)
κ=2
κ=7

2
10

Average samples (n)

1
10

0
10
−20 −15 −10 −5
SNR

Fig. 10. Impact of coherency on the average number of samples.

the quadrature component Y ( N ) contains only noise and it is ignored in the likelihood ratio.
On the contrary, in the case of non-coherent reception one cannot distinguish between the
in-phase and quadrature components and their powers are equally combined to for Z ( N ). In
the intermediate case both components are combined according to (67) with more and more
emphasis put on in-phase component X ( N ) as coherency increases with increase of κ.
In figure 10 we present the impact of non-coherent detection in the number of samples needed
in order to detect a signal with respect to some PD target. Notice that the main repercussion
of non-coherence detection is the increase of samples to nearly twice in comparison to the
coherent detector. In this terms, the non-coherent Wald sequential test procedure can be
thought as having the same efficiency (in terms of number of samples) as the coherent NP-test.

3.4 Optimal fusion rule for distributed Wald detectors

This Section generalized results of Chair & Varshney (1986) to the case of distributed detection
using Wal sequential analysis test. We assume that there are M sensors, making individual
detection according to the Wald algorithm. Once a decision is made at an individual sensor
the decision is send in the binary form to the Fusion Centre for further combining with other
decisions. We assume that the value u = −1 is assigned if the hypothesis H0 is accepted,
u = 1 if the hypothesis H1 is accepted and u = 0 if no decision has been made yet. Only
u = ±1 are communicated to the Fusion Centre.
Since each node utilizes the Wald sequential detection Wald (2004), the decision is made at a
random moments of time. Therefore, at any given moment of time t there is a random number
L(t) ≤ M of decisions which are available at FC as can be seen in figure 11. Probability
distribution of making decision could be approximated either by the two parametric Wald
distribution Wald (2004), or by three parametric generalized inverse Gaussian distribution
Jørgensen (1982) (as seen in fig.9). Parameters of those distributions could be found through
moment/cumulant fitting, using expressions derived in Wald (2004), Filio, Kontorovich
& Primak (2011). Following Chair & Varshney (1986) we would treat this problem as a
two-hypothesis detection problem with individual detector decision being the observation.
For a given number L = L(t) of decisions made by the time t, the optimum decision rule is
equivalent to the following likelihood ratio test
Primary
Multi-Antenna User
CognitiveDetection in Multi-Antenna Cognitive Radio 157
Primary User Detection in
Radio 19

Fig. 11. System model of data fusion system

P (u1 , u2 , · · · , u L | L, H1 ) P( L|H1 ) H1 P0 (C10 − C00 )

≷ . (69)
P (u1 , u2 , · · · , u L | L, H0 ) P( L|H0 ) H0 P1 (C01 − C11 )
Here P( L|H0 ) is the probability of making exactly L decisions assuming that H1 is true and
ul is the decision made by l-th sensor.
Furthermore, assuming the minimum probability of error criteria, i.e. by setting C00 = C11 = 0
and C01 = C10 = 1, introducing notation u L = {u1 , u2 , · · · , u L } and using the Bayes rule one
can recast equation (69) as
P(H1 |u L , l) P(l|H1 ) H1
≷ 1, (70)
P(H0 |u L , l) P(l|H0 ) H0
or, after taking natural log of both side
P(H1 |u L , l) P(l|H1 ) H1
ln + ln ≷ 0, (71)
P(H0 |u L , l) P(l|H0 ) H0
where l is a vector which represents which sensors have made decisions.
Once again following Chair & Varshney (1986), one can calculate probabilities P(H1 |u L , l) and
P(H0 |u L , l) as follows. In the case of the hypothesis H1 one can write

P(H1 , u L |l)
P(H1 |u L , l) = =
P(u L |l)
P1 P1
P(u L |l) ∏ P(ul = +1|H1 ) ∏ P(ul = −1|H1 ) = P(u L |l) ∏(1 − PM,l ) ∏ PM,l , (72)
S+ S− S+ S−

where S+ is the set of all i such that ui = +1 and S− is the set of all i such that ui = −1.
Similarly, in the case of the hypothesis H0 one obtains
P(H0 , u L |l) P0
P(H0 |u L , l) =
P(u L |l)
=
P(u L |l) ∏(1 − PF,l ) ∏ PF,l . (73)
S− S+

Finally, using equations (72) and (73) one obtains the following expression for the conditional
log-likelihood
P(H1 |u L , l) P 1 − PM,l PM,l
ln = ln 1 + ∑ ln + ∑ ln . (74)
P(H0 |u L , l) P0 S PF,l S
1 − PF,l
+ −
158
20 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

0.9
P M=5
MD
PFA M = 5
0.8
PMD M = 7
PFA M = 7
0.7
PMD M = 11
PFA M = 11
0.6
Asymptotic PMD M = 5
Asymptotic PFA M = 5
0.5 Asymptotic PMD M = 7
Asymptotic PFA M = 7
0.4 Asymptotic PMD M = 11
Asymptotic PFA M = 11
0.3

0.2

0.1

0
0 100 200 300 400 500 600 700 800 900 1000
time

Fig. 12. Data fusion scheme considering sequential analysis decision from each sensor
( PM,l = 0.3, PF,l = 0.1).

0.9
PMD M = 5
PFA M = 5
0.8
PMD M = 7

0.7 PFA M = 7
PMD M = 11
0.6 P M = 11
FA
Asymptotic PMD M = 5
0.5 Asymptotic PFA M = 5
Asymptotic PMD M = 7
0.4 Asymptotic PFA M = 7
Asymptotic PMD M = 11
0.3 Asymptotic PFA M = 11

0.2

0.1

0
0 100 200 300 400 500 600 700 800 900 1000
time

Fig. 13. Data fusion scheme considering sequential analysis decision from each sensor
( PM,l = 0.1, PF,l = 0.3).

In order to evaluate the second term in the sum (71) let us first consider an arbitrary node
1 ≤ k ≤ N. Distribution p T,k (τ ) of the decision time at such a node is assumed to be known.
Therefore, probability PD,k (t|Hi ) that the decision is made by the time t = mTs given that the
hypothesis is true is given by

mTS
PD,k (t|Hi ) = pk (τ |Hi )dτ. (75)
0
Primary
Multi-Antenna User
CognitiveDetection in Multi-Antenna Cognitive Radio 159
Primary User Detection in
Radio 21

0
10
Majority Decision
Maximum Likelihood

−1
10

−2
10

−3
10
4 6 8 10 12 14
Sensors

Fig. 14. Total Error Criteria

The probability that no decision has been made by the time t is then simply 1 − PD,k (t|Hi ).
As it has been mentioned earlier, parameters of this distribution could be defined as in Filio,
Kontorovich & Primak (2011). Therefore , the second term in the equation (71).

P(l|H1 ) L P (t|H1 ) M 1 − PD,l (t|H1 )

ln = ∑ ln D,l + ∑ ln . (76)
P(l|H0 ) l =1
PD,l ( t |H 0 ) l = L +1
1 − PD,l (t|H0 )

Finally, the fusion rule in the case of nodes making decision according to Wald’s criteria could
be written as 
1 if a0 + ∑lL=1 al ul > 0
f (u) = (77)
−1 otherwise
where

P1 L P (t|H1 ) M 1 − PD,l (t|H1 )

a0 = ln + ∑ ln D,l + ∑ ln , (78)
P0 l =1 PD,l (t|H0 ) l = L+1 1 − PD,l (t|H0 )
1 − PM,l
al = ln if ul = 1, (79)
PF,l
1 − PF,l
al = ln if ul = −1. (80)
PM,l

Thus, the combining rule is similar to that suggested in Chair & Varshney (1986), however,
with some significant differences in the term of a0 . In figures 12 and 13 it is shown the
performance of the data fusion scheme considering that each one of the sensors takes a
decision based on the sequential detection criteria. In these figures it is plotted the probability
of missdetection ( PMD ) and the probability of false alarm ( PFA ) versus the time in which the
fusion centre gathers the decisions taken from the local observers. Notice that for t → ∞ all
graphs converge to the data fusion rule of Chair & Varshney (1986) for which the theoretical
160
22 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

expression for probability of missdetection and false alarm is derived in appendix 5. It is

obvious that for small values of time, the fusion centre has less information (since not all
the detectors might achieve a decision by then) and the final decision it takes is way less
accurate that for large values of time. Nevertheless, in some practical systems it would be
impossible to wait that long for getting the decision from the fusion centre, so we can use
these results as a trade-off between the performance on the detection and the time of decision
Gosan et al. (2010). Next thing it is possible to observe is the impact that PD,l and PF,l has on
the performance of the data fusion detector.
Notice that for very small values of false alarm probability, al ≈ − ln PM,l if u = −1 in eq. (78)
which means that hypothesis H0 is always less weighted in equation (77) or in other words,
the fusion centre “trusts” more in those sensors who decide that H1 is true. For very small
values of missdetection probability a similar thing occurs but in this case the hypothesis H0
is more weighted in the final sum in equation (77). A special case occurs when PD,l = PF,l ,
P0 = P1 and t → ∞. In this situation, the scheme converts into the more simple majority
decision approach, which just sums all ul and compares with zero. Even though its simplicity,
the maximum likelihood approach performs better than the majority decision scheme in the
minimum probability of error criteria as can be seen in figure 14 Filio et al. (2010). The
perceptive reader must have noticed by now that there might be some confusion at the fusion
centre when there exists an even number of sensors and there is a tie in the decision. This can
be settled by considering the a priori probabilities P0 and P1 which are inherent to the system.

4. Conclusion
In the first part of this chapter, we have investigated the impact the scattering environment on
the performance of primary user detection in multiantennae confinguration. An approximate
expressions for the probability of missed detection in function of the number of antennas,
parameters of the scattering environment and number of observations. It is shown that at low
SNR it is more beneficial to utilize just a single antenna and large number of time samples.
This allowes for a better noise suppression via time averaging. On the contrary, at high SNR, it
is more beneficial to have more antennas in order to mitigate fading which is a dominant cause
of errors in detection in weak noise. It was also shown that for a very strong correlation, i.e.
ρ ≈ 1, the equivalent number of antennas is almost unity, Neq ≈ 1. Therefore this scheme can
be usefully applied in a low SNR situations assuming enough time samples are obtained. As
ρ ≈ 0, the diversity gain is increased, therefore making it suitable for higher SNR situations.
The second part of the chapter was devoted to application of the sequential analysis technique
to a chive a faster spectrum sensing in cognitive radio networks. It was shown that using the
sequential probability ratio test it is possible to detect the presence of a primary user almost
twice as fast as other fixed sample approaches such as Neyman Pearson detectors. This can be
achieved when dealing in the low SNR region which is a quite often operating characteristic in
real life. The effect of error in estimation phase of the carrier on the duration of the sequential
analysis has also been investigated. It was shown that the impact of non-coherent detection in
sensing the presence of primary users using sequential analysis is the increase of almost twice
the samples needed in comparison to a coherent detection approach. Afterwards we derived
an optimal fusion rule using detectors that use sequential analysis for taking decisions. We
assessed the performance of the system in terms of the time that it takes to gather the decision
from all detectors. It was shown that for faster decision, the fusion centre does not consider
the opinions of all sensors and hence the performance gets reduced. On the other hand, as
Primary
Multi-Antenna User
CognitiveDetection in Multi-Antenna Cognitive Radio 161
Primary User Detection in
Radio 23

the time of decision increases the performance is better but the system experiment a higher
latency.

5. Appendix
Performance derivation of data fusion rule
Let us introduce the following notations

ln 1−PPFAMD if ui > 0
ai = , (81)
ln 1P−MD
PFA
= − ln 1P−MD
PFA if ui < 0
and 
ai = ln 1−PPFAMD if ui > 0
ξi = . (82)
bi = − ai = − ln 1P−MD
PFA
if ui < 0
Consider a T (test statistic) given by eq. (77)
K K K
T = a0 + ∑ a k u k = a0 + ∑ ξ k | u k | = a0 + ∑ ξ k . (83)
k =1 k =1 k =1

Here ξ k could be considered as a random variable with PDF

Pξ ( x ) = P+ δ( x − a) + P− δx − b
(84)
= Pδ( x − a) + (1 − P)δ( x − b),

where
a = ai = ln 1−PPFAMD
PFA . (85)
b = − ai = − ln 1P−MD
and P+ is probability of u = +1 decision, equal to

P+ = p(H1 )(1 − PMD ) + p(H0 ) PFA

, (86)
= p(H1 )(1 − PMD ) + [1 − p(H1 )] PFA

The corresponding characteristic function of ξ is then given by

Θξ (s) = P+ e−sa + P− e−sb , (87)

and the characteristic function of T could be evaluated as

 K
ΘT = ΘξK e−sa0 = P+ e−sa + (1 − P+ )e−sb e−sa0
. (88)
k (1 − P )K −k e−s[ka+(K −k )b+ a0 ]
= ∑kK=0 (Kk ) P+ +

Equivalently, the PDF is given by

K  
K k
PT ( x ) = ∑ P (1 − P+ )K −k δ[ x − (ka + (K − k)b + a0 )]. (89)
k =0
k +

If k = 0 then
ka + (K − k)b − a0 = Kb + a0
162
24 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

1 − PFA P(H1 )
= −K ln + ln . (90)
PMD 1 − P(H1 )
If P(H1 ) ≈ 1 such that
 K
P(H1 ) 1 − PFA
> , (91)
1 − P(H1 ) PMD
then the FC makes only H1 decisions i.e.

PMD = 0, PFA = P(H0 ) = 1 − p(H1 ).

If (91) is not satisfied then there is kmax > 0 such that

kmax a + (K − kmax )b + a0 < 0, (92)

and
(kmax + 1) a + (K − kmax − 1)b + a0 > 0. (93)
In this case the scheme suggested in Chair & Varshney (1986) is equivalent to (kmax + 1) out
of K scheme (this is assuming that are statistically equivalent).
Let H1 be true. Then the target is missed if there are no more than kmax positive decisions, or,
equivalently, no less than K − kmax negative decisions. The probability of miss detection at FC
is then given by
kmax  
K k
PMDF = ∑ P (1 − PMD )K −k . (94)
k =0
k MD
To more decisions H1 there should be at least kmax + 1 partial 1. If H0 is true, the probability
of false alarm at the fusion center is then:

H1 : PMDF = k K k K −k
=0 ( k )(1 − PMD ) PMD
∑kmax
. (95)
H0 : PFAF = ∑k=kmax ( k )( PFA ) (1 − PFA )K −k
K K k

6. References
Abdi, A. & Kaveh, M. (2002). A space-time correlation model for multielement antenna
systems in mobile fading channels, 20(3): 550–560.
Akyildiz, I., Lee, W., Vuran, M. & Mohanty, S. (2008). A Survey on Spectrum Management in
Cognitive Radio Networks, 46(4): 40–48.
Almalfouh, S. & Stuber, G. (2010). Interference-aware power allocation in cognitive radio
networks with imperfect spectrum sensing, Proc. of ICC 2010, pp. 1 –6.
Andronov, I. & Fink, L. (1971). Peredacha diskretnykh soobshchenii po parallel’nym kanalam, Sov.
Radio, Moscow.
Chair, Z. & Varshney, P. (1986). Optimal data fusion in multiple sensor detection systems,
22(1): 98 –101.
Chenhui, H., Xinbing, W., Zichao, Y., Jianfeng, Z., Youyun, X. & Xinbo, G. (2010). A geometry
study on the capacity of wireless networks via percolation, 58(10): 2916 – 2925.
Editors, S. (2006). UMTS: Spatial channel model for MIMO simulations, Tech. report 25.996.
Federal Communications Commission (November). Spectrum policy task force, Technical
report, FCC.
Filio, O., Kontorovich, V. & Primak, S. (2011). Characteristics of sequencial detection in
cognitive radio networks, Proc. ICACT 2011, Phoenix Park, Korea.
Primary
Multi-Antenna User
CognitiveDetection in Multi-Antenna Cognitive Radio 163
Primary User Detection in
Radio 25

Filio, O., Kontorovich, V., Primak, S. & Ramos-Alarcon, F. (2010). Collaborative spectrum
sensing for cognitive radio: Diversity combining approach, Wireless Sensor Network
3(1): 24–37.
Filio, O., Primak, S. & Kontorovich, V. (2011). On performance of wald test in partially
coherent channels, Proc. ICCIT 2011, Aqaba, Jordan.
Gosan, N., Jemin, L., Hano, W., Sungtae, K., Sooyong, C. & Daesik, H. (2010). Throughput
analysis and optimization of sensing-based cognitive radio systems with markovian
traffic, 59(8): 4163–4169.
Haghighi, S. J., Primak, S., Kontorovich, V. & Sejdic, E. (2010). Mobile and Wireless
Communications Physical layer development and implementatiom, InTech Publishing,
chapter Wireless Communications and Multitaper Analysis: Applications to Channel
Modelling and Estimation.
Haykin, S. (2005). Cognitive radio: brain-empowered wireless communications,
23(2): 201–220.
Haykin, S., Thomson, D. & Reed, J. (2009). Spectrum sensing for cognitive radio, Proceedings
of the IEEE 97(5): 849 –877.
Jørgensen, B. (1982). Statistical Properties of the Generalized Inverse Gaussian Distribution. Lecture
Notes in Statistics. Vol. 9, Springer-Verlag, New York, Berlin.
Kang, H., Song, I., Yoon, S. & Kim, Y. (2010). A class of spectrum-sensing schemes for cognitive
radio under impulsive noise circumstances: Structure and performance in nonfading
and fading environments, 59(9): 4322 – 4339.
Kay, S. (1998). Statistical Signal Processing: Detection Theory, Prentice Hall PTR, Upper Saddle
River, NJ.
Kontorovich, V., Ramos-Alarcón, F., Filio, O. & Primak, S. (2010). Cyclostationary spectrum
sensing for cognitive radio and multiantenna systems, Wireless Communications and
Signal Processing (WCSP), 2010 International Conference on, pp. 1 –6.
Kotz, S. & Adams, J. W. (1964). Distribution of sum of identically distributed exponentially
correlated gamma-variables, The Annals of Mathematical Statistics 35(1): 277–283.
Li, H. & Han, Z. (2010a). Catch me if you can: An abnormality detection approach for
collaborative spectrum sensing in cognitive radio networks, 9(11): 3554 – 3565.
Li, H. & Han, Z. (2010b). Dogfight in spectrum: Combating primary user emulation attacks in
cognitive radio systems, part i: Known channel statistics, 9(11): 3566 – 3577.
Lin, S.-C., Lee, C.-P. & Su, H.-J. (2010). Cognitive Radio with Partial Channel State Information
at the Transmitter, 9(11): 3402–3413.
Middleton (1960). Introduction to Statistical Communications Theory, first edn, McGraw-Hill,
New York.
Mitola, J. & Maguire, G. (1999). Cognitive radio: making software radios more personal,
6(4): 13–18.
Mitran, P., Le, L. B. & Rosenberg, C. (2010). Queue-aware resource allocation for downlink
ofdma cognitive radio networks, 9(10): 3100 – 3111.
Paulraj, A., Nabar, R. & Gore, D. (2003). Introduction to Space-Time Wireless Communications,
Cambridge University Press, Cambridge, UK.
Poor, H. & Hadjiliadis, O. (2008). Quickest detection, Cambridge University Press.
Rui, Z. (2010). On active learning and supervised transmission of spectrum sharing based
cognitive radios by exploiting hidden primary radio feedback, 58(10): 2960 – 2970.
Shengli, X., Yi, L., Yan, Z. & Rong, Y. (2010). A parallel cooperative spectrum sensing in
cognitive radio networks, 59(8): 4079 – 4092.
164
26 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Simon, M. & Alouini, M.-S. (2000). Digital Communication over Fading Channels:A Unified
Approach to Performance Analysis, John Wiley & Sons, New York.
Slepian, D. (1978). Prolate spheroidal wave functions, Fourier analysis, and uncertainty. V-The
discrete case, Bell System Technical Journal 57: 1371–1430.
Song, M.-G., Kim, D. & Im, G.-H. (2010). Recursive channel estimation method for ofdm-based
cooperative systems, 14(11): 1029 – 1031.
Sun, H., Laurenson, D. & Wang, C.-X. (2010). Computationally tractable model of energy
detection performance over slow fading channels, 14(10): 924 –926.
Tang, P. K. & Han, C. (2010). On the modeling and performance of three opportunistic
spectrum access schemes, 59(8): 4070 – 4078.
Thomson, D. (1982). Spectral estimation and harmonic analysis, 70(9): 1055–1096.
van Trees, H. (2001). Detection, Estimation, and Modulation Theory: Part I, first edn, John Wiley
& Sons, New York.
von Mises, R. (1964). Mathematical theory of probability and statistics, Academic Press, New York.
Wald, A. (2004). Sequential Analysis, 6th edn, Dover Phoenix, Mineola, NY.
Wang, R. & Meixia, T. (2010). Blind spectrum sensing by information theoretic criteria for
cognitive radios, 59(8): 3806 – 3817.
Wenyi, Z. (2010a). Spectrum shaping: A new perspective on cognitive radioŮpart ii:
Coexistence with uncoded legacy transmission, 58(10): 2971 – 2983.
Wenyi, Z. M. (2010b). Spectrum shaping: a new perspective on cognitive radio-part i:
coexistence with coded legacy transmission, 58(6): 1857 – 1867.
Yücek, T. & Arslam, H. (2009). A survey of spectrum sensing algorithms for cognitive radio
applications, IEEE Comm. Surveys & Tutorials 11(1): 116–130.
Zhang, R., Lim, T. J., Liang, Y.-C. & Zeng, Y. (2010). Multi-antenna based spectrum sensing for
cognitive radios: A GLRT approach, 58(1): 84–88.
Zou, Q., Zheng, S. & Sayed, A. (2010). Cooperative sensing via sequential detection,
58(12): 6266 –6283.
 8

Multi-Cell Cooperation for

Future Wireless Systems
A. Silva, R. Holakouei and A. Gameiro
University of Aveiro/Instituto de Telecomunicações (IT)
Portugal

1. Introduction
The wireless communications field is experiencing a rapid and steady growth. It is expected
that the demand for wireless services will continue to increase in the near and medium term,
asking for more capacity and putting more pressure on the usage of radio resources. The
conventional cellular architecture considers co-located multiple input multiple output
(MIMO) technology, which is a very promising technique to mitigate the channel fading and
to increase the cellular system capacity (Foschini & Gans, 1998). On the other hand,
orthogonal frequency division multiplexing (OFDM) is a simple technique to mitigate the
effects of inter-symbol interference in frequency selective channels (Uppala & Li, 2004),
(Bahai et al., 2004). However, the problems inherent to these systems such as shadowing,
significant correlation between channels in some environments and intercell interference
significantly degrade the capacity gains promised by MIMO techniques (Andrews et al.,
2007). Although theoretically attractive, the deployment of MIMO in commercial cellular
systems is limited by interference between neighbouring cells, and the entire network is
essentially interference-limited (Foschini et al., 2006; Mudumbai et al., 2009).
Conventional approaches to mitigate multi-cell interference, such as static frequency reuse
and sectoring, are not efficient for MIMO-OFDM networks as each has important drawbacks
(Andrews et al., 2007). Universal frequency reuse (UFR), meaning that all cells/sectors
operate on the same frequency channel, is mandatory if we would like to achieve spectrally-
efficient communications. However, as it is pointed out in (Foschini et al., 2006), this
requires joint optimization of resources in all cells simultaneously to boost system
performance and to reduce the radiated power. Such systems have the advantage of macro-
diversity that is inherent to the widely spaced antennas and more flexibility to deal with
intercell interference, which fundamentally limits the performance of user terminals (UTs) at
cell edges (Andrews et al., 2007). Different transmit strategies can be considered, depending
on the capacity of the backhaul channel that connects the coordinated base stations.
Recently, an enhanced cellular architecture with a high-speed backhaul channel has been
proposed and implemented, under the European FUTON project (FUTON, 2011), (Diehm et
al., 2010). This project aims at the design of a distributed broadband wireless system (DBWS)
by carrying out the development of a radio over fiber (RoF) infrastructure transparently
connecting the BSs to a central unit (CU) where centralized joint processing can be
performed. Also, multi-cell cooperation is already under study in LTE under the Coordinated
166 Recent Advances in Wireless Communications and Networks

Multipoint (CoMP) concept (3GPP LTE, 2007) that although not included in the current
releases, will probably be specified for the future ones.
In recent years, relevant works on multi-cell precoding techniques have been proposed in
(Jing et al., 2008), (Somekh et al, 2007), (Boccardi & Huang, 2007), (Zhang et al, 2009),
(Marsch & Fettweis, 2009), (Armada et al., 2009), (Kobayashi et al., 2009), (Zhang, 2010),
(Bjornson et al., 2010). The multi-cell downlink channel is closely related to the MIMO
broadcast channel (BC), where the optimal precoding is achieved by the dirty paper coding
(DPC) principle (Costa, 1983). However, the significant amount of processing complexity
required by DPC prohibits its implementation in practical multi-cell processing. Some
suboptimal multi-cell linear precoding schemes have been discussed in (Jing et al., 2008),
where analytical performance expressions for each scheme were derived considering
nonfading scenario with random phases. The comparison of the achievable rates by the
different proposed cooperative schemes showed a tradeoff between performance
improvement and the requirement for BS cooperation, signal processing complexity and
channel state information (CSI) knowledge. In (Somekh et al, 2007) the impact of joint multi-
cellsite processing was discussed through a simple analytically tractable circular multi-cell
model. The potential improvement in downlink throughput of cellular systems using
limited network coordination to mitigate intercell interference has been discussed in
(Boccardi & Huang, 2007), where zero forcing (ZF) and DPC precoding techniques under
distributed and centralized architectures have been studied. In (Zhang et al, 2009) a
clustered BS coordination is enabled through a multi-cellblock diagonalization (BD) scheme
to mitigate the effects of interference in multi-cell MIMO systems. Three different power
allocation algorithms were proposed with different constraints to maximize the sum rate. A
centralized precoder design and power allocation was considered. In (Marsch & Fettweis,
2009), the inner bounds on capacity regions for downlink transmission were derived with or
without BS cooperation and under per-antenna power or sum-power constraint. The
authors showed that under imperfect CSI, significant gains are achievable by BS cooperation
using linear precoding. Furthermore the type of cooperation depends on channel conditions
in order to optimize the rate/backhaul tradeoff. Two multi-cell precoding schemes based on
the waterfilling technique have been proposed in (Armada et al., 2009). It was shown that
these techniques achieve a performance, in terms of weighted sum rate, very close to the
optimal. In (Kobayashi et al., 2009), each BS performs ZF locally to remove the channel
interference and based on the statistical knowledge of the channels, the CU performs a
centralized power allocation that jointly minimizes the outage probability of the UTs. A new
BD cooperative multi-cells scheme has been proposed in (Zhang, 2010), to maximize the
weighted sum-rate achievable for all the UTs. Multiuser multi-cell precoding with
distributed power allocation has been discussed in (Bjornson et al., 2010). It is assumed that
each BS has only the knowledge of local CSI and based on that the beamforming vectors
used to achieve the outer boundary of the achievable rate region was derived considering
both instantaneous and statistical CSI. An overview of the theory for multi-cell cooperation
in networks has been presented in (Gesbert et al., 2010).
In this chapter we design and evaluate linear precoding techniques for multi-cell MIMO-
OFDM cooperative systems. Two approaches are considered: centralized with a high-speed
backhaul channel, where it is assumed that full CSI and data are available at the CU; and
distributed with lower speed backhaul channel, where only some channel information and
data are shared by the BSs. The precoder design aims at two goals: allow spatial users
separation and optimize the power allocation. The two problems can be decoupled leading
Multi-Cell Cooperation for Future Wireless Systems 167

to a two step design: the precoder vectors design and power allocation algorithms. In this
chapter we discuss three centralized power allocation algorithms with different complexities
and per-BS power constraint: one optimal to minimize the average bit error rate (BER), for
which the powers can be obtained numerically by using convex optimization, and two
suboptimal. In this latter approach, the powers are computed in two phases. First the
powers are derived under total power constraint (TPC). Two criterions are considered,
namely minimization of the average BER, which leads to an iterative approach and
minimization of the sum of inverse of signal-to-noise ratio for which closed form solution is
achieved. Then, the final powers are computed to satisfy the individual per-BS power
constraint.
The rest of this chapter is organized as follows: in section 2 the general scenario is described,
section 3 discusses centralized multi-cell MIMO OFDM cooperative precoding schemes,
while in section 4 distributed multi-cell MIMO OFDM cooperative schemes are proposed, in
section 5 the simulation results are presented and discussed. Finally, conclusions are drawn
in section 6.
Throughout this chapter, we will use the following notations. Lowercase letters, boldface
lowercase letters and boldface uppercase letters are used for scalars, vectors and matrices,
respectively. ( .) , ( .) , ( .) represent the conjugate transpose, the transpose and complex
H T *

conjugate operators, respectively. E [.] represents the expectation operator, I N is the identity
matrix of size N × N , CN ( .,.) denotes a circular symmetric complex Gaussian vector, [ A ] i , j
is the (i,j)th element and [ A ] i is the ith column of the matrix A .

2. Scenario description
Multi-cell architectures that assume a global coordination can eliminate the intercell
interference completely. However, in practical cellular scenarios, issues such as the
complexity of joint signal processing of all the BSs, the difficulty in acquiring full CSI from
all UTs at each BS, and synchronization requirements will make global coordination
difficult. Therefore, in this chapter we assume a clustered multi-cell cellular system as
shown in Fig. 1, where the BSs are linked to a central unit (e.g., by optical fiber) as proposed
in (FUTON, 2010). In such architecture the area covered by the set of cooperating BSs is
termed as super-cell. The area defined by all the super-cells that are linked to the same CU is
termed as serving area. The BSs corresponding to a super-cell are processed jointly by a joint
processing unit (JPU). The number of cooperating BSs per super-cell should not be high for
the reasons discussed above. In this chapter, it is assumed that the interference between the
super-cells is negligible. In fact as we are replacing the concept of cell by the one of super-
cell, this means that there will be some interference among the super-cells especially at the
edges. Two approaches can be considered to deal with the inter-super-cell interference. The
precoders are designed to remove both intra-super-cell and inter-super-cell interference, but
as discussed in (Somekh et al., 2007) this strategy reduces the number of degrees of freedom
to efficiently eliminate the intra-super-cell interference. Alternatively, the radio resource
management can be jointly performed for a large set of super-cells (the serving area) at the
CU, and thus the resource allocation can be done in a way that the UTs of each super-cell
edge interfere as little as possible with the users of other super-cells (FUTON, 2010),
justifying our assumption to neglect it. This resource allocation problem is however beyond
the scope of this chapter. In this latter approach all degrees of freedom can be used to
efficiently eliminate the intra-super-cell interference.
168 Recent Advances in Wireless Communications and Networks

Serving Area

Super Cell BS

JPU … JPU
CU

Fig. 1. Enhanced cellular architecture

N tb

BSb
N tb UTk
N tb
... UT2 ...
BS2 BSB
N tb
UTK
UT1 ...
BS1

Backhaul Network

JPU

Fig. 2. Multi-cell system overview

We consider a scenario of B BSs comprising a super-cell; each BS is equipped with N tb
antennas, transmitting to K UTs as shown in Fig. 2. The total number of transmitting
antennas per-super-cell is N t . User k is equipped with single antenna or an antenna array
of N rk elements and the total number of receiving antennas per-super-cell is N r , which is
equal to the number of users K in case of single antenna UTs. Also, we assume an OFDM
based system with N c parallel frequency flat fading channels.
Multi-Cell Cooperation for Future Wireless Systems 169

3. Centralized multi-cell based system

We consider a multi-cell system based on the scenario defined in previous section where the
BSs are transparently linked by optical fiber to a central unit. Thanks to the high speed
backhaul, we can assume that all the information of all BSs, i.e., full CSI and data, belonging
to the same super-cell are available at the JPU. Thus, to remove the multi-cell multiuser
interference we can use a similar linear precoding algorithm designed for single cell based
systems. The major difference between multi-cell and single cell systems is that the power
constraints have to be considered on a per-BS basis instead. The proposed schemes are
considered in two phases: singular value decomposition SVD based precoding and power
allocation.

3.1 System model

To build up the mathematical model we consider that user k , k = 1,..., K can receive up to
N rk data symbols on subcarrier l , l = 1,..., N c i.e., xk ,l = [ x k ,1, l … x k , Nr , l ]T and the global
k

T
symbol vector, comprising all user symbol vectors, is xl =[ x1, l … xKT , l ]T of size N r × 1 .
The data symbol of user k on subcarrier l, is processed by the transmit precoder
N ×N
Wk , l ∈ C t rk in JPU, before being transmitted over BSs antennas. These individual precoders
together form the global transmit precoder matrix on subcarrier l , Wl = ⎡⎣ W1,l W K ,l ⎤⎦ of
size N t × N r . Let the downlink transmit power over the N t distributed transmit antennas
for user k and data symbol i , i = 1,..., N rk on subcarrier l, be pk,i,l , with p k ,l = ⎡ pk ,1, l … pk , N r , l ⎤
⎣ ⎦
and the global power matrix Pl =diag {[p 1, l p K , l ]} is of size N r × N r .
k

Under the assumption of linear precoding, the signal transmitted by the JPU on subcarrier l
is given by zl =Wl Pl1/2 x l and the global received signal vector on subcarrier l can be
expressed by,

y l =H lWl Pl1/2 x l +n l (1)

T
where Hl = ⎡⎣HT1, l HTK ,l ⎤⎦ of size N r × N t is the global frequency flat fading MIMO
channel on subcarrier l . The channel of user k is represented by
H k ,l = ⎡⎣ H 1, k ,l H b , k ,l H B, k ,l ⎤⎦ of size N rk × N t , and H b , k ,l of size N rk × N tb
represents the channel between user k and BS b , b = 1,..., B on subcarrier l . The channel
H b , k ,l can be decomposed as the product of the fast fading H cb , k ,l and slow fading ρb , k
components, i.e., H b , k ,l =H cb , k , l ρb , k , where ρ b , k represents the long-term power gain
between BS b and user k and H cb , k ,l contains the fast fading coefficients with CN ( 0,1 )
T
entries. nl = ⎡⎣nT1, l nTK ,l ⎤⎦ represents the global additive white Gaussian noise (AWGN)
T
vector and n k ,l = ⎡nk ,1,l … nk , N r ,l ⎤ is the noise at the user k terminal on subcarrier l
⎣ k ⎦

with zero mean and power σ 2 , i.e., E[n k ,ln H 2

k , l ]=σ I N r . k

The signal transmitted by the BS b on subcarrier l can be written as zb ,l =Wb ,l Pl1/2 x l , where
Wb ,l of size N tb × N r represents the global precoder at BS b on subcarrier l . The average
transmit power of BS b is then given by,
170 Recent Advances in Wireless Communications and Networks

K N rk N c
E ⎡ z b ⎤ = ∑ ∑∑ pk ,i ,l ⎡⎣ WbH, k ,l Wb , k ,l ⎤⎦
2
(2)
⎣ ⎦ k =1 i =1 l =1 i ,i

where z b is the signal transmitted over the N c subcarriers and Wb , k ,l of size N tb × N rk

represents the precoder of user k on subcarrier l at BS b .

3.2 Centralized precoder vectors

In this section, we consider the SVD based precoding algorithm similar to the one proposed
in (Yu et al., 2004). We assume that N t ≥ N r . Briefly, we define H k ,l as the following
( )
N r - N rk × N t matrix,
T
H k , l = ⎡⎣H1,l ...H k -1,l , H k + 1,l...HK ,l ⎤⎦ (3)

If we denote rank of H k ,l as Lk ,l then the null space of H k ,l has dimension of N t - Lk ,l ≥ N rk .

The SVD of H k , l is partitioned as follows,
H
H k , l =U k ,l D k , l ⎡ Vk(0) V(1) ⎤ (4)
⎣ ,l k ,l ⎦
where Vk(0) , l holds the N t - Lk , l singular vectors in the null space of H k , l . The columns of
Vk(0)
,l are candidate for user k precoding matrix Wk ,l , causing zero gain at the other users,
hence result in an effective SU-MIMO system. Since Vk(0) , l potentially holds more precoders
than the number of data streams user k can support, an optimal linear combination of these
vectors must be found to build matrix Wk ,l , which can have at most N rk columns. To do
this, the following SVD is formed,
H
H k ,l Vk(0) ⎡ (0) (1) ⎤ (5)
, l =U k , l D k , l ⎣ Vk , l Vk , l ⎦

where D k ,l is Lk , l × Lk ,l and Vk(1)

, l represents the Lk , l singular vectors with non-zero
singular values. The Lk ,l ≤ N rk columns of the product Vk(0) (1)
, l Vk , l represent precoders that
further improve the performance subject to producing zero inter-user interference. The
transmit precoder matrix will thus have the following form,

Wl = ⎡ V1,(0)l V1,(1)l ... VK(0),l VK(1),l ⎤ Pl 1/2 = Wl Pl 1/2 (6)

⎣ ⎦
The global precoder matrix with power allocation, Wl = ⎡⎣ W1,l ... WK ,l ⎤⎦ Pl1/ 2 as computed
above, block-diagonalizes the global equivalent channel Hl , i.e.,
{ }
H l Wl = diag ⎡⎣ H e ,1, l ,… , H e , K , l ⎤⎦ and the interference is completely removed considering
perfect CSI.
Let us define H e , k ,l =H k ,l Wk ,l = H k ,l Wk ,l Pk1/,l2 of size N rk × N rk as the equivalent enhanced
channel for user k on subcarrier l , where Pk ,l =diag{pk ,l } is of size N rk × N rk . Rewriting
equation (1) for this user, we have,

y k , l =H e , k , l x k , l +n k , l (7)

To estimate x k ,l , user k processes y k ,l by doing maximal ratio combining (MRC), and the
soft decision variable xˆ k ,l is given by
Multi-Cell Cooperation for Future Wireless Systems 171

xˆ k ,l =H H H H
e , k , l y k , l = H e , k , l H e , k , l x k , l +H e , k , l n k , l (8)

It should be mentioned that channel H e , k ,l can be easily estimated at UT k . It can be shown

that,

HH ⎡
{ ⎤
e , k , l H e , k , l = diag ⎣ pk ,1, l λk ,1, l ,… , pk , N r ,k , l λk , N rk , l ⎦ } (9)

where λk , i ,l is the ith singular value of matrix H k ,l Wk ,l . From equations (8) and (9) is easy to
see that the instantaneous SNR of data symbol i of user k on subcarrier l can be written as

pk ,i ,l λk ,i , l
SNR k ,i ,l = (10)
σ2
From (10), assuming a M-ary QAM constellations, the instantaneous probability of error of
data symbol i of user k on subcarrier l is given by (Proakis, 1995),

Pe , k ,i ,l = ψ Q ( β SNRk , i ,l ) (11)

∞
where Q( x ) = 1 / 2π ( )∫e ( − t 2 /2 )dt , β = 3 / ( M − 1 ) and ψ = ( 4 / log 2 M ) 1 − 1 / M . ( )
x

3.3 Power allocation strategies

Once the multi-cell multiuser interference removed, the power loading elements of Pl can be
computed in order to minimize or maximize some metrics. Most of the proposed power
allocation algorithms for precoded multi-cell based systems have been designed to
maximize the sum rate, e.g., (Jing et al., 2008; Bjornson et al., 2010). In this paper, the criteria
used to design power allocation are minimization of the average BER and sum of inverse of
SNRs, which essentially lead to a redistribution of powers among users and therefore
provide users fairness (which in practical cellular systems may be for the operators a goal as
important as throughput maximization). The aim of these power allocation schemes is to
improve the user’s fairness, namely inside each super-cell.
A. Optimal minimum BER power allocation
We minimize the instantaneous average probability under the per-BS power constraint Ptb ,
K N rk N c
i.e., ∑ ∑∑ pk ,i ,l ⎡ WbH, k ,l Wb , k ,l ⎤ ≤ Ptb , b = 1,..., B . Without loss of generality, we assume a
⎣ ⎦ i ,i
k =1 i =1 l =1
4-QAM constellation, and thus the optimal power allocation problem with per-BS power
constraint can be formulated as,

⎧ K N rk N c
⎛ 1 K N rk N c ⎛ p λ ⎞⎞ ⎪∑ ∑ ∑ p ⎡ W H W ⎤ ≤ Ptb ,b = 1,.., B
min ⎜
{pk ,i ,l } ⎜⎝ KN rk N c
∑ ∑∑ Q ⎜⎜ k ,iσ,l 2k ,i ,l ⎟ ⎟ s.t. ⎨ k = 1 i = 1 l = 1 k ,i ,l ⎣ b , k ,l b , k ,l ⎦ i ,i
⎟⎟
(12)
k =1 i =1 l =1 ⎝ ⎠⎠ ⎪p
⎩ k ,i ,l ≥ 0, k = 1,.., K , i = 1,.., N rk , l = 1,.., N c
Since the objective function is convex in pk ,i ,l , and the constraint functions are linear, this is
a convex optimization problem. Therefore, it may be solved numerically by using for
172 Recent Advances in Wireless Communications and Networks

example the interior-point method (Boyd & Vandenberghe, 2004). This scheme is referred as
centralized per-BS optimal power allocation (Cent. per-BS OPA).
B. Suboptimal power allocation approaches
Since the complexity of the above scheme is too high, and thus it could not be of interest for
real wireless systems, we also resort to less complex suboptimal solutions. The proposed
strategy has two phases: first the power allocation is computed by assuming that all BSs of
each super-cell can jointly pool their power, i.e., a TPC Pt is imposed instead and the above
optimization problem reduces to,

⎧ K N rk N c
⎪ ∑∑ pk ,i ,l ⎡ WkH,l Wk ,l ⎤ ≤ Pt
⎟ ⎟ s.t. ⎨ k∑
⎛ 1 K N rk N c ⎛ p λ ⎞⎞
min ⎜
⎜
{ k ,i ,l } ⎝ KN rk N c
p
∑ ∑∑ Q ⎜⎜ k ,iσ,l 2k ,i ,l ⎟⎟ =1 i =1 l =1
⎣ ⎦ i ,i (13)
k =1 i =1 l =1 ⎝ ⎠⎠ ⎪
⎩ pk ,i , l ≥ 0, k = 1,.., K , i = 1,.., N rk , l = 1,.., N c
K N rk N c K N rk N c
with ∑ ∑∑ pk ,i ,l ⎡ WkH,l Wk ,l ⎤ = ∑ ∑∑ pk , i ,l , note that the N r columns of
⎣ ⎦ i ,i k
w k ,l have
k =1 i =1 l =1 k =1 i =1 l=1
unit norm. Using the Lagrange multipliers method (Haykin, 1996), the following cost
function with μ Lagrange multiplier is minimized,

1 K N rk N c ⎛ p λ ⎞ ⎛ K N rk Nc ⎞
Jc ,1 =
KN rk N c
∑ ∑∑ Q ⎜⎜ k ,iσ,l 2k ,i ,l ⎟ + μ ⎜ ∑ ∑∑ pk ,i ,l − Pt ⎟
⎟ ⎜ k =1 i =1 l =1 ⎟
(14)
k =1 i =1 l =1 ⎝ ⎠ ⎝ ⎠
The powers pk ,i ,l can be determined by setting the partial derivatives of Jc ,1 to zero and as
shown in (Holakouei et al., 2011), the solution is

⎛ λ k2,i ,l ⎞
σ2
pk , i , l = W0 ⎜⎜ ⎟
2 4 ⎟
(15)
λk ,i ,l
⎝
2
(
⎜ 8πμ KN r N c σ ⎟
k
) ⎠
where W0 stands for Lambert’s W function of index 0 (Corless et al., 1996). This function
W0 ( x ) is an increasing function. It is positive for x > 0 , and W0 (0) = 0 . Therefore, μ 2 can be
K N rk Nc
determined iteratively to satisfy ∑ ∑∑ pk ,i ,l = Pt . The optimization problem of (13) is
k =1 i =1 l =1
similar to the single cell power allocation optimization problem, where the users are
allocated the same total multi-cell power, which may serve as a lower bound of the average
BER for the multi-cell with per-BS power constraint. One solution based on Lambert W
function that minimizes the instantaneous BER was also derived in the context of single user
single cell MIMO systems (Rostaing et al., 2002).
The second phase consists in scaling the power allocation matrix Pl by a factor of β in
order to satisfy the individual per-BS power constraints as discussed in (Zhang et al., 2009)
which can be given by

Ptb (16)
β=
⎛ K N rk Nc ⎞
max b =1,...,B ⎜⎜ ∑ ∑∑ pk ,i ,l ⎡⎣ Wb ,k ,l Wb , k ,l ⎤⎦ i ,i ⎟⎟
H

⎝ k =1 i =1 l =1 ⎠
Multi-Cell Cooperation for Future Wireless Systems 173

This scaled power factor assures that the transmit power per-BS is less or equal to Ptb . Note
that this factor is less than one and thus the SNR given by (10) has a penalty of
10log ( β ) dB . This scheme is referred as centralized per-BS suboptimal iterative power
allocation (Cent. per-BS SOIPA).
Although this suboptimal solution significantly reduces the complexity relative to the
optimal one, it still needs an iterative search. To further simplify we propose an alternative
power allocation method based on minimizing the sum of inverse of SNRs, and a closed-
form expression can be obtained. Note that minimizing the sum of inverse of SNRs is similar
to the maximization of the harmonic mean of the SINRs discussed in (Palomar, 2003). In this
case, the optimization problem is written as,

⎧ K N rk Nc
⎛ K N rk Nc σ2 ⎞ ⎪ ∑ ∑∑ p k ,i ,l ⎡ WkH,l Wk ,l ⎤ ≤ Pt
min ⎜ ∑ ∑∑ ⎟ s.t. ⎨ k = 1 i = 1 l = 1 ⎣ ⎦ i ,i (17)
{pk ,i ,l } ⎜⎝ k = 1 i = 1 l = 1 pk ,i ,lλk ,i , l ⎟⎠ ⎪p
⎩ k ,i ,l ≥ 0, k = 1,.., K , i = 1,.., N rk , l = 1,.., N c
Since the objective function is convex in pk ,i ,l , and the constraint functions are linear, (17) is
also a convex optimization problem. To solve it we follow the same suboptimal two phases
approach as for the first problem. First, we impose a total power constraint and the
following cost function, using again the Lagrangian multipliers method, is minimized,

K N rk N c
σ2 ⎛ K rk Nc
N ⎞
Jc ,2 = ∑ ∑∑ + μ ⎜ ∑ ∑∑ pk ,i , l − Pt ⎟ (18)
k = 1 i = 1 l = 1 pk , i , l λk , i , l
⎜ k =1 i =1 l =1 ⎟
⎝ ⎠
Now, setting the partial derivatives of Jc ,2 to zero and after some mathematical
manipulations, the powers pk ,i ,l are given by,

Pt
pk , i , l = (19)
K N rk N c
1
λk ,i ,l ∑ ∑ ∑
j =1 n=1 p =1 λ j ,n , p

The second phase consists in scaling the power allocation matrix Pl by a factor of β , using
(19) instead of (15), in order to satisfy the individual per-BS power constraints. This scheme
is referred as centralized per-BS suboptimal closed-form power allocation (Cent. per-BS
SOCPA).
The above power allocation schemes can also be used, under minor modifications, for the
case where the system is designed to achieve diversity gain instead of multiplexing gain. In
diversity mode the same user data symbol is received on each receiver antenna, increasing
the diversity order. Thus x k ,i ,l = x k , Nr ,l , i = 1...N rk − 1 and then the SNR is given by
k

N rk
pk , l ∑ λk ,i , l
i =1 pk , lα k ,l
SNR k , l = = (20)
2
σ σ2
and the power loading coefficient is computed only per user and subcarrier. In this case to
compute the power allocation coefficients we should replace λk ,i ,l by α k , l and remove the
script i in all equations.
174 Recent Advances in Wireless Communications and Networks

4. Distributed multi-cell based system

As discussed in section 2 due to limitations in terms of delay and capacity on backhaul
network, it is necessary to reduce signalling overhead. For this purpose, in this section the
precoders are designed in a distributed fashion, i.e., based on local CSI at each BS but we
still consider data sharing and centralized power allocation techniques.

4.1 System model

Assuming single antennas UTs and under the assumption of linear precoding, the signal
transmitted by the BS b on sub-carrier l is given by,

K
x b ,l = ∑ pb , k ,l wb , k ,l s k ,l , (21)
k =1

where pb,k,l represents the power allocated to UT k on sub-carrier l and BS b, wb , k ,l ∈

N ×1
tb

is the precoder of user k at BS b on sub-carrier l with unit norms, i.e.,

wb , k ,l = 1, b = 1,..., B, k = 1,..., K , l = 1,..., N c . The data symbol s k ,l , with E ⎡ s k ,l ⎤ = 1 , is
2
⎣⎢ ⎦⎥
intended for UT k and is assumed to be available at all BSs. The average power transmitted
by the BS b is then given by,

Nc K
E ⎡ x b ⎤ = ∑∑ pb , k ,l
2
(22)
⎣ ⎦ l =1 k =1

where x b is the signal transmitted over the N c subcarriers. The received signal at the UT k
on sub-carrier l , yk , l ∈ 1×1 , can be expressed by,

B
yk , l = ∑ hbH, k ,l xb , l + nk ,l (23)
b= 1

where hb , k ,l ∈
N ×1
tb
represents the frequency flat fading channel between BS b and UT k
( )
on sub-carrier l and nk ,l ~ CN 0,σ 2 is the noise.
The channel hb , k ,l , as for the centralized approach, can be decomposed as the product of the
fast fading hbc, k ,l and slow fading ρb , k components, i.e., hb , k , l =hbc , k ,l ρ b , k , where ρ b , k
represents the long-term power gain between BS b and user k and hbc, k ,l contains the fast
fading coefficients with CN ( 0,1) entries. The antenna channels from BS b to user k , i.e. the
components of hbc, k ,l , may be correlated but the links seen from different BSs to a given UT
are assumed to be uncorrelated as the BSs of one super-cell are geographically separated.

4.2 Distributed precoder vectors

As discussed above, to design the distributed precoder vector we assume that the BSs have
only knowledge of local CSI, i.e., BS b knows the instantaneous channel vectors hb , k ,l , ∀k , l ,
reducing the feedback load over the backhaul network as compared with the full centralized
precoding approach. We consider a zero forcing transmission scheme with the phase of the
received signal at each UT aligned. From (21) and (23) the received signal at UT k on sub-
carrier l can be decomposed in,
Multi-Cell Cooperation for Future Wireless Systems 175

B B K
yk , l = ∑ pb , k ,l hbH, k , l wb , k ,l sk ,l + ∑ hbH, k , l ∑ pb , j ,l wb , j ,l s j ,l + nk ,l (24)
b =1 b =1 j = 1, j ≠ k Noise
Desired Signal Multiuser Multicell Interference

where wb , k ,l is a unit-norm zero forcing vector orthogonal to K − 1 channel vectors,

{h } H
b , j ,l
j≠k
. Such precoding vectors always exist because we assume that the number of

antennas at each BS is higher or equal to the number of single antenna UTs, i.e. N tb ≥ K .
Note that here K is the number of users that share the same set of resources. Considering an
OFDMA based system, the total number of users can be significantly larger than K, since
different set of resources can be shared by different set of users. By using such precoding
vectors, the multi-cell interference is cancelled and each data symbol on each subcarrier is
only transmitted to its intended UT. Also, for any precoding vector wb , k ,l in the null space
of {h } H
b , j ,l
j≠k
, wb , k ,l = wb , k , l e jϕ is also in the null space of {h } H
b , j ,l
j≠k
. Thus, we can choose

the precoding vectors such that the terms hbH, k ,l wb , k ,l all have zeros phases, i.e.,
∠( hbH, k ,l wb , k ,l ) = 0, ∀ ( b , k , l ) . These precoding vectors can be easily computed, so if Wb , k ,l is

found to lie in the null space of {h }H

b , j ,l
j≠k
, the final precoding vector wb , k ,l , b = 1,..., B,

k = 1,..., K , l = 1,..., N c , with the phase of the received signal at each UT aligned, is given by,

(h )
H
H
Wb , k ,l
wb , k ,l = Wb , k , l
b , k ,l
(25)
hbH,k ,l Wb , k ,l

( ) holds the N − K + 1 singular vectors in the null space of

where Wb , k ,l ∈
N tb × N tb − K + 1
( tb )
{h } H
b , j ,l
j≠k
. For the case where N tb = K , only one vector lies in the null space of hbH, j ,l { } j≠k
,

but for N tb > K more than one vector lie in the null space of hbH, j ,l {
} . In this latter case, the
j≠k

final wb , k ,l vector is a linear combination of the ( N − K + 1) possible solutions. The

tb

equivalent channel between BS b and UT k , on sub-carrier l can be expressed as,

(h )
H
H
Wb , k , l
hbH, k ,l wb , k , l = hbH, k ,l Wb , k , l = hbH,k ,l Wb , k ,l = hbeq, k ,l (26)
b , k ,l

H
hb ,k ,l Wb , k ,l

From (26) we can observe that the equivalent channel, hbeq, k , l , is a positive real number. By
using the precoding vectors defined in (25) and considering (26), the received signal in (24)
reduces to,
B
yk ,l = ∑ pb , k , l hbeq, k ,l sk ,l + nk , l (27)
b =1
176 Recent Advances in Wireless Communications and Networks

It should be mentioned that at the UT, to allow high order modulations, only the pb , k , l hbeq, k ,l
coefficients are needed to be estimated instead of all the complex coefficients of the channel,
leading to a low complexity UT design.
( ) ( )
2
Since the N tb − K + 1 components of hbH, k ,l Wb , k ,l are i.i.d. Gaussian variables, hbeq, k ,l is a
( )
chi-square random variable with 2 N tb − K + 1 degrees of freedom. Once the hbeq, k , l
variables are independent, each user is expected to achieve a diversity order of
( )
B N tb − K + 1 (assuming that all channels have the same average power, i.e.,
ρb , k = ρ , ∀(b , k ) and pb , k ,l = 1, ∀(b , k , l ) ). Also, because the received signals from different
BSs have the same phase, they are added coherently at the UTs, and thus an additional
antenna gain is achieved.

4.3 Power allocation strategies

In this section the same three criteria considered for the centralized approach are used to
design the power allocation. However, it should be emphasised that for this scenario only
the equivalent channels, i.e., hbeq, k , l , are needed to be known at the JPU.
A. Optimal minimum BER power allocation
From (27) the instantaneous SNR of user k on sub-carrier l can be written as,
2
⎛ B eq ⎞
⎜ ∑ pb , k ,l hb , k ,l ⎟
SNR k ,l = ⎝ b =1 ⎠ (28)
σ2
The instantaneous probability of error for user k is obtained in similar way in section 3. We
minimize the instantaneous average probability under the per-BS power constraint Ptb , i.e.,
Nc K
∑ ∑ pb , k ,l ≤ Pt b
, b = 1,..., B . By assuming a 4-QAM constellation, the optimal power
l =1 k =1
allocation problem with per-BS power constraint can be formulated as,

⎛ ⎛ B eq ⎞ ⎞
⎜ ⎜ ∑ pb , k ,l hb , k ,l ⎟ ⎟ ⎧ Nc K
⎪∑∑ pb , k ,l ≤ Ptb , b = 1,.., B
Nc K
⎜ 1 ⎜ ⎟ ⎟ (29)
min ∑∑ Q
{pb ,k ,l } ⎜ KN c l = 1 k = 1 ⎜
b =1
σ ⎟⎟
s.t. ⎨ l = 1 k = 1
⎜ ⎜ ⎟ ⎟ ⎪p
⎜ ⎜ ⎟⎟ ⎩ b , k , l ≥ 0, b = 1,.., B, k = 1,.., K , l = 1,.., Nc
⎝ ⎝ ⎠⎠
In this distributed approach, the objective function is convex in pb,k,l, and the constraint
functions are linear this is also a convex optimization problem. Therefore, it may be also
solved numerically by using for example the interior-point method. This scheme is referred
as distributed per-BS optimal power allocation (Dist. per-BS DOPA). In this section, the
distributed term is referred to the precoder vectors since the power allocation is also
computed in a centralized manner.
B. Suboptimal power allocation approaches
As for the centralized approach, the complexity of the above scheme is too high, and thus it
is not of interest for real wireless systems, we also resort to less complex suboptimal
solutions. The proposed strategy has two phases: first the power allocation is computed by
assuming that all BSs of each super-cell can jointly pool their power, i.e., a TPC Pt is
imposed instead and the above optimization problem reduces to,
Multi-Cell Cooperation for Future Wireless Systems 177

⎛ ⎛ B eq ⎞ ⎞
⎜ ⎜ ∑ pb , k ,l hb , k ,l ⎟ ⎟ ⎧ B Nc K
⎟ ⎟ s.t. ⎪ ∑∑∑ pb , k ,l ≤ Pt
Nc K
1
min ⎜ ∑∑ Q ⎜⎜ b =1 σ ⎨b = 1 l = 1 k = 1 (30)
{pb ,k ,l } ⎜ KN c l =1 k =1
⎟⎟
⎪p
⎜ ⎜⎜ ⎟⎟ ⎟⎟ ⎩ b , k ,l ≥ 0, b = 1,.., B, k = 1,.., K , l = 1,.., N c
⎜
⎝ ⎝ ⎠⎠
B
with Pt = ∑ Pt and using the Lagrange multipliers method, the following cost function
b
b =1
with μ Lagrange multiplier is minimized,

⎛ B eq ⎞
⎜ ∑ pb , k , l hb , k ,l ⎟
⎟ + μ ⎛⎜ ⎞
Nc K B Nc K
1
J d ,1 = ∑∑ Q ⎜ b =1 ∑∑ ∑ pb , k ,l − Pt ⎟ (31)
KN c l = 1 k = 1 ⎜ σ ⎟ ⎜ ⎟
⎜⎜ ⎟⎟ ⎝ b =1 l =1 k =1 ⎠
⎝ ⎠
The powers pb , k ,l , ∀(b , k , l ) can be determined by setting the partial derivatives of J d ,1 to
zero and as shown in (Silva et al., 2011) the solution is,

⎛⎛ B 2⎞ ⎞
2

( )
σ 2 hbeq, k ,l
2
⎜
( )
⎜ ⎜ ∑ hieq, k ,l ⎟ ⎟
⎟
pb , k ,l = 2
W0 ⎜ ⎝ i = 1 2 2 2 ⎠4 ⎟ (32)
⎛B 2⎞ ⎜ 8πμ N c K σ ⎟
⎜ ∑(h )
eq
i , k ,l ⎟ ⎜ ⎟
⎝
i =1 ⎠ ⎝ ⎠
B Nc K
Therefore, μ 2 can be determined iteratively, using constraint ∑∑ ∑ pb ,k ,l = Pt . The second
b =1 l =1 k =1
phase consists of replacing μ 2 by μb2 , b = 1,..., B in (32), and then computing iteratively
different μb2 to satisfy the individual per-BS power constraints instead, i.e., μb2 are
computed to satisfy,

⎧ Nc K
⎪∑∑ pb , k , l ≤ Ptb , b = 1,..., B (33)
⎨ l =1 k =1
⎪p
⎩ b , k ,l ≥ 0, b = 1,.., B, k = 1,.., K , l = 1,.., N c
This suboptimal scheme is referred as distributed per-BS sub-optimal iterative power
allocation (Dist. per-BS SOIPA). Although this suboptimal solution significantly reduces the
complexity relative to the optimal one, it still needs an iterative search. To further simplify
we also propose for the distributed scenario, an alternative power allocation method based
on minimizing the sum of inverse of SNRs.
In this case, the optimization problem is written as,

⎛ ⎞
⎜ ⎟ ⎧ Nc K
⎜ ⎟ ⎪∑∑ pb , k ,l ≤ Ptb , b = 1,.., B
Nc K
σ2
min ⎜ ∑∑ ⎟ s.t. ⎨ l = 1 k = 1 (34)
{pb ,k ,l } ⎜ l = 1 k = 1 ⎛ B eq ⎞ ⎟
2
⎪p
⎜ ⎜ ∑ pb , k ,l hb , k ,l ⎟ ⎟ ⎩ b , k , l ≥ 0, b = 1,.., B, k = 1,.., K , l = 1,.., N c
⎝ ⎝ b = 1 ⎠ ⎠
178 Recent Advances in Wireless Communications and Networks

The objective function is convex in pb , k ,l , and the constraint functions are linear, (34) is also
a convex optimization problem. To solve it we follow the same suboptimal two phases
approach as for the first problem.
First, we impose a total power constraint and the following cost function, using again the
Lagrangian multipliers method, is minimized,
Nc K
σ2 ⎛ B Nc K ⎞
J d ,2 = ∑ ∑ + μ ⎜ ∑∑ ∑ pb , k ,l − Pt ⎟ (35)
2 ⎜ ⎟
l =1 k =1 ⎛ B ⎞ ⎝ b =1 l =1 k =1 ⎠
⎜∑ pb , k , l hbeq, k , l ⎟
⎝ b =1 ⎠
Now, setting the partial derivatives of J d ,2 to zero and after some mathematical
manipulations, the powers pb , k ,l can be shown to be given by,

(h )
eq 2
b , k ,l
pb , k ,l = (36)
3
⎛ B ⎞
( )
2
β ⎜ ∑ hieq, k ,l ⎟
⎝ i =1 ⎠
where β = μ / σ 2 . As for the first approach, (36) can be re-written by replacing β by
β b , b = 1,..., B , which are computed to satisfy the individual per-BS power constraints and
the closed-form solution achieved is then given by,

( )
2
Ptb hbeq, k ,l
pb , k ,l = (37)
(h )
2
3 eq
⎛ B ⎞ Nc K
( ) ⎟⎠ ∑ ∑
2 b, j,p
⎜ ∑ hi , k ,l
eq
3
⎝ i =1 ⎛ B eq ⎞
( )
p =1 j =1 2
⎜ ∑ hi , j , p ⎟
⎝ i =1 ⎠
This second suboptimal scheme is referred as distributed per-BS closed-form power
allocation (Dist. per-BS SOCPA).
The precoder vectors are designed by assuming that BSs have only knowledge of local CSI.
However, since we consider a centralized power allocation, to compute all powers the
hbeq, k ,l , ∀b,k,l coefficients should be available at the joint processing unit (JPU). In the
distributed multi-cell system each BS should send a real vector of size KN c to the JPU. Note
that in the centralized approach discussed in section 3, each BS should send to the JPU a
complex vector of size N tb KN c , i.e. 2 N tb more information.
Although, in this section single antenna UTs were assumed, the formulation can be
straightforwardly extended for multiple antenna UTs just by considering each antenna as a
single antenna UT. The main difference is that the long term channel power will be the same
for all antennas belonging to the same UT.

5. Results and discussions

5.1 Simulation parameters
In order to evaluate the proposed centralized and distributed multi-cell cooperation
schemes, we assume ITU pedestrian channel model B (Guidelines IMT2000, 1997), with the
Multi-Cell Cooperation for Future Wireless Systems 179

modified taps’ delays, used according to the sampling frequency defined on LTE standard
(3GPP LTE, 2007). This time channel model was extended to space-time by assuming that
the distance between antenna elements of each BS is far apart to assume uncorrelated
channels. To evaluate centralized and distributed schemes, the follwoing scenarios are
considered:
• Scenario 1, we assume that each supercell has 2 BSs, B = 2 which are equipped with 2
antennas, N tb = 2 and 2 UTs, K = 2 , equipped with 2 antennas, N rk = 2 .
• Scenario 2, we assume that each supercell has 2 BSs, B = 2 which are equipped with 2
antennas, N tb = 2 and 2 single antenna UTs, K = 2 .
• Scenario 3, we assume that each supercell has 2 BSs, B = 2 which are equipped with 4
antennas, N tb = 4 and 2 single antenna UTs, K = 2 .
The main parameters used in the simulations are, FFT size of 1024; number of resources, i.e.,
available subcarriers ( N c ) shared by the K users set to 16; sampling frequency set to 15.36
MHz; useful symbol duration is 66.6 μs; cyclic prefix duration is 5.21 μs; overall OFDM
symbol duration is 71.86 μs; subcarrier separation is 15 kHz and modulation is 4-QAM. We
assume that each UT is placed on each cell. The long-term channel powers are assumed to
be ρ b , k = 1, b = k for the intracell links, and ρ b , k , b ≠ k are uniformly distributed on the
interval [ 0.2 , 0.6 ] for the intercell links. All the results are presented in terms of the average
BER as a function of per-BS SNR defined as SNR = Ptb / σ 2 .

5.2 Performance evaluation

5.2.1 Centralized scenario
This section presents the performance results of centralized proposed precoding approaches
for scenario 1. We compare the performance results of four centralized precoding schemes:
one with non power allocation, which is obtained for the single cell systems by setting
Pl = I N r , i.e., the power per data symbol is constrained to one. For multi-cell systems the
power matrix Pl = I Nr should be scaled by β as defined in (16) (setting pk , i ,l = 1, ∀k , i , l ), i.e.,
Pl = β I N r ensuring a per-BS power constraint instead. This scheme is referred as centralized
per-BS non-power allocation (Cent. per-BS NPA). The two suboptimal approaches are Cent.
per-BS SOCPA and Cent. per-BS SOIPA; and the optimal one is Cent. per-BS OPA. Also, we
present results for optimal approach considering total power allocation (Cent. TPC OPA), as
formulated in (13), which may serve as a lower bound of the average BER for the centralized
multi-cell system with per-BS power constraint.
Fig. 3 shows the performance results of all considered precoding schemes for scenario 1,
considering multiplexing mode. It can be observed that the Cent. per-BS SOCPA, Cent. per-
BS SOIPA and Cent. per-BS OPA schemes have significant outperformance comparing to the
Cent. per-BS NPA approach, because they redistribute the powers across the different
subchannels more efficiently. Comparing the two suboptimal approaches we can see that
the iterative one, Cent. per-BS SOIPA, outperforms the closed-form, Cent. per-BS SOCPA
because the former is obtained by explicitly minimizing average probability of error. The
performance of the proposed suboptimal Cent. per-BS SOIPA and Cent. per-BS SOCPA
approaches is close, a penalty less than 0.7 dB for a BER=10-2 can be observed. Also, the
penalty of the Cent. per-BS SOIPA against the lower bound given by the Cent. TPC OPA is
only about 0.5 dB considering also a target BER=10-2.
Fig. 4 shows the performance results of all considered precoding schemes for scenario 1,
considering diversity mode. Comparing these results with the last ones, it can be easily seen
that there is a large gain due to operating in diversity mode. Since now each data symbol is
180 Recent Advances in Wireless Communications and Networks

Cent. per-BS NPA

Cent. per-BS SOCPA
-1
10 Cent. per-BS SOIPA
Cent. per-BS OPA
Cent. TPC OPA
BER

-2
10

-3
10
4 8 12 16 20 24 28
per-BS SNR (dB)

Fig. 3. Performance evaluation of the proposed centralized multi-cell schemes considering

multiplexing mode, for scenario 1

-1
10
Cent. per-BS NPA
Cent. per-BS SOCPA
Cent. per-BS SOIPA
-2
Cent. per-BS OPA
10 Cent. TPC OPA

-3
BER

10

-4
10

-5
10
2 4 6 8 10 12 14
per-BS SNR (dB)

Fig. 4. Performance evaluation of the proposed centralized multi-cell schemes considering

diversity mode, for scenario 1
Multi-Cell Cooperation for Future Wireless Systems 181

collected by each receive antenna of each UT. From this figure we basically can point out the
same conclusions as for the results obtained in the previous one. However, one important
thing that can be found out by comparing multiplexing and diversity modes is that the
difference between Cent. per-BS NPA curves and power allocation based curves (e.g. Cent.
per-BS SOIPA) is bigger in multiplexing mode (approximately 4dB) than diversity mode
(1.5dB) considering a BER=10-2. This can be explained by the fact that in the diversity mode
the equivalent channel gain of each data symbol is the addition of N rk individual channel
gains and thus the dynamic range of the SNRs of the different data symbols is reduced, i.e.,
somewhat leads to an equalization of the SNRs.

5.2.2 Distributed scenario

This section presents the performance results of proposed distributed precoding approaches
for scenario 2. We compare the results of four distributed precoding schemes with different
per-BS power allocation approaches: distributed per-BS equal power allocation (Dist. per-BS
EPA), in this case pb , k ,l = Ptb / KN c , ∀( b , k , l ) ; the two suboptimal approaches Dist. per-BS
SOIPA and Dist. per-BS SOCPA and the optimal one Dist. per-BS OPA. Also, the results for
optimal approach considering total power allocation (Dist. TPC OPA) , as formulated in (30)
are presented. This serves as lower bound for the distributed multi-cell scenario under per-
BS power constraint.
Fig. 5 shows the performance results of all considered distributed precoding schemes for
scenario 2. It can be observed that the Dist. per-BS SOCPA, Dist. per-BS SOIPA and Dist.
per-BS OPA schemes outperform the Dist. per-BS EPA approach, because they redistribute
the powers across the different subchannels more efficiently. For this case the performance

-1 Dist. per-BS EPA

10
Dist. per-BS SOCPA
Dist. per-BS SOIPA
Dist. per-BS OPA
-2 Dist. TPC OPA
10
BER

-3
10

-4
10

-5
10
2 4 6 8 10 12 14 16 18 20
per-BS SNR (dB)

Fig. 5. Performance evaluation of the proposed distributed multi-cell schemes, for scenario 2
182 Recent Advances in Wireless Communications and Networks

of the suboptimal Dist. per-BS SOIPA and optimal Dist. per-BS OPA is very close (penalty
less than 0.1dB), but the gap between these two schemes and the suboptimal Dist. per-BS
SOCPA is considerable. These results show that the Dist. per-BS SOIPA outperforms the
Dist. per-BS SOCPA for large number of subchannels. We can observe a penalty of
approximately 0.6 dB of the Dist. per-BS SOCPA scheme against the Dist. per-BS SOIPA for
a BER=10-3. Also, a gain of approximately 4.2 dB of the suboptimal Dist. per-BS SOIPA
scheme against the Dist. per-BS EPA is obtained, considering BER=10-3.

5.2.3 Performance comparison

This section presents the performance results of both distributed and centralized proposed
precoding approaches for scenarios 2 and 3.
Fig. 6 shows the results for scenario 2, from this figure we can see that the performance of all
power allocation schemes with centralized precoding outperforms the one with distributed
scheme, because there are more degrees of freedom (DoF) to remove the interference and
enhance the system performance. In the distributed case, the performance of the suboptimal
Dist. per-BS SOIPA and optimal Dist. per-BS OPA is very close (penalty less than 0.1dB), but
the gap between these two schemes and the suboptimal per-BS SOCPA is almost increased
to 0.8dB (BER=10-3). In the case of centralized precoding the performances of Cent. per-BS
SOIPA and Cent. per-BS OPA are still very close but both are degraded from Cent. TPC
OPA (about 0.5dB at BER=10-3) and also there is 0.5dB gap among these curves and Cent.
per-BS SOCPA at the same BER. Another important issue that should be emphasized is that
the penalty of the per-BS OPA against the TPC OPA is approximately 0.1 dB (BER=10-3) for
distributed scheme, against 0.5dB for centralized case.
Figure 7 shows the performance results of both distributed and centralized schemes for
scenario 3. By observing this figure almost the same conclusions can be drawn. An
interesting result is that the performances of distributed and centralized schemes are much
closer comparing with scenario 2. This can be explained by the fact that for the centralized
approach the number of DoF, which is given by the number of total transmit antennas
BN tb , increased from 4 (scenario 2) to 8 (scenario 3); while for the distributed approach, the
( )
number of DoF, which is given by B N tb − K + 1 as discussed before; is increased from 2
(scenario 2) to 6 (scenario 3), i.e., the number of DoF of both centralized and distributed
approaches is closer than that in scenario 2. From the presented results two important facts
should be also emphasized: first is that in case of distributed precoding, the performance
improvement achieved with the three proposed power allocation techniques, is higher than
the case of centralized scheme; the second is that in the case of distributed precoding, the
suboptimal techniques are more successful in achieving the lower bound of average BER.

6. Conclusion
In this chapter we proposed and evaluated centralized and distributed multi-cell multiuser
precoding schemes for MIMO OFDM based systems. The proposed precoder vectors were
computed either jointly and centraly at JPU benefiting from high DoF or on each BS in a
distributed manner allowing a low feedback load over the backhaul network, while the
power allocation was computed in a centralized fashion at the JPU.
The criteria considered was the minimization of the BER and two centralized power
allocation algorithms with per-BS power constraint: one optimal that can be achieved at the
expense of some complexity and one suboptimal with lower complexity aiming at practical
Multi-Cell Cooperation for Future Wireless Systems 183

-1
10

Distributed precoding

-2
10
BER

-3
10

Centralized precoding

-4
10
per-BS EPA
per-BS SOCPA
per-BS SOIPA
per-BS OPA
-5
TPC OPA
10
2 4 6 8 10 12 14 16 18 20
per-BS SNR (dB)

Fig. 6. Performance evaluation of the proposed distributed and centralized multi-cell

schemes for scenario 2

-1
10

Distributed precoding
-2
10

-3
BER

10

Centralized precoding

-4
10
per-BS NPA
per-BS SOCPA
per-BS SOIPA
per-BS OPA
TPC OPA
-5
10
-2 0 2 4 6 8 10
per-BS SNR (dB)

Fig. 7. Performance evaluation of the proposed distributed and centralized multi-cell

schemes for scenario 3
184 Recent Advances in Wireless Communications and Networks

implementations. In both the optimal (per-BS OPA) and the suboptimal (per-BS SOIPA), the
computation of the transmitted powers required an iterative approach. To circumvent the
need for iterations further proposed another suboptimal scheme (per-BS SOCPA), where the
power allocation was computed in order to minimize the sum of inverse of SNRs of each UT
allowing us to achieve a closed-form solution.
The results have shown that the proposed multi-user multi-cell schemes cause significant
improvement in system performance, in comparison with the case where no power
allocation is used. Also for both approaches, the performance of the proposed suboptimal
algorithms, namely the per-BS SOIPA approach, is very close to the optimal with the
advantage of lower complexity. Also, the performance of the distributed approach tends to
the one achieved by the centralized, when the number of DoF available tends to the number
of DoF available in the centralized system. Therefore, distributed schemes can be interesting
in practice when the backhaul capacity is limited.
It is clear from the presented results the suboptimal proposed either distributed or
centralized precoding schemes allow a significant performance improvement with very low
UT complexity and moderate complexity at both BS and JPU, and therefore present
significant interest for application in next generation wireless networks for which
cooperation between BSs is anticipated.

7. Acknowledgments
The authors wish to acknowledge the support of the Portuguese CADWIN project,
PTDC/EEA TEL/099241/2008, and Portuguese Foundation for Science and Technology
(FCT) grant for the second author.

8. References
Andrews, J. G.; Choi, W. & Heath, R. W. (2007). Overcoming interference in spatial
multiplexing MIMO cellular networks, IEEE Wireless Communication Magazine, vol.
47, no. 6, pp. 95-104, 2007.
Armada, A. G.; Fernándes, M. S. & Corvaja, R. (2009). Waterfilling schemes for zero-forcing
coordinated base station transmissions, in Proceeding of IEEE GLOBECOM, 2009.
Bahai, A. R. S.; Saltzberg, B. R. & Ergen, M. (2004). Multi-carrier digital communications theory
and applications of OFDM, Springer, New York.
Bjornson, E.; Zakhour, R.; Gesbert, D. & Ottersten, B. (2010). Cooperative multi-cell
precoding: rate region characterization and distributed strategies with
instantaneous and statistical CSI, IEEE Transaction on Signal Processing, vol. 58, no.
8, pp. 4298-4310, 2010.
Boccardi, F. & Huang, H. (2007). Limited downlink network coordination in cellular
networks, In Proeeding IEEE PIMRC’07, 2007.
Boyd, S. & Vandenberghe, L. (2004). Convex optimization, Cambridge: Cambridge University
Press, 2004.
Corless, R. M.; Gonnet, G. H.; Hare, D. E. G.; Jeffrey, D. J. & Knuth, D. E. (1996). On the
Lambert W function, Advances in Computer & Mathematics, vol. 5, pp. 329–359, 1996.
Costa, M. H. M. (1983). Writing on dirty paper, IEEE Transaction on Information Theory, vol.
29, no. 3, pp. 439–441, 1983.
Multi-Cell Cooperation for Future Wireless Systems 185

Diehm, F.; Marsch, P. & Fettweis, G. (2010). The FUTON prototype: proof of concept for
coordinated multi-point in conjunction with a novel integrated wireless/optical
architecture, In Proceeding of IEEE WCNC’10, 2010.
Foschini, G. J. & Gans, M. J. (1998). On limits of wireless communications in a fading
environment when using multiple antenna, Wireless Personal Communication
Magazine, vol. 6, no. 3, pp. 311-335, 1998.
Foschini, G. J. et al. (2006). Coordinating multiple antenna cellular networks to achieve
enormous spectral efficiency, in IEEE Proceedings on Communications, Vol. 153, No.
4, pp. 548-555, August 2006.
FUTON, European ICT project. (2007). Available from: https://www.ict-futon.eu.
Gesbert, D.; Hanly, S.; Huang, H.; Shamai, S.; Simeone, O. & Yu, Wei (2010). Multi-cell
MIMO cooperation networks: A new look at interference, IEEE Journal on Selected
Areas in Communications, vol. 28, no. 9, pp. 1380-1408, Dec. 2010.
Guidelines for the evaluation of radio transmission technologies for IMT-2000. (1997).
Recommendation ITU-R M.1225, 1997.
Haykin, S. (1996). Adaptive Filter Theory, 3rd Ed., Prentice Hall, 1996.
Holakouei, R; Silva, A & Gameiro, A. (2011). Multiuser precoding techniques for a
distributed broadband wireless system, Accepted to Telecommunication System
Journal, special issue in WMCNT, Springer.
Silva, A; Holakouei, R & Gameiro, A. (2011). Power Allocation Strategies for Distributed
Precoded Multicell Based Systems, EURASIP Journal on Wireless Communications and
Networking, special issue in Recent Advances in Multiuser MIMO System, Vol. 2011,
2011.
Jing, S.; Tse, D. N. C.; Sorianga, J. B.; Hou, J.; Smee, J. E. & Padovani, R. (2008). Multicell
downlink capacity with coordinated processing, EURASIP Journal of Wireless
Communicaion Network, 2008.
Karakayali, M.; Foschini, G. & Valenzuela, G. (2006). Network coordination for spectral
efficient communications in cellular systems, IEEE Wireless Communication
Magazine, vol. 13, no. 4, pp. 56-61, 2006.
Kobayashi, M.; Debbah, M. & Belfiore, J. (2009). Outage efficient strategies in network
MIMO with partial CSIT, in Proceeding of IEEE ISIT, pp. 249-253, 2009.
Lozano, A.; Tulino, A. M. & Verdú, S. (2008). Optimum Power Allocation for Multiuser
OFDM with Arbitrary Signal Constellations, IEEE Transactions on Communications,
vol. 56, no. 5, pp. 828-837, 2008.
Marsch, P. & Fettweis, G. (2009). On Downlink network MIMO under a constrained
backhaul and imperfect channel knowledge, in Proceeding of IEEE GLOBECOM,
2009.
Mudumbai, T. R.; Brown, D. R.; Madhow, U. & Poor, H. V. (2009). Distributed transmit
beamforming: challenges and recent progress, IEEE Communication Magazine, vol.
47, no. 2, pp. 102-110, 2009.
Palomar, D. P.; Cioffi, J. M. & Lagunas, M. A. (2003). Joint Tx-Rx beamforming design for
multicarrier MIMO channels: A unified framework for convex optimization, IEEE
Transactions on Signal Processing, vol. 51, no. 9, pp. 2381-2401, 2003.
Proakis, J. (1995). Digital Communications, 3 rd Ed., McGrraw-Hill, New York, 1995.
Rostaing, P.; Berder, O.; Burel, G. & Collin, L. (2002). Minimum BER diagonal precoder for
MIMO digital transmissions, Elsevier Signal Processing, vol. 82, no. 10, pp. 1477-1480,
2002.
186 Recent Advances in Wireless Communications and Networks

Somekh, O.; Zaidel, B. & Shamai, S. (2007). Sum rate characterization of joint multiple cell-
site processing, IEEE Transaction of Information Theory, vol. 53, no. 12, pp. 4473- 4497,
2007.
Third Generation Partnership Project, (2007). LTE Physical Layer - General Description, no
3. 3GPP TS 36.201 v8.1.0, 2007. Available from: http://www.3gpp.org/LTE.
Uppala, L. S. & Li, J. (2004). Designing a mobile broadband wireless access network, IEEE
Signal Processing Magazine, vol. 21, no. 5, pp. 20-28, 2004.
Zhang, J.; Chen, R.; Andrews, J. G.; Ghosh, A. & Heath Jr., R. W. (2009) Networked MIMO
with Clustered Linear Precoding, IEEE Transaction on Wireless Communication, vol.
8, no. 4, pp. 1910-1921, 2009.
Zhang, R. (2010). Cooperative multi-cell block diagonalization with per-base-station power
constraints, in Proceeding of IEEE WCNC, 2010.
 Part 2

Upper Layers
 9

Joint Call Admission Control in Integrated

Wireless LAN and 3G Cellular Networks
Chunming Liu, Chi Zhou, Niki Pissinou and S. Kami Makki
T-Mobile, Illinois Institute of Technology, Florida International University,
Lamar University
U.S.A.

1. Introduction
The fourth-generation (4G) (Liu, 2004) system is expected to support fully integrated
services and ubiquitous access anytime and anywhere. Instead of developing a new uniform
standard for all wireless communication systems, some endeavors in 4G research focus on
the seamless integration of various existing wireless communication networks, such as
integrated Wireless LAN (WLAN) and the third-generation (3G) cellular networks.
3G cellular networks provide wide coverage and universal roaming services with limited
data rate up to 2 Mbps (Liu, 2006, 2007). With careful network planning and mature
admission control algorithms, the achievable Quality of Service (QoS) level of 3G cellular
networks is relatively high. On the other hand, WLANs provide low-cost, high data rate
wireless access within limited hotspot-area. Since WLAN is originally designed for best-
effort data services with contention-based access, it is difficult to achieve strict QoS
provisioning for real-time services, such as voice service (Song et al., 2006).
Due to different network capacities, user mobile patterns, vertical handoffs, and QoS levels,
the integrated WLAN and 3G cellular networks require a new call admission control scheme
to provide QoS provisioning and efficient resource utilization. Currently there are three major
architectures for internetworking between 3G cellular cellular networks and WLAN
(Ahmavaara et al., 2003). But they are all lack of joint resource management and admission
control schemes in integrated environment. Previous research work on admission control in
homogeneous cellular networks and heterogeneous integrated networks are investigated with
technical descriptions on their pros and cons. It is shown that more endeavors are needed on
joint congestion control, load balance, and high-level QoS provisioning in integrated
networks.
In this chapter, a novel joint call admission control (CAC) scheme is proposed to support
both voice and data services with QoS provisioning. Due to different network service
characteristics, 3G cellular network is defined to be a voice-priority network where voice
services have higher priority for resource allocation than data services, while WLAN is
defined as data-priority network where data services have higher priority than voice services.
A joint call admission policy is derived to support heterogeneous network architecture,
service types, QoS levels, and user mobility characteristics. Furthermore, to relieve traffic
congestion in cellular networks, an optimal channel searching and replacement algorithm
and related passive handoff techniques are further developed to balance total system traffic
190 Recent Advances in Wireless Communications and Networks

between WLAN and 3G cellular network, as well as to reduce average system QoS cost,
such as system blocking probability. A one-dimensional Markov model for voice service is
also developed to analyze interworking system performance metrics. Both theoretical analysis
and simulation results show that average system QoS costs, such as overall blocking and
dropping probabilities, are reduced, and our scheme outperforms both traditional disjoint
static CAC scheme and joint CAC without optimization.

2. Technical background
This section briefly describes concepts, architecture and vertical handoffs in integrated
WLAN and cellular networks.

