University of Alberta

Aggregate Interference Analysis for Cognitive Radio Networks

by

Sachitha Panduka Kusaladharma

A thesis submitted to the Faculty of Graduate Studies and Research

in partial fulfillment of the requirements for the degree of

Master of Science
in
Communications

Electrical and Computer Engineering

©Sachitha Panduka Kusaladharma

Fall 2013
Edmonton, Alberta

Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis
and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is
converted to, or otherwise made available in digital form, the University of Alberta will advise potential
users of the thesis of these terms.

The author reserves all other publication and other rights in association with the copyright in the thesis and,
except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or
otherwise reproduced in any material form whatsoever without the author's prior written permission.
Dedicated to my beloved parents...
Abstract

In underlay Cognitive Radio (CR) systems, secondary users may transmit while pri-
mary transmissions take place. However, the interference aggregates on the primary
system, and performance degradation will occur. Thus, quantifying and modeling
of the aggregate interference is important in characterizing the network’s perfor-
mance. An interference analysis for a Poisson process of CR nodes in a finite area
is undertaken while incorporating channel parameters, and a simple Gamma model
for the aggregate interference is proposed. Furthermore, a multiple-ring model to
replace the annular underlay model will be developed. This model is versatile and
mathematically tractable, while being highly accurate. Moreover, the interference
from the nearest interferer node is analyzed, and situations where the nearest inter-
ferer’s interference can approximate the aggregate interference are identified. An
outage and asymptotic analysis is carried out for the models. The effect of different
system and channel parameters are shown within the numerical results.

∼
Acknowledgements

I would like to thank Dr Chintha Tellambura for being an excellent supervisor and
guiding me on my path of research. I also wish to thank all my family and friends
and other teachers who have helped me to come to this stage.

∼
Contents

1 Introduction 1
1.1 Wireless Communications . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Methods to improve spectral efficiency . . . . . . . . . . . . . . . . 2
1.2.1 Cognitive radio . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.2 Heterogeneous networks . . . . . . . . . . . . . . . . . . . 5
1.2.3 Small cell networks . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Contributions and Outline . . . . . . . . . . . . . . . . . . . . . . . 6

2 Background 9
2.1 The wireless channel . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Multipath fading and the Doppler effect . . . . . . . . . . . 9
2.1.2 Small scale fading models . . . . . . . . . . . . . . . . . . 11
2.1.3 Shadowing . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.4 Path loss . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.5 Power control . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Spatial distribution models . . . . . . . . . . . . . . . . . . . . . . 15
2.2.1 Poisson point process . . . . . . . . . . . . . . . . . . . . . 16
2.2.2 Binomial point process . . . . . . . . . . . . . . . . . . . . 17
2.3 Annular underlay networks . . . . . . . . . . . . . . . . . . . . . . 18
2.3.1 Aggregate interference . . . . . . . . . . . . . . . . . . . . 20
2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3 Aggregate Interference Analysis 22

 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Aggregate interference of annular underlay networks . . . . . . . . 23
3.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.2 System model . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.3 Interference Statistics . . . . . . . . . . . . . . . . . . . . 24
3.2.4 Performance Analysis . . . . . . . . . . . . . . . . . . . . 26
3.3 Gamma model approximation . . . . . . . . . . . . . . . . . . . . 28
3.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.2 The Gamma Approximation . . . . . . . . . . . . . . . . . 29
3.4 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4.1 Numerical results of Aggregate interference for annular un-
derlay networks . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4.2 Numerical results for the Gamma approximation . . . . . . 35
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4 Approximation models for the aggregate interference 40

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2 A multiple-ring model for underlay interference . . . . . . . . . . . 41
4.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2.2 Proposed System Model . . . . . . . . . . . . . . . . . . . 42
4.2.3 Interference Statistics . . . . . . . . . . . . . . . . . . . . . 43
4.2.4 Suitability as a stand-alone system model . . . . . . . . . . 47
4.2.5 Performance analysis . . . . . . . . . . . . . . . . . . . . . 49
4.3 Nearest Interferer approximation . . . . . . . . . . . . . . . . . . . 55
4.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.3.2 System model . . . . . . . . . . . . . . . . . . . . . . . . 56
4.3.3 Distribution of the nearest interferer node . . . . . . . . . . 57
4.3.4 Derivation of the MGF . . . . . . . . . . . . . . . . . . . . 60
4.3.5 Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3.6 Outage Analysis . . . . . . . . . . . . . . . . . . . . . . . 64
4.4 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
 4.4.1 Numerical results of the multiple-ring model . . . . . . . . 65
4.4.2 Numerical results for the nearest interferer approximation . 68
4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

5 Conclusions and Future Research Directions 74

5.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.2 Future Research Directions . . . . . . . . . . . . . . . . . . . . . . 75

Bibliography 77
List of Tables

4.1 Error statistics for R = 5 and M = 20. . . . . . . . . . . . . . . . . 54

4.2 Error statistics for R = 1 and M = 100. . . . . . . . . . . . . . . . 54
List of Figures

1.1 Spectral efficiency over the years. Source : Qualcomm, from [1] . . 2

2.1 Random spatial distribution of interferer nodes. . . . . . . . . . . . 17

2.2 Clustered PPP. The right image shows the original set of points, and
the image on the left shows the clustered points after the transfor-
mation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Annular underlay network. Legend: black dot = interferer nodes,
black square = PR, yellow square = PT. . . . . . . . . . . . . . . . 19

3.1 The outage probability vs Pp and Ps , under γT h = 1, α = 2, σ = 2,

σn2 = 0.001, β = 0.0001, R = 30, RG = 15 , RE = 100. . . . . . . 31
3.2 The outage probability vs Pp and Ps , under γT h = 1, α = 2, σ = 0,
σn2 = 0.001, β = 0.0001, R = 30, RG = 15, RE = 100. . . . . . . 32
3.3 The exact and asymptotic outage probability vs the normalized trans-
mit power Pp , for different values of σ and α under σn2 = 0.001,
β = 0.0001, Ps = 30dBm, R = 30, RG = 15, RE = 100, and
γT h = 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.4 The outage probability vs interferer density (β) for different values
of σ and α under Ps = 30 dBm, Pp = 70 dBm, σn2 = 0.001, R =
30, RG = 15, RE = 100, and γT h = 1. . . . . . . . . . . . . . . . 34
3.5 The outage probability vs interferer density (β) for different values
of σ and α under Ps = 30 dBm, Pp R−α = 25 dBm, σn2 = 0.001,
RG = 25, RE = 500, and γT h = 1. . . . . . . . . . . . . . . . . . 35
3.6 The outage probability vs primary transceiver distance (R) for dif-
ferent values of σ and α under Ps = 30 dBm, γT h = 1, Pp =
70 dBm, RG = 15, RE = 100, and σn2 = 0.001. . . . . . . . . . . . 36
3.7 The CDF of the aggregate interference and the Gamma approxi-
mation for different values of β under Ps = 30 dBm, RG = 15,
RE = 100, α = 2, and no shadowing. . . . . . . . . . . . . . . . . 37
3.8 The PDF of the Gamma approximated aggregate interference for
different values of β under Ps = 30 dBm, RG = 15, RE = 100,
α = 2, and no shadowing. . . . . . . . . . . . . . . . . . . . . . . 38
3.9 The PDF of the Gamma approximated aggregate interference for
different values of RG , RE , and σ under Ps = 30 dBm, β = 1 ×
10−3, and α = 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

4.1 System model. Legend: Black circle = interferer nodes, black

square = PR, R2 = 2R, and RM = MR. . . . . . . . . . . . . . . . 42
4.2 Linear interferer network. Legend : black dot = interferer nodes,
black square = PR. . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3 Square shaped interferer network. The shaded area contains the
interferers. Legend : black square = PR. . . . . . . . . . . . . . . . 50
4.4 System model. Legend: black circle = interferer nodes, black square
= PR. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.5 frmin (x) for varying β values when N 6= 0 . . . . . . . . . . . . . 59
RE
4.6 f rmin (y) for varying η values when N 6= 0, RG
= 4. . . . . . . . . 60
RG