2.1 Architecture of integrated WLAN and 3G cellular networks

Driven by the anywhere and anytime mobile service concept, it is expected that 4G wireless
networks will be heterogeneous, integrating different networks to provide seamless Internet
access for mobile users. The integrated WLAN and 3G cellular network takes advantage of
the wide coverage and almost universal roaming support of 3G cellular networks and the
high data rates of WLANs.
Currently, there are three major architectures for internetworking between 3G Universal
Mobile Telecommunications System (UMTS) cellular networks and IEEE 802.11 WLAN.
These are Open Coupling, Tight Coupling, and Loose Coupling (Liu, 2006). The Open
Coupling architecture specifies an open standard and is used for access and roaming
between 802.11 WLAN and UMTS networks. In this approach, both networks are
considered as two independent systems that may share a single billing scheme between
them. An 802.11 WLAN is connected to the Internet through a Gateway Router, and UMTS
network, is connected to the Internet through a Gateway GPRS Support Node (GGSN).
Open Coupling scheme is lack of supports for mobility, resource management, QoS
provisioning, and security in integrated environment.
As a direct integration scheme, Tight Coupling connects the WLAN network to the rest of
the core network in the same manner as other cellular radio access technologies (Liu &
Zhou, 2005a, 2005b; Liu, 2006). As shown in Fig. 1, the WLAN gateway router hides the
details of the WLAN from the 3G UMTS core network by adding a new component, SGSN
emulator, into WLAN. The SGSN emulator connects the gateway router in the WLAN to the
IP core network. It interconnects the UMTS core network at the Gn interface (Liu, 2006), and
implements all UMTS protocols required in a 3G radio access network. In terms of UMTS
protocols, the WLAN service area works like another SGSN coverage area to the UMTS core
network. As a result, all the traffics, including data and UMTS signaling, generated in the
WLAN are injected directly into the UMTS core network through the SGSN emulator. This
increases the traffic load of the UMTS core network. If the operators of the WLAN are
different from those of UMTS network, the new interface between the UMTS and the WLAN
can cause security weaknesses. In addition, the WLAN cards in client devices must
incorporate the UMTS protocol stack, and Universal Subscriber Identity Module (USIM)
authentication mechanism must be used for authentication in the WLAN (Liu & Zhou,
2005a).
In contrast to high cost of Tight Coupling, the Loose Coupling is an IP-based mechanism,
and approach separates the data paths in the 802.11 WLAN and 3G cellular networks (Liu,
2006). The 802.11 WLAN gateway routers connect to the Internet, and all data traffic is
Joint Call Admission Control in Integrated Wireless LAN and 3G Cellular Networks 191

Internet

GGSN
Loose Coupling
UMTS Core
Nework
Tight Coupling
SGSN

RNC
Cell SA

GR: Gateway router WLAN GR

SA: Service Agent NodeB
AP: Access point AP
SGSN: Serving GPRS Suport Node AP AP
GGSN: Gateway GPRS Suport
Node

Fig. 1. Tight coupling and loose coupling

routed to the core Internet, instead of to the cellular core network. To the core network of
the UMTS for example, the 802.11 WLAN appears like a visiting network. The gateway of
the 802.11 WLAN can be connected to a Service Agent (SA), a combined SGSN/GGSN
emulator, which provides not only internetworking protocol for signaling between the
802.11 WLAN and the UMTS 3G core network, but also an interface for data traffics between
the WLAN and IP networks. If the 802.11 WLAN is deployed by the same UMTS operator,
the SA may interface directly to the UMTS Core Network for signaling. Otherwise, the SA is
interfaced to the IP network for both signaling and data traffic. Compared to open or tight
coupling architectures, loose coupling implements the independent deployment and traffic
configuration of both the 802.11 WLAN and UMTS networks. In addition, loose coupling
architecture allows a mobile operator to provide its own private 802.11 WLAN “hotspots”
and interoperate with public 802.11 WLANs and UMTS operators via internetworking
agreements. So generally speaking, loose coupling is most preferable for integrated
WLAN/Cellular network, due to the simplicity and less reconfiguration work.
Though promising, loose coupling have several technical open issues to be addressed before
successful integration, such as integrated location management, seamless vertical handoff,
common QoS provisioning, unified Authentication, Authorization and Accounting (AAA),
joint call admission control and so on. As a part of resouce management, joint call admission
control tightly interacts with vertical handoff and QoS provisioning schemes in integrated
WLAN and 3G cellular networks.
192 Recent Advances in Wireless Communications and Networks

2.2 Vertical handoff

In integrated networks, there are two types of handoff: intra-technology handoff and inter-
technology handoff (Lampropoulos et al., 2005; Shafiee et al., 2011). The intra-technology
handoff is traditional Horizontal Handoff (HHO) in which mobile terminals handoff
between two adjacent base stations or access points using same access technology. In
contrast, inter-technology handoff is called Vertical Handoff (VHO), and happens when
mobile terminals roam between two networks with different access technologies, for
example, between WLAN and 3G UMTS network.

Fig. 2. Handoffs in integrated WLAN and UMTS cellular networks

Vertical handoffs in integrated WLAN / UMTS networks have two scenarios: a mobile
terminal moves out of a WLAN to a UMTS cellular network, and moves from UMTS cellular
network into a WLAN. Considering different service coverage area, the vertical handoff
from WLAN to Cellular network is normally triggered by signal fading when a user moves
out of the service area of the WLAN. However, the vertical handoff from cellular network to
WLAN is regarded as a network selection process, because mobile terminals are in a
wireless overlay area where both cellular access and WLAN access are available to mobile
terminals at same time.
Seamless vertical handoffs face challenges caused by the gap between different QoS levels in
cellular network and in WLAN (Liu, 2006; Shafiee et al., 2011): UMTS cellular networks
provide wide coverage with high QoS provisioning for voice service, but limited-rate data
service. However, WLANs support high-rate data service, but lack of universal roaming
ability and suffer from low QoS level for voice service, due to their original real-time
constraints. Furthermore, call admission control has been implemented in cellular network
to ensure low call dropping probability in system by assigning voice horizontal handoffs
with a higher priority for resource than new voice and data call requests, while WLANs
only support coarse packet-level access without considering handoffs priorities. So in
Joint Call Admission Control in Integrated Wireless LAN and 3G Cellular Networks 193

integrated WLAN and 3G cellular networks, seamless vertical handoffs and call admission
control must be considered as dependent and joint mechanisms to ensure both high-level
call service quality and efficient resource utilization in interworking environment.

3. Call admission control and previous work

In communication system, the call admission control scheme is a provisioning strategy for
QoS provisioning and network congestion reduction (Ahmed, 2005). Arriving calls are
granted or denied based on predefined system criteria. Due to limited spectrum resource
and growing popularity of usage in wireless cellular networks, CAC has been receiving a lot
of attentions for QoS provisioning, and its main features are extended to cover signal
quality, blocking probability of new call, handoff dropping probability, data rate, etc. The
next-generation integrated WLAN and 3G cellular networks pose a great challenge to the
CAC design due to heterogeneous network features, such as varied access techniques,
resource allocation priorities, QoS provisioning levels, vertical handoffs, etc.

3.1 Call admission control in cellular networks

Extensive research work has been done on the CAC schemes in homogeneous cellular
networks (Ahmed, 2005). They can be classified based on various design focuses and
algorithms, and each algorithm has its own advantages and disadvantages. Generally, CAC
in 3G cellular networks give higher priority for voice service than data services for resource
allocation, and higher priority for handoff calls than new call requests. We classify previous
work on CAC into five major categories: signal quality based CAC, guard channel
reservation based schemes, queuing methods, QoS estimation methods, and bandwidth
degradation approaches.
Signal quality based CAC: signal quality in the physical layer is used as a criterion of
admission control (Ahmed, 2005; Liu & Zarki, 1994). Some research work use power level of
received signals or signal-to-noise-ratio (SIR) threshold as call admission requirements (Liu
& Zarki, 1994). An optimal CAC scheme is proposed to minimize the blocking probability
while keeping a good signal quality to reduce the packet error (Ahmed, 2005). However, all
the above schemes only check the signal characteristics in the physical layer without
considering technical features in other layers and service priorities. Furthermore, there are
different criteria for the measurement of signal quality in integrated networks. So it is
difficult for implement a CAC in an interworking environment based on a uniform criterion.
Guard channel reservation based schemes: To prioritize handoff calls over new calls, a
number of channels, guard channels, in each cell are reserved for exclusive use by handoff
calls, while the remaining channels are shared by both new calls and handoff calls. To
decrease the handoff call dropping probability, which is at the cost of increasing the new call
blocking probability, the guard channel must be chosen carefully and dynamically adjusted
so that the dropping probability of handoff call is minimized and the network can support
as many new call requests as possible (Fang & Zhang, 2002; Ahmed, 2005). However, the
intensities of new call requests and handoff requests are time-variant, and it is difficult to
assign appropriate guard channel timely. So the guard channel will reduce the efficiency of
system resource utilization, and may not be suitable for heterogeneous network
environment.
Queuing methods: When there is no channel for incoming call requests, either handoff call
requests are put into a queue while new call requests are blocked, or new call requests are
194 Recent Advances in Wireless Communications and Networks

put into a queue while handoff calls are dropped (Lau & Maric, 1998; Ahmed, 2005).
Although queuing schemes can avoid high blocking probability or dropping probability due
to increased call intensity for a short period, it is not realistic in a practical system in which a
handoff call may not hold in a queue for a long time because of fast signal fading, and new
calls will leave the queue system due to users’ impatience.

Guard Channels

Voice handoff
Calls
Voice New Calls Common Channels

Data Calls

Channel Pool

Fig. 3. Guard channel reservation based scheme for voice and data services
QoS estimation based approaches: CAC in cellular networks calculates the future resource
requirements for new calls and handoff calls based on user mobility and call intensity
estimation (Zhao et al., 2003; Koodli & Puuskari, 2001). A weighted overall handoff failure
probability for all cells is calculated as an indicator for long-term statistics of successful call
completion. The suggested schemes take the overall weighted handoff failure probability as
the criterion for new call admission. Although those schemes can improve the efficiency of
admission control and resource utilization, they cause nontrivial calculation complexity, and
too many real-time control messages among cells may incur large signaling traffic and
communication overhead. Furthermore, rough estimation techniques used in these schemes
may cause erroneous decisions for call requests in a real world scenario, which will
deteriorate the QoS level in the system.
Bandwidth degradation CAC: Some methods are proposed to degrade some connections
adaptively when there are no more resources available for incoming new calls or handoff
calls. For example, longest calls in the system are degraded to free resource for handoff calls
(Jia & Mermelstein, 1996). Another proposal includes an algorithm in which each admitted
connection degrades to a lower bandwidth level according to weights (Ahmed, 2005). Other
proposals reduce the bandwidth of latest admitted connections. However, bandwidth
Joint Call Admission Control in Integrated Wireless LAN and 3G Cellular Networks 195

degradation can only reduce the bandwidth of varied-bit-rate (VBR) and non-real-time
(NRT) services for each individual, and is not suitable for constant-bit-rate (CBR)
connections. Furthermore, though these schemes can reduce the blocking probability, the
QoS level in the network cannot avoid deteriorating after degradation, and the overall
utilization ratio may not be improved.

3.2 Call admission control in integrated WLAN and 3G cellular networks

There have been some works on call admission control in integrated WLAN and 3G cellulr
networks. Most significant ones are WLAN-first approaches, mobility based algorithms and
policy based CAC schemes.
WLAN-first approaches: If mobile terminals locate in a WLAN service area, both new voice
and data calls first request admission to the WLAN. If rejected, the calls overflow to 3G
cellular network. If mobile terminals with on-going voice and data calls move into the
WLAN, the calls always try to handoff to WLAN (Song et al., 2006; Song et al., 2007a). This
unconidional preference to WLAN aims to take advantage of cheaper and higher bandwidth
in WLAN, compared to 3G cellular network. However, these approaches may cause an
over-crowded traffic situation in WLAN, without load balance in both networks. Since user
mobility is not covered in these approaches, frequent handoff requests will happen around
the WLAN boundary, which may cause “Ping-Pong“ effects for too many vertical handoffs
with extra large sigaling traffics generated into networks.
Mobility based algorithms: Some research works consider users with different mobility
speeds and apply different CAC and vertical handoff algorithms for them. Some authors
probabilistically reject vertical handoff requests to WLAN for highly mobile cellular users
(Lampropoulos et al., 2005; Klein & Han, 2004), to reduce unnecessary handoffs. In this
scheme, the processing load and new call blocking probability can be reduced while
maintaining reasonable throughput in the WLAN. A mobility-based predictive call
admission control technique has been proposed for the 4G wireless heterogeneous networks
(Rashad, 2006). In this scheme, local and global mobility profiles for the mobile terminals are
generated and used for call admission decision. However, since randomness of user
mobility, it may be difficult for such algorithms to getting speed estimation timely and
concisely. Besides handoff management based on mobility information, more works are
needed for considering service differentiations, QoS cost, and user preference, to provide
global optimization for resource utilization in integrated networks.
Policy Based CAC Schemes: Some solutions follow policy framework defined by IETF, and
combine call admission control and vertical handoff management together. They use a
mobile-assisted scheme, in which system functionality is controlled by a network policy
engine and a mobile policy engine (Zhuang et al., 2003). As shown in Fig. 4, a pairing of a
policy decision point (PDP) and policy enforcement point (PEP) exist in both engines, along
with policy repositories. PEP is responsible for the execution of a policy that is decided by
PDP, and the policy repositories define the policies that must be followed for a proper
handover decision (Zhuang et al., 2003; Guerrero & Barba 2008). In the call admission
control procedure, PEPs in the mobile terminals consult a PDP residing at the network for
available resources. The PDP will make a decision on call admission, based on network
capacities, QoS level, call types, user preferences as well as estimations on current network
load and performances. This approach gives flexibility to the terminal and the network to
make the best possible handover decision, and implements load balance. However, there are
196 Recent Advances in Wireless Communications and Networks

several drawbacks of this policy method, such as high latencies to fetch context information
during the candidate access point classification procedure, and no optimization policy is
defined for resource allocation in integrated networks.

Nework health
t
monitor
Network policy engine

Network capacities
BER Network selector
Enforce PDP
Call blocking probabilities decision
Call dropping probability Network Policy

PDP PEP
Policy repository

Mobility policy engine

User preferences Mobility
t
QoS levels Management Enforce PDP
Access Cost decision
Network Measurement t Policy
Mobility
Policy repository PDP PEP

Fig. 4. Policy-based admission control framework

4. A novel joint call admission control scheme

In this section, a novel joint call admission control scheme is proposed for both voice and
data services with QoS provisioning in integrated WLAN and 3G UMTS cellular networks.
Due to different network service characteristics, UMTS network is defined to be a Voice-
Priority network where voice services have higher priority for resource allocation than data
services, while WLAN is defined as Data-Priority network where data services have higher
priority than voice services. Instead of using WLAN-first schemes, a novel joint call
admission policy is derived to support heterogeneous network architecture, service types,
QoS levels, mobility characteristics, and user network preferences. Based on the policy, a
channel searching and replacement algorithm (CSR) is designed to relieve traffic congestion
in UMTS cellular network. CSR searches idle channels in WLAN and replacement channels
among mobile terminals based on their location and passive vertical handoffs, and therefore
implements load balance between UMTS networks and WLAN. A one-dimension Markov
model is further developed to compare system performance metrics, such as new call
blocking probability and dropping probability of voice handoff, between the proposed
algorithm and normal disjoint guard channel reservation based scheme.
The CSR algorithm is further improved by considering congestion scenarios in both the
WLAN and UMTS cellular networks. Specifically, a system cost function is derived and
Joint Call Admission Control in Integrated Wireless LAN and 3G Cellular Networks 197

minimized by admitting passive vertical handoffs with a probability, and it is proven that
there exists at least one optimal value for the target passive handoff probability. In this way,
the total traffic is balanced in the interworking environment as well as the resource
utilization is optimized. A linear programming solution is proposed for searching the
optimal admission probability for passive vertical handoff, with performance comparsion to
the traditional disjoint guard channel CAC. Numerical and simulation results demonstrate
that the optimal CSR (oCSR), outperforms disjoint guard channel CAC and original CSR
algorithms in both QoS provisioning and system resource utilization.

4.1 System model

Our system model is based on the loose coupling achitecture in which WLAN is connected
to cellular networks through Internet. All traffics in WLAN are routed to Internet through
gateway routers. Since the coverage area of a UMTS cell is normally much larger than
WLAN area, the cellular cell is called a “macro-cell,” while the WLAN region is regarded as
a “micro-cell” inside (Liu et al., 2007). The overlaid service area between the “macro-cell”
and the “micro-cell” provides mobile terminals with opportunities to connect to either
UMTS network or the WLAN, as shown in Fig. 5.

Phone A NodeB

Phone B
AP
UMTS Cell Phone C Phone E
VH_U
Phone D
VH_W
WLAN

VH_W: Vertical handoff from UMTS cellular network to WLAN

VH_U: Vertical handoff from WLAN to UMTS cellular network

Fig. 5. System model for integrated UMTS cell and WLAN

There are one class of voice service and N classes of data service considered in the integrated
networks. Since voice is sensitive to the transmission delay, the QoS requirement for voice
198 Recent Advances in Wireless Communications and Networks

service is represented as an average voice packet generation rate bv and the maximum
tolerable transmission delay Dv. On the other hand, QoS constraint for data service i is
mainly defined as the minimum throughput Ti, since data services are assumed as best
effort services and not sensitive to the transmission delay.
With careful network planning and mature admission control algorithms, the UMTS
network is assumed to support both voice and data services with QoS provisioning.
However, due to limited bandwidth and high cost per bit, voice calls are assigned with
higher priority than data services in cellular networks, and data services are regarded as
best-effort services. So we set the proportion Rc (0< Rc < 1) as the maximum ratio of
resource usage assigned to all data services in the cellular network.
On the other hand, the traditional WLAN is designed to support only best-effort data
services. To match the QoS difference between the cellular network and the traditional
WLAN, we assume that the IEEE 802.11e WLAN standard is adopted and our polling
algorithm named IIT (Liu & Zhou, 2006) is used to guarantee bandwidth and delay
requirements for voice services in WLAN while enhanced distributed coordination function
(EDCF) to support data services. Since normally mobile users connect to WLAN for
high-speed data transmission with low cost, we assume that data services in WLAN have
higher priority than voice services. Compared to cellular network, the proportion Rw (0 <
Rw < 1), is set as the maximum ratio of resource usage assigned to all voice services in the
WLAN.
Based on user mobility characteristics, voice call requests in a UMTS cell can be classified
into the following two categories: vertical handoffs from the overlaid WLAN to the UMTS,
which is denoted as VH_U, and new calls from this UMTS cell. Since a WLAN network is
assumed to be overlaid by a UMTS cell, there are also two categories in admission requests:
new call requests for WLAN access, and vertical handoffs from the surrounding UMTS cell,
denoted as VH_W, as shown in Fig. 5. To reduce complexity, the horizontal handoffs from
nearby cellular cells to the target cell are assumed to be equal to horizontal handoffs out of
the cell to neighbor cells so that we do not consider horizontal handoffs in this work. We
further assume that QoS level in the integrated networks are mainly bottlenecked by the
total bandwidth in the integrated networks. Therefore, the QoS performance considered in
this work includes blocking probability of new call requests and dropping probability of
handoff call requests.

4.2 Call admission control flow and policy

The admission control flow for the integrated networks is classified into two sub-schemes
according to the types of call requests, as shown in Fig. 6. If the call request is a new call and
mobile terminal is out of the WLAN service area, the mobile terminal will send the new call
request to the UMTS base station directly. On the other hand, if the call is a new call and
mobile terminal is in the WLAN area, the mobile terminal will first send a new call request
to the WLAN as the first choice, since the WLAN is much less expensive per call. The new
call request will be handled by the CAC in the WLAN. If the WLAN cannot accommodate
the new call request, the request will be forwarded to the UMTS base station.
The situation becomes complex when we consider vertical handoff call requests. Since voice
service is real-time service, a user is much more sensitive to voice dropping than dropping
of data service during vertical handoff. So voice vertical handoff from WLAN to UMTS
Joint Call Admission Control in Integrated Wireless LAN and 3G Cellular Networks 199

cellular network should be assigned a higher priority than new call requests and data
vertical handoff from WLAN to UMTS, to avoid possible voice dropping in the cellular
service area. Here, we adopt guard channel scheme (Fang & Zhang, 2002) to reserve some
bandwidth for voice VH_U handoff. However, for any data VH_U handoff, no matter how
high data rate it gets in WLAN, it become best-effort service when the user move from
WLAN to cellular network. So data VH_U is assigned same priority as any new call requests
is in cellular network.

New call
handoff call
request
request

checking
service area

VH_W ?
No

In overlay Yes
area ? Yes

CAC
in UMTS CAC in
No CAC WLAN
in WLAN
CAC
in UMTS Yes
Reject ? No
Yes Reject ?
Reject ?
Yes
Reject ? No
Yes
Drop the call
No Keep original
Block the connection with
No
call Admitted in Admitted in Admitted in UMTS
WLAN UMTS WLAN
Admitted in
UMTS

(a) New call admission control flow (b) Handoff call admission control flow

Fig. 6. Call admission control flow

Furthermore, for all VH_W handoff requests from cellular to WLAN, users can still keep
original connections with the cellular network even when the handoff request is denied by
the WLAN. Since there is no real connection dropping for VH_W handoffs, both VH_W and
new calls in the WLAN are assigned the same priority in our system.
Based on the above analysis of user mobility and service characteristics, the pseudocode of
proposed admission policy for interworking system are developed in Fig. 7. We assume
total bandwidth in cellular network and WLAN as Bc and Bw, respectively. The numbers of
admitted voice call in WLAN and cellular network are denoted as Vw and Vc, respectively.
The guard channel in cellular network is set as Gc. The number of admitted data service of
class-i is denoted as Di. When WLAN receive a call request, the average throughput, ti, for
current WLAN can be calculated based on enhanced DCF mode method (Liu & Zhou,
2006).
200 Recent Advances in Wireless Communications and Networks

Fig. 7. Call admission control policy

4.3 Channel searching and replacement (CSR) algorithm

Although the above proposed CAC can handle call requests in both WLAN and cellular
networks, all admission decisions are made based on the situation of each individual
network. To improve the whole system performance, we propose a channel searching and
replacement (CSR) algorithm based on passive vertcial handoff to implement joint resource
management.
Due to different capacities and user densities, the traffic intensities and QoS levels are often
unbalanced in the WLAN and overlaid cellular network. When WLAN becomes congested,
the traffic will be routed to the cellular network automatically. On the other hand, when the
3G cellular network has no resource available for an incoming call requests, our CSR
algorithm is used to find available resources in the WLAN by switching some 3G
Joint Call Admission Control in Integrated Wireless LAN and 3G Cellular Networks 201

connections staying in WLAN area to the WLAN, as shown in Figure 8. Specifically, if there
exists an ongoing cellular connection and the mobile terminal residing in the WLAN area,
and there is still bandwidth available in the WLAN at the same time, the cellular connection
will be switched to the WLAN by vertical handoff, and then the incoming call request will
take the released bandwidth in cellular network to avoid being blocked or dropped. This
kind of vertical handoff is called “passive“ because it is initiated by the system resource
management instead of by users or signal fading.
To achieve the fairness among different service connections, CSR checks the difference of
QoS provisioning in both networks before switching a cellular connection to WLAN. If there
is no QoS degradation during switching and WLAN can guarantee QoS provisioning for all
existing ongoing calls, then the bandwidth or channel is released.
Considering the CSR algorithm may increase the blocking probability in the WLAN (i.e.,
deteriorate the QoS in the WLAN by forwarding more traffics from the cellular network to
WLAN). We further assume that there is a call admission probability for passive vertical
handoff, which is determined by the system status of cellular network and WLAN, and QoS
levels. The pseudocode of the CSR is shown in Fig. 8.

switch (call request in cellular network)

case (data-call-arrival):
if (CAC for data::admitted) & (QoS provisioning )
admit the call;
else if (Channel_Searching() == 1) & (No degradation)
switch the cellular connection to WLAN;
admit the call request & assign a channel with a probability P;
else { reject the call request ;}
break;
case (voice-call-arrival):
if (CAC for voice::admitted) & (QoS provisioning )
admit the call ;
else if (Channel_Searching() == 1) & (No degradation)
switch the cellular connection to WLAN;
admit the call request & assign a channel with a probability P;
else { reject the call request ; }
break;
default: break;
end
------------------------------------------------------------------------
#Channel_searching() :
Search for cellular connections but mobile terminal staying in WLAN;
if (at least one cellular connection in WLAN) & (QoS provisioning in
WLAN ) { return 1; }
else { return 0; }

Fig. 8. Channel searching and replacement (CSR) algorithm

4.4 Analysis and comparsion

In this section, the proposed CSR algorithm is compared with traditional disjoint guard
channel (DGC) scheme with system performance metrics, including new call blocking
202 Recent Advances in Wireless Communications and Networks

prabability and handoff dropping probability. To reduce the complexity, we focus on voice
services in the integrated WLAN and 3G UMTS cellular networks, with fixed total channels
in UMTS cell and bandwidth in WLAN.

4.4.1 DGC algorithm

First the traditional DGC algorithm is considered. Assume that the arrival process for both
new calls and vertical handoff follows Poisson distributions, and the channel holding time
for both vertical handoffs and new calls are exponentially distributed. Let λn and 1 / μn
denote the arrival rate and the average channel holding time for new voice call in the UMTS
cell, respectively. Let λv and 1 / μ v denote the the arrival rate and average channel holding
time for voice vertical handoff from WLAN to UMTS cell, respectively. The arrivals of new
calls and vertical handoffs are independent of each other. To simplify, assume the avarage
channel holding time for both new voice call and handoff call are same: μn = μ v .
Assume total C available channels in UMTS cellular network for voice service. An
approximate one-dimension Markov model (Fang & Zhang, 2002; Liu et al., 2007) is derived
to present state transitions in UMTS network, as shown in Fig. 9(a). The state space in
cellular network can be denoted as {(m , n)| 0 ≤ m + n ≤ C} , where m and n are the numbers of
admitted new calls and admitted vertical handoffs in the cell, respectively. The traffic
intensity of vertical handoffs ωv and traffic intensity of new calls ωn are specified as
ωv = λv μ v and ωn = λn μn , respectively.
Based on the stationary state distribution, the vertical handoff dropping probability Pv and
new call blocking probability Pn , for disjoint guard channel scheme can be expressed as
follows,

(ω n + ω v )G ⋅ (ω v ) C −G
Pv = π c (C ) = C!
(1)
G
(ω n + ω v ) i C
(ω n + ω v )G (ω v ) i−G
∑i =0 i!
+ ∑
i =G +1 i!

C
(ω n + ωv )G ⋅ (ω v ) i −G
C ∑ i!
Pn = ∑ π c (i ) = i =G
(2)
G
(ω n + ω v ) i C
(ω n + ωv )G (ω v ) i −G
∑ ∑
i =G
+
i =0 i! i =G +1 i!
where π c ( i ) represents the stationary state of occupied channel i. The detailed derivations
for above equations are shown in our previous work (Liu & Zhou, 2007).

4.4.2 CSR algorithm

In the proposed CSR scheme, the total number of occupied channels in the cell and the idle
channels in the WLAN are the keys to deciding whether a new voice calls or a vertical
handoffs need intersystem channel switching through a passive handoff to the WLAN.
When the total channel number i in the cell is larger than Gc, an incoming new call request
can get admission if there is an ongoing cellular connection residing the WLAN and there is
still bandwidth available in the WLAN. When the total occupied UMTS channel number
Joint Call Admission Control in Integrated Wireless LAN and 3G Cellular Networks 203

equals to C, an incoming vertical handoff from WLAN can also be admitted in cellular
network if there is a successful channel replacement in the WLAN. To avoid over-utlization
on WLAN, it is assumed that a call request can get admission with probability δ that is
determined by the total number of occupied channels in the cell, the probability for mobile
terminals using ongoing cellular connection while located in the WLAN, and the state of
current occupied channels in the WLAN. Based on the above descriptions, we can get a
Markov chain model for the cellular network, shown in Fig 9(b).
Using CSR, call request blocking or dropping in a cellular network will happen in following
two scenarios:
Scenario 1: There is no idle channel available in cellular network, and no cellular
connections residing in the WLAN;
Scenario 2: There is no idle channel available in cellular network, and no channel within the
WLAN, although there is a cellular connection residing in the WLAN.
So Let Pf be the probability of an ongoing cellular call remaining in a WLAN, which is
assumed to be determined by a user’s preference for vertical handoff and mobility velocity.
Let ψ c (i ) be the probability that there is no cellular connection within the WLAN when the
number of total occupied channels in the cellular network is i.

⎛i⎞
ψ c ( i ) = ⎜ ⎟ ⋅ ( p f )0 ⋅ (1 − p f )i (3)
⎝0⎠
If the probability for finding a cellular connection staying in the WLAN is set as 1, which
means always finding available cellular connection successfully, the traffic intensity in the
WLAN depends on not only original traffic inside, but also on passive handoffs from the
cell. So the traffic intensity ρ (i ) in the WLAN is a function of state i in UMTS cell and can be
expressed as,

ρ (i ) = I 1 (i ) ⋅ ( ρn + ρ v ) + I 2 (i ) ⋅ ( ρn + ρ v + ωn ) + I 3 (i ) ⋅ ( ρn + ρ v + ωn + ωv ) (4)

where ρ n is original traffic intensity of new call requests in WLAN, ρ v is original call
intensity of vertical handoff requests from UMTS to WLAN. Ii() are state indicator functions:
I 1 (i ) equals to 1 when state i smaller than guard channel Gc, otherwise equals to zero.
I 2 (i ) equals to 1 when state i larger than Gc-1 and smaller than total channels C in UMTS
cell, otherwise equals to zero. I 3 (i ) equals to 1 when state i equals to total channels C in
UMTS cell, otherwise equals to zero.
Since in WLAN vertical handoffs and new calls are assigned with same priorities for
resource, the blocking probability of new call is same to dropping probability of vertical
handoffs. Considering voice service, the blocking probability pbw in WLAN is determined
by incoming traffic intensity ρ (i ) , which is affected by traffic intensities in both UMTS cell
and WLAN, the probability of an ongoing cellular call remaining in a WLAN, as well as
admission probability of passive handoffs.
According to above definitions of the two scenarios, the blocking probability for new call
requests and dropping probability for vertical handoffs from WLAN to cellular network can
be approximated as,
C
{ }
Pn = ∑ ψ c (i ) + [ 1 − ψ c (i )] ⋅ pbw (i ) ⋅ π c ( i )
i =G
(5)
204 Recent Advances in Wireless Communications and Networks

{ }
Pv = ψ c (C ) + [ 1 − ψ c (C )] ⋅ pbw (C ) ⋅ π c (C ) (6)

where π c ( i ) represents the stationary state of occupied channel i in UMTS cell.

Since probability that there is no cellular connection within the WLAN is alway smaller than
1, and same for blocking probability pbw in WLAN, it is proved (Liu & Zhou, 2007) that
value of blocking probability for new call requests and dropping probability of vertical
handoffs in UMTS cell through CSR algorithm are both smaller than the probability values
using disjoint guard channels shown in equations (1) and (2).

Notations:

ωn : Traffic intensity of new voice calls in UMTS cellular network

ω v : Traffic intensity of voice vertical handoff from WLAN to UMTS cellular network
Gc : Guard channels in UMTS cellular network

ωn + ωv ω n + ωv ω n + ωv ωv ωv

0 1 Gc C
1 Gc C

(a) State -transition model for Disjoint Guard Channel scheme in UMTS

δ ⋅ ωn δ ⋅ ωn
ω n + ωv ω n + ωv ω n + ωv ωv ωv ωv
δ ⋅ (ωn + ωv )

0 1 G G+1 C
1 2 G
G+1 G+2 C
(b) State -transition model for Channel Searching and Exchange scheme in UMTS

Fig. 9. State-transition diagram for DGC and CSR algorithms

4.5 Optimization on joint call admission control

Although the blocking probability of new calls and dropping probability of handoff calls in
UMTS cellular network get reduced by using CSR algorithm, the cost is load balance traffics
to WLAN and therefore may deteriorate QoS in WLAN, such as increasing blocking
probability in WLAN. So the joint call admission control needs to be optimized to achieve
the minimum blocking probability per Erlang in the integrated networks.
A weitghted system cost function is derived based on blocking probability, dropping
probability, call intensities, and probability of passive vertical handoffs. Our goal is to
Joint Call Admission Control in Integrated Wireless LAN and 3G Cellular Networks 205

minimize average weighted system cost with constraint on probability of passive vertical
handoffs, as shown in follows:

W1 ⋅ Pn ⋅ ωn + W2 ⋅ Pv ⋅ ωv + W3 ⋅ Pbw ⋅ ( ρn + ρ v )
Minimize Pave =
ωn + ωv + ρn + ρ v
s.t. 0 ≤ δ ≤ 1
where W1, W2, and W3 are cost weights for the blocking probability in the cellular network,
the dropping probability in cellular network, and the blocking probability in the WLAN,
respectively.
It is easy to prove that blocking probability in WLAN is a monotonically increasing
continuous function of δ , while blocking probability and dropping probability in UMTS
cell are continuous decreasing functions over δ in the interval between zero and one. So the
weighted cost function is also a continuous function over the same interval. According to
the Extreme Value Theorem, target cost function has a minimum and a maximum value
over the interval 0 ≤ δ ≤ 1 . So it is feasible to find out a optimal admission probability for
passive handoff which minimizes the integrated system cost with linear programming. Here
we should notice that there may be more than one optimal value for the admission
probability.

5. Numerical and simulation results

In this section, the performances of CSR are testified through numerical results and
simulations. Referred from (Fang & Zhang, 2002; Liu, 2006; Liu et al., 2007), the system
parameter values are shown in Table 1, and results are shown as below. We focus on voice
service and assume that the traffic intensity of data service in both WLAN and cellular
network are kept constant. The step searching method of linear programming (Liu, 2006) is
used to find the optimal admission probability for passive vertical handoff.

Bc Bw Gc bv Rc Rw pf W1 W2 W3 Ti
20 30ms 18 30kb 0.2 0.2 0.3 1.0 2.0 1.0 30kb
Table 1. System parameters
Fig. 10 shows the changes in the optimal admission probability for passive vertical handoff
as handoff intensity in the cell varies. We set new call intensity in UMTS cell ω n = 10, new
call intensity in WLAN ρn = 10, vertical handoff intensity ρ v = 5. Since the weight of
handoff dropping is larger than both the weights of blocking calls in cellular network and in
WLAN, the optimal admission probability increases quickly for W3 = 1.3 and W3 = 2.0, and
is 1 when the handoff intensity is larger than 45. In other words, the integrated system
attempts to allocate each idle resource in the WLAN to handoff in cellular network to avoid
larger system cost caused by dropping probability.
In contrast, when new call intensity ρn in the WLAN increases ( ωv is set as 5), the
admission probability for W3 = 2.0 and W3 = 1.3 is reduced to zero, but remains 1 for W3 = 1,
as shown in Figure 11. Again, it is shown that CSR can adjust the traffic intensity among the
two networks to avoid overloaded situation in the WLAN. For W3 = 1.0, since the cost for
blocking a passive handoff is no more than the costs of blocking a new call or dropping a
connection in cellular network, the passive handoff always get an admission into the WLAN.
206 Recent Advances in Wireless Communications and Networks

Optimal admission probability

W3 = 1.3
W3 = 2.0
0.8 W3 = 1.0

0.6

0.4

0.2

0
20 40 60 80 100
Handoff intensity in cellular network
Fig. 10. Optimal admission probability for passive handoff vs handoff intensity in cellular
Optimal Admission Probability

W3 = 1.3
0.8 W3 = 2.0
W3 = 1.0
0.6

0.4

0.2

0
20 40 60 80 100
New Call Intensity in WLAN

Fig. 11. Optimal admission probability for passive handoff vs new call intensity in WLAN
To validate the analytical results, simulations were performed based on the OPNET tool, an
efficient discrete event-driven simulator. Fig. 12 shows the average system cost for DGC,
CSR, and optimal CSR (oCSR), when new call intensity in UMTS, ωn , is set as 30. In this
case, the optimal admission probibility for passive handoff δ can be obtained as 0.078. DGC
has the highest system cost due to its disjoint resource allocation, while oCSR can achieve
the optimal resource allocation with minimum average system cost. Since the cost of oCSR is
less than that of CSR, original CSR in UMTS cellular network is a sub-optimal solution for
the overall resource allocation for integrated networks.
Joint Call Admission Control in Integrated Wireless LAN and 3G Cellular Networks 207

0.7
DGC
CSR

Average System Cost

0.6 oCSR

0.5

0.4

0.3

0.2

0.1

0
0 2 4 6 8 10 12
Time (seconds) x 10
4

Fig. 12. System cost of DGC, CSR, and optimal CSR

0.81
DGC
oCSR
0.80

0.79
Utilization

0.78

0.77

0.76

0.75

0.74

0.73
20 30 40 50 60
New call intensity in cellular network

Fig. 13. Utilization with new call intensity in UMTS

Similarly, Fig. 13 shows the simulation result of utilization of system resource as new call
requests ωn in cellular network increases. We can see that optimal CSR has larger resource
utilization than DGC does because optimal CSR uses idle resource in each network when
traffic intensity in a network increases.
Fig. 14 shows the blocking probability when new call intensity in cellular network increases.
When ωn equals 20, 30, 40, 50, and 60, the optimal admission probability for passive
handoffs are 0.496, 0.302, 0.216, 0.167, and 0.136, respectively. It is shown that the blocking
probability of new call of oCSR scheme is always less than in the DGC scheme, due to
optimal passive handoffs in oCSR scheme.
208 Recent Advances in Wireless Communications and Networks

1
DGC
oCSR
0.9

Blocking probability
0.8

0.7

0.6

0.5

0.4

0.3
20 30 40 50 60
New call intensity in cellular network
Fig. 14. Blocking probability with optimal CSR and DGC

0.7
DGC
oCSR
0.6
Dropping probability

0.5

0.4

0.3

0.2

0.1

0
10 20 30 40 50
Handoff call intensity
Fig. 15. Dropping probability with optimal CSR and DGC
Similarly, Fig. 15 shows the handoff dropping probability in the cell as the handoff intensity
increases. Due to limited resources in the cellular network, both dropping probabilities
increase. However, the dropping probability of the DGC is always greater than the
dropping probability of the oCSR, since some handoffs are transferred to the WLAN, except
in the case vertical handoff equals to 10. Since the optimal admission probability is equal to
zero when ωv = 10, there is no passive handoff from the cellular network to the WLAN and
both dropping probabilities are the same.

6. Conclusion
In this chapter, we introduce the next-generation call admission control schemes in
integrated WLAN / 3G cellular networks. Technical background and previous works on call
Joint Call Admission Control in Integrated Wireless LAN and 3G Cellular Networks 209

admission control in homogeneous and heterogeneous networks are investigated. Then a

novel joint call admission control scheme is proposed to support both voice and data
services with QoS provisioning in next-generation integrated WLAN / 3G UMTS networks.
A joint admission policy is first derived with considering heterogeneous network
architecture, service types, QoS levels, and user mobility characteristics. To relieve traffic
congestion in networks, a channel searching and replacement algorithm, CSR, is further
developed and optimized to balance total system traffics between WLAN and 3G cellular
network, as well as to reduce average system QoS cost. A one-dimensional Markov model
for voice traffic is further developed to analyze interworking system performance metrics.
Both theoretical analysis and simulation results show that our scheme outperforms both
traditional disjoint guard channel scheme and non-optimized joint call admission control
scheme.
Our feature work will focus on more real-time services, such as video services, and
investigate interactions between resource management and user mobility in integrated
WLAN / 3G cellular networks.

7. References
Ahmavaara, K.; Haverinen, H. & Pichna, R. (2003), Interworking Architecture between
3GPP and WLAN Systems, IEEE Communications Magazine, Vol. 41, No.11, (Nov
2003), pp. 74 – 81, ISSN 0163-6804
Ahmed, M. (2005), Call admission control in wireless networks: a comprehensive survey,
IEEE Communications Surveys & Tutorials, Vol. 7, No. 1, May 2005, pp. 50-69, ISSN
1553-877X.
Fang, Y. & Zhang, Y. (2002), Call admission control schemes and performance analysis in
wireless mobile networks, IEEE Transactions on vehicular Technology, vol. 51, No.2,
(March 2002), pp. 371-382, ISSN 0018-9545
Guerrero, J. & Barba, A. (2008), Policy-based Network Management Reference Architecture
for an Integrated Environment WLAN-3G, IEEE Latin America Transactions, Vol. 6,
No. 2, (June 2008), pp. 229-234, ISSN 1548-0992
Jia, D. & Mermelstein, P. (1996), Adaptive Traffic Admission for Integrated Services in
CDMA Wirelessaccess Networks, IEEE Journal on Selected Areas in Communications,
Vol. 14, No. 9, (Dec. 1996), pp.737–47, ISSN 0733-8716.
Klein, T. & Han, S. (2004), Assignment strategies for mobile data users in hierarchical
overlay networks: Performance of optimal and adaptive strategies, IEEE Journal on
Selected Areas in Communications, vol. 22, No. 5, (June 2004), pp. 849–861, ISSN 0733-
8716
Koodli, R. & Puuskari, M. (2001), Supporting packet-data QoS in next generation cellular
networks, IEEE Communication Magazine, Vol.39, No.2, (Feb. 2001), pp.180-188, ISSN
0163-6804
Lampropoulos, G.; Passas, N. & Merakos, L. (2005), Handover management architectures in
integrated WLAN / Cellular Networks, IEEE Communications Surveys & Tutorials,
Vol. 7, No. 4, (May 2005), pp. 30-44, ISSN 1553-877X.
Lau, K. & Maric, S. (1998), Mobility of Queued Call Requests of a new call-queueing
technique for cellular systems, IEEE Transactions on vehicular Technology, vol. 47,
no.2, (May 1998), pp.480-488, ISSN 0018-9545
Liu, C. & Zhou, C. (2004), Challenges and Solutions for Handoff Issues in 4G Wireless
Systems An Overview, Proceedings of International Latin American and Caribbean for
210 Recent Advances in Wireless Communications and Networks

Engineering and Technology 2004, Paper No. 047, Miami, Florida, USA, June 2-4,
2004.
Liu, C. & Zhou, C. (2005), HCRAS: A novel hybrid internetworking architecture between
WLAN and UMTS cellular networks, Proceedings of IEEE Consumer Communications
& Networking Conference 2005, pp. 374-379, ISBN 0-7803-8784-8, Las Vegas, Nevada,
January 1- 10, 2005.
Liu, C. & Zhou, C. (2005), An improved architecture for UMTS-WLAN Tight Coupling,
Proceedings of IEEE Wireless Communications & Networking Conference 2005, pp. 1690-
1695, ISBN 0-7803-8966-2, New Orleans, LA, March 13-17, 2005.
Liu, C. & Zhou, C. (2006), Providing Quality of Service in IEEE 802.11 WLAN, Proceedings of
IEEE Consumer Communications & Networking Conference 2005, pp. 374-379, ISBN 0-
7803-8784-8, Las Vegas, Nevada, USA, January 1-10, 2006.
Liu, C. (2006). System Design and Resource Management in the Next-Generation Integrated
WLAN / 3G Cellular Networks, Doctoral dissertation, Florida International University,
Miami, Florida, USA, August 2006.
Liu, C.; Zhou, C.; Pissinou, N. & Makki, K. (2007), Resource Management in the Next-
Generation Integrated WLAN / 3G Cellular Networks, Proceedings of IEEE Wireless
Communications & Networking Conference 2007, pp. 3343-3348, ISBN 1-4244-0658-7,
Hong Kong, China, March 11-15, 2007.
Liu, Z. & Zarki, M. (1994), SIR-based Call Admission Control for DS-CDMA Cellular
Systems, IEEE Journal on Selected Areas in Communications, Vol.12, no. 4, (May 1994),
pp. 638–44, ISSN 0733-8716
Rashad, S. (2006), Mobility-based predictive call admission control and resource reservation
for next-generation mobile communications networks, Doctoral dissertation,
University of Louisville, In ACM digital library, Available from
http://portal.acm.org/citation.cfm?id=12929
Shafiee, K. ; Attar, A. & Leung, V. (2011), Optimal Distributed Vertical Handoff Strategies in
Vehicular Heterogeneous Networks, IEEE Journal on Selected Areas in
Communications, Vol. 29 , No.3, (March 2011), pp. 534 – 544, ISSN 0733-8716
Song, W.; Jiang, H. & Zhuang, W. (2007), Performance analysis of the WLAN first scheme in
cellular/WLAN interworking, IEEE Transactions on Wireless Communications, vol. 6,
No. 5, (May 2007), pp. 1932-1943, ISSN 1536-1276
Song, W.; Jiang H.; Zhuang W. & Saleh, A. (2006), Call Admission Control for Integrated
Voice/Data Services in Cellular/WLAN Interworking, Proceedings of IEEE
International Conference on Communications 2006, PP. 5480 - 5485, ISBN 1-4244-0355-
3, Istanbul, TURKEY, June 11-15, 2006
Song W.; Zhuang W. & Cheng Y. (2007). Load balancing for cellular/WLAN integrated
networks, IEEE Network, Vol. 21, No. 1, (Feb 2007), pp. 27-33, ISSN 0890-8044.
Zhao, D.; Shen, X. & Mark, J. (2003), Radio Resource Management for Cellular CDMA
Systems Supporting Heterogeneous Services, IEEE transactions on Mobile Computing,
Vol.2, No.2, (June 2003), pp. 147 – 160, ISSN 1536-1233
Zhuang W.; Gan Y.; Loh K. & Chua K. (2003). Policy-based QoS-management architecture in
an integrated UMTS and WLAN environment, IEEE Communications Magazine, Vol.
41, No. 11, (Nov. 2003), pp. 118-123, ISSN 0163-6804
 0
10

Near-Optimal Nonlinear Forwarding Strategy for

Two-Hop MIMO Relaying
Majid Nasiri Khormuji and Mikael Skoglund
Royal Institute of Technology (KTH)
Sweden

1. Introduction
Relaying (1–3) has been considered as a paradigm for improving the quality of service (i.e.,
bit-error-rate, data rate and coverage) in wireless networks. In this work, we study a two-hop
relay channel in which each node can have multiple antennas. It is well-known that utilizing
multiple-input multiple-output (MIMO) links can significantly improve the transmission rate
(see e.g. (4; 5) and references therein). Thus, one can expect a combination of a MIMO gain and
a relaying gain in a MIMO relay link. We focus on one-shot transmission, where the channel is
used once for the transmission of one symbol representing a message. This is often referred to
as uncoded transmission. The main motivation for such a scenario is in considering applications
requiring either low-delays or limited processing complexity.
The capacity of the MIMO relay channel is studied in (6). The work in (9) establishes the
optimal linear relaying scheme when perfect CSI is available at the nodes. The work in (7; 8)
investigates linear relay processing for the MIMO relay channel. In this paper, in contrast
to (6–9), we study an uncoded system, and we propose a nonlinear relaying scheme which is
superior to linear relaying and performs close to the theoretical bound. Our proposed scheme
is based on constellation permutation (10; 11) at the relay over different streams obtained by
channel orthogonalization.
We investigate a two-hop MIMO fading Gaussian relay channel consisting of a source, a
relay and a destination. We assume that all three nodes have access to perfect channel
state information. We propose a nonlinear relaying scheme that can operate close to the
optimal performance. The proposed scheme is constructed using channel orthogonalization
by employing the singular value decomposition, and permutation mapping. We also
demonstrate that linear relaying can amount to a significant loss in the performance.

1.1 Organization
The remainder of the chapter is organized as follows. Section 2 first introduces the two-hop
relay channel model and then explains the transmission protocol and the assumptions on the
channel state information (CSI) at the nodes and finally formulates an optimization problem.
Section 3 simplifies and reformulates the optimization problem introduced in the preceding
section, by channel orthogonalization using SVD. Section 4 introduces a novel relaying
strategy in which the relay first detects the transmitted message and employs permutation
coding over different streams obtained by channel orthogonalization. This section also
212
2 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

z1 z2

x1 y1 x2 y2
w α H1 f (y1 ) H2 β ŵ

Fig. 1. Gaussian two-hop MIMO relaying.

provides some performance bounds. Section 5 finally provides some simulation results and
concludes the chapter.

2. System model and problem formulation

In this section, we first introduce the two-hop Gaussian vector relay channel in detail and
then formulate the general problem of finding an optimal relaying strategy for the underlying
channel.
We consider Gaussian two-hop communication between a source and a destination, as
illustrated in Fig. 1. The communication is assisted by a relay node located between the source
and the destination. We assume that the relay node has no own information to transmit and
its sole purpose is to forward the information received from the source to the destination. We
additionally assume that all nodes may have different number of antennas. It is assumed that
there is no direct communication between the source and the destination. (This is reasonable
when e.g., the destination is located far away from the source or there is a severe shadow
fading between the source and the destination.) The communication between the source and
the relay takes place in two phases as described in the following.
First–Hop Transmission: During the first phase, the source transmits its information and the
relay listens to the transmitted signal. The received signal vector at the relay, denoted by y1 ,
is given by
y1 = H1 x1 + z1 (1)
where H1 ∈ C [ L× M] denotes the channel between the source and the relay, x1 ∈ C [ M×1]
denotes the transmitted signal vector from the source and z1 ∈ C [ L×1] denotes the additive
circularly symmetric Gaussian noise. The signal vector x1 is the output of the modulator α
which is defined as

α : W −→ C M
x1 = α ( w )

where w ∈ W  {1, 2, 3, . . . , 2q } denotes a message to be transmitted over the channel.

Some particular choices for defining α are, for example, the 2q -QAM and 2q -PSK modulation
schemes. We assume an average power constraint at the source, such that trE {x1 x1† } ≤ P1 .
Second–Hop Transmission: During the second phase, only the relay transmits and the source
is silent. We assume that the relay uses a forwarding strategy given by the following
deterministic function

f : C L −→ C L
x2 = f (y1 )

Since the function f (·) is arbitrary, our model includes linear as well as nonlinear mappings.
We assume an average power constraint at the relay such that trE {x2 x2† } ≤ P2 . The received
Near-Optimal
Near-Optimal NonlinearNonlinear Forwarding
Forwarding Strategy Strategy
for Two-Hop MIMO Relaying for Two-Hop MIMO Relaying 2133

signal at the destination, denoted by y2 , is then given by

y2 = H2 x2 + z2 (2)

where H2 ∈ C [ N × L] denotes the channel between the relay and the destination, x2 ∈ C [ L×1]
denotes the transmitted signal vector from the relay and z2 ∈ C [ N ×1] denotes the additive
circularly symmetric Gaussian noise. Finally, the destination, upon receiving y2 , detects the
transmitted message using the function (demodulator or detector) β defined as

β : C N −→ W
ŵ = β(y2 )

where ŵ ∈ W denotes the detected message at the destination.

Channel Statistics: We assume that the entries of the channel matrices H1 and H2 are
i.i.d. Rayleigh fading, distributed according to CN (0, 1). The entries of the noise vectors
z1 and z2 are assumed to be independent zero-mean circularly symmetric Gaussian noise.
The covariance matrices of the noise vectors are given by Rz1 z1 = E [z1 z1† ] = N1 I L and
Rz2 z2 = E [z2 z2† ] = N2 I N , where I N and I M denote the identity matrices of size N and
M, respectively. Additionally, we assume that the channels stay unchanged during the
transmission of one block but they vary independently from one block to another.
Channel State Information (CSI): We assume that the source, the relay, and the destination know
H1 and H2 perfectly. The CSI of backward channels at the relay and the destination can be
obtained using training sequences and the CSI of the forward channels at the source and the
relay can be obtained either using reciprocity of the links or feedback. When the channel
matrices are constant or varying slowly, one can obtain accurate CSI at the nodes. Satellite
MIMO link and wireless LAN are two practical examples in which this model is applicable.

2.1 Problem formulation

The goal is to minimize the average message error probability. Thus for a given message set W,
we need to find the triple (α∗ , β∗ , f ∗ ) under the average power constraint such that

(α∗ , β∗ , f ∗ ) = arg min Pr{ŵ = w}. (3)

α,β, f

We desire to find a structured solution to the optimization problem in (3). Imposing structure
on a communication strategy results in loss of performance in general. On the other hand, a
structured strategy however facilitates the design. We first utilize the channel knowledge to
orthogonalize each hop using the SVD and then propose a nonlinear scheme that performs
close to the theoretical bound.

3. Channel orthogonalization via SVD

In the following, we employ the singular value decomposition (SVD) to obtain an equivalent
parallel channel for each hop. We then rewrite the optimization problem given by (3) for the
equivalent channel.
Using the SVD, any channel realizations of H1 and H2 can be written as

H1 = U1 D1 V1†
H2 = U2 D2 V2†
214
4 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Relay

z1 z2

x V1 U1 D1 V1† U1† g(·) V2 U2 D2 V2† U2† ỹ

Fig. 2. Processing using the SVD of the channel matrices.

√ √
λ11 λ21 z̃21
z̃11

ỹ11 x21
x11 ỹ21

√ .. √ ..
λ1r . z̃1r g(·) λ2t . z̃2t

ỹ1r x2t
x1r ỹ2t

Fig. 3. Equivalent parallel channel.

where U1 ∈ C [ L× L], V1 ∈ C [ M× M], U2 ∈ C [ N × N ] and V2 ∈ C [ L× L] are unitary matrices,

and D1 ∈ R [ L× M] and D2 ∈ R [ N × L] are non-negative and diagonal matrices. Note that
since U1 , V1 , U2 and V2 are invertible, linear operations of the form of AG or GA (where
G ∈ {U1 , V1 , V2 , U2 } and A is an arbitrary matrix with an appropriate size) impose no loss of
information. Thus we can preprocess the transmitted signal vectors from the source and the
relay and postprocess the received signal vectors at the relay and the destination as illustrated
in Fig. 2. Consequently, the received signal at the relay after the linear postprocessing is given
by

ỹ1 = U1† y1
= U1† H1 V1 x1 + U1† z1
= U1† U1 D1 V1† V1 x1 + U1† z1
= D1 x1 + z̃1

where the last equality follows from the identities U1† U1 = I L and V1† V1 = I M and the
definition z̃1 = U1† z1 . The random vector z̃1 ∼ CN (0, N1 I L ) since U1 is a unitary matrix. In
a similar fashion, we can obtain

ỹ2 = D2 x2 + z̃2

where z̃2  U2† z2 ∼ CN (0, N2 I M ). See also Fig. 2. Because D1 and D2 are diagonal matrices,
we have

ỹ1i = λ1i x1i + z̃1i , i ∈ {1, 2, . . . , min( M, L )}

ỹ2j = λ2j x2j + z̃2j , j ∈ {1, 2, . . . , min( L, N )}
Near-Optimal
Near-Optimal NonlinearNonlinear Forwarding
Forwarding Strategy Strategy
for Two-Hop MIMO Relaying for Two-Hop MIMO Relaying 2155
√ 
where λ1i is the ith entry on the main diagonal of D1 and λ2j is the jth entry on the main
diagonal of D2 . The equivalent channel obtained by the SVD operation is shown in Fig. 3.
The function g(·) in Fig. 3 denotes the forwarding strategy at the relay, defined as

g : Cr −→ C t
x2 = g(ỹ1 )

where r = min( M, L ) and t = min( L, N ). We consider both linear as well as nonlinear

mappings. One can thus optimize the mapping according to

(α∗ , g∗ (ỹ1 ), β∗ ) = argmin Pr{ŵ = w}. (4)

{α:trE [x1 x1† ]≤ P1 },{g(ỹ1 ) :trE [ g(ỹ1 ) g† (ỹ1 )]≤ P2},β

4. Transmission strategies and performance bounds

4.1 Lower bound on Pe
We next give a simple lower bound on the average message error probability, which we use
as a benchmark to evaluate different transmission strategies in the sequel.
Lemma 1. For the two-hop vector channel shown in Fig. 1, the average message error probability Pe
is lower bounded by
Pe ≥ max{ Pe1 , Pe2 } (5)
where Pe1 and Pe2 denote the average message error probability of the first- and the second hop,
respectively.

Proof. Consider a two-hop channel where the first hop is noise-free and the second hop is
identical to the original channel in Fig. 1. Denote the average error probability of this new
channel by P̄e . It is easy to see that Pe ≥ P̄e = Pe2 . In a similar manner we can obtain Pe ≥
P̃e = Pe1 , where P̃e denotes the error probability of a two-hop channel with identical first hop
to that in Fig. 1 and a noise-free second hop. This yields (5).

4.2 Linear relaying

One of the fundamental strategies in the literature is linear relaying, commonly known as
amplify-and-forward (AF). Using AF in our setting, the relay function is given by

x2i = gi (ỹ1i ) = κi μ i ỹ1i , i ∈ {1, ..., min{r, t}} (6)


P2
where μ i = † ]+ N is
λ1i E [ x1i x1i
a power normalization factor and 0 ≤ κi ≤ 1 is a power
1

allocation factor where ∑ti=1 κ2i = 1. Note that the number of parallel channels that can be
utilized is min{r, t}, i.e., the minimum number of parallel streams of the first- and second hop.
In (9), it is shown that the strategy given by (6) is optimal if the relay mapping is constrained
to be linear. However as we show, AF is in general suboptimal for the underlying channel.
The received signal-to-noise ratio (SNR) of the ith stream at the destination is given by

κ2i λ1i λ2i P1i P2

γiAF = (7)
N1 N2 + λ1i P1i N2 + κ2i λ2i P2 N1

where P1i  E [ x1i x1i

† ]. The fact that the received noise at the relay is forwarded to the

destination is the main drawback of AF relaying.

216
6 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Relay Destination
q
{w j }2j=1,j=i 1 − i q
{w j }2j=1,j=i
( i) i
Pe1

wi

( i)
( i) Pe2
1 − Pe1
wi wi
( i)
1 − Pe2
Fig. 4. Error probability transition in DF relaying.

Case r = 1: In order to maximize γi , one should choose the strongest mode (the stream with
largest singular value) with full power when r = 1. Note that the use of weaker streams at
the relay does not improve the performance of AF since all streams are transmitting the same
signal, thus allocating all power to the strongest mode is the optimal solution. Therefore, the
maximum possible achievable SNR for linear relaying when r = 1, is given by
λ11 λ21 P1 P2
γ ∗AF = (8)
N1 N2 + λ11 P1 N2 + λ21 P2 N1
where λ11 and λ21 are the largest eigenvalues of the first- and second hop, respectively.

4.3 Relaying via Detect-and-Forward (DF)

Another approach for forwarding the received signals is to first detect the transmitted message
and then re-modulate it. That is
x2i = gi (ỹ1 ) = κi αri (ŵˆ ) = κi αri ( βr (ỹ1 )) (9)
where ŵˆ = βr (ỹ1 ) is the detected message and βr denotes the detector at the relay. The
modulator for generating x2i is denoted by αri . We also have trE [x2 x2† ] = P2 .
The following proposition derives a simple upper bound on the average message error
probability of DF relaying.
Lemma 2. The average message error probability is upper bounded by
( i) ( i)
Pe ≤ Pe1 + Pe2 − min q Pe1 Pe2 (10)
1≤ i ≤2

( i) ( i)
where Pe1 and Pe1 respectively denote the ith message error probability of the first- and the second hop
and Pe1 and Pe1 respectively are the average message error probabilities of the first- and the second hop.

Proof. Consider the transmission of wi from the source. The relay either detects the
transmitted message correctly or declares another message. This is illustrated in Fig. 4. Using
Fig. 4, the ith message error probability can be bounded as
( i) ( i) ( i) ( i)
Pe = (1 − Pe1 ) Pe2 + Pe1 (1 − i ) (11)
( i) ( i) ( i) ( i)
≤ Pe1 + Pe2 − Pe1 Pe2 (12)
Near-Optimal
Near-Optimal NonlinearNonlinear Forwarding
Forwarding Strategy Strategy
for Two-Hop MIMO Relaying for Two-Hop MIMO Relaying 2177

q
where i denotes the detection probability of wi at the destination when {w j }2j=1,j=i is
transmitted from the relay, under the constraint that the source is transmitted wi . The
inequality in (11) follows from the fact that 0 ≤ i ≤ 1. By taking average over all possible
messages, we have
2q
( i)
Pe = ∑ Pe p(wi ) (13)
i =1
2q  
( i) ( i) ( i) ( i)
≤ ∑ Pe1 + Pe2 − Pe1 Pe2 p(wi ) (14)
i =1
2q
( i) ( i)
= Pe1 + Pe2 − ∑ Pe 1
Pe2 p(wi ) (15)
i =1
  2q
( i) ( i)
≤ Pe1 + Pe2 − min q Pe1 Pe2
1≤ i ≤2
∑ p(wi ) (16)
i =1
( i) ( i)
= Pe1 + Pe2 − min q Pe1 Pe2 (17)
1≤ i ≤2

This completes the proof.

Proposition 1. DF relaying achieves the same performance as that of a single hop (i.e., max{ Pe1 , Pe2 })
at high SNR when N = M.

Proof. For given modulator and optimal demodulator, the error probability at the destination
is upper bounded as
 
a1 1 a2 1
Pe ≤ Pe1 + Pe2 = NL + O
DF
+ LM + O
γ1 γ1NL+1 γ2 γ2LM+1
⎧  
⎪
⎪
⎨ γaNL
1
+ O γ NL1 +1 if N < M
= 1
 1  (18)
⎪
⎪ a2
⎩ γ LM + O 1
LM + 1 if M < N
γ2
2

P1 P2
where we used Lemma 2 and γ1  N 1
, γ2  N 2
, and a1 and a2 are two constants depending
on the number of antennas and the modulation scheme.
We also have the following lower bound using Lemma 1
⎧  
⎪
⎪
⎨ γaNL
1
+O 1
NL + 1 if N < M
Pe ≥ max{ Pe1 , Pe2 } = 1  γ1  (19)
⎪
⎪
⎩ γaLM
2
+O 1
LM + 1 if M < N
γ2
2

Comparing (18) and (19), we see that the upper bound and lower bound meet each other at
high SNR. This therefore establishes the optimality of DF at high SNR.

Proposition 2. DF achieves the optimal diversity order d∗ = min{ NL, ML }.

218
8 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Proof. From Lemma 1, we conclude that d∗ ≤ min{ NL, ML }. But, from Lemma 2 we know
that PeDF ≤ Pe1 + Pe2 . Thus the diversity order is bounded as d DF ≥ min{ NL, ML }. Therefore,
DF achieves the optimal diversity order.

In the following we comment further on the conventional DF and a propose a novel DF

relaying scheme.
Conventional DF: One way to simplify the problem is to use the same modulator over all
streams. That is αri = αr for all streams. By doing so, with a similar argument as that in
the linear relaying case, the optimal power allocation would be to use all the power on the
strongest mode.
Proposition 3. Relaying using conventional DF (i.e., transmission using the strongest mode) is
optimal at high SNR when N > M.

Proof. The proof follows from the observation that using only the stream with the strongest
mode of the second hop, one can obtain higher diversity gain compared to the first hop for
any source modulator. Since M < N, we have
 
a1 1 a1 1
Pe ≤ LM + O
DF
, and Pe ≥ LM + O . (20)
γ1 γ1LM+1 γ1 γ1LM+1
This completes the proof.

Proposed DF: A more sophisticated approach at the relay is to use different modulators over
distinct streams. In the following, we propose a structured method for obtaining different
modulators based on a given modulator, say αr . Let π denote a permutation operation on a
given finite sequence. For example, if a = (1, 2, 3, 4) the operation π ( a) produces a different
ordering of the elements in the sequence, such as π1 ( a) = (4, 3, 1, 2). In the following let ᾱr
denote the list of letters produced by the modulator αr , in the default order. Now we construct
the ith modulator using ᾱr as
ᾱri = κi π i (ᾱr ) (21)
where κi is a power allocation factor used at ith stream such that {κi } meets the power
constraint trE [x2 x2† ] ≤ P2 . Thus, the transmitted signal from the relay over the ith stream
is given by
x2i = gi (ỹ1 ) = κi αri (ŵˆ ) = κi αri ( βr (ỹ1 )) (22)
Here βr denotes the detector used at the relay and the modulator αri is constructed using
the ith permutation used over the ith stream, i.e., π i . Now designing a relaying strategy
specializes to finding the optimal permutations and the power allocation factors. That is
({κi∗ }ti=1 , {π i∗ }ti=1 ) = arg min Pr{ŵ = w} (23)
{κ i }ti=1 ,{πi }ti=1

The proposed DF scheme includes conventional DF as a special case, by choosing κi = 0 for

i = 1. Thus, the error probability achieved by the proposed DF scheme is upper bounded
by that of conventional DF. The main advantage of the proposed scheme is that it enjoys a
structured design based on a given modulator. From Proposition 3, one can conclude that
this scheme does not bring any advantage at high SNR when N > M. However, in the
following section we show that the proposed DF approach can attain considerable gain over
conventional DF and linear relaying at moderate SNR’s, that is, in an SNR regime where
diversity gain is not a useful performance measure.
Near-Optimal
Near-Optimal NonlinearNonlinear Forwarding
Forwarding Strategy Strategy
for Two-Hop MIMO Relaying for Two-Hop MIMO Relaying 2199

5. Numerical results and concluding remarks

In the following we present numerical results for the case when the source has only one
single antenna and the relay and the destination have 10 antennas each. This scenario is of
importance, for example in the uplink transmission in cellular networks, where the mobile
node has only a single antenna. Under this constraint, the relay has only one incoming
stream and multiple outgoing streams (see Fig. 3). Fig. 5 shows the average message error
probability for three different relaying schemes; linear relaying, conventional DF relaying, the
proposed DF relaying approach based on permutation mappings using two streams. We use
16-QAM as the modulator and an optimal ML detector at the relay and the destination. For
the proposed scheme we use two streams in the second hop. The optimal permutation is
obtained using exhaustive search. We also plotted a lower bound on the performance for any
relaying scheme, using Lemma 1. Here we set P1 = P2 = P, N1 = N2 = 1. From Fig. 5, we
see that linear relaying performs worst, and the proposed DF relaying scheme provides the
best performance. Surprisingly, the performance of the proposed DF is very close to the lower
bound.

−1

−2
log10 ( Pe )

−3

AF
DF: 1 Stream
−4 DF: 2 Streams
Lower Bound

−5
0 3 6 9 12
P [dB]

Fig. 5. Average message error probability (Pe ) using 16-QAM modulation for different
forwarding strategies (AF, conventional DF (i.e., one stream) and proposed DF (i.e., two
streams with permuted modulations)). Here we set P1 = P2 = P, N1 = N2 = 1, number of
antennas at the source is N = 1 and number of antennas at the relay and the destination are
L = M = 10.
220
10 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

We can see from Fig. 5 that the performance of conventional DF approaches that of the
proposed scheme at high SNR. This is in accordance with Proposition 3. However, we also
see that the new scheme gives considerable gains in the low- and moderate SNR regime, and
it achieves the optimal performance at lower SNR compared to conventional DF.

6. References
[1] E. C. van der Meulen, “Three-terminal communication channels,” Adv. Appl. Probab.,
vol. 3, pp. 120-154, 1971.
[2] T. M. Cover and A. El Gamal, “Capacity theorems for the relay channel,” IEEE Trans.
Inf. Theory, vol. 25, no. 5, pp. 572-584, Sep. 1979.
[3] J. N. Laneman, G. W. Wornell, and D. N. C. Tse, “Cooperative diversity in wireless
networks: efficient protocols and outage behavior,” IEEE Trans. Inf. Theory, vol. 50, no.
12, pp. 3062-3080, Dec. 2004.
[4] E. Telatar, “Capacity of multi-antenna Gaussian channels,” Technical Memorandum, Bell
Laboratories (Published in European Transactions on Telecommunications, Vol. 10, No.6, pp.
585-595, Nov/Dec 1999), 1995.
[5] D. Palomar and S. Barbarossa, “Designing MIMO communication systems:
Constellation choice and linear transceiver design,” IEEE Trans. Signal Processing, vol.
53, no. 10, pp. 3804-3818, Oct. 2005.
[6] B. Wang, J. Zhang, and A. Høst-Madsen, “On the capacity of MIMO relay
channels,” IEEE Trans. Inf. Theory, vol. 51, no. 1, pp. 29-43, Jan. 2005.
[7] A. S. Behbahani, R. Merched, and A. M. Eltawil, “Optimization of a MIMO relay
network” IEEE Trans. on Signal Processing, vol. 56, no. 10, Oct. 2008.
[8] N. Khajehnouri and A. H. Sayed, “Distributed MMSE relay strategies for wirless sensor
networks” IEEE Trans. on Signal Processing, vol. 55, no. 7, July 2007.
[9] X. Tang and Y. Hua , “Optimal design of non-generative MIMO wireless relays” IEEE
Trans. on Wireless Communications, vol. 6, no. 4, April 2007.
[10] M. N. Khormuji and E. G. Larsson, “Improving collaborative transmit diversity using
constellation rearragment,” In proceedings IEEE WCNC, March 2007.
[11] M. N. Khormuji and E. G. Larsson, “Rate-optimized constellation rearrangement for the
relay channel,” IEEE Communication Letters, vol. 12, no. 9, pp. 618-620, Sept. 2008.
 11

Connectivity Support in
Heterogeneous Wireless Networks
Anna Maria Vegni1 and Roberto Cusani2
1University
of Roma Tre
Department of Applied Electronics, Rome;
2University of Roma “La Sapienza”

Department of Information Engineering,

Electronics and Telecommunications, Rome;
Italy

1. Introduction
Recent advances in wireless technology and decreasing costs of portable devices strongly
contributed to increase the popularity of mobile communications.
Wireless communication and device integration have lead to the so-called nomadic computing
(or mobile computing) where portable devices (such as laptop and handheld computers) allow
users to access Internet and data on their home or work computers from anywhere in the
world. Multimedia services requirements nowadays encompass not only large bandwidths,
but also on-the-move facilities. Future 4th generation wireless communications systems will
provide seamless mobility support to access heterogeneous wired and wireless networks
(Makhecha & Wandra, 2009), (Lin et al., 2010).
Emerging and pre-existing wireless technologies exhibit different characteristics, access
technologies, available services and network performances. For example GSM, UMTS,
WLAN and WiMAX have different bandwidth (70 Mbps for WiMAX and 9.6 kbps for GSM),
cell diameter ( 50 km in LoS for WiMAX and 100 m for WLAN), or handover latency (3 s for
WLAN and 50 µs for WiMAX).
The increasing demand for services with high QoS requirements and novel mobility
scenarios, like on-the-move business users, home and office networks, on-the-move
entertainment, info-mobility etc., provide users to be connected to the Internet anytime and
anywhere, as well as user services and connectivity be maintained, and kept alive. Mobility
management in heterogeneous networks is the essential support for roaming nomadic devices
switching from one access technology to another, at the same time maintaining seamless
connectivity at high QoS services (i.e. video-streaming).
New emerging multimode mobile devices are equipped with multiple wireless network
interface cards, providing Vertical Handover capability to autonomously select the best access
network. The design of innovative handover mechanisms —sometimes called as handoff—
between heterogeneous mobile devices (e.g. PDA, laptop, smart phones) and seamless
integration of different integrated network (e.g. GSM, UMTS, HSDPA, GPS, WLAN, Bluetooth
and so on) is an open research issue.
222 Recent Advances in Wireless Communications and Networks

In this way, a mobile user can seamlessly switch between different networks, supporting the
same services. This process must be performed to automatically adapt to change access
networks and environments, without any user participation. In order to do this, cross layer
design for multimedia communications is required. Mobile computing then becomes more
feasible, e.g. a mobile user performing a videoconference using UMTS maintains this service
even though the link breaks down, accessing into a WLAN network.
Vertical Handover (VHO) is a mechanism allowing heterogeneous connectivity by enabling
switches from a serving network to a candidate network, whenever users or network
requirements (i.e. power level, network congestions, or other QoS constraints) impose or
suggest it. Notice that VHO allows switching from one access technology to another, thus
offering additional functionalities with respect to classic horizontal handover where mobile
nodes move from an access point to another without changing the serving access network
(Balasubramaniam & Indulska, 2004), (McNair & Fang, 2004).
In this chapter we show how heterogeneous networks for next generation multimedia
systems can cooperate in order to provide seamless mobility support to mobile users
requiring high multimedia Quality-of-Service (QoS) constraints (Knightson et al., 2005).
We describe the traditional techniques of Vertical Handover in heterogeneous wireless
networks. Basically, in Section 2 we introduce the main characteristics of handover process
and our effort is addressed on a first handover classification, which distinguishes between
horizontal and vertical, hard and soft, upward and downward procedures, and more.
Beyond several handover algorithms, in Subsection 2.1 we give an overview of current IEEE
802.21 standard for seamless connectivity in heterogeneous environments. In Section 3 we
describe different decision metrics for handover mechanisms. Various metrics triggering
handover decisions, including multi-parameters QoS, and mobile terminal location
information, will be described in details in Subsection 3.1, and 3.2, respectively. Moreover, a
hybrid approach which exploits both power measurements and location information will be
presented in Subsection 3.3. Finally conclusions are drawn in Section 4.

2. Vertical handover procedures overview

New-generation wireless networks adopt a heterogeneous broadband technology model
aiming to guarantee seamless connectivity to mobile users, anytime and anywhere. Different
network characteristics are expected for different multimedia applications, each of them
requiring a specific QoS level. Ubiquitous access through a single network technology could
not always guarantee seamless connectivity, due to geographical coverage limitations, so
that the cooperation of different access networks represents an important feature for
heterogeneous environments.
A general definition of handover assumes it as the process by which a mobile terminal keeps
its connection active when migrating from the coverage of one network Access Point (AP) to
another. Basically, different types of handovers can occur in wireless overlay networks.
Network switching can be performed not only to maintain user connectivity but also to keep
high QoS. There are some decision handover parameters based on QoS, available resources,
channel quality or preference consumer.
In GSM, handover decision is based on the perception of channel quality, reflected by the
received signal strength and the availability of resources in neighbour cells. The Base Station
(BS) usually measures the quality of the radio link channels used by Mobile Nodes (MNs) in
its service area. Measures are periodically updated so that degradations in signal strength
Connectivity Support in Heterogeneous Wireless Networks 223

going below a prescribed threshold can be detected and handover toward another radio
channel or cell can be initiated.

Fig. 1. Heterogeneous networks scenario

Horizontal handover (HHO) occurs between the APs of the same network technology, while
vertical handover (VHO) occurs between APs belonging to different networks. Several kind of
VHO can be envisaged, as described as follows. According to Figure 1, upward vertical
handover is a handover to a wireless overlay with a larger cell size and generally lower
bandwidth per unit area. It makes a mobile device disconnect from a network providing
faster but smaller coverage (e.g. WLAN) to a new network providing slower but broader
coverage.
Viceversa, a mobile device performing a downward VHO disconnects from a cell providing
broader coverage to one providing limited coverage but higher access speed. In this case, a
link layer trigger can inform the mobile device that it is now under the coverage of a new
network (e.g. WLAN) and the mobile node may wish to execute the handover.
Downward VHOs may be anticipated or unanticipated, such that a mobile device may already
be under the coverage of the new network but may prefer to postpone the handover based
on requirements of the applications running on the mobile node. Handover is then
performed later, being already aware of the coverage status of the new network.
A main issue is to decide if or when to start the handover, and who performs it. Handover
policies are based on different metrics for handover decision. Traditional solutions simply
consider RSSI (Received Signal Strength Indication) and channel availability. More
sophisticated handover policies also consider: (i) Quality-of-Service, as different types of
services require various combinations of reliability, latency, and data rate; (ii) costs, i.e.
different networks may employ different billing strategies; (iii) network conditions like
traffic, available bandwidth, network latency, and congestion; (iv) system performance,
224 Recent Advances in Wireless Communications and Networks

such as channel propagation characteristics, path loss, inter-channel interference, Signal-to-

Noise ratio and Bit Error Rate; (v) mobile terminal conditions like battery power and
dynamic factors such as speed, moving pattern, moving histories, and location information.
In the latter case, if the battery level of a MN is low, the handover commutes toward a
network that guarantees lower power consumption. In the case when the user requires a
guaranteed QoS level for her applications, handover switches to a network meeting such
requirements.
A more detailed description of handover decision metrics will be given in Section 3.

2.1 IEEE 802.21 media-independent handover

The IEEE 802.21 group is developing standards to enable handover and interoperability
between heterogeneous network types, including both 802 and non 802 networks.
The standard provides quick handovers of data sessions across heterogeneous networks
with small switching delays and minimized latency. The handover in heterogeneous
networks could become more flexible and appropriate with this standard, through the use
of innovative IEEE 802.21 mobile devices. The standard considers both wired and wireless
technologies such as 802.3, 802.11, 802.16, 3GPP2, and 3GPP.
The analysis of IEEE 802.21 standard aims to understand the scope of this protocol. Seamless
handover of data sessions is the main target, based on Media Independent Handover (MIH)
functional model. IEEE 802.21 specification classifies the function that enhances handovers
across heterogeneous media. The MIH protocol entity is to every extent a new protocol layer
located between the Network Layer (Layer 3) and the interface-specific lower layers (MAC
and PHY in the case of IEEE interfaces, RRC and LAC in the case of 3GPP or 3GPP2
interfaces, respectively).
The main entities of IEEE 802.21 are (Gupta et al., 2006):
1. The Media Independent Information Service (MIIS) that includes policies and directives
from the Home Network (HN). The mobile terminal refers to the HN policies when
performing handover decisions;
2. The Service Access Points (SAPs), exchanging service primitives between the MIH layer
and its adjacent layers and functional planes;
3. A Decision Engine (DE) within the MIH instance, residing in the mobile terminal, which
identifies the best available access technology to support the current connectivity. The
DE is a state machine that selects a preferred link based on available interfaces, policies,
QoS and security parameter mapping;
4. A Transport Mechanism to facilitate the communication between the mobile terminal
MIH and the Information Service (IS) instance to access in the network.
The MIH function at the mobile terminal is continuously supplied with information
regarding the network conditions, measured to perform the access into one available
heterogeneous network. The MIH function receives the information through dedicated
interfaces by exchanging messages with the IS entity positioned in the HN. Generally, the
MIHF defines three main services to perform handovers between heterogeneous networks,
such as (i) the Media Independent Event Service (MIES), (ii) the Media Independent
Command Service (MICS), and (iii) the Media Independent Information Service (MIIS).
MIES provides event reporting, event filtering and event classification corresponding to
dynamic changes in link characteristics, link quality and link status. It acts all the instances
to make event detection and notify, still maintaining the actual link connection to the MN.
Connectivity Support in Heterogeneous Wireless Networks 225

Some of these events employed are “Link Up”, “Link Down”, “Link Detect”, “Link
Parameter Reports” and “Link Going Down”, (Gupta, et al. 2006).
MICS uses the MIHF primitives to send commands from higher layers to lower ones. It
determines the status of the connected links, performing mobile and connectivity decisions
of the higher layers to the lower ones. Then, MIIS is a mechanism to discover available
neighboring network information in order to facilitate handover process. It assures a set of
information entities, both static and dynamic. In the first case, there are the names and
services providers of the MN’s current network, while the dynamic information include link
layer parameters such as channel information, MAC addresses, security information and
other higher layer service information.
Three main handover schemes have been developed within IEEE 802.21 standardization
process, defined as follows:
1. Serving Network-Initiated and Candidate Network-Canceled Handover: the Service Access
Network (SAN) sends messages about information request to the IS in order to know if
a handover mechanism can be initiated. The Candidate Network (CN) could not be an
available resource, because of link quality level or network traffic status;
2. Serving Network-Initiated and MN-Canceled Handover: after sending information request
messages to the IS, the MN could be not available to perform handover, because of MN
movement or user intervention or low battery that let the MN renouncing handover
mechanism. In this case, the handover could be canceled directly by interaction between
SAN and CN;
3. MN-Initiated and MN-Canceled Handover: in this case, the MN communicates only with
the SAN, which sends messages of Resources-Request to the CN. The MN could be no
more available to perform handover, (e.g. movement or time out or user intervention).
The MN sends HandoverCancel message to SAN which is asked to interrupt the
handover mechanism. Figure 2 shows the flowchart of this approach.

3. Techniques for connectivity support in heterogeneous networks

Vertical Handover preserves user connectivity on–the–move (Pollini, 1996). It is applied when
network switching is expected in order to (i) preserve host connectivity, (ii) optimize QoS as
perceived by the end user, and (iii) limit the number of unnecessary vertical handover
occurrences.
Different VHO schemes can be classified on the basis of the criteria and parameters adopted
in the handover initialization phase. The following list collects the main metrics whose
monitoring can drive handover decisions:
• Received Signal Strength (RSS)-based VHO algorithms are largely used in cellular
networks (i.e. 2G and 3G networks). The handover process is initiated on the basis of a
decreasing level of measured RSS (Ayyappan & Dananjayan, 2008); (Inzerilli & Vegni,
2008);
• Signal-to-Noise and Interference ratio (SINR)-based VHO algorithms are typically used in
UMTS networks. SINR factor directly impacts achievable goodput in a wireless access
network. The handover is driven by a reduction of measured SINR below a fixed
threshold (Yang et al., 2007); (Vegni et al., 2009);
• Multi-parameter QoS-based VHO algorithms: this approach is based on the overall
quality assessment for the available networks obtained balancing various parameters —
subjective and objective quality metrics— (Vegni et al., 2007);
226 Recent Advances in Wireless Communications and Networks

• Location-based VHO algorithms estimate network QoS levels on the basis of the MN’s
location relatively to the serving access point (Kibria et al., 2005); (Inzerilli et al., 2008);
(Inzerilli et al., 2010).
The above-mentioned VHO algorithms are analyzed here in the following.
The RSS-based VHO is the traditional technique for connectivity switching in heterogeneous
networks. It is driven on power level measurements, such as when the measured RSS
coming from the SN drops below a predefined threshold. The RSS of the monitored set of
CNs is evaluated, and a vertical handover will be executed towards the best—most
appropriate—candidate network. This approach represents the primitive and simplest
handover mechanism, which however does not aim to optimize communication performance
but only focuses on maintaining seamless connectivity (Ayyappan & Dananjayan, 2008).
Moreover, since RSS value suffers from severe fluctuations due to the effects of shadowing
and fading channels, filtering techniques (e.g. exponential smoothing average (Inzerilli &
Vegni, 2008)) should be considered to estimate the trend of RSS signal.
On the other hand, the SINR-based approach compares received power with noise and
interference levels in order to obtain a more accurate performance assessment. SINR factor
represents a valid handover decision metric, as it directly affects the maximum data rate
compatible with a given Bit Error Rate (BER). SINR-based VHO approach is more suitable to
meet QoS requirements, since a reduction of SINR factor produces a reduction of data rate
and QoS level (Yang et al., 2007). Both RSS and SINR-based schemes are reactive approaches,
whose aim is to compensate for performance degradation.
The multi-parameter QoS-based VHO scheme in (Vegni et al., 2007) represents a proactive
approach performing regular assessment of the QoS level offered by the current SN, as well
as by other CNs. In general, a multi-parameter QoS-based VHO technique is well suited for
multimedia applications like real-time video streaming.
In location-based VHO solutions, the knowledge of MN’s location information is exploited
to assess the quality of the bidirectional link between SN and MN (Inzerilli et al., 2008).
Moreover, the estimation of MN’s position can drive the initiation of a reactive handover
mechanism. Information about MN’s position can be determined in several ways (Kibria et
al., 2005), including Time of Arrival, Direction of Arrival, RSS, as well as A-GPS (Assisted
Global Positioning System) techniques.
Notice that in general each time a vertical handover is initiated the traffic overhead
increases. The limitation of handover occurrences is an issue specially when unwanted and
unnecessary vertical handovers are executed. This represents the case of a mobile node
moving back and forth between the two neighbouring wireless networks—or in general
around a corner that involves three or more wireless networks—. This aspect in known in
literature as ping-pong effect (Kim et al., 2007).
Repeated vertical handover attempts lead to frequent location and registration updates
(with network resource consumption), frequent connectivity interruptions, as well as serious
affections to MN’s QoS (i.e., decreasing battery life). Frequent handovers lead the user to
experience many unpleasant transients of service interruption.
Techniques to prevent unnecessary and unwanted handovers have been proposed (Kim et
al., 2007); (Inzerilli & Vegni, 2008); (Inzerilli et al., 2008). A hysteresis cycle or a hard
limitation in maximum handover frequency can mitigate this phenomenon.
The above descriptions have shown the main vertical handover approaches, which are
based on single metrics (i.e., RSS, SINR, QoS, and location). Still, many handover techniques
are based on the combination of two or more metrics, which generate most effective VHO
Connectivity Support in Heterogeneous Wireless Networks 227

decisions, and avoid unnecessary and unwanted vertical handover occurrences (Vegni et al.,
2009). These approaches are called as hybrid (or combined) vertical handovers.
The following Subsection 3.1 and 3.2 offer a more detailed description of QoS- and location-
based vertical handover algorithms, respectively.

Mobile Serving Access Network Candidate Access Information

Node (3G) Network (WLAN) Server
( )
1. MIH_Get_Information Request
2. MIH_Get_Information Response

WLAN Available. Power on

WLAN Radio. Listen to Beacon

3. MIH_Link_Detected Indication
Link Type = WLAN
MACNewPoA = MAC of AP(x)
MIH Capability = 1 (Supported)

Network Selection
Switch to WLAN

4. MIH_Net_HO_Candidate_Query Request
CurrentLinkType = 3G
SuggestedNewPoA = MAC of AP(x)
HandoverMode = Make before Break
OldLinkAction = LINK_NO_ACTION

5. MIH_N2N_HO_Query_Resources Request
QueryResourceList = QoS:??

6. MIH_N2N_HO_Query_Resources Response
ResourceStatus = 1 (Available)
AvailableResourceList = QoS:??
7. MIH_MN_HO_Candidate_Query Response
CurrentLinkType = 3G
HandoverAck = 1 (Initiate Handover)
PreferredLinkType = WLAN
PreferredPoA = MAC of AP(x)
ErrorCode = 0

8. MIH_MN_HO_Candidate_Commit Request
CurrentLinkType = 3G
NewLinkType = WLAN
MACNewPoA = MAC of AP(x)
OldLinkAction = HANDOVER_CANCEL

9. MIH_MN_Candidate_Commit Responset
CurrentLinkType = 3G
OldLinkAction = HANDOVER_CANCEL
HandoverStatus = 0 (Abort Handover)

10. MIH_N2N_Candidate_Commit Request

CurrentLinkType = 3G
OldLinkAction = HANDOVER CANCEL

Optional. Handover-Cancel 11. MIH_Net_Candidate_Commit Request

transaction between CurrentLinkType = 3G
Serving Network & OldLinkAction = HANDOVER CANCEL
Candidate Network.