4.7 System model for a CR setup employing power control. Legend:

black circle = nearest interferer node, black square = PR, red square
= CR receiver. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.8 The exact and asymptotic outage probability vs the primary system
power level Pp , for different values of σ and α under βl = 0.01,
R = 20, and M = 5. . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.9 Comparison of the outage probability vs primary system power
level Pp with different values of σ and α, for the multiple-ring
model and the annular underlay model under βl = 0.01, R = 20,
and M = 5. The values for β, RG and RE have been chosen ac-
cordingly. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.10 The error percentage of the outage of the multiple-ring model vs
the primary system power level Pp , for different values of M and
R under β = 0.001, α = 2, and σ = 0. The value of βl has been
chosen appropriately for each case. . . . . . . . . . . . . . . . . . 68
4.11 The outage probability vs the primary system power level Pp , un-
der different values of β, for the annular underlay model and the
multiple-ring model. α = 2, σ = 0, R = 5, and M = 20. . . . . . . 69
4.12 Outage probability vs Ps for power controlled interferer nodes un-
der RCR = 0 and RCR = 275 for β = 0.0001, α = 2, RE = 200,
RG = 25, R = 30, Prec = −30 dBm, γT h = 1, and σn2 = 0.001. . . 70
4.13 Outage probability vs Pp for different values of R, with RE = 500,
RG = 50, Ps = 20 dBm, β = 0.0001, α = 2, γT h = 1, and
σn2 = 0.001. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.14 Outage probability vs RG for different values of β, with RE = 200,
R = 40, Pp = 50 dBm, Ps = 30 dBm, α = 2, γT h = 1, and
σn2 = 0.0005. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.15 Outage probability vs Ps for different values of β and α, with RE =
200, RG = 20, R = 10, Pp = 60 dBm, γT h = 1, and σn2 = 0.001. . . 73
Abbreviations

Acronym Meaning
BPP Binomial Point Process
CAGR Cumulative Annual Growth Rate
CCDF Complementary Cumulative Distribution Function
CDF Cumulative Distribution Function
CLT Central Limit Theorem
CR Cognitive Radio
CVRMSE Coefficient of Variation of Root Mean Square Error
DSA Dynamic Spectrum Access
DySPAN Dynamic Spectrum Access Networks
FSL Free Space Loss
GPS Global Positioning System
IEEE Institute of Electrical and Electronic Engineers
ISI Inter Symbol Interference
ITU International Telecommunication Union
MGF Moment Generating Function
MIMO Multiple Input Multiple Output
NEM Normalized Error of the Mean
PDF Probability Density Function
PDP Power Delay Profile
PPP Poisson Point Process
PR Primary Receiver
PT Primary Transmitter
QoS Quality of Service
RMSE Root Mean Square Error
SINR Signal to Interference and Noise Ratio
WiMAX Worldwide interoperability for Microwave Access
List of Symbols & Mathematical
Notation

Notation Definition
k!
factorial of k
n
k
binomial coefficient, n choose k
min (a1 , . . . , an ) minimum of all scalars ai for relevant i
O(·) the
R ∞ remainder in series
x−1 −t
Γ(x, a) a
t e dt
Γ(x) Γ(x, 0)
2 F1 (, ; ; ) Gauss’ Hypergeometric function [2, (eq. 9.10)]
2 F2 (, ; ; ) generalized Hypergeometric function [2, (eq. 9.14)]
Kν (·) modified Bessel function of the second kind [2, (eq.
8.407)]
bxc the largest integer less than x
R ∞ e−xt
En (x) 1 tn
dt
fX (·) probability density function (PDF)
FX (·) cumulative distribution function (CDF)
MX (·) moment generating function (MGF)
EX [·] expectation with respect to X
Chapter 1

Introduction

1.1 Wireless Communications

Wireless communications is one of the most revolutionary engineering breakthroughs

of all time, and has grown exponentially in terms of technology and users over the
past decades. Today, wireless communications has enabled a variety of applications
and standards including [3]:

1. Mobile telephony 2. Broadcast technology

3. Wireless Local area networks 4. Fixed wireless access
5. Cordless telephony 6. Wireless sensor networks
7. Satellite communications 8. Radar and navigation
9. Body area networks 10. Personal area networks

Cisco states that the global wireless data traffic grew 133% in 2011, and that it
was higher than anticipated [4]. Furthermore, the International telecommunication
Union (ITU) states that by 2012, the global mobile penetration was at 6 billion,
which corresponds to 86% of the world’s population [5]. Therefore, the demand
for services and bandwidth is extremely high. But, the radio frequency spectrum
is a finite resource, and has to be utilized efficiently to allow all these services and
technologies to operate.
However, the spectral efficiency gains (data rate that can be sent over a unit

1
 Figure 1.1: Spectral efficiency over the years. Source : Qualcomm, from [1]

bandwidth) are slowing over the years [1]. This slowdown is clearly evident in
Fig. (1.1). Due to this, optimizing spectrum usage or developing ways to access
pre-allocated spectrum is essential for new technologies.