Fig. 2. MN-initiated and MN-cancelled handover

228 Recent Advances in Wireless Communications and Networks

3.1 QoS-based vertical handover

As defined in IEEE 802.21 standard, QoS-based HO decision is based on current and
expected network conditions, according to the application QoS requirements (Golmie et al.,
2006). Current network conditions are measured using network performance parameters
from various layers, e.g. signal strength from layer 1, packet loss from layer 2, throughput
and delay from layer 2+, etc.
According to the ITU-T Y.1540, the applications QoS requirements are defined by:
• Packet Transfer Delay (PTD), maximum end-to-end tolerated delay [s];
• Packet Delay Variation (PTV), i.e. jitter: maximum packet jitter [s];
• Packet Loss Ratio (PLR): maximum tolerated packet loss;
• Throughput: required data rate of successful packets [bit/s].
QoS-based Decision Engine (QDE) is a main network entity that implements a QoS-based
handover for assigned application QOS requirements (Golmie et al., 2006). QDE is a MIH
user that considers application QoS requirements and network performance measurements
provided by the MIH. The MIH function exchanges information between network entities
and the QDE, including technology, protocol types and network measurements. Network
performance conditions, such as instantaneous measurements for current conditions, are
evaluated from past observations and previous connections, or as default estimates.
QDE entity is located as a remote entity, as part of the MN, or the AP/BS. For our scope, we
consider it as a network entity, as illustrated in Figure 3. No limitation occurs if QDE
function is distributed over remote entities of each network. In this way, the MN receives
Video Quality Metrics VQMs from QDEs of neighbouring candidate networks.

Fig. 3. Proposed network architecture for QoS-based handover

Connectivity Support in Heterogeneous Wireless Networks 229

During the last decade several techniques to assess the quality of multimedia services
without performing a subjective test have been investigated, leading to objective quality
metrics that approximate the MOS (Mean Opinion Score). Such metrics are usually classified
as full-reference, reduced reference and no-reference metrics. They differ in the degree of
knowledge about the original multimedia flow used in the quality assessment. Full
reference methods evaluate the difference between the original signal and the received one
and are, thus, rarely employed in real time assessment of video quality. On the other hand,
reduced reference metrics require a channel for sending side information concerning some
characteristics of the original signal, while No-Reference (NR) metrics do not require any
knowledge of the original. On the other hand, the perceived video quality highly depends
on the end-user, according to her a priori knowledge of the topic, level of attention while
looking at the video, and assigned task.
In general a QoS-based VHO technique focuses on the maximization/enhancement of QoS
level experienced by the mobile user. The connectivity is switched from the serving network
to a selected candidate network, which provides high QoS. In (Vegni et al., 2007) an
innovative NR-Video Quality Metric (NR-VQM) has been assumed. This metric incorporates
both spatial and temporal resolution reduction, packet losses, latency, and delay jitter, as
well as indicators for the evaluation of (i) blocking, blurring and ringing introduced by
current video coders, and (ii) jerkiness and other effects produced by packet losses. Jerkiness
is evaluated by a neural network feed with estimated dynamics of objects composing the
scene, trained by means of subjective tests.
The impact of packet losses on perceived quality is based on the analysis of the inter-frame
correlation measured at the receiver side. Presence of error concealment algorithms
employed by the receiver is also taken into account. The packet losses degradation
assessment presents best performances for long sequences with slow motion.
Since the NR algorithm directly processes the rendered video, no information about the
kind of errors, delays and latencies affecting the links is required. In addition, the NR
techniques easily account for continuous increase of computing power of both mobile and
wired terminals, and allow a wide spread of error concealment techniques to increase the
perceived quality.
Subjective test evinced good agreement between NR metric and MOS, regardless of intrinsic
video characteristic and spatial-temporal resolution.
In the scheme presented in (Vegni et al., 2007) the NR-QoS is combined with a packet
classification originally designed for DiffServ (DS) protocol, allowing different QoS grades
to be mapped into different classes of aggregated traffic flows, (Shin et al., 2001). It is an
adaptive packet forwarding mechanism for DS networks. It allows mapping mechanism of
video packets onto different DS levels based on Relative Priority Index (RPI), that is the
relative preference per each packet in terms of loss and delay. The packets are classified,
conditioned, and remarked to certain network DS levels by considering the traffic profile
based on the Service Level Agreement (SLA) and the current network status.
The proposed QoS-based handoff scheme in (Vegni et al., 2007) is well suited for real-time
applications in IEEE 802.21 scenarios. The overall vertical handover process is organized in
five phases, such as (i) measurement phase, (ii) QoS prioritization, (iii) initialization, (iv)
candidate networks scanning, and (v) VQM conversion. Figure 4 depicts the pseudo-code of
the QoS-based vertical handover algorithm. Basically, the multi-parameter QoS-based
handover algorithm considers three inputs, such as (i) the RSS samples, (ii) the mobile node
distance and (iii) the current QoS level (i.e. Lev). As output it returns the number of vertical
230 Recent Advances in Wireless Communications and Networks

handovers executed by the mobile node (i.e. nVHO). In the following we shall describe the
overall mechanism in more details.
Given a video application, the QoS mapping process is accomplished by considering the
relative prioritized packets to maximize end-to-end video quality. In the measurement phase,
each T seconds, the MN gets samples of RSS and position, as well as monitors Lev parameter
for the video stream received from the current serving network. QoS monitoring is
performed on the basis of the NR metrics1. In our scope, NR technique addresses on audio
and video flows evaluations to the receiver side, although it could be also tuned and
optimized by means of a full-reference metric applied to some low rate probe signals.
In the QoS prioritization phase, the probability to perform a handover is evaluated (i.e. PVHO).
By opportunistically weighting the QoS-Lev parameter the handover probability will be
mainly driven by QoS factors. The received video-streaming quality is monitored according
to a subjective evaluation. On the basis of user preferences, two appropriate QoS thresholds
are defined, called as Th1 and Th2, with Th1 > Th2.

Input : {RSS ; d ; Lev}

Output : nVHO
while T > 0 do
Measurement phase in SN
if Lev < Th1 then
First alarm to QDE
QoS priorization phase
Calculation of PrVHO ( x )
end
Candidate network scanning phase
List of target networks
if L ev < Th 2 then
Handoff initiation phase
VQM Conversion phase
Selection of a target network
nVHO ← 1 ( VHO executed )
end
end
end
end

else
nVHO ← 0 ( no VHO executed )
end
end

Fig. 4. Multi-parameter QoS-based vertical handover pseudo-code

1 NR method presents some basic indicators for temporal and spatial analysis: block distortion is

evaluated by applying first a coarse temporal analysis for each frame, to extract blocks potentially
affected by artefacts produced by lost packets.
Connectivity Support in Heterogeneous Wireless Networks 231

Traffic congestions, transmission errors, lost packets or delay can keep QoS level lower than
a first threshold, i.e. Lev < Th1. If Lev keeps on decreasing, the handoff initiation phase can be
required by the MN whenever Lev < Th2.
The MN alerts to change the serving network and sends this alarm message to a closer QDE.
So, the candidate networks scanning phase occurs on the basis of VQM parameters, such as
throughput, link packet error rate, Packet Loss Probability (PLP), supported number of
Class of Service (CoS), etc. All these parameters are sent inside the information message
“LINK QoS PARAMETER LIST”.
Based on the statistics computed on previous NR-QoS reports produced by the served MNs,
when the QDE communicates with a MN, it can operate a conversion of VQM parameters for
each network in NR parameters. The MN evaluates which candidate network is appropriate
for its video application. As an instance, let us suppose that a MN is in a WLAN area. When
it realizes a QoS reduction, it sends a first alarm to the QDE, which will start a candidate
network scanning process in order to select a target network providing a QoS enhancement
to the MN (i.e. Lev1 > Lev). A set of target networks to hand over is selected, and the best
network is chosen on the basis of MN preferences and handover policies. Finally, the handover
is performed according to the IEEE 802.21 message exchange in the scheme of Figure 2.
Finally, in order to determine the probability to perform a vertical handover (i.e. Pr(VHO)),
we shall provide the following assumptions:
1. A mobile node is locating at position P = (x, y), in the middle of two active networks (i.e.
Ni, i = 1,2);
2. The averaged received signal levels over N1 and N2 radio links are assumed as lognormal
distributions, respectively rN 1 ( P ) and rN 2 ( P ) , with mean signal levels μ N 1 and μ N 2 ,
and the shadowing standard deviations σ N2 1 and σ N2 2 ;
3. The distance dN i from the MN’s position to the reference BS of network Ni can be
assumed as a stationary random process with mean value d and variance c 2σ t2dn , where
c the speed of light and σ t2dn is the standard deviation of the signal delay measurement 2;
4. On the basis of each single parameter (i.e. RSSI, distance, and QoS) different thresholds
are assumed, called as R, D and Q for RSSI, distance and quality criterions, respectively.
Each threshold is typical for a single access network (i.e. RW and RU are the RSS
thresholds for WLAN and UMTS, respectively).
Let us suppose to perform a handover from UMTS to WLAN. The handover decision occurs
when both (i) the RSS measurement on WLAN is higher than RW (i.e. rW ( P ) ≥ RW ), (ii) the
distance from MN to WLAN AP is lower than DW (i.e. dW ≤ DW), and (iii) the QoS-Lev in
WLAN is upper than QW (i.e. qW ≥ QW) 3. Thus, the probability to initiate the handover from
UMTS to WLAN in the position P, is

{⎣ ⎦}
PrU → W ( P ) = P ⎡rW ( P ) ≥ RW ⎤ ⋅ P {[ dW ≤ DW ]} ⋅ P {[ qW ≥ Th2 ]} . (1)

On the other hand, the handover decision from WLAN to UMTS is taken only when it is
really necessary, such as when (i) the RSS measurement on WLAN is lower than RW (i.e.

2 Basically, the delay measurement of the signal between the MN and the BS is characterized by two

terms, (i) the real delay and (ii) the measurement noise tdn. It is assumed to be a stationary zero-mean
random process with normal distribution.
3 Notice that due to chip WLAN monetary cost the handover decision does not take account to the RSS,

the distance and the QoS criteria on UMTS network.

232 Recent Advances in Wireless Communications and Networks

rW ( P ) ≤ RW ), (ii) the RSS measurement on UMTS is higher than RU (i.e. rU ( P ) ≥ RU ), (iii) the
distance from MN to UMTS BS is lower than DU (i.e. dU ≤ DU), and (iv) the QoS-Lev in UMTS
is upper than QU (i.e. qU ≥ QU). Thus, the probability to initiate the handover from WLAN to
UMTS in the position P, is

{
⎣ ⎦ } {⎣ ⎦ }
PrW →U ( P ) = P ⎡rW ( P ) ≤ RW ⎤ ⋅ P ⎡rU ( P ) ≥ RU ⎤ ⋅ P {[ dU ≤ DU ]} ⋅ P {[ qU ≥ Th2 ]} . (2)

3.1.1 Analytical model

In this subsection we introduce the analytical model behind the QoS-based VHO technique
described in (Vegni et al., 2007). Particularly, we shall define two main network parameters
for handover decision from a serving network to a candidate network, such as (i) the
average time delay and (ii) the average packet rate. Based on these parameters, the handoff
mechanism shall be performed only if it is necessary to maintain the connection on.
We recall the average time delay [s] for the k-th network from the Pollaczeck-Kinchin formula,
which considers the average time delay as the sum of average time delay for the service and
waiting one, such as

τk =
⎡
1 ⎢ 1 + ρ k 1 + Cb
⋅
2
( ) ⎤⎥ . (3)
μk ⎢ 2 ( 1 − ρ k ) ⎥
⎣ ⎦
From (3) we consider the average time delay for a single packet sent from a N1 to N2 (i.e.
TN 1 → N 2 [s]) such as

N
TN 1 → N 2 = ∑ τ kγ N( k1)→ N 2 , (4)
k =1

where γ N( k )→ N is the probability that packets are sent from N1 to N2, on the k-th link with
1 2

capacity Ck [bit/s].
The average packet rate represents how many packets are sent from N1 to N2, i.e. rN 1 → N 2
[packets/s]. Considering all the available networks (i.e. N1, N2, …, Nn), the total average
packet rate Λ tot [packets/s] is

N N
Λ tot = ∑∑ rN i → N j , (5)
i =1 j =1

and the total mean time delay ΔTMean [s] is

N N
∑∑ TN →N i j
⋅ rN i →N j
i =1 j =1
ΔTMean = N N
. (6)
∑∑ rNi →N j
i =1 j =1

In the case of handover occurrence from Ni to Nj, the mobile user moves from Ni to Nj with a
probability β i → j . So, the probability ν j that an user moves from her own serving network is:
Connectivity Support in Heterogeneous Wireless Networks 233

ν j = ∑ β i → j = 1 − β i →i , (7)
i≠ j

where β i →i represents the probability a user stays in her serving network. In this way, we
can find the average packet rate from Ni to Nj during handover, as

→ N = ∑ rN → N β j →m + β i → j ∑ rN → N .
rN( HO ) ( HO ) ( HO ) (8)
i j j m i h
m h

So, the total packet rate Λ(totHO ) [packets/s] will be:

N N
Λ(totHO ) = ∑∑ rN( HO
→ N = ∑∑∑ rN → N β i → m + ∑∑ β i → h ∑ rN → N =
) ( HO ) ( HO )
i j i m i h
i =1 j =1 i j m i j h (9)
= Λ tot + ∑∑ β i → h ∑ rN( HO )
→N . i h
i j h

Let us assume ρj [packets/s] as the average rate of packets sent to Nj. By replacing (7),
HO
the expression of Λ tot becomes
HO
Λ tot = Λ tot + ∑ν j ρ j . (10)
j

If we consider an uniform handover probability (i.e. ν j = ν ) then Λ(totHO ) becomes

Λ(totHO ) = Λ tot ( 1 + ν ) . (11)

Finally, let χ be the ratio between the average time delay in case and in absence of
handover, τ ( HO ) and τ respectively

⎡
1 ⎢1 +θ
( HO )
1 + Cb 2 ( ) ⎤⎥
χ=
τ ( HO )
=
μ k ⎢ 2 1 − θ ( HO )
⎣ ( ) ⎥
⎦ = 1 +θ
( HO )
(1 + Cb 2
⋅
)
(1 − θ ) =
τ ⎡
1 ⎢ 1 + θ 1 + Cb ⎥
2 ⎤
( ) (
1 + θ 1 + Cb 2 ) (
1 − θ ( HO ) ) (12)
μk ⎢ 2 ( 1 − θ ) ⎥
⎣ ⎦

=
(
1 + θ 1 + Cb 2 ( 1 + ν ) ) ⋅
(1 − θ ) .
1 + θ 1 + Cb ( 2
) 1 − θ (1 +ν )

where θ ( HO ) [Bit/s] is the throughput experienced by a mobile user during handover.

3.2 Location-based vertical handover

In this subsection we shall introduce a location-based vertical handover approach (Inzerilli
et al., 2008) which aims at the twofold goal of (i) maximizing the goodput and (ii) limiting
the ping-pong effect. The potentialities of using location information for VHO decisions,
especially in the initiation process is proven by experimental results obtained through
computer simulation. Leveraging on such results, in this subsection we shall introduce only
the handover initiation phase since it represents the core of our location-based VHO
technique. A detailed description of the proposed algorithm is in (Inzerilli et al., 2008).
234 Recent Advances in Wireless Communications and Networks

The mobile node’s location information is used to initiate handovers, that is, when the
distance of the MN from the centre of the cell of the candidate network towards which a
handover is attempted possesses an estimated goodput, i.e. GPCN, significantly greater than
the goodput of the current serving network, i.e. GPSN. The handover initiation is then followed
by a more accurate estimate (handover assessment) which actually enables or prevent
handover execution (Inzerilli et al., 2008).
In the handover initiation phase, the algorithm evaluates the goodput experienced by a MN in
a wireless cell. The goodput depends on the bandwidth allocated to the mobile for the
requested services and the channel quality. When un-elastic traffic is conveyed (e.g. real-
time flows over UDP) the goodput is given by:

GP = BW ⋅ ( 1 − Pout ) , (13)

where BW [Bit/s] is the bandwidth allocated to the mobile node and Pout is the service
outage probability. When elastic traffic is conveyed (typically when TCP is used),
throughput tends to decrease with increasing values of Pout. BW is a function of the nominal
capacity, of the MAC algorithm which is used in a specific technology and sometimes of the
experienced Pout. We consider the maximum value of BW, i.e. BWmax which is obtained in the
case of a single MN in the cell and with a null Pout 4.
Pout is a function of various parameters. In UMTS network it can be calculated theoretically,
using the following formula:

⎧ ⎫
EbUMTS
UMTS
Pout = Pr ⎨
⎪ ,Tx
( ) ⎪
⋅ Ad−1 rUMTS ≤ μ UMTS ⎬ , (14)
I
⎩⎪ 0
+ γσ ( 2
N
UMTS
) ⎭⎪
where EbUMTS
,Tx is the bit energy in the received signal, µ and γ are parameters dependent on
the signal and interference statistics, σ N2 is the receiver noise power, Ad (r UMTS) is the signal
attenuation factor dependent on the MN’s distance r UMTS from the centre of the cell, and I0 is
the inter and intra-cell interference power. The service outage probability for a WLAN
WLAN
network Pout can be calculated theoretically in a similar fashion using the following
formula:

⎧ WLAN ⎫
⎪ Eb ,Tx
WLAN
Pout = Pr ⎨ ⋅ A −1 WLAN
r (≤ μ )
WLAN ⎪
⎬. (15)
( 2
⎪⎩ γσ N WLAN) d
⎭⎪
We define as the radius of a wireless cell Rcell the distance from the cell centre beyond which
the signal-to-noise ratio or the signal-to-interference ratio falls below the minimum
acceptable value (i.e. μ). Rcell can be obtained resolving the above equations or empirically,
through measurement on the network. As an alternative, typical value for well-known
WIFI
technologies can be used, e.g. Rcell ≈ 120 m for IEEE 802.11a outdoor, and
UMTS
100 m ≤ Rcell < 1 km for a UMTS micro-cell.

4 In an IEEE 802.11a link, the maximum theoretical BWWLAN is equal to 23 Mbps (out of a nominal
capacity of 54 Mbps), although it decreases rapidly with the number of users because of the contention-
based MAC. In HSDPA network, the maximum BWUMTS is equal to 14.4 Mbps, which decreases rapidly
with Pout.
Connectivity Support in Heterogeneous Wireless Networks 235

Since the path loss Ad (r ) is approximately proportional to rγ, the SNR(r) can be written as

⎡⎛ R ⎞γ ⎤
SNR(r ) = μ ⎢⎜⎜ cell ⎟⎟ + δ Ad ⎥ . (16)
⎢⎝ r ⎠ ⎥
⎣ ⎦
Maximum GP in a WLAN and UMTS cell can be calculated with the following approximated
formulas, respectively

⎧ ⎧⎛ UMTS ⎞γ ⎫
⎪GPUMTS = BW UMTS ⋅ Pr ⎪⎨⎜ Rcell ⎟ + δ A < 1⎪⎬
⎪ max max ⎜ r UMTS ⎟ d
⎪ ⎩⎪⎝ ⎠ ⎭⎪
⎨ (17)
⎧⎛ R WLAN ⎞ γ ⎫
⎪ WLAN WLAN ⎪ cell ⎪
⎪GPmax = BWmax ⋅ Pr ⎨⎜ WLAN
⎜ ⎟ + δ Ad < 1⎬
⎟
⎪ ⎪⎩⎝ r ⎠ ⎪⎭
⎩
which will be regarded as zero out of cells.
Handover initiation will be performed when the estimated goodput of the new network is
greater than the current one. Namely, in the case of vertical handover from WLAN to
UMTS, the following equations applies:

UMTS WLAN
GPmax < GPmax . (18)

It is worth noticing that when handover executions are taken too frequently, the quality as
perceived by the end user can degrade significantly in addition to wasting battery charge.

3.2.1 Simulation results

In this section we report on network performance of the Location-based Vertical Handover
algorithm (also called as LB-VHO). Particularly, we investigate the Cumulative Received
Bits (CRB [Bits]), and the number of vertical handovers performed by the user moving in the
grid, obtained using our event-driven simulator. Details of the simulator can be found in
(Vegni, 2010).
We modelled movements of a MN over a grid of 400 x 400 square zones, each with an edge
of 5 m, where 3 UMTS cells and 20 IEEE 802.11b cells are located. Typical data rate values
have been considered for UMTS and WLAN. The location of each wireless cell has been
generated uniformly at random, as well as the the MN’s path.
Table 1, shows the statistics on the CRB collected for S = 20 randomly generated scenarios,
each of them differs from the other in terms of the UMTS/WLAN cell location and the path
of the MN on the grid. Performance have been compared to a traditional Power-based
Vertical Handover (PB-VHO), which uses power measurements in order to initiate VHOs
instead of mobile location information (Inzerilli & Vegni 2008).
For each approach LB and PB three parameters are reported related to the CRB, i.e. the mean
value, the standard deviation and the dispersion index, defined as the ratio of the standard
deviation over the mean value. The three value for LB and PB are reported versus different
values of the waiting time parameter Twait 5.

5 Notice that if the MN moves at 1 m/s, a 10 s waiting time results to 10 m walked.

236 Recent Advances in Wireless Communications and Networks

The LB approach brings about a reduction of CRB between 6.5% for a null waiting time and
20% for waiting time equal to 60 s. It follows that the waiting time constraint is not suitable
for LB approach in order to reduce the number of vertical handovers while keeping a
limited reduction of CRB.
Table 2 shows results of the number of VHO experienced with the LB and PB approach, still
in terms of the mean value, standard deviation and dispersion index for various waiting
time values. It can be noticed that the number of vertical handover with LB is on average
significantly smaller, i.e. ranging in [9.65, 3.70] than that experienced with PB approach, i.e.
ranging in [9.15, 329.85]. This remarks that the PB approach requires a constraint on
handover frequency limitations, while this approach is counterproductive with LB.

Waiting LB Mean LB Stand. LB Disp. PB Mean PB Stand. PB Disp.

Time [s] [Gb] Dev [Gb]. Index [Gb]. Dev. [Gb]. Index
0 5.82 2.38 40.91% 6.23 2.30 36.90 %
60 4.59 2.34 50.88% 5.76 2.14 37.13 %
Table 1. Statistics on the CRB for LB and PB approach

Waiting LB Mean LB Stand. LB Disp. PB Mean PB Stand. PB Disp.

Time [s] [Gb] Dev. [Gb] Index [Gb] Dev. [Gb] Index
0 9.65 2.00 20.73 329.85 794.50 240.87
10 7.25 1.15 15.93 30.20 46.36 153.51
20 5.85 2.31 39.48 19.90 22.54 113.26
30 5.15 1.15 22.42 14.10 16.29 115.53
40 4.35 1.15 26.54 11.80 12.49 105.85
50 4.20 2.00 47.62 9.80 10.58 107.99
60 3.70 1.15 31.21 9.15 7.57 82.75
Table 2. Statistics on the Number of VHO for LB and PB approach

25
PB-VHO
LB-VHO

20
Vertical handovers (VHO)

15

10

-10 0 10 20 30 40 50 60 70

waiting time, [s]

Fig. 5. Number of vertical handover occurrences for PB and LB VHO algorithm

Connectivity Support in Heterogeneous Wireless Networks 237

In Figure 5, the mean values of vertical handovers for LB and PB vs. the waiting time
constraint are depicted. This shows even more clearly how the LB approach, providing a
more accurate assessment for handover initiation, limits handover initiations, resulting in
about a little performance gain. In contrast, PB approach is unstable even for high values of
waiting time, as it can be noticed from the fact that the PB curve is not monotone.
Finally, in Figure 6 (a) and (b) are reported the dynamics of the CRB over the mobile node
steps during the simulation (a step is performed every 5 seconds) for a null waiting time and
a waiting time of 60 s, respectively. The instability of PB approach when no waiting time
constraint is applied is clearly shown in Figure 6 (a).

9 9
x 10 x 10
10 10
LB-VHO, Twait = 0 LB-VHO, Twait = 60
9 9
PB-VHO, Twait = 0 PB-VHO, Twait = 60
8 8

7 7

6 6
Bits

Bits
5 5

4 4

3 3

2 2

1 1

0 0
0 500 1000 1500 2000 2500 0 500 1000 1500 2000 2500

MN’s steps MN’s steps

(a) (b)

Fig. 6. CRB during a simulated scenario with PB and LB-VHO approaches, for (a) Twait = 0 s,
(b) Twait = 60 s

3.3 Hybrid vertical handover technique

In this section we complete the overview of the main vertical handover techniques in
heterogeneous wireless networks, by introducing a hybrid scheme for connectivity support6.
Different wireless networks exhibit quite different data rate, data integrity, transmission
range, and transport delay. As a consequence, direct comparison between different wireless
links offering connectivity to a MN is not straightforward. In many cases VHO requires a
preliminary definition of performance metrics for all the visited networks which allows to
compare the Quality-of-Service offered by each of them and to decide for the best.
VHO decisions can rely on wireless channel state, network layer characteristics and
application requirements. Various parameters can be taken into account, for example: type
of the application (e.g. conversational, streaming, interactive, background), minimum
bandwidth and maximum delay, bit error rate, transmitting power, current battery status of
the MN, as well as user’s preferences.
In this section we present a mobile-controlled reactive Hybrid VHO scheme ―called as
HVHO― where handover decisions are taken on the basis of an integrated approach using
three components: (i) power map building, (ii) power-based (PB) VHO, and (iii) enhanced

6 An extended version of this technique is described in (Inzerilli et al., 2010).

238 Recent Advances in Wireless Communications and Networks

location-based (ELB) VHO. The HVHO technique is suitable for dual-mode mobile
terminals provided with UMTS and WLAN network interface cards, exploiting RSS
measurements, MN’s location information, and goodput estimation as discussed in Section
3. The overall procedure is mobile-driven, soft and includes measures to limit the ping-pong
effect in handover decisions. The flowchart of HVHO is depicted in Figure 7.
Basically, the HVHO approach proceeds in two phases:
1. In the initial learning phase when the visited environment is unknown, the RSS based
approach is used, i.e. hereafter referred to as Power-Based (PB) mode. In the meanwhile,
the MN continuously monitors the strength of the signals received from the SN, as well
as from the other candidate networks. By combining RSS samples with location data
provided by the networks or some auxiliary navigation aids, like GPS, the MN builds a
path losses map for each discovered network in the visited environment;
2. At the end of this phase the MN enters the ELB-VHO mode and it can exploit the path
losses map to take handover decision using its current location.

INIT

N
start PB-VHO
MRIn > MRIth
algorithm
Y
Build cell
radius list

start ELB-VHO stop PB-VHO

algorithm algorithm

Fig. 7. Flowchart of hybrid vertical handover algorithm

In the initial learning phase, the new environment is scanned in order to detect the UMTS
and WLAN access networks eventually present and, then to build a path loss map for each of
them. The path losses associated to the UMTS base stations in the monitored set and to the
access points of the WLAN network are estimated by taking the difference between the
nominal transmitted power and the short term average of the received signal strength.
Averaging is required in order to smooth fast fluctuations produced by multipaths, and can
be performed by means of a mean filter applied to the RSS time series multiplied by a
sliding temporal window (Inzerilli & Vegni, 2008).
Let n be the discrete time index and pn be the power measure at time tn. The moving average
estimate PN of the received power on a sliding window of length K is

n
1
PN = ∑
K i = N −K + 1
pi , N ≥ K . (19)

Though averaging over the last K samples, it allows reducing the impact of instantaneous
power fluctuations in power detection and reduce the power error estimation. On the other
Connectivity Support in Heterogeneous Wireless Networks 239

hand, as the MN is assumed to be moving, the length of moving windows depends on the
actual MN speed. However, moving average filters are prone to outlayers. A more robust
estimate can therefore be computed by replacing the linear mean filter with the (non-linear)
median filter.
Let us suppose a mobile node is moving in an area approximated with a lattice of M x M
square zones, each with a width Lzone [m]. While moving on the lattice, the MN calculates
the power received for each visited zone Zj. Let Pn Zj denote the average of the power
samples collected inside Zj and associates it with the planar coordinates of the centre of the
zone (xj, yj). Namely, power level calculated at time n for the zone (xj, yj) is given by

Z
(
Pn j = Pn xi , y j = ) 1
Z( j)
∑ pi , (20)
i∈Z ( j )

where Z (j) is the set of the power samples pi collected in the last visited zone j up to time n,
and Z ( j ) is the cardinality of the set Z (j).
Equation (20) provides a criterion to assess received power from both the UMTS and WLAN
networks on which handover decisions of the PB approach are based. In addition, it allows
assessing the power P(xj, yj) for each Zj zone which can be stored in the terminal and
populate a power map for the visited area. Once each zone of the lattice were visited at least
once, the power map would be completed. However, it is possible that the complete visit of
all the zones of the map can take long, and perhaps never occurs, especially if the number of
zones M2 is big. As a consequence, in order to accelerate power map building we can use
polynomial interpolation to assign a power value to zones which has not been visited yet.
Namely, let us assume that zone Zj has not been assigned a power value yet. Moreover, let
Zj1 and Zj2 be the nearest zones and aligned to Zj (as depicted in the examples of Figure 8),
with a power value assigned. We can use linear interpolation to assess the power value
P (xj, yj) of Zj as follows. When the zone j-th is between zone j1 and j2 (Figure 8(a) and (b)),
assessed power value of P(xj, yj) of zone j-th is given by:

D jj2 D jj1
( )
P xi , y j =
D jj1 + D jj2
(
P x j1 , y j1 + ) D jj1 + D jj2
( ) j j
P x j2 , y j2 , with D j 1 < D j 2 , (21)

where

(xj − xj ) + (y j − y j )(xj − xj ) + (y j − yj )
2 2 2 2
D jj1 = 1 1
, D jj2 = (22)
2 2

are the Euclidean distances between the centers of the zones Z j = ( x j , y j ) and Z j = ( x j , y j )
with Z j = ( x j , y j ) , respectively. Conversely, when Z is not between Z j and Z j , as
1 2
j 1 2
depicted in Figure 7 (c) and (d), the assessed power value P (xj, yj) of the zone j-th is given
by:

D jj2 D jj2
(
P xj ,yj = ) D jj2 − D jj1
(
P x j1 , y j1 − ) D jj2 − D jj1
( )
P x j2 , y j2 . (23)

It is worth highlighting that linear interpolation through (22) and (23) brings about errors in
the power map. In addition, the exploitation of (22) and (23) starting from the power values
of all visited zones does not guarantee completion of the power map. In general, a sufficient
240 Recent Advances in Wireless Communications and Networks

y y
j2
j2 (a) (b)
D = 2
j1
D =2 2
j1
j j
j j

j1 D =2 2
j2
j D jj2 = 3 2
j1

y x y x

(c) j (d)
D =2 j1 j1
D = 5
j2 j1 j j
j1 j

D =3 j2
j D jj2 = 2 5
j2

x x
Fig. 8. Power Map built according to the displacement of zones
number of visited zones has to be achieved prior completion of the power map is possible.
Such number is also dependent on the actual path of the MN in the lattice.
Let us introduce a coefficient to denote the degree of reliability of the power map at time n.
Let VZn be the set of visited zones up to time n. We introduce the Map Reliability Index
MRIn at time n as follows:

VZn
MRI n = , (24)
M2
where M2 is the total number of zones in the square lattice. We fix empirically a threshold
value MRITH for the index in (24) beyond which the knowledge of the visited environment is
regarded acceptable. Only when this threshold value is reached MRITH, polynomial
interpolation with (22) and (23) is started. As in (Wang et al., 2001), a lookup table of power
profiles in each visited area is stored in the MN’s database.
Basically, each visited zone size depends on the rate of change of the received power signals.
For example, Lzone = 50 m is typical for a macro-cell with a slow average power variation, and
Lzone = 10 m for a microcell with a fast signal change. Then, each zone size has pre-measured
signal means and standard deviations for the serving cell and the neighbouring cells.
Figure 9 shows how a visited zone is built. The MN is in position P1 =(x1, y1), while
P2 = (xc, yc) is the centre of a WLAN/UMTS cell. We can evaluate the angle α between the
line from P1 and P2 and the horizontal plane, as:

⎛ y − yc ⎞
α = arcsin ⎜⎜ (25)
⎟⎟ ,
⎝ P1 P2 ⎠
Connectivity Support in Heterogeneous Wireless Networks 241

where P1 P2 is the distance from P1 and P2 obtained according to the Euclidean formula. The
angle α is adopted to get the power attenuation, as we assume that the WLAN/UMTS cell
radius rWLAN/UMTS strictly depends on a factor γ (α), that modifies the cell radius value, as:

rWLAN /UMTS (α ) = γ (α ) ⋅ rWLAN /UMTS . (26)

The factor γ (α) is expressed as:

⎛ α − ϕ kWLAN /UMTS ⎞
γ (α ) = 1 + 0.8 ⋅ sin ⎜ ⎟⎟ , (27)
⎜ 2
⎝ ⎠
where ϕ k represents the k-th WLAN/UMTS cell down-tilt value, such as

360
ϕ kWLAN /UMTS = k ⋅ , (28)
N WLAN /UMTS
which depends on the number of WLAN/UMTS cells NWLAN/UMTS, (i.e. 10 and 3 WLAN
access points and UMTS base stations, respectively). So, the factor γ (α) is in the range [0.2,
1.8], and rWLAN/UMTS (α) will be decreased or increased of 80% of rWLAN/UMTS.

P1 = ( x , y )

P2 = ( xc , yc )

wireless cell

Fig. 9. Trigonometric approach for path loss map building in a (circular) wireless cell
environment

3.3.1 Power-based approach for hybrid vertical handover

The Power-based VHO approach is exploited by the HVHO technique during the power
maps building phase. Particularly, from mobile switch-off up to the completion of the power
maps of both the UMTS and WLAN networks, the mobile node uses the PB-VHO approach
to guarantee seamless connectivity (Inzerilli & Vegni, 2008). It performs handover using
power measurements only, and does not take account of location information.
With the PB-VHO scheme the MN selects a network access, either UMTS or WLAN, and
keeps it till the received power from the current network drops below the receiver
sensitivity. Hence, the other network is scanned in order to verify if a handover to the other
network can be done. Namely, if the power from the other network exceeds the receiver
sensitivity, a handover to the new network is executed. In case power from both networks is
below the minimum sensitivity, power scanning in both networks is continued repeatedly
till one of the two networks exhibit a power value above its sensitivity threshold.
242 Recent Advances in Wireless Communications and Networks

Power scanning frequency is limited in order to preserve battery charge as well as to

prevent the ping-pong effect. When the mobile switches on it attempts selecting the WLAN
network interface. Namely, if the measured power from the WLAN network interface card
is above the value of MN WLAN receiver sensitivity, then the WLAN connectivity is
available and the WLAN access is selected. Otherwise, if the measured power from the
UMTS network interface card is above the value of MN UMTS receiver sensitivity, UMTS
connectivity is available and the UMTS access is selected. When both checks fail, the mobile
node waits a waiting time pause before re-trying the WLAN network scanning again.

3.3.2 Enhanced location-based approach for hybrid vertical handover

In the location-based approach presented in Subsection 3.2 we have assumed WLAN/UMTS
circular cells. When the mobile terminal accesses a power map of the visited area
representing the WLAN coverage area, as well as the UMTS coverage area, it is possible to
derive a more accurate assessment of the GP in the UMTS and WLAN links, respectively. In
this subsection we introduce a modified version of the location-based approach as described
in Subsection 3.2, that is suitable for non-circular wireless cells. In particular, we consider a
variable radius for each cell k-th, such as the cell radius becomes a function of the angle αk
between the horizontal axe and the axe connecting the border of the cell with its centre. This
approach is then called as Enhanced Location-based (ELB) VHO, for non-circular wireless
cells. The ELB-VHO approach is exploited in order to obtain a characterization of the wireless
cell geometry coming from the power map building phase, as depicted in Figure 10.

PM ( xM , yM )

Pc ( xc , yc ) α

wireless cell

Fig. 10. Trigonometric approach for path loss map building in (anisotropic) wireless cell
environment
The boundary of a cell in the power map is identified by a set of values of the power
approaching the network sensitivity μ. When such values are not available in the power
map they can be obtained through polynomial interpolation. The centre of the cell is instead
identified with the maximum power value.
Once the boundary and centre of each cell k-th in the power map has been identified as a set
of points, it is possible to assign an angle αk to each point of the boundary with respect of the
centre of the cell and its distance Rcell (αk). The list of radius Rcell (αk) for each cell k-th is
exploited by the ELB scheme.
Connectivity Support in Heterogeneous Wireless Networks 243

It follows that the maximum GP in a WLAN and UMTS cell can be calculated with the
following approximated formulas, which replace (17) as:

⎧ ⎧⎛ UMTS γ ⎫
⎪GPUMTS = BW UMTS ⋅ Pr ⎪⎜ Rcell (α ) ⎟ + δ A < 1⎪
⎞
⎪ max max ⎨ ⎜ UMTS ⎟ d ⎬
⎪ ⎪⎩⎝ r ⎠ ⎭⎪
⎨ (29)
⎧⎛ R WLAN α ⎞ γ ⎫
⎪ ⎪ ( ) ⎪
WLAN
⎪GPmax WLAN
= BWmax ⋅ Pr ⎨⎜ cellWLAN ⎟ + δ Ad < 1⎬
⎪ ⎜ r ⎟
⎪⎩⎝ ⎠ ⎭⎪
⎩
However, the handover decisions are still taken on the basis of (19).

4. Conclusions
In this chapter we described the main aspects of vertical handover procedure. This
mechanism is oriented to ensure and maintain service continuity for mobile users in
heterogeneous wireless network environments. Three different vertical handover strategies
have been investigated, such as (i) the multi-parameter QoS-based approach, (ii) the
location-based algorithm and (iii) the hybrid vertical handover technique.
The multi-parameter QoS-based VHO assumes both subjective and objective video quality
metrics as handover decision criterion, such as a vertical handover is initiated whenever the
QoS level is decreasing under a fixed threshold. In the location-based VHO the mobile node
position is exploited in order to estimate some network performance (i.e. goodput figure). A
handover is then initiated whenever a selected candidate network guarantees higher
performance than the serving network.
Finally, we illustrated the third vertical handover technique (i.e. HVHO), that is an hybrid
approach based on both power measurements and location information. The HVHO
develops an enhanced location-based approach to build and maintain path loss maps, which
provides an updated description of the wireless cells in a visited environment. The use of
combined location and power information to drive handover decisions brings about
goodput enhancements, while assuring a limited VHO frequency with respect to simple
single-parameter techniques.

5. References
Makhecha K. P. & Wandra K. H. (2009). 4G Wireless Networks: Opportunities and Challenges,
Annual IEEE India Conference (INDICON), pp.1-4, December 2009.
Lin M.; Heesook Choi; Dawson T. & La Porta T. (2010). Network Integration in 3G and 4G
Wireless Networks, Proceedings of 19th International Conference on Computer
Communications and Networks (ICCCN), pp.1-8, August 2010.
Balasubramaniam S. & Indulska J. (2004). Vertical handover supporting pervasive computing in
future wireless networks, Computer Communications, Vol. 27, Issue 8, pp. 708–719, 2004.
Knightson K.; Morita N. & Towle T. (2005). NGN architecture: generic principles, functional
architecture, and implementation, IEEE Communication Magazine, Vol. 43, Issue 10,
pp. 49–56, October 2005.
McNair J. & Fang Z. (2004). Vertical handovers in fourth-generation multinetwork
environments, IEEE Wireless Communications, Vol. 11, Issue 3, pp. 8–15, June, 2004.
244 Recent Advances in Wireless Communications and Networks

Pollini G. P. (1996). Trends in handover design, IEEE Communication Magazine, Vol. 34, No.
3, March 1996, pp. 82–90.
Inzerilli T. & Vegni A. M. (2008). A reactive vertical handover approach for WiFi-UMTS
dual-mode terminals, Proceeding of 12th Annual IEEE International Symposium on
Consumer Electronics, April 2008, Vilamoura (Portugal).
Ayyappan, K. & Dananjayan, P. (2008). RSS Measurement for Vertical Handover in
Heterogeneous Network, Journal of Theoretical and Applied Information Technology,
Vol. 4, Issue 10, October 2008.
Vegni A. M.; Carli M.; Neri A. & Ragosa G. (2007). QoS-based Vertical Handover in
heterogeneous networks, Proceeding on 10th International Wireless Personal
Multimedia Communications, December 2007, Jaipur (India).
Yang K.; Gondal I.; Qiu B. & Dooley L. S. (2007). Combined SINR based vertical handover
algorithm for next generation heterogeneous wireless networks, Proceeding on IEEE
GLOBECOM 2007, November 2007, Washinton (USA).
Vegni A. M.; Tamea G.; Inzerilli T. & Cusani R. (2009). A Combined Vertical Handover
Decision Metric for QoS Enhancement in Next Generation Networks, Proceedings of
IEEE International Conference on Wireless and Mobile Computing, Networking and
Communications 2009, pp. 233–238, October 2009, Marrakech (Morocco).
Kibria M. R.; Jamalipour A. & Mirchandani V. (2005). A location aware three-step vertical
handover scheme for 4G/B3G networks, Proceeding on IEEE GLOBECOM 2005, Vol.
5, pp. 2752–2756, November 2005, St. Louis (USA).
Kim W. I.; Lee B. J.; Song J. S.; Shin Y. S. & Kim Y. J. (2007). Ping-Pong Avoidance Algorithm
for Vertical Handover in Wireless Overlay Networks, Proceeding of IEEE 66th
Vehicular Technology Conference, pp. 1509-1512, September 2007.
Inzerilli T.; Vegni A. M.; Neri A. & Cusani R. (2008). A Location-based Vertical Handover
algorithm for limitation of the ping-pong effect, Proceedings on 4th IEEE International
Conference on Wireless and Mobile Computing, Networking and Communications,
October 2008, Avignon (France).
Gupta V.; Williams M. G.; Johnston D. J.; McCann S.; Barber P. & Ohba Y. (2006) IEEE 802.21
Overview of Standard for Media Independent Handover Services, IEEE 802 Plenary,
San Diego, CA, USA, July 2006.
Golmie N.; Olvera-Hernandez U.; Rouil R.; Salminen R. & Woon S. (2006). Implementing
Quality of Service based handovers using the IEEE 802.21 framework, IEEE 802.21
session 15 San Diego, California, July 2006.
Shin J.; Kim J. W. & Kuo C. C. J. (2001). Quality-of-Service Mapping Mechanism for Packet
Video in Differentiated Services Network, IEEE Transactions on Multimedia, Vol. 3,
no. 2, June 2001.
Vegni A. M. (2010). Multimedia Mobile Communications in Heterogeneous Wireless Networks -
Part 2, PhD thesis, University of Roma Tre, March 2010, available online at
http://www.comlab.uniroma3.it/vegni.htm
Wang S. S.; Green M. & Malkawi M. (2001). Adaptive Handoff Method Using Mobile
Location Information, Proceedings on IEEE Emerging Technology Symp.
Broadband Comm. for the Internet Era Symposium, pp. 97-101, September 2001.
Inzerilli T.; Vegni A. M.; Neri A. & Cusani R. (2010). A Cross-Layer Location-Based
Approach for Mobile-Controlled Connectivity, International Journal of Digital
Multimedia Broadcasting, vol. 2010, 13 pages, 2010.
 12

On the Use of SCTP in Wireless Networks

Maria-Dolores Cano
Department of Information Technologies and Communications
Technical University of Cartagena
Spain

1. Introduction
Communications networks, particularly Internet, allow starting new businesses, to improve
the current ones, and to offer an easiest access to new markets. Nowadays, Internet connects
millions of terminals in the world, and it is a goal that this connection could be done with
anyone, at any moment, and anywhere. In order to achieve this target, new lax and varied
access requirements are needed. It is expected that a user would be able to access network
services in a transparent way disregarding the location. The user terminal could seamlessly
use the best available access technology (e.g., WLAN (Wireless Local Area Networks), LTE
(Long Term Evolution), or PLC (Power Line Communications)), and service provisioning
should agree with the user contract. This convergence of communications networks is
giving rise to new challenges. The Internet Protocol (IP) has been selected to provide the
necessary interconnection among all wireless and wired existing technologies. However, the
use of IP does not solve all drawbacks. Multimedia applications show that current transport
protocols like TCP (Transmission Control Protocol) or UDP (User Datagram Protocol) are
not good enough to meet the new quality requirements.
To face these new challenges, the IETF (Internet Engineering Task Force) defined a new
transport protocol called Stream Control Transmission Protocol (SCTP) (Stewart, 2007),
whose main features are multihoming and multistreaming. Multistreaming allows
transmission of several data streams within the same communication, splitting the
application data into multiple streams that have the property of independently sequenced
delivery, so that message losses in any one stream will only initially affect delivery within
that stream, and not delivery in other streams. On the other hand, multihoming allows
binding one transport layer’s association to multiple addresses at each end of the SCTP
association. The binding allows a sender to transmit data packets to a multihomed receiver
through one of those different destination addresses. Therefore, SCTP is not only intended
for signaling, but it can be used for any data application transport. The first studies about
the performance of SCTP showed promising results. For instance, in (Kamal et al., 2005),
authors evaluate the benefits of using SCTP instead of TCP as the underlying transport
protocol for a MPI (Message Passing Interface) middleware. Darche et al. (2006) presented a
network architecture to enhance the cooperation of mobile and broadcast networks using
SCTP as the transport layer protocol. In (Shaojian et al., 2005), authors study the suitability of
SCTP for satellite networks. Kim et al. investigate in (Kim et al., 2006) the applicability of
SCTP in MANET (Mobile Ad hoc NETworks). In (Kozlovszky et al., 2006), authors carry out
246 Recent Advances in Wireless Communications and Networks

performance measurements with TCP and SCTP as protocols to be used in distributed

cluster environments. Finally, in (Natarajan et al., 2006) authors propose the use of SCTP for
HTTP-based applications, showing the benefits with real web servers compatible with
SCTP. All these works showed the notable performance of SCTP as a multipurpose
transport layer protocol.
This chapter reviews the specific use of SCTP in wireless networks and illustrates how to
implement a multipurpose SCTP client/server application, compatible with IPv6, from a
practical point of view. We describe how to enable multistreaming and multihoming
capabilities. Through experimental tests in wired and wireless networks, we measure the
SCTP performance regarding multistreaming and multihoming operation, compare it with
the TCP protocol, and discuss its advantages and drawbacks. Therefore, the main
contribution of this chapter is to present a survey in the work carried so far to turn the SCTP
into a feasible transport-protocol option for wireless networks and to show the practical
aspects of the design of a SCTP’s open source client/server application, including some
basic, but explanatory, experimental results in a single server – single client scenario. This
work reveals that SCTP may be a competitive transport protocol for multimedia
applications.
The rest of the chapter is organized as follows. Section 2 reviews the SCTP characteristics
and its applicability in wireless networks. Section 3 explains how to make a SCTP
client/server application. Experimental results are shown and discussed in Section 4. The
chapter ends with conclusions in Section 5.

2. Related work
The SCTP features are described in this section. In addition, a survey about the applicability
of SCTP in wireless environments has been also included. Among the advantages of using
SCTP in wireless networks, mobility and multimedia transmission are highlighted,
reviewing the most relevant works in these two areas. Other improvements like security or
the introduction of redundancy for data delivery are also mentioned.

2.1 Stream control transmission protocol

SCTP is a message oriented transport protocol. Like TCP, SCTP provides a reliable transport
service ensuring that data arrives in sequence and without errors. Like TCP, SCTP is a
session-oriented mechanism, meaning that a relationship is created between the endpoints
of a SCTP association prior to data being transmitted, and this relationship is maintained
until all data transmission has been successfully completed. However, SCTP includes some
new features (see Table 1) that evidence the advantages of using it in applications needing
transport with additional performance and reliability.
Multihoming. A SCTP endpoint has the ability to work with more than one IP address, thus
a session can remain active even in the presence of network failures. One of the main
advantages is that in a conventional single-homed session, the failure of a local area network
access can isolate the end system, but with multi-homing, redundant local area networks
can be used to reinforce the local access. Multi-homing is not used for redundancy, as
indicated in (Stewart, 2007). A pair of IP addresses <source, destination> is defined as the
primary path, being used for data transmission. The other combinations of source and
destination addresses will be considered as alternative paths, and will be employed in case
of a primary path failure, which is detected by using the heartbeat mechanism (monitoring
On the Use of SCTP in Wireless Networks 247

function). The IP addresses of the SCTP association could be exchanged even if the
association is already in use, i.e., it is possible to include new IP addresses during the
communication (Stewart et al., 2007). This feature is known as Dynamic Address
Reconfiguration or Mobile SCTP.

Characteristics TCP UDP SCTP

Unicast Yes Yes Yes
Byte oriented Yes No No
Message oriented No No Yes
Reliable transport service Yes No Yes
Multi-homing No No Yes
Multi-stream No No Yes
Cookie mechanisms No No Yes
Rate adaptive Yes No Yes
Heartbeat mechanism No No Yes
Table 1. TCP, UDP, and SCTP comparison
Heartbeat Mechanism. A SCTP source should check if it is possible to reach the remote
endpoint. This is done by means of the heartbeat mechanism. Alternative paths are
monitored with heartbeat messages. Heartbeat messages are small messages with no user-
data periodically sent to the destination addresses, and immediately acknowledged by the
destination. The sender of a heartbeat message should increment a respective error counter
of the destination address each time a heartbeat is sent to that address and not
acknowledged within the corresponding time interval (RTO, Retransmission TimeOut). If
this counter reaches a maximum value, the endpoint should mark this address as inactive.
On the contrary, upon the receipt of a heartbeat acknowledgement, the sender of the
heartbeat should clear the error counter of the destination address to which the heartbeat
was sent, and mark the destination address as active.
Multistreaming. This feature allows splitting the application data into multiple streams that
have the property of independent sequenced delivery, so that message losses in any one
stream will only initially affect delivery within that stream, and not delivery in other
streams. This is achieved by making independent data transmission and data delivery.
SCTP uses a Transmission Sequence Number (TSN) for data transmission and detection of
message losses, and also a Stream ID/Stream Sequence Number pair, which is used to
determine the sequence of delivery of received data. Therefore in reception, the end point
can continue to deliver messages to the unaffected streams while buffering messages in the
affected stream until retransmission occurs.
Initiation. SCTP initiation procedure requires four messages. A cookie mechanism was
incorporated to avoid Denial of Service (DoS) attacks. A SCTP client sends an init message
to the SCTP server. The server replies with an init ack message that includes a cookie (a TCB
(Transmission Control Block), a validity period, and a signature for authentication). Since
the init ack is addressed to the source IP address of the init message, an attacker cannot get
the cookie. A valid SCTP client would get the cookie, and send it back in a cookie echo
message to the server. When this packet is received, the server starts giving resources to the
client. The procedure finishes with a cookie ack message.
Data Exchange. Data exchange in SCTP is very similar to the TCP SACK procedure (Stewart,
2007). SCTP uses the same congestion and stream control algorithms as TCP.
248 Recent Advances in Wireless Communications and Networks

Shutdown. SCTP shutdown procedure uses three messages: shutdown, shutdown ack, and
shutdown complete. Each endpoint has an ack of the data packets received by the remote
endpoint before closing the connection. SCTP does not support a half open connection, but it
is assumed that if the shutdown initiates, then both endpoints will stop transmitting data.

2.2 SCTP in wireless networks

Seamless mobility is one of the challenges in wireless networks. With the proliferation of
new types of wireless access technologies (e.g., WiFi, WiMAX, 3G, vehicular networks, etc.),
a user, through his/her mobile device, should be able to change his/her location
maintaining the Quality of Service (QoS) performance disregarding the roaming, either
horizontal (under the same technology) or vertical (crossing different technologies). SCTP is
a competitive solution for mobility due to its multihoming capability. Multimedia
transmission is another challenge in wireless networks due to the higher likelihood of
packet losses (error-prone channels). In this case, SCTP multistreaming improves the data
rate throughput since streams are independently delivered; hence, the multimedia
application is less sensitive to packet losses. Finally, some new modifications to SCTP have
been presented in the related literature to increase its performance, e.g., allowing
redundancy in multihomed devices. This section reviews the most relevant works in these
areas.

2.2.1 Mobility and handovers in wireless networks

Several works in the related literature had demonstrated the advantages of using SCTP to
improve both vertical or horizontal handovers and signaling in wireless networks. Authors
in (Afif et al., 2006a) proposed to include a new type of chunk in SCTP able to send QoS
transmission parameters over the radio interface from an EGPRS mobile to the SCTP peer.
By doing so, SCTP could adapt the transmission rate depending on the radio transmission
conditions (e.g., LLC error rate, RLC/MAC block error rate, etc.). The reason to incorporate
this new chunk, as stated by the authors, can be explained as follows. Even though SCTP is
able to change the IP addresses in use, data packets are sent to old IP address before the
alternative addresses become the primary ones. Therefore, there are packet losses during the
exchange process. The simulation study in an EGPRS network with handovers between cells
showed that the achieved throughput is higher with this modification than with the
standard SCTP implementation because fewer packets are lost during handovers. From a
similar perspective, same authors verified in (Afif et al., 2006b) that their modification is also
useful for handovers between EGPRS and Wireless Local Area Networks (WLAN).
Honda et al. proposed a new handover mechanism based on SCTP and a new data
retransmission feature for smooth handover. In their work, authors state that the exchange
of addresses in SCTP, assuming the new addresses to use are unknown at the beginning of
the SCTP association (i.e., using Dynamic Address Reconfiguration), suffers a high delay
mainly due to the multiple RTO expirations required to identify the failure. To overcome
this situation, authors propose to include two algorithms called FastAssociation
Reconfiguration and Fast Transmission Recovery. The former minimizes the RTO needed to
substitute the addresses in use, whereas the latter allows sending data just after the
establishment of the new addresses. Observe that in the standard, it was necessary to wait
an RTO after a new path is configured to send data. The evaluation, carried out in an
experimental network with WLAN links, showed that the handover latency was notably
reduced using the authors’ approach.
On the Use of SCTP in Wireless Networks 249

Focusing on vertical handover between WLAN and cellular networks, particularly UMTS
(Universal Mobile Telecommunication System), authors in (Ma et al., 2007) proposed a very
interesting error recovery scheme called Sending-buffer Multicast-Aided Retransmission
with Fast Retransmission that increases the throughput achieved during the SCTP
connection in the presence of forced vertical handovers from WLAN to UMTS. A forced
vertical handover occurs when the mobile node leaves the WLAN coverage due to the loss
of signal and switches to the cell network. The advantages of using SCTP for vertical
handovers were clearly identified in (Ma et al., 2004): higher throughput, shorter delay, a
simpler network architecture, and ease to adapt network congestion and flow control
parameters to the new network; but a scenario with forced handovers involves important
packet losses. Ma, Yu & Leung (2007) categorized these packet losses as dropping
consecutive packets because of the loss of signal (WLAN) and random packet losses over the
cellular link. To deal with these different types of errors, the authors propose to use two
solutions. First, packet losses due to the loss of signal enable the Sending-buffer Multicast-
Aided Retransmission algorithm, which multicast all buffered data on both the primary and
the alternate address (observe that in a standard implementation SCTP only retransmits
data to the alternate address if the error was due to a time out). The same applies to new
data that needs to be sent. Second, packet losses likely due to random packet losses over the
link (detected by the reception of duplicated acknowledgments) activate the Fast
Retransmission algorithm, which force the retransmission to be done to the same destination
IP address. With these two algorithms, long waiting delays are avoided, thus increasing the
achieved throughput. Working on the same heterogeneous scenario with WLAN and UMTS
networks, Shieh et al. (2008) detected that SCTP significantly decreases the congestion
window when new primary addresses are used in the SCTP association (i.e., during a
handover). Therefore, they proposed to assign an adequate initial congestion window
according to the bandwidth available in the new path, so the association can skip the slow-
start phase and enter the congestion avoidance phase directly. Packet-pair bandwidth
proving is used to estimate the available bandwidth in the new path. Authors demonstrated
the feasibility and goodness of their proposal through simulation. From an experimental
point of view, authors in (Bokor et al., 2009) designed and implemented a real native IPv6
UMTS-WLAN testbed to evaluate the effect of SCTP parameter configuration in terms of
handover effectiveness, link changeover characteristics, throughput, and transmission delay.
Among the most important parameters that have an effect on handover are: RTO.Min,
RTO.Max, Path.Max.Retransmission, and HB.Interval. Authors verified that with the standard
parameters, the handover delay would rise exponentially due to RTO redoubling, but using
a more appropriate setting the handover delay rises linearly when the RTO is incremented.
They also recommended keeping the HB.Interval (the time that elapses between
consecutive heartbeat monitoring messages) as low as possible. Finally, they found that the
SCTP performance in terms of delay, jitter, and throughput was better in UMTS than in
WLAN.
From another perspective, authors in (Lee et al., 2009) studied a mobile web agent
framework based on SCTP. Typical web agents use TCP as transport protocol. However,
mobile web agents using TCP present the following drawbacks: performance degradation,
head-of-line (HOL) blocking, and unsupported mobility (as identified by IEEE Std 802.11-
1997 and IEEE 802.16e-2005). By transmitting each object in a separate stream, SCTP solves
the HOL problem. Mobility is achieved by the SCTP multihoming capability. To improve
the performance, authors assumed that mean response time between HTTP requests and
250 Recent Advances in Wireless Communications and Networks

replies is the most important performance parameter in a web environment. Therefore, they
proposed to use SCTP to decrease the response time compared to the classical TCP
implementation of web agents. Authors described the complete architecture for the mobile
SCTP web agent framework. By simulation, they found that the mean response time
decreased notably (around 30%) by using SCTP. The mean packet loss was also smaller with
SCTP, and the faster the moving speed the better the SCTP performance in terms of packet
loss compared to TCP.
Regarding the option of introducing crosslayer techniques to combine the SCTP features
with information available at lower levels, the IEEE introduced the IEEE 802.21-2008 Media
Independent Handover (MIH) as a way to provide link layer intelligence and other related
network information to upper layers. MIH does not carry out the network handover, but it
provides information to allow handover within a wide range of networks (e.g., WiFi,
WiMAX, 3G, etc.). In (Fallon et al., 2009) authors proposed to separate path performance
evaluation (i.e., how SCTP detects that a path is no longer available) from path switching
(i.e., update the new addresses of the primary path in the SCTP association). Whereas the
first task will be done with MIH, SCTP will only be in charge of the second task (path
performance is disabled in SCTP). By simulation, authors demonstrated that the
combination of SCTP and MIH reacts to sudden performance degradation resulting from
obscured line of sight in a heterogeneous scenario with WiMAX and HSDPA technologies.
Indeed, the throughput of the SCTP connection improved notably (from 5% to 45%)
compared to the standard SCTP implementations.
Network Mobility (NEMO), commonly used in military or vehicular applications, has been
also studied from a SCTP perspective. In host mobility, a network in which terminals
change their location, mobility is managed through the mobile node itself. In a mobile
network, mobility is managed by a central node (e.g., a bus providing a WLAN service that
moves around a city, hence changing the access point from which obtains Internet access).
Leu & Ko (2008) proposed a method that combines SIP and SCTP with the aim of
minimizing delay and packet losses during the handovers of a mobile network. With the
authors’ proposal, packet losses decreased significantly. Similarly, Huang & Lin (2010)
presented a method to improve the bandwidth use and the achieved throughput in
vehicular networks by using SCTP. Their approach is explained as follows. In a Vehicle to
Infrastructure network (V2I), moving vehicular nodes communicate with Road Side Units
(RSU) deployed in a specific area. RSU are connected to the wired infrastructure, e.g.,
providing Internet access to mobile vehicular nodes. Usually, several RSU share the same
gateway to access the infrastructure. Therefore, authors proposed to use this gateway as a
SCTP-packet monitoring station, buffering all SCTP packets containing data chunks. In the
event of a packet loss, the gateway (not the destination node, which is assumed to be in the
wired part of the network) will be in charge of retransmitting lost packets in the wireless
link. With this scheme, the wired part of the communication is used more efficiently because
no retransmissions are sent (unless the packet loss occurs in the wired part of the network).
Moreover, since the destination node is not informed about packet losses in the wireless part
of the network, its congestion window does not decrease as much, keeping a higher
throughout rate in average. The performance of this proposal was done through simulation.
Authors verified that the achieved throughput, the transmission time, and the congestion
window behaved better with their approach than with the standard SCTP
implementation.
On the Use of SCTP in Wireless Networks 251

2.2.2 Multimedia transmission over wireless networks

The use of multimedia services and applications over wireless links is another important
research area. Authors in (Wang et al., 2003) presented one of the first works evaluating the
performance of Partial Reliability SCTP (PR SCTP), a modification of SCTP that provides
unreliable transmission service to part of the data to be sent, as the transport protocol for
video (MPEG-4) transmission in a wireless local area network. Results showed an
improvement in the video quality comparing PR SCTP with UDP. Another interesting
works regarding MPEG-4 video transmission over wireless technologies are presented in
(Nosheen et al., 2007) and (Chughtai et al., 2009). In the first work, authors compared SCTP
with UDP and DCCP (Datagram Congestion Control Protocol) (Kohler et al., 2006). By
simulation, they found that the throughput achieved by UDP could be more than 20%
smaller than the throughput achieved by SCTP or DCCP in a wireless environment.
However, the delay was higher in SCTP due to the congestion control mechanism. In the
presence of background traffic, the results also showed that SCTP and DCCP outperformed
UDP. As an extension to this work, Chughtai et al. (2009) carried out a similar study to
compare the QoS performance of SCTP, UDP, and SCTP transmitting video in a WiMAX
network. The simulation scenarios included downloading or uploading MPEG-4 video
traffic using a different number of subscribers, different packet sizes, and a variable video
rate. Results showed that delay and jitter were lower with SCTP than with UDP or DCCP. In
terms of throughput, DCCP performed slightly better than SCTP, and both exceeded UDP
performance.
Wang et al. (2008) also studied video delivery over wireless networks using SCTP. They
focused on the multistreaming feature of SCTP, and how to use it to optimize video quality.
Previous works from the literature such as (Balk et al., 2002) showed the benefits of using
multistreaming for MPEG-4 video transmission in wired network by applying a differential
treatment among streams in a SCTP association. Differing from previous works, Wang et al.
(2008) proposed MPEG-4 transmission with optimized partial reliability among streams in a
heterogeneous scenario with error-prone 802.11 wireless channels. Their proposal was based
on retransmitting packets belonging to stream of I-frames until packets are eventually
received, while no retransmissions are attempted for packets in stream of B- and P- frames.
In terms of retransmission overhead delay, simulation results showed that adjusting SCTP
fast retransmit threshold can reduce the retransmission overhead delay, hence increasing
the I-frame data rate, and the video quality. Furthering the results obtained in this work, the
same authors introduced in (Wang et al., 2009) an extension to the SCTP protocol. The goal
was to improve the transmission of delay sensitive multimedia data by including a selective
retransmission of lost packets depending on whether the lost packets would still arrive
before the schedule time. Assuming that there is clock synchronization between the SCTP
associated peers, authors included a new field to the SCTP header with the time a packet is
sent, so that the endpoint after reception can estimate the one-way delay. This value is sent
to the sender from the receiver in the acknowledgement packet. Then, in the receiver side,
the time of each frame of MPEG-4 to be played out is calculated, so if the frame is not
received before this schedule time will be considered as non-useful and its retransmission
will not be necessary. By simulation, authors achieved interesting results, confirming the
improvement in the MPEG-4 video transmission performance.
Voice over IP (VoIP) is another important application that is gaining momentum. Chang et
al. (2009) presented a middleware to transfer the session initiation protocol (SIP) signaling
and real-time transmission protocol (RTP) messages from using UDP or TCP to SCTP.
252 Recent Advances in Wireless Communications and Networks

Switching from UDP or TCP to SCTP (with Dynamic Address Reconfiguration) provides a
seamless way for the user to roam maintaining the QoS level of the VoIP call. Authors
analyzed their proposal in a real testbed. Nevertheless, results showed that although
mobility was achieved, the delay was higher with their proposal.
Live TV broadcasting over wireless technologies could also benefit from the use of SCTP.
Liu et al. (2010) introduced a method to provide an economic way of live news broadcasting
by using SCTP. Satellite News Gathering (SNG) vehicles, which usually use satellite links
for transmission, are an expensive service for TV companies, mainly due to the required
equipment. In this case, the current deployment of WiMAX networks is a feasible alternative
to satellite communication, but the bandwidth offered by WiMAX is not enough to provide
a live TV service with QoS demands. Therefore, the authors proposed to take advantage of
all available wireless networks, not only WiMAX but also HSDPA or WiFi, thus increasing
the available bandwidth. A SCTP multi-link connection with both multihoming and multi-
streaming was a key point for this implementation. SCTP Concurrent Multipath Transfer,
which will be explained in next section, is also needed. With an experimental testbed,
authors demonstrated the feasibility of their proposal, not only achieving a cost-effective
system to provide live TV broadcasting but also increasing the coverage of previous SNG
systems.

2.2.3 Other SCTP improvements

Concurrent Multipath Transfer (CMT) consists of simultaneously sending data over all
available paths, hence, increasing the bandwidth of the SCTP association (Iyengar et al.,
2006). In environments where the paths of the SCTP association exhibit very different
network conditions (e.g., round trip times or bandwidth), packet reordering is required in
the receiver side, and this might cause retransmission, lowering the connection rate. To
avoid this situation, authors in (Perotto et al., 2007) compared the performance of two SCTP
modifications: Sender-Based Packet Pair SCTP (SBPP-SCTP) and Westwood SCTP (W-
SCTP). The former uses the sender-based packet pair technique, mentioned in the previous
section, to estimate the bottleneck bandwidth of each path. The latter uses the same
algorithm as in TCP Westwood (Mascolo et al., 2004) for the bandwidth estimation. Both aim
at minimizing packet reordering. In presence of intermittent interfering cross-traffic, authors
showed that W-SCTP achieves a higher throughput than SBPP-SCTP. Aydin & Shen (2009)
studied the performance of CMT SCTP over 802.11 static multihop wireless networks. They
compared CMT SCTP with three different techniques: i) standard SCTP using just one path
(the best one in terms of bandwidth) to send data, ii) standard SCTP using just one path (the
worst one in terms of bandwidth) to send data, and iii) standard SCTP using all available
paths to send data (splitting the traffic into the different available paths of the SCTP
association). Results showed that in a multihop wireless scenario the achieved throughput is
higher with CMT SCTP than with any of the three alternatives used for comparison.
Nevertheless, CMT SCTP still presents a drawback to be completely useful for wireless
networks: the received buffer blocking problem. This problem was clearly stated in (Wang et
al., 2010): “In SCTP transmission, data streams between each other are logically
independent, if receiver has received all data chunks of a certain stream. The data of this
stream can be delivered to the application layer. But in traditional CMT, because data
chunks of the same stream maybe transferred to different paths, the data chunks could not
arrive at the receiver orderly and duly, so the receive buffer blocking problem happens. This
problem can seriously influence network performance, especially in high error rate and
On the Use of SCTP in Wireless Networks 253

delay wireless network.” Consequently, authors proposed a new modification of the SCTP
called Wireless Concurrent Multipath Transfer SCTP (WCMT SCTP). With this modification,
each SCTP path delivers packets belonging to the same stream (one or more than one). For
instance, if there are three paths available and there are five streams, then the first path only
transmits packets from the first stream, the second path only transmits packets from streams
two and three, and the third path only transmits packets from streams four and five.
Authors also added other changes to the standard CMT implementation: a per-path
congestion control mechanism, a new congestion control mechanism and a new
retransmission mechanism that takes into account the type of error. Results obtained by
simulation showed that WCMT SCTP performs better than CMT SCTP in ad hoc networks.
In a similar way, Yuan et al. (2010) improved the CMT SCTP mechanism by categorizing the
streams depending on their specific QoS requirements, and grouping those streams with
similar QoS needs in subflows that are sent through the more appropriate paths available in
the SCTP association. Finally, the work done in (Xu et al., 2011) showed how to optimize
CMT SCTP for video and multimedia content distribution.
Another interesting works that improve the performance of SCTP in wireless environments
from different perspectives are (Cui et al., 2007; Cano et al., 2008; Lee & Atiquzzaman, 2009;
Cheng et al., 2010; Funasaka et al., 2010). Cui et al. (2007) proposed to use a hierarchical
checksum method that improves the retransmission procedure, thus increasing the achieved
throughput in links with high packet losses. Cano et al. (2008) investigated how to combine
the use of IPSec (Internet Protocol Security) with SCTP to enhance the security of the
wireless communication. The work done in (Lee & Atiquzzaman, 2009) presented an
analytical model to estimate the delay of HTTP over SCTP in wireless scenarios. Last, Cheng
et al. (2010) proposed to use two new methods for bandwidth estimation and per-stream-
based error recovery.

Library Description
netinet/sctp.h It contains definitions for SCTP primitives and data
structures.
netdb.h It contains definitions for network database operation, e.g.
translation.
sys/socket.h It defines macros for the Internet Protocol family such as the
datagram socket or the byte-stream socket among others.
netinet/in.h It contains definitions of different types for the Internet
Protocol family, e.g. sockaddr_in to store the socket
parameters (IP address, etc.).
arpa/inet.h To manage numeric IP addresses, making available some of
the types defined in netinet/in.h
Table 2. Description of the libraries related to SCTP network communication

3. Implementation
For the sake of simplicity, we implement three SCTP client/server applications. The first one
is called single SCTP, the second one is called multistream SCTP, and the last one is called
multihomed SCTP. Single SCTP is very similar to TCP, since it will be able to transmit just
254 Recent Advances in Wireless Communications and Networks

one data stream between source and destination endpoints. Multistream SCTP includes
multistreaming, and finally, multihomed SCTP incorporates multihoming. The three
implementations are written in C code. We use the libraries provided by the Berkeley Socket
Application Programming Interface, which are briefly described in Table 2. Next sections
detail the practical SCTP implementation issues.

1 int main(int argc, char *argv[])

2 {
3 int sockfd;
4 struct hostent *host;
//Structures to manage IP address
5 struct sockaddr_in remote_addr; //IPv4
6 host = gethostbyname(argv[1]);
7 ra_family = host->h_addrtype; //AF_INET
//IPv4 socket
8 sockfd = socket( ra_family, SOCK_STREAM, IPPROTO_SCTP);
9 if(sockfd == -1)
10 {perror("Socket:");exit(1);}
//Set server IP address
11 remote_addr.sin_family=AF_INET;
12 remote_addr.sin_port=htons(REM_PORT);
13 remote_addr.sin_addr=*((struct in_addr *)host->h_addr);
14 bzero(&(remote_addr.sin_zero),8);
//Connect to server
15 if(connect(sockfd,(struct sockaddr*)&remote_addr,sizeof(struct sockaddr))==-1)
16 {perror("connect:"); exit(1);}
//Omitting lines of code to receive a file
//Close socket
17 close(sockfd);
18 return 0;
19 }

Fig. 1. Extract of the original SCTP client code in a single file transmission

3.1 Single SCTP

SCTP server and SCTP client structures are very similar to those used in TCP. Fig. 1 shows
how to implement a SCTP client. The only difference with TCP is in the socket() function,
where the protocol type field should be IPPROTO_SCTP instead of the common parameter 0
used for TCP or UDP transport protocols (see code line 8 in Fig. 1). The rest of the
implementation is done as in TCP; i.e., once the socket is created, the server IP address is set
(see code lines 11-14 in Fig. 1), and the client connects to the server (see code line 15 in Fig.
1). Observe that we use the server IP address as an argument in the command line (see code
line 6 in Fig. 1). If we want to use IPv6 instead of IPv4, some simple changes included in
Table 3 are needed. First, it is necessary an appropriate structure to store an IPv6 address.
Second, the gethostbyname() function, needs to know that the IP address is an IPv6 one, and
the same applies to all lines of code where we use the IP address.
On the Use of SCTP in Wireless Networks 255

In Fig. 2, we define how to implement a SCTP server. In this case, to execute the server, no
parameters are needed in the command line. We define a constant called MYPORT to
include the port number associated to the server IP address (see code line 12 in Fig. 2). The
server IP address is automatically set to any local IP address available (see code line 13 in
Fig. 2). Then, we follow the usual sequence to set up the server. First, we create the socket
with the socket() function. As indicated before, the socket protocol is set to IPPROTO_SCTP
(see code line 15 in Fig. 2). Then, we set the socket parameters with the bind() function (see
code line 18 in Fig. 2). Afterwards, we execute listen() so that the server can receive a specific
number of client requests (see code line 20 in Fig. 2). The accept() function makes the server
to wait for client requests (see code line 25 in Fig. 2). Finally, if a client request is received,
the client is served by a child process due to the fork() function (see code line 27 in Fig. 2). To
make it compatible with IPv6, lines indicated in Table 4 should be replaced.

1 int main(int argc, char *argv[])

2 {
3 int sockfd, newfd;
4 socklen_t sin_size;
5 struct sockaddr_in local;
6 struct sockaddr_in remota;
7 struct hostent *host;
8 sa_family_t la_family;
9 la_family = host->h_addrtype;
10 host = gethostbyname(argv[1]);
11 local.sin_family = AF_INET;
12 local.sin_port = htons(MY_PORT);
13 local.sin_addr.s_addr = htonl(INADDR_ANY);//Any local IP address
14 bzero(&(local.sin_zero),8);
15 sockfd = socket( la_family, SOCK_STREAM, IPPROTO_SCTP);
16 if(sockfd == -1)
17 {perror("Socket:"); exit(1);}
18 if((bind(sockfd, (struct sockaddr*)&local, sizeof(struct sockaddr)))==-1)
19 {perror("bind");exit(1);}
20 if(listen(sockfd,5) == -1)
21 {perror("listen");exit(1);}
22 for(;;)
23 {
24 sin_size=sizeof(struct sockaddr_in);
25 if((newfd = accept(sockfd, (struct sockaddr*)&local,&sin_size)) == -1)
26 {perror("accept");exit(1);}
27 if (!fork()
//Omitting lines of code to send a file
28 while(waitpid(-1,NULL,WNOHANG)>0);}

Fig. 2. Extract of the original SCTP server code in a single file transmission
256 Recent Advances in Wireless Communications and Networks

Line# in Fig. 1 New code for IPv6

struct sockaddr_in6 remote_addr6; //IPv6
5
struct in6_addr ipv6; //To store IPv6 address
6 host = gethostbyname2(argv[1], AF_INET6); //get IP address
sockfd = socket( ra_family, SOCK_STREAM, IPPROTO_SCTP);
if(sockfd == -1) {perror("Socket:");exit(1);}
remote_addr6.sin6_family = AF_INET6;
8 to 14 remote_addr6.sin6_flowinfo = 0;
remote_addr6.sin6_port = htons(REM_PORT);
inet_pton(AF_INET6, argv[2], ipv6.s6_addr);
remote_addr6.sin6_addr = ipv6;
if(connect(sockfd,(struct sockaddr*)&remote_addr6,sizeof(struct
15
sockaddr))==-1)
Table 3. How to make the SCTP client implementation compatible with IPv6. Lines
indicated in the first column should be replaced with lines shown in the second column

Line# in Fig. 1 New code for IPv6

5-6 struct sockaddr_in6 local6;
struct sockaddr_in6 remota6;
10 host = gethostbyname2(argv[1], AF_INET6);
11 to 14 local6.sin6_family = AF_INET6;
local6.sin6_flowinfo = 0;
local6.sin6_port = htons(MY_PORT);
local6.sin6_addr = in6addr_any;
18 if((bind(sockfd, (struct sockaddr*)&local6, sizeof(struct
sockaddr))) == -1)
24-25 sin_size=sizeof(struct sockaddr_in6);
if((newfd = accept(sockfd, (struct sockaddr*)&local6,&sin_size))
== -1)
Table 4. How to make the SCTP server implementation compatible with IPv6

3.2 Multistream SCTP

A SCTP client/server application with multistreaming allows sending/receiving multiple
streams simultaneously. For instance, these different streams could belong to different files,
so it would be possible transferring several files with the same SCTP association. Thus, the
client only uses one request to the SCTP server. Nowadays, file downloading (music, games,
software, etc.) is one of the most important services driving the usage of Internet. With the
multistreaming SCTP feature, a unique association between SCTP client and SCTP server
may accept many multimedia file transmissions, resulting in bandwidth saving as it will be
shown in Section 4. The less traffic in the network, the more efficient the use.
From the SCTP client point of view, the multistream operation has to be enabled by setting
some particular properties. The sequence is as follows. First, we create all data structures.
Second, we create the SCTP socket as explained in the previous section (see code line 10 in
Fig. 3). Then, the maximum number of ingoing and outgoing streams should be indicated.
Accordingly, setsockopt() is used to set the number of flows or streams in the client/server
On the Use of SCTP in Wireless Networks 257

1 int main()
2 {
3 int connSock, in, ret;
4 struct sockaddr_in servaddr;
5 struct sctp_status status;
6 struct sctp_sndrcvinfo sndrcvinfo;
7 struct sctp_event_subscribe events;
8 struct sctp_initmsg initmsg;
9 int numElem=0, firstTime=1;
10 connSock = socket( AF_INET, SOCK_STREAM, IPPROTO_SCTP ); //IPv4
11 memset( &initmsg, 0, sizeof(initmsg) );
12 initmsg.sinit_num_ostreams = 30; //max streams
13 initmsg.sinit_max_instreams = 30; //max streams
14 initmsg.sinit_max_attempts = 5; //max attempts
15 ret = setsockopt( connSock, IPPROTO_SCTP, SCTP_INITMSG,&initmsg,
sizeof(initmsg) );
16 bzero( (void *)&servaddr, sizeof(servaddr) ); //server to connect to
17 servaddr.sin_family = AF_INET;
18 servaddr.sin_port = htons(MY_PORT_NUM);
19 servaddr.sin_addr.s_addr = inet_addr("192.168.1.10" );
20 ret = connect( connSock, (struct sockaddr *)&servaddr, sizeof(servaddr) ); //connect
to the server
21 memset( (void *)&events, 0, sizeof(events) );
22 events.sctp_data_io_event = 1;
23 ret = setsockopt( connSock, SOL_SCTP, SCTP_EVENTS,(const void *)&events,
sizeof(events) );
24 //File transfer
25 //Loop to receive the different streams
26 in = sctp_recvmsg( connSock, (void *)buffer, sizeof(buffer),(struct sockaddr *)NULL,
0, &sndrcvinfo, &flags );
27 if(in==0)
28 break;
29 //Store each stream in its corresponding file
30 if (sndrcvinfo.sinfo_stream == STREAM1)
31 { if(firstTime)
32 {fp=fopen("reciboweb.txt","wb"); firstTime=0;}
33 numElem=fwrite(buffer, 1, 1024, fp);
34 if(num_elementos<1024) {fclose(fp); break;} }
35 else if (sndrcvinfo.sinfo_stream == STREAM2) 42
36 {//Save this stream in its corresponding file, lines 31-41}
37 else if ...//Save each stream in its corresponding place
38 //End loop to receive different streams
39 fclose(fp);
40 close(connSock);
41 return 0; }

Fig. 3. Extract of the original SCTP client code in a multistream transmission

258 Recent Advances in Wireless Communications and Networks

SCTP association. Both client and server agree on this parameter (see code lines 12-15 in
Fig. 3). Afterwards, both the server IP address and the port number to connect to are
indicated (see code lines 16-19 in Fig. 3). Next, the client connects to the server (see code line
20 in Fig. 3). Finally, we enable data delivery with the function setsockopt() (see code lines 21-
23 in Fig. 3). By doing so, the client is able to use the primitive sctp_recvmsg() for data
delivery. At this point, the client is ready to receive data in multiple streams within the same
SCTP association.
On the other hand, the SCTP server multistream implementation also needs some variations
compared to the SCTP server single implementation. First, we declare data structures. After
that, we create the SCTP socket as explained in the previous section (see code line 8 in Fig.
4). Then, the server IP address is automatically set to any local IP address available, the port
is assigned, and the bind() function is called (see code line 9-13 in Fig. 4). The maximum
number of ingoing and outgoing streams is specified now (see code lines 15-18 in Fig. 4).
Observe that it is the same value used previously for the client implementation. Next, the
server remains listening for client requests (see code line 19 in Fig. 4). If there is a client
request, then the server accepts the connection, and it starts sending the corresponding files.
Once a client is connected to the server, the information sent from the server to the client
should be identified, so that the client knows what file (stream) the data belong to. Whereas
the source (the server in this case) is in charge of assigning an identifier to each stream,
which is done with the sctp_sendmsg()function and a stream number (see line 28 and 35 in
Fig. 4), each stream is identified using the sndrcvinfo.sinfo_stream field (see line 30 and 35 in
Fig. 3) in the receiving side (the client in this case).

3.3 Multihomed SCTP

In this section, we describe the additional code necessary to facilitate the SCTP multihomed
feature. After calling the bind() function and before the SCTP association is established, any
additional address should be enabled. Otherwise, multihoming cannot be used unless
Dynamic Address Reconfiguration is set. Enabling addresses is done with the sctp_bindx()
function. sctp_bindx() links any IP address (IPv4 or IPv6) to the SCTP association. It can be
also used to delete an IP address from an association. Table 5 shows the new lines of code.

New code for multihoming

hst_adicional = gethostbyname(argv[3]);//get additional address/es
sctp_bindx( sockfd, (struct sockaddr*)ip4, 1, SCTP_BINDX_ADD_ADDR);
Table 5. How to make the SCTP client/server implementation with multihoming

4. Experimental results
Three different scenarios are evaluated to compare the performance of SCTP vs. TCP. In the
first scenario, our SCTP application transfers a single text file, a single mp3 file, or a single
mpeg file from server to client. We called it the single operation. In the second scenario, our
SCTP application transmits different types of files simultaneously from server to client. We
called it the multistream operation. The former is like a normal TCP transfer file operation.
The latter could emulate a web loading, where usually different types of multimedia files
are involved. In the third scenario, we test the multihoming feature in what we called the
multihomed operation. Next we describe the experimental topology, and discuss the
experimental results.
On the Use of SCTP in Wireless Networks 259

1 int main()
2 {
3 int listenSock, connSock, ret, msglen;
4 struct sockaddr_in servaddr;
5 struct sctp_initmsg initmsg;
6 FILE *fp;
7 int num_bytes=0;
8 listenSock = socket( AF_INET, SOCK_STREAM, IPPROTO_SCTP );
9 bzero( (void *)&servaddr, sizeof(servaddr) );
10 servaddr.sin_family = AF_INET;
11 servaddr.sin_addr.s_addr = htonl( INADDR_ANY );
12 servaddr.sin_port = htons(MY_PORT_NUM);
13 ret = bind( listenSock, (struct sockaddr *)&servaddr, sizeof(servaddr) );
14 memset( &initmsg, 0, sizeof(initmsg) );
15 initmsg.sinit_num_ostreams = 30;
16 initmsg.sinit_max_instreams = 30;
17 initmsg.sinit_max_attempts = 5;
18 ret = setsockopt( listenSock, IPPROTO_SCTP, SCTP_INITMSG, &initmsg,
sizeof(initmsg) );
19 listen( listenSock, 5 );
20 int i=0;
21 while( 1 )
22 {
23 connSock = accept( listenSock, (struct sockaddr *)NULL, (int *)NULL );
24 fp = fopen("textoweb.txt","rb");
25 do
26 {
27 num_bytes=fread( (void *)buffer, 1,1024, fp);
28 ret = sctp_sendmsg( connSock, (void *)buffer, (size_t)strlen(buffer),NULL, 0, 0, 0,
STREAM1, 0, 0 );
29 }while(!feof(fp));
30 fclose(fp);
31 fp = fopen("vaquero.jpg","rb");
32 do
33 {
34 num_bytes=fread( (void *)buffer, 1,1024, fp);
35 ret = sctp_sendmsg( connSock, (void *)buffer, (size_t)strlen(buffer),NULL, 0, 0, 0,
STREAM2, 0, 0 );
36 }while(!feof(fp));
37 fclose(fp);
38 ...//Send each file with its corresponding stream identifier
39 }
40 return 0; }

Fig. 4. Extract of the original SCTP server code in a multistream transmission

260 Recent Advances in Wireless Communications and Networks

4.1 Experimental scenario

The experimental topology is illustrated in Fig. 5. We measure both the time required to
initialize the TCP or SCTP socket(s), and the time that it takes to transfer the file(s) with TCP
or SCTP. Tests are carried out with two laptops in a 10 Mbps wired Ethernet local area
network. Both laptops also have wireless cards to verify the multihoming feature. During
the tests, there was no other traffic in the network, but the one from these experiments.
Likewise, the only application running on the laptops is our TCP or SCTP application.
In the single operation tests, we transmit a 1 MB file from the server to the client through the
wired local area network, and repeat the experiment for a 3MB file, and a 50MB file. Each
transmission is repeated 100 times. In the multistream operation tests, the client should load
a multimedia web page from the server. Therefore, the client should download a variety of
multimedia files. Since we have not implemented a web server compatible with SCTP, we
carry out experiments assuming that the client downloads two or four multimedia files of
different sizes. Both tests (downloading two or four multimedia files) are performed 100
times. Experimental results have a confidence interval of 95% that has been calculated with
a normal distribution function using 100 samples.

Fig. 5. Experimental topology. Laptops have Intel Centrino platforms, Intel Pentium M
740/1.73 GHz processors, and 1GB RAM. Operating system is Linux (SuSe 10.0)

4.2 Results
Results from the single operation tests show that TCP is slightly faster than SCTP in a single
file transmission. Table 6 includes the average transmission time for single-file transmissions
with TCP and SCTP and the corresponding confidence intervals. For instance, we observe
that the transmission of a 3 MB file with SCTP lasts 2.73 seconds compared to the 2.6
seconds of TCP. SCTP is slower than TCP for two reasons. Firstly because its socket
initiation time is 1ms larger (it uses four packets, adding the effect of the cookie mechanism).
Secondly, the monitoring of the path that the SCTP carries out periodically (heartbeat
mechanism) also introduces some overhead. As a result, the SCTP transmission lasts
approximately 3% more than the TCP one.
Regarding the multistream operation, the first clear conclusion is that TCP requires more IP
packets to proceed with these transmissions. A TCP connection requires three packets for
negotiation and four packets for shutdown. Therefore, the more files to transmit with TCP
the more packets, because it is necessary to establish a different connection to download
each file (each stream) with TCP. Likewise, a SCTP association needs four packets for
negotiation and three for shutdown, however, SCTP will only require an association for
downloading multiple files. Fig. 6 shows the overhead amount produced with SCTP and
TCP, where the x axis represents the number of files to be transmitted and the y axis the
number of bytes used. We represent in this figure the number of bytes used in TCP for
initiation and shutdown, as well as the number of bytes consumed by SCTP in initiation,
shutdown, and heartbeat packets. For the heartbeat mechanism, we consider sending the
On the Use of SCTP in Wireless Networks 261

heartbeat signal every 100ms, 250ms, 500ms, and 1 s. Observe that the time interval for
sending the heartbeat is an adjustable parameter. Clearly, the more frequent the heartbeat
the more bandwidth consumed. Assuming the minimum possible packet sizes for TCP and
SCTP, and taking into account the SCTP heartbeat mechanism, the overhead introduced by
TCP would be smaller than the one introduced by SCTP only if the heartbeat is very
aggressive. Otherwise, the fact of establishing one TCP connection for each file transmission
produces higher bandwidth consumption.

1 MB file 3 MB file 50 MB file

TCP SCTP TCP SCTP TCP SCTP
Average transmission time (s) 1.06 1.09 2.60 2.73 47.11 48.52
Confidence interval 0.24 0.29 0.32 0.20 3.78 2.26
Table 6. SCTP vs. TCP average transmission times in single operation tests

Fig. 6. Overhead introduced by TCP and SCTP. For TCP, the packet size is 20 bytes (we
assume no data is sent with the first ACK packet). For SCTP, we take the following minimum
packet sizes as indicated in (Stewart, 2007): INIT 20 bytes, INIT ACK 20 bytes, COOKIE ECHO
8 bytes, COOKIE ACK 4 bytes, HEARTBEAT REQUEST 4 bytes, HEARTBEAT ACK 4 bytes
On the other hand, socket initiation is still faster in TCP. However, since more sockets need
to be used in TCP, the total initiation time difference between TCP and STCP is shorter and
shorter as the number of files to be transmitted increases. Fig. 7 and Fig. 8 represent the
duration of initiating sockets in TCP versus initiating sockets in SCTP. Indeed, when two
multimedia files are transmitted (Fig. 7), the average time dedicated to sockets initiation in
TCP is 1.69 ms, while the average time is 1.73 ms for SCTP. However, if we send four
multimedia files, the average time increases to 3.4 ms average in TCP whereas
approximately the same value remains in SCTP (Fig. 8). Thus, when four files are
transmitted, SCTP total initiation time is half of the TCP total initiation time. Consequently,
results show that not only the SCTP multiple file transmission is faster than the TCP one,
262 Recent Advances in Wireless Communications and Networks

but it consumes less bandwidth. Table 7 includes the average times for a multiple-file
transmission and the corresponding confidence intervals.

Fig. 7. Socket initiation time in TCP and SCTP in two-file downloading

Fig. 8. Socket initiation time in TCP and SCTP in four-file downloading

2 multimedia files 4 multimedia files

TCP SCTP TCP SCTP
Average Transmission time (s) 6.20 3.10 18.02 6.90
Confidence intervals 1.55 0.86 1.61 1.07

Table 7. SCTP vs. TCP average transmission times in multistream operation tests
On the Use of SCTP in Wireless Networks 263

Finally, we test the multihoming SCTP feature in the topology shown in Fig. 5, where two
PCs are connected to each other through two interfaces (one is wired, the other is wireless).
We use the multistreaming SCTP client and server implementations shown in Fig. 3 and
Fig. 4 respectively, including the new lines shown in Table 5. At first, client and server are
using the wired network (primary IP addresses). Then, one of the wired network interface
card is disabled. Experimental results show that in less than 1 second SCTP reacts in the
presence of the network failure by replacing primary IP addresses with the alternative one
(wireless one) to continue with the transmission. The time to change the IP addresses in use
includes the ARP resolution, which is almost negligible in this scenario. Table 8 shows the
exchange of IP addresses in use.

No. Time Source Destination Protocol Info

55559 185.79019 192.168.1.10 192.168.1.11 SCTP DATA
55560 185.79022 192.168.1.11 192.168.1.10 SCTP SACK
55561 185.79108 192.168.1.10 192.168.1.11 SCTP DATA
55562 185.79215 192.168.1.10 192.168.1.11 SCTP DATA
55563 185.79218 192.168.1.11 192.168.1.10 SCTP SACK
55564 185.79304 192.168.1.10 192.168.1.11 SCTP DATA
55565 185.99060 192.168.1.11 192.168.1.10 SCTP SACK
55570 186.79958 linuxpedro.local ARP who has 192.168.2.33?
Tell 192.168.2.34
55571 186.79959 192.168.2.33 ARP 192.168.2.33 is at
00:80:5a:32:cb:c0
55572 186.80009 linuxpedro.local ARP who has 192.168.2.33?
Tell 192.168.2.34
55573 186.80009 192.168.2.33 ARP 192.168.2.33 is at
00:80:5a:32:cb:c0
55574 186.81128 192.168.2.34 192.168.2.33 SCTP DATA
55575 186.81132 192.168.2.33 192.168.2.34 SCTP SACK
55576 186.83170 192.168.2.34 192.168.2.33 SCTP DATA
• Frame 55565 (64 bytes on wire, 64 bytes captured)
• Linux cooked capture
• Internet Protocol, Src: 192.168.1.11 (192.168.1.11), Dst: 192.168.1.10 (192.168.1.10)
• Stream Control Transmission Protocol, Src Port: 5200 (5200), Dst Port: 20000 (20000)
Source port: 5200
Destination port: 20000
Verification tag: 0x 52c9c5b0
Checksum: 0xe5b04b29 [correct CRC32C]
• SACK chunk (Cumulative TSN: 261016901, a_rwnd: 112640, gaps:0, TSNs: 0)
Table 8. Extract of the traffic captured with Wireshark (Wireshark, 2011). The first 6 SCTP
packets use the primary IP addresses. After the network failure (packet# 55565), alternative
addresses are used
264 Recent Advances in Wireless Communications and Networks

5. Conclusion
In this work, we have presented a survey with the most relevant works on the applicability
of SCTP in wireless networks. We have categorized the benefits of SCTP for wireless
technologies in the following categories: mobility and handovers, multimedia transmission,
and other improvements related to multiple path transmission or security. We have also
shown the practical aspects of the design of a SCTP client/server application. In our
example, the SCTP application is used to download files from a server. We have described
the basics of how to enable multihoming and multistreaming capabilities in SCTP. We have
observed that it is quite easy to adapt current applications to the SCTP protocol. When
comparing to TCP, the advantages of SCTP are numerous (e.g., faster average transmission
times and resources saving), above all in applications that require the transmission of
multiple files. Moreover, multihoming allows increasing reliability, a key additional
requirement in multimedia applications over wireless networks.

6. Acknowledgment
This research has been supported by the MICINN/FEDER project grant TEC2010-21405-
C02-02/TCM (CALM).

7. References
Afif, M., Martins, P., Tabbane, S., & Godlewski, P. (2006a). Radio aware SCTP extension for
handover data in EGPRS. Proceedings 17th Annual IEEE International Symposium on
Personal, Indoor and Mobile Radio Communications PIMRC'06, pp. 1-5.
Afif, M., Martins, P., Tabbane, S., & Godlewski, P. (2006b). SCTP Extension for
EGPRS/WLAN Handover Data. Proceedings 31st IEEE Conference on Local Computer
Networks, pp. 746-750.
Aydin, I., & Shen, C.-C. (2009). Performance Evaluation of Concurrent Multipath Transfer
Using SCTP Multihoming in Multihop Wireless Networks. Proceedings 8th IEEE
International Symposium on Network Computing and Applications, pp. 234-241.
Balk, A., Sigler, M., Gerla, M., & Sandidi, M. Y. (2002). Investigation of MPEG-4 video
streaming over SCTP. Proceedings 6th World Multiconference on Systemics. Cybernetics.
and Informatics SCI’02, pp. 1-4.
Begg, C. L., Pawlikowski, K., Sirisena, H., & De Silva, P. (2007). Suitability of SCTP for High
Quality Video Streaming over CDMA2000. Proceedings Australasian
Telecommunication Networks and Applications Conference, pp. 496-502.
Bokor, L., Huszák, A., & Jeney, G. (2009). On SCTP Multihoming Performance in Native
IPv6 UMTS–WLAN Environments. Proceedings 5th International Conference on
Testbeds and Research Infrastructures for the Development of Networks & Communities
and Workshops TridentCom’09, pp. 1-10.
Cano, M.-D., Romero, J.A., & Cerdan, F. (2008). Experimental Tests on SCTP over IPSec.
Proceedings IFIP International Conference on Network and Parallel Computing NPC’08,
pp. 96-102.
Chang, L.-H., Huang, P.-H., Chu, H.-C., & Tsai, H.-H. (2009). Mobility Management of VoIP
services using SCTP Handoff Mechanism. Proceedings Symposia and Workshops on
Ubiquitous, Autonomic and Trusted Computing, pp. 330-335.
On the Use of SCTP in Wireless Networks 265

Cheng, R.-S., D.-J., Chao, H.-C., & Chen, W.-E. (2010). An Adaptive Bandwidth Estimation
Mechanism for SCTP over Wireless Networks. Proceedings 5th International
Conference on Future Information Technology, pp. 1-5.
Chughtai, H. M. O., Malik, S. A., & Yousaf, M. (2009) Performance Evaluation of Transport
Layer Protocols for Video Traffic over WiMax. Proceedings IEEE 13th International
Multioptic Conference, pp. 1-6.
Cui, X., Cui, L., & Koh, S. J. (2007). A Hierarchical Checksum Scheme for SCTP over Wireless
Networks with Worse Channel Condition. Proceedings International Conference on
Wireless Communications, Networking and Mobile Computing WiCom’07, pp.1845-1848.
Darche, D., Kopp, R., Mazieres, B., Lepage, F., & Gnaedinger, E. (2006). Using SCTP to improve
performances of hybrid broadcast/telecommunication network system. Proceedings of
IEEE Consumer Communications & Networking Conference, Vol. I, pp. 371-375.
Fallon, E., Murphy, L., & Murphy, J. (2009). Optimizing Metropolitan Area Wireless Path
Selection Using Media Independent Handover. Proceedings Second International
Workshop on Cross Layer Design IWCLD'09, pp. 1-5.
Honda, M., Sakakibara, H., Nishida, Y., & Tokuda, H. (2007). SmSCTP: A Fast Transport
Layer Handover Method Using Single Wireless Interface. Proceedings 12th IEEE
International Symposium on Computers and Communications ISCC’07, pp.319-324.
Huang, C.-M., & Lin, M.-S. (2010). RG-SCTP: Using the Relay Gateway Approach for
Applying SCTP in Vehicular Networks. Proceedings IEEE International Symposium on
Computers and Communications ISCC’10, pp. 139-144.
IEEE Std 802.11-1997. (1997). IEEE 802.11 wireless LAN medium access control (MAC) and
physical layer (PHY) specifications.
IEEE Std. 802.16e-2005. (2005). IEEE Standard for Local and metropolitan area networks.
Part 16: Air interface for fixed broadband wireless access systems.
IEEE Std. 802.21-2008. (2008). IEEE Standard for Local and metropolitan area networks. Part
21: Media Independent Handover Services.
Iyengar, J. R., Amer, P., & Stewart, R. (2006). Concurrent multipath transfer using SCTP
multihoming over independent end-to-end paths. IEEE/ACM Transactions on
Networking, Vol. 14, No. 5, pp. 951–964.
Kamal, H., Penoff, B., & Wagner, A. (2005). SCTP versus TCP for MPI. Proceedings of
ACM/IEEE SuperComputing Conference SC’05, pp. 30-44.
Kim, D., Song, J., Kim, J., Yoo, H., Park, J., & Cano, J.C. (2006). The Applicability of SCTP to
Mobile Ad Hoc Networks. Proceedings International Conference on Advanced
Communication Technology ICACT’06, Vol. 3, pp. 1979-1984.
Kohler, E., Handley, M., & Floyd, S. (2006). Datagram Congestion Control Protocol. RFC 4340.
Kozlovszky, M., Berceli, T., & Kutor, L. (2006). Analysis of SCTP and TCP based
communications in high speed clusters. Nuclear Instruments and Methods in Physics
Research Section A, Vol. 559, Issue 1, pp.85-896.
Lee, Y.-J. & Atiquzzaman, M. (2009). Mean Waiting Delay for Web Object Transfer in
Wireless SCTP Environment. Proceedings IEEE International Conference on
Communications ICC’09, pp. 1-5.
Lee, Y.-J., Lee, D.-W., & Atiquzzaman, M. (2009). Novel web agent framework to support
seamless mobility for data networks. IET Communications Journal, Vol. 3, No. 12, pp.
1861–1869.
Leu, F.-Y. & Ko, Z.-J. (2008). A Novel Network Mobility Scheme Using SIP and SCTP for
Multimedia Applications. Proceedings of International Conference on Multimedia and
Ubiquitous Engineering, pp. 564-569.
266 Recent Advances in Wireless Communications and Networks

Liu, H.-S., Hsieh, C.-C., Chen, H.-C., Hsieh, C.-H., Liao, W., Chu, P.-C., & Wang, C.-H.
(2010). Exploiting Multi-link SCTP for Live TV Broadcasting Service. Proceedings
IEEE 71st Vehicular Technology Conference, pp. 1-6.
Ma, L., Yu, F., Leung, V. C. M., & Randhawa, T. (2004). A new method to support
UMTS/WLAN vertical handover using SCTP. IEEE Wireless Communications, Vol.
11, No. 4, pp. 44-51.
Ma, L., Yu, F. R., & Leung, V. V. M. (2007). Performance Improvements of Mobile SCTP in
Integrated Heterogeneous Wireless Networks. IEEE Transactions on Wireless
Communications, Vol. 6, No. 10, pp. 3567-3577.
Mascolo, S., Grieco, L. A., Ferorelli, R., Camarda, P., & Piscitelli, G. (2004). Performance
evaluation of Westwood+ TCP congestion control. Performance Evaluation, Vol. 4,
No. 55, pp. 93–111.
Natarajan, P., Iyengar, J. R., Amer, P. D., & Stewart, R. (2006). SCTP: An innovative transport
layer protocol for the web. Proceedings 15th International World Wide Web Conference,
WWW’06, pp. 615-624.
Nosheen, S., Malik, S. A., Zikria, Y. B., & Afzal, M., K. (2007). Performance Evaluation of
DCCP and SCTP for MPEG4 Video over Wireless Networks. Proceedings IEEE 11th
International Multitopic Conference, pp. 1-6.
Perotto, F., Casetti, C., & Galante, G. (2007). SCTP-based Transport Protocols for Concurrent
Multipath Transfer. Proceedings IEEE Wireless Communications and Networking
Conference WCNC’07, pp. 2969-2974.
Shaojian, F., Atiquzzaman, M., & Ivancic, W. (2005). Evaluation of SCTP for space networks.
IEEE Wireless Communications, Vol. 12, No. 5, pp. 54-62.
Shieh, C.-S., Lin, I-C., & Lai, W. K. (2008). Improvement of SCTP Performance in Vertical
Handover. Proceedings of Eighth International Conference on Intelligent Systems Design
and Applications, pp. 494-498.
Stewart, R. (2007). Stream Control Transmission Protocol. RFC 4960.
Stewart, R., Xie, Q., Tuexen, M., Maruyama, S., & Kozuka, M. (2007). Stream Control
Transmission Protocol (SCTP) Dynamic Address Reconfiguration. RFC 5061.
Wang, B., Feng, W., Zhang, S.-D., & Zhang, H.-K. (2010). Concurrent multipath transfer protocol
used in ad hoc networks. IET Communications Journal, Vol. 4, No. 7, pp. 884–893.
Wang, H., Jin, Y., Wang, W., Ma, J., & Zhang, D. (2003). The performance comparison of
PRSCTP, TCP and UDP for MPEG-4 multimedia traffic in mobile network. Proceedings
International Conference on Communication Technology (ICCT), Vol. 1, pp. 403-406.
Wang, L., Kawanishi, K., & Onozato, Y. (2008). MPEG-4 Optimal Transmission over SCTP
Multi-streaming in 802.11 Wireless Access. Proceedings 7th Asian-Pacific Symposium
on Information and Telecommunication Technologies, pp. 172-177.
Wang, L., Kawanishi, K., & Onozato, Y. (2009).Achieving Robust Fairness of SCTP Extension
for MPEG-4 Streaming. Proceedings 20th Personal, Indoor and Mobile Radio
Communications Symposium PIMRC '09, pp. 2970-2974.
Wireshark. <http://www.wireshark.org>. Last visited March 20th, 2011.
Xu, C., Fallon, E., Qiao, Y., Zhong, L., & Muntean, G.-M. (2011). Performance Evaluation of
Multimedia Content Distribution Over Multi-Homed Wireless Networks. IEEE
Transactions on Broadcasting, Vol. PP (99), pp. 1-12.
Yuan, Y., Zhang, Z., Li, J., Shi, J., Zhou, J., Fang, G., & Dutkiewicz, E. (2010). Extension of
SCTP for Concurrent Multi-Path Transfer with Parallel Subflows. Proceedings IEEE
Wireless Communications and Networking Conference WCNC’10, pp. 1-6.
 0
13

Traffic Control for Composite Wireless Access

Route of IEEE802.11/16 Links
Yasuhisa Takizawa
Kansai University
Japan

1. Introduction
The expansion and diversification of wireless communications are proceeding rapidly
with the diffusion of cellular phones, WiFi and WiMAX. However, concern is increasing
that the growth of wireless systems will exhaust finite wireless resources. Cognitive
radio technology(Mitorall & Maguire, 1999; Mitoralll, 1999; Harada, 2005), which has been
proposed as a solution to this problem, aims to optimize the utilization of diverse
wireless resources. Furthermore, AIPN (All-IP Network) (3GPP, 2005) and NGN (Next
Generation Network)(ITU, 2006) investigate the network architecture that accommodates
diverse communication media. Accordingly, we expect that in the near future, wireless access
networks will be composed of diverse wireless medias.
To exploit wireless media diversity in expected access networks, some bandwidth-aggregation
methods in wireless media have recently been proposed. Bandwidth-aggregation combines
diverse communication links in parallel and suitably distributes packets to communication
links. The works(Phatak & Goff, 2002; Snoeren, 1999; Shrama et al., 2007) aggregate wireless
links in IP to improve IP throughput. The work(Chebrou & Rao, 2006) also aggregates
wireless links in IP to decrease IP delay based on wireless media that provide a bandwidth
guarantee. The works(Hsieh et al., 2004; Zhang et al., 2004) aggregate communication links
in a transport layer to improve TCP throughput. Meanwhile, wireless access networks
process traffic of diverse application, and the traffic is classified by the following two types of
application traffic:
• Traffic of throughput-oriented application such as FTP and Web on TCP.
• Traffic of delay-oriented application such as VoIP and Video Conference on UDP.
Therefore, wireless access networks are required to provide high throughput and low
delay by diverse applications. The above works do not consider delay except for
the work(Chebrou & Rao, 2006), and the work(Chebrou & Rao, 2006) does not consider
IEEE802.11 that no bandwidth guarantee is provided. Furthermore, the works(Phatak & Goff,
2002; Snoeren, 1999; Shrama et al., 2007; Chebrou & Rao, 2006) improve IP performance, but
can not provide effective improvement of application performance because they do not
consider out-of-order packets which occur by the packet distribution to multiple links. The
works(Hsieh et al., 2004; Zhang et al., 2004) consider the out-of-order packet, and can improve
the performance of TCP application, but can not improve that of UDP application such as VoIP
and Video Conference.
268
2 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

802.11a/b 802.16
Transmission Rate 54Mbps/11a, 11Mbps/11b 75Mbps
Coverage 50m/11a, 100m/11b 1000m
CSMA/CA TDD/FDD
Access Control
(Decenteralized) (Centeralized)
Bandwidth Guarantee No Yes
Table 1. Performance of wireless systems.

In this chapter, assuming the expected wireless access network to be composed of IEEE802.11,
which is a popular wireless system, and IEEE802.16, which is expected to spread, a IP
packet distribution on the access route, which combines IEEE802.11-link and IEEE802.16-link
in parallel, is proposed to improve the application performance. The proposed packet
distribution increases IP throughput and decreases IP delay. Furthermore, it reduces
out-of-order packets and provides high throughput and low delay to both UDP applications
and TCP applications simultaneously.
Our works(Takizawa et al., 2008; Takizawa, 2008) have proposed the packet distribution
for combining IEEE802.11/16 wireless upload links. We expand the above packet
distribution to reduce out-of-order packets and to apply download traffic, and show
its essential characteristics of packet distribution for composite wireless access route of
IEEE802.11/16-links (call M-route) , then propose a packet distribution method for M-route.
Furthermore, we evaluate the method’s performance by multiple application traffic on both
UDP and TCP in a wireless access network composed of 802.11a, 802.11b and 802.16, which
have the different characteristic from each other (see Table 1).
The configuration of wireless access networks by wireless media diversity is assumed as
follows (see Fig. 1).

Network

16-Coverage Base Station

16-antenna
16-link
11a-link
11b-link
11b-Coverage

11a/b-antenna 11a/b-antenna

11a-Coverage 11a-Coverage

11b-Coverage

Fig. 1. Assumed wireless access network.

• Base station provides an access point function of IEEE802.11a/b-wireless systems and a

base station function of 16-wireless system, and accommodates IEEE802.11a/b-antennas
and an IEEE802.16-antenna by wired connecting. It also provides the function of gateway.
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 2693

• Each terminal is equipped with IEEE802.11a/b-interfaces and IEEE802.16-interface, and

can communicate with base station by using each interface.
• IEEE802.11a/b-antennas and terminals are randomly deployed within coverage of
IEEE802.16-antenna.
• The access network is IP network.

2. Characteristics of IEEE802.11 link for packet distribution

In this section, based on Media Access Control (MAC) of IEEE802.11 DCF, the characteristics
of IEEE802.11 wireless link (11-link) for packet distribution is analyzed.

2.1 IEEE802.11 link cost

Based on queuing theory(Gross & Harris, 1985), a link load is shown as the number of packets
in a link, including waiting packets in the queue and the currently processed packet. d( i,k) ,
which is cost of link k between a terminal i and a base station, is defined as the link load, and
it is expressed using Little’s theorem(Little, 1961) as follows.

d( i,k) = F( i,k) · T( i,k) (1)

where F( i,k) is the packet arrival rate of link k in terminal i and T( i,k) is the average delay of
link k in terminal i. Delay is the time from packet arrival at the terminal to completion of
packet transmission, therefore the delay is composed of a waiting delay in queue and an air
time. The air time is composed of MAC delay and transmission delay, which take the MAC
retransmission into consideration.
Based on Eq. (1), T( i,k) decreases if d( i,k) decreases on constant F( i,k) and on maximum of d( i,k) ,
that is, link capacity, F( i,k) can increase if T( i,k) decreases. F( i,k) corresponds to a throughput on
condition that no packet loses. Therefore, when d( i,k) decreases, a throughput increases and a
delay decreases on a link.
The dependence of the link cost on the packet arrival rate, which corresponds to the number
of distributed packets in unit time to a link, is shown. Based on Eq. (1), the link cost depends
on the average delay. The average delay is composed of the waiting delay in queue and the
packet service time. Therefore, in regard with 11-link, the dependence of the above elements
on the packet arrival rate are shown, and in summarizing them, the dependence of the link
cost on the packet arrival rate is shown.

2.1.1 Dependence of packet service time on packet arrival rate

In (Bianchi, 2000), throughput analysis of IEEE802.11 DCF is shown, and in
(Carvalho & Garcia, 2003), the packet service time analysis of that is shown based on
(Bianchi, 2000). According to these, the dependence of the average packet service time on the
packet arrival rate is shown.
DCF adopts an exponential backoff scheme, and employs a discrete-time backoff timer. The
timer immediately following a Distributed InterFrame Space (DIFS) starts, and a terminal,
which is a terminal or a base station, is allowed to transmit only at the beginning of each
Slot Time. The Slot Time size σ is set equal to the time needed at any terminal to detect the
transmission of a packet from any other terminal. At each packet transmission, the backoff
timer is randomly chosen in the range (0, CW − 1). CW is called Contention Window, and
depends on the number of transmissions failed for the packet. At the first transmission
attempt, CW is set equal to CWmin called minimum contention window. After each failed
transmission, CW is doubled, up to a maximum value CWmax = 2r CWmin (r is a maximum
270
4 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

Transmission Attempt Period

Transmission Waiting Period in range (0, 20CWmin-1)

i
i
i
i
i
i
i
i
f
stage 0

b
c
i

Transmission Attempt Period

Transmission Waiting Period in range (0, 21CWmin-1)

i
f
stage 1

Transmission Attempt Period

Transmission Waiting Period in range (0, 2iCWmin-1)

i
s
stage i

i: idle b: busy c: collision f: transmission fail s: transmission success

Fig. 2. Exponential binary backoff in IEEE802.11.

number of retransmissions). Each transmission attempt is referred to as a bakoff stage. The

packet service time is the sum of time for each backoff stage. Each backoff stage is composed of
the transmission waiting period and the transmission attempt period (see Fig. 2). The backoff
stage starts in the transmission waiting period, and the backoff timer is initialized to a random
value in the range (0, CWi − 1) at the backoff stage i start. CWi is the contention window size
of the backoff stage i. In the period, the backoff timer is decremented only when the channel
is idle, and it is frozen when the channel is busy. The duration of the period is the time until
the backoff timer becomes zero from initial value. The transmission attempt period starts
when the backoff timer reaches zero, and a packet transmission takes place. The duration of
period is the time to transmit a packet. In the model of (Bianchi, 2000) and (Carvalho & Garcia,
2003), a fixed number of terminals is assumed, and the backoff stage is repeated until a packet
transmission success using CWi until stage r and using CWr beyond stage r. The stage r is
called maximum backoff stage. Furthermore, using the probability τ that a terminal transmits
in a randomly chosen slot time, the following probabilities in an exponential backoff scheme
are expressed.

ptr = 1 − (1 − τ )n−1
( n − 1) τ (1 − τ ) n − 2
psuc =
ptr
pi = 1 − ptr (2)
ps = ptr · psuc
pc = ptr (1 − psuc )
q = (1 − τ ) n −1
where n is the number of terminal in the channel coverage, ptr is the probability that there
is at least one transmission in the slot time of the transmission waiting period, psuc is the
probability that a transmission occurring on the channel is successful, pi is the probability that
the slot time is idle in the transmission waiting period, ps is the probability that the channel
is busy due to a packet transmission success in the transmission waiting period, pc is the
probability that the channel is busy due to a collision in the transmission waiting period, and
q is the probability that a packet transmission success in the transmission attempt period. Let
B be the average time which the transmission waiting period takes until a packet transmission
succeeds, and let A be the average time which the transmission attempt period takes until a
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 2715

Source
RTS Data

SIFS DIFS
Destination
CTS SIFS SIFS
ACK

Others

NAV(RTS)
Others(Hidden)

NAV(CTS)

Fig. 3. RTS/CTS access control sequence in IEEE802.11.

packet transmission succeeds, B and A are derived from a binary exponential backoff scheme
as follows(Carvalho & Garcia, 2003). (Note: In this section, "time" is the duration in slot time
units σ of IEEE802.11)

tb (ηCWmin − 1)
B= (3)
2q
tb = pi ti + ps ts + pc tc
q − 2r (1 − q )r +1 (4)
η=
1 − 2(1 − q )
1−q
A= tc + ts (5)
q
ti = 1
ts = RTS + SIFS + δ + CTS + SIFS + δ + H
(6)
+ PL + SIFS + δ + ACK + DIFS + δ
tc = RTS + DIFS + δ
where ti is the time of idle (i.e., one backoff slot), ts is the average time that the channel is
sensed busy due to a packet transmission success, tc is the average time that the channel is
busy due to a collision in the channel, RTS, CTS and ACK are time that RTS, CTS and ACK
frame is transmitted respectively, SIFS and DIFS are the interval time (see Fig. 3), δ is the
propagation delay, H is the time that a packet header is transmitted, and PL is the time the
payload is transmitted. According to Eq. (2), q = 1 − ptr , therefore, tb /q expresses the average
time that the backoff timer is decreased by one, and (ηCWmin − 1)/2 expresses the average
of sum of backoff timer in all stage. In Eq. (5), (1 − q )/q expresses the average number of
collision in the transmission attempt priod.
Then, the average packet service time S is argued using the above analysis. S is shown as
follows.

S = B+A (7)
When the number of terminal is constant, the dependence of S on τ is shown using the first
and second derivative of S at τ as follows.

dS d2 S
>0 >0 (8)
dτ dτ 2
Therefore, S is a convex monotonically increasing function of τ. Figure 4(a) illustrates the
dependence of S on τ by using Eq. (7) in 11b MAC parameter, transmission rate 11Mbps, a
272
6 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

number of terminals 10∼40, maximum backoff stage 5, and the payload size 1500 bytes, and
it also shows the same characteristics.
4
1x10 0.02
8
F in n=10
1x10

Packet Arrival Rate (packet/slot)

F in n=20
Maximum Backoff Stage 5

Packet Service Time (slot /packet)

Payload Size 1500bytes
Average Packet Service Time

8000
7
0.015 8x10

2
F in n=30

(slot/packet)

2nd Moment of
6000 7
6x10

0.01 n=40 n=30 n=20 n=10

F in n=40

4000 7
4x10
n=40 n=30 n=20 n=10

0.005
2000 2x10
7

1/S in n=10

Maximun Backoff Stage 5
1/S in n=20

Payload Size1500bytes 1/S in n=40
1/S in n=30

0 0 0
0 0.02 0.04 0.06 0.08 0.1 0 0.01 0.02 0.03 0.04 0.05 0 0.01 0.02 0.03 0.04 0.05
Tx Attempt Rate (Tx/slot) Tx Attempt Rate (Tx/slot) Tx Attempt Rate (Tx/slot)

(a) Dependence of S on τ. (b) Dependence of F on τ. (c) Dependence of S2 on τ.

Fig. 4. Dependence of each element on τ.

In (Bianchi, 2000) and (Carvalho & Garcia, 2003), the transmission queue is assumed to be
always non-empty, thus, the dependence of τ on the packet arrival rate F is not considered.
Let F be the number of arrival packets at a link in a slot time, the dependence is argued. The
average number of arrival packets in period S is FS, and the average number of transmission
attempts on a successfully transmitted packet is (1 − q )/q + 1. Then, the average number of
that a packet transmission attempts in period S is FS/q. Therefore, τ is shown as follows.

FS F
τ= = (9)
qS q
Figure 4(b), which illustrates the dependence of F on τ using Eq. (9) in the same parameter as
Fig. 4(a). In Fig. 10, when F < 1/S (1/S is the packet service rate), that is, when the load does
not exceed the link capacity, and when the number of terminal is constant, F for τ is concavely
and monotonically increasing. Therefore, within link capacity, the dependence of F on τ is
shown using the first and second derivative of F at τ as follows.

dF d2 F
>0 <0 (10)
dτ dτ 2
˙ (10) as follows.
Furthermore, the first and second derivative of S on F is shown using Eqs.(8),

dS dS dτ dS 1
= =   >0 (11)
dF dτ dF dτ dF
dτ
2 2  2
d S d S dτ dS 1
= −   >0 (12)
dF 2 dτ 2 dF dτ d2 F
dτ 2
Therefore, within link capacity, S is a convex monotonically increasing function of F.

2.1.2 Dependence of waiting delay in queue on packet arrival rate

The dependence of W which is the waiting delay in queue on F is argued. NQ , which is the
number of waiting packets in queue, is F × W using Little’s theorem. W is composed of the
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 2737

packet service time for NQ packets and R, which is the sum of the residual service time in each
packet arrival. Consequently, W is shown as follows.

W = NQ · S + R = F · W · S + R (13)
Each residual service time in a packet arrival is S2 /2S(Bertsetkas
& Gallager, 1992), where S2
is the second moment of S. The average number of packet arrivals in S is FS; accordingly, R is
FS2 /2. Applying the above relations to Eq. (13), W is given as

FS2
W= (14)
2(1 − FS )
Let V [ S ] be the variance of S, and it is shown as follows(Carvalho & Garcia, 2003)
 2
tb (CWmin γ − 1) 1−q
V [S] = + tc
2 q2 (15)
[2q − 4q + 1 − r (−1 + 2q )q ][2(1 − q )]r + 2q2
2
γ=
(−1 + 2q )2
Using Eq. (15), S2 is shown as follows.

S2 = S + V (S ) (16)
Furthermore, using Eq. (16), the first and second derivatives of S2 at τ are shown, respectively,
as follows.

dS2 d2 S 2
>0 >0 (17)
dτ dτ 2
Figure 4(c) illustrates the dependence of S2 on τ using Eq. (16) in the same parameter as
Fig. 4(a), and it also shows the same characteristics. Furthermore, applying Eq. (10) to Eq. (17),
the first and second derivatives of S2 at F are shown, respectively, as follows.

dS2 d2 S 2
>0 >0 (18)
dF dF 2
Using Eqs. (14) (18), the first and second derivatives of W at F are shown, respectively, on the
condition of FS < 1, as follows.

dW d2 W
>0 >0 (19)
dF dF 2
FS < 1, that is, F < 1/S expresses the condition that a link load is with a link capacity.
Therefore, within a link capacity, W is also a convex monotonic increasing function of F.

2.1.3 Dependence of 11-link cost on packet arrival rate

Finally, the dependence of the 11-link cost on the packet arrival is argued. The average delay
T is also a convex monotonic increasing function of F because of T = W + S. Applying the
dependence of T on F to Eq. (1), the first and second derivatives of a 11-link cost d at F are as
follows.
274
8 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

dd d2 d
>0 >0 (20)
dF dF 2
Consequently, a 11-link cost d is also a convex monotonic increasing function of F within a
link capacity and in a fixed number of terminals.

2.2 Cost of M-route compositing multiple 11-links for upload traffic

On communications using a M-route which aggregates multiple 11-links from terminal to
a base station , the cost of M-route for upload traffic is the sum of cost of each 11-uplink
composing M-route because the number of packets in a M-route is the sum of the number of
packets in each link composing M-route. Therefore, mi which is the cost of M-route for upload
traffic in terminal i is shown as follows (see Fig. 5(a)).

mi = ∑ d( i,x ) (21)
x ∈Ui
Ui is the set of an uplink which is provided by a 11-wireless interface equipped with terminal i.
Here, in steady packet arrival rate, the packet distribution from an 11-uplink k to an 11-uplink
j in M-route of terminal i, is argued. In this case, the packet distribution to the other 11-uplinks
is constant, thus the dependence of F( i,j) on F( i,k) is shown as follows.

dF( i,j) d2 F( i,j)

= −1 =0 (22)
dF( i,k) d( F( i,k))2
Using Eqs.(20) and (22), the first and second derivatives of d( i,j) at F( i,k) are shown as follows.

dd( i,j) dd( i,j) dF( i,j)dd( i,j)

= =− <0
dF( i,k) dF( i,j) dF( i,k)dF( i,j)
 2 (23)
d2 d( i,j) d2 d( i,j) dF( i,j) dd( i,j) d2 F( i,j)
= + >0
d( F( i,k))2 d( F( i,j))2 dF( i,k) dF( i,j) d( F( i,k))2
Consequently, d( i,j) is a convex monotonically decreasing function of F( i,k). According to
Eq.(21), mi is the sum of d( i,k) , which is a convex monotonically increasing function of F( i,k),
and d( i,j) , which is a convex monotonically decreasing function of F( i,k), and the uplink cost
of the others, which are constant for F( i,k). Therefore, mi is a convex function of F( i,k) (see
Fig.5(b)), and mi has a optimal solution for F( i,k).
Because mi is a convex function of F( i,k), the optimal solution can be searched by the packet
distribution which aim to descend the gradient in the convex function. When packets are
distributed from a 11-uplink k to 11-uplink j in M-route, the condition of the gradient descent
on M-route cost is shown as follows using Eq.(22).

dmi dd( i,k) dd( i,j)

= − >0 (24)
dF( i,k) dF( i,k) dF( i,j)
Applying Eq.(1) to Eq.(24), and transforming Eq.(24) into difference equation, thus the first
derivative of mi at F( i,k) is shown as follows.
   
dmi ΔT( i,k) ΔT( i,j)
= lim T( i,k) + F( i,k) − T( i,j) + F( i,j) >0 (25)
dF( i,k) ΔF( i,k ) →0 ΔF( i,k) ΔF( i,j)
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 2759

terminal i

mi
Composite Access Route of multiple links

mi = d(i,k ) + d(i, j )
interface a

F(i,a ) T(i,a ) link a

d(i,a ) Base Station

Cost
Queue

interface k
d(i, j )
Fi F(i,k ) T(i,k ) link k
d(i,k )
d(i,k ) Base Station

Queue

interface z

F(i,z) T(i,z) link z

d(i,z ) Base Station

Queue

F
(i,k)

(a) M-route cost for upload traffic. (b) Dependence of M-route cost
on packet distribution.

Fig. 5. M-route for upload traffic.

Furthermore, applying finite difference approximation to Eq.(25), the following is derived.

dmi (n )
≈ T( i,k) (n + 1) − T( i,j) (n + 1) > 0 (26)
dF( i,k)
Where, mi (n ) is M-route cost of terminal i in packet distribution of n time and T( x,y) (n + 1) is
average delay of 11-link y in terminal x in packet distribution of n + 1 time.
Consequently, when the packet distribution meets Eq.(26) which means the average delay
of source 11-uplink on packet distribution becomes larger than that of destination 11-uplink
on packet distribution, the M-route cost for upload traffic decreases and approaches the
optimal solution. Such packet distribution is repeated with the decrease in the amount of the
distributing packets (ΔF( i,k) → 0), and finally the average delay of source 11-uplink becomes
equal to that of destination 11-uplink, the M-route cost for upload traffic reaches its optimal
solution.
Furthermore, the search for the optimal solution of M-route cost has the additional
effectiveness which decreases the arrival of out-of-order packets because of the equalization
of the delay of source 11-link and destination 11-link.

2.3 Cost of M-route compositing multiple 11-Links for download traffic

A base station associates its 11-interface with multiple terminals in its coverage. Thus,
its interface is composed of multiple 11-downlinks according to multiple terminals in its
11-coverage, that is, its topology is point-to-multipoint. In this subsection, the cost of M-route
for download traffic (i.e. in a base station) in steady packet arrival rate is argued.
In queueing theory, a link has a queue of packets to be transmitted, and has an independent
server on other links within the same interface. However, an 11-downlink is different from
a link reserved the resource such as WiMAX (TDD or FDD) link and CDMA link, and
an 11-downlink shares the resource of interface among other downlinks within the same
interface. Conceptually, we can also view an 11-downlink within an interface as follows.
• Each 11-downlink has a queue which is independent on the other downlinks.
• Each 11-downlink has a common server as an interface among the other downlinks.
That is, in 11-downlink k[i ] to terminal i, which is provide by interface k of base station,
F( bs,k[i]) which is packet arrival rate of link k[i ] in base station, is independent on the others,
276
10 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

and T( bs,k[i]) which is average delay of link k[i ] in base station, is common among the others.
Therefore, d( bs,k[i]) which is cost of 11-link k[i ] in base station, is shown as follows (see Fig. ??).

d( bs,k[i]) = F( bs,k[i]) · T ( bs,k[i]) (27)

Base Station

interface K

F(bs,k[a ]) link k[a]

d(bs,k[a ]) Terminal a

Queue

F(bs,k[i]) T(bs,k )
d(bs,k[i]) link k[i]

Terminal i

Queue

F(bs,k[z]) link k[z]

d(bs,k[z ]) Terminal z

Queue

(a) Downlink cost associated by 11-interface.

mbs[i]

d(bs, j[i]) d(bs,k[i]) T (bs,J [i]) T (bs,K [i])

0 F(bs,K [i]) Fbs[i] 0 F(bs,K [i]) Fbs[i]

(b) Dependence of m bs[i ]. (c) Dependence of T (bs,k [i ]) &
T (bs,j [i ]).

Fig. 6. Downlink and M-route for download traffic.

where T ( bs,k[i]) is the average delay of 11-interface k providing downlink k[i ] in base station.
That is, T ( bs,k[i]) is the average delay based on all the packets which are distributed to
11-interface k.
To argue the dependence of d( bs,k[i]) on F( bs,k[i]), the first derivative of d( bs,k[i]) at F( bs,k[i]) is
shown. Using Eq.(27), it is shown as follows.

dd( bs,k[i]) dT ( bs,k[i])

= T ( bs,k[i]) + F( bs,k[i]) (28)
dF( bs,k[i]) dF( bs,k[i])
dT ( bs,k [ i])
It is difficult to derive dF( bs,k [ i]) ,
which is the dependence of of T ( bs,k[i]) on F( bs,k[i]), because
T ( bs,k[i]) is dependent on not only F( bs,k[i]) but also the packet distribution of the other
downlinks provided by 11-interface k. To simplify this difficulty, the following condition is
assumed.
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 277
11

dT ( bs,k[i])
>0 (29)
dF( bs,k[i])
dd ( bs,k [ i])
According to the condition Eq.(29), its is dF( bs,k [ i]) > 0, then d( bs,k[i]) is a monotonically
increasing function of F( bs,k[i]).
In the condition, the packet distribution from a 11-downlink k[i ] to a 11-downlink j[i ], is
argued. These 11-downlinks are contained in the M-route which aggregates 11-downlinks
to terminal i, and are respectively provided by different 11-interface (11-interface k and j).
The same as the packet distribution of M-route for upload traffic, the packet distribution to
the other 11-downlinks to terminal i, which is respectively provided by different 11-interface
except for 11-interface k and j, is constant, thus the dependence of F( bs,j[i]) on F( bs,k[i]) is shown
as follows.

dF( bs,j[i])
= −1
dF( bs,k[i])
d2 F( bs,j[i])
=0 (30)
d( F( bs,k[i]))2

Therefore, the first derivative of d( bs,j[i]) at F( bs,k[i]) in the condition Eq.(29) is shown as follows.

dd( bs,j[i]) dd( bs,j[i])

=− <0 (31)
dF( bs,k[i]) dF( bs,j[i])
Consequently, in the condition Eq.(29), d( bs,j[i]) is a monotonically decreasing function of
F( bs,k[i]). Because Eq.(21) can be applied to M-route for download traffic, mbs [i] which is
the cost of M-route to terminal i is the sum of d( bs,k[i]), which is a monotonically increasing
function of F( bs,k[i]), and d( bs,j[i]), which is a monotonically decreasing function of F( bs,k[i]),
and the 11-downlink cost of the others, which is constant for F( bs,k[i]). Therefore, mbs [i] is
a multioptimization function of F( bs,k[i]), and it has some local minimums for F( bs,k[i]) (see
Fig. 6(b)).
Here, argue the dependence of mbs [i] on F( bs,k[i]). it is shown as follows using Eq.(30) and (31).

dmbs [i] dd( bs,k[i]) dd( bs,j[i])

= − (32)
dF( bs,k[i]) dF( bs,k[i]) dF( bs,j[i])
Furthermore, Eq.(32) is transformed into difference equation, and is applied finite difference
approximation based on Eq.(29), then the condition that the M-route cost for download traffic
decreases is shown as follows.

dmbs [i] (n )
≈ T ( bs,k[i])(n + 1) − T ( bs,j[i])(n + 1) > 0 (33)
dF( bs,k[i])
Where, mbs [i](n ) is the cost of M-route to terminal i from base station in packet distribution of
n time and T ( bs,y[ x ])(n + 1) is average delay of interface y in packet distribution of n + 1 time,
and the interface y provides 11-downlink to terminal x.
278
12 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

Furthermore, based on Eqs. (29) and (30), the dependence of T( bs,j[i]) on F( bs,k[i]) is shown as
follows.

dT ( bs,j[i])
<0 (34)
dF( bs,k[i])
That is, T ( bs,j[i]) is a monotonically decreasing function of F( bs,k[i]). Therefore, a cost of each
link in M-route should be considered a monotonically increasing function of the packet arrival
rate, and the cost of M-route is the sum of each link cost, is a multioptimization function of
F( bs,k[i]) (see Fig.6(b)). That is, mbs [i] has some local minimums for F( bs,k[i]) and the packet
distribution meeting Eq.(33) may not bring mbs [i] to the optimal solution.
On the other hand, T ( bs,k[i]) and T ( bs,j[i]) is respectively a monotonically increasing/decreasing
function for F( bs,k[i]), and then, in 0 ≤ F( bs,k[i]) ≤ Fbs [i], the number of solutions which
makes T ( bs,k[i]) equal to T ( bs,j[i]) is 1 in the maximum (see Fig. 6(c)). Consequently, the packet
distribution which meets Eqs.(29) and (33) is repeated, and finally it reaches T ( bs,k[i])(n + 1) −
T ( bs,j[i])(n + 1) = 0, then the M-route cost mbs [i] reaches its optimal solution. Furthermore, the
search for the optimal solution of M-route cost mbs [i] has the additional effectiveness which
decreases the arrival of out-of-order packets because of the equalization the delay of source
11-link and destination 11-link.

3. Characteristics of IEEE802.16 link for packet distribution


&(



)(

$
#

#


 

 

'
 

'
 

!
$ 

& 

& 

 

 

 

 



#



'


'





 !


Fig. 7. IEEE802.16 MAC frame.

The performance of IEEE802.16 is actively analyzed. (Nakaya & Hossain, 2006) investigates
the delay analysis based on queueing theory, but it does not consider MAC of IEEE802.16.
(Cho et al., 2005; Lin et al., 2007; Iyengar et al., 2005; He et al., 2007; Ni et al., 2007) investigate
the performance analysis based on MAC of IEEE802.16. Cho et al. (2005) analyzes the
utilization and throughput and (Lin et al., 2007) analyzes the utilization for BW request based
on polling. These analyses do not investigate the delay. On the other hand, (Iyengar et al.,
2005; He et al., 2007; Ni et al., 2007) analyze the delay, but does not consider waiting time in
queue. In this section, in regard with IEEE802.16 link (16-link), considering the waiting time
in queue and MAC of IEEE802.16, the dependence of average delay on traffic is analyzed in
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 279
13

accordance with its four QoS classes. Furthermore, based on the analyzed dependence, the
characteristics of 16-link for packet distribution is shown.
Figure. 7 shows 16-frame in TDD. The frame consists of DL-subframe and UL-subframe.
Each subframe consists of time slots. Base station (BS) sends DL-MAP and UL-MAP in
DL-subframe, and all terminals listen to the DL-subframe, and know that they should listen
to slots in DL-subframe, and know that they should use slots in UL-frame to transmit data.
In such communications between BS and terminals, IEEE802.16(IEEE std. 802.16-2004, 2004;
IEEE std. 802.16e-2005, 2005) supports four class for QoS, which are UGS, rtPS, nrtPS, BE.
In UGS class, BS assigns fixed-size periodic data grants to both of uplink and downlink in
terminals. In rtPS class and nrtPS class, BS assigns data grants to downlink, and polls to
terminals in accordance with the reserved capacity for uplink in each terminal, and in nrtPS
class, terminals are additionally allowed to use contention requests for uplink bandwidth
(BW). In BE class, terminals are allowed to use contention requests only for both of uplink
and downlink, and BS does not poll to terminals.
On the analysis, the assumptions are as follows.
• 16-frame length is constant.
• The multiplexing is TDD.
• The DL-subframe and DL-subframe length in frame is the ratio of 1:1.
• The modulation for each link is unchanged after the communication is arranged
• A time is normalized by slot.

3.1 16-link in UGS

In UGS class, BS assigns fixed-size periodic data grants to both of uplink and downlink
in terminals. The fixed-sized periodic data grants is slots of which map is in
DL-MAP or UL-MAP. The data arrival process at slot can be approximated to poisson
process(Bertsetkas & Gallager, 1992) (Note. data arrival at link means transmission data
occurrence in link). Based on the above, argue the average time that a packet waits in queue
of downlink, which is Wdl.UGS. Wdl.UGS consists of the follows.
• The average residual time Rdl.USG. When a new packet arrives at 16-downlink, a 16-frame
is already being processed. Rdl.USG is a remaining average time until the current 16-frame
is processed completely.
• The queued packet average processing time for UGS of downlink, Qdl.UGS. Qdl.UGS is a
average time to process the all queued packets in UGS of downlink on a packet arrival.
• The average advance time Adl.USG. In 16-frame, Adl.USG is a average time to process the
other packets before a packet in USG of downlink is processed.
Rdl.USG consists of Rds.UGS, which is the average residual time for the packet in USG of
downlink, and Rother , which is the average residual time for the packet in frame except for
UGS of downlink. Let Cdl.UGS be the reserved slots in frame for UGS of downlink, Rds.USG is
Cdl.UGS/2. Let V dl.UGS and V 2 dl.UGS be respectively the first and second moment of process
time for a packet in frame except for UGS of downlink, Rother is V 2 dl.UGS/2V dl.UGS. Let L F be
the number of slots in 16-frame, then Rdl.UGS is derived as follows.

Cdl.USG C
Rdl.USG = Rds.UGS + (1 − dl.USG ) Rother
LF LF
2 (35)
Cdl.UGS V 2 dl.UGS
= + ( L F − Cdl.UGS)
2L F 2V dl.UGS L F
280
14 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

Argue Qdl.UGS. Based on Little’s theorem(Gross & Harris, 1985), the number of queued
packets in UGS of downlink, which is Ndl.UGS, is derived as follows.

Ndl.UGS = Fdl.UGS · Wdl.UGS (36)

Where Fdl.UGS is a packet arrival rate at UGS of downlink, which is average number of arrival
packets within a slot in UGS of downlink, Wdl.UGS is the average time that a packet waits in
queue in UGS of downlink. Let m be a data grants period which is expressed by the number
of frames, and Qdl.UGS is derived as follows.

Qdl.UGS = Fdl.UGS · Wdl.UGS · m · L F (37)

Adl.USG is equal to the residual time of DL-subframe, and is derived as follows.
2
Cdl.UGS V 2 ds.UGS
Adl.UGS = + ( L dl − Cdl.UGS) (38)
2L dl 2V ds.UGS L dl
L dl is the number of slots in DL-subframe, V ds.UGS and V 2 ds.UGS are respectively the first
and second moment of process time of a packet in DL-subframe except for UGS. Accordingly,
Wdl.UGS is expressed as follows.

Wdl.UGS = Rdl.USG + Fdl.UGS · Wdl.UGS · m · L F + Adl.UGS

R + Adl.UGS (39)
Wdl.UGS = dl.UGS
1 − mFdl.UGS L F
Based on Eq. (39), the average delay in UGS of downlink, which is Tdl.UGS, is derived as
follows.

Tdl.UGS = Wdl.UGS + Cdl.UGS (40)

V2
Assuming the modulation for each link to be unchanged, Cdl.UGS, V dl.UGS, dl.UGS, V ds.UGS,
and V 2 ds.UGS are constant even if Fdl.UGS changes, and they are independent on Fdl.UGS. That
is, Rdl.UGS and Adl.UGS are independent on Fdl.UGS. Therefore, using Eq. (40), the first and
second derivative of Tdl.UGS at Fdl.UGS are derived respectively as follows.

dTdl.UGS d2 Tdl.UGS
>0 2
>0 (41)
dFdl.UGS dFdl.UGS
Consequently, Tdl.UGS is a convex monotonically increasing function of Fdl.UGS.
Argue Wul.UGS, which is the average time that a packet waits in queue of uplink. Similar to
Wdl.UGS, Wul.UGS consists of Rul.UGS, which is the average residual time for frame on a packet
arrival at USG of uplink, Qul.UGS, which is the queued packet processing time for UGS of
uplink, and Aul.UGS which is the average advance time for UGS of uplink. Rul.UGS is common
to Rdl.UGS, and Qul.UGS is Ful.UGSWul.UGSmL F based on Little’s theorem. Aul.UGS is the sum
of L dl and the residual time for UL-subframe because UL-subframe is arranged to following
DL-subframe. Let Tul.UGS and Cul.UGS be respectively the average delay in USG of uplink and
the number of reserved slots for UGS of uplink, Wul.UGS and Tdl.UGS are respectively dervied
as follows

Rul.UGS + Aul.UGS
Wul.UGS =
1 − mFul.UGS L F (42)
Tul.UGS = Wul.UGS + Cul.UGS
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 281
15

Similar to downlink, Rul.UGS and Aul.UGS are independent on Ful.UGS. Accordingly, Tul.UGS is
a convex monotonically increasing function of Ful.UGS.

3.2 16-downlink in rtPS and nrtPS

In rtPS, BS periodically assigns data grants to downlink of terminals based on the reserved
capacity for the link. Similar to UGS, delay of 16-downlink in rtPS, which is Tdl.rtPS, is derived
as follows.

X2 dl.rtPS V 2 dl.rtPS
Rdl.rtPS = + ( L F − X dl.rtPS)
2L F 2V dl.rtPS L F
X2 dl.rtPS V 2 ds.rtPS
Adl.rtPS = + ( L dl − X dl.rtPS)
2L dl 2V ds.rtPS L dl
Rdl.rtPS + Adl.rtPS (43)
Wdl.rtPS =
1 − mFdl.rtPS L F
Tdl.rtPS = Wdl.rtPS + X dl.rtPS
X dl.rtPS + V dl.rtPS = L F
X dl.rtPS + V ds.rtPS = L dl
X dl.rtPS and X2 dl.rtPS are respectively the first and second moment of the number of granted
slots, which is a process time of a packet, for rtPS of downlink, V dl.rtPS and V 2 dl.rtPS be
respectively the first and second moment of process time of a packet in frame except for rtPS of
downlink, V ds.rtPS and V 2 ds.rtPS be respectively the first and second moment of process time
of a packet in DL-subframe except for rtPS, Fdl.rtPS is a rtPS packet arrival rate at 16-downlink,
and Wdl.rtPS is the average time that a packet waits in queue in rtPS of downlink,
Argue the dependence of X rtPS and X2 dl.rtPS on Fdl.rtPS. Assuming the modulation for each
link to be unchanged, X rtPS increases in the linear for the increase in Fdl.rtPS. Therefore, the
dependence of X rtPS and X2 dl.rtPS on Fdl.rtPS are respectively expressed as follows.

dX dl.rtPS d2 X dl.rtPS
>0 2
=0
dFdl.rtPS dFdl.rtPS
(44)
dX2 dl.rtPS d2 X2 dl.rtPS
>0 2
=0
dFdl.rtPS dFdl.rtPS
Based on Eq. (43) and (44), the dependence of Tdl.rtPS on Fdl.rtPS is derived as follows.

dTdl.rtPS d2 Tdl.rtPS
>0 2
>0 (45)
dFdl.rtPS dFdl.rtPS
The difference of nrtPS form rtPS is the length of data grant periods, and the data grants
period in nrtPS is longer than that in rtPS. Then the depenadence of delay Tdl.nrPS on Fdl.nrtPS,
which is nrtPS packet arrival rate at uplink, is the same as that in rtPS. Consequently, Tdl.rtPS
and Tdl.nrtPS are a convex monotonically increasing function of the each packet arrival rate.

3.3 16-uplink in rtPS

In rtPS, BS periodically polls to terminals in accordance with the reserved capacity for uplink,
and terminals reply by sending BW requests with allocated space (i.e., contention free). In
next frame, BS assigns data grants which is mapped by UL-MAP to terminals, and terminals
use data grant to transmit data. The difference of rtPS of uplink from that of downlink is that
282
16 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

two frames is necessary to transmit a packet. Let Rul.rtPS, Aul.rtPS, m, and Ful.rtPS, X ul.rtPS be
respectively the average residual time for rtPS packet of uplink, the average advance time for
rtPS packet of uplink, the polling period in rtPS, the packet arrival rate at rtPS of uplink, and
the average process time for packet in rtPS of uplink, Wul.rtPS, which is the queued packet
processing time for UGS of uplink, and Tul.rtPS, which is the average delay in rtPS of uplink,
are respectively expressed as follows.

Rul.rtPS + Aul.rtPS
Wul.rtPS =
1 − 2mFul.rtPS L F (46)
Tul.rtPS = Wdl.rtPS + X ul.rtPS
Rul.rtPS is common to Rdl.rtPS, and Aul.rtPS is the sum of Adl.rtPS and L F because the rtPS of
uplink is necessary to additional a frame to poll to terminal and to request BW to BS with
contention free. Therefore, Rdl.rtPS and Adl.rtPS are independence on Ful.rtPS, and then Tul.rtPS
is a convex monotonically increasing function of Ful.rtPS the same as rtPS of downlink.

3.4 16-uplink in nrtPS and 16-link in BE

In 16-uplink of nrtPS and 16-link of BE, also the arrival packets are enqueued and wait to
be processed with FCFS. Let the waiting time be Wbw (argue later in detail). The packet is
dequeued with FCFS, and then, is processed. The packet processing in nrtPS is based on
the polling from BS the same as uplink of rtPS. Furthermore, uplink of nrtPS is additionally
allowed to use contention BW request. In BE, the link is allowed to use contention BW request
only. In such contention mode, terminals send BW request during the contention period
in UL-subframe. Depending on the number of contention BW request, the collision of BW
request occurs. In contention BW request, each terminal resolves and avoids the collision as
follows.
• Each terminal waits the random number of slots before sending BW request in the
contention period. The number of waiting slots, which is back-off counter, is generated
based on exponential binary backoff mechanism.
• The backoff counter is decreased during the contention period.
• When the counter is zero, terminal sends BW request in the contention period.
• The terminal sending BW request waits data grants in DL/UL-map from BS.
• When the terminal does not receive data grants from BS in duration of the timer,
terminal increases the contention window size, and generates the backoff counter based
on exponential binary backoff mechanism, and then waits the opportunity sending BW
request when the counter is zero. That is the retransmission process.
The contention BW request is analyzed based on the following model.
• The packet processing time consists of BW request opportunity waiting period, BW request
attempt period, and packet transmission period.
• A BW request opportunity waiting period is the number of slots to be spent until the
back-off counter becomes zero.
• A BW request attempt period is the number of slots to be spent by BW request
transmission. In BW request attempt period, BW request transmission succeeds or collides.
The collision causes the timeout in receipt of data grant, and spends the number of slots
corresponding to the timeout. The success spends the number of slots to be spent from BW
request accepted by BS to complete transmission of a packet in terminal.
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 283
17

• In each terminal, Let τbw be the BW request attempt rate (req/slot) in the contention period
of UL-subframe, and then the probability q bw that BW request is transmitted successfully
is (1 − τbw )n−1 , where n is the number of terminals transmitting BW request .
• In each terminal, the packet arrival process (i.e., upload traffic) and packet request process
(i.e., download traffic) is poisson process(Bertsetkas & Gallager, 1992). Let Fbw be an packet
arrival/request rate (packets/slot), which need the contention BW request.
• The contention period ratio, which is the ratio of the number of slots in the contention
period in a frame, is constant. Let Uc be the contention period ratio.
• The process of the BW request that BS receives is assumed to FCFS, and the allocating data
grants rate (slot/packet) in DL-subframe or UL-subframe for BW request in BS is Sdg , and
is constant.
The contention BW request process is the same as the model described in 2.1.1 except for tb
in Eq.(4), ts and tc in Eq.(6). tb is 1 because the contention BW request process decrements
the backoff counter without carrier sensing. tc is the number of slots to be spent by timeout
of data grant receipt from BS, and is a constant. ts is the number of slots to be spent from
the success transmitting of BW request to the complete transmission of packet, and then it
depends on Fbw . ts is divided into tss , which is the air time of BW request from terminal to BS,
and tbs , which is the time from the receipt of BW request in BS to the complete transmission
of packet in terminal, and tss is a constant.
Here, argue the dependence of tbs on Fbw . In Sbw which is the average time from first
transmission attempt of contention BW request to successful transmission of that, the average
number of arrival/request packets for contention BW request is Fbw Sbw , and, in Sbw , the
average number of BW request transmission attempts is (1 − q bw )/q bw + 1. Therefore, τbw
is expressed as follows.

Fbw Sbw F
τbw = = bw (47)
Uc Sbw q bw Uc q bw
And, based on Eqs.(3), (4) and (5), Sbw is shown as follows.

Sbw = Bbw + Abw

ηCWmin −1
Bbw = 2q (48)
1− q
Abw = q tc + tss
Furthermore, let Fbw_bs be the arrival rate of BW request at BS, Fbw_bs is shown as follows.

Fbw_bs = q bw nFbw (49)

Based on Eqs.(47), (48) and (49), on condition of Fbw < 1/Sbw , the dependence of Fbw , Fbw_bs
and Sbw on τbw is respectively shown as follows.

dFbw dFbw_bs dSbw

>0 >0 >0 (50)
dτbw dτbw dτbw
Figure 8(a) and 8(b) respectively illustrates the dependence of Fbw and Fbw_bs on τbw by using
Eqs. (47), (49), and each also shows the same characteristics. Therefore, on condition of Fbw <
1/Sbw , the dependence of Fbw_bs and Sbw on Fbw is respectively shown, by using Eq.(50), as
follows.
284
18 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

dFbw_bs dSbw_bs
>0 >0 (51)
dFbw dFbw
-3 -3
2x10 5x10
Fbw in n=10
Fbw_bs in n=10

1/Sbw in n=10
1/Sbw in n=10

BW Request Arrival Rate at BS

Fbw in n=20
-3 Fbw_bs in n=20

4x10
Packet Arrival Rate at SS

-3 1/Sbw in n=20
1/Sbw in n=20

1.5x10
Fbw in n=30
Fbw_bs in n=30

(packet/slot)

1/Sbw in n=30
-3 1/Sbw in n=30

(req/slot)
3x10
Fbw in n=40
Fbw_bs in n=40

-3
1x10 1/Sbw in n=40
1/Sbw in n=40

-3
2x10

-4
5x10
-3
1x10

0 0
0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2
BW Request Attempt Rate (req/slot) BW Request Attemp Rate (req/slot)

(a) Dependence of Fbw on τbw . (b) Dependence of Fbw_bs on τbw .

Fig. 8. Dependence of each element on τbw .

Argue Wbw_bs which is the waiting time in queue of BS for data grant. The process of the
received BW requests in BS is assumed to be FCFS, and conceptually it can be view as
queueing system of which the packet arrival rate is Fbs_bs and the packet service rate is Sdg .
Therefore, Wbw_bs is shown, based on Eq.(14), as follows

Fbw_bs S2dg
Wbw_bs = (52)
2(1 − Fbw_bs Sdg )
Sdg is constant for Fbw_bs , and then, on condition of Fbw_bs < 1/Sdg , the dependence of Wbw_bs
on Fbw_bs is shown as follows.

dWbw_bs
>0 (53)
dFbw_bs
tbs is the sum of Wbw_bs and Sdg , and then, based on Eqs.(51) and (56), the dependence of tbs
on Fbw on condition of Fbw < 1/Sbw and Fbw_bs < 1/Sdg , that is, within the link capacity, is
shown as follows.

dtbs
>0 (54)
dFbw
Argue Wbw which is the packet waiting time in queue of terminal. According to the
exponential binary backoff model described in 2.1.1, and applying tb = 1 and the constance
of tc for τbw to Eq.(15), Wbw is derived as follows.

Fbw S2bw
Wbw = (55)
2(1 − Fbw Sbw )
According to Eqs.(16), (50), on condition of Fbw < 1/Sbw , that is, within link capacity, the
dependence of Wbw on Fbw is derived as follows.

dWbw
>0 (56)
dFbw
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 285
19

Finally, Tbw , which is the average delay for contention BW request, is the sum of Wbw , Sbw , tss
and tbs , the dependence of Tbw on Fbw is derived, based on Eqs.(51), (54) and (56), as follows.

dTbw d
= (Wbw + Sbw + tss + tbs ) > 0 (57)
dFbw dFbw
Therefore, Tbw is monotonically increasing function of Fbw.

3.5 Packet distribution for 16-link

The average delay on 16-link, except uplink in nrtPS and link in BE, is a convex monotonically
increasing function of packet arrival rate, therefore, its characteristics on packet distribution
corresponds to that of 11-downlink. On the other hand, 16-uplink in nrtPS and 16-link in BE
are a monotonically increasing function of packet arrival rate, therefore, their characteristics
on packet distribution corresponds to that of 11-uplink.

4. IP packet distribution for M-route compositing IEEE802.11/16 links

Based on the analyzed characteristics of 11/16-link for packet distribution, the characteristics
of the access route compositing multiple 11-links or 16-links is the same. Therefore,
the characteristics of M-route compositing 11links and 16-links for the packet distribution
corresponds to that of the access route compositing multiple 11-links or 16-links.
According to the above, IP packet distribution method for the M-route compositing 11-links
and 16-links be described.
The characteristics of M-route compositing 11/16-link for the packet distribution corresponds
to that of the access route compositing multiple 11-links or 16-links because that of 11-link and
16-link are the same.

4.1 Restriction condition

According to Eqs.( 26) and ( 33), the optimal solution of the M-route cost can be searched by
the repeating packet distribution that the average delay of distribution source link becomes
larger than that of distribution destination link, and that the average delay of both source link
and destination link become equal finally. Additionally, the packet distribution for download
traffic needs to meet the condition Eq.(29) when the source link on the packet distribution is
an 11-link.
Here, argue the condition Eq.(29). Transforming Eq.(29) into finite difference approximation,
it is shown as follows.

dT ( bs,k[i]) ΔT ( bs,k[i])
≈ >0 (58)
dF( bs,k[i]) ΔF( bs,k[i])
Because 11-link k[i ] is a source link on packet distribution, ΔF( bs,k[i]) < 0. Therefore, to meet
Eq.(58), ΔT ( bs,k[i]) < 0. In other words, it is that the average delay of source 11-interface on
packet distribution decreases. The increase in average delay of source 11-interface k does not
meet the condition and it occurs in the following unsteady state.
• The packet arrival rate at other links provided by 11-interface k increases.
• The number of links provided by 11-interface k increases.
The first item in the above list means the increase in contention with other terminals, thus
it also causes the increase in average delay of source link when source link is 11-uplink or
16-uplink in nrtPS or 16-link in BE. The second item means the increase in a number of
286
20 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

terminals, thus it causes the increase in average delay of source link because of the same reason
as the first item. Then it also occurs when source link on packet distribution is 11-uplink or
16-uplink in nrtPS or 16-link in BE. In above cases, M-route cost also loses the monotonically
increasing characteristics for packet arrival rate. Therefore, in consideration of the unsteady
state that traffic fluctuates, the restriction condition which is the decrease in the average delay
of source link on packet distribution is a necessary condition to bring the M-route cost to the
optimal solution.

4.2 Search for optimal solution of M-route cost with packet distribution
Argue the search for optimal solution of M-route cost with Packet Distribution in unsteady
state by the following packet distribution.
• M( x,y) is a M-route from x to y. On x and y, one is a base station and the other is a terminal.
• Packets transmitted to y at x are distributed.
• K denotes a source interface on the packet distribution. J denotes a destination interface
on the packet distribution.
• K is either an 11-interface or 16-interface and J is also either an 11-interface or 16-interface.
• ( x, Z [ y]) denotes a certain link to y in x, which link is provided by a certain interface Z.
• F( x,Z [y]) denotes a packet arrival rate at ( x, Z [ y])
• T ∗ ( p( x,Z [y])) denotes interface average delay T ( p( x,Z [y])) if Z is 11-interface, and denotes
link average delay T ( p( x,Z [y])) if Z is 16-interface.
Based on subsection 4.1, the search for optimal solution of M-route cost in unsteady state is
the search for the packet distribution meeting the following conditions.

T(∗x,K [y])(n ) − T(∗x,J [y])(n ) > 0 ΔT(∗x,K [y])(n ) < 0 (59)

where ΔT(∗x,K [y])(n ) denotes the difference between T(∗x,K [y])(n ) and T(∗x,K [y])(n − 1).
According
to Eq. (59), the proposed packet distribution method implements the search for the optimal
solution in IP layer using the measured average delay in MAC layer as the following iteration.
Step1: In the initial period, packets are distributed equally to each link in M-route with a
round robin manner.
Step2: At end of the initial period, T(∗x,Z [y])(0) of each link in M-route is derived,
and ( x, Max [ y])(0) which has maximum average delay in the initial (0-th) period, and
( x, Min [ y])(0) which has minimum average delay in the initial (0-th) period, is respectively
selected in M( x,y). On the packet distribution, ( x, Max [ y])(0) and ( x, Min[ y])(0) is
respectively assigned to the source link ( x, K [ y])(1) in the next (1-th) period and the
destination link ( x, J [ y])(1) in that period. ΔF( x,K [y])(1), which is the amount of packet
distribution from ( x, K [ y])(1) to ( x, J [ y])(1) in the next (1-th) period, is derived as follows.
where r( x,y) denotes the packet distribution rate of M( x,y), and r0 denotes the initial packet
distribution rate.

ΔF( x,K [y])(1) = r( x,y) (1) · F( x,K [y])(0)

(60)
r( x,y) (1) = r0
Step3: According to ΔF( x,K [y])(1), the packet distribution in the 1-th period is carried out.
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 287
21

Step4: At end of n-th period (n ≥ 1), T(∗x,Z [y])(n ) of each link in M( x,y) is derived. The delay
of each packet is a period when the packet arrives at IP layer, and is enqueued in queue of an
interface, and is dequeued by an interface, and is sent and interface receives its ACK based
on the media access control. Therefore, it can be measured within packet distributing side x.
Based on the relation of T(∗x,K [y])(n ) and T(∗x,J [y])(n ), ΔF( x,K [y])(n + 1) is derived as follows.

• In T(∗x,K [y])(n ) > T(∗x,J [y])(n ) and in ΔT(∗x,K [y])(n ) < 0, Eq. (59) is met. Therefore,
ΔF( x,K [y])(n + 1) is allocated the same as ΔF( x,K [y])(n ), and it is shown as follows.

ΔF( x,K [y])(n + 1) = r( x,y) (n + 1) · ΔF( x,K [y])(n )

(61)
r( x,y) (n + 1) = r( x,y) (n )

• In T(∗x,K [y])(n ) < T(∗x,J [y])(n ) and in ΔT(∗x,K [y])(n ) < 0, M( x,y) cost goes beyond the optimal
solution and ascents the gradient. Because it is caused by the excessive packet distribution
from source link to destination link, ΔF( x,K [y])(n + 1) is allocated smaller than ΔF( x,K [y])(n )
as follows. where α is the decrement rate (0 < α < 1).

ΔF( x,K [y])(n + 1) = r( x,y) (n + 1) · ΔF( x,K [y])(n )

(62)
r( x,y) (n + 1) = α · r( x,y) (n )

• In ΔT(∗x,K [y])(n ) > 0, the traffic among the source link increases as shown in subsection
4.1. Because ΔF( x,K [y])(n ) is underestimated, and because the monotonically increasing
characteristics of the source link cost for the packet distribution is regained, ΔF( x,K [y])(n +
1) is allocated larger than ΔF( x,K [y])(n ) as follows. where β is the increment rate (β > 1).

ΔF( x,K [y])(n + 1) = r( x,y) (n + 1) · ΔF( x,K [y])(n )

(63)
r( x,y) (n + 1) = β · r( x,y) (n )
Step5: ( x, Max [ y])(n ) and ( x, Min[ y])(n ) are respectively selected in M( x,y), and are
respectively assigned to ( x, K [ y])(n + 1) and ( x, J [ y])(n + 1). According to ( x, K [ y])(n + 1),
( x, J [ y])(n + 1), and ΔF( x,K [y])(n + 1), the (n + 1)-th packet distribution is carried out, then
return to Step4.
In each M-route of both a base station and terminals, the above iteration gradually updates
the amount of packet distribution, and brings M-route cost to the optimal solution, reducing
the out-of-order packets occurred by distributing packets to multiple links.

5. Performance evaluation
In this section, the simulation evaluation of the packet distribution method for M-route
compositing 11/16-links is shown.

5.1 Simulation scenario

For the simulation evaluation, OPNET 12.0A PL3 was used, and the network configuration
was as follows (see Fig. 9):
• Base station is equipped with an 16-interface and 4×11a/b-interfaces. 16-interface and
11a/b-interface respectively connects to 16-antenna and 11/ab-antenna.
• The number of terminals is 100, and each terminal is equipped with 16-interface and
11a/b-interface.
288
22 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

11b-Coverage 11a-Coverage

11a-Coverage
Antenna-B
11b-Coverage

Antenna-A

Server

560m

Base station
11b-Coverage
11b-Coverage
Antenna-B
Antenna-B
11a-Coverage

11a-Coverage

560m

Fig. 9. Example of access network topology.

• An antenna-A which equips with 16- and 11a/b-antena, three antenna-B which equip with
11a/b-antena, and 100 terminals without mobility are randomly deployed in 560m × 560m
space with a 1/10 scale of 16-coverage with 1000m radius.
• A FTP server and a Video Conference (VC) server, which are outside the wireless access
network, are connected to the base station by a wired network.
In the above access network, M-route between each terminal and a base station combines
available links as follows.
• The M-route between a base station and a terminal in 11a-coverage (area-A) combines
11a/b-link and 16-link.
• The M-route between a base station and a terminal in 11b-coverage and outside
11a-coverage (area-B)combines 11b-link and 16-link.
• The M-route between a base station and a terminal outside 11b-coverage (area-C) uses only
16-link.
The performance of 11a/b-wireless system and 16-wireless system shown in Table 1 is applied,
and each the capacity reservation of 16-link is shown in Table 2. Assuming the evaluation
environment to be a suburban area in line of sight, the 11a/b-radio propagation model is
a two-ray model and Ricean fading with Ricean factor 6dB(Takada, 2004), and the 16 radio
propagation model is a Erceg (TerrainA).
According to (3GPP2, 2006), the VC traffic on UDP is generated at each terminal as follows:
• The average video rate in the incoming and the outgoing is 32 Kbps.
• The distribution in video rate is a truncated pareto distribution with maximum 8Kbits
• The frame rate in the incoming and outgoing is 10fps. A frame corresponds to a data
packet in VC.
• As the sequence control of frame, VC waits for the frame with expected sequence number
for a period of 100 msec that is equal to frame interval. The frame that arrived on excess of
the period is destroyed.
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 289
23

• In 16-link, VC is mapped to rtPS for QoS class.

Furthermore, FTP traffic on TCP is also generated at each terminal as follows:
• In 10 sec period, FTP session which transfers a file of the size of 1K∼400Kbytes starts.
• 50% of the FTP sessions are download session.
• Each FTP session is established between each terminal and a FTP server.
• In 16-link, FTP is mapped to nrtPS for QoS class.
The evaluation items are as followings.
• IP average delay (sec/packet), is the average delay between terminal and servers in an IP
packet.
• IP throughput (bps), is the average arrival amount of IP packets at terminals and servers
during a unit time.
• FTP response time (sec/file), is the average delay to transfer a file in end-to-end between a
terminal and an FTP server.
• FTP throughput (bytes/sec), is the average amount of arrival data packets at terminals and
an FTP server during a unit time.
• VC average delay (sec/frame), is the average delay of end-to-end between terminal and a
VC server in a data frame.
• VC throughput (bytes/sec), is the average arrival amount of data frames at terminals and
a VC server during a unit time.
The end-to-end delay is composed of the delay in wireless access network and that in wired
communication between the base station and server. The delay in wired communication is
common without depending on any packet distribution in wireless access network because
the wired communication is out of scope of wireless access network. Therefore, the delay in
wired communication can be assumed to be constant to any packet distribution in wireless
access network, and the delay in wireless access network depends on packet distribution in
wireless access network. In viewpoint of packet distribution, the trend of the end-to-end
delay corresponds with that of the delay in wireless access network. Thus the delay in
wired communication can be logically ignored. Furthermore, assuming the access speed of
a future core network to be Gigabits order(Konishi et al., 2008), the delay in WiFi corresponds
to 102 ∼ 103 order of that in wired core network because the bandwidth of WiFi is Mbps.
Then, the delay between the base station and server is left out of consideration because it is
independent on the performance of the wireless access network. Furthermore, to demonstrate
the effectiveness of the proposed method, it is compared with the following methods.
• Single link (SL) uses a link. The terminals in area-A use 11a-link, the terminals in area-B
use 11b-link, and the other terminals use 16-link.
• Round robin (RR) uses available links and distributes packets equally to each link.
• Actual transmission rate (TR) uses available links and distributes packets to each link in
proportion to the measured transmission rate at each link in every 10 sec.
In the search for minimal solution, r0 is 0.1, α is 0.5, β is 1.5, and the update period of packet
distribution is 10 sec.
Furthermore, the link combination in IP is transparent to the upper layer. Therefore, the upper
layer is provided with the M-route as a single link view.
290
24 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

QoS Class rtPS nrtPS

Maximum Sustained Transmission Rate 384Kbps 384Kbps
Minimum Reserved Transmission Rate 80Kbps 1Kbps
Table 2. Capacity reservation for 16-link.
5
0.006 7.5x10
6
2x10
FTP file size 1 Kbytes session interval 10 sec FTP file size 1 Kbytes session interval 10 sec
IP Average Delay (sec/packet)

FTP file size 1 Kbytes session interval 10 sec

IP Out-of-Order Packets (bps)

VC video rate 32 Kbps frame rate 10 fps VC video rate 32 Kbps frame rate 10 fps
VC video rate 32 Kbps frame rate 10 fps
6
7x10

IP Throughput (bps)
5
0.0055 1.5x10

6
6.5x10
Proposal
0.005 RR 1x10
5

SL
6 TR Proposal
6x10
Proposal RR
RR TR
0.0045 SL 5x10
4

TR 5.5x10
6

6
0.004 5x10 0
300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 1000
Simulation Time (sec) Simulation Time (sec) Simulation Time (sec)

(a) Average delay. (b) Throughput. (c) Out-of-order.

Fig. 10. Transition of IP on FTP file size 1K bytes.

6
1.5x10
6
3x10
6 5x10

Proposal Proposal 6
RR 2.5x10
6
RR 4x10
11b Load (bps)
11a Load (bps)

16 Load (bps)
6 SL SL
1x10 TR TR

6 6
2x10 3x10

5
Proposal
5x10 RR
6 6 SL
1.5x10 2x10
TR
FTP file size 1 Kbytes session interval 10 sec
VC video rate 32 Kbps frame rate 10 fps
FTP file size 1 Kbytes session interval 10 sec FTP file size 50 Kbytes session interval 10 sec
VC video rate 32 Kbps frame rate 10 fps 6
VC video rate 32 Kbps frame rate 10 fps
6
0 1x10 1x10
300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 1000
Simulation Time (sec) Simulation Time (sec) Simultion Time (sec)

(a) 11a load. (b) 11b load. (c) 16 load.

Fig. 11. Distributed traffic load to each wireless system on FTP file size 1K bytes.
TCP Retransmissions (number/5sec)

8
FTP file size 1 Kbytes session interval 10 sec 0.22 4
FTP file size 1 Kbytes session interval 10 sec 1.55x10
VC video rate 32 Kbps frame rate 10 fps
FTP Response Time (sec/file)

VC video rate 32 Kbps frame rate 10 fps

FTP Throughput (bytes/sec)

7 0.2

0.18 1.5x10
4
6
Proposal
0.16 RR
SL
5 TR
4
Proposal 0.14 1.45x10
RR Proposal
SL RR FTP file size 1 Kbytes session interval 10 sec
4
TR SL VC video rate 32 Kbps frame rate 10 fps
0.12
TR

4
3 0.1 1.4x10
300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 1000
Simulation Time (sec) Simulation Time (sec) Simulation Time (sec)

(a) TCP retransmissions. (b) FTP response time. (c) FTP throughput.

Fig. 12. Transition of TCP and FTP on FTP file size 1K bytes.

5.2 Transition of delay and throughput in low traffic load

Figures 10(a) and 10(b) show, respectively, the transition of IP average delay and IP
throughput, when file size in FTP is 1K bytes. As the packet distribution proceeds, the IP
average delay of the proposal decreases rapidly, and becomes much lower than that of the
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 291
25

5
0.008 9x10
FTP file size 1 Kbytes session interval 10 sec

VC average Delay (sec/frame)

FTP file size 1 Kbytes session interval 10 sec

VC Throughput (bytes/sec)
VC video rate 32 Kbps frame rate 10 fps VC video rate 32 Kbps frame rate 10 fps

5
0.007 Proposal 8x10
RR
SL
TR
Proposal
5
0.006 7x10 RR
SL
TR

5
0.005 6x10

5
0.004 5x10
300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 1000
Simulation Time (sec) Simulation Time (sec)

(a) Average delay. (b) Throughput.

Fig. 13. Transition of VC on FTP file size 1K bytes.

others. Figures 11(a), 11(b) and 11(c) show, respectively, the transition of distributed load to
11a-wireless system (11a-load), that to 11b-wireless system (11b-load) and that to 16-wireless
system (16-load), when file size in FTP is 1K bytes. The decrease in IP average delay of the
proposal corresponds to the increase in 11a-load of the proposal (see Fig. 10(a) and Fig. 11(a)).
In area-A, 11a accommodates a few terminals because of its narrow coverage, and the proposal
distributes almost packets to 11a-link the same as SL, and saves the capacity of 11b and 16 for
many terminals outside area-A. RR and TR in the area distributes packets to other link as
well, thus RR and TR can not use 11a capacity effectively to save the capacity of 11b and 16.
Consequently, RR and TR bring the large load to 16 (see Fig. 11(c)), which of links have low
transmission rate (see Tab. 2), and it causes the inferior IP average delay of RR and TR to that
of the proposal. In area-B, SL distributes all packets to 11b-link (see Fig. 11(b)), and then the
packet collision in 11b occurs frequently. Thus, it causes the inferior IP average delay of SL to
that of the proposal. In comparison with SL, the packet distribution of the proposal and TR
improve IP performance, but that of RR lowers IP performance.
The IP out-of-order packets of the proposal decreases the same as the decrease in its IP average
delay, consequently, its out-of-order packets becomes much lower than that of RR and TR (see
Fig. 10(c)). Therefore, its packet distribution effects the decrease in IP average delay and the
decrease in out-of-order packets. Figures 12(a) shows the number of TCP retransmissions for
a period of 5 sec. The TCP retransmissions of the proposal is nearly equal to that of SL and
RR, and that of TR is larger than that of the others. The cause of TCP retransmission in SL
is packet loss. In area-B, SL distributes all packets to 11b, thus the packet collision occurs
frequently in 11b and then it causes the TCP retransmission. The cause of TCP retransmission
in the proposal, RR and TR is out-of-oder packets. The number of TCP transmissions in RR is
lower than that of TR. RR loads larger mount of packets with 16 than the others (see Fig. 11(c)).
Because the 16-link has the low transmission rate, the IP average delay of RR is inferior to that
of the others (Fig. 10(a)). Then TCP congestion window size of RR is smaller than that of TR
and the proposal, and the amount of distributed packets to multiple links for a period is fewer
than that of TR and the proposal, thus the probability of occurrence of out-of-order packets
is lower. Consequently, the TCP retransmissions of RR is lower than that of TR. That of the
proposal is also lower than that of TR, then the delay equalization between multiple links
in the proposal effects the decrease in the occurrence of out-of-order packets, and effects the
decrease in TCP retransmissions.
Figures 12(b) and 12(c) show, respectively, the transition of FTP response time and FTP
throughput. The FTP response time of SL and the proposal are superior to that of RR and
TR. The IP average delay of TR is superior to that of SL, however, the FTP response time of TR
is inferior to that of SL. The inversion is caused by the large number of TCP retransmissions in
292
26 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

TR, and the packet distribution of TR lowers the FTP performance. The cause of the inferior
FTP response time of RR to that of SL is not the TCP retransmissions, but is the small amount of
TCP flow based on TCP congestion window size, then the packet distribution in RR distributes
the large number of packets to 16-link, which is narrow bandwidth, and originally lowers IP
performance. The number of TCP retransmissions and the FTP response time of the proposal
is the same as those of SL. As the above mentioned, the cause of TCP retransmission in SL is
the packet loss in 11b-link, but the cause of that in the proposal is the out-of-order packet, that
is, the proposal offsets the improvement of IP performance against the out-of-order packets,
and does not improve the FTP performance, but does not lower it.
Figures 13(a) and 13(b) show, respectively, the transition of VC average delay and VC
throughput. The VC average delay of SL is equal to the IP average delay because a VC
frame corresponds to a IP packet and because out-of-order packet does not occur. In the
proposal, RR, and TR, the VC average delay is larger than that of IP because the sequence
control in VC waits for frame with the expected sequence on the occurrence of out-of-order
packet. Therefore, VC average delay of TR is higher than that of SL though IP average delay of
TR is lower than that of SL, i.e., the packet distribution of TR lowers the VC performance. On
the other hand, that of the proposal is lower than that of SL, therefore, the effect of the packet
distribution in the proposal overcomes the ill of it, and can improve the VC performance. That
of RR is higher than that of the others because RR originally lowers IP performance.