1.2 Methods to improve spectral efficiency

Several methods to improve spectral efficiency will be discussed next.

1.2.1 Cognitive radio

Many spectrum bands have already been allocated for different services and providers.
However, most of pre-allocated frequency spectrum has been found to remain idle
for much of the time [6]. This is a significant under utilization of available spec-

2
trum, and can be mitigated via cognitive radio (CR) [7]. A CR node is a transmitting
entity which adapts its system parameters (e.g. frequency, power, multiple access
method, modulation) to suit its environment [3]. However, the most well known at-
tribute of CR is the ability to access spectrum pre-allocated to a licensed user. This
ability improves spectral efficiency.
In CR terminology, the licensed users are also called primary users. They have
reserved the particular frequency block from the governing authorities after paying
a fee, and are thus licensed to transmit. The users who opportunistically access
primary users’ spectrum are termed the secondary users or cognitive radios. While
they are permitted to opportunistically access the spectrum, their interference on the
primary network should be either eliminated or managed below a prescribed level.
Cognitive radios can be broadly classified based on the access technology into
interweave CR networks, spectrum overlay CR networks, and spectrum underlay
CR networks [8–12]. However, other literature [10] classify CR networks into just
underlay and overlay networks.

• Interweave networks
In interweave CR networks, the secondary users sense the spectrum, and
transmit only when primary user transmissions are absent. The spectrum ac-
cess is thus opportunistic, and requires advanced spectrum sensing schemes
in order to begin and end transmissions immediately as necessitated [11]. In-
stead of spectrum sensing, CR nodes may also use an out-of-band beacon
transmission [13] from the primary network to detect the presence or absence
of primary transmissions.

• Underlay networks
In spectrum underlay CR networks, the secondary users are allowed to trans-
mit even while primary users are transmitting [8, 10, 11]. To minimize or
eliminate interference to the primary network, several schemes have been
proposed. The first method is to have a centralized database with dynamic
information on the transmitters/receivers of the target frequency in a given
area. Therefore, the secondary users can transmit below a certain power level

3
 if they are beyond a given distance. Due to this distance, the secondary user
signals attenuate sufficiently so that interference on the primary receiver is
negligible. The second method is to use a beacon transmitter from either the
primary transmitter or receiver, where the secondary users transmit only when
the beacon signal power is less than a certain amount (so that the secondary
user is beyond a pre-determined distance). Alternatively, Global Position-
ing System (GPS) based schemes and spread spectrum schemes have been
proposed for underlay networks [8, 10]. Interference generated by underlay
networks is the main focus of this thesis.

• Overlay networks
Concurrent primary and secondary transmissions are also allowed in overlay
CR networks. However, the differing assumption from underlay networks is
that the CR nodes have prior information on primary transmissions such as
codebook information and channel gains [10, 11]. Moreover, the CR nodes
can actively assist the primary transmissions by using a portion of their power
to relay primary user messages.

Standards

Since CR is a new technology, no previous standard existed for opportunistic spec-

trum access. While elements of CR are already included in standards such as IEEE
802.11 (WiFi),IEEE 802.15.4 (Zigbee), and IEEE 802.16 (WiMAX) [14], IEEE
802.22, the Wireless Regional Area Network (WRAN) standard [15] is the first to
fully incorporate CR techniques. Furthermore, the IEEE Dynamic spectrum access
networks (DySPAN) committee has developed the IEEE P1900.X standards.
IEEE 802.22 has been developed to access white spaces in television frequency
bands in order to provide wireless broadband access to rural areas. With this stan-
dard, the CR users must be aware about the availability of spectrum at a given
instance. The two methods used for spectrum awareness are geo location/database
and spectrum sensing [15]. In addition, they should be able to dynamically adapt
transmissions to not interfere with licensed TV transmissions [15].
The IEEE P1900.x standards develop better dynamic spectrum access (DSA),

4
which includes new techniques for managing interference, sensing, network man-
agement and coordination of wireless networks [14]. For example, the IEEE P1900.6
standard is regarding spectrum sensing techniques for CR [16]. Furthermore, poli-
cies regarding DSA are being developed for current technologies such as 3G/4G,
and WiFi [14].