5.3 Transition of delay and throughput in high traffic load

2.5 6
9x10
FTP file size 350 Kbytes session interval 10 sec 4x10
7
IP Average Delay (sec/packet)

FTP file size 350 Kbytes session interval 10 sec

IP Out-of-Order Packets (bps)
2 VC video rate 32 Kbps frame rate 10 fps FTP file size 350 Kbytes session interval 10 sec
VC video rate 32 Kbps frame rate 10 fps VC video rate 32 Kbps frame rate 10 fps
6
1.5 8x10
IP Throughput (bps)

1 3.5x10
7
6
7x10
0.5

0.3 7
6
6x10 Proposal
3x10 RR
TR
6
0.2 5x10
7
Proposal 2.5x10 Proposal
0.1 RR RR 6
4x10
SL SL
TR TR
6
0 2x10
7 3x10
300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 1000
Simulation Time (sec) Simulation Time (sec) Simualtion Time (sec)

(a) Average delay. (b) Throughput. (c) Out-of-oder.

Fig. 14. Transition of IP on FTP file size 350K bytes.

7
3x10
8x10
6 FTP file size 350 Kbytes session interval 10 sec FTP file size 350 Kbytes session interval 10 sec FTP file size 350 Kbytes session interval 10 sec
6
VC video rate 32 Kbps frame rate 10 fps 8x10 VC video rate 32 Kbps frame rate 10 fps VC video rate 32 Kbps frame rate 10 fps
6
7x10
7
2.5x10
11b Load (bps)

Proposal
11a Load (bps)

6
6 7x10
16 Load (bps)

6x10 RR
SL
6 TR
5x10 6 7
6x10 2x10 Proposal
6 RR
4x10 SL
TR
6
3x10
6 5x10 7
1.5x10
Proposal
2x10
6 RR
4x10
6 SL
TR
6 7
1x10 1x10
300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 1000
Simulation Time (sec) Simulation Time (sec) Simulation Time (sec)

(a) Average delay. (b) Throughput. (c) Out-of-oder.

Fig. 15. Distributed traffic load to each wireless system on FTP file size 350K bytes.
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 293
27
TCP Retransmissions (number/5sec)

1400
120 6







 

 FTP file size 350 Kbytes session interval 10 sec 4x10
FTP file size 350 Kbytes session interval 10 sec

FTP Response Time (sec/file)

VC video rate 32 Kbps frame rate 10 fps
1200 

 


 
 



FTP Throughput (bytes/sec)

VC video rate 32 Kbps frame rate 10 fps
100
Proposal
1000 RR
SL
80 TR 6
3x10
800

60
600

400 40 2x10
6

Proposal Proposal
200 RR 20 RR
SL SL
TR TR
0 6
300 400 500 600 700 800 900 1000 0 1x10
300 400 500 600 700 800 900 1000 300 400 500 600 700 800 900 1000
Simulation Time (sec) Simulation Time (sec) Simulation Time (sec)

(a) TCP retransmissions. (b) FTP response time. (c) FTP throughput.

Fig. 16. Transition of TCP and FTP on FTP file size 350K bytes.
4
FTP file size 350 Kbytes session interval 10 sec 8x10
5
VC Average Delay (sec/frame)

3 VC video rate 32 Kbps frame rate 10 fps

VC Throughput (bytes/sec)
2
Proposal
1
RR
5
SL 7x10
0.03 TR

0.02 Proposal
5
RR
6x10 SL
TR
0.01

FTP file size 350 Kbytes session interval 10 sec

VC video rate 32 Kbps frame rate 10 fps
0 5
5x10
300 400 500 600 700 800 900 1000
300 400 500 600 700 800 900 1000
Simulation Time (sec) Simulation Time (sec)

(a) Average delay. (b) Throughput.

Fig. 17. Transition of VC on FTP file size 350K bytes.

Figures 14(a) and 14(b) show, respectively, the transition of IP average delay and IP
throughput, when file size in FTP is 350K bytes, furthermore, Fig. 15(a), 15(b) and 15(c) show,
respectively, the transition of 11a load, 11b load and 16 load, when file size in FTP is 350K
bytes. The IP average delay of the proposal is low, and is stable. On the other hand, that of
the others increase as linear, and become much higher than that of the proposal. Furthermore,
their IP throughput are lower than that of the proposal. In area-A, the packet distribute to
11a-link brings low delay to IP because of wide bandwidth and few accommodated terminals
in 11a, as mentioned in 5.2. In area-B, the packet collision and loss in 11b further increase
because of the increase in traffic, and the large number of retransmissions in MAC brings the
increase in delay to IP. Furthermore, the packet loss in 11b brings the decrease in throughput
to IP. Each 16-link has the narrow bandwidth, but does not cause the collision because of
TDD. i.e., The delay of 16-link is lower than that of 11b-link because of no retransmission
process in MAC, which of delay in 11b exponentially increases based on a binary back-off
mechanism. Therefore, the large number of packet distribute to 11b brings the increase in
delay and the decrease in throughput to IP. Consequently, IP average delay of the proposal,
which distributes the smaller number of packets to 11b than the others (see Fig. 15(b)), is
lowest, and its IP throughput is highest.
Figures 14(c) and 16(a) show, respectively, the transition of IP out-of-order packets and TCP
retransmissions, when file size in FTP is 350K bytes. The IP out-of-order packets of the
proposal decreases rapidly as the packet distribute proceeds the same as the case that FTP
file size is 1K bytes, i.e., the delay equalization between the multiple links in the proposal
effects the decrease in IP out-of-order packets. That of RR also decreases, but the decrease in
294
28 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

the amount of TCP flow based on TCP congestion window size, which becomes small rapidly
by the increase in IP delay of RR, brings it. TCP retransmission is caused by the IP packet
loss and IP out-of-order packets. The TCP retransmissions in SL is caused only by IP packet
loss, and IP packet loss is caused by the large number of distributed packets to 11b. That of
RR, TR and the proposal is caused by IP packet loss and IP out-of-order packets. That of RR
is caused largely by IP packet loss, because RR distributes the large number of packets to 11b
and IP out-of-order packets decreases by the decrease in TCP flow. Therefore, the trend of
TCP retransmissions of RR is similar to that of SL. TR also distributes the large number of
packets to 11b, but distributes the larger number of packets than RR to 11a and 16, which of
packet loss probability is much lower than 11b, i.e., the TCP retransmissions in TR is caused
mainly by out-of-order packets and it reduces the upward trend of TCP retransmissions in
comparison with SL and TR. On the other hand, the TCP retransmissions of the proposal is
low stable in comparison with the others. The proposal distributes the much smaller number
of IP packets than the others to 11b and reduces IP packet loss, furthermore, it equalizes the
delay of each link in M-route, thus reduces also IP out-of-order packets. That brings the low
and stable retransmissions to TCP.
Figures 16(b) and 16(c) show, respectively, the transition of FTP response time and FTP
throughput, when file size in FTP is 350K bytes. The FTP response time of RR and TR increase
as linear. In RR and TR, FTP session can not complete in a period of 10 sec, which is FTP
session start interval, because the amount of TCP flow is restrained low by the large number
of retransmissions. The active FTP session accumulates. Therefore, the access network causes
the congestion. In the proposal, FTP session can complete within 10 sec, and the delay not
increase and is stable. Furthermore, the throughput reaches the input load 4M bytes/sec.
Therefore, the proposal controls avoids the congestion.

5.4 Dependence of delay on throughput

0.05
IP Average Delay (sec/packet)

VC Average Delay (sec/frame)

Proposal
FTP Response Time (sec/file)

20 0.014 Proposal
RR
SL RR
0.04 TR SL
TR
0.012
15
0.03
0.01
10
0.02
0.008

5
0.01 Proposal 0.006
RR
SL
TR
0 0 0.004
7 7 7 7 6 6 6 6 6 6 6 6 6
1x10 2x10 3x10 4x10 0 1x10 2x10 3x10 4x10 1x10 2x10 3x10 4x10 5x10
IP Throughput (bps) FTP Throougput (bytes/sec) Sum Throughput of VC and FTP (bytes/sec)

(a) IP. (b) FTP. (c) VC.

Fig. 18. Dependence of delay on throughput.

Figure 18(a), 18(b), and 18(c) shows, respectively, the dependence of IP average delay on IP
throughput, the dependence of FTP response time on FTP throughput , and the dependence
of VC average delay on VC throughput when FTP file size increases from 1K bytes to 400K
bytes. The average delay and throughput are each the averages for 10 topologies in which the
antennas and terminals are deployed randomly in the evaluation space.
When the FTP traffic is low, the performance of SL and the proposal is superior to that of
RR and TR. In low load, if packets are distributed to a widest band link, that is, if the packet
distribution is equalized to that of SL, the performance becomes high. The packet distribution
of the proposal becomes equal to that of SL, but that of RR and TR do not. As FTP traffic
Traffic Control
Traffic Control forWireless
for Composite Composite Wireless
Access Route Access
of IEEE802.11/16 Links Route of IEEE802.11/16 Links 295
29

increases, the 11b-link load of M-route in 11b-coverage and outside 11a-coverage becomes
high, then M-route including 11b-link needs to distribute packets to 11a-link or 16-link. SL can
not distribute packets of 11b-link to other links, then SL is saturated first by the exhaustion
of 11b-link capacity. By the same cause, RR and TR are saturated in FTP file size 300K bytes
and 400K bytes respectively. The proposal distributes packets from 11b-link to 16-link and
11a-link, and avoids the saturation until FTP file size exceeds 400K bytes.
Summarizing, in any FTP traffic, the proposal can distribute packets effectively in comparison
with other methods, and it produces low delay and hight throughput on both TCP application
and UDP application, and simultaneously.

6. Conclusion
In this chapter, the packet distribution characteristics in IEEE802.11-link and that in
IEEE802.16-link was respectively shown, and, based on these characteristics, the packet
distribution method for access route compositing IEEE802.11/16-links was proposed.
Furthermore, its performance through evaluation with IEEE802.11a/b and IEEE802.16 was
shown. Consequently, the proposed method was found to have the following effectiveness.
• It can greatly effectively distribute packets to IEEE802.11/16 links according to link load.
• And, it can also reduce out-of-packets caused by distributing packets to multiple links.
• Then, It can decrease delay and can increase throughput on both TCP application and UDP
application, and simultaneously.

7. References
Arkin, D. (2004). My Life, Arkin Publishing, Arkinson.
Mitoralll, J. & Maguire, G. (1999). Cognitive Radio: Making Software Radios More Personal,
IEEE Pers. Comm., Vol. 6, No. 4, pp. 13–14 1999.
Mitoralll, J. (1999). Cognitive Radio for Flixible Multimedia Communications, Proc. MoMuC99,
pp. 3–10, 1999.
Harada, H. (2005). A Study on a new wireless communications system based on cognitive
radio technology, IEICE Tech. Rep., Vol. 105, No. 36, pp. 117–124, 2005.
3GPP TS 22.258. (2005). Service Requirements for the All-IP Network (AIPN); Stage 1, v2.0.0,
2005.
ITU-T. (2006). Y.2021: NGN Release 1, 2006.
Phatak, D. S. & Goff, T. (2002). A novel mechanism for data streaming across multiple IP
links for improving throughput and reliability in mobile environments, Proc. IEEE
INFOCOM, pp. 773–781, 2002.
Snoeren, A. C. (2002). Adaptive Inverse Multiplexing for Wide-Area-Wireless Networks, Proc.
IEEE GlobCom’99, Vol.3, pp.1665–1672, 1999.
Shrama, P.; Lee, S.; J. Brassil, J. & Shin, K. (2007). Aggregating Bandwidth for Multimode
Mobile Collaborative Communities, IEEE Tans. on MC, Vol. 6, No. 3, pp. 280–296,
2007.
Chebrolu, K. & Rao, R. (2006). Bandwidth Aggregation for Real-Time Applications in
Heterogeneous Wireless Networks, IEEE Tans. on MC, Vol. 5, No. 4, pp. 388–403, 2006.
Hsieh, H.; Kim, K. & Sivakumar, R. (2004). An end-to-end approach for transparent mobility
across heterogeneous wireless networks, Mob. Netw. Appl., Vol. 9, No. 4, pp. 363–378,
2004.
296
30 Recent Advances in Wireless Communications andCommnucations
Wireless Networks

Zhang, M.; Lai, J.; Krishnamurthy, A.; Peterson, L. & Wang, R. (2004). A Transport Layer
Approach for Improving End-to-End Performance and Robustness Using Redundant
Paths, USENIX 2004, 2004.
D. Gross, D. & C. Harris, C. (1985). Fundamentals of Queueing Theory 2nd ed, John Wiley &
Sons, 1985.
Little, J. (1961). A Proof of the Queueing Formula L = λW", Opre Res J., 18:172–174, 1961.
Bianchi, G. (2000). Performance analysis of the IEEE 802.11 distributed coordination function,
IEEE JSAC, Vol. 18, No. 3, pp. 535–547, 2000.
Carvalho, M. M. & Garcia-Luna-Aceves, J. J. (2003). Delay analysis of IEEE 802.11 in single-hop
networks, Proc. of ICNP, pp. 146 –155, 2003.
Takizawa, Y.; Taniguchi, N.; Yamanaka, S.; Yamaguchi, A. & Obana, S. (2008). Characteristics
of Packet Distribution in Wireless Access Networks Accommodating IEEE802.11 and
IEEE802.16, IPSJ Journal, Vol. 49, No. 9, pp. 3245–3256, 2008.
Takizawa, Y.; Taniguchi, N.; Yamanaka, S.; Yamaguchi, A. & Obana, S. (2008). Packet
Distribution Control for Wireless Access Networks Accommodating IEEE802.11 and
IEEE802.16, IPSJ Journal, Vol. 49, No. 10, pp. 3576–3587, 2008.
Bertsekas, D. & Gallager, R. (1992). Data Networks, Prentice Hall, 1992.
IEEE Std 802.16-2004. (2004). Local and Metropolitan Area Networks, Part 16: Air Interface
for Fixed Broadband Wireless Access Systems, 2004.
IIEEE Std. 802.16e-2005. (2005). Local and Metropolitan Area Networks, Part 16: Air Interface
for Fixed and Mobile Broadband Wireless Access System, 2005.
Takada, J. (2004). Radiowave Propagation for Mobile Satellite Communications, Tech. Rep. of
IEICE, Vol. 104, No. 671, pp. 13–16, 2004.
3GPP2. (2006). C30-20060823-004A Evaluation methodology V4.0, 2006.
Konishi, S.; Wang, X.; Kitahara, T.; Nakamura, H. & Suzuki, T. (2008). A Study on Ultra
Low-Latency Mobile Networks, Wireless Personal Comm.:An Int. Journal, Vol. 44, No.1,
pp.57–73, 2008.
Nakayo, D. J. and Hossain, E. (2006). A Queuing-Theoretic and Optimization-Based Model
for Radio Resource Management in IEEE 802.16 Broadband Wireless Networks, IEEE
Trans Comp., Vol. 55, No.11, pp. 1473–1488, 2006.
Cho, D.; Song, J.; Kim, M. and Han, K. (2006). Performance Analysis of the IEEE 802.16
Wireless Metropolitan Area Network, Proc. IEEE DFMA 2005., 2005.
Lin, L.; Jia, W. and Lu, W. (2007). Performance Analysis of IEEE 802.16 Multicast and Broadcast
Polling based Bandwidth Request, Proc. IEEE WCNC 2007., 2007.
Iyengar, R.; Iyer, P. and Biplab Sikdar, B. (2005). Delay Analysis of 802.16 based Last Mile
Wireless Networkst, Proc. IEEE GlobeCom 2005., 2005.
He, J.; Guild, K.; Yang, K. and Chen, H. (2007). Modeling Contention Based Bandwidth
Request Scheme for IEEE 802.16 Networks, IEEE Comm. Letter, Vol. 11, No.8, pp.
698–700, 2007.
Ni, Q.; Xiao, Y.; Turlikov, A. and Jiang, T. (2007). Investigation of Bandwidth Request
Mechanisms under Point-to-Multipoint Mode of WiMAX Networks, IEEE Comm.
Magazine, Vol. 4, No.4, pp. 477–486, 2007.
 Part 3

Applications and Realizations

 14

Wireless Sensor Network: At a Glance

A.K. Dwivedi1 and O.P. Vyas2
1School of Studies in Computer Science & Information Technology,
Pandit Ravishankar Shukla University, Raipur, C.G.,
2Indian Institute of Information Technology-Allahabad (IIIT-A),

Deoghat, Jhalwa, Allahabad, U.P.,

India

1. Introduction
Wireless Sensor Network is a technology which has capability to change many of the
Information Communication aspects in the upcoming era. From the last decade Wireless
Sensor Networks (WSNs) is gaining magnetic attention by the researchers, academician,
industry, military and other ones due to large scope of research, technical growth and nature
of applications etc. Wireless Sensor Networks (WSNs) employ a large number of miniature
disposable autonomous devices known as sensor nodes to form the network without the aid
of any established infrastructure. In a Wireless Sensor Network, the individual nodes are
capable of sensing the environments, processing the information locally, or sending it to one
or more collection points through a wireless link. Day to day applications of WSNs is
increasing from domestic use to military use and from ground to space.
The objective of this book chapter is to explore all aspects of WSNs under different modules
including these as well in a systematic flow: Sensor nodes, Existing hardware, Sensor node’s
operating systems, node deployment options, topologies used for WSN, architectures, WSN
lifecycle, Resource constraint nature, Applications, Existing experimental tools, Usability &
reliability of experimental tools, Routing challenges and Protocol design issues, Major existing
protocols, Protocol classifications, Protocols evaluation factors, Theoretical aspects of major
Energy Efficient protocols, Security issues, etc.
This chapter contains from very basic to high level technical issues obtained from highly
cited research contribution in a concluding manner but presenting whole aspects related to
this field.

2. Wireless sensor nodes and existing hardware

Wireless sensor nodes are tiny, light weight sensing devices consists of a constrained
processing unit, little memory, EEPROM or Flash memory for tiny operating systems and
other desired programs, one or more sensors, a limited range transceiver, battery or solar
based power unit and optionally a mobility subsystem for mobile sensor nodes (Dwivedi &
Vyas, 2010).
Tatiana Bokareva presented a mini hardware survey related to wireless sensor nodes
(Tatiana), except this a comprehensive listing of existing wireless sensor nodes is presented
300 Recent Advances in Wireless Communications and Networks

Communication
Subsystem
Mobility
Subsystem
Power Processing
Subsystem Unit
Sensor
Subsystem(s)

EEPROM &/or Program Accelerator Meter

Flash Memory Memory Humidity Meter
Magneto Meter
Pressure Meter
Temperature Meter . . . . .

Fig. 1. Block diagram of wireless sensor node

and maintained by Imperial College London (ICL, 2007), Embedded WiSeNts Platform Survey
(Embedded WiSeNts, 2006) presents an in-depth survey of five popular wireless sensor
nodes (ESB/2, BTnode, uNode, Tmote Sky, and EYES IFXv2), another pretty listing is
presented by University of California’s Sensor Network Systems Laboratory (Senses, 2005).
As well as Sensor Network Museum (SNM, 2010) maintained by TIK computer Engineering
and Networks Laboratory, ETH Zurich presents a collection of reference data and links for
commonly used wireless sensor nodes and related links. In a research contribution
(Manjunath, 2007), technical specifications of some well known wireless sensor nodes are
presented in tabular format, as here in its original (Table 1).
Resource footprint (Tatiana; ICL, 2007; Embedded WiSeNts, 2006; Senses, 2005; SNM, 2010;
Manjunath, 2007) for various currently available Wireless Sensor nodes provides us a
summary that most of the Nodes belongs to within the following configuration:
• 4-bit to 8-bit processor
• 512 Byte to 512 KB RAM (Program and Data Memory)
• 4 KB to 4 MB Flash/External Memory
• 250 Kbps 2.4 GHz IEEE 802.15.4 or Bluetooth 2.0 or 10 Kbps etc. as radio transceiver
On the basis of above mentioned resource footprint it can be concluded that each and every
currently available sensor nodes face limited resource problems such as narrow address
space and slow clock cycle of micro controller, small program and data memory as well as
external memory, low bandwidth and low range of transceivers.
Table 2 presents a wider look on technical aspects of some hardware systems for WSNs,
because hardware designing requires a holistic approach for WSNs, looking at all areas of
the design space. Expanding the uses of WSNs for various applications, expect more
performance for less power out of the hardware platforms. Envision a future of WSNs made
up of ultra low power nodes that provide high power computation and can be deployed for
decades is possible only with more research effort (Hempstead et al., 2008).

3. Operating systems for wireless sensor nodes

WSNs are composed of large numbers of tiny-networked devices that communicate
untethered. Operating systems are at the heart of the sensor node architecture. In terms of
 S.N. Platform MCU RAM Code RF Transceiver Frequency Radio range
Memory (feet)

1. Mica Atmel ATMega128L 4KB 128KB TR1000 433, 916 MHz 200

2. Mica2 Atmel ATMega128L 4KB 128KB CC1000 315, 433, 916 MHz 500

3. Mica2Dot Atmel ATMega128L 4KB 128KB CC1000 315, 433, 916 MHz 500

4. MicaZ Atmel ATMega128L 4KB 128KB CC2420 2.4 GHz 410

5. Cricket Atmel ATMega128L 4KB 128KB CC1000 433 MHz 500

6. TelosA TIMSP430 2KB 60KB CC2420 2.4 GHz 410

Wireless Sensor Network: At a Glance

7. TelosB TIMSP430 10KB 48KB CC2420 2.4 GHz 410

8. BTnode3 Atmel ATMega128L 64KB 128KB Zeevo-BT/CC1000 2.4 GHz/868 MHz 328/500

9. EYES TIMSP430 4KB 60KB TR1001 868 MHz 984

10. Intel mote ARM7TDMI (Core) 64KB 512KB Zeevo-BT 2.4 GHz 328

11. Intel mote2 PXA27x (Core) 256KB 32MB CC2420 2.4 GHz 410

12. MANTIS nymph Atmel ATMega128L 4KB 128KB CC1000 315, 433, 868, 915 MHz 500

13. XYZ mote ARM7TDMI (Core) 32KB 256KB CC2420 2.4 GHz 410

14. ECR TIMSP430 2KB 60KB TR1001 868 MHz 984

15. ESB TIMSP430 2KB 60KB TR1001 868 MHz 984

16. Smart-Its mote Atmel ATMega103L 4KB 128KB Ericsson-BT/TR1001 2.4 GHz/ 868 MHz 328/984

17. Tmote Sky TIMSP430 10KB 48KB CC2420 2.4 GHz 410

Table 1. A summarized list of some popular wireless sensor node (Manjunath, 2007)
18. TinyNode 584 TIMSP430 10KB 48KB Xemics XE1205 868 MHz 200

19. ZebraNet H/W TIMSP430 2KB 60KB 9XStream 902-928 MHz 328
301
 302

Event
Data path Throughput Energy
SN System Arch Style driven Circuit Techniques Accelerators Memory (KB) Process Voltage (V)
width (MIPS) (pJ/ins)
(Y/N)

Atmel GP Off-the-
1. 8 N N N 132KB 350nm 3.0 7.3 MHz 3200
ATmega128L shelf

(Hempstead et al., 2008)

GP Off-the-
2. TI MSP430 16 N N N 10KB NA 3.0 8 MHz 750
shelf

Timer, message 1.8 200 218

3. SNAP/LE GP RISC 16 Y Asynchronous 8KB 180nm
interface 0.6 23 24

GP RISC
Timer, message 1.8 54 152
4. BitSNAP Bit-serial 16 Y Asynchronous 8KB 180nm
interface 0.6 6 17
datapath

Synchronous - 2
5. Smart Dust GP RISC 8 N None 3.125KB 250nm 1.0 0.5 (500KHz) 12
clock

1.0V (high) 150µW

Protocol Two power Custom radio
6. Charm NA N 68KB 130nm 0.3-1.0V 8 MHz 53.6 µW
processor domains stack
(low) leakage

7. Michigan 1 GP 8 Y Sub-threshold None 0.25KB 130nm 0.360 833 KHz 2.6

8. Michigan 2 GP 8 Y Sub-threshold None 0.3125KB 130nm 0.350 354 KHz 3.52

Event driven Timer, filter,

9. Harvard 8 Y VDD-gating 4KB 130nm 0.55-1.2 12.5 MHz 680 pJ/task
accelerator message proc

Table 2. Technical specification for some hardware systems for Wireless Sensor Network
Recent Advances in Wireless Communications and Networks
Wireless Sensor Network: At a Glance 303

Wireless Sensor Networks we need these things in operating system architectures:

Extremely small footprint, extremely low system overhead and extremely low power
consumption. When designing or selecting operating systems for tiny-networked sensors,
our goal is to strip down memory size and system overhead because typical wireless sensor
nodes are equipped with a constrained processing unit, little memory, EEPROM or Flash
memory, battery or solar based power unit. In a research contribution (Hempstead et al.,
2008) and in a technical report (Fröhlich & Wanner, 2008) three classifications of O. S.
architectures are described for wireless sensor nodes: Monolithic, Modular/Micro and
Virtual Machine.
After evaluating various research contributions specifically devoted to operating systems
used for wireless sensor nodes (Fröhlich & Wanner, 2008, Reddy et al., 2007; Dwivedi et al.,
2009a; Manjunath, 2007) total 39 operating systems are identified:

1. TinyOS 2. Contiki 3. Mantis OS

YATOS
4. Microsoft .NET Micro 5. 6. BTnutOS or NutOS
(Yet Another Tiny OS)
7. PeerOS 8. Embedded Linux 9. NanoRK
10. μCOS 11. Squawk VM 12. SensorOS
13. MagnetOS 14. CORMOS 15. Bertha
16. kOS 17. VMSTAR 18. Maté
19. CVM 20. EYES 21. SenOS
22. DCOS 23. t-Kernel 24. Nano-QPlus
25. SmartOS 26. AVRX 27. Pixie
28. LiteOS 29. T2 30. OSSTAR
31. Jallad 32. CustomOS 33. GenOS
34. MoteWorks 35. NanoVM 36. ParticleVM
37. KVM 38. AmbiCompVM 39. SOS
Table 3. List of operating systems available for Wireless Sensor Nodes
D. Manjunath presents a review of current operating systems for WSNs (Manjunath, 2007)
whose aims were to explicate “why sensor operating systems are designed the way they
are”. This technical report questions every design decision, and provide a detail reasoning
for why these decisions.

4. Node deployment options in wireless sensor networks

As we know that WSN is deployed to measure environment parameters in Region of
Interest (ROI) and to send it to a controller node or base station. In WSNs how nodes will
deployed is basically application specific and totally dependent on environment. The node
deployment option affects the performance of routing protocol basically in terms of energy
consumptions. Basically there are three ways in which tiny sensor nodes can be deployed in
a wireless sensor network environment:
• Regular Deployment - Sensor nodes can be deployed in a well planned, fixed manner;
not necessarily geometric structure, but that is often a convenient assumption. In this
type of deployment data is routed through a predefined path.
Area of Use: Medical and health, Industrial sector, Home networks, etc.
304 Recent Advances in Wireless Communications and Networks

• Random Deployment – Sensor nodes are scattered over finite area. When the
deployment of nodes is not predefined optimal positioning of cluster head becomes a
critical issue to enable energy efficient network operation. Random deployment is
generally used in rescue operations.
Area of Use: Environmental and Habitual monitoring, etc.
• Sensor Nodes with Mobility - Can move to compensate for deployment shortcomings;
can be passively moved around by some external force (wind, water, vehicle); can
actively seek out “interesting” areas.
Area of Use: Battle field surveillances, Emergency situations (Fire, Volcano, Tsunami),
etc.

5. Topologies used for wireless sensor networks

Wireless sensor nodes are typically organized in one of three types of network topologies:
• In a star topology, each node connects directly to a gateway.
• In a cluster tree topology, each node connects to a node higher in the tree and then to
the gateway, and data is routed from the lowest node on the tree to the gateway.
• Finally, to offer increased reliability, mesh networks feature nodes that can connect to
multiple nodes in the system and pass data through the most reliable path available.

Star

Gateway
Mesh Sensor node
Cluster Tree
Fig. 2. Topologies used for Wireless Sensor Networks
Three phases related to topology maintenance and changes has been presented in a research
contribution (Akyildiz et al., 2002a):
• Pre-deployment and Deployment phase
• Post-deployment phase
• Redeployment of additional nodes phase

6. Architectures for wireless sensor networks

In a technical report (Karl & Willig, 2003) Holger Karl and Andreas Willig present views on
WSN architectures in the light of principle differences in application scenarios and
underlying communication technology. The architecture of WSNs will be drastically
different both concerning a single node and the network as a whole. Wide range of sensor
node architectures has been presented till today but as a general design principle all of them
have targeted the following objectives: energy efficiency, small size and low cost. The
architecture for network as a whole is a set of principles that guide where functionality
should be implemented along with a set of interfaces, functional units, protocols, and
physical hardware that follows those guidelines.
Wireless Sensor Network: At a Glance 305

In another research paper (Dulman & Havinga, 2005) the characteristics of wireless sensor
networks from an architectural point of view is presented. Since WSNs are designed for
specific applications so there is no precise architecture to fit them all but rather a common
set of characteristics that can be taken as a starting point. In same paper a data-centric
architecture is also presented.
A research paper (NeTS-NOSS, 2007) presents six aspects of architecture for WSN: Design
Principles, Functional Architecture, Programming Architecture, Protocol Architecture,
System Support Architecture and Physical Architecture. This paper also states that “The
situation today in sensor networks is that none of these six levels of network system
architecture are ‘solved’ or even clearly established. The vast majority of the studies fall in
the category of protocol architecture”.
In a research paper (Vazquez et al., 2009), an architecture for integrating Wireless Sensor
Networks into the Internet of Things called “Flexeo” is presented. In another research paper
(Schott et al., 2007) a flexible protocol architecture “e-SENSE” for WSNs has been
introduced, which is well-suited for capturing the context surrounding service users in
order to enable a variety of advanced context-aware applications in beyond 3G mobile
communication systems.

7. Wireless sensor networks lifecycle

Characteristically, there are four phases in the lifecycle of a wireless sensor network (the
implementation phase is omitted because the sensor code is frequently reused). Researchers
are usually involved in the planning and deployment phase, while the final customers are
more interested in monitoring and control the WSN.

Fig. 3. Wireless sensor network lifecycle

Planning WSNs Planning phase usually involves the inspection of the deployment area
and the selection of the correct locations to position the sensors in a way
that accomplishes the intended goal.
Deploying WSNs In the deployment phase, sensor nodes continually send their wireless
connection quality and route to the base.
Monitoring WSNs In this phase, the user interest is mainly focuses on the values read by
network sensors.
Controlling WSNs The application can also be used to control WSNs by sending commands
to the network. These commands can tell the network devices to stop
sending messages, increase the time between messages or even reset the
network (restart the Multi-Hop algorithm). In future, WSNs could be
controlled via a web interface or a handheld device, being easier to stop
and restart the network as needed.
306 Recent Advances in Wireless Communications and Networks

8. Resource constraint nature of wireless sensor networks

Wireless Sensor Networks (WSNs) employ a large number of miniature disposable
autonomous devices known as sensor nodes to form the network without the aid of any
established infrastructure. In a Wireless Sensor Network, the individual nodes are capable
of sensing their environments, processing the information locally, or sending it to one or
more collection points through a wireless link. Sensor node may fail due to lack of energy,
physical damage, communications problem, inactivity (a node becomes suspended), or
environmental interference. Resource footprint for various currently available Wireless
Sensor nodes is presented in section 2, obtained from (Tatiana; ICL, 2007; Embedded
WiSeNts, 2006; Senses, 2005; SNM, 2010; Manjunath, 2007). Here is a table focuses on
resource constraint nature of Wireless Sensor Nodes and obviously WSNs:

Node CPU Memory Radio

Rene ATMEL 8535 512 Byte RAM 10 Kbps
1999 8 KB Flash
Mica-2 ATMEGA 128 4 KB RAM 76 Kbps
2001 128 KB Flash
Telos Motorola HC 508 4 KB RAM 250 Kbps
2004
Mica-Z ATMEGA 128 4 KB RAM 250 Kbps
2004 128 KB Flash
BT Node ATMEL Mega 128L 128 KB Flash 4 KB EEPROM Bluetooth
2001 4 KB SRAM
Imote 1.0 ARM 7TDMI 64 KB SRAM Bluetooth
2003 512 KB Flash
Stargate Intel PXA 255 64 KB SRAM Serial
2003 Connection
Insysnc Cerfoube Intel PXA 255 32 KB Flash to Sensor
2003 64 KB SRAM Network
PC 104 X86 Processor 32 KB Flash
64 KB SRAM
Table 4. Presenting resource constraint nature of some popular wireless sensor nodes

9. Applications of wireless sensor networks

WSNs can be applied in a wide range of areas, such as: habitat monitoring and tracking,
disaster relief, emergency rescue operation, home networks, detecting
chemical/biological/radiological/nuclear/explosive material, monitoring patents and
elderly people, asset and warehouse management, building monitoring and control, fleet
monitoring, military battlefield awareness and surveillance, security and surveillance,
environmental monitoring, pipeline corrosion monitoring, homeland security, monitoring
conditions of buildings and bridges, industrial process monitoring and control, machine
health monitoring, healthcare applications, home automation, traffic control, etc.
With the help of research contributions (Biradar et al., 2009; Katiyar et al., 2010) a table is
presented here, which systematically summarized some applications for different areas:
Wireless Sensor Network: At a Glance 307

Area Applications
Military • Military situation awareness.
• Sensing intruders on basis.
• Detection of enemy unit movements on land and sea.
• Battle field surveillances
Emergency • Disaster management.
situations • Fire/water detectors.
• Hazardous chemical level and fires.
Physical world • Environmental monitoring of water and soil.
• Habitual monitoring.
• Observation of biological and artificial systems.
• Marginal Farming.
Medical and health • Sensors for blood flow, respiratory rate, ECG (electrocardiogram),
pulse oxymeter, blood pressure and oxygen measurement.
• Monitoring people’s location and health condition.
Industry • Factory process control and industrial automation.
• Monitoring and control of industrial equipment.
• Machine health monitoring.
Home networks • Home appliances, location awareness (blue tooth).
• Person locator.
Automotive • Tire pressure monitoring.
• Active mobility.
• Coordinated vehicle tracking.
Area monitoring • Detecting enemy intrusion
• Geo-fencing of gas or oil pipelines.
• Detecting the presence of vehicles.
Environmental • Air pollution monitoring.
monitoring • Forest fires detection.
• Greenhouse monitoring.
• Landslide detection.
• Volcano monitoring.
• Flood detection.
Water/Wastewater • Landfill ground well level monitoring and pump counter.
monitoring • Groundwater arsenic contamination assessment.
• Measuring water quality.
Cognitive sensing • Bio-inspired sensing.
• Swarm intelligence.
• Quorum sensing.
Underwater • Oceanographic data collection.
acoustic sensor • Pollution monitoring.
systems • Disaster prevention.
• Assisted navigation.
• Tactical surveillance.
Traffic Management • Traffic congestion control.
& Monitoring • Road Surface Condition Monitoring (BusNet in Sri Lanka).
Table 5. Some applications of WSNs in different areas
308 Recent Advances in Wireless Communications and Networks

Deploying nodes in an unattended environment will provide more possibilities for the
exploration of new applications. WSNs will be ubiquitous in the near future, due to new
opportunities for the interaction between humans and their physical world also WSNs are
expected to contribute significantly to pervasive computing.

10. Existing standards for wireless sensor networks

WSNs fascinate a number of standardization bodies to develop standards, due to a smaller
amount of standards exists for WSNs in comparison to other wireless networks. A number
of standards are currently under development or ratified for WSNs. Some standardization
bodies working in the specific field of WSNs to setup standards, such as:

Standardization body Specific work area for WSN

Institute of Electrical and Physical layer and MAC sub layer of Data link layer.
Electronics Engineers
Internet Engineering Data link layer and all above layers of WSN protocol stack.
Task Force
International Society of All layers of WSN protocol stack
Automation
DASH7 Alliance Promotes the use of the ISO 18000-7 standard for wireless
sensor networks.
Table 6. Some main Standardization bodies and their specific work area
Apart from these several non-standard, proprietary mechanisms and specifications also
exist. The most commonly used predominant standards in WSNs include:

IEEE 802.15.4 Standard for low-rate, wireless personal area networks, defines
the "physical layer" and the "medium access layer".
Zigbee ZigBee builds upon the 802.15.4 standard to define application
profiles that can be shared among different manufacturers.
IEEE 802.11 Standards efforts for low-power Wi-Fi.
IEEE 1451 The objective of this standard is to make it easier for different
manufacturers to develop smart sensors and to interface those
devices to networks.
ISA100 Addresses wireless manufacturing and control systems in the
areas of the: Environment, Technology and life cycle, and
Application of Wireless technology.
6LoWPAN IPv6 over low-power wireless networks, defines an adaptation
layer for sending IPv6 packets over IEEE 802.15.4.
uIPv6 uIPv6 is the world's smallest certified open source IPv6 stack
provides TCP/IP connectivity to tiny embedded 8-bit micro
controllers for low-cost networked device such as sensors and
actuators with maintained interoperability and RFC standards
compliance.
Table 7. Predominant standards in field of WSNs
Wireless Sensor Network: At a Glance 309

11. Existing experimental tools for wireless sensor networks

Research activities in the area of Wireless Sensor Networks (WSNs) need expositive
performance statistics about scenario, systems, protocols, gathered data, applications and
many more. There are various experimental tools for fulfilling these requirements, someone
are in practical use while other one are in literatures. In this part of chapter a glance on
currently available simulation tools/frameworks, emulators, visualization tools, testbeds,
debuggers, code-updaters and network monitoring tools used for wireless sensor networks
is presented (Dwivedi & Vyas, 2011).

11.1 Simulator/simulation framework

A simulator is a software that imitates selected parts of the behavior of the real world.
Depending on the intended usage of the simulator, different parts of the real-world system
are modeled and imitated. The parts that are modeled can also be of varying abstraction
level. A wireless sensor network simulator imitates the wireless media and the constraints
nodes in the network. Some sensor network simulators have a detailed model of the
wireless media including effects of obstacles between nodes, while other simulators have a
more abstract model.
Type of simulation
Simulators either run as in an asynchronous mode, event triggered mode, or in synchronous
mode, where events happen in parallel in fixed time slots (DCG’s Sinalgo, 2009):
• Synchronous simulation
The synchronous simulation is purely based on rounds.
• Asynchronous Simulation
The asynchronous simulation is purely event based.
Categorization of simulators
A large number of sensor network simulators have been proposed by researchers. In a
research contribution WSN Simulators are categorized (Eriksson, 2009) as:
• Generic Network Simulators
• Code Level Simulators
• Firmware Level Simulators
In another research contribution (Shu et al., 2009), simulators have been classified into the
following three major categories based on complexity:
• Algorithm Level Simulators
• Packet Level Simulators
• Instruction Level Simulators
Several simulators exist that are either adjusted or developed specifically for wireless sensor
networks. Here is a table presenting 63 simulators/simulation frameworks (Table 8).

11.2 Emulator or emulation environment

As a networked embedded system, a WSN application involves sensor node hardware, its
drivers, operating systems, and networking protocols. As a result, the performance of the
WSN application depends on all of these factors in addition to its implementation. An
emulator is a special type of simulator whose aims is to enable realistic performance
evaluation for WSN applications. Emulation environment or emulators are good choice, in
310 Recent Advances in Wireless Communications and Networks

1. Network Simulator (NS) 2. Mannasim (NS-2 3. DiSenS (Distributed

Extension for WSNs) SENsor network
Simulation)
4. (J) Prowler 5. LecsSim 6. WISDOM
7. TOSSIM 8. OPNET 9. Sinalgo
10. TOSSF 11. SENS 12. SENSORIA
13. PowerTOSSIMz 14. EmStar/Em* 15. Capricorn
16. ATEMU 17. EmTOS 18. SIDnet-SWANS
19. COOJA 20. SenQ 21. Stargate Simulator
(starsim)
22. GloMoSim (Global 23. H-MAS 24. JiST/SWANS (Java in
Mobile Information (Heterogeneous Mobile Simulation Time/
Systems Simulation) Ad-hoc Sensor-Network Scalable Wireless Ad
Simulation Environment) hoc Network Simulator)
25. QualNet 26. SensorSim 27. SNSim
28. SENSE 29. Shawn 30. SNIPER-WSNSim
31. VisualSENSE 32. NetTopo 33. SNAP
34. AlgoSenSim 35. Atarraya 36. SimPy
37. Georgia Tech Network 38. SSFNet (Scalable 39. Mule
Simulator (GTNetS) Simulation Framework)
40. OMNet++ 41. WiseNet 42. CaVi
43. Castalia 44. SimGate 45. Ptolemy
46. J-Sim (formerly JavaSim) 47. SimSync 48. Maple
49. Mote simulator 50. SNetSim 51. WISENES (WIreless
(motesim) SEnsor NEtwork
Simulator)
52. JiST/SWANS++ 53. SensorMaker 54. WSNet-Worldsens and
WSim
55. Avrora 56. TRMSim-WSN 57. LSU SensorSimulator
58. Sidh 59. PAWiS 60. WSNGE
61. Prowler 62. OLIMPO 63. TikTak
Table 8. Simulator/simulation frameworks specifically designed for WSNs
which WSN applications can be directly run for testing, debugging, and performance
evaluation. Additionally, studies on the lower layers (e.g., hardware drivers, OS, and
networking) as well as cross-layer techniques can also be done in this environment by
plugging the target modules into the emulator. Here is a table which presents 14 emulators:

1. VMNET 2. Freemote 3. UbiSec&Sens

4. ATEMU 5. EmPro 6. Emuli
7. Emstar 8. NetTopo 9. MSPSim
10. TOSSIM 11. OCTAVEX 12. MEADOWS
13. AvroraZ/Avrora 14. SENSE
Table 9. Emulators specifically designed for WSNs
Wireless Sensor Network: At a Glance 311

11.3 WSN data visualization tools

With the increase in applications for sensor networks, data manipulation and representation
have become a crucial component of sensor networks. The data gathered from WSNs is
usually saved in the form of numerical form in a central base station. There are many
programs that facilitate the viewing of these large amounts of data. These special programs
are called data visualization tool for WSNs. Visualization tools can support different data
types, and visualize the information using a flexible multi-layer mechanism that renders the
information on a visual canvas. Here is a table presenting 19 data visualization tools (Parbat
et al., 2010) that are especially designed and developed for WSNs applications:

1. SpyGlass 2. TOSGUI 3. Oscilloscope

4. MoteView 5. MSR Sense 6. GSN
7. TinyViz 8. Trawler 9. WiseObserver
10. XbowNet 11. SNAMP 12. SenseView
13. MonSense 14. Surge Network Viewer 15. MeshNetics WSN Monitor
16. NetTopo 17. Mica Graph Viewer 18. MARWIS
19. Octopus
Table 10. Data visualization tools specifically designed for WSNs

1. Motelab 2. NetEye 3. Sharesense

4. NESC-Testbed 5. INDRIYA 6. Trio
7. WUSTL 8. CLARITY 9. sMote
10. CitySense 11. GNOMES 12. CTI-WSN Testbed
13. Kansei 14. WSNTB 15. FEEIT WSN Testbed
16. MISTLAB 17. TWIST 18. Roulette
19. Orbitlab 20. X-sensor 21. BigNet
22. Emulab 23. ENL Sensor Network 24. UCR Wireless Networking
Testbed Research Testbed
25. WISEBED (Wireless 26. Imote2 Sensor 27. SWOON (Secure Wireless
Sensor Network Network Testbed Overlay Observation
Testbeds) Network)
28. REALnet 29. PICSENSE 30. WHYNET
31. KonTest 32. SOWNet 33. CENS-Testbed
34. SANDbed 35. IP-WSN Testbed 36. SCADDS WSN Testbeds
37. BANAID 38. SenseNet 39. Crossbow WSN Testbed
40. Motescope 41. Omega 42. GaTech Testbed
43. Tutornet: A Tiered 44. CENSE (A Century of 45. Intel Research Berkeley’s 150-
Wireless Sensor Sensor nodes) mote SensorNet Testbed
Network Testbed
46. WINTeR (Wireless Industrial Sensor 47. FireSenseTB: A wireless sensor networks
Network Testbed for Radio-Harsh testbed for forest fire detection (Kosucu et
Environments) al., 2009)
Table 11. Testbeds used for experimental usage specifically for WSNs
312 Recent Advances in Wireless Communications and Networks

11.4 Testbeds for WSN

To achieve high-fidelity in WSN experiments use of testbed is very productive. Testbeds are
an environment that provides support to measure number of physical parameters in
controlled and reliable environment. This environment contains the hardware,
instrumentations, simulators, various software and other support elements needed to
conduct a test. Generally, testbeds allow for rigorous, transparent and replicable testing. By
providing the realistic environments for testing the experiments, the testbeds bridge the gap
between the simulation and deployment of real devices. The testbeds thus deployed can
improve the speed of innovation and productive research. Here is a table presenting 47
testbeds, used for experimental purposes in various universities, colleges, research
institutions or by individuals (Table 11).

11.5 Debugging tools/services/concepts

Due to extreme resource constraints nature, deployment in harsh and unattended
environments, lack of run-time support tools and limited visibility into the root causes of
system and application level faults make WSNs notoriously difficult to debug. Currently,
most debugging systems in WSNs are aimed at diagnosing specific faults, such as detection
of crashed nodes, sensor faults, or identifying faulty behavior in nodes. There are few
debugging solutions for WSNs available, with a fairly wide range of goals and feature sets.
Debuggers for WSNs have been categorized (Tavakoli, 2007) into three distinct categories:
source-level debuggers, query-oriented debuggers, and decision-tree debuggers. Here is a
table presenting 26 debuggers/debugging concepts/debugging concepts:

1. Clairvoyant 2. S2DB 3. ActorNet

4. Dustminer 5.
Envirolog 6. ANDES
7. Sympathy 8.
NodeMD 9. EvAnT
10. FIND 11.
StackGaurd 12. KleeNet
13. Passive Distributed 14.
Storage-centric 15. Model-based diagnosis for
Assertions (PDA) method for Debugging WSNs
16. Chowkidar 17. Marionette 18. Post-Deployment Performance
Debugging (PD2)
19. Nucleus-NMS 20. REDFLAG 21. Declarative Tracepoints
22. Debugging WSNs 23. Monitored External 24. SNTS: Sensor Network
Using Mobile Actors Global State (MEGS) Troubleshooting Suite
25. Wringer 26. MDB
Table 12. Debugging tools/services/concepts specifically useful for WSNs

11.6 Code-updation/reprogramming tool

Large scale WSNs may be deployed for long periods of time during which the requirements
from the network or the environment in which the nodes are deployed may change. This
may necessitate modifying the executing application or re-tasking the existing application
with different sets of parameters, which will collectively refer to as code-
updation/reprogramming. The relevant forms of code-updation/reprogramming are (Panta
et al., 2009):
• Remote Multi-hop Reprogramming
• Incremental Reprogramming
Wireless Sensor Network: At a Glance 313

Incremental Reprogramming poses several challenges. A class of operating systems, including

the widely used TinyOS, does not support dynamic linking of software components on a
node. SOS and Contiki, do support dynamic linking, however, their reprogramming
support also does not handle changes to the kernel modules. Here is a table presenting 10
code-updaters/reprogramming (Table 13).

1. Trickle 2. Deluge 3. Hermes

4. FlexCup 5. Stream 6. FIGARO
7. Zephyr 8. MNP (Multi-hop network 9. Multihop Over-the-Air
reprogramming) Programming (MOAP)
10. MARWIS (Management ARchitecture for WIreless Sensor Networks)
Table 13. Code-updaters/Reprogramming tools specifically designed for WSNs

11.7 Network monitoring tools

WSNs are typically composed of low cost tiny hardware devices and tend to be unreliable,
with failures a common phenomenon. Accurate knowledge of network health status,
including nodes and links of each type, is critical for correctly configuring applications on
really deployed WSN and/or WSN testbeds and for interpreting the data collected from
them. Here is a table presenting 8 networks monitoring:

1. Memento 2. Sympathy 3. LiveNet

4. NUCLEUS 5. HERMES 6. Chowkidar
7. DiMo 8. MARWIS (Management Architecture for heterogeneous
Wireless Sensor Networks)
Table 14. Network monitoring tools specifically designed for WSNs

12. Usability & reliability of experimental tools

The statistics gathered from experimental tools can be realistic and convenient, but due to
cost of large number of sensors most researches in wireless sensor networks area is
performed by using these experimental tools in various universities, institutes, and research
centers before implementing real one. These experimental tools provide the better option for
studying the behavior of WSNs before and after implementing the physical one.
Simulators are commonly used for rapid prototyping and also used for the evaluation of
new network protocols and algorithms as well as enable repeatability because they are
independent of the physical world and its impact on the objects. Moreover, simulations
enable nonintrusive debugging at the desired level of detail. In a research contribution
various factors have been presented that influences simulation results (Dwivedi et al., 2010).
For successful WSN development cooperation not only between test-beds and simulators
but also between simulators is required, however, simulators are usually not designed with
cooperation in mind (Li et al., 2010).

13. Routing challenges & protocol design issues in WSNs

Routing in WSNs is very challenging due to unique inherent characteristics (energy
efficiency and awareness, connection maintenance, minimum resource usage limitation, low
 SN Main Category Sub Categories
314

1. Classification • Flat-based or Data Centric routing: In flat-based routing algorithm, all nodes play the same role and
based on mainly apply flood based data transferring.
Network • Hierarchical-based or Cluster based routing: Hierarchical protocols aim at clustering the nodes so that
Structure cluster heads can do some aggregation and reduction of data in order to save energy. Hierarchical
(Al-Karaki & routing is mainly two-layer routing where one layer is used to select cluster heads and other for
Kamal, 2004) routing.
• Location-based routing: Location-based protocols utilize the position information to relay the data to the
desired regions rather than the whole network.
2. Classification • Multipath-based routing: This type of routing protocols uses multiple paths instead of a single path in
based on order to enhance network performance.
Protocol • Query-based routing: In this type of routing protocol destination nodes propagate a query for data
Operation (sensing task) from a node through the network, and a node with this data sends the data that matches
(Al-Karaki & the query back to the node that initiated the query.
Kamal, 2004) • Negotiation-based routing: These protocols use high-level data descriptors in order to eliminate
redundant data transmissions through negotiation. Communication decisions are also made based on
the resources available to them.
• QoS-based routing: In QoS-based routing protocols, the network has to balance between energy
consumption and data quality. In particular, the network has to satisfy certain QoS metrics (delay,
energy, bandwidth, etc.) when delivering data to the base station.
• Non-coherent & Coherent data-processing based routing: In non-coherent data processing routing, nodes
will locally process the raw data before it is sent to other nodes for further processing.
3. Classification • Gossiping and agent-based unicast forwarding: These schemas are an attempt of working without routing
based on tables in order to minimize the overflow needed to build the tables, as much as result of the initial
Packet stages in which the tables were not built yet.
Destination • Energy-efficient unicast: These techniques analyze the network nodes distribution to set the cost of
(Karl & Willig, transmitting over the link between two nodes and select an algorithm to calculate the minimum cost.

Table 15. Protocol classifications and sub-classifications for WSNs

2006) • Broadcast and multicast: Many nodes must collect or distribute the information to every node in the
network (broadcast). In a similar way, sometimes it is necessary to distribute data to a subset of
previously known nodes. This process is called multicast.
• Geographic routing: This kind of routing appeared due to two main motivations: (1) sending data
randomly to every node in a given region is called geo-casting; (2) the destination node location must be
specified geographically or relatively (with a location service).
• Mobile nodes: These aspects with motion ability should be considered for wireless sensor networks:
mobile sensor nodes, mobile base station, mobile sensed phenomenon or combination of these.
4. Crossbow (Xbow) • Basic routing (with normal or improved variants)
classification • Reliable routing
(Olivares et al., • Low Power routing
2007) • XMesh routing
latency, load balancing in terms of energy used by sensor nodes, etc.) that distinguish this
network from other wireless networks such as mobile ad hoc networks, cellular networks,
Recent Advances in Wireless Communications and Networks
 SN Main Category Sub Categories

5. Classification • Stateful Ad Hoc routing: Stateful ad hoc routing protocols require node to maintain some routing
based on State information that is collected using the routing protocol (e.g., through route request propagation or by
(Eriksson, 2009) reversing paths taken by the query).
• Stateless Geometric Ad Hoc routing: These kinds of protocols only track the position of their neighbors
and select among them a neighbor that is likely to be closer to the destination.
6. Classification • Pull based epidemic algorithm: A node asks a selected neighbor for new information. The node will receive
based on new information only if the neighbor has new information.
Epidemic • Push based epidemic algorithm: A node with new information sends the information to a selected
behavior (Akdere neighbor.
et al., 2006) • Pull-push based epidemic algorithm: This algorithm is a combination of two models described above.
7. Classification • Protocols operating on flat topology (WSN consisting Homogeneous nodes)
based on Sensor • Protocols operating on hierarchical topology(WSN consisting Heterogeneous nodes).
Wireless Sensor Network: At a Glance

Node Architecture
(Al-Karaki &
Kamal, 2004)
8. Classification • Source-initiated (Src-initiated): A source-initiated protocol sets up the routing paths upon the demand of
based on the source node, and starting from the source node. Here source advertises the data when available and
Protocol’s initiates the data delivery.
initialization • Destination-initiated (Dst-initiated): A destination initiated protocol, on the other hand, initiates path
point (Biradar et setup from a destination node.
al., 2009)
9. Classification • Proactive: A proactive protocol sets up routing paths and states before there is a demand for routing
based on how the traffic. Paths are maintained even there is no traffic flow at that time. This approach is best suited for
source finds the applications having fixed nodes
destination • Reactive: In reactive routing protocol, routing actions are triggered when there is data to be sent and
(Biradar et al., disseminated to other nodes. Here paths are setup on demand when queries are initiated. This
2009) approach is best suited for applications mobile nodes
• Hybrid: This approach combines both techniques.
10. Classification • Protocols that control the transmission power level at each node by increasing network capacity while
based on the basis keeping the network connected.
of how to reduce • Protocols that make routing decisions based on power optimization goals.

Table 15. (continues) Protocol classifications and sub-classifications for WSNs

useful energy • Protocols that control the network topology by determining which nodes should participate in the
consumption network operation (be awake) and which should not (remain asleep).
(Younis & Fahmy,
2004)
11. Cooperative • In this approach, sensor nodes send data to a central node that join the data to reduce the cost in terms
routing (Castillo of energy consumption.
et al., 2007)
315

and wireless mesh networks. Major Constraints while designing protocols for WSNs are:
Energy, Processing power, Memory. In various literatures or research contributions, related
316 Recent Advances in Wireless Communications and Networks

to WSNs these design challenges are identified (Dwivedi et al., 2009a; Eriksson, 2009;
Al-Karaki & Kamal, 2004; Karl & Willig, 2006; Akyildiz et al., 2002b; Akkaya & Younis, 2005;
Wachs et al., 2007).
• Due to the relatively large number of sensor nodes, it is not possible to build a global
addressing scheme for the deployment of a large number of sensor nodes as the
overhead of ID maintenance is high. Thus, traditional IP based protocols may not be
applied to WSNs.
• In contrast to typical communication networks, almost all applications of sensor networks
require the flow of sensed data from multiple sources to a particular Base Station.
• Sensor nodes are tightly constrained in terms of energy, processing, and storage
capacities. Thus, they require careful resource management.
• In most application scenarios, nodes in WSNs are generally stationary after deployment
except for, may be, a few mobile nodes.
• Sensor networks are application specific, i.e., design requirements of a sensor network
change with application.
• Position awareness of sensor nodes is important since data collection is normally based
on the location.
• Finally, data collected by various sensors in WSNs is typically based on common
phenomena; hence there is a high probability that this data has some redundancy.
Visibility (Wachs et al., 2007) is a new metric for WSNs protocol design. The objective of this
visibility metric is that “Minimize the energy cost of diagnosing the cause of a failure or
behavior”.

14. Major existing protocols for wireless sensor networks

A lot of protocols has been proposed in various research contributions, some of them are as
follows: Rumor, DSR, SER (Stream Enabled Routing), AODV, SPIN (Sensor Protocols for
Information via Negotiation) (SPIN-PP, SPIN-EC, SPIN-BC, SPIN-RL), GRAB, Direct
Diffusion, GAF, SEER (Simple Energy-Efficient Routing), GBR, ARPEES, TIDD, TEEN,
CADR, APTEEN, ACQUIRE, CEDAR, COUGAR, SAR, TinyAODV, PEQ (Periodic Event-
driven and Query-based), GEAR, HPEQ (Hierarchical PEQ), MECN, CPEQ (Cluster PEQ),
SMFCN, HEAP (Hierarchical Energy Aware Protocol for Routing & Aggregation in Sensor
Networks), GF, PEGASIS (Power Efficient Gathering in Sensor Information System), GF-
RSST, HPEGASIS (Hierarchical PEGASIS), LEACH, etc.
Some good research contributions (Al-Karaki & Kamal, 2004; Wachs et al., 2007) presents
survey on existing WSN Protocols, whereas some other good one are dedicated to
comparison, classification and other aspects of WSN Protocols (Dwivedi & Vyas, 2010;
Biradar et al., 2009; Al-Karaki & Kamal, 2004; Wachs et al., 2007; Castillo et al., 2007).

15. Existing protocol classifications for wireless sensor networks

A careful attention is needed while selecting or proposing a new routing protocols for
wireless sensor networks because WSNs are challenging due to the inherent characteristics
such as energy efficiency and awareness, connection maintenance, minimum resource usage
limitation, low latency, load balancing in terms of energy used by sensor nodes, etc. Various
classifications for WSNs are presented in different literatures, at a glance these are (Table
15).
Wireless Sensor Network: At a Glance 317

16. Protocol evaluation factors

These are the some parameters on which routing protocols must be evaluated during
designing new one:

Evaluation Parameter Description

Power Usage Sensor node’s lifetime is clearly dependent on its power
source, thus useful power usage must be which involves:
transmitting/receiving data, processing query requests,
forwarding queries/data to neighboring nodes.
Data Aggregation Substantial energy savings and traffic optimization can be
obtained through data aggregation.
Scalability The possibility to enlarge and reduce the network.
Reliability or Fault Fault tolerance is the ability to sustain WSN functionalities
Tolerance without any interruption due to node failures.
Latency (delay) and Multi-hop relays and data aggregation cause data latency,
Overhead these important factors influences routing protocol design.
Data Delivery Model Data delivery model (Continuous, Event-driven, Query-
driven , Hybrid) (Ahvar & Fathy, 2010) determines when the
data collected by the sensor node has to be delivered.
Quality of Service (QoS) Quality service required by the application, involves: length
of life time, data reliability, energy efficiency, location-
awareness, collaborative-processing, etc. QoS factors will
affect the selection of routing protocols for a particular
application.
Security Security concerns needs special attention in current era
where data stealing and data diddling becomes major issue.
Node Deployment Node deployment option affects the performance of routing
option protocol basically in terms of energy consumptions.
Topology Topology of a WSN affects many of its characteristics like;
latency, capacity, and robustness. As well as, the complexity
of data routing and processing depends on the network
topology.
Sensor Density and Sensor density of nodes affects the degree of coverage area of
Network Size interest whereas networks size affects reliability, accuracy,
and data processing algorithms.
Environment or Scenario A critical parameter, because node and network lifetime is
directly dependent on it.
Byte Overhead (Saaranen Byte overhead means the total number of bytes in the routing
& Pomalaza-Ráez, 2004) control messages needed to find a route to the sink. For
flooding, byte overhead means the total number of bytes in
the extra messages flooded throughout the network. In both
cases the bytes in the data packets transmitted by nodes
along the route from the originating node to the sink node
are not counted as overhead.
Table 16. WSN Protocol evaluation factors
318 Recent Advances in Wireless Communications and Networks

Except these there are exist some common performance metrics, including latency,
throughput, success rates, energy consumption and energy load, that must be calculated for
the evaluation of routing algorithms.

17. Theoretical aspects of major energy efficient protocols

A classification on energy efficient/aware routing protocols is available in a research
contribution (Ahvar & Fathy, 2010) which classified this type of protocols into: Energy Saver
and Energy Manager. Energy saver protocols decrease energy consumption totally because
most of them try to find the shortest path between source and destination to reduce energy
consumption. The objective of energy manager protocols is to balance energy consumption
in networks to avoid network partitioning. In first approach finding best route is totally
based on energy balancing consideration, it may lead to long path with high delay and
decreases network lifetime whereas in later approach finding best route only with the
shortest distance consideration may lead to network partitioning. A lot of researches were
conducted on the energy efficiency/awareness issue, some are presented here (Table 17)

18. Security issues in wireless sensor networks

In a survey paper (Dwivedi et al., 2009b) different classes of adversaries, and considers
security goals in each scenario (indoor and outdoor) of WSNs, including: sensor nodes,
networks of sensor nodes, operating systems, applications, middleware, and internet, are
presented. This paper also presents valuable, in-depth recommendations of how to design
and implement a security strategy for WSN. A procedure for protecting systems makes sure
that the facility is physically secure, provides a recovery/restart capability, and has access to
backup files establishing a priority sequence, one would probably want to start from within
the firm and work out. Threats and their usual defenses are illustrated in (Figure 4)
Most WSN routing protocols are quite simple thus sometimes even more susceptible to
attacks. Most network layer attacks against sensor networks falls under one of the following
categories: Selective forwarding, Sinkhole attacks, Sybil attacks, Wormholes, HELLO flood
attacks, Spoofed/Altered/Replayed routing information, Acknowledgement spoofing.
Some security issues that must need attention in wireless sensor networks, are as follows:
Secure routing, Secure discovery and verification of location, Key establishment and trust
setup, Attacks against sensor nodes, Secure group management, and Secure data
aggregation.
In the ideal world, a secure routing protocol should guarantee the integrity, authenticity,
and availability of messages in the presence of adversaries of arbitrary power. Every eligible
receiver should receive all messages intended for it and be able to verify the integrity of
every message as well as the identity of the sender. Several countermeasures and design
considerations are also proposed in a research contribution (Karlof & Wagner, 2003).
Some mechanisms for authentication and security are based on public key cryptography.
Public key cryptography is too expensive for sensor nodes. Security protocols for sensors
networks must rely exclusively on efficient symmetric key cryptography. These protocols
are too expensive in terms of node state and packet overhead and are designed to find and
establish routes between any pair of nodes - a mode of communication not prevalent in
sensor networks. Tackling with natural and manmade disasters is only possible with proper
planning.
 S.N. Energy Efficient Major Theoretical Aspects
Protocol
1. TEEN - First protocol for reactive networks with enhanced efficiency.
(Threshold - Time critical data reaches the user almost instantaneously. Eminently well suited for time critical data
sensitive Energy sensing applications.
Efficient sensor - Message transmission consumes much more energy than data sensing. So, even though the nodes sense
Network continuously, the energy consumption in this scheme can potentially be much less than in the proactive
protocol) network, because data transmission is done less frequently.
(Manjeshwar & - The soft threshold can be varied, depending on the criticality of the sensed attribute and the target
Agarwal, 2001) application.
- A smaller value of the soft threshold gives a more accurate picture of the network, at the expense of
increased energy consumption. Thus, the user can control the trade-off between energy efficiency and
accuracy.
- At every cluster change time, the attributes are broadcast afresh and so, the user can change them as
required.
Wireless Sensor Network: At a Glance

- The main drawback of this scheme is that, if the thresholds are not reached, the nodes will never
communicate; the user will not get any data from the network at all and will not come to know even if
all the nodes die. Thus, this scheme is not well suited for applications where the user needs to get data
on a regular basis.
- Another possible problem with this scheme is that a practical implementation would have to ensure that
there are no collisions in the cluster.
2. APTEEN - A Protocol for Hybrid network (inherit best characteristics of both proactive and reactive network).
(Adaptive - To provide periodic data collection as well as near real-time warnings about critical events.
Periodic - By sending periodic data, it gives the user a complete picture of the network. It also responds
Threshold- immediately to drastic changes, thus making it responsive to time critical situations. Thus, it combines
sensitive Energy both proactive and reactive policies.
Efficient Sensor - It offers a flexibility of allowing the user to set the time interval (TC) and the threshold values for the
Network attributes.
Protocol) - Energy consumption can be controlled by the count time and the threshold values.
(Manjeshwar & - The hybrid network can emulate a proactive network or a reactive network, by suitably setting the count
Agarwal, 2002) time and the threshold values.
- The main drawback of this scheme is the additional complexity required to implement the threshold
functions and the count time. However, this is a reasonable trade-off and provides additional flexibility
and versatility.
3. HEED (Hybrid - An energy-efficient clustering protocol, using residual energy as primary parameter and network
Energy-Efficient topology features (e.g. node degree, distances to neighbors) as secondary parameters.
Distributed - Here all nodes are assumed to be homogenous nodes (with same initial energy).
clustering) - It extends the basic scheme of LEACH.

Table 17. Major theoretical aspects of some major energy efficient protocols for WSNs
(Younis & - The clustering process is divided into a number of iterations, as well as in each iteration nodes which are
Fahmy, 2004) not covered by any cluster head doubles their probability of becoming a cluster head.
- Since it enable every node to independently and probabilistically decide on its role in the clustered
network, thus cannot guaranteed optimal elected set of cluster heads.
319
 S.N. Energy Efficient Major Theoretical Aspects
320
Protocol
4. H-HEED - A protocol for heterogeneous WSN.
(Heterogeneous - Cluster head selection is primarily based on the residual energy of each node. Since the energy
- HEED) (Kour & consumed per bit for sensing, processing, and communication is typically known, and hence residual
Sharma, 2010) energy can be estimated.
- Intra cluster communication cost is considered as the secondary parameter to break the ties, tie means
that a node might fall within the range of more than one cluster head.
- Different level of heterogeneity is introduced: 2-level, 3-level and multi-level in terms of the node
energy.
- In 2-level H-HEED, two types of sensor nodes, i.e., the advanced nodes and normal nodes are used.
- In 3-level H-HEED, three types of sensor nodes, i.e. the super nodes, advanced nodes and normal nodes
are used.
- In this heterogeneous approach all the sensor nodes are having different energy as a result nodes will
die randomly.
- Multi-level H-HEED prolongs lifetime and shows better performance than other level of H-HEED and
HEED protocol.
5. Reactive Energy - To solve the problem of limited energy, the loading of nodes have to be distributed as possible as it can.
Decision Routing - If the energy consumption can be shared averagely by most nodes, then the lifetime of sensor networks
Protocol will be enlarged.
(REDRP) (Ying- - This protocol will create the routes in reactive routing method to transmit the data node gathered.
Hong et al., 2006) - It uses the residual energy of nodes as the routing decision for energy-aware.
6. PEGASIS - A near optimal chain-based protocol and an enhanced descendant of LEACH.
(Power-Efficient - It has two main objectives: increases the lifetime of each node by using collaborative techniques and
Gathering in allow only local coordination between nodes that are close together so that the bandwidth consumed in
Sensor communication is reduced.
Information - Nodes route data destined ultimately for the base station through intermediate nodes.
Systems) - In determining the routes only consider the energy of the transmitter and neglect the energy dissipation
(Lindsey & of the receivers.
Raghavendra, - It assumes that each sensor node can be able to communicate with the base-station directly and all nodes
2002) maintain a complete database about the location of all other nodes in the network.
- The method of which the node locations are obtained is not outlined.
- It also assumes that all sensor nodes have the same level of energy and they are likely to die at the same
time.
7. Hierarchical- - Its objective is to decrease the delay incurred for packets during transmission to the base-station.
PEGASIS - In its concept only spatially separated nodes are allowed to transmit at the same time.
(Savvides - This chain-based protocol with CDMA capable nodes, constructs a chain of nodes, that forms a tree like
et al., 2001) hierarchy, and each selected node in a particular level transmits data to the node in the upper level of
the hierarchy, that ensures data transmitting in parallel and reduces the delay significantly.
- Results shows that this hierarchical extension of PEGASIS performs better than the regular PEGASIS
scheme by a factor of about 60.
8. SHPER - Enhanced integration of a hierarchical reactive routing protocol.
(Scaling - It supposes the coexistence of a base station and a set of homogeneous sensor nodes which are randomly
Hierarchical distributed within a delimited area of interest.
Power Efficient - Consists of two phases: the initialization phase and the steady state phase.
Routing) - Hard and soft thresholds are utilized in the SHPER protocol as with TEEN.
(Kandris et al., - Best suited in real life applications where imbalance in energy distribution is the common case.

Table 17. (continues) Major theoretical aspects of some major energy efficient protocols for WSNs
2009) - Network scalability is retained because it adopts both multi-hop routing and hierarchical architecture.
Recent Advances in Wireless Communications and Networks
 S.N. Energy Efficient Major Theoretical Aspects
Protocol
9. TREnD - A novel cross-layer WSN protocol for control applications.
(Timely, - The routing algorithm of TREnD is hierarchically subdivided into two parts: a static route at inter
Reliable, clusters level and a dynamical routing algorithm at node level. This is supported at the MAC layer by
Energy-efficient hybrid TDMA/CSMA solution.
and Dynamic) - The protocol parameters are adapted by an optimization problem, whose objective function is the
(Marco network energy consumption, and the constraints are the reliability and latency of the packets.
et al., 2010) - It uses a simple algorithm that allows the network to meet the reliability and latency while minimizing
for energy consumption.
- It is best fit for industrial environments.
10. LEACH (Low - A most popular cluster-based protocol, which includes distributed cluster formation.
Energy Adaptive - The idea is to form clusters of the sensor nodes based on the received signal strength and use local
Clustering cluster heads as routers to the sink.
Hierarchy) - It randomly selects a few sensor nodes as cluster-heads and rotates this role to evenly distribute the
Wireless Sensor Network: At a Glance

(Heinzelman et energy load among the sensors in the network.

al., 2000) - Its operation is separated into two phases: setup phase where clusters are organized and CHs are
selected and steady state phase where the actual data transfer to the base station takes place.
- It uses a TDMA/CDMA MAC to reduce inter-cluster and intra-cluster collisions.
- Optimal number of cluster heads is estimated to be 5% of the total number of nodes.
- This protocol is most appropriate for the applications when there is a need for constant monitoring.
11. SEER (Simple - A protocol that considers energy saving and balancing, with poor idea about energy balancing.
Energy - Once the network has been deployed in the area where it is to operate, the sink transmits a broadcast
Efficient packet.
Routing) - Each node in the network is assumed to have a unique address within the network.
(Hancke & - When a node observes new data, it initiates the process of routing. Two types of data packets can be
Leuschner, sent: normal data and critical data.
2007) - When nodes receive a data message they update the remaining energy value in the neighbor table for
the neighbor that sent the message. Nodes that forward data messages follow the same process, except
for minor differences.
- If node's remaining energy falls below a certain threshold, it transmits an energy message to all of its
neighbors to inform them of its energy level.
- The sink node periodically sends a broadcast message through the network so that nodes can add new
neighbors that joined the network to neighbor tables and remove neighbors that have failed from the
neighbor tables.
- Nodes also update remaining energy values stored in the neighbor tables.
12. BEAR - An extended version of SEER protocol with some visible difference specially in forwarding data
(Balanced procedure that saves and balance energy consumption in WSNs.
Energy-Aware - Finds optimal route in energy level and hop count both.
Routing) - Routing decisions in BEAR are based on the distance to the base-station as well as on remaining battery
(Ahvar & Fathy, energy level of nodes on the path towards the base station.
2010) - BEAR is better than the SEER protocol in energy managing, due to the fact that BEAR sends data packet

Table 17. (continues) Major theoretical aspects of some major energy efficient protocols for WSNs
along a balanced path.
321
322 Recent Advances in Wireless Communications and Networks

Fig. 4. Security threats and their usual defenses in Wireless Sensor Networks (Dwivedi et al.,
2009b)

19. Reference
Dwivedi, A. K. & Vyas, O.P. (2010). Network Layer Protocols for Wireless Sensor Networks:
Existing Classifications and Design Challenges, International Journal of Computer
Applications (IJCA), Vol.8, No.12, Article 6, pp. 30-34.
Tatiana, M. (2010). Mini Hardware Survey. Available from
http://www.cse.unsw.edu.au/~sensar/hardware/hardware_survey.html
Wireless Sensor Network: At a Glance 323

Imperial college London, U.K. (2007). Body Sensor Networks. Available from
http://ubimon.doc.ic.ac.uk/bsn/index.php?m=206
Embedded WiSeNts Platform Survey (2006). Available from
http://www.embedded-wisents.org/studies/survey_wp2.html
Senses (2005). Available from http://senses.cs.ucdavis.edu/resources.html
The Sensor Network Museum (2010). Available from
http://www.snm.ethz.ch/Main/Homepage
Hempstead, M.; Lyons, M.J.; Brooks D. & Wei, G.-Y. (2008). Survey of Hardware
Systems for Wireless Sensor Networks, Journal of Low Power Electronics, Vol.4,
pp. 1-10.
Fröhlich, A.A. & Wanner L.F. (2008). Operating System Support for Wireless Sensor
Networks, Journal of Computer Science, Vol.4, No.4, pp. 272-281.
Reddy, A.M.; Kumar, V.A.V.U.P.; Janakiram, D, & Kumar, G.A. (2007). Operating Systems
for Wireless Sensor Networks: A Survey, Technical Report, IIT Madras, Chennai,
India.
Dwivedi, A. K.; Tiwari, M.K. & Vyas, O.P. (2009). Operating Systems for Tiny Networked
Sensors: A Survey, International Journal of Recent Trends in Engineering, Vol.1, No.2,
pp. 152-157.
Manjunath, D. (2007). A Review of Current Operating systems for Wireless Sensor
Networks, Technical Report, Department of ECE, Indian Institute of Science,
Bangalore, INDIA.
Akyildiz, I.; Su, W.; Sankarasubramaniam, Y. & Cayirci, E. (2002). A survey on Sensor
Networks, IEEE Communications Magazine, Vol.40, Issue:8, pp. 102-114.
Karl, H. & Willig, A. (2003). A Short Survey of Wireless Sensor Networks, TKN Technical
Report TKN-03-018, Technical University Berlin, Germany.
Dulman, S. & Havinga, P. (2005). Architectures for Wireless Sensor Networks, Proceedings of
the IEEE ISSNIP 2005, pp. 31-38.
NeTS-NOSS: Creating an Architecture for Wireless Sensor Networks, (2007). Available from
http://snap.cs.berkeley.edu/documents/architecture.pdf
Vazquez, J.; Almeida, A.; Doamo, I.; Laiseca, X. & Orduña, P. (2009). Flexeo: An Architecture
for Integrating Wireless Sensor Networks into the Internet of Things, Proceedings of
3rd Symposium of Ubiquitous Computing and Ambient Intelligence 2008, Springer
Berlin/Heidelberg, Vol.51, pp. 219-228, 2009.
Schott, W.; Gluhak, A.; Presser, M.; Hunkeler U. & Tafazolli, R. (2007). e-SENSE Protocol
Stack Architecture for Wireless Sensor Networks. Proceedings of 16th IST Mobile and
Wireless Communication Summit, pp. 1-5.
Biradar, R.V.; Patil, V.C.; Sawant, S.R. & Mudholkar, R.R. (2009). Classification and
Comparison of Routing Protocols in Wireless Sensor Networks, Special Issue on
Ubiquitous Computing Security Systems, UbiCC Journal, Vol.4, pp. 704-711.
Katiyar, V.; Chand, N. & Chauhan, N. (2010). Recent advances and future trends in Wireless
Sensor Networks, International Journal of Applied Engineering Research, Vol.1, No.3,
pp. 330-342, ISSN 0976-4259.
Distributed Computing Group’s Sinalgo-Simulator for Network Algorithms. (2009).
Available from http://disco.ethz.ch/projects/sinalgo/tutorial/tuti.html
324 Recent Advances in Wireless Communications and Networks

Eriksson, J. (2009). Detailed Simulation of Heterogeneous Wireless Sensor Networks,

Dissertation for Licentiate of Philosophy in Computer Science, Uppsala University,
Sweden, ISSN 1404-5117.
Shu, L.; Hauswirth, M.; Zhang, Y.; Mao, S.; Xiong N. & Chen, J. (2009). NetTopo: A
Framework of Simulation and Visualization for Wireless Sensor Networks,
Proceedings of the ACM/Springer Mobile Networks and Applications.
Parbat, B.; Dwivedi A.K. & Vyas, O.P. (2010). Data Visualization Tools for WSNs: A
Glimpse, International Journal of Computer Applications, Vol.2, No.1, pp.14-20, ISSN
0975-8887.
Kosucu, B.; Irgan, K.; Kucuk, G.; & Baydere, S. (2009). FireSenseTB: A Wireless Sensor
Networks Testbed for Forest Fire Detection, Proceedings of the IWCMC.
Tavakoli, A. (2007). Wringer: A Debugging and Monitoring Framework for Wireless Sensor
Networks, Proceedings of the ACM SenSys Doctoral Colloquium.
Panta, R.K.; Bagchi, S. & Midkiff, S.P. (2009). Zephyr: Efficient Incremental Reprogramming
of Sensor Nodes using Function Call Indirections and Difference Computation,
Proceedings of the USENIX. Available from
http://www.usenix.org/events/usenix09/tech/full_papers/panta/panta_html
Dwivedi, A.K.; Patle, V.K. & Vyas, O.P. (2010) Investigation on Effectiveness of Simulation
Results for Wireless Sensor Networks, CCIS, Springer-Verlag Berlin Heidelberg,
Vol.70, pp. 202-208.
Li, Q.; Österlind, F.; Voigt, T.; Fischer, S. & Pfisterer, D. (2010). Making Wireless Sensor
Network Simulators Cooperate, Proceedings of the PE-WASUN’10, Bodrum, Turkey,
pp. 95-98.
Al-Karaki, J.N. and Kamal, A.E. (2004). Routing Techniques in Wireles Sensor Networks:
A Survey, IEEE Wireless Communications (J), pp. 06-28.
Karl, H. and Willig, A. (2006). Protocols and Architectures for Wireless Sensor Networks,
Editorial John Wiley & Sons Ltd., ISBN 13978-0-470-09510-2.
Akyildiz, I.F.; Su, W.; Sankarasubramaniam, Y. & Cayirci, E. (2002). Wireless Sensor
Networks: A Survey, Computer Networks (Elsevier) (J), Vol.38, pp.393-422.
Akkaya, K. and Younis M. (2005). A Survey on Routing Protocols for Wireless Sensor
Networks, Ad Hoc Network (Elsevier) (J), Vol.3, pp. 325-349.
Wachs, M.; Choi; Jung, J. II; Lee, W.; Srinivasan, K.; Chen, Z.; Jain, M. and Levis, P. (2007).
Visibility: A New Metric for Protocol Design. Proceedings of ACM SenSys.
Olivares, T.; Tirado, P.J.; Royo, F.; Castillo, J.C. & Orozoco-Barbosa, L.: (2007). IntellBuilding:
A Wireless Sensor Network for Intelligent Buldings. Poster. Proceedings of 4th
European Conference on Wireless Sensor networks (EWSN).
Akdere, M.; Bilgin, C.C.; Gerdaneri, O.; Korpeoglu, I.; Ulusoy, Ö. and Cetintemel, U. (2006).
A Comparison of Epidemic Algorithms in Wireless Sensor Networks. Elsevier
Journal of Computer Communications, Vol.29, pp. 2450-2457.
Younis, O. & Fahmy, S. (2004). HEED: A Hybrid, Energy-Efficient, Distributed Clustering
Approach for Ad Hoc Sensor Networks. IEEE transactions on Mobile Computing,
Vol.3, pp. 366-379.
Castillo, J.C.; Olivares, T. & Orozco-Barbosa, L. (2007). Routing Protocols for Wireless Sensor
Networks-Based Network. Technical Report, Albacete Research Institute of
Informatics, University of Castilla, SPAIN.
Wireless Sensor Network: At a Glance 325

Tilak, S.; Abu-Ghazaleh, N.B. & Heinzelman, W. (2002). A Taxonomy of Wireless

Microsensor Network Models, ACM SIGMOBILE Mobile Computing and
Communications Review, Vol.6, Issue 2, pp. 28–36.
Saaranen, A. & Pomalaza-Ráez, C.A. (2004). Comparison of Reactive Routing and
Flooding in Wireless Sensor Networks, Proceedings of Nordic Radio Symposium,
Oulu, Finland.
Ahvar, E. & Fathy, M. (2010). BEAR: A Balanced Energy-Aware Routing Protocol for
Wireless Sensor Networks. Wireless Sensor Network, Vol.2, pp. 793-800.
Manjeshwar, A. & Agarwal, D.P. (2001). TEEN: a Routing Protocol for Enhanced
Efficiency in Wireless Sensor Networks. Proceedings of 1st International Workshop
on Parallel and Distributed Computing Issues in Wireless Networks and Mobile
Computing.
Manjeshwar, A. & Agarwal, D.P. (2002). APTEEN: A Hybrid Protocol for Efficient Routing
and Comprehensive Information Retrieval in Wireless Sensor Networks.
Proceedings of International Parallel and Distributed Processing Symposium (IPDPS), pp.
195-202.
Kour, H. & Sharma, A.K. (2010). Hybrid Energy Efficient Distributed Protocol for
Heterogeneous Wireless Sensor Network. International Journal of Computer
Applications, Vol.4, pp. 1-5.
Ying-Hong, W.; Yi-Chien, L.; Ping-Fang, F. & Chih-Hsiao, T. (2006). REDRP: Reactive
Energy Decisive Routing Protocol for Wireless Sensor Networks. Ubiquitous
Intelligence and Computing, LNCS, Vol.4159, pp. 527-535, Springer
Berlin/Heidelberg.
Lindsey, S. & Raghavendra, C. (2002). PEGASIS: Power-Efficient Gathering in Sensor
Information Systems, Proceedings of IEEE Aerospace Conference, Vol.3, pp. 1125-
1130.
Savvides, A.; Han, C-C & Srivastava, M. (2001). Dynamic Fine-grained localization in Ad-
Hoc Networks of Sensors. Proceedings of 7th ACM Annual International Conference on
Mobile Computing and Networking (MobiCom), pp. 166-179.
Kandris, D.; Tsioumas, P.; Tzes, A.; Nikolakopoulos, G. & Vergados, D.D. (2009). Power
Conservation through Energy Efficient Routing in Wireless Sensor Networks.
Sensors, 9, pp. 7320-7342, ISSN 1424-8220.
Marco, P.D.; Park, P.; Fischione, C. & Johansson, K.H. (2010). TREnD: a Timely, Reliable,
Energy-efficient and Dynamic WSN Protocol for Control Applications. Proceedings
of Information Communication Conference.
Heinzelman, W.; Chandrakasan, A. & Balakrishnan, H. (2000). Energy-Efficient
Communication Protocol for Wireless Microsensor Networks. Proceedings of 33rd
Hawaii International Conference on System Sciences (HICSS '00).
Hancke, G.P. & Leuschner, C.J. (2007). SEER: A Simple Energy Efficient Routing Protocol for
Wireless Sensor Networks, South African Computer Journal, Vol.39, pp.17-24.
Dwivedi, A.K.; Tiwari, M.K. & Vyas, O.P. (2009). A Review of Security in Wireless Sensor
networks for Indoor Application Scenario: Prospects and Challenges, Proceedings of
National Conference on Wireless Communication and Networking (WINCON), pp. 138-
148.
326 Recent Advances in Wireless Communications and Networks

Karlof, C. & Wagner, D. (2003). Secure routing in wireless sensor networks: Attacks and
countermeasures. Proceedings of 1st IEEE International Workshop on Sensor Network
Protocols and Applications.
Dwivedi, A.K. & Vyas, O.P. (2011). An Exploratory Study of Experimental Tools for Wireless
Sensor Networks. Wireless Sensor Network,Vol. 3, ISSN 1945-3078 (Print), 1945-3086
(Online). Available from http://www.scirp.org/journal/wsn
 0
15

Software Defined Radio Platform for Cognitive

Radio: Design and Hierarchical Management
Amor Nafkha, Christophe Moy, Pierre Leray,
Renaud Seguier and Jacques Palicot
SUPELEC/IETR, Avenue de la Boulaie,
Cesson Sévigné Cedex,
France

1. Introduction
Cognitive Radio (CR) Mitola (2000) is a promising technology to improve spectrum utilization
of wireless communication systems. Current investigations in CR have been focused on the
physical layer functionality. The cognitive radio, built on a software-defined radio, assumes
that there is an underlying system hardware and software infrastructure that is capable of
supporting the flexibility needed by the cognitive algorithms. As already foreseen by Mitola
Mitola & Maguire (1999), a Cognitive Radio is the final point of Software Defined Radio (SDR)
platform evolution: a fully reconfigurable radio that changes its communication functions depending
on network and/or user demands. Mitola’s definition on reconfigurability is very broad and we
only focus here on the reconfigurability of the hardware platform for Cognitive Radio. SDR
basically refers to a set of techniques that permit the reconfiguration of a communication
system without the need to change any hardware system element. As explained in the
schematic of figure 1, this relies on a cognitive circle. Figure 1 (a) is from Mitola (2000) and
figure 1 (b) is a simplified view of the cycle summarized in three main steps:
• Observe: gathers all the sensing means of a CR,
• Decide: represents all that implies some intelligence including learning, planning decision
taking,
• Adapt: reconfigures the radio, designed with SDR principles, in order to be as flexible as
possible.
The figure 2 draw the general approach that can help the radio to better adapt its functionality
for a given service in a given environment without restriction on the sensors nature.
Sensors are classified in function of the OSI layers they correspond to, with a rough division
in three layers. Corresponding to the lower layers of the OSI model, we find specifically
all the sensing information related to the physical layer: propagation, power consumption,
coding scheme, etc. At the intermediate level are all the information that participate to
vertical handover, or can help to make a standard choice, as a standard detection sensor
for instance. The network load of the standards supported by the equipment may also be
of interest. It also includes the policies concerning the vicinity, the town or the country.
The highest layer is related to the applications and all that concerns the human interaction
328
2 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Fig. 1. (a) Mitola’s cognitive cycle, (b) simplified version

Fig. 2. Simplified OSI model for cognitive radio context

with the communicating device. It is related to everything that concerns the user, his habits,
preferences, policies, profile. If a user has the habit to connect to a video on demand service
every evening while coming back home from office by metro, a CR terminal should be
aware of it to plan all the requirements in terms of battery life, sufficient quantity of credit
on his contract, vertical handover succession depending on each area during the trip, etc.
The equipment can be aware of its environment with the help of sensors like microphone,
video-camera, bio-sensors, etc. As we are at the early beginning of such technology, it is
difficult to foresee all the possibilities. We can think, for instance, that user’s biometric
information and/or facial recognition will ensure equipment security. Video-camera could
also be used to indicate if the terminal is outside or inside a building. This may impact
propagation features, but also the capability or not to receive GPS signals. Another example
could be given in the context of video conferencing, a separation between the face of the
speaker and the background could help decreasing the data rate while refreshing slowly the
background of the image Nafkha et al. (2007).
Note that this classification is also related to three well-known concepts of the literature:
• Context aware for higher layers Chen & Kotz (2000),
• Interoperability for intermediate layers Aarts et al. (2001),
• Link adaptation for lower layers Qiu & Chuang (1999).
All this may be combined to achieve cross-layer optimizations. This is one of the
responsibilities of the cognitive engine in our mind. However, due to the high financial
pressure on spectrum issues, CR is often restricted in the research community to spectrum
management aspects as in Fan et al. (2008); Ghozzi et al. (2006). Opportunistic spectrum
Software Defined
Software Defined Radio
Radio Platform Platform
for Cognitive for and
Radio: Design Cognitive Radio: Design and Hierarchical Management
Hierarchical Management 3293

access approaches are explored to increase the global use of the spectrum resources. FCC has
been already opening the door for several years, in the TV broadcasting bands, and permits
secondary users (e.g. not licensed) to occupy primary users spectrum when available.
More futuristic CR scenarios may also be considered concerning the spectrum management.
We may even imagine in the very long term a fully deregulated spectrum access where all
radio connections features would be defined on-the-fly: carrier frequency, modulation, data
rate, coding scheme, etc. But this means also to overcome regulatory issues in addition to
technological challenges.

2. Background and related work

The objective behind this section is to highlight other cognitive radio platforms and to give
our architecture purpose.

2.1 Related work

There are a large number of experimental SDR platforms that have been developed to support
individual research projects. The various experimental SDR platforms have made different
choices in how they are addressed the issues of flexibility, partitioning and application. To
highlight the variety of architectures, five popular platforms will be discussed briefly prior to
introducing our platform.
• NICT SDR Platform: The Japanese National Institute of Information and Communications
Technology (NICT) constructed a software defined radio platform to trial next generation
mobile networks. The platform has two embedded processors, four Xilinx Virtex2 FPGA
and RF modules that could support 1.9 to 2.4 and 5.0 to 5.3 GHz. The signal processing
was partitioned between the CPU and the FPGA, with the CPU taking responsibility for the
higher layers. An objective of this platform was to explore selection algorithms to manage
handover between existing standards. To this end, a number of commercial standards
were implemented, for example 802.11a/b/g, digital terrestrial broadcasting, WCDMA and
a general OFDM communication scheme.
• Berkeley Cognitive Radio Platform: This platform is based around the Berkeley Emulation
Engine (BEE2) which is a platform that contains five high-powered Xilinx Virtex2 FPGAs
and can connect up to eighteen daughter-boards. In the Cognitive Radio Platform radio,
daughter-boards have been designed to support up to 25 MHz of bandwidth in an 85 MHz
range in the 2.4 GHZ ISM Band. The RF modules have highly sensitive receivers and to
avoid self-generated noise operate either concurrently at different frequencies (FDD) or at
the same frequency in a time-division manner. This cognitive radio platform requires only
a low-bandwidth connection to a supporting PC as all signal processing is performed on
the platform.
• Kansas University Agile Radio (KUAR): The KUAR platform was designed to be a low-cost
experimental platform targeted at the frequency range 5.25 to 5.85 GHz and a tunable
bandwidth of 30 MHz. The platform includes an embedded 1.4 GHz general purpose
processor, Xilinx Virtex2 FPGA and supports gigabit Ethernet and PCI-express connections
back to a host computer. This allows for all, or almost all processing to be implemented on
the platform.
330
4 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

• OpenAirInterface: The mobile communications department at EURECOM proposed an

open-source hardware/software development platform and open-forum for innovation
in the area of digital radio communications. OpenAirInterface implements in software
the Physical and Medium-access layers for wireless communications as well as providing
a IPv4/IPv6/MPLS network device interface under Linux. The initiative targets
4th generation wireless systems (UMTS Longterm-evolution (LTE), 802.16e/j) and
rapidly-deployable MESH networks using a similar radio interface technologies. The
development can be seen as an open-source testbed for advanced algorithmic prototyping
and performance evaluation.
• Universal Software Radio Peripheral (USRP): The USRP is one of the most popular SDR
platforms currently available and it provides the hardware platform for the GNU Radio
project. The first USRP system, released in 2004, was a USB connected to a computer with
a low-performance FPGA. The FPGA was used primarily for routing information but also
allowed some limited signal processing. The USRP could realistically support about 3
MHz of bandwidth due primarily to the performance restrictions of the USB interface. The
second generation platform was released in September 2008 and utilizes gigabyte Ethernet
to allow support for 25 MHz of bandwidth. The system includes a medium range Xilinx
Spartan3 device which allows for a local processing. The radio-frequency performance of
the USRP was limited and is more directed towards experimentation rather than matching
any communications standard.
Our proposed platform has been developed in order to achieve high flexibility and
reconfigurability of the wireless baseband processing. For the hardware part, for example,
we exploited the ability to reconfigure partial areas of an FPGA anytime after its initial
configuration. Our development concerns all processing blocs: from the video treatment to
the intermediary frequency signal generation. Our intention is not to develop any commercial
platform, but just to test and verify our approach to achieve baseband flexibility using:
• Partial Reconfiguration Nafkha et al. (2007) and Common Operator Alaus et al. (2008).
• Hiearchical Reconfiguration Management Delahaye et al. (2005).
• Hierarchical and Distributed Cognitive Radio Architecture Management Godard. (2009).

2.2 The proposed solution

The proposed solution is a design approach and not a hardware platform itself so that it is not
restricted to a specific hardware platform. It intends to answer the design issue of SDR in the
following context:
• flexible processing including partial FPGA reconfiguration and Common Operator
approach.
• heterogeneous processing, including processors (GPP, DSP), FPGA and ASICs,
• portability from a HW device target to another.
In order to cope with these characteristics, a modular-based approach is privileged. This is the
main support of flexibility. It permits indeed to separate the radio application into sub-pieces
that can be split in any sub-set depending on the HW devices that compute their processing
needs. This also favorites changes in the repartition of the processing modules on the HW
devices. As all processing modules are designed independently in a modular-based approach,
Software Defined
Software Defined Radio
Radio Platform Platform
for Cognitive for and
Radio: Design Cognitive Radio: Design and Hierarchical Management
Hierarchical Management 3315

this also guarantees the non dependence of processing modules in terms of operating rhythm.
One can just not make them run faster than their fastest speed, but anything lower is
compatible.
This is very straightforward in a processor environment as the processing modules varies with
the processor frequency (or it architecture after compilation). But this is generalized to the
reconfigurable HW world while using Globally Asynchronous Locally Synchronous (GALS)
principles. It turns HW processing as SW in the sense that the exchanges between processing
modules are asynchronous from the data rate they have to process. The consequence is
that these processing modules can be ported to several designs at different speeds, with no
dependence with the speed of other blocks. Another major effect is that it becomes transparent
to replace a SW processing module, e.g. running in a processor, by a HW processing module,
e.g. running in a FPGA, and vice versa. Moreover, a HW processing module can be easily
moved to a processor instantiated inside a FPGA (such as a NIOS for Altera or a MicroBlaze
for Xilinx) without reconsidering the global behavior of the processing modules it is connected
to.
This design approach is completely compatible with an Intellectual property (IP) oriented
design strategy. Re-usability has several major advantages: gains of time at all development
stage, debug and validation stages, and integration stage. It permits also to benefit from third
party expertise to speed-up or complete the proprietary designs. To sum-up the proposed
solution consists in declaring rules for the design of IPs or processing modules so that they
can be easily assembled in the design framework that is detailed below.

3. System structure
The presented real-time platform provides a simple wireless video stream broadcasting
system to verify and test our approach. It consists of one transmitter as the base station and
one receiver as the terminal. The system architecture is depicted in figure 3. Basically, the
transmitter and receiver hosts can communicate and exchange their data through an existing
TCP/IP networks or Intermediate Frequency (IF) link. The transmitter host utilizes USB port
to communicate with the video camera. At the receiver side, any standard display monitor
allows us to display the incoming video stream.

3.1 Hardware architecture

The digital hardware setup of the whole system is based on Sundance modules. The
transmitter and receiver side contain a Sundance SMT310Q carrier boards, plugged via
Peripheral Component Interconnect (PCI) bus to a standard PC. The hardware architecture
is depicted in figure 4.
At the transmitter side shown at figure 5, the Sundance SMT310Q carrier-board is used
to carry the processing modules (SMT395, SMT348 and SMT350 ADC/DAC) in the four
available TIM-40 slots. The Sundance SMT395 module is placed in the first TIM-slot and
controls the operation of other modules. It consists principally a Texas Instruments (TI) 6416T
fixed-point Digital Signal Processor (DSP) running at 1 GHz, a Xilinx Virtex II Pro FPGA, and
two Sundance High-speed Bus (SHB, up to 400MB/s) for fast data exchange with the other
modules. In our platform the DSP is used as a control device for the ADC/DAC and memory
modules and to set the parameters for the pre-distortion filter running in real-time on the
FPGA at the module SMT350. Based on the Xilinx Virtex4 range, the SMT348 features 16MB
332
6 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

of blistering fast QDRII memory, ensuring ample capacity to develop todays demanding
applications. The SMT348 includes SHB and SLB (Sundance LVDS Bus) interfaces. It provides
quick and easy connection to rapid ADC and DAC modules for data acquisition or software
radio systems. The SMT350 module, is composed of:
• Two DACs DAC5686 from Texas Instrument with 16 bits of resolution and a maximum
sampling rate of 500MSPS with interpolation filters
• Two ADCs ADS5500 from Texas Instrument with 14bits of resolution and a maximum
sampling frequency of 125MHz
• A CDCM7005 from Texas Instrument which provides individual sample frequency to each
converters

Fig. 3. Hardware Architecture

The stream server program encapsulates the video data into Internet Protocol (IP) stream and
saves the IP stream in the buffer allocated in the main memory of the host PC. The DSP module
fetches the data in the buffer through the PCI interface provided by the above mentioned
carrier board and then executes the partial part of the digital baseband and Intermediate
Frequency (IF) signal processing algorithms of the transmitter. The driver of the carrier board
offers the DSP module the methods to access the main memory of the host PC through PCI
interface by providing C/C++ Application Program Interface (API) functions. The Xilinx
FPGA on the DSP module takes care of the Sundance High speed Bus (SHB) interfacing
between the DSP module SMT395 and the FPGA module SMT348. The SHB interface is able
to transfer 32-bit data at a 100 MHz clock. Via SHB the digital IF signal is forwarded to the
SMT350 to generate the analog signal using its integrated Digital to Analog Converter (DAC).
The analog IF signal goes through the low-pass filter.
The hardware setup of the receiver is similar to the transmitter, as shown in figure 5. In this
case, the SMT350 is configured as an Analog to Digital Converter (ADC) module and the
signal experiences the reciprocal of the transmitter. The IF signal coming from the transmitter
is sampled synchronously by the ADC on the SMT350 module. The FPGA module SMT348
receives the digital samples from the SHB interface and accomplishes a high parallel part of
Software Defined
Software Defined Radio
Radio Platform Platform
for Cognitive for and
Radio: Design Cognitive Radio: Design and Hierarchical Management
Hierarchical Management 3337

digital signal process algorithms of the receiver. The simplest part of the baseband process is
sent to the SMT395 module via the SHB.
The final received IP packets are saved to the buffer in the main memory of the host PC
through the PCI interface. The network layer program fetches the IP packets from the buffer
and emits them to the certain IP port by IP socket programming. The video stream player
always listens to the IP port and plays the video back.
In both side, transmitter/receiver, the testbed platform contains a Graphics Processor Unit
(GPU). The main reason behind is that the GPU is specialized in compute intensive, highly
parallel computation and therefore is designed in such a way that more transistors are devoted
to data processing rather than data caching and flow control.

Fig. 4. Transmitter and receiver overview

3.2 Software architecture

The two host stations are a standard personal computers running Microsoft Windows XP
and Microsoft Visual C++. Several hardware and software tools, as depicted in figure 6, are
necessary for the completion of our testbed. These tools include the physical DSP/FPGA and
their associated development board that allowed for continual reprogramming of test systems
as well as many features for data storage and output display. Xilinx has also supplied a suite
of tools that are used in our platform. These Xilinx software tools are used for developing the
hardware and software aspects of the system. Although many of these tools have included
documentation from Xilinx, their support of partially reconfigurable systems is currently
somewhat lacking. Therefore, the integration of these tools into a working tool flow to achieve
the goal of a partial reconfiguration for example required research from numerous sources and
some experimentation with the tools. For the partial reconfiguration implementation, we need
the following Xilinx tools:
334
8 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

• Xilinx EDK provides a framework for design of hardware/software components of

the embedded processor systems on programmable logic. Appropriate tools for each
stage of the design in addition to IBM PowerPC and Xilinx MicroBlaze processor cores
infrastructure and peripheral IP cores facilitate hardware/software partitioning and
design reuse.
• Xilinx ISE allows for complete FPGA development. It can automatically interpret the HDL
syntax, synthesize the description, place and route the logic elements and then provide a
software BIT file description to connect these logic elements together to create the circuit
described in the HDL. All these tool flow steps require their own respective application
program to perform the function.
• Xilinx PlanAhead is a floor-planning tool provided by Xilinx to allow developers flexibility
on how there synthesis designs should be placed on the FPGA floorplan. This tool is
useful in ASIC designs where locations of logic elements are an important factor to the
performance of the application. For the scope of our platform, PlanAhead has partial
reconfiguration options which make it a required tool in the partial reconfiguration tool
flow.
The programming of the TI 6416T fixed-point DSP is achieved using the Code Composer
Studio (CCS) Integrated Development Environment (IDE). The CCS IDE allows a user to
connect, program in C, and run the DSP through a graphical user interface. furthermore,
a user is able to view the memory contents of the DSP and profile the execution time for
pieces of their code all in real-time. For high level synthesis of the digital signal process
algorithms, we use the SynDEx tool environment Grandpierre et al. (1999) which provides a
formal framework based on graphs and system-level computer-aided design (CAD) software.
On the one hand, this tool specifies the functions of the applications, the distributed resources
in terms of processors and/or specific integrated circuits, and communication media. On
the other hand, it assists the designer in implementing the functions onto the resources
while satisfying timing requirements and, as far as possible, minimizing the resources. The
results is a real-time behavior of the application functions executed on various resources,
like processors, integrated circuits or communication media. For the software part of the
application, code is automatically generated as a dedicated real-time executive.
As a result of demand for video treatment at the both side (transmitter/receiver), the CUDA
programming model is used. It is ANSI C extended by several keywords and constructs. The
GPU is treated as a co-processor that executes data-parallel kernel code. The user supplies a
single source program encompassing both host (CPU) and kernel (GPU) code. Each CUDA
program consists of multiple phases that are executed on either the CPU or the GPU. The
phases that exhibit little or no data parallelism are implemented in host (CPU), which is
expressed in ANSI C and compiled with the host C compiler. The phases that exhibit rich data
parallelism are implemented as kernel functions in the device (GPU) code. A kernel function
defines the code to be executed by each of the massive number of threads to be invoked for a
data-parallel phase. These kernel functions are compiled by the NVIDIA CUDA C compiler
and the kernel GPU object code generator. There are several restrictions on kernel functions:
there must be no recursion, no static variable declarations, and a non-variable number of
arguments. The host code transfers data to and from the GPU’s global memory using API
calls. Kernel code is initiated by performing a function call.
Software Defined
Software Defined Radio
Radio Platform Platform
for Cognitive for and
Radio: Design Cognitive Radio: Design and Hierarchical Management
Hierarchical Management 3359

SynDEx tools Grandpierre et al. (1999) provides a formal framework based on graphs and
system-level software. On the one hand, these specify the functions of the applications,
the distributed resources in terms of processors and/or specific integrated circuit and
communication media, and the non-functional requirements such as real-time performances.
On the other hand, they assist the designer in implementing the functions onto the resources
while satisfying timing requirements and, as far as possible, minimizing the resources. This
is achieved through a graphical environment (see figure 5 ), which allows the designer
to explore manually and/or automatically the design space solutions using optimization
heuristics. Exploration is mainly carried out through timing analysis and simulations. The
results of these prediction’s is a real-time behavior of the application functions executed on
various resources, like processors, integrated circuits or communication media. This approach
conforms to the typical hardware/software co-design process. Finally, for the software part of
the application, code is automatically generated as a dedicated real-time executive.

Fig. 5. SynDEx utilization global view

4. Hardware development
Even though the prototyping effort is focused on an FPGA-based design, we are also
exploring the architectural benefits of custom integrated circuitry, primarily related to power
consumption and the silicon area, which are important performance parameters for hardware
designs used in mobile/portable platforms. The approach we have chosen to take involves
identifying the hardware architecture appropriate for low-power configurable design based
on heterogeneous blocks (i.e. blocks that are highly optimized for a particular function,
yet flexible enough to support a variety of configuration parameters) as a compromise for
the trade-off between programmability and power consumption/area. In addition to fast
prototyping, the additional benefits of using modern FPGAs (e.g. Xilinx Virtex 4) are the
availability of highly optimized features implemented as non-standard configurable logic
blocks (CLB) like phase-locked loops, low-voltage differential signal, clock data recovery, lots
of internal routing resources, hardware multipliers for DSP functions, memory, programmable
336
10 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

I/O, and microprocessor cores. These advantages simplify mapping from hierarchical blocks
to FPGA resources.

4.1 Common operator

At the foundation of our study is the ’common operators’ technique to the design of
reconfigurable equipment. Its main principle is the identification and (re)use throughout the
design of common components that can each match several processing contexts, via a simple
parameter adjustment. This technique can greatly increase the efficiency of a multi-standard
software-defined radio, both in terms of its cost, and of the speed of reconfiguration during
operation. The common operator technique belongs to the parameterization techniques
firstly proposed by Jondral et al. (2002). The common operators’ technique is discussed more
extensively in Alaus et al. (2008).
A part of the theoretical approach of this technique is presented in Rodriguez et al. (2007).
Two different parameterization techniques have been proposed in the literature:
• The Common Function (CF) Technique consists in seeking an optimized generic function
(the expected common function) like coding, mapping, among others which can replace
the initial task present in a predefined set of standards. This Common Function (CF)
technique was historically the first one proposed by several articles from Karlsruhe
University (Germany) Jondral et al. (2002), Rhiemeier (2002)
• The Common Operator (CO) technique claims to be independent of the standards by
finding the smallest set of highest-level operators like MAC, FFT, etc., which are used
by the maximum functions number. It is an open technique. The foundation paper was
Palicot & Roland (2003) which identified the FFT as a common operator
The CO technique is implemented in our platform in order to optimize both the area and
the reconfiguration time on the FPGA.These operators are very small regards to the needed
bit-stream, therefore the time to reconfigure a function using an operator is very small
too. These operators are managed by the lowest level of our hierarchical reconfiguration
manager(see section 5). This manager level is very close to the operator itself and in most
cases embedded in the same resource. Furthermore, thanks to the partial reconfiguration
approach, it is very easy to modify either a complete specific operator or parameters of an
operator, providing a huge gain in reconfiguration time.

4.2 Partial Dynamic Reconfiguration

Partial Dynamic Reconfiguration is the capability to reconfigure specific areas of the FPGA
at run-time after its initial configuration. PDR is carried out to allow the FPGA to adapt
to changing hardware algorithms, improve resource utilization, to enhance performance
or to reduce power consumption. In March of 2006, Xilinx introduced the early access
partial reconfiguration flow along with the introduction of slice based bus macros which are
pre-routed intellectual property (IP) cores. The restriction of full column modular PDR was
removed allowing reconfigurable modules of any arbitrary rectangular size to be created.
The EAPR flow also allows signals from the static regions to cross through the partially
reconfigurable regions via the bus macros. Using the principle of glitch-less reconfiguration,
no glitches will occur in signal routes as long as they are implemented identically in every
reconfigurable module for a region. The only limitation of this approach is that all the partial
bit-streams for a module, to be executed on a reconfigurable region, must be predetermined.
Software Defined
Software Defined Radio
Radio Platform Platform
for Cognitive for and
Radio: Design Cognitive Radio: Design and Hierarchical Management
Hierarchical Management 337
11

The Virtex-II and the Virtex-II Pro are the first Xilinx architectures that support Internal
configuration access port (ICAP) Blodget et al. (2003) which is a subset of the Xilinx SelectMAP
interface having fewer signals because it only deals with partial configurations and does not
have to support different configuration modes. For Virtex-II and Virtex-II Pro series, the ICAP
furnishes an 8 bit input data bus and an 8 bit output data bus while with the Virtex-4 Series,
the ICAP interface has been updated with 32 bit input and output data buses to increase its
bandwidth. The ICAP allows the internal logic of the FPGA to reconfigure and to read-back
the configuration memory. With combination of either a hard or a soft microprocessor as a
controller, dynamic reconfiguration is carried out through the ICAP interface Blodget et al.
(2003)
We consider two possible scenarios for dynamically reconfiguring the partial reconfiguration
modules: exo-configuration and endo-configuration as shown in figure 8. The exo-configuration
constitutes the traditional way to configuring an FPGA. A configuration bit-stream is
controlled by an external processor like DSP. In this way, new modules, or upgraded versions
of them, can be created and used at any moment. This approach exhibits upgrade-ability,
but the platform is totally dependent on the processor for modifying its function. The
endo-reconfiguration, also known as self-reconfiguration, considers a different scenario.
An FPGA reconfigures itself using its own local resources. The platform is thus totally
independent as it does not require an external source to provide a bit-stream and to decide
whether to self-reconfigure. The main draw-back is that partial bit-streams need to be
previously generated by a host computer. This approach benefits, thus, of an autonomous
reconfiguration with very limited upgrade-ability.

Fig. 6. Exo and Endo FPGA configuration

Our Platform supports both of the previous cited configuration techniques. In fact, at
boot time, the initial full configuration bit-stream file is sent to the FPGA. This file
includes an internal configuration controller, the internal reconfiguration interface, the initial
instantiations of processing bloc units. Depending on the granularity level. One stays internal
to the FPGA in case of limited-scale reconfiguration (for co-accelerators configuration) or
design parameterization. This implies to interconnect the Micro-blaze to the ICAP internal
configuration interface. In this case, small partial bit-streams can be stored inside the FPGA,
and the use of endo-configuration lets free the other HW resources of the platform. At a
338
12 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

larger scale, configuration for the HW accelerator is external. This implies to interconnect
the DSP to the external SelectMap or internal ICAP reconfiguration interfaces. The bit-stream
corresponding to the design of HW accelerators are stored in an external SRAM memory.

5. Software development
The great diversity of processing types in a multi-standard application implies a large number
of processing configurations to be managed. The configuration management is complex
and we believe that a hierarchical approach of configuration control and management could
simplify it Delahaye et al. (2005). We combine two configuration features, as presented in
section 5.1 and 5.2, in order to create the complete configuration framework for CR testbed.
The proposal of a hierarchical view enables to manage multi-granularity of configurations,
which is of particular interest for heterogeneous architectures. The proposed model is
composed of three levels of hierarchy detailed in figure 9. A system architecture compliant
with this functional model includes one Configuration Manager Unit at level 1 (L1_CMU),
several Configuration Manager Units at level 2 (L2_CMU), each of them being responsible for
one or several Configuration Manager Units of level 3 (L3_CMU), which directly manage the
processing components.

Fig. 7. Hierarchical model of configuration management

5.1 Configuration data-path

As the communication applications are data-flow oriented Delahaye et al. (2005), our
approach is based on a data-path model. The functions of the baseband blocks chain are
mapped into several Processing Block Units (PBU). Each PBU is optimized using specific
reconfigurable hardware resources. In addition, a configuration path, also split into several
configuration manager units, controls the reconfigurable processing path. Each CMU,
dedicated to a type of PBUs, manages the configuration of a type of baseband function in the
chain. The split configuration path offers the possibility to partially reconfigure the baseband
chain by an independent reconfiguration of each PBUs.
Software Defined
Software Defined Radio
Radio Platform Platform
for Cognitive for and
Radio: Design Cognitive Radio: Design and Hierarchical Management
Hierarchical Management 339
13

5.2 Hierarchical management

The hierarchical configuration management model presented in Delahaye et al. (2005) is
based on the configuration data path approach. This model is necessary to manage the
multi-granularity of configuration required by the different contexts. It is composed of three
levels of hierarchy that are detailed below:
• level 1: This first high level classification allows a control of category-specific functions
to manage parameters at the highest level. The L1_CMU works at the standard level as a
host towards the underlying levels of management. This entity is in charge of choosing the
functional units which will constitute the entire configuration of the baseband processing
chain. At this level, generic functions are handled as generic components. Any hardware
implementation is not yet considered.
• level 2: The generic functions selected at level 1 are parameterized at the middle level in
accordance to standard specifications. The set of attributes of each function is handled by
the L2_CMU in order to create each functional context of the entire processing chain.
• level 3: The processing data path architecture at this lowest level depends on the
reconfigurable computing resources of the hardware architecture. The main task of the
L3_CMU in the configuration path is to find the available processing resources and
configure them to enable the execution of the functional context created at the middle level.
As presented in figure 8, an hierarchical configuration management is proposed to map
processing elements. The CPU and the external storage memories are resources used from
a standard PC station. the video coder is implemented in software by a Graphics Processing
Unit (GPU). The L1_CMU is a task running on the CPU which controls the configuration of all
testbed resources (DSP, PFGA). The DSP works as the master of the (DSP/FPGA) subpart of
the platform. The position of master allows the DSP to manage the overall configuration
of the functions that run on the SW/HW resources. The FPGA partial reconfigurability
is, of course, a mandatory feature to allow reconfiguration of a single component. The
L3_CMUs responsible for the configuration of the co-accelerators are implemented as a task
into the Micro-blaze soft processor. It allows to perform fine grain reconfiguration of the
FPGA without involving any external resource. The hardware and software designs of the
processing functions are stored in the external storage memory of the platform where the
configuration management can reach them.

6. Application and results

We have designed and implemented a real-time multi-standard application composed of an
Active Appearance Models (AAM) schema Cootes et al. (2009) and a digital communication
layer (802.11g, UMTS or GSM). The coding application feeds a video coded bit-stream in
the transmitter whereas the associated video decoder is connected to the receiver (802.11g,
UMTS or GSM). In this section, we give some results of our development approach with the
Sundance platform.

6.1 The overall scenario

The platform illustrates the adaptation of the radio link according to the compression of the
source in a video-telephony context. A person switches-on his terminal in order to perform
an audio-video conversation with another person. At the beginning of the communication,
340
14 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

Fig. 8. example of the hierarchical configuration management mapping

the face of the speaker and the background of the image are transmitted using a traditional
compression mode. This requires a relatively high data-rate over the time, a model of the
person’s face is generated at the transmitter’s side, and sent to the receiver. Once this model
is understood by the receiver, the transmitted parameters of the face’s model (orientation,
opening of the mouth, of the eyes, direction of the glance) are enough to reproduce the face
behavior at the receiver. This permits to save the data amount required to transmit the face of
the speaker, by reducing very significantly the data to be transmitted through the air. The data
rate variations by step as well as the dynamic reconfigurations of the radio link are illustrated
in figure 9.

Fig. 9. Standard adaptation as function of the video compression

Software Defined
Software Defined Radio
Radio Platform Platform
for Cognitive for and
Radio: Design Cognitive Radio: Design and Hierarchical Management
Hierarchical Management 341
15

At the start, the person switches-on his terminal and starts a video-conference service. Video
coder starts learning the face model, as well as models for eyes and mouth reconstruction.
Then the radio link goes through the following steps.
Step 1: The image is transmitted using a traditional compression mode. The terminal learns
the 3D model of the person face and performs a 802.11g modulation with standard error
coding.
Step 2: The face model is learned: only high level parameters of the face are transmitted
(location, size, orientation) so the receiver can reconstruct the 3D model of the face with its
texture on the already sent background. In order to improve the reconstruction at the receiver,
errors between the model and the real image are also transmitted by the means of an UMTS
modulation with standard error coding.
Step 3: The mouth variations are modeled: The mouth characteristics, as well as high level
parameters of the face model are transmitted through UMTS with a very robust error coding
on the data for the mouth model.
Step 4: In this last step all face features’ models were already learn only high level parameters
of all three face, mouth, and eyes models are transmitted, as well as the errors with respect
to the real image to help the reconstruction process. GSM modulation with standard error
coding can then be used.
The last step is the longer period of the video-call, which permits to reduce very efficiently the
global mean throughput necessary for the communication. This justifies the efforts accepted
at the beginning of the call in terms of adaptation complexity. Changing from one data rate
to another is possible, while permanently reconfiguring the air link characteristics to up a
significant degree.

6.2 Experimental results

The matching step of SynDEx consists in performing mapping and scheduling of the
algorithm’s operations and data transfers onto the architecture processing components and
communication media. It is carried out by a heuristic which takes into account durations
of computations and inter-component communications to optimize the global application
latency.
• Algorithm graph: Application algorithm is represented by a data-flow graph (DFG) to
exhibit the potential parallelism between operations. The algorithm model is a direct
data dependence graph. An operation is executed as soon as its inputs are available, and
this DFG is infinitely repeated. SynDEx includes a hierarchical algorithm representation,
conditional statements and iterations of algorithm parts. The application can be described
in a hierarchical way by the algorithm graph. The lowest hierarchical level is always
composed of indivisible operations. Operations are composed of several input and
output ports. Special inputs are used to create conditional statements. Hence an
alternative sub-graph is selected for execution according to the conditional entry value.
Data dependencies between operations are represented by valued arcs. Each input and
output port has to be defined with its length and data type. These lengths are used to
express either the total required data amount needed by the operation before starting its
computation or the total amount of data generated by the operation on each output port.
• Architecture graph: The architecture is also modeled by a graph, which is a directed graph
where the vertices are computation operators (e.g processors, DSP, FPGA) or media (e.g
342
16 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

SHB, SDB, PCI, Ethernet) and the edges are connections between them. So the architecture
structure exhibits the actual parallelism between operators. Computation vertices have no
internal computation parallelism available. An example is shown in figure 10. In order to
perform the graph matching process, computation vertices have to be characterized with
algorithm operation execution times. Execution times are determined during the profiling
process of the operation. The media are also characterized with the time needed to transmit
a given data type.

Fig. 10. Syndex model

The output files generated by SynDEx are exploited by our platform to manage correctly
the hardware reconfiguration platform. These text files are managed by the L1_CMU of the
transmitter (the platform in charge to send video stream) and by L1_CMU of the receiver to be
standard compatible with the current or future transmission. In the next section , we present
the hardware platform used and its specifications.
This hardware architecture is represented by an architecture graph under SynDEx and in
the same manner a DFG is done for the application task (telecommunication chain to be
used). Then, the heuristic of SynDEx realizes the adequation between the graph and the
hardware and generates constraint files for the DSP and the FPGA that give information for
the reconfiguration of the platform. Then the host of the platform sends a new source-code
to the DSP and a new order of partial reconfiguration to the FPGA using the architecture
described above.

7. Conclusion
In this chapter, we introduce a heterogeneous reconfigurable Sundance platform to support
Cognitive Radio in the context of emergency networks. The heterogeneous reconfigurable
architecture includes heterogeneous processing elements such as general purpose processors
(GPU), DSPs and FPGAs. A key element in this heterogeneous reconfigurable architecture
is the run-time partial reconfiguration of the hardware part, which can achieve the
Software Defined
Software Defined Radio
Radio Platform Platform
for Cognitive for and
Radio: Design Cognitive Radio: Design and Hierarchical Management
Hierarchical Management 343
17

reconfigurability in combination with the energy efficiency. A design methodology is needed

to map applications onto a heterogeneous platform which has two new features: transaction
level modeling of applications and run-time spatial mapping. In the future, we aim to
validate the HDCRAM approach, test our PAPR and spectrum sensing algorithms in the
case of MIMO system, and sets of algorithms for cognitive radio. The ultimate goal is
to build a heterogeneous reconfigurable radio platform to demonstrate the cognitive radio
functionalities.

8. References
J. Mitola, ”Cognitive Radio: An Integrated Agent Architecture for Software Defined Radio,” Ph.D.
dissertation, Royal Inst. of Tech., Sweden, May 2000.
J. Mitola, G. Maguire, ”Cognitive radio: making software radios more personal, Personal
Communications,” IEEE Wireless Communications, Vol. 6, No. 4. (1999), pp. 13-18
A. Nafkha, R. Seguier, J. Palicot, C. Moy, J.P. Delahaye, "A Reconfigurable BaseBand
Transmitter for Adaptive Image Coding", IST Mobile and Wireless Communications
Summit, 1-5 July 2007,
G. Chen, D. Kotz, "A Survey of Context-Aware Mobile Computing Research", Technical Report
TR2000-381, Dept. of Computer Science, Dartmouth College, Nov. 2000
E. Aarts, H. Harwig, and M. Schuurmans, "Ambient Intelligence: The Invisible Future", J.
Denning, ed., McGraw Hill, New York, 2001.
X. Qiu, J. Chuang, "Link adaptation in wireless data networks for throughput maximization
under retransmissions", in Proceedings of International Conference on Communications
(ICC), Vancouver, 1999.
W. Fan, M. Krunz, C. Shuguang; "Price-Based Spectrum Management in Cognitive Radio
Networks", IEEE Journal of Selected Topics in Signal Processing, vol.2, 2008.
M. Ghozzi, M. Dohler, F. Marx, J. Palicot, "Cognitive radio: methods for the detection of
free bands, Towards reconfigurable and cognitive communications", Comptes rendus
Physique, Paris, vol. 7, no7, 2006.
F. Jondral, "Software Defined Radio Enabling Technologies" edited by W.Tuttlebee, Wiley, 2002.
L.Alaus, J. Palicot, C. Roland, Y. Louet, D.Noguet, "Promising technique of parametrization
for reconfigurable radio, the Common Operators Technique : fundamentals and
examples", Signal Processing For Software Defined Radio Handsets, Springer, 2008.
A-R. Rhiemeier, "Benefits and Limits of Parameterized Channel Coding for Software Radio",
2nd Karlsruhe Workshop on Software Radios, Germany, 2002.
J. Palicot, C. Roland, "FFT: a Basic Function for a Reconfigurable Receiver", International
Conference on Telecommunications, Papeete, Tahiti, 2003.
V. Rodriguez, C. Moy, J. Palicot, "Install or invoke?: The optimal trade-off between
performance and cost in the design of multi-standard reconfigurable radios," Wiley
Inter-science, Wireless Communications and Mobile Computing Journal, Volume 7 Issue
9, Pages 1143 - 1156, 2007.
J.P. Delahaye, J. Palicot, P. Leray, "A Hierarchical Modeling Approach in Software Defined
Radio System Design", IEEE Workshop on Signal Processing Systems, Athens (Greece),
Nov. 2005.
344
18 Recent Advances in Wireless CommunicationsWill-be-set-by-IN-TECH
and Networks

B. Blodget and S. McMillan and P. Lysaght, "A lightweight approach for embedded
reconfiguration of FPGAs," in Design, Automation and Test in Europe Conference and
Exhibition, 2003.
T. Grandpierre, C. Lavarenne, and Y. Sorel, "Optimized Rapid Prototyping for Real-Time
Embedded Heterogeneous Multiprocessors",in CODES, Rome, Italy, May 1999.
J.P. Delahaye, C. Moy, P. Leray, J. Palicot, "Managing Dynamic Partial Reconfiguration on
Heterogeneous SDR Platforms", Sdr Forum, November, Los Angeles, USA, 2005.
L. GODARD, C. MOY, J. PALICOT, ”An Executable Meta-Model of a Hierarchical
and Distributed Architecture Management for the Design of Cognitive Radio
Equipments,” Annals of Telecommunications, Special issue on Cognitive Radio,
Volume 64, Numbers 7-8, 2009.
T.F.Cootes, G.J.Edwards and C.J.Taylor, ”Active Appearence Models”, Europeen Conference
on Computer Vision,1998.
 16

Dealing with VoIP Calls

During “Busy Hour” in LTE
Angelos Antonopoulos1, Elli Kartsakli2,
Luis Alonso2 and Christos Verikoukis1
1Telecommunications Technological Centre of Catalonia (CTTC)
2Department of Signal Theory and Communications (TSC),
Technical University of Catalunya (UPC)
Spain

1. Introduction
Long Term Evolution (LTE) is an evolving wireless standard developed by the 3rd
Generation Partnership Project (3GPP) which, along with 3GPP HSPA+, 3GPP EDGE
Evolution and Mobile WiMAX (IEEE 802.16e), opens the road to 4G technologies. The
standard is focused on delivering high data rates for bandwidth-demanding applications
and on improving flexibility and spectral efficiency, thus constituting an attractive solution
for both end users and mobile operators. An important feature of LTE that differentiates it
from conventional mobile standards is the all-IP packet based network architecture, which
further ensures the seamless integration of internet applications and facilitates the
convergence between fixed and mobile systems.
The radio interface of LTE is based on Orthogonal Frequency Division Multiplexing
(OFDM) and supports Multiple-Input-Multiple-Output (MIMO) technology. The standard
defines asymmetrical data rates and modulations for uplink and downlink, using different
access schemes for each link. In particular, Orthogonal Frequency Division Multiple Access
(OFDMA) is employed in the downlink, while the technically similar but less power-
demanding Single Carrier – Frequency Division Multiple Access (SC-FDMA) is used in the
uplink. In terms of the wireless spectrum allocation, LTE supports variable channel
bandwidths that vary from 1.4 to 20 MHz and can be deployed in different frequency bands.
The LTE architecture, referred to as Evolved Packet System (EPS) comprises the Evolved
Radio Access Network (E-UTRAN) and the Evolved Packet Core (EPC), illustrated in Fig. 1
(3GPP, 2010). The E-UTRAN consists of a network of enhanced base stations called evolved
Nodes B (eNBs) whose main role is to manage the radio resource and mobility in the cell in
order to optimize the communication among all User Equipments (UEs). The eNBs can
communicate with each other through the X2 interface and can access the EPC by means
of the S1 interface. On the other hand, the EPC consists of a control plane node called
the Mobility Management Entity (MME) and two user plane nodes, the Serving Gateway
and the Packet Data Network Gateway (PDN Gateway or P-GW). These control planes
handle the data packet routing within the LTE and towards non-3GPP data networks,
respectively.
346 Recent Advances in Wireless Communications and Networks

Fig. 1. LTE Architecture and functional split between E-UTRAN and EPC
LTE provides service differentiation by adopting a class-based Quality of Service (QoS)
concept. In particular, each data flow between the user equipment and the P-GW (called EPS
bearer) is assigned a QoS profile. A total of nine profiles are defined in the specification
(3GPP, 2011) that can be mapped to different types of applications such as real time video
and voice services, online gaming, etc. Each profile involves the bearer type, the flow
priority, an upper bound for the packet delay and the packet error rate. The bearer type
indicates whether a Guaranteed Bit Rate (GBR) will be provided to the bearer by
permanently allocating network resources during the data session.
The essential difference between GBR and non-GBR bearers is that, in the first case, a
connection may be blocked if the network does not have the resources to guarantee the
desired QoS of these connections. This concept is known as Call Admission Control (CAC)
and it is an important component of radio resource management. CAC algorithms are
usually implemented in eNBs and their role is to determine whether a new connection
request should be accepted or rejected, depending on the available network resources.
This chapter is focused on CAC policies for handling the admission of Voice over IP (VoIP)
calls in an LTE system. Particular interest is laid on the “busy hour” phenomenon, which is
defined as the “uninterrupted period of 60 minutes during the day when the traffic offered
is the maximum” (Weber, 1968). It is during these intervals of increased traffic that CAC
mechanisms play a significant role in the system performance and stability. However, most
works in the literature either implement fixed bandwidth reservation schemes or give
priority to real time services once they have been admitted to the system. In this chapter, the
authors present two different approaches that take into consideration the “busy hour”
phenomenon, namely, a dynamic bandwidth reservation scheme and a dynamic CAC
mechanism that adapts to the incoming traffic load.
Dealing with VoIP Calls During “Busy Hour” in LTE 347

The chapter is organized as follows. Section 2 outlines the related work on admission
control schemes for 4G networks found in the literature. The two proposed CAC algorithms
are described in Section 3 and their performance is discussed in Section 4. Finally, Section 5
is devoted to conclusions.

2. Related work
Although the admission control concept has been extensively studied in the literature, only
a limited number of contributions are developed within the context of 4G networks. The
objective of this section is to highlight the recent work on this field in order to provide the
reader with an up to date State of the Art.
One of the first attempts towards introducing admission control in the fourth-generation
cellular mobile networks has been made by Jeong et al. (Jeong et al., 2005). In their work, the
authors present a CAC scheme that supports the QoS requirements of the accepted
connections in IEEE 802.16e wireless systems. The objective of the proposed CAC is to
maximize the utilization of the resources, considering as basic parameter the capacity
estimation of the cell. Furthermore, during the admission and scheduling process, the base
station distinguishes the delay-sensitive real-time (RT) from the delay-tolerant non-real-time
(NRT) connections. The proposed scheme achieves to fulfil the QoS demands of the
connections, but in temporary overloaded situations, only NRT class connections can be
admitted, thus excluding entirely the RT traffic.
Qian et al. (Qian et al., 2009) propose a novel radio admission control scheme for multiclass
services in LTE systems. The authors introduce an objective function to maximize the
number of admitted users and propose a CAC algorithm that implements a service
degradation scheme whenever a limitation of resources occurs. In their paper, there is a
service differentiation approach, with different portions of bandwidth devoted to each
traffic class. However, in the presented numerical results there is no plot that distinguishes
the blocking probabilities for the different types of traffic, thus not providing any
information about the actual handling of the multiclass services.
Anas et al. propose an admission control algorithm for LTE utilizing the fractional power
control (FPC) formula agreed in 3GPP (Anas et al., 2008). In their work, GBR is the only
considered QoS of the bearer, while each user is assumed to have a single-bearer. The main
idea of their proposed algorithm is that the current resource allocation can be modified in
order for the new user to be admitted without violating the power restriction for the
physical uplink shared channel (PUSCH).
Lei et al. (Lei et al., 2008) introduce a resource allocation algorithm along with a connection
access control scheme for LTE systems with heterogeneous services. Their proposed CAC
assigns different portions of bandwidth for real-time and non-real-time connections, thus
balancing the ongoing connections of different traffic classes and facilitating the support to
potential handoff users. However, the results show that the cell throughput remains the
same whether the proposed admission control scheme is applied or not.
In (Kwan et al., 2010) a novel predictive admission control scheme is presented. The authors
propose a new cell load measurement method and mechanisms for predicting the load
increase due to the acceptance of new connections. In the same content, a resource-estimated
CAC algorithm is proposed in (Bae et al., 2009). Specifically, whenever a service request
occurs, the resource-estimated CAC algorithm calculates the required amount of resources
in order for the request to be served. This amount is determined based on the service type,
348 Recent Advances in Wireless Communications and Networks

the modulation and the coding scheme level of the particular user. However, the results
show that the proposed CAC is beneficial only in terms of packet delay, since the average
data rate and the cell utilization are decreased.
Regarding the bandwidth reservation, a downlink CAC algorithm with look-ahead calls for
3GPP LTE mobile networks is presented in (Sallabi and Shuaib, 2009). The proposed
algorithm handles the advance resource reservations, providing a high probability that the
advance calls will be immediately served once their session is ready to start. Nevertheless, it
is hard to derive useful conclusions since there is no reference or comparison to other
admission control methods.
Finally, there are various works on the scheduling phase of LTE that grant priority to the
VoIP traffic service (Choi et al., 2007; Puttonen et al., 2008; Saha and Quazi, 2009). The main
idea behind these contributions is that the prioritization of the voice packets takes place
once the connection has been admitted in the system. This results in higher satisfaction of
the VoIP users since the QoS of the voice traffic remains in high levels.

3. Admission control schemes

As mentioned in the introduction, LTE defines a class-based QoS concept, thus providing a
simple but effective solution to operators in order to offer differentiation between packet
services. Furthermore, recent studies have shown that the proportion of VoIP users show a
continuous growth from 28% of users in 2008 (up from 20% of users in 2007) to more than
50% in 2010 (Report Study, 2009). Due to this fact, the proposed schemes focus on voice
flows, giving them higher priority comparing to the other types of traffic of the standard.
The problem becomes more intense if we take under consideration the variation of daily
traffic volume, where there is a peak during the “busy hour”. In Fig. 2 the mean number of
calls per minute to a switching centre taken as an average for periods of 15 minutes during
10 working days (Monday-Friday) is depicted (Iversen, 2010).

Fig. 2. Typical 24-hour traffic variation

In this section, two CAC algorithms to handle the admission of VoIP calls are presented. The
target of the both schemes is to provide enhanced Grade of Service (GoS) to voice traffic
flows by improving the acceptance rate of the VoIP calls. Grade of Service is defined as the
Dealing with VoIP Calls During “Busy Hour” in LTE 349

probability of a call being blocked (BP) or delayed more than a specified interval. From a
practical aspect it could be also defined as the probability of a user receiving a network busy
signal in a telephone service and can be measured using the following equation:

Number _ of _ lost _ calls

GoS = BP = (1)
Number _ of _ offered _ calls

3.1 Bandwidth reservation-based CAC mechanism (BR CAC)

Our first proposed admission control mechanism is based on the bandwidth reservation
concept and is executed under “busy hour” conditions. Under these conditions (i.e. for high
arrival rate of VoIP calls), once a connection request arrives at the system, it is mapped onto
the corresponding service class. Three main service classes are considered in our scheme:
i) the voice GBR ii) the non-voice GBR and iii) the non-GBR traffic types. The two first
classes are included in the GBR family, while the third includes the connections that do not
require any Guaranteed Bit Rate. In case of voice connections, the request is accepted if the
total available bandwidth (BWT) suffices to serve the incoming connection. On the other
hand, restricted bandwidth (BWT - BWR) is provided to the other GBR classes, as the
algorithm’s aim is to prioritize VoIP calls over other types of connections. In order to deal
with the connections that do not require any QoS guarantees (non-GBR), the requests are
always admitted, but no bandwidth allocation is considered. The portion of the reserved
bandwidth for voice traffic is dynamically changed according to the traffic intensity of the
VoIP calls:

BWR = ⎣⎢ ρ 1 × β ⎦⎥ × BW1 (2)

In the above expression, the traffic intensity ρ1 is a measure of the average occupancy of the
base station during a specified period of time. It is denoted as ρ = λ1 / μ1, where λ1 is the
mean arrival time for VoIP connections and μ1 represents their mean service rate (duration).
Furthermore, BW1 is the bandwidth needed for each VoIP call, while β ∈ [ 0,1] denotes the
bandwidth reservation factor.
Formula (2) implies that traffic intensity has an impact on the blocking probabilities of both
voice and non-voice connections. It makes sense that applying this bandwidth reservation
scheme, the blocking probability for the VoIP connections is decreased, since a portion of
bandwidth is exclusively dedicated to this service type. On the contrary, the available
bandwidth for the connections of the other service types is decreased and consequently the
blocking probability for the specific types increases.
In bandwidth reservation schemes, one of the main difficulties is to avoid the inefficient
utilization of system resources. However, in our case, the daily traffic variation establishes
the ability to predict an increase in VoIP calls, thus enabling us to tackle this problem.
Therefore, our scheme outperforms classic bandwidth reservation mechanisms.

3.1.1 Analytical model

In this section an analytical model for the proposed bandwidth reservation scheme is
developed, to derive the blocking probabilities for the different class types. The results are
further verified by extensive simulations, presented in the following section.
In order to simplify the analysis, the non-voice connections (e.g. video, data etc) are treated
as a single class type with the same characteristics (i.e. arrival rate, bandwidth demand). In
350 Recent Advances in Wireless Communications and Networks

this point we must clarify that this simplification takes place only in the admission control
process since, after being accepted, the connections are treated according to their different
priorities. Furthermore, non-GBR connections are not included in the model as they are
always accepted without any QoS guarantees.

λ1 λ1 λ1
0,n 1,n 2,n … ...
μ1 2 ⋅ μ1 3 ⋅ μ1

λ2 n ⋅ μ2 λ2 n ⋅ μ2 λ2 n ⋅ μ2
… ...

… ...

… ...

λ2 3 ⋅ μ2 λ2 3 ⋅ μ2 λ2 3 ⋅ μ2
λ1 λ1 λ1
0,2 1,2 2,2 … ...
μ1 2 ⋅ μ1 3 ⋅ μ1

λ2 2 ⋅ μ2 λ2 2 ⋅ μ2 λ2 2 ⋅ μ2 2 ⋅ μ2
λ1 λ1 λ1 λ1 λ1
0,1 1,1 2,1 … ... k,1 … ...
μ1 2 ⋅ μ1 3 ⋅ μ1 k ⋅ μ1 (k + 1) ⋅ μ1

λ2 μ2 λ2 μ2 λ2 μ2 μ2
λ1 λ1 λ1 λ1 λ1 λ1
… ... k,0 … ... m,0
0,0 1,0 2,0

μ1 2 ⋅ μ1 3 ⋅ μ1 k ⋅ μ1 (k + 1) ⋅ μ1 m ⋅ μ1

r,s+1

λ2 (s + 1)⋅ μ2
λ1 λ1

r-1,s r,s r+1,s

r ⋅ μ1 (r + 1)⋅ μ1

λ2 s ⋅ μ2

r,s-1

Fig. 3. The two-dimensional Markov model's state transition diagram

Thus, the 2-dimensional continuous Markov model (Fig. 3) can be used to analyze the
performance of the proposed scheme. The state space of this Markov model is

S = {( r , s ) 0 ≤ r ≤ m ,0 ≤ s ≤ n , r ⋅ BW1 + s ⋅ BW2 ≤ BWT } , (3)

Dealing with VoIP Calls During “Busy Hour” in LTE 351

⎢ BWT ⎥ ⎢ BWT − BWR ⎥

where m = ⎢ ⎥ and n = ⎢ ⎥ . The number of VoIP and non-VoIP connections
⎣ BW1 ⎦ ⎣ BW2 ⎦
is represented by r and s, respectively. Additionally, BWT and BWR represent the overall
and the reserved bandwidth, while BW1 and BW2 represent the bandwidth that is needed
in order to serve each VoIP and non-VoIP connection, respectively. We also define other
parameters as follows:
λ1 Arrival rate of VoIP connections
λ2 Arrival rate of non-VoIP connections
1 μ1 Service time for VoIP connections
1 μ2 Service time for non-VoIP connections
The state transmission diagram of the Markov model is shown in Fig. 3.
Its steady state equation is the following:

pr , s ⋅ ( λ1 ⋅ ϕr + 1, s + λ2 ⋅ ϕr , s + 1 ⋅ θ r , s + 1 + r ⋅ μ1 ⋅ ϕr − 1, s + s ⋅ μ 2 ⋅ ϕr , s − 1 ) =
(4)
= λ1 ⋅ pr − 1, s ⋅ ϕr − 1, s + λ2 ⋅ pr , s − 1 ⋅ ϕr , s − 1 ⋅ θ r , s + ( r + 1 ) ⋅ μ1 ⋅ pr + 1, s ⋅ ϕr + 1, s + ( s + 1 ) ⋅ μ2 ⋅ pr , s + 1 ⋅ ϕr , s + 1

where pr , s denotes the steady state probability of the system lying in the state ( r , s ) and
φr , s , θ r , s denote characteristic functions:

⎧1, (r , s ) ∈ S
ϕr , s = ⎨ (5)
⎩0, otherwise

⎧1, r ⋅ BW1 + s ⋅ BW2 ≤ BWT − BWR

θr ,s = ⎨ (6)
⎩0, otherwise

The above functions are used in order to prevent a transition into an invalid state, according
to the previously defined restrictions. Furthermore, considering the normalization condition
∑ pr ,s = 1 , the steady state probability for each possible state can be obtained.
( r , s )∈S

The blocking probabilities for VoIP and non-VoIP connections are given by:

BPVoIP = ∑ pr , s (7)
( r + 1) ⋅ BW1 + s ⋅ BW2 > BWT

BPnon −VoIP = ∑
r ⋅ BW1 + ( s + 1) ⋅ BW2 > BWT − BWR
pr , s (8)

3.1.2 Operational example

In order to clarify the mathematical analysis above, we provide two possible states of the
system’s Markov Chain. Fig. 4 depicts the exact form of the chain in each of the two cases.
The first represents the state where there is no available bandwidth for non-voice
connections, hence not permitting the transition from s to s+1. On the other hand, the
second represents an equivalent situation along with the assumption that only voice
connections are served in the system (s=0), thus not allowing the transition from s to s-1
and vice versa.
352 Recent Advances in Wireless Communications and Networks

r,s+1

(s + 1)⋅ μ2
λ1 λ1

r-1,s r,s r+1,s r,s+1

r ⋅ μ1 (r + 1)⋅ μ1

λ2 s ⋅ μ2
(s + 1)⋅ μ2
λ1 λ1

r,s-1 r-1,s r,s r+1,s

r ⋅ μ1 (r + 1)⋅ μ1
Fig. 4. Two examples of possible states of the system
First case: We assume that the system lies in the state (r, s), subject to the following
constraints:

{( r , s ) , ( r + 1, s ) , ( r , s + 1) , ( r − 1, s ) , ( r , s − 1)} ∈ S (9)

r ⋅ BW1 + ( s + 1 ) ⋅ BW2 > BWT − BWR (10)

r ⋅ BW1 + s ⋅ BW2 < BWT − BWR (11)

Under these assumptions and using the definitions of φr , s and θ r , s , we derive the steady
state equation for the specific case:

pr , s ⋅ ( λ1 + r ⋅ μ1 + s ⋅ μ 2 ) = λ1 ⋅ pr − 1, s + λ2 ⋅ pr , s − 1 + ( r + 1 ) ⋅ μ1 ⋅ pr + 1, s + ( s + 1 ) ⋅ μ 2 ⋅ pr , s + 1 (12)

Second case: In this case we assume that the system lies in the state (r,s), subject to the
following constraints:

{( r , s ) , ( r + 1, s ) , ( r , s + 1) , ( r − 1, s )} ∈ S (13)

( r , s − 1) ∉ S (14)

r ⋅ BW1 + ( s + 1 ) ⋅ BW2 > BWT − BWR (15)

r ⋅ BW1 + s ⋅ BW2 = BWT − BWR (16)

Considering again the definitions of φr , s and θ r , s , we derive the respective steady state
equation for this case, that is:

pr , s ⋅ ( λ1 + r ⋅ μ1 ) = λ1 ⋅ pr − 1, s + ( r + 1 ) ⋅ μ1 ⋅ pr + 1, s + ( s + 1 ) ⋅ μ2 ⋅ pr , s + 1 (17)
Dealing with VoIP Calls During “Busy Hour” in LTE 353

3.2 Dynamic call admission control algorithm (DCAC)

In the same context, we propose a second CAC algorithm that gives priority to the VoIP
calls during the “busy hour”. In this scheme, unlike the previous one, no bandwidth
reservation takes place, while there is an effort towards a fairer handling of all connections.
According to this CAC scheme, the eNB accepts all the VoIP flows if the available
bandwidth suffices in order for the calls to be served. In the case of non-VoIP flows there is
an outage probability that depends both on the arrival rate of VoIP requests as well as on
the available bandwidth. The requests of non-GBR connections are always admitted, but no
bandwidth allocation is considered, since non-GBR flows do not need any QoS guarantees.
The proposed algorithm has two main parameters: the arrival rate of VoIP requests and the
available bandwidth of the system. The outage probability for the non-VoIP connections
increases either when the arrival rate of the VoIP calls grows or when the available
bandwidth decreases. The capacity required in order to serve all the upstream connections
can be approximated with the following expression:

Cneed = ∑ ρ × BW
i =1,2
i i
(18)

All the parameters in the above expression have been already defined. However, it should
be stressed that the index i corresponds to different service types and can take values 1 and
2 for VoIP and non-VoIP traffic, respectively.
In case that the system bandwidth suffices to serve the flows of all service types, the outage
probability is equal to zero. Due to this fact, the proposed admission control has the same
output as classic admission control schemes under light traffic conditions in the network.
On the contrary, in overloaded environments where the bandwidth is not sufficient for all
connections, an admission control algorithm is required in order to provide different levels
of priority to the various connections.
Let us consider the arrival rate of the VoIP requests, defined as λ1. If this rate is higher than a
specific threshold there will be an outage probability for the requests of the other GBR
service types. This threshold is defined by the administrator/operator of the network, by
considering the network parameters, e.g. the arrival rate of VoIP calls during “busy hour”.
The value of the outage probability fluctuates between Poutmin and Poutmax, depending on
the available system bandwidth. In the extreme case that we have no available bandwidth,
the overall outage probability becomes Poutmax. Adversely, when the total bandwidth of the
system is available and no connections are being served, i.e., BWavailable/BWT = 1, the outage
probability becomes Poutmin, since there is enough bandwidth in order for the connections of
all types to be served. These borderline values are selected by the system’s operator
according to each traffic class’ desired level of priority. On the other hand, whenever the
arrival rate of VoIP connections is smaller than this arrival rate threshold, we assume that
we are out of “busy hour” and, therefore, the outage probability equals zero.
The flowchart in Fig. 5 depicts the connection acceptance/rejection procedure in the
proposed Dynamic Connection Admission Control (DCAC) algorithm. The basic process of
the connection request flow has been described above. In the last part of the algorithm, there
is an estimation of the available bandwidth ratio in order to derive the exact value of the
outage probability (the higher the ratio, the lower the probability). In particular, the Poutmin
is a system parameter, designated by the operator, which determines the desirable level of
priority to be assigned to the voice calls. By adding this value to the normalized bandwidth
ratio, the outage probability for the specific connection is derived.
354 Recent Advances in Wireless Communications and Networks

Admission Request

Arrival Rate No Accept

>Threshold Poutage=0
Connection
?

Yes

Total BW > Yes Accept

Poutage=0
Cneed Connection
?
No

Estimate Ratio of
Poutage =
Available Bandwidth Accept/Reject
Pout min +
(Available BW / Connection
Total BW) Normalized Ratio

Fig. 5. Dynamic connection admission control (flowchart)

4. Performance evaluation
In order to evaluate the performance of the proposed CAC schemes and verify the validity
of the analytical formulation, corresponding event-driven C++ simulators that execute the
rules of the algorithms have been developed. In this section, the simulation set up is
described, followed by a discussion of the obtained results.

4.1 Simulation scenario

Based on the physical capabilities of the LTE technology, we assume that the overall
bandwidth for the uplink traffic is 4 Mb/s. Assuming that the non-VoIP traffic consists
mainly of audio and video data, an average bandwidth of 128 kb/s for each connection is
considered (Koenen, 2000). The codec chosen to generate VoIP traffic is the G.711, resulting
to a constant bit rate of 64 kb/s. Each result was produced by running the simulation 100
times using different seeds, while we simulate 3600 seconds of real time in order to be in
accordance with the definition of “busy hour”.
In order to evaluate the efficiency of the proposed algorithms, a research on the state-of-the-
art admission control mechanisms for the LTE standard has been conducted. Several
schemes in the literature accept a new connection when the following condition is satisfied:

C reserved + TRiservice ≤ Ctotal (19)

where Creserved represents the capacity reserved by the already admitted connections in the
system, TRiservice denotes the traffic rate that should be guaranteed to the new connection i
of service type service and Ctotal is the total available capacity.
We refer to these methods as capacity-based (CB) algorithms in order to distinguish from
our proposed algorithms which are either based on the bandwidth reservation (BR) concept
Dealing with VoIP Calls During “Busy Hour” in LTE 355

or follow a dynamic approach (DCAC). In order to study the performance of our

mechanisms we have carried out simulation tests by varying the VoIP requests arrival rate,
thus providing a large range of voice traffic that fluctuates between 15 and 240
connections/min. However, it should be clarified that the rate request of the voice
connections remains constant during the busy hour. The system parameters that are
presented in Table 1, define that the arrival rate of all connections follows a Poisson
distribution, while the mean service time for the connections is exponentially distributed.

Parameter Value
Bandwidth 4 Mb/s
λ2 Poisson (1 connection/s)
1/ μ1 Exponential (mean 50 s)
1/ μ2 Exponential (mean 50 s)
BW1 64 kb/s (G.711)
BW2 128 kb/s
Threshold 0.2 calls/s
β (BR) 1/3
Poutmin (DCAC) 0.6
Poutmax (DCAC) 0.85
Table 1. System parameters
Under these assumptions and considering λ1 = 1 connection/s, the system can serve about
98% of the VoIP calls if all the requests of the other classes are rejected, which means that the
network is overloaded. Furthermore, in the specific case we use a single admission control
based on bandwidth availability (CB) where all the requests are accepted if there is enough
bandwidth to serve them, regardless of the class that they belong to, the system serves about
57% of the VoIP flows and 34% of the other flows.
Finally, before proceeding to the simulation results, let us recall that the aim of the proposed
schemes is to serve more voice traffic by reducing the GoS, and consequently the blocking
probability, of VoIP calls.

4.2 Performance results

Simulation results are compared to those obtained with the mathematical model presented
in section 3.1.1. First, it can be observed that the simulation results verify the mathematical
analysis, with the difference varying in a range of less than 2% (Fig. 6). Comparing the first
proposed admission control to traditional schemes for different values of arrival rates for the
VoIP connections, we observe that the BR CAC outperforms single admission control
methods in terms of GoS, without any deterioration in the overall system performance. Fig.
6 depicts the GoS among various arrival rates of VoIP calls. It is observed that, using our
proposed CAC, a better system performance in terms of voice communication is achieved,
as there is a significant enhancement in GoS (10-40%) of VoIP traffic.
On the other hand, the GoS of the other types of connections is increased as expected, but
examining the system considering the total number of connection requests (both VoIP and
non-VoIP) we achieve a more efficient utilization of system resources as we observe an
enhancement in the total GoS ratio for high arrival rates of VoIP connections (i.e. rates
greater than 1 connection/s).
356 Recent Advances in Wireless Communications and Networks

Fig. 6. GoS vs. VoIP Calls Arrival Rate (proposed Bandwidth Reservation (BR) CAC vs.
Capacity-based (CB) CAC including analytical results)

Fig. 7. GoS vs. VoIP Calls Arrival Rate (proposed DCAC vs. Capacity-based (CB) CAC)
Dealing with VoIP Calls During “Busy Hour” in LTE 357

The simulation results of the proposed Dynamic Call Admission Control (DCAC) algorithm
comparing to the Capacity-based (CB) algorithm are presented in Fig. 7. This algorithm not
only improves the voice traffic service, but also enhances the overall system performance.
However, in this case the level of prioritization of the VoIP calls over the other type of traffic
is lower compared to the bandwidth reservation scenario, thus resulting in a fairer
distribution of the system resources.
Furthermore, it is interesting to observe that even for the lower arrival rates of VoIP calls
(i.e. 0.25 and 0.5 calls/s) the DCAC handles efficiently the system’s bandwidth, due to its
flexibility, while the BR scheme fails to overcome the Capacity-based algorithm. The
comparison between the two proposed schemes is given in Fig. 8. In this figure, even if there
is no further information provided, it can be clearly seen how the two proposed schemes
deal with the different types of traffic, as well as their overall performance. An interesting
observation is that, in this particular scenario, the curves for the total GoS for the two
schemes cross when the arrival rate is approximately 1.3 connections/s. Below this
threshold (i.e. for relatively low traffic conditions) the DCAC outperforms the proposed BR
scheme, while above this threshold (i.e. for relatively high traffic conditions) the BR scheme
handles the total connections in a more efficient way.
The system’s bandwidth is a main parameter of the DCAC. In Fig. 9 the provided Grade of
Service for various values of bandwidth is plotted. As far as networks with restricted
bandwidth capabilities are considered, we observe that our proposed dynamic admission
control algorithm outperforms single methods, as it improves the GoS of both VoIP calls (11-
27%) and of the total number of connections (8-10%) as well.

Fig. 8. GoS vs. VoIP Calls Arrival Rate (proposed Bandwidth Reservation (BR) CAC vs.
proposed DCAC)
358 Recent Advances in Wireless Communications and Networks

Fig. 9. GoS vs. Total System’s Bandwidth (proposed DCAC vs. Capacity-based (CB) CAC)

5. Conclusion
In this chapter, two new admission control schemes for the LTE architecture have been
presented. The first mechanism (BR CAC) is based on bandwidth reservation concept, while
the second (DCAC) reacts dynamically, depending on the available system’s bandwidth.
Compared to simple, Capacity-based (CB) admission control methods for 4G networks, the
proposed solutions improve the Grade of Service of the voice traffic, without deteriorating
the total system performance. The main idea of the proposed schemes is that the base station
serves more VoIP calls by considering the “busy hour” phenomenon. Finally, although both
the proposed algorithms have been designed with LTE infrastructure in mind, the flexibility
of the schemes enables their adaptation to other similar technologies such as IEEE 802.16
(WiMAX).

6. Acknowledgment
This work has been funded by the Research Projects GREENET (PITN-GA-2010-264759),
CO2GREEN (TEC2010-20823) and CENTENO (TEC2008-06817-C02-02).

7. References
3GPP (2010). Evolved Universal Terrestrial Radio Access (E-UTRA) and Evolved Universal
Terrestrial Radio Access Network (E-UTRAN) (Release 10); Overall description;
Dealing with VoIP Calls During “Busy Hour” in LTE 359

Stage 2, 3rd Generation Partnership Project (3GPP), TS 36.300, v. 10.2.0, Dec. 2010.
Available at http://www.3gpp.org/ftp/Specs/html-info/36300.htm
3GPP (2011). Policy and Charging Control Architecture (Release 11) ; Evolved Universal
Terrestrial Radio Access (E-UTRA) and Evolved Universal Terrestrial Radio
Access (E-UTRAN); 3rd Generation Partnership Project (3GPP), TS 23.203, v.
11.0.1, Jan. 2011. Available at
http://www.3gpp.org/ftp/Specs/html-info/23203.htm
Anas, M.; Rosa, C.; Calabrese, F.D.; Michaelsen, P.H.; Pedersen, K.I. & Mogensen, P.E.
(2008). QoS-Aware Single Cell Admission Control for UTRAN LTE Uplink,
Proceedings of IEEE Vehicular Technology Conference (VTC Spring 2008), pp.2487-2491,
Marina Bay, Singapore, May 11-14, 2008.
Bae, S. J.; Lee, J. J.; Choi, B. G.; Kwon, S & Chung, M. Y. (2009). A Resource-Estimated Call
Admission Control Algorithm in 3GPP LTE System, Proceedings of ICCSA 2009,
Suwon, Korea.
Choi, S.; Jun, K.; Shin, Y.; Kang, S. & Choi, B. (2007). MAC Scheduling Scheme for VoIP
Traffic Service in 3G LTE, Proceedings of IEEE Vehicular Technology Conference Fall
(VTC-2005-Fall), pp.1441-1445, Baltimore, USA, Sept. 30-Oct. 3, 2007.
Iversen, V.B. (2010). Teletraffic Engineering Handbook, Technical University of Denmark.
Available at: http://oldwww.com.dtu.dk/teletraffic/handbook/telenook.pdf
Jeong, S. S.; Han, J. A. & Jeon, W. S. (2005). Adaptive Connection Admission Control Scheme
For High Data Rate Mobile Networks, Proceedings of IEEE Vehicular Technology
Conference Fall (VTC-2005-Fall), vol.4, pp. 2607- 2611, Dallas, Texas, USA, Sept. 25-
28, 2005.
Koenen, R. (2000). Coding of Moving Pictures and Audio, ISO/IEC JTC1/SC29/WG11
N4668, March 2000.
Kwan, R.; Arnott, R. & Kubota, M. (2010). On Radio Admission Control for LTE Systems,
Proceedings of Vehicular Technology Conference Fall (VTC 2010-Fall), pp.1-5, Ottawa,
Canada, Sept. 6-9, 2010.
Lei, H.; Yu, M.; Zhao, A.; Chang, Y. & Yang, D (2008). Adaptive Connection Admission
Control Algorithm for LTE Systems, Proceedings of IEEE Vehicular Technology
Conference (VTC) 2008, pp.2336-2340, Marina Bay, Singapore, May 11-14, 2008.
Puttonen, J.; Kolehmainen, N.; Henttonen, T.; Moisio, M. & Rinne, M. (2008). Mixed Traffic
Packet Scheduling in UTRAN Long Term Evolution Downlink, Proceedings of IEEE
19th International Symposium on Personal, Indoor and Mobile Radio Communications
(PIMRC) 2008, pp.1-5, Cannes, France, Sept. 15-18, 2008.
Qian, M.; Huang, Y.; Shi, J.; Yuan, Y.; Tian, L. & Dutkiewicz, E. (2009). A Novel Radio
Admission Control Scheme for Multiclass Services in LTE Systems, Proceedings of
IEEE Global Telecommunications Conference (GLOBECOM) 2009, pp.1-6, Honolulu,
Hawaii, USA, Nov. 30-Dec. 4, 2009.
Report Study (2009). Enterprise VoIP Market Trends 2009-2012, Osterman Research Inc., Feb.
2009.
Saha, S. & Quazi, R. (2009). Priority-coupling-a semi-persistent MAC scheduling scheme for
VoIP traffic on 3G LTE, Proceedings of 10th International Conference on
Telecommunications (ConTEL 2009), 2009, pp.325-329, Zagreb, Croatia, June 8-10,
2009.
360 Recent Advances in Wireless Communications and Networks

Sallabi, F. & Khaled Shuaib, K. (2009). Downlink Call Admission Control Algorithm with
Look-Ahead Calls for 3GPP LTE Mobile Networks, Proceedings of IWCMC’09,
Leipzig, Germany, June 21–24, 2009.
Weber, J. (1968). Dictionary of English Language Traffic Terms, , IEEE Transactions on
Communication Technology, vol.16, no.3, pp.365-369, June 1968.
 17

A Semantics-Based Mobile
Web Content Transcoding Framework
Chichang Jou
Tamkang University
Taiwan

1. Introduction
With the rapid development of wireless communication technology, in addition to desktop
computers, many users are accessing the internet from hand-held appliances, such as
tablets, PDAs, and cellular phones. Many new computation paradigms, such as pervasive
computing and ubiquitous computing, have been proposed to embrace this emerging
portable computation framework. However, because of the difference in users’ speed in
adopting new technologies, hand-held devices in use have miscellaneous hardware
limitations, such as CPU speed, power, memory, bandwidth, and image resolutions. They
also have various restrictions in software support, such as operating system, installed
programs, real-time processing capability, and rendering functionality. These ad hoc
limitations have become barriers in human-computer interaction. Most web contents, such
as web pages and images, are mainly in the HTML format, which is designed for desktop
computers. Without proper modification, most rendered web contents in hand-held devices
encounter distorted or fragmented user interface, broken images, slow responses, etc. These
ad hoc characteristics of hand-held devices have become a barrier in enhancing web
availability.
In this chapter, we illustrate how a semantics-based content adaptation framework could be
utilized to fill up the computational gap between mobile devices and desktop computers.
The transcoding mechanism of our framework, called Content Adaptation Proxy Server
[CAPS], resides behind web servers. In CAPS, web pages and image files are transcoded
according to: (1) RDF (Resource Description Framework) (Manola et al., 2004) of web
content; and (2) semantics extracted from the CC/PP (Composite Capability Preferences
Profiles) (Klyne et al., 2004) client device configuration. These semantic properties will be
stored and interpreted inside the Jena Inference System (Carroll et al., 2004) as knowledge
facts to obtain proper transcoding parameters for each particular device. Then web pages
with proper layout modification and images with proper rendering parameters for each
particular device will be constructed and delivered. This technology will recreate contents
suitable for resource-limited devices to balance information loss and information
availability.
For the rest of this chapter, we will review related work in Section 2. In Section 3, we
describe the design principle and system architecture of CAPS. Semantics extraction and
knowledge base construction for the Jena Inference System are discussed in Sections 4 and 5.
362 Recent Advances in Wireless Communications and Networks

Section 6 demonstrates web content transcoding process. Section 7 illustrates the

implementation of CAPS and demonstrates transcoding examples on various mobile
devices. Section 8 concludes this paper.

2. Related work
According how many web content types are handled, web content adaptation could be
classified into: universal and specific web content adaptation. Universal web content
adaptations are mostly proxy-based and could be applied to web pages and various
multimedia types. Their functions include integrating various web content types for the
rendering. Section 2.1 will cover proxy-based universal web content adaptation. Specific
web content adaptation focuses on the algorithm design. The most frequently studied web
content type is HTML files. Due to the complexity in analyzing HTML files, proper
adaptation of HTML files in mobile devices is very difficult. Section 2.2 will cover works in
using heuristics to adjust layouts of HTML documents to fit a particular mobile device. We
will cover how the semantic web technology has been applied to web content adaptation in
Section 2.3.

2.1 Proxy-based universal web content transcoding

Nagao et al. (2001) proposed constructing on the Web a system framework using XML and
external annotations to Web documents. They proposed three approaches for annotating
documents—linguistic, commentary, and multimedia. With annotated documents that
computers can understand and process more easily, their framework allowed content to
reach a wider audience with minimal overhead.
Lum and Lau (2002) built a quality-of-service oriented decision engine for content
adaptation. They designed flows for content negotiation and processing for multimedia
contents.
Ardon et al. (2003) prototyped a proxy-based web transcoding system based on network
access control, user preferences, and displaying capability of equipments. Since all
transcoding procedures were finished in the content provider’s server, this server-centric
framework avoided potential copyright problems.
Sacramento et al. (2004) designed the Mobile Collaboration Architecture [MoCA], a
middleware for developing and deploying context-aware collaborative applications for
mobile users. It comprises client and server APIs, core services for monitoring and inferring
the mobile devices' context, and an object-oriented framework for instantiating customized
application proxies.
Hua et al. (2006) integrated content adaptation algorithm and content caching strategy for
serving dynamic web content in a mobile computing environment. They constructed a
testbed to investigate the effectiveness of their design in improving web content readability
on small displays, decreasing mobile browsing latency, and reducing wireless bandwidth
consumption.
Hsiao et al. (2008) proposed the architecture of versatile transcoding proxy (VTP). The VTP
architecture can accept and execute the transcoding preference script provided by the client
or the server to transform the corresponding data or protocol according to the user's
specification. They adopted the concept of dynamic cache categories and proposed a new
replacement algorithm for caches.
A Semantics-Based Mobile Web Content Transcoding Framework 363

Nimmagadda et al. (2010) presented a content adaptation method for multimedia

presentations constituting media files with different start times and durations. They
performed adaptation based on preferences and temporal constraints specified by authors
and generate an order of importance among media files. Their method can automatically
generate layouts by computing the locations, start times, and durations of the media files.

2.2 Web page transcoding

Bickmore and Girgensohn (1999) designed a “Digestor System” which was capable of
automatic filtering and re-authoring so that WAP-enabled cellular phones could read HTML
contents. Their basic idea was to extract plain texts in the HTML document by discarding all
formatting elements and unnecessary information. The result was then divided into a
navigation page and several plain text sub-pages. They also utilized transcoding cache to
diminish the run-time overhead.
Huang and Sundaresan (2000) tried the semantics approach in transcoding web pages to
improve web accessibility for users. Their system was designed to improve the interface of
e-business transactions and to extend interoperable web forms to mobile devices. They used
XML/DTD to specify the semantic and grammatical relationship among web contents, so
that web forms could achieve consistency, simplicity and adaptability. The advantage of this
system was its ability to provide concept-oriented content adaptation, but it was difficult to
be extended.
Buyukkokten et al. (2001) used an “accordion summarization” transcoding strategy where
an HTML page could be expanded or shrunk like an accordion. The HTML page was
restructured as a tree according to the semantic relationships among its textual sections. All
textual sections were split into several Semantic Textual Units, which were automatically
summarized. Users could check each summary to expand the node for detailed information.
However, this framework only worked in the browser they designed for digital libraries.
Hwang et al. (2003) also treated web page layout as a tree according to the tag hierarchy.
They defined a grouping function to transform such a tree into sub-trees, and introduced a
filtering mechanism to modify the sub-trees for adequate display in the target device. They
analyzed specific web page layout structure and re-authored, according to heuristics, web
pages for several mobile devices. Each of their transcoding method could handle only
specified layout structures of web pages and did not consider mobile device characteristics.

2.3 Semantic web technologies in web content transcoding

In addition to Huang and Sundaresan (2000), several researchers have tried to incorporate
semantic web technologies into web content transcoding. DELI (Butler, 2002), an HP
Semantic Lab project, adopted simple negotiation algorithms for rewriting web pages based
on context information, like user preference and device capabilities. Due to lack of
implementation, its applications were restricted.
Hori et al. (2000) proposed an annotation-based system for Web content transcoding. They
introduced a framework of external annotation, in which existing Web documents were
associated with content adaptation hints as separate annotation files. This annotation-based
transcoding system was then extended with particular focus on the authoring-time
integration between a WYSIWYG annotation tool and a transcoding module.
Glover and Davies (2005) used heuristic algorithms to find proper pre-defined web page
templates according to device attributes. Their focus was in applying XML/XSLT styles to
database contents retrieved in dynamic web pages.
364 Recent Advances in Wireless Communications and Networks

Hsu at el. (2009) proposed a hybrid transcoding approach to combine the traditional
transcoding technologies based on ontology-based metadata to improve the rendering
problem caused by heterogeneous devices. This heterogeneous markup document
transcoding platform was then presented to serve as a transcoding service broker to
facilitate interoperability between distributed heterogeneous transcoders.

3. Design principle and system architecture of CAPS

This section illustrates the design principle and system architecture of CAPS.

3.1 Design principle of CAPS

Butler (2001) describes the capabilities of mobile devices in the following categorizations: (1)
output: screen, resolution, color, relative size of fonts, sound, etc. (2) input: touchscreen,
mouse, keyboard, voice, joystick, etc. (3) processor (4) memory (5) multimedia objects: GIFs,
JPGs, WAVs, MP3s, etc. (6) application language: native code, intermediate code. (7)
browser language: content markup, client side scripting, applet, and styling. When a
particular mobile device receives a web content that it could not handle, it should consider
web content adaptation regarding the above aspects. However, from the users’ point of
view, a more important issue is whether the adapted content could be comprehended. To
tackle the above two issues, CAPS follows the following guidelines of the device
independence principle of W3C (Gimson et al., 2003):
• For some web content or application to be device independent, it should be possible for
a user to obtain a functional user experience associated with its web page identifier via
any access mechanism.
• A web page identifier that provides a functional user experience via one access
mechanism should also provide a user experience of equivalent functionality via any
other access mechanism.
• It should be possible for a user to provide or update any adaptation preferences as part
of the delivery context.

3.2 System architecture of CAPS

In this section, we explain the system architecture and the data flows of Content Adaptation
Proxy Server [CAPS] in Figure 1. Its adaptation mechanism resides behind web servers. The
Proxy Listener is responsible for receiving HTTP requests (message 1) from miscellaneous
mobile devices and for dispatching these requests to the Web Content Fetcher. The client’s
device information as well as user’s personal preferences will be embedded inside these
requests through CC/PP diff, which is a modified version of predefined CC/PP profile from
the hardware manufacturers.
When Proxy Listener accepts a request from a client, it will spawn a working thread in the
Web Content Fetcher to handle the request (message 2). Web Content Fetcher performs the
standard task of proxy servers. If the requested web content is already in the cache, it will be
fetched from Cached Web Content (messages 3.3 and 3.4). If not, then it will be fetched from
the source through internet clouds (messages 3.1 and 3.2), and then be saved in the Cached
Web Content. The working thread is also responsible for resolving the CC/PP profile diff.
The web content and CC/PP diff will then be sent (message 4) to the Semantics Extractor to
acquire the implicit RDF semantic information within the CC/PP diff, HTML web pages
and metadata of image files.
A Semantics-Based Mobile Web Content Transcoding Framework 365

CAPS
Client
1. HTTP Request

Proxy Listener

2. Dispatched Request
3.1 Content Request
Internet
3.3 web content request Cached web
Web Content Fetcher content

3.4 cached web content request

3.2 web content

5.1 device info

Device Info DB
4. Web Content and
CC/PP diff
5.2 semantics
7. HTTP response properties
Jena
Semantics Extractor
Inference
5.3 transcoding parameters System

6. Web Content and Transcoding Service Request

Transcoder

Fig. 1. System architecture of content adaptation proxy server

These semantics information will be sent (messages 5.2) to the Jena Inference System as basic
facts. Jena Inference Engine will combine these facts with the knowledge base of CC/PP
UAProfile RDFS model, Transcoding Rules and Web Page Auxiliary Vocabulary to
determine the proper transcoding parameters in the format of sequential RDF predicates for
the requests.
The web content and transcoding parameters will then be passed (message 6) to the
Transcoder. Besides the layout rewriting mechanisms, the Transcoder is equipped with
transcoding toolkits for image resolution adjustment. The results of the Transcoder
processing consist of web pages with modified layout and images with proper resolution
and parameters suitable for the requesting mobile device. They will be returned to the client
(message 7) and displayed in its browser.

4. Semantics extractor of CAPS

This section illustrates details about semantics extracted from the CC/PP device
configurations and the web content. Semantics of device characteristics will be collected
through CC/PP diff. The CC/PP semantics of device configurations are RDF compatible
already. They could be sent to the Jena Inference System as facts. The total customized
device description can be translated into a graph model within the Jena Inference System,
which will be described in Section 5.
366 Recent Advances in Wireless Communications and Networks

4.1 Semantics extracted from device configurations

CC/PP is a two-layered user preferences and device capabilities description based on
XML/RDF (Ohto & Hjelm, 1999). CC/PP consists of the following three categories:
hardware platform, software platform, and browser user agent. Figure 2 shows example
detailed attributes in the XML/RDF format (Klyne et al., 2004) for one mobile device.
In CAPS, all predefined CC/PP profiles are stored within a profile directory. A CC/PP diff
is manually configured by the user, and is normally used to reflect the user preferences. It is
dynamic and can be further modified in later sessions. Many protocols have been proposed
to enhance HTTP 1.1 protocol to include CC/PP profile diff. Two of such protocols are
CC/PP-ex (Ohto & Hjelm, 1999) and W-HTTP (WAP, 2001). We adopt CC/PP-ex in this
framework.

Fig. 2. Example CC/PP for a mobile device

4.2 Semantics extracted from web content

Since most HTML pages are not well-formed, it is hard to extract semantics from them
directly. The semantics extractor module first would transform HTML pages into the well-
formed XHTML format through the JTidy1 toolkits. The following two file types of the

1JTidy, http://jtidy.sourceforge.net
A Semantics-Based Mobile Web Content Transcoding Framework 367

requested URL will be handled in CAPS: (1) XHTML files: Their metadata are about layouts
of the document, possibly with hyperlinks to external textual or binary files. (2) Image files:
These are binary files with adjustable parameters, like color depths and resolution.
Currently, CAPS could handle JPEG, PNG, and GIF images. For files encoded in
indestructible formats, like Java applets, since they could not be adjusted, CAPS would
directly forward them to the Transcoder for delivery to the client device.
To extend CAPS to new file types, we need to specify just the metadata about the new file
type, and to build the semantics extraction component and transcoding rules for web
contents of the new file types.
For XHTML files, the semantic extractor module collects the following schema information:
the identification of each XHTML element, and the layout of the XHTML page. We apply
XHTML Document Object Model [DOM] (Le Hégaret et al., 2005) tree node scanning to
extract node information and relationships among XHTML elements. We solve the element
identification problem in an XHTML page by XPath (Clark & DeRose, 1999) so that each
node in the DOM tree could be specified accurately.
Statistical or inferred semantics data for the following Web Page Auxiliary Vocabulary will
be extracted for each XHTML DOM tree node:
1. NumberOfWords: This data indicates number of words in a paragraph. It is used to
determine whether the paragraph corresponding to the XHTML node should be split.
2. NumberOfImages: This data indicates number of the <IMG> tags in a specific XHTML
node. It is used to decide whether a tabular cell is an advertisement banner.
3. AverageLink: This is the quotient of the number of links and the number of words
within a XHTML node. In web contents with useful information, this value tends to be
very high, and all contents in the node should be preserved.
4. Title: For XHTML nodes with the <H1> or <H2> tag, or with texts surrounded by pairs
of the <B></B> or <STRONG> </STRONG> tags and followed by <BR> immediately,
the collected content is treated as a title. This could be used as the title of the sub-page
corresponding to this node.
5. Layout: This information indicates whether the node is used for layout composition.
For example, to determine whether a <TD> is a layout element or an actual tabular cell,
we calculate the number of words for the element. If its number of words exceeds a
specific threshold, we mark such a <TD> element as a layout element.
Consider the following simplified XHTML page:
<HTML>
<BODY>
<TABLE>
<TR>
<TD>Gentoo Linux is a totally new linux distribution.
</TD>
</TR>
</TABLE>
</BODY>
</HTML>
We can describe the <TD> tag in the above page with RDF, XPath and Web Page Auxiliary
Vocabulary as follows:
368 Recent Advances in Wireless Communications and Networks

<rdf:Description rdf:about="/HTML/BODY/TABLE/TD[1]">
<rdf:type rdf:resource="html:ELEMENT_NODE" />
<html:NumberOfWords>8</html:NumberOfWords>
<html:IsLayout>false</html:IsLayout>
<html:NumberOfImage>0</html:NumberOfImage>
<html:NodeName>TD</html:NodeName>
<html:ChildNodeNumber>0</html:ChildNodeNumber>
<html:ParentNode rdf:parseType=”Resource”
rdf:resource=”_:/HTML/BODY/TABLE/TR[1]”/>
</rdf:Description>

5. The Jena inference system of CAPS

Knowledge Base

XHTML Schema + Transcoding Rules

Auxiliary Vocabulary

Jena Inference Engine

Facts

Web Pages Merged CC/PP

Semantics Profile

Fig. 3. Architecture of the Jena inference system

We use Jena (Carroll et al., 2004), a semantic web toolkit of “Device Independent” ideal, to
determine the transcoding parameters, which are represented as a sequence of RDF
predicates. The Jena Inference System, displayed in Figure 3, has three main components.
These components are utilized in CAPS as follows:
1. Knowledge Base – It contains the acquired knowledge and rules in deciding the
content adaptation parameters. XHTML schema is derived by mapping from XHTML
XML schema to RDF/RDFS as one knowledge base. Transcoding Rules contain rules
using web content ontology and device characteristics ontology for transcoding.
Auxiliary Vocabulary for transcoding parameter decision are also described by
RDF/RDFS and serialized into Jena knowledge base. All RDF knowledge is serialized
in the XML format to provide more flexibility and interoperability in content
adaptation.
2. Jena Inference Engine – This is the decision engine to inference and to generate
transcoding parameters. We make use of the engine without any modification.
A Semantics-Based Mobile Web Content Transcoding Framework 369

3. Facts – These are facts supported in the form of instantiated predicates. In CAPS, they
are the semantic data collected by the Semantic Extractor.
The transcoding rules of CAPS are to link the web content description and device
information in CC/PP, which are transmitted via CCPP diff. In other words, these rules
would map the user’s or device’s requirements into parameter settings of web content. We
follow the semantic network model (Hayes & McBride, 2004) defined by W3C, and define
the following sets of facts to explain formal meaning of the transcoding rules:
1. D: CC/PP Facts transmitted from the client side. These are used to describe
characteristics of devices or user preferences.
2. Ct: All web content semantics for MIME type t obtained by “semantics extractor”.
3. L: All constraints and limitations of web content. For example, image width less than
screen width, image color depth less than or equal to 24 bit, etc.
In addition, we define several vocabularies to represent the actions in removing an element
in text/html document, and in changing the coding of a web page. For MIME type t,
suppose the set of transcoding operations for type t is Ot. The union of the above sets forms
the set of all statements of our RDF model:

SRDF = ∪ (Ct ∪ Ot ) ∪ L ∪ D
t

The preconditions of the transcoding rules is a subset of ∪Ct ∪ L ∪ D . The result of these
t
transcoding rules is a sequence of operations. An example sequence is: update image size,
change coding, etc. Suppose the set of possible transcoding actions for MIME type t is called
Mt. In CAPS, Mt is formally defined as:

Mt = Ot ∪ C t

We call Mt the transcoding module for MIME type t. The result of transcoding rules for web
content with type t could be represented a subset of Mt. Finally, the set of all possible
transcoding rules for MIME type t could be defined as:

Rt = P(Ct ∪ D ∪ L ) × P( Mt )

where P( ) is the power set function. Thus, the input of the transcoding rules is a set of
statements about web content (Ct), device CC/PP configurations (D), and constraints (L).
The output of the transcoding rules then is a member of P(Mt) that are the required actions
for the transcoding.
We apply heuristics to design the transcoding rules in the “IF…THEN…” format. If the
precedent parts of a rule are all true, then the consequent part of the rule would be added as
a statement of transcoding parameters into the knowledge base. We provide rules regarding
device characteristics by defining restriction rules using first order predicate logic. Rules are
categorized to back up each other. For example, if rules for HP 6530 PDA are not sufficient,
then rules for PPC Pocket PC could be used. The following example rule is to reduce the
width of a JPG image file to fit the screen width of the mobile device:

Accept(DV, “image/jpeg”) ^ Format(I, “image/jpeg“) ^ ScreenWidth(DV,DW) ^

ImageWidth(I, IW) ^ LessThan(DW, IW) Æ ScaleImageByWidth(I, DW)
370 Recent Advances in Wireless Communications and Networks

In the above RDF rule, DV, I, DW, IW are variables for device, image, device width, and
image width. Accept(DV, “image/jpeg”) is a statement of the RDF model to express that the
device could render the multimedia type “image/jpeg”. The other predicates before the
“→“ symbol could be easily interpreted similarly. So the predicates before the “→“ symbol
are to check the conditions about web content (Format and ImageWidth), device CC/PP
configurations(Accept), and constraints(LessThan). The predicate after the “→“symbol
indicate the action to be performed: ScaleImageByWidth. The above inference rule could be
expressed by the following Jena rule:
[ ScaleImageByWidth:
(system:Content content:Width ?image_width),
(system:HardwarePlatform ccpp:DisplayWidth ?display_width),
lessThan( ?display_width, ?image_width ) ->
(system:Content content:ScaleImageByWidth ?display_width) ]
The above rule is named ScaleImageByWidth. In Jena rules, names prefixed with ‘?’ are
variables. The namespaces “system”, “ccpp”, and “Content” point to the content adaptation
proxy server, the standard UAProf Schema (WAP, 2001), and web content under
transcoding, respectively. The above rule means that if the width of the device
(?dispaly_width) is less than the width of the image (?image_width), then set width of the
image as width of the device, and set height of the image proportionally with respect to the
adjustment ratio of the width.

6. Transcoder of CAPS
The Transcoder is composed of several transcoding modules corresponding to file types of
XHTML, JPEG, etc. It could be extended to handle other file types. According to the
transcoding actions and parameters produced by Jena, it performs detailed content
adjustment and filtering. For image files, currently the system not only transforms image
files into the same format with different parameters, but also transforms image files into
different formats.
According to file type of the web content, the Transcoder dispatches them to the
corresponding adaptation component. To perform the required content transcoding
operations, it will query the inferred RDF model through Jena’s RDF query language RDQL
to obtain the required transcoding parameters. The use of RDQL could prevent tight
coupling of the transcoding components. An example RDQL query is demonstrated as
follows:
SELECT ?predicate, ?object
WHERE ( system:Content, ?predicate, ?object)
USING system FOR http://www.im.abc.edu/~def/proxy.rdfs#
USING is to specify the name space. This query could obtain all RDF statements with subject
system:Content, where system is the name space and Content represents web content
currently under transcoding. Transcoding query results are represented as instantiated
predicates. For example, if the transcoding predicate for an image file is
ScaleImageByWidth, then the Transcoder would adjust the image width and height
proportionally. After all RDQL query results are handled, the resulting web content would
be returned to the client.
A Semantics-Based Mobile Web Content Transcoding Framework 371

7. Implementation of CAPS
We implement the content adaptation proxy server in the Fedora Linux 13 operating system
by the Java Language J2SDK 1.6.0. We use the package org.w3c.dom for handling XHTML
DOM trees of the web pages, and use the java.awt.image.BufferedImage package for
handling the JPEG, PNG, and GIF image files.

7.1 Implementation details

The Semantics Extractor is implemented by a Java interface, a factory for semantic
extraction, and one class for each supported file type. The obtained semantic attributes for
the implemented MIME types are listed in Table 1.

MIME type Semantic attributes extracted

image/jpeg Image Height

Image Width
image/png
Image Color Depth
image/gif Image Format

Encoding of the web page

text/html Number Of Words
In line Document Type
Table 1. The extracted Web content attributes in CAPS
The transcoding rules for images are listed in Table 2, and some of the transcoding rules for
HTML pages are listed in Table 3.
The transcoding modules are responsible for receiving the transcoding parameters and
performing the actual content adaptation. In CAPS, we have implemented modules for
HTML, JPEG, PNG, and GIF files. The JPEG, PNG, and GIF image files are handled by the
java.awt.image.BufferedImage package, while the HTML files are handled by the
org.w3c.dom package. The web content extraction and parsing component obtains the
requested content from the internet, and use JTidy to reformat the web page into the
XHTML format and then build the DOM tree for the page. Most of the transcoding modules
are implemented by built-in Java classes. The only module that we do use non-built-in Java
classes are for the transcoding of images to ASCII files, which is completed through open
source tool Asciizer2.

7.2 System test of CAPS

To demonstrate the functionalities of this framework, we tested three client mobile devices:
HP iPAQ hx2400, Symbian S80 Simulator, and Panasonic EB-X700. We would like to show
the effect of the following two CC/PP parameters: supported file types and display size. The
goal of the adaptation is to avoid the use of horizontal scroll bar, so as to increase the
readability of transcoded pages and images. The related specifications and restrictions of
these devices are listed in Table 4.

2 Asciizer, http://asciizer.sourceforge.net
372 Recent Advances in Wireless Communications and Networks

Transcoding Rule Comment

[ScalingImageByWidth: (system:Content content:Width ?image_width), Adjust image

(system:HardwarePlatform ccpp:DisplayWidth ?display_width), width to fit
lessThan( ?display_width, ?image_width ) -> in screen size
(system:Content content:ScaleImageByWidth ?display_width),
(system:ScaleImageByWidth system:TranscodeType "text/jpeg")]

[ExtractColor: Modify
(system:Content content:Width ?image_color_depth), image color
(system:HardwarePlatform ccpp:ScreenColorDepth ?device_color_depth), depth to
lessThan( ?device_color_depth, ?image_color_depth ) match the
->(system:Content content:ReduceColorDepth ?device_color_depth), display
(system:ReduceColorDepth system:TranscodeType "text/jpeg")] capability of
the device.
[PngToJpeg: Transform
(system:Content content:Type “image/png”), PNG files to
(system:SoftwarePlatform ccpp:CcppAccept ?Bag), JPEG.
noValue(?Bag ?li "image/png"),
(?Bag ?li "image/jpeg")
-> (system:Content system:TransformTo "image/jpeg"),
(system:TransformTo system:TranscodeType "text/jpeg")]

[JpegToPlainText: Transform
(system:Content content:Type “image/jpeg”), JPEG to plain
(system:SoftwarePlatform ccpp:CcppAccept ?Bag), text.
noValue(?Bag ?li "image/jpeg"),
(?Bag ?li "text/plain")
-> (system:Content system:TransformTo "text/plain"),
(system:TransformTo system:TranscodeType "text/jpeg")]

[JpegToHTML: (system:Content content:Type “image/jpeg”), Transform

(system:SoftwarePlatform ccpp:CcppAccept ?Bag), JPEG to
noValue(?Bag ?li "image/jpeg"), (?Bag ?li "text/html") HTML.
-> (system:Content system:TransformTo "text/html"),
(system:TransformTo system:TranscodeType "text/jpeg")]

Table 2. Transcoding rules for image files

Figure 4 shows the upper part of the tested web page (http://www.amazon.com) in a
Microsoft IE 8 browser in a desktop computer. The upper parts of the transcoded pages in
the built-in browser for the three tested mobile devices are displayed by two screen shots in
Figures 5 to 7. All resulting transcoded web pages satisfy the goal of avoiding the use of
horizontal scroll bar by adjusting the page layout, image size, and image resolution.
Unsupported CSS, Javascripts, flashes, div’s and tables are filtered out.
A Semantics-Based Mobile Web Content Transcoding Framework 373

Transcoding Rule Comment

[ExtractTableContent: If the browser in

(?node content:NodeName "table"), the mobile device
(system:BrowserUA ccpp:TablesCapable "No") does not support
->(system:Content system:ExtractTableContent ?node) , table, then extract
(system:ExtractTableContent system:TranscodeType "text/html") ] the content.
[FilterCSSScript: If the browser in
(?node content:NodeName "style"), the mobile device
( system:BrowserUA ccpp:StyleSheetCapable "No" ) does not support
-> (system:Content system:RemoveNode ?node)] CSS, then filter the
[FilterCSSScript: CSS content.
(?node content:NodeName "style"),
(system:SoftwarePlatform ccpp:CcppAccept ?Bag),
noValue(?Bag ?li "text/css")
-> (system:Content system:RemoveNode ?node) ,
(system:RemoveNode system:TranscodeType "text/html")]
[FilterFlash: If the browser in
(?node content:NodeName "object"), the mobile device
(system:SoftwarePlatform ccpp:CcppAccept ?Bag), does not support
noValue(?Bag ?li "x-application/flash"), Flash, then filter
(?node content:InlineDocumentType “x-application/flash”) the Flash content.
-> (system:Content system:RemoveNode ?node) ,
(system:RemoveNode system:TranscodeType "text/html")]
[TransformToPlainText: Transform HTML
(system:SoftwarePlatform ccpp:CcppAccept ?Bag), to plain text.
noValue(?Bag ?li "text/html"), (?Bag ?li "text/plain")
-> (system:Content system:TransformTo "text/plain"),
(system:TransformTo system:TranscodeType "text/html") ]
Table 3. Transcoding rules for HTML files

Device HP iPAQ hx2400 Symbian S80 Panasonic EB-

X700
Category PDA Smart Phone Smart Phone
Operating Windows Mobile Symbian Series80 Symbian
System 5.0 Series60
Browser IE Mobile Built in Built in
text/html text/xhtml text/chtml
Supported file text/css text/css image/jpeg
types image/jpeg image/jpeg
image/png image/png
image/gif image/gif
Display size 480 x 320 (pixels) 220 x 640 (pixels) 176 x 148 (pixels)
Connection Bluetooth WLAN GPRS
Table 4. Specifications and restrictions of tested mobile devices
374 Recent Advances in Wireless Communications and Networks

Fig. 4. Test web page in a desktop computer

Fig. 5. Results of the test page in HP iPAQ hx2400

A Semantics-Based Mobile Web Content Transcoding Framework 375

Fig. 6. Results of the test page in Symbian Series80

Fig. 7. Results of the test page in Panasonic EB-X700

7.3 Comparison of CAPS with related work

In Table 5, we compare CAPS with related works mentioned in Section 2.1 in the following
aspects: purpose, implementation levels, server deployment, dynamic adaptor loading, and
transcoding parameter selection method. Some results are from the comparison about
proxy-based web content adaptation system in Endler et al. (2005).
In the aspect of purpose, systems focused on Quality-of-Service [QoS] would emphasize on
the reduction of transmission time and on improving users’ browsing experiences. Systems
focused on multimedia would emphasize on the handling of multimedia files, like summary
of sound and images, and reduction of sampling frequencies. Systems with general purpose
would have a flexible framework and emphasize on the dynamic deployment capability.
To handle the fast growing number of accepted web content types in mobile devices, web
content adaptation systems normally would be equipped with the dynamic deployment of
376 Recent Advances in Wireless Communications and Networks

transcoding modules to improve the scalability of the system. Only MARCH and CAPS
provide dynamic adaptor loading of transcoding modules. In MARCH, the transcoding
server was dynamically determined by the dynamic transcoding path, where each node
represents a transcoding server. After traversing all nodes in the path through message
passing, in which web content is encapsulated in HTTP/1.1, the web content transcoding is
also finished. In CAPS, the dynamic adapter loading is implemented through Java’s
dynamic binding and Run-Time Type Identification [RTTI]. The transcoding modules are
decided on-the-fly and thus require less loading than MARCH.

MoCA
MARCH Lum and Lau Nagao et al. CAPS
Systems (Sacramento
(Ardon et al.) (2002) (2001) (this study)
et al., 2004)
QoS for universal,
Purpose universal multimedia universal
multimedia caching
Implementa
application - middleware application application
tion level
Server server-driven,
- server-driven proxy-based proxy-based
deployment proxy-based
Dynamic
adapter yes no yes no yes
loading
Transcoding
parameter
rules heuristics rules heuristics rules
selection
method
Table 5. Comparison of CAPS with related work
In parameter selection, there are two approaches: heuristics and rules. The benefits of
heuristics-controlled parameter selection are easy to be implemented in programs and faster
response time. It drawbacks include less number of handled cases, and less flexibility. Once
logic used in parameter selection is changed, the programs must be re-compiled. On the
other hand, use of rules in parameter selection provides better flexibility. Additionally, with
the capability of reasoning, more transcoding strategies could be obtained than heuristics.
However, the drawback of rules-controlled parameter selection is the difficulty to write
correct and proper rules.
CAPS uses light-weight transcoding components through flexible API. It could be easily
deployed as distributed processing by RMI or Web Services. Therefore, the costs for
deployment, allocation, and scalability could be greatly reduced.

8. Conclusion
We designed and implemented a flexible and robust framewrok, called CAPS, for web
content adaptation using RDF semantics from CC/PP device characteristics, XHTML web
pages, and JPEG, PNG, and GIF image files. Past researches in this area either did not take
device characteristics into consideration, or were not a general purpose solution for
A Semantics-Based Mobile Web Content Transcoding Framework 377

miscellaneous mobile devices. We made use of the Jena Inference System to obtain the
transcoding parameters through the fact and knowledge base built from the collected
semantics.
In CAPS, a single copy of the web pages could serve many different mobile devices.
Previous tedious web page rewriting labour for mobile devices could be saved. Our
framework could be easily extended to new file types by importing related semantics and
transcoding modules.
We plan to incorporate support of style sheet and web form specifications, such as
CSS/Mobile (Schubert & Berjon, 2008) and XHTML for Mobile (McCarron et al., 2010), into
this semantics-based content adaptation framework. By supporting these dynamic web
pages, the increased user interactivity would accelerate user acceptance of pervasive
computing.

9. Acknowledgment
This work was supported in part by Taiwan’s National Science Council under Grant NSC
95-2221-E-032 -027.

10. References
Ardon S.; Gunningberg P.; Landfeldt B.; Ismailov Y.; Portmann M. & Seneviratne A. (2003).
MARCH: a distributed content adaptation architecture. International Journal of
Communication Systems, Vol. 16, No. 1, pp. 97-115, ISSN 1074-5351
Bickmore T. & Girgensohn A. (1999). Web page filtering and re-authoring for mobile users.
The Computer Journal, Vol. 42, No. 6, pp. 534-546, ISSN 1460-2067
Butler M. (2001). Current Technologies for Device Independence, HP Laboratories Technical
Report, No. 83
Butler M. (2002). DELI: A Delivery context library for CC/PP and UAProf, External technical
report, HP Semantic Lab
Buyukkokten O.; Garcia-Molina H. & Paepcke A. (2001). Accordion summarization for end-
game browsing on PDAs and cellular phones, Proceeding Of the 2001 SIGCHI
Conference on Human Factors in Computing Systems, pp. 213-220, ISBN 1-58113-327-8
Carroll J.; Dickinson I.; Dollin C.; Reynolds D.; Seaborne A. & Wilkinson K. (2004). Jena:
implementing the semantic web recommendations, Proceedings of the 13th World
Wide Web Conference, pp. 74-83, ISBN 1-58113-844-X
Clark J. & DeRose S. (1999). XML Path Language (XPath) v. 1.0, W3C, Available from
http://www.w3.org/TR/xpath
Endler M.; Rubinsztejn H.; Rocha R. C. A.; Sacramento V. (2005). Proxy-based Adaptation
for Mobile Computing, Technical Report, Departamento de Informática, Pontifícia
Universidade Católica do Rio de Janeiro, Available from http://ftp.inf.puc-
rio.br/pub/docs/techreports/05_24_endler.pdf
Gimson R.; Finkelstein S. R.; Maes S. & Suryanarayana L. (2003). Device Independence
Principles, W3C Working Group Note, Available from
http://www.w3.org/TR/di-princ/
Glover T. & Davies J. (2005). Integrating device independence and user profiles on the web.
BT Technology Journal, Vol. 23, No.3, pp. 239-248, ISSN 1358-3948
378 Recent Advances in Wireless Communications and Networks

Hayes P. & McBride B. (2004). RDF Semantics, W3C Recommendation, Available from
http://www.w3.org/TR/rdf-mt/
Hori M.; Kondoh G.; Ono K.; Hirose S. & Singhal S. (2000). Annotation-based web content
transcoding, Computer Networks, Vol. 33, No. 1-6, pp. 197-211, ISSN 1389-1286
Hsiao J. L.; Hung H. P. & Chen M. S. (2008). Versatile transcoding proxy for internet content
adaptation. IEEE Transactions on Multimedia, Vol. 10, No. 4, pp. 646 - 658, ISSN 1520-
9210
Hsu I. C.; Chi L. P. & Bor S. S. (2009). A platform for transcoding heterogeneous markup
documents using ontology-based metadata. Journal of Network and Computer
Applications, Vol. 32, No. 3, pp. 616-629, ISSN 1084-8045
Hua Z.; Xie X.; Liu H.; Lu H. & Ma W. (2006). Design and performance studies of an
adaptive scheme for serving dynamic web content in a mobile computing
environment, IEEE Transactions on Mobile Computing, Vol. 5, No. 12, pp. -1662, ISSN
1536-1233
Huang A. W. & Sundaresan N. (2000). A semantic transcoding system to adapt web services
for users with disabilities, Proceeding of the 4th international ACM conference on
Assistive technologies, pp. 156-163, ISBN 1-58113-313-8
Hwang Y.; Kim J. & Seo E. (2003) Structure-aware web transcoding for mobile devices. IEEE
Internet Computing, Vol. 7, No. 5, pp. 14-21, ISSN 1089-7801
Klyne G.; Reynolds F.; Woodrow C.; Ohto H.; Hjelm J.; Butler M. & Tran L. (2004).
Composite Capability/Preference Profiles (CC/PP): Structure and Vocabularies 1.0,
W3C, Available from http://www.w3.org/TR/CCPP-struct-vocab
Le Hégaret P.; Whitmer R. & Wood L. (2009). Document Object Model (DOM), W3C,
Available from http://www.w3.org/DOM/
Lum W. Y. & Lau F. C. M. (2002). A context-aware decision engine for content adaptation.
IEEE Pervasive Computing, Vol. 1, No. 3, pp. 41-49, ISSN 1536-1268
Manola F. & Miller E. (2004). RDF Primer, W3C, Available from
http://www.w3.org/TR/rdf-primer/
McCarron S.; Ishikawa M.; Baker M.; Matsui S.; Stark P.; Wugofski T. & Yamakami T. (2010).
XHTML™ Basic 1.1 - Second Edition, W3C Recommendation, Avail able from
http://www.w3.org/TR/2010/REC-xhtml-basic-20101123/
Nagao K.; Shirai Y. & Squire K. (2001). Semantic annotation and transcoding: making web
content more accessible, IEEE MultiMedia, Vol. 8, No. 2, pp. 69-81, ISSN 1070-986X
Nimmagadda Y.; Kumar K. & Lu Y. H. (2010). Adaptation of multimedia presentations for
different display sizes in the presence of preferences and temporal constraints,
IEEE Transactions on Multimedia, Vol. 12, No. 7, pp. 650 - 664, ISSN 1520-9210
Ohto H. & Hjelm J. (1999) CC/PP exchange protocol based on HTTP Extension Framework,
W3C, Available from http://www.w3.org/TR/NOTE-CCPPexchange
Schubert S. & Berjon R. (2008). CSS Mobile Profile 2.0, W3C Candidate Recommendation,
Available from http://www.w3.org/TR/2008/CR-css-mobile-20081210/
Wireless Application Protocol Forum, Ltd. (2001). Wireless Profiled HTTP, Available from
http://read.pudn.com/downloads19/doc/comm/67224/WAP-229-HTTP-
20010329-a.pdf
Wireless Application Protocol Forum, Ltd. (2001). WAG UAProf, Available from
http://read.pudn.com/downloads19/doc/comm/67224/WAP-248-UAProf-
20011020-a.pdf
 18

Power Supply Architectures for Wireless

Systems with Discontinuous Consumption
Jose Ignacio Garate and Jose Miguel de Diego
Bilbao School of Engineering-University of the Basque Country/Trelec BTC S.L.L.
Spain

1. Introduction
Many wireless systems develop in present days relay on discontinuous transmission and
reception or TDD access technology, (Time Division Duplex), as this way of operation has
several advantages, provided it is not required a dedicated link based only on FDD access,
(Frequency Division Duplex). The used of TDD access improves the spectral efficiency,
which increases the number of communication channels available sharing the same
spectrum resources, besides, it enhances the energetic efficiency of the system as the
hardware only works during a limited period of time and, in spite of the FDD access
technique, which requires a continuous power transmission, is becoming more popular
through communication systems like WIFI, MIMO and the like, the road maps for 4G radio
access technologies are based on both TDD and FDD (E. Dahlman et al.,2006). The new LTE,
(3GPP Long Term Evolution), standard for 3GPP flexible spectrum usage is supported
through FDD/TDD harmonisation, where there is a convergence between paired spectrum
and unpaired spectrum solutions (K. Fazel & S. Kaiser, 2008). To illustrate the basic time
domain behaviour of the TDD and FDD channels, Fig. 1 represents their simplified time
versus frequency channel distribution.
The Fig. 1 provides hints of how the communication access scheme affects several
parameters of the wireless system. The present chapter discusses those issues concerning the
power supply system which reverts back to the general performance of wireless systems.

Fig. 1. TDMA, TDD and FDD time versus frequency and code channel distribution
In spite of there are many architectures of power supplies to cope with discontinuous
consumption, due to the special characteristics of the wireless systems, not all of them are
suitable or appropriate to all types of wireless systems. This fact is especially true or
significant for battery powered wireless devices, as the restrictions in size and autonomy
imposed are key factors or introduce new variables or constrains.
380 Recent Advances in Wireless Communications and Networks

The research work that will be presented in this chapter is devoted to developing generic
architectures of power supply systems for wireless systems, which possess the current
consumption pattern of a discontinuous load. It also tries to answer, or at least eases to
understand and face the design, development and production challenges related with the
performance of wireless devices whenever they face with this type of current consumption.

2. Discontinuous consumption in wireless systems

As the discontinuous consumption concept is a generic topic, it requires a reference frame
linked with wireless systems. This chapter considers two types of discontinuous
consumption in wireless devices; a random one not directly involved in the communication
process, for example, the activation of the backlights, the speaker, servos and the like, and a
periodic one that will be addressed as discontinuous which is the subject of the research.
This periodic consumption is linked with the access technology employ in the wireless
system and leads to the transmission and reception time periods. In spite of such
classification, it is interesting to highlight that almost all tasks performed by a wireless
systems processor are controlled and previously programmed, therefore, the magnitude of
the current consumption, demanded by a particular event, it is predefined.

2.1 Characteristics
From the power supply perspective, one of the main attributes of a wireless system with
TDD access scheme is its periodic consumption pattern, Fig. 2. The characteristics
parameters of the consumption are represented in the picture and are the following:
- Period and duty cycle of the consumption (t1, t2).
- Magnitude of the consumption (IPEAK, ILOAD, ISTANDBY).
- Time mask and slopes of the communication burst.

ILOAD
Q1 = t1·I1
T=t2 + t1
I1 IPEAK I2 / I1= kI
Q2 = t2·I2
Time
mask detail

t t1
tON=t1 t2 = kt · t1 I2 = ISTANDBY
Fig. 2. Power versus time load current consumption for wireless system with discontinuous
transmission, and detail of the current pulse
These parameters are the tools to determine or dimension the power supply system of a
wireless system. Period, duty cycle and magnitude set the energy demands place upon the
power supply. Meanwhile, the time mask and slopes of the communication burst are
relevant to control the switching harmonics of the signal and, at the same time, maintain the
signal spectrum within its assigned bandwidth. Fast transitions mean switching harmonics
of high frequency difficult to be restrained within regulation specifications, particularly at
extreme conditions of temperature and voltage.
Power Supply Architectures for Wireless Systems with Discontinuous Consumption 381

2.2 Effects
The noticeable effects of discontinuous consumption in wireless systems are fluctuations
and drops in the supply voltage, applied to the terminals of the load, around the nominal
value; this fluctuation follows the consumption pattern. Voltage drop is ruled by the Ohm
law, but not only must be considered the distributed resistive component of electric path
between load and source, but also its reactive part. The resistive component conditions or
determines the magnitude of voltage drop, meanwhile; the reactive one defines the shape
and damping of consumption rise and fall slopes.

2.2.1 Voltage ripple

In wireless systems, the direct outcomes of voltage ripple are two; switching harmonics, and
voltage level out of operational ranges.
A) Switching harmonics
The frequency bandwidth available for a wireless system is a scarce resource and must be
optimized to allocate as many communication channels as possible. The TDD strategy to
achieve this goal is multiplex in time a number of channels at the same frequency within a
specific bandwidth. To make the communication systems work it is required that the
transmission is produced in a specific timing. Transceiver activation, on its assigned time
slot, is not produced instantaneously, which implies, before the information is received or
transmitted, that there are two periods of time for conditioning the signal. These two time
periods constitute the rise and fall ramp time. To this extent there are two situations to be
considered:
- If ramps are too fast implies high-frequency interferences, switching harmonics.
Switching harmonics reduce the amount of channel spectral density energy available
for communication, consequently, they degrade the link traffic capacity and its overall
performance, in other words, it means that could be set less communication links.
- If slopes are too slow, they widen the bandwidth and corrupt the spectral modulation
mask, which occupy the adjacent channel reducing the traffic maximum rate and the
sensitivity of adjacent receivers as their SINAD, (signal to noise ratio), is diminish.
B) Voltage ripple
The voltage level apply to the load varies between two values that correspond to minimum
a maximum load. It is likely that the voltage operative range of the wireless device is
exceeded in certain situations, particularly at extreme conditions of temperature.
Moreover, whenever wireless systems are battery powered, voltage drift increases as the
power source voltage varies, between maximum and minimum load, due to the battery
internal resistance. This is also applicable, to a certain extent, if a converter is placed
between the power source and the load, as voltage drift could set the converter out of its
regulation input voltage range.

2.2.2 Discontinuous current and electromagnetic compatibility

Seemingly, discontinuous consumption and voltage drops imply that the current is also
variable. On the other hand, the discontinuous current drain from the power source has a
direct impact on it, particularly for battery powered devices, which means energy losses in
the internal battery resistance that are not uniform, as the load impedance presented varies
382 Recent Advances in Wireless Communications and Networks

following the consumption pattern. Besides, existence of discontinuous current implies

current flux through a wire, which induces magnetic fields on the power lines.
There are three basic mechanisms or arrangements that produce magnetic fields; a signal
track with a variable current, a current loop, and two parallel lines. The strength of magnetic
fields varies with the level of current consumption, and their effects increase if there is any
current loop involving the power lines that connect the source and load. These loops may
produce interferences in any element of the wireless system, within or close to them. To
make the phenomena challenging, usually, the frequency of magnetic field is a low-
frequency one.
It is known that a drawback of low frequency magnetic fields is their mechanism of
attenuation. Magnetic fields require an absorptive shield, (ferrite), instead of the reflective
one use for high frequency electric fields, which reduces its capability to shield them.
Consequently, existence of magnetic fields implies side effects, in terms of the
electromagnetic compatibility, EMC, of wireless systems, which should be avoided to fulfil
the applicable regulation. Thus, design requires not only a careful routing and layout of
power lines but also conditions the distribution of the wireless system architecture on PCB
(M. I. Montrose, 1996).

3. Power supplies and discrete components for wireless systems

From the power supply perspective, once is stated that the classification of wireless systems
starts with the type of access technology employ, which also defines if the consumption is
continuous or periodic, for the power supply is the subject of this chapter, wireless systems
will be sorted in two generic groups based on the type of power source they employ, in spite
of inherit characteristics of portable wireless systems, like cellular terminals, impose certain
restrictions over the power supply architecture and the devices it made of.

3.1 Types of power sources

Power sources are sensitive to the consumption patterns of wireless systems, but the power
source itself conditions the architecture of both wireless device and power supply.
Consequently, wireless systems are sorted in two groups; the first are systems directly
connected to the power source, and the second is made of those that require a conditioning
of the power source voltage and current.

3.1.1 Direct connection to power source

Apparently, the ideal scenario may be a power supply directly connected to the wireless
systems or the load. As there is no electronic between source and load, the energy losses are
reduced to those in the electric paths. This is true meanwhile the energy that the load drains
from the battery is constant and correctly dimensioned to its internal resistance. This ideal
situation is not such, as the energy drain is not always constant, the battery discharges over
time and its capacity varies over the whole operational temperature range.
Battery powered electronic devices such cellular terminals, PDAs, Ebook readers and the
like are typical examples of wireless systems directly connected to the power source.

3.1.2 Voltage and current adapter

If the voltage and current levels of the source need to be conditioning, it is required a
voltage converter between source and load. It does not matter if the power source is a solar
Power Supply Architectures for Wireless Systems with Discontinuous Consumption 383

panel, a battery or the mains AC power lines, this fact will only affect the architecture of the
voltage converter. There are tree generic alternatives: AC-DC isolated converter, DC-DC
isolated converter and DC-DC converter (B. Sahu & G.A. Rincon-mora, 2004).
Whenever AC power source is used, it is mandatory an AC-DC isolated converter, but the
need of isolation between DC power source and the wireless systems is only a matter of
electromagnetic compatibility standards, electrostatic discharges and security regulation.

3.2 Systems, component and devices for wireless power supply

Unless there is a wide range of components for power supplies and sources, the next lines
summarize the requirements upon key components and devices of the power supply.

3.2.1 Battery
The main power source of portable or battery powered wireless systems is the battery cell
itself (Saft, 2008). The battery could be primary or secondary, i.e., rechargeable or not
rechargeable, respectively. From the point of view o the chapter, the battery equivalent
circuit is made of its internal resistance, RIN. It use to be of low value and depends on the
technology, tenths of milliohms for 1 Ahour capacity Ion-Lithium battery.

Fig. 3. Detail of an Ion-Lithium battery internal protection circuit and its true table
Due to the characteristics of wireless systems stress onto battery voltage supply level, size
and weight the battery technologies more suitable are, among others, the following:
- Niquel-Metal-Hydrite (NiMH) and Niquel-Cadmium (NiCd), both require fuse for
safety.
- Ion-Lithium and Ion-Lithium-Polymer, both need a protection circuit plus the fuse.
The basic circuit architecture of a Lithium battery is shown in the following picture, Fig. 3.
The schematic shows that the equivalent resistance of the cell is made of the internal
resistance of the battery, plus the resistance of contact and the resistance of the protection
circuit. The protection circuit is made of the resistance of the fuse, recommended a
polyswitch type, and a couple of mosfets. The contribution of all these electronic elements
must be considered as they increase the ripple of the voltage supply.

3.2.2 Converters for wireless systems. Types of converters

The performance of wireless systems is sensitive to the power supply voltage ripple and its
fluctuation between maximum and minimum values. Consequently, it is highly
384 Recent Advances in Wireless Communications and Networks

recommended suppress or attenuate the voltage ripple with filtering and voltage regulation.
Filtering is achieved by means of high-value capacitors of low ESR and inductors;
meanwhile, regulation is obtained through DC-DC converters, linear or witched ones.
As long as it is not always feasible a direct connection to the power source, power
converters are used to adapt the power supply voltage and current level to those of the
wireless systems, even if the power source is a battery. Moreover, depending on the systems
architecture, may be required a second regulator to stabilize the output of the former one.
There is a wide range of power supply architectures available, switched or linear (R. W.
Erikson, 1997). If AC-DC conversion is required, in spite of it is possible its integration
within the wireless systems, is better employing an external one of a plug-in type. External
AC-DCs are widespread as they ease the design and certification of the equipment
electronics. This is true because external plug-in are already certified. Besides, in the
particular case of wireless modules, their manufactures usually translate the discontinuous
consumption impact to the application integrator or to the converter manufacturer. The
Fig. 4 shows an example of such a problem; the manufacturer provides a small size chipset,
already certified, but on its application note highlight that it requires to work a capacitor of
the same size plus a voltage regulator.

Fig. 4. Comparison between a communication module and the capacitor it requires

Summarizing the line of reasoning, the selection of power supply technologies for wireless
systems should be guided by the following factors:
- Type of converter
- Isolation.
- Control scheme of the switched converter
- Control architecture of the feedback loop
Once is certified the need of power conversion, remains without answer the topic of
switched or linear conversion. The advantages and drawbacks of linear regulation versus
switched regulation are exposed in the following lines.
1) Linear regulation is obtained through a voltage control loop that samples the output
voltage. The main device of a linear regulator works on its active operation region, so the
voltage drop across its terminal produces power losses in the form of heat sink.
The advantages of linear regulation are its simple architecture, and the lack of
electromagnetic interference. Also, it does not require inductive elements, and its current
consumption under no-load conditions is low. On the other hand, it has low efficiency when
the difference between input and output voltage are significant.
2) A switched converter employs an active device that works between cut and saturation
regions; therefore, the dissipation losses are lower and cause, mainly, by switching losses
Power Supply Architectures for Wireless Systems with Discontinuous Consumption 385

and the voltage drop in the active device over cut and saturation. The power is delivered to
the load through the energy store in an inductor, which charging cycle is a function of the
energy demanded by the load. So, the energy drained from the source is used mostly to
feeding the load, which reduces the power losses that are limited to those of the control
circuit and the component leakages. Therefore, a performance analysis of switched
converters shows that they provide a better balance between input and output voltages than
the linear ones. They are, also, smaller and lighter than its linear counterparts for the same
power rating, mainly because the isolation transformer is smaller. Furthermore, the size and
value of the transformer or the switching inductance and the capacitors are reduced as the
switching frequency is increased. Lower value capacitors contribute to reduce the voltage
ripple, because it is possible used ceramic capacitors of low ESR, in the order of tenths
milliohms or lower.
On the other hand, a switched power supply introduces electromagnetic fields, radiated and
conducted, that make the technical requirements restrictive, as the complexity of electronic
design increases. Switched regulators are, also, more complex to design due to they require
a higher number of discrete components, which reduces the electronic liability.
Moreover, switched converter has another issue that must be bear in mind for green design
applications. As long as the current consumption is discontinuous, the load remains inactive
for some periods of time; during those periods its current consumption may reach zero.
Hence, switched converter has poor efficiency under no-load conditions as there is a
quiescent current in the electronic of the power supply. For example, standard 12 V and 4 W
commercial DC-DC have a quiescent current consumption between 30 and 50 mA.
Unless solutions switched regulation based may appear the most suitable, many
manufactures employ linear regulation, especially when; there is available a power source
with voltage levels close to those required by the wireless system, and size it is not a
restriction. Doing so it is avoided EM fields, which increase cost and technical requirements.

3.2.3 Capacitors
Power supply of wireless systems employs capacitors to store energy and filtering. The
challenges to face are finding capacitors of high value, small size and low ESR that
withstand the voltage levels applied to the electronics.
Sometimes, the equipment size does not allow the use of high-value capacitors; the
alternative is employ capacitors of hundred microfarads that only help to smooth voltage
transitions. This is the case of GSM cellular terminals that when transmitting at maximum
power, the peak current consumption may reach 3 A.
Furthermore, capacitor ESR produces load voltage ripple, and its leakage resistance
introduces a continuous discharge of the battery. For example, an standard tantalum
capacitor, AVX model TPCL106M006#4000, has 10 µF nominal capacitance and ESR of 4000
mΩ. An electrolytic capacitor provides higher capacitance value on a bigger size and with
more ESR. On the other hand, a ceramic one has small size and low ESR, but there are not
feasible for high capacitance. Table 1 highlights the differences between technologies for the
same capacitance value.
Then the main limiting factors of capacitors are their ESR and size. The Table 1 provides a
comparison between different types of capacitors. High value capacitors are intended to be
used in the equipment, close to the load. To reduce the impact of the size it is possible;
redistribute several capacitors in parallel, or use the technology of super-capacitors.
386 Recent Advances in Wireless Communications and Networks

C ESRMAX VMAX Size

Technology Supplier Code
(mF) (mΩ) (V) (mm)
Tantalum Kemet A700X227M006ATE015 220 15 6,3 7.3×4.3×4.0
Tantalum AVX TPSD477*006-0100 470 100 6,3 7.3×4.3×2.8
Electrolytic Nichicon UUG1A102MNL1MS 470 790 25 Ø12.5×13.5
Tantalum Kemet A700X477M002ATE015 470 15 2 7.3×4.3×4.0
Tantalum AVX TAJD477*002-NJ 470 200 2.5 7.3×4.3×2.9
Electrolytic Nichicon UUG1A471MNL1MS 1000 371 10 Ø12.5×13.5
Electrolytic Nichicon UUG0J222MNL1MS 2200 183 6,3 Ø12.5×16.5
Electrolytic Nichicon UUG0J472MNL1MS 4700 100 6,3 Ø16×16.5
Electrolytic Nichicon UUG0J682MNL1MS 6800 77 6,3 Ø18×16.5
Super-cap AVX ES48301 60000 190 6,3 48×30×4.0
Table 1. Comparison between high-value capacitors technologies
Super-capacitor employs new technology developed in recent years. They combine high
capacitive values with small size and low ESR, which provide good performance against
high current surges, making them suitable for applications with high-peak currents. As an
example, the technical parameters of some super-capacitors are summarized on Table 2.

C ESRMAX VMAX Ileakage Size

Supplier Code
(mF) (mΩ) (V) (µA max) (LxWxH mm)
AVX BZ015B603Z_B 60 96 5,5 10 28 x 17 x 6,5
AVX BZ02CA903Z_B 90 108 12 20 48 x 30 x 6,8
Cooper FC-3R6334-R 330 250 3,6 - 2 x 17 x 40
Maxwell PC10-90 10 180 2,5 40 29,6x23,6x 4,8
Table 2. Comparison between super-capacitor technologies

4. Wireless systems powered through passive components

Wireless systems powered through passive components have in common the type of power
source, which is often a battery. At this point, the key issue is how to increase the autonomy
of these electronic devices, in doing so, the following items should be consider, balancing
the tradeoffs of each one:
- Limit the load active times by reducing TX and RX periods.
- Increase the efficiency of the power supply system.
- Smooth current and voltage transitions.
- Reduce standby and quiescent current consumption.
The characteristics of battery powered wireless devices reduce the range of alternatives of
power supply systems exclusively based on passive components, especially if the
restrictions are combined with small size requirements. The most widespread architectures
of power supply systems with passive components are described in the following topics.
Unless the conclusion and results could be extrapolated to any wireless communication
system with discontinuous consumption, in order to homogenize the description, and allow
Power Supply Architectures for Wireless Systems with Discontinuous Consumption 387

the comparison of different architectures, the reference wireless communication system is a

GSM cellular terminal that transmits and receives only in one time slot. In this framework,
the characteristics parameters of the terminal are the following:
- Frequency of the GSM pulse = 216 Hz.
- Transmission time, tON = 1/8 of the period, or time slot that last 578 µs.
- Maximum current peak, ILOAD, 2 A for a nominal 3,6 V Ion-Lithium battery.
- Standby current consumption, ISTANDBY 20 mA @ 3,6 V.
- Mean current consumption, IMEAN, equals to 2 A · 1/8 + 0,02 A · 7/8=267,5 mA @ 3,6 V,
Ec. 1.

I MEAN = I LOAD ⋅
tON
+ I STANDBY ⋅
(T − tON ) (1)
T T

4.1 Direct connection

Direct connection between the battery and load reduces the voltage drop in the electrical
path between both elements of the systems (W. Schroeder, 2007). The Fig. 5 represents the
elements that must be considered when scaling a direct connection power supply system,
and it also shows the equivalent circuit of the power supply, the source and the load.
A small capacitor, C, could be included to smooth the voltage ripple of transitions between
the load states ON and OFF, and it also filters some conducted emissions. For this purpose,
wireless device manufactures commonly employ ceramic capacitor of around 10 µF. This
capacitor only has effect on the first microseconds of the transient; consequently, the voltage
drop in the load, VLOAD, is the same independently of the consumption peak, Fig. 6. It could
be appreciated in the figure how the voltage ripple increases proportionally to the current
consumption and depends on the distributed resistance between source and load.

Fig. 5. Schematic diagram of a wireless system directly connected to the battery

4.2 High value load capacitor

Direct connection presents sharp transitions in the waveforms of current and voltage at both
ends of load and source, Fig. 6. A straightforward regulation system uses a high-value
capacitor in parallel with the load to smooth both, current and voltage, waveforms. The
capacitor acts as a low-pass filter damping the slopes of the consumption transitions, in
other words, it delivers a fraction of the energy that the wireless systems demands to the
power source. The energy that a capacitor drives depends on its parameters, and the load
consumption characteristics. Capacitor stores energy over inactive cycle of load and delivers
388 Recent Advances in Wireless Communications and Networks

energy when the load is active. The higher the capacitor or super-capacitor value the lower
the load voltage ripple. The impact of capacitor on the power supply performance will differ
depending on where is located. It could be placed in two different locations:
- In the battery cell or at the ends of the battery terminals.
- Close to the load, within the wireless electronics.

(a) (b)
Fig. 6. Load voltage and battery current waveforms of a wireless system with discontinuous
consumption for (a) maximum consumption and (b) mean consumption
Unless it may appear a satisfactory technique, it has some drawbacks. The main limiting
factors of capacitors are their ESR value and size. The first produces voltage ripples,
consequently. The second may lead to a capacitor size that does not fit within the wireless
device. This inconvenient could be overcome, to a certain extent, by means of distribute the
capacity in several capacitors in parallel or by using the technology of super-capacitors.

4.2.1 Minimum capacitor value

Before to start describing the technical alternatives of power supply systems with passive
components, it is necessary made some insight concerning the minimum capacitance, C,
required to absorb the current peaks at the load, which is a function of the maximum current
consumption peak, its tON and the period. A straightforward way to estimate the C value is
through the following reasoning line. The equivalent circuit of the power supply system
plus the load, (wireless system), is presented in Fig. 7.
The circuit of the figure is valid no matter the capacitor is placed at the load or the battery,
and it is made of:
- The battery of nominal voltage E.
- The distributed resistance between load and battery plus the battery internal resistance,
R2.
- The ideal capacitor, C1.
- The discontinuous load made of a resistance R1 and ideal switch, S1.
- The final charge voltage, V2.
- The minimum discharge voltage, V1.
- ΔV=V2-V1 is the load voltage ripple, Vripple, or the magnitude of capacitive discharge.
Power Supply Architectures for Wireless Systems with Discontinuous Consumption 389

Fig. 7. Un-loaded equivalent circuit of battery, distributed resistance, capacitor and load,
and detail of the load voltage ripple showing the capacitor charge and discharge
Whenever the load, or wireless system, is not activated, the capacitor voltage is equal to V1,
so the capacitor discharge time is a function of RLOAD=VLOAD/IPEAK, through Ec. 2.

⎡ E − V (0 + ) ⎤
tCHARGE = R 2 ⋅ C 1 ⋅ ln ⎢ ⎥ (2)
⎣⎢ E − V (t ) ⎥⎦
Where V1 and V2 are equal to:

V (0 + ) = V 1 = E ⋅ (1 − ε ) − ΔV (3)

V (t ) = V 2 = ⋅E ⋅ (1 − ε ) (4)

Replacing V1 and V2, results:

⎡ E − E ⋅ (1 − ε ) + ΔV ⎤ ⎡ E ⋅ ε + ΔV ⎤
tCHARGE = R 2 ⋅ C 1 ⋅ ln ⎢ ⎥ = R 2 ⋅ C 1 ⋅ ln ⎢ (5)
⎣ E − E ⋅ (1 − ε ) ⎦ ⎣ E ⋅ ε ⎥⎦
Being the capacitor load at VC(0+) = V2, it starts a discharge cycle that last a maximum time
of tON. So, the new equivalent circuit of the load plus the power supply is represented in
Fig. 7. Solving the Thevenin, the circuit is simplified as it shows Fig. 8, being the Thevening
voltage, Ee, equals to:

1 (6)
Ee = E ⋅
R2
R1 + 1

Fig. 8. Equivalent circuit of battery, distributed resistance, capacitor and load over the
capacitor discharge cycle
390 Recent Advances in Wireless Communications and Networks

Fig. 9. Equivalent circuit of battery, distributed resistance, capacitor and load including the
capacitor ESR
In these conditions, the capacitor discharge time is defined with the expression Ec. 7.

⎡ E − V (0 + ) ⎤
t DISCHARGE = t ON = R 1 // R 2 ⋅ C 1 ⋅ ln ⎢ ⎥ (7)
⎣⎢ E − V (t ) ⎥⎦
Where V1 and V2 are equal to:

V (t ) = V 1 = ⋅E ⋅ (1 − ε ) − ΔV (8)

V (0 + ) = V 2 = ⋅E ⋅ (1 − ε ) (9)

Replacing V1 and V2, results:

⎡ Ee − E ⋅ (1 − ε ) ⎤
(10)
t DISCHARGE = t ON = R 1 // R 2 ⋅ C 1 ⋅ ln ⎢ ⎥
⎣ Ee − (E ⋅ (1 − ε ) − ΔV ) ⎦
The mathematical expressions obtained may further complicated by adding to the circuits of
Fig. 7 and Fig. 8 the capacitor ESR, which is a function of the capacitance through the loss
tangent, Fig. 9. The expression that relates the ESR with the capacitance is, approximately:

1 (11)
R ESR = tgδ ⋅
2π ⋅ f ⋅ C 1
Consequently, the total voltage ripple, Fig. 10, is the sum of the one that causes the
capacitive discharge, plus the one produce in the ESR of the capacitor is:

ΔV = Vripple = ΔVESR + ΔVC (12)

Bearing mind the reasoning followed on the previous lines, and replacing Ec. 5 and 12 in Ec.
5, the capacitor discharge time, with its ESR effect, is qual to:

⎡ E ⋅ ε + ΔV ⎤
t CHARGE = (R 2 + R ESR )⋅ C 1 ⋅ ln ⎢ (13)
⎣ E ⋅ ε ⎥⎦
In the same way, replacing in Ec. 10, the discharge time is equal to:

⎡ Ee − E ⋅ (1 − ε ) ⎤
t DISCHARGE = (R ESR + R 1 // R 2 ) ⋅ C 1 ⋅ ln ⎢ ⎥ (14)
⎣ Ee − (E ⋅ (1 − ε ) − ΔV ) ⎦
Power Supply Architectures for Wireless Systems with Discontinuous Consumption 391

This lasts equations estimate the capacitance as a function of the targeted voltage ripple.

Fig. 10. Ideal waveform detail of the load voltage ripple showing the capacitor charge and
discharge, and including the capacitor ESR contribution

4.2.1 At the battery ends

The first place to locate a high-value capacitor is in the battery pack. Fig. 11 shows two
wireless control applications that use a high-value capacitor at the battery terminals. In (a)
the maximum value is limited by the size of the mechanic, it employs two aluminium
organic capacitors of 470 μF in parallel. Meanwhile in (b) the size of the equipment allows
the use of a 33 mF super-capacitor.

(a) (b) (c)

Fig. 11. Pictures of wireless control systems with capacitor place at the battery terminals, (a)
high-value aluminium organic and (b) super-capacitor. (c) Schematic diagram of power
supply system with high-value capacitor at the battery ends
The Fig. 11(c) shows the power supply schematic of a wireless system directly connected to
the battery with a capacitor at the battery ends. The equivalent circuit is made of the
resistive elements of the PCB tracks, connectors, and the equivalent resistance of the battery,
which includes internal resistance, fuse resistance and protection electronics if required.
The waveforms of current and voltage at the ends of the battery represented in the Fig. 12
illustrate the behaviour of this architecture for three capacitors, it could be seen the
following; the current drain from the battery, IBATT, is lower than the load current demand,
ILOAD, the voltage ripple, VBATT at the battery is lower than the ripple at the load, and, at the
instant of battery connexion, the current through the connector is zero.
Therefore, place a super-capacitor or a high-value capacitor in the battery helps to reduce
the space it occupied in the wireless device, but increases the size of the battery pack. Super-
capacitors also presents manufacturing disadvantages as they are not suitable for automatic
surface mount assembly, SMD, because do not withstand a standard lead-free oven
soldering profile.
392 Recent Advances in Wireless Communications and Networks

They also have some technical disadvantages. Whenever the load drains current, it creates a
resistive path between the battery and load, which is made of the battery contact resistance,
ΣRCONN, sense resistance, RSENSE and the distributed serial resistance of the PCB tracks,
RPCB_TRACKS. These increase the voltage drop at the load terminals.

Fig. 12. Load voltage ripple, battery current, capacitor current and current load for
capacitors of 2200, 4700 and 6800 µF at the battery terminals

4.2.2 At the load ends

To prevent voltage drops between the battery and load, super-capacitor or high-value
capacitor should be place as close as possible to the load, as it is depicted in Fig. 13 (a). (b) is
a picture of M2M wireless module with high-value capacitors at the load ends. The picture
illustrates how the capacitance is distributed in several capacitors to eases fit it in the device.
The total capacitance is the sum of four special tantalum capacitor of 1000 µF value each in
parallel. This arrangement, not only gets high-value capacitance (4000 µF), but also reduces
the equivalent ESR, as it is the sum of the ESR resistance of each capacitor in parallel.
The behaviour of this architecture is represented on Fig. 14 and Fig. 15. The first group of
traces shows the input and output voltage and currents for three super-capacitors of 500, 200
and 60 mF respectively. The output voltage, VLOAD, represents the magnitude of the ripple,
which is a function of each capacitor ESR, as theirs ESR value is such that their charge and
discharge could not be appreciated because they never fully discharge. The load current
consumption, ILOAD, is the result of adding battery, IBATT, and capacitor, ICload, currents.
Power Supply Architectures for Wireless Systems with Discontinuous Consumption 393

Capacitors

ILOAD

(a) (b)
Fig. 13. (a) Schematic diagram of high-value capacitor at the load terminals. (b) Picture
Detail of an M2M application with tantalum capacitors at the load terminals (4000 µF)

Fig. 14. Voltage ripple in the load, V(LOAD), battery, capacitor and load current with super-
capacitors of 60, 200 and 500 mF
The second group reproduces the same waveforms when three electrolytic capacitors of
2200, 4700 and 6800 µF are used instead of super-capacitors. The voltage ripple is depicted
as VLOAD in the first trace; it shows the charge and discharge of the capacitors, and the
contribution of theirs ESR to the voltage ripple.
The behaviour represented in Fig. 13 and Fig. 15 could be summarizing as follows; the
current through the battery is lower than the current drain by the load, and voltage drops at
the load ends are further diminished because most of the energy demanded is extracted
directly from the capacitor. Unfortunately, place high-value capacitor at the load ends
394 Recent Advances in Wireless Communications and Networks

causes a high current peak each time the battery is replaced, Fig. 16. Eventually, this current
surge may destroy or damage the connectors after a certain number of battery replacements.

Fig. 15. (a) Voltage ripple in the load, V(LOAD), (b) battery, (c) capacitor and (d) load
current with capacitors of 2200, 4700 and 6800 µF

Fig. 16. Instantaneous voltage and current at the connexion of a 60 mF with the battery
Power Supply Architectures for Wireless Systems with Discontinuous Consumption 395

4.2.3 LC network
This alternative is based on a LC network made of a series inductor and followed by a
parallel capacitor, at is shown in the Fig. 17(b), and it is seldom used in older cellular
terminals and radio modules with discontinuous consumption. The LC network of
Fig. 17(a), from the frequency point of view, constitutes a low-pass filter, although, it also
should be analyzed in the time domain to completely characterize its behaviour.
The series inductor limits the capacitor charge current; this fact smoothes the input current
fluctuations, but it is required a minimum value of inductance and capacitance to be
effective. The technology constrains the inductance values available through size and
current parameters, which may make not feasible the required values, and consequently, the
impact of an LC network is reduced to a small smooth of the transitions slopes. The table 3
illustrates SMD inductors availability of inductance higher than 680 µH that withstand
currents above 2 A.

Pogo-pin
Battery
Inductance connector
(22 µH)

(a) (b)
Fig. 17. (a) Schematic detail of the power supply system and load with LC network.
(b) Detail of and LC network in the power supply of a cellular terminal

Core Manufacturer Code L (µH) ISAT(A) RDC(mΩ) LxWxH (mm)

Close Bourns SRR1240-470M 2 2 135 12,5x12,5x4,0
Close Vishay IHLP-4040DZ-11 0,5 0,5 270 11,3x11,5x4,0
Close Coilcraft MSS1246T-104 1,84 1,84 210 12,3x12,3x4,8
Open Coilcraft DO3340P-104 2,5 2,5 220 12,9x9,4x11,4
Open Coilcraft DO5040H-684 2 2 780 18,5x15,2x12,0
Open Pulse PF0504.104NL 2,5 2,5 153 18,5x15,2x11,4

Table 3. Comparison between inductor technologies with and without shielding

Furthermore, it must be consider that and inductor generates lines of EM fields that closes
trough the air, unless the inductor posses a magnetic shield. As the EM field generated on
the inductance has the same low frequency pattern of the discontinuous consumption, it
implies EM interferences that could not be avoided unless shielded inductances are used.
The table 3 provided also shows, for the same value and manufactures, the differences
between open and shielded inductors. For example, a 100 µH shielded inductor implies a
40% increase of volume and 660 mA reduction of maximum current withstand. The space
occupied by the inductor is increases by the one the capacitor requires, and in spite of
capacitor may be smaller that its counterpart architectures with a single capacitor, it must be
396 Recent Advances in Wireless Communications and Networks

bigger that hundreds of microfarads to make the LC network effective in reducing the
voltage output ripple and the input current fluctuation.
As a final remark, an LC network may generate oscillations, periodic or damped, which
depend on the values of the LC network and the equivalent load resistance, as it constitutes
a RLC network affected by a pulsed signal. This also means that the input current could have
negative values, which implies a current send back to the battery, i.e., a charging current.
The behaviour of the LC network is represented in the following picture for three LC
combinations. The first group of waveforms, Fig. 18, represents the load current, ILOAD. The
trace of IBATT shows, how the inductance L charges over the consumption pulse, and how it
delivers the energy to the capacitor over the inactive part of the cycle. At the same time,
ILOAD illustrates the capacitor providing current to the load meanwhile the pulse current last.
When the current pulse ends the capacitor is being charged through the inductance until the
next pulse came. The waveform at the capacitor ends, VC, is equal to the load voltage, VLOAD.
This voltage shows the charge and discharge cycles of the capacitor which follows the
current cycles of the load, ILOAD. On Fig. 19 could be appreciated the same behaviour for
different pairs of LC.

Fig. 18. Current and voltage for the power supply of a wireless system with LC filtering

5. Wireless systems powered through converters

If the power source requires a conditioning of its voltage level to those of the wireless
systems internal electronics, in terms of discontinuous load, it is required a different
approach to the one used for battery powered wireless systems (B. Arbetter et al., 2006). An
example of this kind of wireless devices is the communication modules, called M2M
(machine to machine). Whenever the power supply system employs voltage converters, no
Power Supply Architectures for Wireless Systems with Discontinuous Consumption 397

Fig. 19. Load voltage ripple, battery and load current for a capacitor of 470 µF and de 22µ,
100µ and 1mH inductors
matter if they are linear or switched, the discontinuous consumption produces current
peaks at the input of the power supply. A current peak causes voltage drops in the load
input voltage due to the internal resistance of the power lines and the power source
connectors. If linear regulation is used, the regulators input and output current peaks are
equals. On the other hand, if the converter is switched, the bigger the differences between
input and output voltage, the lower the current peaks are. Nevertheless, independently of
the type of power converter used, it must be design to deliver the maximum current peak.
To cope with the effects of a discontinuous consumption, in M2M modules for wireless
communications, manufactures such as Wavecom, Sony Ericsson, Telit, Freescales or
Siemens, recommend on theirs application notes employ high value and size capacitors,
which lead to, in many design conditions, a size of recommended capacitor bigger or of the
same size as the M2M module itself, at is shown in Fig. 4.
Bearing in mind what was exposed above, the technical alternatives of power supply
systems with converters for wireless systems are detailed in the following headings.

5.1 Capacitor calculation

Before starting with the analysis of power supply architecture, it is required to define, on a
first step, the maximum mean current that the load demands to the power source as a
function of the period and duty cycle. The value obtained is used to program the current
limit of the power converter. This current corresponds to the maximum current that the
power converter drives to the capacitor. If it is set that the current limit must be tON/T times
the maximum current peak, IP or IPEAK, that the load demands, this current limit is 1/N, (1/8
for a GSM cellular terminal). With this pattern the maximum input current of the power
supply system, IIN, is equal to the mean value of the maximum peak current consumption
over a period. The voltage converter provides the energy that the load demands maximum
398 Recent Advances in Wireless Communications and Networks

consumption. Consequently, unless the maximum mean current will never overcome, IIN
follows the load consumption fluctuations, at it is described in the following expression:

I IN = I PEAK ⋅ t ON /T (15)
Where,
- tON o tdischarge is the duty cycle of consumption, equal to the capacitive discharge time.
- T is the consumption period.
- IPEAK is the maximum peak consumption over a period.
Thus, the required capacitance is obtained through the following reasoning. Being the
current through the capacitor:

I = C ⋅ dV / dt (16)
And considering that, on an ideal situation, the charge and discharge of a capacitor is lineal.
This is feasible as the maximum drive current of the voltage converted is limited to a fixed
value. Thus, doing the differential voltage equal to the voltage increment, Vripple, and the
differential time equal to the time increment, tON, the current is equal:

I = C ⋅ ΔV / Δt (17)
Solving for the capacitance value, and considering that the voltage ripple ΔVC, for an ideal
capacitor, has only a capacitive discharging contribution, the capacitance results:

C = I ⋅ t ON / ΔVC (18)

For example, if the mean load current is 250 mA, for a maximum peak current of 2 A, and a
period time of 4,64 ms, the capacitance value, (without including the effect of its ESR), for a
maximum load voltage ripple of 0,4 V, is:

C = 0,250 A· 580 µs / 0,4 V ≅ 3600 µF (19)

5.2 Constant input current power supply

Once it is stated that a high-value capacitor smoothes current and voltage transitions and
reduces its magnitude, but does not maintain constant the voltage excursion around the
nominal voltage values of the power source. The first improvement could be add a voltage
regulator to the capacitor, Fig. 20. The power supply system of Fig. 20 is made of a lineal or
switched regulator with a fixed current limit, plus a high-value capacitor close to the
wireless system load.
The Fig. 20 includes the connector and power lines resistance, RIN. Following this equivalent
resistance it is placed the current limited voltage regulator. The current limit of the regulator
must be adjusted, approximately, to the maximum peak current averaged by N for a period.
For an EGSM cellular terminal, this current is equal to the number of time slots used for
transmission, IPEAK/8. With this power supply architecture the input current, IIN, always has
a value close to the average current consumption. At any time, the regulator is able to
provide the maximum power that the wireless system load may demand. For example, on a
M2M GSM module, the input current, IIN, varies following the consumption fluctuations,
and never overcomes the maximum average consumption, IPEAK/8. The RSENSE resistance is
used to measure the current that the capacitor drains or supplies, and, at the same time,
limits the maximum current that the converter could provide. If the wireless systems
Power Supply Architectures for Wireless Systems with Discontinuous Consumption 399

requires electronics to monitoring o control the TX power it could be done by means of a

current sensor, RSENSE, this element increase the voltage drop and must be consider.
Unless linear conversion is an option, a switched regulator is a better solution, meanwhile
the electromagnetic fields generated are under control. Linear regulation reduces the power
efficiency of the supply system as the input voltage of the power source may vary over a
wide range of values. As a design rule, whenever the magnitude difference between input
and output voltage are not relevant a linear regulator could be used.

Fig. 20. Block diagram power supply system with regulated input and output capacitor
The high-value capacitor is placed at the output of the voltage converter. The capacitor
stores the energy that requires the discontinuous load of the wireless system over its active
time, for GSM this time corresponds to a time slot. The capacitor charge and discharge
produces voltage ripple, VC, at its terminals. Fig. 21(a) shows this voltage ripple and the
regulator input current, when the load current peak is maximum, i.e., when the wireless
system transmits at maximum power, and if the capacitor charge and discharge is produced
at constant current.
To ease the computation of the voltage ripple an ideal capacitor is used, and two ideal
sources of charge and discharge, that represents the converter and the load, respectively.
Thus, the voltage ripple is a function of the current peak and capacitor value. If the capacitor
is not ideal, the capacitor ESR introduces an extra voltage drop that must be added to the
total voltage ripple, as it could be seen in the VC detail of Fig. 10.
The expression for a generic wireless system with discontinuous consumption could be
written as states Ec. 20, being T the consumption period, and Δt or tON the time the
consumption last.

dVC IC I
IC = C ⋅ , si I C = Cte ⇒ ΔV = ⋅ Δt = C ⋅ tON (20)
dt C C
Over the capacitor discharge the instantaneous voltage, VC, could be represent as Ec. 21:

I PEAK − I IN I −I
Discharge : ΔV = ⋅ Δt = PEAK IN ⋅ tON (21)
C C
And the capacitor charging time, VC(t), is:

I IN I
Charge : ΔV = ⋅ (T − tON ) ⇒ ΔV = IN ⋅ 7 ⋅ Δt(GSM ) (22)
C C
400 Recent Advances in Wireless Communications and Networks

Fig. 21. PA GSM cellular terminal voltage and current waveforms for three power levels,
(a) maximum and (b) (c) intermediate
At the load terminals, there is a ripple proportional to the capacitor discharge, plus the
voltage drop at the current sensor, RSENSE. If the sense resistance is small enough, VC is
almost equals to VLOAD. The Fig. 21(a) shows the shape of the voltage ripple, VLOAD, at the
load ends, for different TX output power levels or discontinuous consumption.
The circuit of Fig. 20 allows to reduce the current peaks at the wireless systems input to the
average value, IIN, as it could be seen in Fig. 21(a). Considering the two current situations
represented in the figure it could be deduce that:
- When the load is maximum, (PA transmits at its maximum power), the current
consumption through the input is constant, IIN, Fig. 21(a).
- If the power transmission is reduced, so the current consumption, and the current
through the output, IIN, is discontinuous but its maximum peak value never overcome
the average current consumption for the maximum power, Fig. 21(b) and (c).
Consequently, on a GSM terminal the input current never reach a value greater than 1/8,
(tON/T for a generic application), of the current peak demanded by the wireless system at
maximum consumption rating.

5.3 Constant current and constant input current power supply

The Constant Input Current power supply architecture has voltage ripple at the load which
varies as a function of the wireless device current consumption, Fig. 20. Unless the load
voltage ripple is lower than the achieved with direct connection, this ripple could be
reducing further by means of the architecture depicted in Fig. 22. This figure represents a
Constant Current and Constant Input Current power supply system.
The architecture of Fig. 22 stabilizes the capacitor voltage, VC, by means of a second
regulator. This second converter could be lineal because as its input voltage range is within
the same range as the output voltage, so its dropout or voltage drop in the active component
Power Supply Architectures for Wireless Systems with Discontinuous Consumption 401

is low and will not affect the efficiency of the overall power systems. The purpose of the
output converter is suppress the capacitor voltage ripple. Fig. 23 shows the load voltage,
VLOAD, and its drive current, ILOAD, obtained when using this power supply architecture. The
second voltage conversion element absorbs the voltage fluctuation of charge and discharge
of the capacitor in the power supply voltage, allowing a capacitor of lower value.

Fig. 22. Block diagram of constant current and constant input current power supply

Fig. 23. PA GSM cellular terminal voltage and current waveforms for power levels,
(a) maximum and (b) intermediate
If the ripple requirements for input current and load voltage are quite restrictive, this
alternative reduces its effects and, if there is also current and efficiency restrictions the
linear, converter could be replace by a standard switched one of any manufacturer, for
402 Recent Advances in Wireless Communications and Networks

example, a MAX1678. This device has some advantage as it is specially design for GSM and
UMTS, with the remark that is only efficiency in boost mode.

5.4 Summary
The ideas exposed are summarized in the Fig. 24 and Table 5. Fig. 24 provides a graphic
comparison between input and output voltages and currents for the different architectures
of power supply to cope with wireless systems discontinuous consumption.
The first group of waveforms, Fig. 24(a), represent the most unfavourable conditions, where
the power source and the wireless systems are directly connected. The input and output
current and voltages exhibit ripple with abrupt slopes.
Fig. 24(b) shows the alternative of a high value capacitor place in the load. Its charge and
discharge smoothes the slopes of input and output current and voltages.

Fig. 24. Comparison between voltage and current waveforms of power supply architectures
for wireless systems with discontinuous consumption
On Fig. 24(c), the alternative represented is an LC network; the graphics unveils that the
input current could become negative over part of the consumption cycle, also presents over-
damp, in voltage and current, that may produce oscillation.
The waveforms for regulated supply systems with fixed input current limit and a high-value
capacitor are represented in Fig. 24(d). The fixed input current limit is the average value of
the maximum peak current consumed in the load.
On Fig. 24(e), it is represented the voltage and current for a Constant Current and Constant
Input Current Power Supply.
The Table 5 summarizes those parameters to balance in order to choose the more suitable
power supply architecture for wireless systems. The table sorted the architectures in two
columns, one for battery powered systems and the second for systems that require output
voltage level conditioning. From the data shown, it could be inferred that: the preferred
Power Supply Architectures for Wireless Systems with Discontinuous Consumption 403

With passive components With voltage converters

Constant Constant
Direct C in the C in the LC
Current Output
connexion battery load Network
Limit Voltage
Medium Medium Medium Medium
VIN ripple High High
/Low /Low /Low /Low
Medium Medium
VOUT ripple High Medium High Negligible
/Low /Low
Overvoltage No No No Yes No No
Medium Medium
Efficiency High High Alto High
/Low /Low
EMC
Poor Average Average Poor Average Average
behaviour
Complexity Low Low Low Medium Medium Medium
Medium Medium Medium Medium Medium
Size Small
/High /High /High /High /High
IIN pulses High Medium Medium High Low Low
IIN peaks Yes Yes Yes Yes No Low
Table 5. Comparison between power supply architectures for wireless systems
architecture for battery powered systems is a high-value capacitor in the load terminals. If
the wireless device needs voltage conversion, the recommended alternative is the Constant
Current and Constant Input Current power supply system, made of a double voltage
conversion and a high-value capacitor between the two regulators.

6. Conclusions
Wireless systems and communication electronics have their functionality and performance
conditioning by the type of power consumption they present. This chapter highlights the
effects of discontinuous consumption on wireless systems. It also provides keys and
guidelines to identify the phenomena, and how they restrict wireless device functionality.
The effects identified are:
- Power supply voltage drops produced by the current through the equivalent series
resistance between source and load.
- Existence of variable electromagnetic fields, generated by the discontinuous current flux
that affects the EMC performance of the electronics
Bearing in mind these two issues, the characterization of discontinuous consumption is
made through the study of power supply systems suitable for such type of consumption.
This is the reason why it has been proposed and analyzed two generic types of power
supply systems for wireless systems that encompass all: systems directly battery powered,
and systems that required supply voltage levels that differ from those provided by power
source.
Commonly, power supply systems are dimensioned for the required output voltage and the
maximum peak power consumption, which do not guaranty wireless systems proper
operation whenever their consumption or presented load is discontinuous. Therefore, the
effects of discontinuous current consumption and its solution are presented. The study
analyses two power supply scenarios, direct connection between source and load through
404 Recent Advances in Wireless Communications and Networks

passive components, and voltage regulation. In such conditions, the most common
architectures could be restricted to:
- High-value capacitor in the load or the source.
- Voltage converter with input current limit plus high-value capacitor al the load ends.
- Current Limit and Constant Input Current power supply system made of two
converters, plus a high-value parallel capacitor between both converters.
Comparing and analyzing the different architectures studied, summarized in Table 5, could
be concluded that suitable architectures are:
- A high-value capacitor in the load or power source.
- Current Limited and Constant Input Current power supply system if it is required
conditioning the voltage levels.
Unless two regulators architecture may appear cumbersome, it is interesting highlight that
increasing the switching frequency of DC-DC converters make feasible the use of such
regulators, as high-frequency allows the reduction the size of the inductor and capacitors
required.
In spite of there were identified valid architectures, the effects of the discontinuous
consumption are not eliminated completely, as shows Fig. 24. None of the architectures gets
an input current drain from the source constant.
If it is placed a high-value capacitor close to the load, there is a high current surge at the
connexion instant between power source and wireless system. This current surge damages
the connector reducing its lifetime. Moreover, high-value capacitor or super-capacitor may
not be feasible, for they do not fit in the mechanics. Consequently, a trade-off between size
and maximum available capacity and performance must be achieved.
If Current Limited and Constant Input Current power supply is used, the input current is
discontinuous for consumptions below the maximum, Fig. 24(e). The direct consequence is
variable current flux that produces electromagnetic fields.

7. References
B. Arbetter, R. Erikson & D. Maksimovic, DC-DC converter design for battery-operated systems, in
Proceeding of IEEE Power Electronic Specialist Conference, 1995, vol. 1, pp. 103-109.
B. Sahu & G.A. Rincon-mora, A Low Voltage, Non-Inverting, Dynamic, Synchronous Buck-Boost
Converter for Portable Applications, IEEE Transactions on Power Electronics, vol. 19,
no. 2, Feb. 2004, pp. 443-452.
E. Dahlman, H. Ekström, A. Furuskär, Y. Jading, J. Karlsson, M. Lundevall & S. Parkvall, The
3G Long-Term Evolution - Radio Interface Concepts and Performance Evaluation, IEEE
Vehicular Technology Conference (VTC) , Melbourne, Australia, May 2006.
K. Fazel & S. Kaiser, 2008, Multi-Carrier and Spread Spectrum Systems: From OFDM and MC-
CDMA to LTE and WiMAX, 2nd Edition, John Wiley & Sons, ISBN 978-0-470-99821-2.
M. I. Montrose, 1996, Printed circuit Board Design techniques for EMC Compliance, Piscataway,
NJ, IEEE Press.
R. W. Erikson, 1997, Fundamentals of Power Electronics, Chapman and Hall, 1st ed., New
York.
Saft Rechargeable Battery Systems, 2008, Rechargeable Battery Systems Handbook, Available
from: http://www.saftbatteries.com.
W. Schroeder, July 2007, Direct Battery Connection Benefits Portable Designs, White Paper,
Semtech.
 19

Wireless Sensor Networks in

Smart Structural Technologies
Yang Wang1 and Kincho H. Law2
1Georgia Institute of Technology, Atlanta, Georgia,
2Stanford University, Stanford, California,
USA

1. Introduction
Recent advances in wireless communication, as well as embedded computing, have opened
many new exciting opportunities for wireless sensor networks. Miniature and low-cost
wireless sensors are expected to become available in the next decade, offering countless
possibilities for a wide range of applications. Among them is smart structural technology, an
active research domain that holds significant promise for enhancing infrastructure
management and safety. A smart structure refers to a specially equipped structure (e.g.
buildings, bridges, dams, etc.) that can monitor and react to surrounding environment and
the structure’s own conditions, in a pre-designed and beneficial manner.
Smart structural technology encompasses at least two major fields, i.e. structural health
monitoring and structural control. A structural health monitoring (SHM) system measures
structural responses and predicts, identifies, and locates the onset of structural damage, e.g.
due to deterioration or hazardous events. Structural sensors, such as micro-electro-
mechanical system (MEMS) accelerometers, metal foil strain gages, fiber optic strain sensors,
among others, have been developed and employed to collect important information about
civil structures that could be used to infer the safety conditions of the structure (Farrar, et al.
2003, Sohn, et al. 2003, Chang 2009). On the other hand, structural control technology aims to
mitigate adverse effects due to excessive dynamic loads (Yao 1972, Soong 1990, Housner, et
al. 1997, Spencer and Nagarajaiah 2003).
Structural monitoring and control both involve acquiring response data in real time. In
order to transmit real-time data, coaxial cables are normally employed as the primary
communication link. Cable installation is labor intensive and time consuming, and can cost
as much as $5,000 US dollars per communication channel (Çelebi 2002). To eradicate the
high cost incurred by the use of cables, wireless systems could serve as a viable alternative
(Straser and Kiremidjian 1998). Wireless communication standards, such as Bluetooth (IEEE
802.15.1), Zigbee (IEEE 802.15.4), Wi-Fi (IEEE 802.11b), are now mature and reliable
technologies widely adopted in many industrial applications (Cooklev 2004). Potential
applications of wireless technologies in structural health monitoring have been explored by
a number of researchers, as reviewed by Lynch and Loh (2006). By incorporating a control
interface, wireless sensors have also been extended to potentially command control devices
for structural control applications (Wang, et al. 2007b).
406 Recent Advances in Wireless Communications and Networks

Compared to cable-based systems, wireless structural monitoring and control systems have
a unique set of advantages and technical challenges. Besides the desire for portable long-
lasting energy sources, such as batteries, reliable data communication is a key issue for
implementation. The purpose of this chapter is to review the important issues and metrics
for adopting wireless sensor networks in smart structural systems. In a structural health
monitoring system, sensors are typically deployed in a passive manner, primarily for
measuring structural responses. Structural control systems, on the other hand, need to
respond in real time to mitigate excess dynamic response of structures. Typical feedback
control systems require real-time information and measurements to instantly determine
control decisions. Although structural monitoring and control applications pose different
needs and requirements, efficient information flow plays a key and critical role in both
implementations. For example, the transmission latency and limited bandwidth of wireless
devices can impede real-time operations as required by control or monitoring systems. In
addition, communication in a wireless network is inherently less reliable than that in cable-
based systems, particularly when node-to-node communication range lengthens. These
information constraints, including bandwidth, latency, range, and reliability, need to be
considered carefully using an integrated system approach and pose many challenges in the
selection of hardware technologies and the design of software/algorithmic strategies.
The chapter adopts a previously designed wireless structural monitoring and control system
as an example to discuss various intriguing research challenges (Wang, et al. 2005, Wang
2007). The system contains wireless sensing and control units that can be used for both
wireless structural health monitoring and real-time feedback structural control.
Modularized software is designed for the wireless units, so that application programs can be
conveniently embedded into the units. The architectural details of the wireless structural
monitoring and control system are presented. For different structural applications, including
health monitoring and control, special communication protocols have been designed to
efficiently manage the information flow among the wireless units. Finally, laboratory and
field validation tests have been conducted to assess the performance of the prototype
wireless structural monitoring and control system.

2. Design and implementation of a wireless sensing and control unit

Sensing and control units are the fundamental components of a wireless monitoring and
control system. The prototype wireless unit is designed in such a way that the unit can serve
as either a sensing unit (i.e. a unit that collects data from sensors and wirelessly transmits
the data), a control unit (i.e. a unit that calculates optimal control decisions and commands
control devices), or a unit for both sensing and control. Fig. 1 shows the functional diagram
of the prototype wireless sensing and control unit. The wireless sensing unit shown in the
top part of Fig. 1 serves as the core component, with which off-board modules for signal
conditioning and signal generation can be easily incorporated.

2.1 Hardware and software of the wireless sensing and control unit
The wireless sensing unit consists of three functional modules: sensor signal digitization,
computational core, and wireless communication. The sensing interface converts analog
sensor signals into digital data, which is then transferred to the computational core through
a high-speed Serial Peripheral Interface (SPI) port. Besides a low-power 8-bit Atmel
ATmega128 microcontroller, external Static Random Access Memory (SRAM) is integrated
Wireless Sensor Networks in Smart Structural Technologies 407

Wireless Sensing Unit

Computational Core

128kB External Wireless

SRAM Communication
UART
Sensor Signal CY62128B Port
Parallel Wireless Transceiver:
Digitization SPI 20kbps 2.4GHz
Port Port
8-bit Micro- 24XStream, or 40kbps
4-channel 16-bit 900MHz 9XCite
controller
Analog-to-Digital ATmega128
Converter ADS8341
Structural
Sensors SPI
Port
Sensor Signal Control Signal Structural
Conditioning Generation Control Devices

Amplification, Filtering, 16-bit

and Voltage-offsetting Digital-to-Analog
Converter AD5542

Fig. 1. Functional diagram detailing the hardware design of the wireless sensing unit.
Additional off-board modules can be interfaced to the wireless sensing unit to condition
sensor signals and issue control commands
with the computational core to accommodate local data storage and analysis. The
computational core communicates with a wireless transceiver (24XStream or 9XCite models
currently provided by Digi International) through a Universal Asynchronous Receiver and
Transmitter (UART) interface. The auxiliary sensor signal conditioning module assists in
amplifying, filtering, and offsetting analog sensor signals prior to digitization. The auxiliary
control signal generation module offers an interface through which the wireless sensor can
send analog control commands to structural control devices. Hardware design of the
wireless unit and auxiliary modules have been described in details elsewhere (Wang, et al.
2005, Wang 2007, Wang, et al. 2007a). The key parameters of the prototype wireless sensing
unit are summarized in Table 1. Peer-to-peer communication among wireless units is
supported for collaborative data analysis.
In order to manage the hardware components in a wireless sensing unit, software modules
are implemented and embedded in the ATmega128 microcontroller. For the ATmega128
microcontroller, software can be written in a high-level programming language, such as C,
compiled into binary instructions, and loaded into the non-volatile flash memory of the
microcontroller. When the wireless unit is powered on for normal operation, the
microcontroller automatically starts executing the embedded instructions. The software
design of the wireless sensing and control units follows the hierarchical structure as shown
in Fig. 2. At the bottom level are the software modules that manage the basic peripherals of
the microcontroller. The middle layer consists of software modules that manage other
onboard hardware components. Specific software modules for structural health monitoring
and control are implemented in the top level application layer.
408 Recent Advances in Wireless Communications and Networks

Design Parameter Specification

Computing Core
8-bit RISC1 architecture, up to 16MIPS2 throughput at
Microcontroller
16MHz
Flash Memory 128K bytes
Internal SRAM3 4K bytes
External SRAM 128K bytes
EEPROM4 4K bytes
Power Consumption 30mA active, 55μA standby
Wireless Transmission 9XCite 24XStream
Operating Frequency ISM 902-928 MHz ISM 2.4000 - 2.4835 GHz
Data Transfer Rate 38.4 kbps 19.2 kbps
Communication Up to 300' (90m) indoor, Up to 600' (180m) indoor, 3
Range 1000' (300m) at line-of-sight miles (5km) at line-of-sight
Power Consumption 55mA transmitting, 35mA 150mA transmitting, 80mA
receiving, 20μA standby receiving, 26μA standby
Sensing Interface 4 channels, 16-bit, up to 100 kHz
Control Interface 1 channel, 16-bit, up to 1 MHz
Physical Size 10.2cm × 6.5cm × 4.0cm
Table 1. Key parameters of the wireless sensing unit

Structural Health Monitoring Structural Control

Application Autoregressive Fast Fourier Linear Quadratic Kalman Damper

Software Analysis Transform Regulator Control Filter Hysteresis Model

Computing /
Communication Sensing Actuation
Data Storage
External Static Analog-to- Digital- to-
System Wireless Digital(A2D) Analog (D2 A)
Random Access
Management Communication Conversion Conversion
Memory (SRAM)

Microcontroller Universal Asynchronous Internal Static Serial

Timer
Peripherals Receiver and Random Access Peripheral
Interrupt
Transmitter (UART) Memory (SRAM) Interface (SPI )

Fig. 2. Three-layer software architecture for the ATmega128 microcontroller in the wireless
sensing and control unit

1 RISC: reduced instruction set computer.

2 MIPS: million instructions per second.
3 SRAM: static random access memory.

4 EEPROM: electrically erasable programmable read-only memory.

Wireless Sensor Networks in Smart Structural Technologies 409

As shown in Fig. 2, the lowest level of the embedded software manages the peripherals of
the ATmega128 microcontroller and serves as the fundamental modules to support the
functions of other hardware components. Embedded modules include: timer interrupt
functions, byte-by-byte communication through the UART and SPI ports, and internal
memory management. The timer interrupt service is implemented to achieve a constant time
step for sensor data sampling. The interrupt function is also a powerful feature that allows
the software to momentarily pause an executing task (such as data processing or wireless
communication) to sample data from the sensing interface according to a precise timing
schedule. Immediately after servicing the sensing interface, the paused task is resumed and
the program continues its execution. This timer interrupt feature is utilized to implement
continuous data streaming from multiple wireless sensing units, where sensor data
sampling has to occur at a constant sampling step amidst the execution of the wireless
communication or data interrogation program. In effect, the software supports concurrency
thereby allowing multiple software tasks to execute at the same time.
Building on top of the microcontroller peripherals are the software drivers that manage
other hardware components in the wireless unit. Utilizing the UART peripheral, the
wireless communication driver provides the following functions interfacing the
microcontroller with the wireless transceiver: 1) reading or setting the radio parameters of
the attached wireless transceiver; 2) sending or receiving data through the wireless
transceiver; 3) implementing the state machine representing the wireless communication
protocol. A driver module is implemented to manage the 128kB external Static Random
Access Memory (SRAM). This module includes functions to enable and disable the external
SRAM, as well as functions that allow access to the lower 64kB half or higher 64kB half of
the memory chip. The other two hardware drivers, the A2D and the D2A modules, manage
the interfaces with the structural sensors and control devices. The ATmega128
microcontroller provides only one SPI port, which is shared by both the A2D converter
(ADS8341) for sensing and the D2A converter (AD5542) for control. The A2D module
commands the ADS8341 to convert a 0 to 5V analog sensor signal into a 16-bit integer.
Knowing the sensitivity and offset of the sensor signal, the microcontroller can then
compute a floating-point number quantifying the physical parameter being measured by the
sensor. Conversely, the D2A module takes a floating-point number between -5V and 5V as
input, converts the number into a 16-bit integer, and pushes the integer to the AD5542 to
output the corresponding control voltage signal.
Utilizing the hardware drivers for communication, computing, sensing, and control,
software can be developed to support structural health monitoring and control applications.
A number of engineering algorithms, such as Fast Fourier Transform (FFT), autoregressive
(AR) analysis, linear quadratic regulator (LQR) control, and Kalman Filter, have been
implemented and embedded in the wireless units. The ability to execute embedded
application software allows the wireless sensing units to make and execute decisions.
Onboard data processing also helps save energy resources (i.e. preserving limited battery
power) by reducing wireless transmission of large amounts of raw sensor data. With the
application software executing in the wireless unit, each unit acts as an autonomous node in
a wireless monitoring and control network. This architecture of distributed sensing and
control represents a new paradigm in structural health monitoring and control, as
opposed to traditional centralized systems, where data are processed in a centralized
location.
410 Recent Advances in Wireless Communications and Networks

2.2 Communication constraints

As noted in Table 1, the sensing unit is designed to support two wireless transceivers: 900-
MHz 9XCite and 2.4-GHz 24XStream (MaxStream 2004, MaxStream 2005). This dual
transceiver support allows the wireless sensing and actuation unit to operate in different
regions around the world. Wireless communication poses four major constraints to the
information flow within a structural monitoring and control network: bandwidth, latency,
reliability, and range. It is thus important to assess the communication constraints of the
transceivers.

Data packet sent from

ATmega128 to 24XStream

Sending Unit time

TLatency TUART

Receiving Unit time

Data packet coming out of 24XStream

and going into ATmega128

Fig. 3. Three-layer software architecture for the ATmega128 microcontroller in the wireless
sensing and control unit
Bandwidth and latency are about the timing characteristics of the communication links.
Bandwidth refers to the data transfer rate once a communication link is established. Using
the MaxStream 24XStream transceiver as an example, the anticipated transmission time for a
single data packet is illustrated in Fig. 3. The transmission time consists of the communication
latency, TLatency, of the transceivers and the time to transfer data between the microcontroller
and the transceiver using the universal asynchronous receiver and transmitter (UART)
interface, TUART. Assume that the data packet to be transmitted contains N bytes and the
UART data rate is TUART bps (bits per second), which is equivalent to RUART /10 bytes per
second, or RUART /10000 bytes per millisecond. It should be noted that the UART is set to
transmit 10 bits for every one byte (8 bits) of sensor data, including one start bit and one
stop bit. The communication latency in a single transmission of this data packet can be
estimated as:

10000N (ms) (1)

TSingleTransm = TLatency +
RUART
In the prototype wireless sensing and control system, the setup parameters of the 24XStream
transceiver are first tuned to minimize the transmission latency, TLatency. Then experiments
are conducted to measure the actual achieved TLatency, which turns out to be around
15±0.5ms. The UART data rate of the 24XStream radio, RUART, is selected as 38400 bps in the
implementation. For example, if a data packet sent from a sensing unit to a control unit
contains 11 bytes, the total time delay for a single transmission is estimated to be:

10000 × 11
TSingleTransm = 15 + ≈ 17.86 (ms) (2)
38400
Wireless Sensor Networks in Smart Structural Technologies 411

This amount of latency typically has minimal effect in most monitoring applications, but has
noticeable effects to the timing-critical feedback control applications. This single-
transmission delay represents one communication constraint that needs to be considered
when calculating the upper bound for the maximum sampling rate of the control system. A
few milliseconds of safety cushion time at each sampling step are a prudent addition that
allows a certain amount of randomness in the wireless transmission latency without
undermining the reliability of the communication system. Although the achievable
transmission latency, TLatency, is around 15ms for the MaxStream 24XStream transceiver, it
can be as low as 5ms for the 9XCite transceiver. This lower latency makes the 9XCite
transceiver more suitable for real-time feedback control applications compared with the
24XStream transceiver. However, the 9XCite transceiver may only be used in countries and
regions where the 900MHz band is for free public usage, such as the North America, Israel,
South Korea, among others. On the other hand, operating in the 2.4GHz international ISM
(Industrial, Science, and Medical) band, the 24XStream transceiver can be used in most
countries in the world.
The other two constraints, reliability and range, are related to the attenuation of the wireless
signal traveling along the transmission path. The path loss PL (in decibel) of a wireless
signal is measured as the ratio between the transmitted power, PTX [mW] , and the received
power, PRX [mW] (Molisch 2005):

PTX [mW]
PL [ dB] = 10log10 (3)
PRX [mW]

Path loss generally increases with the distance, d, between the transmitter and the receiver.
However, the loss of signal strength varies with the environment along the transmission
path and is difficult to quantify precisely. Experiments have shown that a simple empirical
model may serve as a good estimate to the mean path loss (Rappaport and Sandhu 1994):

PL( d) [ dB] = PL ( d0 )[dB] + 10n log10 ⎛⎜ d ⎞⎟ + Xσ [ dB] (4)

⎝ d0 ⎠
Here PL ( d0 ) is the free-space path loss at a reference point close to the signal source (d0 is
usually selected as approximately 1 meter). Xσ represents the variance of the path loss,
which is a zero-mean log-normally-distributed random variable with a standard deviation
of σ. The parameter n is the path loss exponent that describes how fast the wireless signal
attenuates over distance. Basically, Eq. (4) indicates an exponential decay of signal power:

−n
PRX [ mW ] = P0 [ mW ] ⎛⎜ d ⎞⎟ (5)
⎝ d0 ⎠
where P0 is the received power at the reference distance d0. Typical values of n are reported
to be between 2 and 6. Table 2 shows examples of measured n and σ values in different
buildings for 914 MHz signals (Rappaport and Sandhu 1994).
A link budget analysis can be used to estimate the range of wireless communication
(Molisch 2005). To achieve a reliable communication link, it is required that

PTX [dBm]+AG[dBi] ≥ PL ( d )[dB] + RS[dBm] + FM[dB] (6)

412 Recent Advances in Wireless Communications and Networks

where AG denotes the total antenna gain for the transmitter and the receiver, RS the receiver
sensitivity, FM the fading margin to ensure quality of service, and PL(d) the realized path
loss at some distance d within an operating environment. Table 3 summarizes the link
budget analysis for the 9XCite and 24XStream transceivers, and their estimated indoor
ranges.

Building n σ [dB]
Grocery store 1.8 5.2
Retail store 2.2 8.7
Suburban office building – open plan 2.4 9.6
Suburban office building – soft partitioned 2.8 14.2
Table 2. Values of path loss exponent n at 914MHz

9XCite 24XStream
PTX [dBm] 0.00 16.99
AG [dBi] 4.00 4.00
RS [dBm] -104.00 -105.00
FM [dB] 22.00 22.00
PL = PTX +AG-RS-FM [dB] 86.00 103.99
PL ( d0 ) [dB], d0 = 1 m 31.53 40.05
PL − PL ( d0 ) [dB] 54.47 63.94
n 2.80 2.80
d [m] 88.20 192.18
Table 3. Link budget analysis to the wireless transceivers
The path loss exponent n is selected to be 2.8, which is the same as the soft-partitioned office
building in Table 2. Generally, 2.4GHz signals typically have higher attenuation than
900MHz signals, and, thus, a larger path loss exponent n. The transmitter power PTX ,
receiver sensitivity RS, and fading margin FM of the two wireless transceivers are obtained
from the MaxStream datasheets. A total antenna gain AG of 4 is employed by assuming that
low-cost 2dBi whip antennas are used by both the transmitting and the receiving sides. The
free-space path loss at d0 is computed using the Friis transmission equation (Molisch 2005):

PL(d0 ) [ dB] = 20log10 ( 4π d0 λ ) (7)

where λ is the wavelength of the corresponding wireless signal. Finally, assuming that the
variance Xσ is zero, the mean communication range d can be derived from Eq. (4) as:

( PL − PL ( d0 ) ) (10 n )
d = d010 (8)

Table 3 shows that the transceivers can achieve the communication ranges indicated in
Table 1. It is important to note the sensitivity of the communication range with respect to the
path loss exponent n in Eq. (8). For instance, if the exponent of 3.3 for indoor traveling
(through brick walls, as reported by Janssen & Prasad (1992) for 2.4 GHz signals) is used for
the 24XStream transceiver, its mean communication range reduces by half to 87m.
Wireless Sensor Networks in Smart Structural Technologies 413

3. Wireless structural health monitoring

The prototype wireless unit is first investigated for applications in wireless structural health
monitoring. A structural health monitoring system measures structural performance and
operating conditions with various types of sensing devices, and evaluates structural safety
using damage diagnosis or prognosis methods. Eliminating lengthy cables, wireless sensor
networks can offer a low-cost alternative to traditional cable-based structural health
monitoring systems. Another advantage of a wireless system is the ease of relocating
sensors, thus providing a flexible and easily reconfigurable system architecture. This section
first provides an overview to the wireless structural health monitoring system, and then
introduces the communication protocol design for reliable data management in the
prototype system. A large-scale field deployment of the wireless structural health
monitoring system is summarized at the end of the section.

3.1 Overview of the wireless structural health monitoring system

A simple star-topology network is adopted for the prototype wireless sensing system. The
system includes a server and multiple structural sensors, signal conditioning modules, and
wireless sensing units (Fig. 4). The server is used to organize and collect data from multiple
wireless sensing units in the sensor network. The server is responsible for: 1) commanding
all the corresponding wireless sensing units to perform data collection or interrogation
tasks, 2) synchronizing the internal clocks of the wireless sensing units, 3) receiving data or
analysis results from the wireless network, and 4) storing the data or results. Any desktop or
laptop computer connected with a compatible wireless transceiver can be used as the server.
The server can also provide Internet connectivity so that sensor data or analysis results can
be viewed remotely from other computers over the Internet. Since the server and the
wireless sensing units must communicate frequently with each other, portions of their
software are designed in tandem to allow seamless integration and coordination.

Structural Structural Structural Structural

Sensors Sensors Sensors Sensors

Signal Signal Signal Signal

Conditioning Conditioning Conditioning Conditioning

Wireless Wireless Wireless Wireless

Sensing Unit Sensing Unit Sensing Unit Sensing Unit

Wireless Sensor
Network Server

Fig. 4. An overview of the prototype wireless structural sensing system

At the beginning of each wireless structural sensing operation, the server issues commands
to all the units, informing the units to restart and synchronize. After the server confirms that
all the wireless sensing units have restarted successfully, the server queries the units one by
one for the data they have thus far collected. Before the wireless sensing unit is queried for
its data, the data is temporarily stored in the unit’s onboard SRAM memory buffer.
414 Recent Advances in Wireless Communications and Networks

A unique feature of the embedded wireless sensing unit software is that it can continue
collecting data from interfaced sensors in real-time as the wireless sensing unit is
transmitting data to the server. In its current implementation, at each instant in time, the
server can only communicate with one wireless sensing unit. In order to achieve real-time
continuous data collection from multiple wireless sensing units with each unit having up to
four analog sensors attached, a dual stack approach has been implemented to manage the
SRAM memory (Wang, et al. 2007a). When a wireless sensing unit starts collecting data, the
embedded software establishes two memory stacks dedicated to each sensing channel for
storing the sensor data. For each sensing channel, at any point in time, only one of the stacks
is used to store the incoming data stream. While incoming data is being stored into the
dedicated memory stack, the system transfers the data in the other stack out to the server.
For each sensing channel, the role of the two memory stacks alternate as soon as one stack is
filled with newly collected data.

3.2 Communication design of the wireless structural health monitoring system

To ensure reliable wireless communication between the server and the wireless units, the
communication protocol needs to be carefully designed and implemented. The commonly
used network communication protocol is the Transmission Control Protocol (TCP) standard.
TCP is a sliding window protocol that handles both timeouts and retransmissions. It
establishes a full duplex virtual connection between two endpoints. Although TCP is a
reliable communication protocol, it is too general and cumbersome to be employed by the
low-power and low data-rate communication such as in a wireless structural sensing
network. The relatively long latency of transmitting each wireless packet is another
bottleneck that may slow down the communication throughput. For practical and efficient
application in a wireless structural sensing network, a simpler communication protocol is
needed to minimize transmission overhead. Yet the protocol has to be designed to ensure
reliable wireless transmission by properly addressing possible data loss. The
communication protocol designed for the prototype wireless sensing system inherits some
useful features of TCP, such as data packetizing, sequence numbering, timeout checking,
and retransmission. Based upon pre-assigned arrangement between the server and the
wireless units, the sensor data stream is segmented into a number of packets, each
containing a few hundred bytes. A sequence number is assigned to each packet so that the
server can request the data sequentially.
To simplify the communication protocol, special characteristics of the structural health
monitoring application are exploited. For example, since the objective in structural
monitoring application is normally to transmit sensor data or analysis results to the server,
the server is assigned the responsibility for ensuring reliable wireless communication. As
the server program normally runs on a computer and the wireless unit program runs on a
microcontroller, it is also reasonable to assign the responsibility to the server since it has
much higher computing power. For example, communication is always initiated by the
server. After the server sends a command to the wireless sensing unit, if the server does not
receive an expected response from the unit within a certain time limit, the server will resend
the last command again until the expected response is received. However, after a wireless
sensing unit sends a message to the server, the unit does not check if the message has
arrived at the server correctly or not, because the communication reliability is assigned to
the server. The wireless sensing unit only becomes aware of the lost data when the server
queries the unit for the same data again. In other words, the server plays an “active” role in
the communication protocol while the wireless sensing unit plays more of a “passive” role.
Wireless Sensor Networks in Smart Structural Technologies 415

Init. and Timeout

Sync . Resend
i= 1 The unit is 01Inquiry
expected to
State 1 State 2 Condition
be ready
Wait for i-th Wait for
unit ready Send 01 Inquiry reply Action
to the i -th unit

State 3 Received Received

Timeout 03 DataReady 02 NotReady
Wait for
Resend Send Resend
reply
04 PlsSend 04 PlsSend 01 Inquiry

Received one packet, and Collected all data State 4

more data to be collected from the i-th unit Wait for
Send 04 PlsSend Send 05 EndTransm reply

Receive 06 AckEndTransm Timeout

Resend
If i == N (the last unit ), then let i = 1; otherwise let i = i + 1
05 EndTransm
(a) State diagram of the server

Received 10 Restart Received 05 EndTransm

Received 04 PlsSend
Send 11 AckRestart Send 06 AckEndTransm
Send requested packet

Received 01 Inquiry
State 1 State 2
Init. and and data is ready
Wait for Wait for
Sync . Send 03 DataReady
01 Inquiry 04 PlsSend

Received 01 Inquiry Received 01 Inquiry

but data is not ready Send 03DataReady
Send 02 NotReady

Received 05 EndTransm Condition

Send 06 AckEndTransm Action

(b) State diagram of a wireless sensing unit.

Fig. 5. Communication state diagrams for wireless structural health monitoring
416 Recent Advances in Wireless Communications and Networks

Finite state machine concepts are employed in designing the communication protocol for the
wireless sensing units and the server. A finite state machine consists of a set of states and
definable transitions between the states (Tweed 1994). At any point in time, the state
machine can only be in one of the possible states. In response to different events, the state
machine transits between its discrete states. The communication protocol for initialization
and synchronization can be found in (Wang, et al. 2007a). Fig. 5(a) shows the communication
state diagram of the server for one round of sensor data collection, and Fig. 5(b) shows the
corresponding state diagram of the wireless units. During each round of data collection, the
server collects sensor data from all of the wireless units; note that the server and the units
have separate sets of state definitions.
At the beginning of data collection, the server and all the units are all set in State 1. Starting
with the first wireless unit in the network, the server queries the sensor for the availability of
data by sending the ‘01Inquiry’ command. If the data is not ready, the unit replies
‘02NotReady’, otherwise the unit replies ‘03DataReady’ and transits to State 2. After the
server ensures that the data from this wireless unit is ready for collection, the server transits
to State 3. To request a data segment from a unit, the server sends a ‘04PlsSend’ command
that contains a packet sequence number. One round of data collection from one wireless
unit is ended with a two-way handshake, where the server and the unit exchange
‘05EndTransm’ and ‘06AckEndTransm’ commands. The server then moves on to the next
unit and continuously collects sensor data round-by-round.

3.3 Field validation tests at Voigt Bridge

Laboratory and field validation tests have been conducted to verify the performance of the
wireless structural monitoring system. Field tests are particularly helpful in assessing the
limitations of the system, and providing valuable experience that can lead to further
improvements in the system hardware and software design. This section presents an
overview of the validation tests conducted on the Voigt Bridge located on the campus of the
University of California, San Diego (UCSD) in La Jolla, California (Fraser, et al. 2006). Voigt
Bridge is a two lane concrete box girder highway bridge. The bridge is about 89.4m long and
consists of four spans (Fig. 6). The bridge deck has a skew angle of 32º, with the concrete
box-girder supported by three single-column bents. Over each bent, a lateral diaphragm
with a thickness of about 1.8m stiffens the girder. Longitudinally, the box girder is
partitioned into five cells running the length of the bridge (Fig. 6b).
Girder cells along the north side of the bridge are accessible through four manholes on the
bridge sidewalk. As a testbed project for structural health monitoring research, a cable-
based system has been installed in the northern-most cells of the box girder. The cable-based
system includes accelerometers, strain gages, thermocouples, and humidity sensors. For the
purpose of validating the proposed wireless structural monitoring system, thirteen
accelerometers interfaced to wireless sensing units are installed within the two middle spans
of the bridge to measure vertical vibrations. One wireless sensing unit (associated with one
signal conditioning module and one accelerometer) is placed immediately below the
accelerometer associated with the permanent wired monitoring system. While the wired
accelerometers are mounted to the cell walls, wireless accelerometers are simply mounted
on the floor of the girder cells to expedite the installation process. The installation and
calibration of the wireless monitoring system, including the placement of the 13 wireless
sensors, takes about an hour. The MaxStream 9XCite wireless transceiver operating at
900MHz is integrated with each wireless sensing unit.
Wireless Sensor Networks in Smart Structural Technologies 417

Lateral diaphragm Longitudinal diaphragm

1 2 3 4 5 6 7 8 9 10 11 12 13

6.1
Abut. 1

m
Bent 1 Bent 2 Bent 3

6 .1
m
Abut. 2

16.2 m 29.0 m 29.0 m 15.2 m

N
Wireless network server One pair of wireless and wired accelerometers

(a) Plan view of the bridge illustrating locations of wired and wireless sensing systems

Wired accelerometer 1.8

m

Wireless accelerometer
10 .7 m

Section A- A

(b) Elevation view to section A-A (c) Side view of the bridge over Interstate 5

Fig. 6. Voigt Bridge test comparing the wireless and wired sensing systems
Two types of accelerometers are associated with each monitoring system. At locations #3, 4,
5, 9, 10, and 11 in Fig. 6(a), PCB Piezotronics 3801 accelerometers are used with both the
cabled and the wireless systems. At the other seven locations, Crossbow CXL01LF1
accelerometers are used with the cabled system, while Crossbow CXL02LF1Z
accelerometers are used with the wireless system. Table 4 summarizes the key parameters of
the three types of accelerometers. Signal conditioning modules are used for filtering noise,
amplifying and shifting signals for the wireless accelerometers. The signals of the wired
accelerometers are directly digitized by a National Instruments PXI-6031E data acquisition
board (Fraser, et al. 2006). Sampling frequencies for the cable-based system and the wireless
system are 1,000 Hz and 200 Hz, respectively.

Specification PCB3801 CXL01LF1 CXL02LF1Z

Sensor Type Capacitive Capacitive Capacitive
Maximum Range ± 3g ± 1g ± 2g
Sensitivity 0.7 V/g 2 V/g 1 V/g
Bandwidth 80 Hz 50Hz 50Hz
RMS Resolution (Noise Floor) 0.5 mg 0.5 mg 1 mg
Minimal Excitation Voltage 5 ~ 30 VDC 5 VDC 5 VDC
Table 4. Parameters of the accelerometers used by the wire-based and wireless systems in
the Voigt Bridge test
418 Recent Advances in Wireless Communications and Networks

-3
x 10
5
Acceleration (g)

Wired #6
-5
0 2 4 6 8
-3
x 10
5

Wired #12
-5
0 2 4 6 8
Time (s)

-3
x 10
5
Acceleration (g)

Wireless #6
-5
0 2 4 6 8
-3
x 10
5

Wireless #12
-5
0 2 4 6 8
Time (s)

(a) Comparison between wired and wireless time history data

Wireless Sensor Networks in Smart Structural Technologies 419

FFT Magnitude
Wired #6

0
0 5 10 15
4
Wired #12
2

0
0 5 10 15
Frequency (Hz)
FFT Magnitude

Wireless #6
0.5

0
0 5 10 15

Wireless #12
0.5

0
0 5 10 15
Frequency (Hz)
(b) Comparison between FFT to the wired data, as computed offline by a computer, and FFT
to the wireless data, as computed online by the wireless sensing units
Fig. 7. Comparison between wired and wireless data for the Voigt Bridge test
The bridge is under normal traffic operation during the tests. Fig. 7(a) shows the time
history data at locations #6 and #12, collected by the cable-based and wireless monitoring
systems when a vehicle passes over the bridge. A close match is observed between the data
collected by the two systems. The minor difference between the two data sets can be mainly
attributed to two sources: 1) the signal conditioning modules are used in the wireless system
but not in the cabled system; 2) the wired and wireless accelerometer locations are not
exactly adjacent to each other, as previously described. Fig. 7(b) shows the Fourier spectra
determined from the time history data. The FFT results using the data collected by the
cabled system are computed offline, while the FFT results corresponding to the wireless
data are computed online in real-time by each wireless sensing unit. After each wireless
sensing unit executes its FFT algorithm, the FFT results are wirelessly transmitted to the
420 Recent Advances in Wireless Communications and Networks

network server. Strong agreement between the two sets of FFT results validates the
computational accuracy of the wireless sensing units. It should be pointed out that because
the sampling frequency of the cabled system is five times higher than that of the wireless
system, the magnitude of the Fourier spectrum for the wired data is also about five times
higher than those for the wireless data.
One attractive feature of the wireless sensing system is that the locations of the sensors can
be re-configured easily. To determine the operating deflection shapes of the bridge deck, the
configuration of the original wireless sensing system is changed to attain a more suitable
spatial distribution. Twenty wireless accelerometers and the wireless network server are
mounted to the bridge sidewalks (Fig. 8). The communication distance between the server
and the farthest wireless sensing unit is close to the full length of the bridge. The installation
and calibration of the wireless monitoring system, including the placement of all the
wireless sensors, again takes about an hour. Sampling frequency for the wireless monitoring
system is kept at 200 Hz.

A
1 2 3 4 5 6 7 8 9 10

6.1
Abut. 1

m
Bent 1 Bent 2 Bent 3

6 .1
m
11 12 13 14 15 16 17 18 19 20 Abut. 2

16.2 m 29 .0 m A 29 .0 m 15 .2 m

N
Wireless network server Wireless accelerometer Hammer location

(a) Plan view of the bridge illustrating locations of wireless accelerometers

Wireless accelerometer
1.8
m

10 .7 m

Section A -A

(b) Elevation view to section A-A (c) Side view of the bridge over Interstate 5
Fig. 8. Wireless accelerometer deployment for the operating deflection shape analysis to
Voigt Bridge
The communication protocol described before is implemented in the server and the wireless
sensing units. For the tests described in this chapter, the server collects sensor data or FFT
results from all 20 wireless units. Due to the length of the bridge and continuous traffic
conditions, the wireless communication experienced some intermittent difficulty during the
two days of field testing. However, the wireless monitoring system proved robust by
recognizing communication failures and successfully retransmitting the lost data according
to the communication protocol rules.
Wireless Sensor Networks in Smart Structural Technologies 421

Fig. 9 shows the operating deflection shapes (ODS) extracted from one set of test data
collected during a hammer excitation test. The hammer excitation is applied at the location
shown in Fig. 8(a) and during intervals of no passing vehicles. DIAMOND, a modal analysis
software package, is used to extract the operating deflection shapes (ODS) of the bridge
deck (Doebling, et al. 1997). Under hammer excitation, the operating deflection shapes at or
near a resonant frequency should be dominated by a single mode shape (Richardson 1997).
Fig. 9 presents the first four dominant operating deflection shapes of the bridge deck using
wireless acceleration data. The ODS #1 (4.89 Hz), #2 (6.23 Hz), and #4 (11.64 Hz) show
primarily flexural bending modes of the bridge deck; a torsional mode is observed in ODS
#3 (8.01 Hz). Successful extraction of the ODS shows that the acceleration data from the 20
wireless units are well synchronized.

ODS #1, 4.89Hz ODS #2, 6.23Hz

0.5 0.5
0 0
0 0
-10 -10
-0.5 -0.5
-20 -20
-60 -40 -20 0 20 40 60 -60 -40 -20 0 20 40 60
ODS #3, 8.01Hz ODS #4, 11.64Hz

0.5 0.5
0 0
0 0
-10 -10
-0.5 -0.5
-20 -20
-60 -40 -20 0 20 40 60 -60 -40 -20 0 20 40 60

Fig. 9. Operating deflection shapes extracted from wireless sensor data

4. Wireless structural control

A feedback structural control system contains an integrated network of sensors, controller,
and control devices. When external excitation (such as an earthquake or typhoon) occurs,
structural response is measured by sensors and immediately collected by the controller. The
controller makes optimal decisions for the control devices, which then exert appropriate
forces to the structure so that undesired structural vibrations are effectively mitigated. A
wireless sensing/control unit can serve as both the sensor and the controller modules of a
structural control system. Each wireless unit, in addition to collecting and communicating
sensor data in real time, can also make optimal control decisions and command control
devices. This section first provides an overview to the prototype wireless structural control
system, and then describes the communication protocol design of the system. Laboratory
wireless structural control experiments are also reported.

4.1 Overview of the wireless structural control system

Fig. 10 illustrates the communication patterns of a centralized control system using cabled
communication and the prototype decentralized structural control system using wireless
communication. In a centralized structural control system, one centralized controller collects
data from all the sensors in the whole structure, computes control decisions, and then
dispatches command signals to control devices. This centralized control strategy implemented
with cabled communication requires high instrumentation cost, is difficult to reconfigure,
422 Recent Advances in Wireless Communications and Networks

and potentially suffers from single-point failure at the controller. Wireless decentralized
control architectures can offer an alternative solution. In a decentralized architecture,
multiple sensors and controllers can be distributively placed in a large structure, where the
controller nodes can be closely collocated with the control devices. As each controller only
needs to communicate with sensors and control devices in its vicinity, the requirement on
communication range can be significantly reduced, and the communication latency
decreases by reducing the number of sensors or control devices that each controller has to
communicate with.

Sensor Sensor Sensor Sensor Sensor

Controller

Control Control Control Control Control

device device device device device

Centralized Cabled Control

Sensor Sensor Sensor Sensor Sensor

Controller & Controller & Controller & Controller & Controller &
Ctrl. device Ctrl. device Ctrl. device Ctrl. device Ctrl. device

Decentralized Wireless Control

Fig. 10. Centralized and decentralized control systems

For application in wireless feedback structural control, real-time communication is

important for system performance. Limited wireless communication range poses another
challenge while instrumenting a large-scale structure with the wireless sensing and
control system. Particularly, in discrete-time feedback control, a steady sampling time
step and low communication latency are essential for the system performance. The
feedback control loop designed for the prototype wireless sensing and control system is
Wireless Sensor Networks in Smart Structural Technologies 423

illustrated in Fig. 11(a), and the pseudo code implementing the feedback loop is presented
in Fig. 11(b). As shown in the figures, sensing is designed to be clock-driven, while control
is designed to be event-driven. The wireless sensing nodes collect sensor data at a preset
sampling rate, and transmit the data during an assigned time slot. Upon receiving the
required sensor data, the control nodes immediately compute control decisions and apply
the corresponding command signals to the control devices. If due to occasional data
packet loss, a control node doesn’t receive the expected sensor data at one time step, the
control node may use a projected data sample for control decisions, or doesn’t take any
action at this time step.

4.2 Communication protocol design for the wireless structural control system
Similar to the structural monitoring application, a reliable communication protocol must be
properly designed for the wireless structural control system. Fig. 12 illustrates the
communication state diagrams of a coordinator unit and other wireless units within a
wireless sensing and control subnet. To initiate the system operation, the coordinator unit
first broadcasts a start command ‘01StartCtrl’ to all other sensing and control units. Once the
start command and its acknowledgement ‘03AcknStartCtrl’ are received, the system starts
real-time feedback control operation, i.e. both the coordinator and other units are in State 2.

Wireless Sensor Nodes Wireless Control Nodes

Wireless
Collect and send Communication Receive
sensor data sensor data

Sensor Controller

Structural Control
System device

(a) Feedback control loop between the wireless sensing nodes and control nodes

Wireless Sensing Nodes Wireless Control Nodes

(Clock-driven) (Event-driven)
ITERATE { ITERATE {
IF (sensor data arrived on time)
Wait for the assigned time slot. Compute control decisions.
Apply control command signal.
Sample sensor data. ELSE
Use projected data sample or no action.
Wirelessly transmit sensor data. Wait for the wireless sensor data.
} }

(b) Pseudo code for the feedback control loop

Fig. 11. Illustration of the feedback control loop in a wireless decentralized control system
424 Recent Advances in Wireless Communications and Networks

At every sampling time step, the coordinator unit broadcasts a beacon signal ‘02BeaconData’
together with its own sensor data, announcing the start of a new time step. Upon receiving
the beacon signal, other sensing units broadcast their sensor data following a preset
transmission sequence, so that transmission collision is avoided. The wireless control units
responsible for commanding the control devices receive the sensor data, calculate desired
control forces, and apply control commands at each time step. In order to guarantee a
constant sampling time step and to minimize feedback latency, timeout checking or
retransmission is not recommended during the feedback control operation. This design is
suitable for both centralized control and decentralized control.

Coordinator Unit

At every sampling time step

Timeout
Send 02BeaconData; make control
Init. Resend 01StartCtrl decision; wait other sensor data
Send 01StartCtrl

State1 State2
Got 03AcknStartCtrl
wait wait
AcknStart sensor
Ctrl data

Other Sensing/Control Units

Got 02BeaconData
Wait assigned slot; send latest
Init. sensor data; make control decision
Got 01StartCtrl

Reply 03AcknStartCtrl State2

State1
wait wait
StartCtrl 02Beacon
Data
Got 02BeaconData
Got 01StartCtrl
Wait assigned time slot and
Reply 03AcknStartCtrl
send latest sensor data

Fig. 12. Communication state diagram of a coordinator unit and other sensing/control units
in one wireless subnet
For illustration purpose, a 3-story structure instrumented with the prototype wireless
control system is shown in Fig. 13. The steel frame structure is designed and constructed
by researchers affiliated with the National Center for Research on Earthquake Engineering
(NCREE) in Taipei, Taiwan. The prototype wireless system consists of wireless sensors
and controllers that are mounted on the structure for measuring structural response data
and commanding MR dampers in real-time. Besides the wireless sensing and control units
Wireless Sensor Networks in Smart Structural Technologies 425

that are necessary for data collection and the operation of the control devices, a remote
command server with a wireless transceiver is also included for experimental purpose. In
a laboratory setup, the server is designed to initiate the operation of the control system
and to log the data flow in the wireless network. To initiate the operation, the command
server first broadcasts a start signal to all the wireless sensing and control units. Once the
start command is received, the wireless units that are responsible for collecting sensor
data start acquiring and broadcasting data at a preset time interval. Accordingly, the
wireless units responsible for commanding the MR dampers receive the sensor data,
calculate desired control forces, and apply control commands within the specified time
interval.

S3 V3
Ci : Wireless control unit (with one
wireless transceiver included) Floor-3

Si : Wireless sensing unit (with one

C2
wireless transceiver included) 3m

Ti : Wireless transceiver D2 V2
Floor-2
Di : MR Damper

Vi : Velocity meter C1
3m

Floor plan: 3m x 2m D1 V1
Floor weight: 6,000kg Floor-1
Steel I-section beams and
columns: H150 x 150 x 7 x 10
C0
3m
T0
Lab experiment
command server D0 V0
Floor-0

(a) A 3-story test structure (b) Deployment of the wireless sensors,

mounted on the shake table controllers, and control devices

Fig. 13. Laboratory setup of the wireless structural control system

To coordinate the wireless transmissions during the feedback control, a pre-specified

communication sequence should be observed by all the wireless units. For example, if all
three wireless control units need velocity data from all the floors to compute control
decisions, a communication sequence illustrated in Fig. 14 can be adopted by the prototype
system. The control sampling step, which is 80ms in this example, is mostly decided by the
total time required for transmitting all four data packets. For the 24XStream wireless
transceiver adopted in the system, wireless transmission of each velocity measurement takes
about 18ms. During every control time step, the wireless unit C0 first samples the velocity
data V0 at its own floor, and then sends out the data together with a beacon signal to other
wireless units. Upon receiving the beacon signal, units C1, C2, and S3 sequentially broadcast
their sensor data. Last, a period of 8ms is designed as a safety cushion for each control
sampling time step, allowing certain randomness in the wireless transmission time. The
control units C0, C1, and C2 compute control decisions and apply actuation signals during
the intervals of wireless transmissions.
426 Recent Advances in Wireless Communications and Networks

80ms

18ms 18ms 18ms 18ms 8ms

Compute
C0

Compute
C1
Beacon with data
Compute
C2

S3
Data only

3ms 12ms 3ms

Fig. 14. Communication sequence in a wireless structural control network

4.3 Validation experiments for the wireless structural control system

Validation experiments for the wireless control system were conducted at NCREE in Taipei,
Taiwan, using the structure shown in Fig. 13. The floor plan of this structure is 3m × 2m,
with each floor weight adjusted to 6,000 kg using concrete blocks; inter-story heights are 3m.
The three-story structure is mounted on a 5m × 5m 6-DOF shake table. For this study, only
longitudinal excitation in one degree of freedom is applied. Besides wireless sensors, a
separate set of accelerometers, velocity meters, and linear variable displacement transducers
(LVDT) are installed on each floor of the structure; this set of sensors are interfaced to a
high-precision tethered data acquisition (DAQ) system native to the NCREE facility.
For this experimental study, three 20 kN MR dampers are deployed. Each damper is
installed under a V-brace upon one of the three floors (Fig. 13b). The damping coefficients of
the MR dampers can be changed by issuing a command voltage between 0V to 1.2V. This
command voltage determines the electric current of the electromagnetic coil inside the MR
damper, which, in turn, generates a magnetic field that sets the viscous damping properties
of the MR damper fluid (Lin, et al. 2005). Two control systems, the wireless control system
and a traditional wire-based control system, are installed in the test structure. For the
wireless system, a total of four wireless sensors are installed to measure floor velocities (Fig.
13). Velocity feedback control algorithms presented in a previous paper are used by both the
wired and the wireless control systems (Wang, et al. 2007b). In a centralized feedback
pattern, real-time data from all sensors are required for making the control decisions for
every MR damper. For this test structure, the wire-based system can achieve a sampling rate
of 200Hz; as shown in Fig. 14, the wireless system can achieve a sampling rate of 12.5Hz.
In order to ensure that appropriate control decisions are computed by the wireless control
units, one necessary condition is that the real-time velocity data used by the control units are
reliable. Rarely experiencing data losses during the experiments, our prototype wireless
sensor network proves to be robust. As reported by Lynch, et al. (2008), data losses less than
2% are experienced. Should data loss be encountered, the wireless control unit is currently
designed to simply use the data sample from the previous time step. To illustrate the
reliability of the velocity data collected and transmitted by the wireless units, Fig. 15(a)
presents the Floor-1 time history data during a centralized wireless control test. The data is
Wireless Sensor Networks in Smart Structural Technologies 427

collected by both the wired DAQ system and the three wireless control units. During the
test, unit C1 measures the data from the associated velocity meter directly, stores the data in
its own memory bank, and transfers the data wirelessly to units C0 and C2. After the test run
is completed, data from all the three control units are sequentially streamed to the
experiment command server, where the results are plotted as shown in Fig. 15(a). These
plots illustrate strong agreement among data recorded by the three wireless control units
and by the wired system using a separate set of velocity meters and data acquisition system.
It is shown that the velocity data are not only reliably measured by unit C0, but also
properly transmitted to other wireless control units in real-time.

Floor-1 Absolute Velocity

0.1
Velocity (m/s)

Measured by Cabled System

0

-0.1
0 5 10 15 20 25 30 35 40
Time (s)
0.1
Velocity (m/s)

Recorded by Unit C0
0

-0.1
0 5 10 15 20 25 30 35 40
Time (s)
0.1
Velocity (m/s)

Recorded by Unit C1
0

-0.1
0 5 10 15 20 25 30 35 40
Time (s)
0.1
Velocity (m/s)

Recorded by Unit C2

-0.1
0 5 10 15 20 25 30 35 40
Time (s)
(a) Floor-1 absolute velocity data recorded by the cabled and wireless sensing systems
428 Recent Advances in Wireless Communications and Networks

Floor 3/2 Inter-Story Drift under El Centro Excitation (Peak 1 m/s2)

0.02
Uncontrolled Structure
Wireless Centralized
0.01 Wired Centralized
Drift (m)

-0.01

-0.02
0 2 4 6 8 10
Time (s)
Floor 2/1 Inter-Story Drift under El Centro Excitation (Peak 1 m/s2)
0.02

0.01
Drift (m)

-0.01

-0.02
0 2 4 6 8 10
Time (s)
Floor 1/0 Inter-Story Drift under El Centro Excitation (Peak 1 m/s2)
0.02

0.01
Drift (m)

-0.01

-0.02
0 2 4 6 8 10
Time (s)
(b) Inter-story drifts of the structure with and without control

Fig. 15. Experimental time histories

Wireless Sensor Networks in Smart Structural Technologies 429

The time histories of the inter-story drifts from the same centralized wireless control test are
plotted in Fig. 15(b), together with the drifts of a centralized wired control test and a bare
structure test when the structure is not instrumented with any control system (i.e. the MR
dampers are not installed). The same ground excitation (1940 El Centro NS earthquake
record scaled to a peak ground acceleration of 1m/s2) is used for all the three cases shown in
Fig. 15(b). The results show that both the wireless and wired control systems achieve
considerable performance in mitigating inter-story drifts. Running at a much shorter
sampling time step, the wired centralized control system achieves slightly better control
performance than the wireless centralized system in terms of mitigating inter-story drifts.
To further study different decentralized schemes with different communication latencies,
three wireless control architectures are compared: (#1) decentralized, (#2) partially
decentralized, and (#3) centralized. Fig. 16 illustrates the information feedback pattern of
each control architecture. The fully decentralized pattern (Wireless #1) specifies that when
computing control decisions, the MR damper at each floor only needs the inter-story
velocity difference at Story 1. The partially decentralized pattern (Wireless #2) specifies that
the control decisions require inter-story velocity from a neighboring floor. Finally, the
centralized pattern (Wireless #3) indicates all velocities relative to ground are required by
the control decisions. Different information patterns result in different sampling frequencies
for each control architecture. Compared with the centralized scheme, the advantage of a
decentralized architecture is that fewer communication and data processing are needed at
each sampling time step, thereby reducing sampling time step length. As shown in Fig. 16,
the wireless system can achieve a sampling rate of 16.67Hz for partially decentralized
control and 50Hz for fully decentralized control.

Story 3 V3- V2 D3 V3- V2 D3 V3- V0 D3

Story 2 V2- V1 D2 V2- V1 D2 V2- V0 D2

Story 1 V1- V0 D1 V1- V0 D1 V1- V0 D1

Wireless #1 (50 Hz) Wireless #2 (16.67 Hz) Wireless #3 (12 .5 Hz)
Fully Decentralized Partially Decentralized Centralized
Fig. 16. Various decentralized and centralized information feedback
Fig. 17 shows the peak inter-story drifts and floor accelerations for the original uncontrolled
structure and the structure controlled by the three different wireless schemes, as well as the
wired centralized control scheme. The 1940 El Centro NS record is employed as the ground
excitation, with peak ground acceleration scaled to 1m/s2. Compared with the uncontrolled
structure, all wireless and wired control schemes achieve significant reduction with respect
to maximum inter-story drifts and absolute accelerations. Among the four control cases, the
wired centralized control scheme shows good performance in mitigating both peak drifts
and peak accelerations. This result is expected because the wired system has the advantages
of lower communication latency and utilizes sensor data from all floors. The wireless
schemes, although running at longer sampling steps, achieve control performance comparable
to the wired system. For all three earthquake records, the fully decentralized wireless
430 Recent Advances in Wireless Communications and Networks

control scheme (Wireless #1) results in low peak inter-story drifts and the smallest peak
floor accelerations at most of the floors. This result illustrates that in the decentralized
wireless control cases, the higher sampling rate (achieved due to lower communication
latency) potentially compensates for the lack of data from faraway floors.

Maximum Inter-story Drifts

3
No Control
Wireless #1
Wireless #2
Wireless #3
Wired
Story

1
0 0.005 0.01 0.015 0.02 0.025
Drift (m)

Maximum Absolute Accelerations

3
No Control
Wireless #1
Wireless #2
Wireless #3
Wired
Floor

1
0 0.5 1 1.5 2 2.5 3 3.5
2
Acceleration (m/s )

Fig. 17. Experimental results of different control schemes under 1940 El Centro NS earthquake
excitation with peak ground accelerations (PGA) scaled to 1m/s2
Wireless Sensor Networks in Smart Structural Technologies 431

5. Summary and discussion

This chapter discusses the various issues of applying wireless sensor networks to modern
smart structural technologies, including structural health monitoring and structural control.
Autonomous wireless sensing and control units with embedded computing can serve as the
building blocks of a smart structural system. For different structural applications, design
concepts have been proposed to address the information constraints in a wireless sensor
network, such as bandwidth, latency, range, and reliability. Robust communication protocol
design for centralized and decentralized information architectures is proposed for efficiently
managing the information flow in the wireless network. State machine concepts prove to be
effective in designing simple yet efficient communication protocols for wireless structural
sensing and control networks. Large-scale laboratory and field validation tests have been
conducted to validate the efficacy and robustness of the information management schemes
implemented in the wireless structural monitoring and control system. Most recently, the
prototype wireless sensing system has been successfully tested for long-range measurement
of low-amplitude and low-frequency vibrations at Canton Tower, a.k.a. Guangzhou TV and
Sightseeing Tower, the world’s tallest TV tower upon construction (Ni, et al. 2011).
A common trend in both structural monitoring and structural control application is the
increasingly dense deployment of system nodes, i.e. sensors in a structural monitoring
system, or sensors, controllers, and control devices in a structural control system. For
example, in structural monitoring systems for cable-supported bridges, hundreds of sensors
are often deployed for recording loading conditions and bridge responses (Wong 2004, Ko
and Ni 2005, Çelebi 2006). Among many modern structural control systems, hundreds of
semi-active hydraulic dampers have been installed in high-rise buildings (Kurino, et al. 2003,
Spencer and Nagarajaiah 2003, Shimizu, et al. 2004). With rapid advancement in wireless
sensor networks, there will be an inevitable trend of reduced system cost yet increased
system nodal densities. Particularly in recent years, more and more large-scale wireless
structural health monitoring (Lynch, et al. 2006, Kim, et al. 2007, Weng, et al. 2008, Whelan
and Janoyan 2009, Rice, et al. 2010) and wireless structural control (Swartz and Lynch 2009,
Wang and Law 2011) studies have been reported. Furthermore, researchers have started
interesting exploration on mobile sensor networks, as the next-generation wireless sensor
networks, for structural health monitoring applications (Zhu, et al. 2010). Such a mobile
sensor network involves miniature autonomous mobile robots that carry wireless sensors
and automatically move upon a large structure. In summary, it is believed that future
monitoring and control systems will enjoy tremendous opportunities provided by the
continuing advancements in wireless sensor technologies.

6. Acknowledgment
This research was partially funded by the National Science Foundation under grants CMS-
9988909 and CMMI-0824977 (Stanford University), as well as CMMI-0928095 (Georgia
Institute of Technology). The first author was supported by an Office of Technology
Licensing Stanford Graduate Fellowship. Prof. Jerome P. Lynch at University of Michigan
kindly provided insightful advices to the development of the prototype wireless sensing
and control system. Prof. Chin-Hsiung Loh, Dr. Pei-Yang Lin, and Dr. Kung-Chun Lu at the
National Taiwan University offered generous support for conducting the shake table
experiments at NCREE, Taiwan. The authors would also like to express their gratitude to
432 Recent Advances in Wireless Communications and Networks

Prof. Ahmed Elgamal and Dr. Michael Fraser of the University of California, San Diego, for
their generous assistance throughout the field validation tests at Voigt Bridge.

7. References
Çelebi, M. (2006). Real-time seismic monitoring of the new Cape Girardeau Bridge and
preliminary analyses of recorded data: an overview. Earthquake Spectra, Vol. 22, No.
3, pp. 609-630
Çelebi, M. (2002). Seismic Instrumentation of Buildings (with Emphasis on Federal Buildings).
Report No. 0-7460-68170, United States Geological Survey, Menlo Park, CA
Chang, F.-K. (Ed.) Structural Health Monitoring 2009: From System Integration to
Autonomous Systems, Proceedings of the 6th International Workshop on Structural
Health Monitoring, Lancaster, PA, USA, September 9-11, 2009
Cooklev, T. (2004). Wireless Communication Standards : a Study of IEEE 802.11, 802.15, and
802.16, Standards Information Network IEEE Press, New York
Doebling, S. W., Farrar, C. R. & Cornwell, P. J. (1997). DIAMOND: A graphical interface
toolbox for comparative modal analysis and damage identification, Proceedings of
the 6th International Conference on Recent Advances in Structural Dynamics,
Southampton, UK, July 14 - 17, 1997
Farrar, C. R., Sohn, H., Hemez, F. M., Anderson, M. C., Bement, M. T., Cornwell, P. J.,
Doebling, S. W., Schultze, J. F., Lieven, N. & Robertson, A. N. (2003). Damage
Prognosis: Current Status and Future Needs. Report No. LA-14051-MS, Los Alamos
National Laboratory, Los Alamos, NM
Fraser, M., Elgamal, A. & Conte, J. P. (2006). UCSD Powell Laboratory Smart Bridge Testbed.
Report No. SSRP 06/06, Department of Structural Engineering, University of
California, San Diego, La Jolla, CA
Housner, G. W., Bergman, L. A., Caughey, T. K., Chassiakos, A. G., Claus, R. O., Masri, S. F.,
Skelton, R. E., Soong, T. T., Spencer, B. F., Jr. & Yao, J. T. P. (1997). Structural control:
past, present, and future. Journal of Engineering Mechanics, Vol. 123, No. 9, pp. 897-971
Janssen, G. J. M. & Prasad, R. (1992). Propagation measurements in an indoor radio
environment at 2.4 GHz, 4.75 GHz and 11.5 GHz, Proceedings of IEEE 42nd Vehicular
Technology Conference, Denver, CO, May 10 - 13, 1992
Kim, S., Pakzad, S., Culler, D., Demmel, J., Fenves, G., Glaser, S. & Turon, M. (2007). Health
monitoring of civil infrastructures using wireless sensor networks, Proceedings of the
6th International Conference on Information Processing in Sensor Networks (IPSN '07),
Cambridge, MA, April 25 - 27, 2007
Ko, J. M. & Ni, Y. Q. (2005). Technology developments in structural health monitoring of
large-scale bridges. Engineering Structures, Vol. 27, No. 12, pp. 1715-1725
Kurino, H., Tagami, J., Shimizu, K. & Kobori, T. (2003). Switching oil damper with built-in
controller for structural control. Journal of Structural Engineering, Vol. 129, No. 7, pp.
895-904
Lin, P.-Y., Roschke, P. N. & Loh, C.-H. (2005). System identification and real application of
the smart magneto-rheological damper, Proceedings of the 2005 International
Symposium on Intelligent Control, Limassol, Cyprus, June 27 - 29, 2005
Lynch, J. P. & Loh, K. J. (2006). A summary review of wireless sensors and sensor networks
for structural health monitoring. The Shock and Vibration Digest, Vol. 38, No. 2, pp.
91-128
Wireless Sensor Networks in Smart Structural Technologies 433

Lynch, J. P., Wang, Y., Loh, K. J., Yi, J.-H. & Yun, C.-B. (2006). Performance monitoring of the
Geumdang Bridge using a dense network of high-resolution wireless sensors. Smart
Materials and Structures, Vol. 15, No. 6, pp. 1561-1575
Lynch, J. P., Wang, Y., Swartz, R. A., Lu, K.-C. & Loh, C.-H. (2008). Implementation of a
closed-loop structural control system using wireless sensor networks. Structural
Control and Health Monitoring, Vol. 15, No. 4, pp. 518-539
MaxStream, Inc. (2004). 9XCite™ OEM RF Module Product Manual v1.1. Lindon, UT
MaxStream, Inc. (2005). XStream™ OEM RF Module Product Manual v4.2B. Lindon, UT
Molisch, A. F. (2005). Wireless Communications, John Wiley & Sons, IEEE Press, Chichester,
West Sussex, England
Ni, Y. Q., Li, B., Lam, K. H., Zhu, D., Wang, Y., Lynch, J. P. & Law, K. H. (2011). In-
construction vibration monitoring of a super-tall structure using a long-range
wireless sensing system. Smart Structures and Systems, Vol. 7, No. 2, pp. 83-102
Rappaport, T. S. & Sandhu, S. (1994). Radio-wave propagation for emerging wireless
personal-communication systems. Antennas and Propagation Magazine, IEEE, Vol. 36,
No. 5, pp. 14-24
Rice, J. A., Mechitov, K., Sim, S.-H., Nagayama, T., Jang, S., Kim, R., B. F. Spencer, J., Agha,
G. & Fujino, Y. (2010). Flexible smart sensor framework for autonomous structural
health monitoring. Smart Structures and Systems, Vol. 6, No. 5, pp. 423-438
Richardson, M. H. (1997). Is it a mode shape, or an operating deflection shape? Sound and
Vibration Magazine, Vol. 31, No. pp. 54-61
Shimizu, K., Yamada, T., Tagami, J. & Kurino, H. (2004). Vibration tests of actual buildings
with semi-active switching oil damper, Proceedings of the 13th World Conference on
Earthquake Engineering, Vancouver, B.C., Canada, August 1 - 6, 2004
Sohn, H., Farrar, C. R., Hemez, F. M., Shunk, D. D., Stinemates, D. W. & Nadler, B. R. (2003).
A Review of Structural Health Monitoring Literature: 1996-2001. Report No. LA-13976-
MS, Los Alamos National Laboratory, Los Alamos, NM
Soong, T. T. (1990). Active Structural Control: Theory and Practice, Wiley, Harlow, Essex,
England
Spencer, B. F., Jr. & Nagarajaiah, S. (2003). State of the art of structural control. Journal of
Structural Engineering, Vol. 129, No. 7, pp. 845-856
Straser, E. G. & Kiremidjian, A. S. (1998). A Modular, Wireless Damage Monitoring System for
Structures. Report No. 128, John A. Blume Earthquake Eng. Ctr., Stanford
University, Stanford, CA
Swartz, R. A. & Lynch, J. P. (2009). Strategic network utilization in a wireless structural
control system for seismically excited structures. Journal of Structural Engineering,
Vol. 135, No. 5, pp. 597-608
Tweed, D. (1994). Designing real-time embedded software using state-machine concepts,
Circuit Cellar Ink, (53), pp. 12-19.
Wang, Y., Lynch, J. P. & Law, K. H. (2005). Design of a low-power wireless structural
monitoring system for collaborative computational algorithms, Proceedings of SPIE,
Health Monitoring and Smart Nondestructive Evaluation of Structural and Biological
Systems IV, San Diego, CA, March 9, 2005
Wang, Y. (2007). Wireless Sensing and Decentralized Control for Civil Structures: Theory and
Implementation. PhD Thesis, Department of Civil and Environmental Engineering,
Stanford University, Stanford, CA
434 Recent Advances in Wireless Communications and Networks

Wang, Y., Lynch, J. P. & Law, K. H. (2007a). A wireless structural health monitoring system
with multithreaded sensing devices: design and validation. Structure and
Infrastructure Engineering, Vol. 3, No. 2, pp. 103-120
Wang, Y., Swartz, R. A., Lynch, J. P., Law, K. H., Lu, K.-C. & Loh, C.-H. (2007b).
Decentralized civil structural control using real-time wireless sensing and
embedded computing. Smart Structures and Systems, Vol. 3, No. 3, pp. 321-340
Wang, Y. & Law, K. H. (2011). Structural control with multi-subnet wireless sensing
feedback: experimental validation of time-delayed decentralized H∞ control design.
Advances in Structural Engineering, Vol. 14, No. 1, pp. 25-39
Weng, J.-H., Loh, C.-H., Lynch, J. P., Lu, K.-C., Lin, P.-Y. & Wang, Y. (2008). Output-only
modal identification of a cable-stayed bridge using wireless monitoring systems.
Engineering Structures, Vol. 30, No. 7, pp. 1820-1830
Whelan, M. J. & Janoyan, K. D. (2009). Design of a robust, high-rate wireless sensor network
for static and dynamic structural monitoring. Journal of Intelligent Material Systems
and Structures, Vol. 20, No. 7, pp. 849-863
Wong, K.-Y. (2004). Instrumentation and health monitoring of cable-supported bridges.
Structural Control and Health Monitoring, Vol. 11, No. 2, pp. 91-124
Yao, J. T. P. (1972). Concept of structural control. Journal of Structural Division, ASCE, Vol. 98,
No. 7, pp. 1567-1574
Zhu, D., Yi, X., Wang, Y., Lee, K.-M. & Guo, J. (2010). A mobile sensing system for structural
health monitoring: design and validation. Smart Materials and Structures, Vol. 19,
No. 5, pp. 055011
 20

Extending Applications of
Dielectric Elastomer Artificial
Muscles to Wireless Communication Systems
Seiki Chiba and Mikio Waki
Chiba Science Institute
Wits Inc.
Japan

1. Introduction
Electro active polymers (EAPs) are used for actuators that can electrically control their
motions to resemble those of actual muscles. Thus, they are called artificial muscles. In
addition, since EAPs are often made of flexible materials, they have also come to be called
“soft actuators” in recent years. There are many types of EAPs such as dielectric elastomers
(Perline & Chiba, 1992a), ionic polymer-metal composites (Oguro et al., 1999),
electroconductive Polymers (Otero & Sansinera, 1998), and ion polymer gels (Osada et al.,
1992b). Figure 1 shows typical EAPs.

Fig. 1. Typical electro active polymers (EAPs)

EAP can be generally classified into two categories: elctrochemical polymers and fileld
activated polymers (Kornbluh et al., 2004a). Electrochemical polymers use electrically driven
mass transport of ions or electrically charged species to effect a charge in the shape (or vice
versa). Field-activated polymers use an electric field to effect a shape change by acting
directly on charges within the polymer (or vice versa). Each type of EAP has advantages and
disadvantages for the application to wiress communications. Electrochemical polymers
436 Recent Advances in Wireless Communications and Networks

typically can exert relatively high pressures and can be driven by low voltages. However,
they are relatively slow and limited in size (since they are dependent on molecular
trnasport), require high current and relatively energy inefficient. They can operate best over
a narrow range of temperatures and must often be kept moist (Kornbluh et al., 2004a). In
contrast, field-activated polymers can be fast, efficient ,and relatively insensitive to
temperature and humidity fluctuations. These polymers can operate at relatively high
voltages and low currents, that usually requires additional voltage conversion components
but makes the size and capacity of wires and interconnects lighter and less ctritical
(Kornbluh et al., 2004a).
A type of field-activated EAP transducer that embodies the desirable proprties of polymer is
dielectric elastomers (Pelrine et al., 2000).
Dielectric elastomer is a new transducer technology uses rubber like polymer (elastomer) as
actuator materials. They have been gaining attention as technologies that have reached the
practical use level as actuators and even as devices that can generate electricity efficiently
(Chiba et al., 2008a).
The present paper examines the possibilities of frequency-variable antennas that utilize the
actuator mode of dielectric-type artificial muscles, and sensor networks that utilize this
electric generator mode (Chiba et al., 2007a; Chiba et al., 2008a).

2. Background on dielectric elastomer artificial muscles

Dielectric elastomer is a new smart material with characteristics and properties not seen in
other materials. The basic element of dielectric erastomer is a very simple structure
comprised of thin polymer films (elastomers) sandwiched by two electrodes made of a
flexible and elastic material, and can operate as an electric control actuator.

Fig. 2. Performance of dielectric elastomer is similar to that of natural muscle

Extending Applications of Dielectric Elastomer
Artificial Muscles to Wireless Communication Systems 437

Using a dielectric elrastomer actuator makes it possible to achieve a highly efficient

transduction from electric energy into mechanical energy (the theoretical transduction
efficiency is 80-90%, which translates into a considerable energy saving compared to other
actuator technologies such as electric motors with gearboxes. At the material level, this
material has fast speed of response (over 50,000 Hz has been demonstrated for small
strains), with a high strain rate (up to 380% as shown in Photo 1), high pressure (up to 8
MPa), and power density of 1 W/g (for comparison, human muscle is 0.2 W/g and an
electric motor with gearbox is 0.05 W/g) (Stanford et al., 2004a).
The energy density of dielectric elastomer has reached 3.4J/g, about 21 times that of single-
crystal piezoelectrics and more than two oreders of magnitude greater than that of most
commercial actuators (Pelrine et. al., 2000a). As can be seen in Figure 2, dielectric elastomers
not only outperform existing actuator technologies in various areas but also are similar to
natural muscle in that they fill the “actuator gap” between other actuation technologies,
(Chiba, 2002). That is, dielectric elastomers have an actuation pressure/density that is bigger
than that of electrostatic actuators and magnetic actuators, and cause strains that are bigger
than that of piezo electric actuators and magneto strictive actuators.

Photo 1. Acrylic elastomers showing 380% linear strain

2.1 Principle of operation of dielectric elastomers

Dielectric elastomer tranducers are based on the electromechanical response of an
elastomeric dielectric film with compliant electrodes on each surface. Actuators based on
dielectric elastomers technology operate on the simple principle shown in Figure 3. When a
voltage is applied across the compliant electrodes, the polymer shrinks in thickness and
expands in area.
The net volume change of the polymer materials that we investigate is small because of their
high bulk moduli. Therefore, the electrodes must be compliant, to allow the film to strain.
The observed response of the film is caused primarily by the interaction between the
electrostatic charge on the electrodes. Simply put, the opposite charges on the two electrodes
attract each other, while the like charges on the electrodes repel each other. Using this
simple electrostatic model, we can derive the effective pressure produced by the electrodes
on the film as function of the applied voltage. The pressure, ρ, is

ρ = ε εoE2 = ε εo(V/t)2 (1)

438 Recent Advances in Wireless Communications and Networks

where εo and ε are the permittivity of free space (8.85 x 10-12 F/m) and the relative permittivity
(dielectric contact) of polymer, respectively; E is the applied electric field in V/m; V is the
applied voltage; and t is the film thickness. The response of the polymer is functionally
similar to that of electrostrictive polymers, in that the response is directly related to the
square of the applied electric field,

Fig. 3. Principle of operation of dielectric elastomers

Two observation made from Equation 1 clarify the difference between Maxwell stress
actuation and the use of conventional air-gap electrostatic actuators. First, ε for polymers is
typically in the range 2-12, whereas for air ε is 1. Thus the actuation pressure is increased
substantially via polymers rather than air the same electric field. Another difference is that
typical air-gap actuators have an additional factor of 0.5 in their equivalent pressure
expression, i.e., the polymers double the actuation pressure independent of the dielectric
constant. The reason for this difference is that the polymers can stretch in area rather just
contract in thickness. Polymers have two modes of converting electrical to mechanical
energy. In contrast to polymers, air-gap actuators are typically made of rigid materials that
can convert electrical to mechanical energy via only one mode of motion, such as the
convergence of opposite electrodes.
Dielectric elastomers also have other advantages over air-gap electrostatic actuators, even
though both are based on electrostatic force. Several polymers have been identified with
breakdown strength of 300 MV/m or more in thin films, but breakdown strength this high
are difficult to achieve consistently in air-gap electrostatic devices.
As mentioned above, the three effects, i.e., “two-mode coupling,” “high dielectric
susceptibility,” and “high electric strength,” greatly contribute to the actuation pressure of
the dielectric elastomers.

2.2 Development summary of dielectric elastomer actuators

The elastomer has excellent workability which enables the shape design of devices with
sizes from micrometers to several meters. Also, as elastomers are light and deform like
rubbers, they can show flexible movements like bionic actions. They can express “flexible
and natural feeling” which systems with motors cannot imitate. A wide array of proof-of-
Extending Applications of Dielectric Elastomer
Artificial Muscles to Wireless Communication Systems 439

principle devices for use in leg robots (see Fig. 4), swimming robots, snakelike robots,
compact inspection robots, geckolike robots for climbing up perpendicular walls or across
ceilings, and flying robots, as well as in achieving compatibility with living organisms are
currently developed (Stanford et. al., 2004b). The main feature of the dielectric elastomers is
that they do not use any gears and cams, thus enabling high efficiency and safe and smooth
driving even if the speed or direction of movement are suddenly changed.

Linear strain Bend Rotation

(a)

(b)
Fig. 4. Biologically inspired robots powered by dielectric elastomer rolls (Pei et al, 2003;
Chiba et al, 2006a). (a) Role Actuator Having 3-DOF (b) Application example to a robot: it
enables sideways stepping like a crab without turning around, when it collides with wall
The 3-DOF actuator may be used as actuator for variable antenna of wireless communication
device (see section 3 “Proof-of-principle experiment on a frequency-variable antenna
utilizing the actuator mode of dielectric-type artificial muscles”).
Moreover, as this actuator has a wide dynamic range (DC to several tens of kHz), its
applications to speakers and vibrational devices have been advanced (see Fig. 5) (Chiba et
al., 2007a).
This device may be suitable for vibrators and speakers of wireless communication devices.
In addition, as there is a direct proportionality between the change in the capacitance and
elongation of dielectric elastomer actuators, they can be used for pressure- and position-
sensors (see Fig. 6). It may be possible to use the sensor function of dielectric elastomers to
pick up electric waves for wireless communication devices.
440 Recent Advances in Wireless Communications and Networks

Fig. 5. Structure of speaker using dielectric elastomer (The black shaped part is dielectric
elastromer) (Chiba et al., 2007a)

Fig. 6. Linear relation between capacitance and stroke of actuator (Kornbluh et al., 2004b)

3. Proof-of-principle experiment on a frequency-variable antenna utilizing the

actuator mode of dielectric-type artificial muscles
The popularization of mobile telephones has brought wireless technology even closer to our
daily lives. In recent years, improvements in integrated technology of electronic circuits and
the increasing multi-functionality of mobile terminals have led to the inclusion of a
multitude of diverse formats such as 3GPP, wireless LAN, Bluetooth, digital TV, etc., in
single mobile communication devices. Since these communication formats all use different
frequencies, it is necessary either to install a separate antenna for each wavelength, or use
one antenna that can accommodate multiple frequencies.
Extending Applications of Dielectric Elastomer
Artificial Muscles to Wireless Communication Systems 441

Methods to create an antenna that is compatible for multiple frequencies include integrating
antenna elements that can respond to multiple frequencies, and using an antenna that is
shaped so that it can tune to a broad range of frequencies. The easiest method is to change
the length of the antenna element, but because this changes the length of the antenna device,
it requires equipment such as motors and gears. This makes it difficult to use as a small,
lightweight frequency-variable antenna.
One way to resolve these problems may be to create a lightweight frequency-variable
antenna with a simple structure by utilizing dielectric-type artificial muscles in the actuator
part of a variable antenna.It may be possible to change the position of the reflection element
and/or changing the length of dipolar- or monopolar antenna elements. Furthermore, by
forming this structure onto polymers, it is possible to create a changeable-type planar
antenna that can be installed in small, lightweight portable devices.
The present experiment corroborated the possibility of creating such variable-type antennas
by using artificial muscle to change the length and tuning frequency of a monopolar
antenna.
The variable-type monopolar antenna used in this experiment had a very simple structure. It
was composed of a radial section that was attached to the dielectric artificial muscle
actuator, and an antenna element section that was installed vertically on the core. (see
Photo 2)

Photo 2. A frequency-variable antenna utilizing the actuator mode of dielectric elastomer

artificial muscles
By changing the control voltage that was applied to the artificial muscle, a structure was
created in which it was possible to change both the length of the antenna element part that
was thrust out from the radial section and the tuning frequency.
In actuators that use dielectric artificial muscles, a thin (0.05 mm) elastomer film was
attached to a 10 cm-diameter circular frame. By attaching two of these elastomers onto this
frame, it became a diaphragm type with the cores of the elastomers attached to one another.
The total weight, including the structural parts, was about 20 g.
The frequencies used in the experiment were in the 2.45 GHz band that is currently used in
3GPP, wireless LAN, and so on. The length L of the monopolar antenna element at a
frequency of 2.45 GHz was 1/4 of the wavelength λ (122.4 mm), or 122.4/4 = 30.6 mm, and
442 Recent Advances in Wireless Communications and Networks

the changeable width of the actuator was 4 mm. This made it possible to change the tuning
frequency within a range of about 300 MHz.
The change in the tuning frequency was confirmed by measuring V. S. W. R. (Voltage
Standing Wave Ratio) using a network analyzer (Photo 3).

(a) Before change (b) At the time of the maximum change

Photo 3. Measurement of V. S. W. R. (The setting frequency range of a network analyzer:
start frequency, 1.8 GHz and stop frequency, 2.9 GHz)
In this experiment, a diaphragm actuator for artificial muscle speakers was used, but this
system was not smart, because the muscle part was too large. However, since the purpose of
this experiment was to make the resonant frequency of a non-directional antenna variable
by changing the length of the antenna element, a monopole antenna, which has the simplest
structure, and artificial diaphragm muscles were used.
In our next experiment, we plan to change the direction of electric wave radiation by
varying the installation angle of a directional antenna with roll-type artificial muscles.
In another words, the plan call for conducting an experiment to vary the directivity inside
the vertical face of the antenna by making a model (ground plane) antenna by changing the
wire in the radial part, and enabling the angle of attachment to the radial part to be changed
by the roll-type artificial muscle. If such a variable antenna can be put to practical use, then
it might be possible to create a system where the antenna can automatically be varied to
match a more optimal electric wave environment, and even a small amount of electric
power can be used to construct a suitable electric wave environment.
Furthermore, plans are being drawn for conducting an experiment on a planar antenna
whose directivity and tuning frequency can be changed by using the dielectric-type artificial
muscle to transform the antenna formed on the polymer. In the near future, by using
variable antennas whose shape changes to match the use in mobile telephones, personal
computers, etc., it may be possible to create a pleasant wireless communications
environment with just a little bit of electrical power.

4. Sensor network that utilizes the power generation mode of a dielectric

elastomer artificial muscle
Another working mode of the dielectric elastomer artificial muscle is the power generation
mode. This is operatively the opposite of the actuator function. By adding external power to
the dielectric type artificial muscle, the shape can be changed, and the increased static
Extending Applications of Dielectric Elastomer
Artificial Muscles to Wireless Communication Systems 443

electrical energy produced therefrom can generate electricity. Since this power generation
phenomenon is not dependent on the speed of transformation, its power generation device
can generate electric energy by utilizing natural energies such as up-and-down motions of
waves, slowly flowing river water, human and animal movements, and vibration energies
produced from vehicles and buildings.

4.1 Principal of the power generation mode

The operation principle in the generator mode is the transformation of mechanical energy
into electric energy by deformation of the dielectric elastomer (Ashida et al,:2000b).
Functionally, this mode resembles piezoelectricity, but its power generation mechanism is
fundamentally different. With dielectric elastomer, electric power can be generated even by
a slow change in the shape of dielectric elastomer, while for piezoelectric devices impulsive
mechanical forces are needed to generate the electric power. Also, the amount of electric
energy generated and conversion efficiency from mechanical to electrical energy can be
greater than that from piezoelectricity (Chiba et al,. 2007a). Fig.7 shows the operating
principal of dielectric elastomer power generation.

Fig. 7. Operating principle of dielectric elastomer power generation

Application of mechanical energy to dielectric elastomer to stretch it causes compression in
thickness and expansion of the surface area. At this moment, electrostatic energy is
produced and stored on the polymer as electric charge. When the mechanical energy
decreases, the recovery force of the dielectric elastomer acts to restore the original thickness
and to decrease the in-plane area. At this time, the electric charge is pushed out to the
electrode direction. This change in electric charge increases the voltage difference, resulting
in an increase of electrostatic energy.

C =ε0εA/t =ε0εb/t2 (1)

where ε0 is the dielectric permittivity of free space, ε is the dielectric constant of the polymer
film, A is the active polymer area, and t and b are the thickness and the volume of the
polymer. The second equality in Equation (1) can be written because the volume of
elastomer is essentially constant, i.e., At = b = constant.
The energy output of a dielectric elastomer generator per cycle of stretching and contraction
is

E = 0.5C1V b 2 (C1/C2-1) (2)

444 Recent Advances in Wireless Communications and Networks

where C1 and C2 are the total capacitances of the dielectric elastomer films in the stretched
and contracted states, respectively, and V b is the bias voltage.
Considering then changes with respect to voltages, the electric charge Q on a dielectric
elastomer film can be considered to be constant over a short period of time and in the basic
circuit. Since V = Q/C, the voltages in the stretched state and the contracted state can be
expressed as V1 and V2, respectively, and the following equation is obtained:

V2 = Q/C2 = (C1/C2) (Q/C1) = (C1/C2) V1 (3)

Since C2 < C1, the contracted voltage is higher than the stretched voltage, corresponding to
the energy argument noted above. The higher voltage can be measured and compared with
predictions based on the dielectric elastomer theory. In general, experimental data based on
high impedance measurements are in excellent agreement with predictions. When the
conductivity is assumed to be preserved in the range of electric charging, Q remains
constant.

(a)

(b)
Fig. 8. Voltage for compression of dielectric elastomer and measurement circuit. (a) Typical
scope trace from contraction of dielectric elastomer. Voltage spike occurs at contraction and
gradually back to (stretched) voltage due to load resistance. (b) Measurement circuit of
generated energy
Extending Applications of Dielectric Elastomer
Artificial Muscles to Wireless Communication Systems 445

Figure 8(a) shows a typical scope trace from contraction of dielectric elastomer. Figure 8(b)
shows a simplified circuit for oscilloscope measurement of voltage. The voltage peak
generated for one cycle is typically on the order of a few ms to several tens of ms for a
piezoelectric element. However, in the case of dielectric elastomer, the peak width is on the
order of 150-200 ms or longer (Chiba et al., 2008a). The long power-generation pulse
duration of dielectric elastomer can allow for the direct use of generated energy for activities
such as lighting LEDs. This can even power wireless equipment that is evolving today at a
rapid pace. In continuous cyclical motions, it is easy to continuously obtain electrical energy
by using flat and smooth circuits, even with gentle kinetic energy below a few Hz (Chiba et
al., 2007b)

4.2 Application of dielectric elastomer generator to wireless communication system

In a power generation experiment, a thin artificial muscle film (25 cm long x 5cm wide,
weight about 0.5 g) attached a human arm was able to generate 20 mJ of electrical energy
with one arm movement. It is also possible to make them generate electricity putting up
dielectric elastomers besides the arm to the side and the chest of the body (See Fig. 9a).

(a)

Streched state Relaxed states

(b)

Fig. 9. Harvesting energy system from human body. (a) Conceptual rendering of dielectric
elastomers put up to side and chest of arm and body: (b) Stretched state of dielectric
elastomer (left) and Relaxed state of the elastomer (right)
446 Recent Advances in Wireless Communications and Networks

Furthermore, in an experiment using different power generation equipment, artificial muscle

film attached to the bottom of a shoe was verified to generate electricity when the artificial
muscle was distorted while walking. When an adult male took one step per second, one
shoe was able to produce about 1 W of electrical power. (Harsha et al., 2005)

Fig. 10. Shoe generator

This confirmed that by utilizing human movement, enough electrical power could be
obtained to recharge batteries for mobile telephones and similar devices (Chiba et al., 2008).
In addition, electrical energy from the movements of animals could be used to construct
livestock management systems. Other applications of animal-generated energy being
investigated include scientific surveys of ecosystems of migratory birds and fish, among
others.
In an experiment using a diaphragm actuator, electrical power output of about 0.12 – 0.15W
was obtained by pressing the center of a roughly 1 g, 8 cm-diameter EPAM a few
millimeters one time per second (Chiba et al., 2007a). Using the same equipment, the electric
power generated was able to illuminate 6 LEDs, and by combining this with a wireless
system, it became possible to turn a device on and off from a remote location.
In such ways, dielectric elastomer artificial muscles can supply electrical power only when
mechanical energy is obtained, and it is possible to simultaneously act as a switch that
detects power sources and motion. Consequently, it may possible to easily create wireless
networks, with simple components that do not require batteries (Chiba et al., 2007a).
In recent years, global warming and accompanying abnormal weather have begun to have
an impact on our daily lives. To protect ourselves from the disasters brought about by
abnormal weather, it is important to thoroughly understand the current situation, that is,
how the global environment is changing.
The monitoring of the global environment has been done by various countries on their own,
but to monitor environmental changes on a global scale it will be necessary to build wide-
ranging sensor networks. One of the major issues with that, however, is that there is no
good method for obtaining electrical energy for running this system. Presently, many if not
most of these sensor systems are powered by solar batteries, but in some locations and
during some seasons the daylight hours are extremely short, and in maritime and desert
Extending Applications of Dielectric Elastomer
Artificial Muscles to Wireless Communication Systems 447

areas salt and dust can dramatically reduce the electrical output. All this makes it difficult to
maintain a stable sensor system.

(a)

(b)
Photo 4. Small scale power generation device. a) Cartridge of used for small generator.
The black ring-shaped part is dielectric elastomer. b) A power of approximately 0.12 W can
be generated, by pushing the central part of dielectric elastomer by 3- 4 mm once a second
As one way of resolving these issues, power generation systems that utilize artificial
muscles to generate power through transformation alone are attracting attention. Already,
experiments using wave power to generate electricity have been able to produce a few watts
of electrical energy with small artificial muscle power generation equipment loaded onto
448 Recent Advances in Wireless Communications and Networks

weather observation buoys, (see photo 6 and fig. 11) and this has also been confirmed to
recharge batteries (Chiba et al., 2009).

Photo 5. Small scale power generation device & LED controlled by wireless signals

Photo 6. Dielectric elastomer generator on the test buoy

Extending Applications of Dielectric Elastomer
Artificial Muscles to Wireless Communication Systems 449

Fig. 11. Electricity generated by ten centimeter-high waves

In other experiments (see photo 7), electrical energy has been obtained from flowing water
in a laboratory (Chiba et al., 2007a). The flow of water rotates the water-mill, and the
rotational motion induces the deformation of the dielectric elastomer to generate electrical
energy. Figure 12 shows the conceptual rendering of water mill generator using dielectric
elastomer (Chiba et al., 2007a).
Furthermore, the results of simulations based on conceptual designs of flag-type power
generation equipment using artificial muscles have indicated that there is little loss from the
fluttering of flags and that it is possible to generate electric power with a high rate of
efficiency (Chiba et al., 2007b).

Photo 7. Water mill generator using dielectric elastomer

450 Recent Advances in Wireless Communications and Networks

Fig. 12. Conceptual rendering of water mill generator using dielectric elastomer (Chiba et al.,
2010)

Fig. 13. Conceptual rendering of flag-type power generation

4.3 Analysis of power generation cost

Even without dielectric elastomer technology, ocean wave power is beginning to flourish in
several countries. These ocean wave power systems typically use hydraulic pistons that are
pumped by the wave action. The hydraulic fluid flows through a transmission and then a
turbine to spin a rotary electromagnetic generator. When these systems are successfully
developed for commercial use, the unit price of a power generation of kWh is estimated to
be about 20 US Cents (Chiba et al., 2008b). These wave power systems are typically designed
for ocean waves exceeding 2 - 3 m in height. At significantly smaller wave heights, the
systems become less economically attractive (Miyazaki et al, 2007).
Extending Applications of Dielectric Elastomer
Artificial Muscles to Wireless Communication Systems 451

Fig. 14. Conceptual rendering of wave power generator system using dielectric elastomers
Because of its simplicity, efficiency, and size scalability, we believe that dielectric elastomer-
based wave generator systems can be attractive not only for large wave applications but for
many applications where the waves are much smaller. An estimate based on data from our
sea trial demonstration experiments has shown that even in seas where the wave height is
only 1 m throughout the year (e.g., the sea close to Japan), if there are spaces of
approximately 500 m in length and 10 m in width, the establishment of a sea-based facility
generating 6 MW of power is possible (Chiba et al., 2008b). This is a useful amount of
power, be it for general use or for providing energy for nearby residential or industrial
needs. The ability to produce the power where it is needed can eliminate the losses and costs
associated with power transmission over long distances and make wave power even more
attractive. The power generation efficiency estimated on the basis of the data obtained from
in-tank experiments in 2006 (Chiba et al, 2006b) and ocean demonstration experiments in
2007 (Chiba et al, 2008a) and 2008 (Chiba et al, 2009) is approximately 19 US cents/kWh. In
the near future, we expect that the electric power generation per unit mass or volume of
dielectric elastomer material can double, and that the expected power generation cost per
452 Recent Advances in Wireless Communications and Networks

kilowatt-hour is 5 - 7.5 US cents. This value is comparable to that for fossil fuel thermal
power plants. Of course, the wave power systems have the additional benefit of not
releasing any pollution or greenhouse gasses.

5. Future of dielectric elastomer systems to wireless communication

The variable antenna technologies with artificial muscles have high expectations to apply to
not only data communications for mobile phones and personal computers but also wireless
sensor systems which monitor various data concerning weather conditions and environments.
In the future, the combination of these artificial muscle power-generating systems with
various sensing systems will make it possible to conduct sensing on a global scale, and may
even make a significant contribution to the creation of systems that will protect human lives
from natural disasters that have so far been difficult to predict.
Various power generating systems can be set up in each place on the earth as shown in
Figure 15 in order to create wire sensor networks.

Fig. 15. Sites where power generation using dielectric elastomers is possible and conceptual
rendering of the generation systems: (a) Wind Power Generator on tops of buildings (Chiba
et al., 2007b) (b) Water Mill Generators (Chiba et al., 2007a) (c) Waste energy Generators
(Chiba et al., 2011) (d) Drain Generators (Chiba et al., 2011) (e) Wind Power Generators for
Personal Houses (f) Solar Heat Generators (Chiba et al., 2007b) (g) Wave Generators (Chiba
et al., 2006; Chiba et al., 2008a) (h) Wave Generators in Ocean (Chiba et al., 2008a) (i) Hydrogen
Production Plant (Chiba et al., 2008b)

6. References
Chiba S., et al. (2007a). Extending Applications of Dielectric Elastomer Artificial Muscle.
Proceedings of SPIE, San Diego, March 2007.
Extending Applications of Dielectric Elastomer
Artificial Muscles to Wireless Communication Systems 453

Chiba S.; Stanford S., Pelrine R., Kornbluh R., and Prahlad H. (2006a), Electroactive Polymer
Artificial Muscle, JRSJ, Vol. 24, No.4, pp 38-42. 2006.
Chiba S,; Prahad H, Pelrine R, Konbluh R, Stanford S and Eckerle J. (2006b). Electro Power
Generation Using Electro active Polymers (EPAM). Proceedings of 15th Japan
Institute of Energy Conference (Kogakuuin University, Japan) JIE pp 297-298, July
2006.
Chiba S.; Pelrine R., Kornbluh R., Prahlad H., Stanford S., & Eckerle J. (2007b). New
Opportunities in Electric Generation Using Electroactive Polymer Artificial Muscle
(EPAM). J. Japan. Inst. Energy, Vol. 86, No. 9, pp. 38-42, 2007.
Chiba S (2002), Dielectric Elastomer for MEMS and NEMS and Toward the Future. Electro
Packing Technology, Vol.18, No. 1, pp 33-38, 2002.
Chiba S.; Waki M., Kormbluh R., & Pelrine R. (2008a). Innovative Power Generators for
Energy Harvesting Using Electroactive Polymer Artificial Muscles, Electroactive
Polymer Actuators and Devices (EAPAD), ed. Y. Bar-Cohen. Proceedings of SPIE.
Vol. 6927, 692715 (1-9), San Diego, March 2008.
Chiba, S., Kornbluh R., Pelrine R., and Waki M. (2008b) “Low-cost Hydrogen Production
From Electroactive Polymer Artificial Muscle Wave Power Generators”, Proceedings
of World Hydrogen Energy Conference, Brisbane, Australia, June 16-20, 2008.
Chiba, S., Waki M., Kornbluh K., and Pelrine R.. (2009). Innovative Wave Power Generation
System Using EPAM. Proceedings of Oceans’ 09, Bremen, Germany, May 2009.
Chiba S. and Waki M. (2011). Recent Progress in Dielectric Elastomers (Harvesting Energy
Mode and High Efficient Actuation Mode), To be published in Clean Tech, Nihon
Kogyo Shuppan, Tokyo, Japan, April, 2011.
Harsha P, Kornbluh R, Pelrine R, Stanford S, Eckerle J and Oh S. (2005). Polymer Power:
Dielectric elastomers and their applications in distributed actuation and power
generation. Proceedings of ISSS 2005, International Conference on Smart Materials
Structures and Systems. Bangalore, India.
Kornbluh R., Pelrine R., and Chiba S. (2004b). Silicon to Siliocon: Stretching the Capabilities
of Micromachines with Electroactive polymers, IEEJ, Vol.124, No. 8, 2004, ISSN
1341-8939.
Miyazaki T and Osawa H. (2007). Search Report of Wave Power Devices Proceedings of
Spring Conference of the Japan Socity of Naval Architects and Ocean Engineers, No.4
pp43-46, April 2007.
Pelrine R., and Chiba S. (1992a). Review of Artificial Muscle Approaches. Proceedings of
Third International Symposium on Micromachine and Human Science, Nagoya, Japan,
June 1992.
Pelrine R.; Kornbluh K., Pei Q., & Joseph J. (2000a). High Speed Electrically Actuated
Elastomers with Over 100% Strain. Science 287: 5454, pp 836–839, 2000.
Pei Q., Rosenthal M., Pelrine R., Stanford S., and Kornbluh R (2003) Multifunctional
electroelastmer roll actuators and their application for biomimetic walking robots,
proceedings of SPIE, Smart Structures and mterials, Electroactive Polymer
Actuators and Devices (EAPAD), ed. Y. Bar-Cohen, San Diego, CA, March 2003.
Stanford S, Bonwit N, Pelrine R, Kornbluh R, Pei Q and Chiba S (2004b). Electro Polymer
Artificial Muscle (EPAM) for Biomimetics Robots. Proceedings of 2nd Conference on
Artificial Muscles. AIST Kansai Center, Osaka, Japan, 2004.
454 Recent Advances in Wireless Communications and Networks

Oguro K., Fujiwara N., Asaka K., Onishi K. and Sewa S. (1999). Polymer electrolyte actuator
with gold electrodes. Proceedings of the SPIE’s 6th Annual International Symposium on
Smart Structures and Materials, SPIE Proc. Vol. 3669,(1999), pp. 64-71.
Otero F. and Sansiñena M. (1998). Soft and wet conducting polymers for artificial muscles”,
Advanced Materials, 10 (6), (1998) pp. 491-494.
Osada Y., Okuzaki H. and Hori H. (1992b). A polymer gel with electrically driven motility”,
Nature, Vol. 355, pp. 242-244, (1992).
Ashida A., Ichiki M., Tanaka T. and Kitahara T. (2000b). Power Generation Using Piezo
Element: Energy Conversion Efficiency of Piezo Element”, Proc. of JAME annual
meeting, pp.139-140, (2000).