1.2.2 Heterogeneous networks

Heterogeneous networks, another approach to improve spectral efficiency, is a broad

concept where different cells co-exist dynamically while using similar frequen-
cies [17]. The cells differ in terms of sizes and power levels drastically. For ex-
ample, large macro cells (covering a radius of many kilometres) are needed for
basic coverage needs and moving subscribers. In contrast, picocells (put in place
by the operators to cover a radius of a few hundred metres or less) may be deployed
in dense urban areas to increase capacity. Femtocells (which have a small cover-
age area, and are restricted to a certain building or customer premises) which are
even smaller may be used in individual buildings where an extremely high capac-
ity is needed [18]. When services from different cells are available, the user may
choose the best cell or use multiple cells to increase their data rate and lower the
outage. The performance of heterogeneous networks critically depend on mutual
interference levels of the cells.

1.2.3 Small cell networks

The concept of small cells deals with the dense deployment of cells within a given
geographical area to increase the capacity, spectral efficiency, and power usage [19].
They are especially useful in urban environments where the demand for wireless
data traffic is high. Apart from the obvious spectrum efficiency gains, small cells
can potentially provide significant energy savings [19]. However, administration,
organization, and maintenance of small cell networks are a challenge. Furthermore,

5
interference mitigation among different cells and guaranteeing a required Quality
of Service (QoS) are demanding [19].

1.3 Problem Statement

While all these techniques improve the overall spectrum efficiency, this thesis will
concentrate on underlay CR networks. The advantages of underlay networks are
their ability to access the frequency spectrum simultaneously with primary users,
and not requiring the secondary users to have a prior knowledge of primary trans-
missions [11]. Despite these advantages, interference on the primary users due to
simultaneous primary and secondary transmissions must be minimized. Therefore,
the aggregate interference generated by underlay networks must be characterized
for differing operational conditions.
A comprehensive analysis of interference for different channel models, CR node
densities, and their spatial distributions is therefore necessary. Furthermore, due
to the complexity of existing interference models, simpler, and accurate models
are desirable. Therefore, characterizing the aggregate interference and developing
approximate models are the main focuses of this thesis.

1.4 Contributions and Outline

The main contributions of the thesis are regarding the analysis, approximation, and
modeling of aggregate interference (I). They are broadly listed below:

• Analysis of I for an underlay network consisting of a Poisson process of

interfering nodes in a finite annular area around the primary network.

• Development of a Gamma model for I under a composite fading channel via

a moment matching method.

• Development of a multiple-ring model for I, and investigation of its accuracy.

6
 • Characterization of the interference caused by the nearest interfering node,
and investigation of when the nearest interferer can approximate I.

The outline of the thesis is as follows:

Chapter 2

In Chapter 2, basic background concepts and models for small scale fading, shad-
owing, path loss, spatial distribution, and others will be presented.

Chapter 3

The first part of Chapter 3 comprises a comprehensive interference analysis for in-
terfering nodes spatially distributed as a Poisson point process in an annular region.
The moment generating function (MGF) of the aggregate interference is derived for
generic path loss exponent values and shadowing levels. The cumulative distribu-
tion function (CDF) of the signal to interference and noise ratio (SINR) is obtained,
and an asymptotic analysis is performed. Subsequently, the aggregate interference
is modeled by a Gamma distribution using a moment matching method. The accu-
racy of this approximation is confirmed, and further calculations are facilitated by
this simple probability density function (PDF) to represent the aggregate interfer-
ence.

Chapter 4

Chapter 4 proposes a new model for interference, and also analyzes the effect of
the nearest interfering node. In the first part of the chapter, a new system model to
approximate the conventional annular underlay model is proposed. The new model,
named the multiple-ring model, consists of interferers constrained on multiple rings
around the primary receiver (PR) instead of being spread spatially. This model is
shown to be accurate, simple, and versatile. It can even be used as a stand-alone
system model for other types/shapes of node distributions. In order to evaluate
the accuracy, an MGF based performance analysis and an asymptotic analysis is
performed for Rayleigh fading, and composite fading and shadowing.

7
 In the second part, the nearest node approximation for the aggregate interfer-
ence is investigated. The PDF of the distance of the nearest interfering node to the
PR, and the MGF of the interference from the nearest node are derived for Rayleigh
fading. The nearest node dominates the aggregate interference under certain condi-
tions, which include lower node densities and higher path loss exponent values.

8
Chapter 2

Background

This chapter provides some brief mathematical background on key concepts used
in the thesis. They include wireless channel characterization, spatial distribution
models, and interference characterization.

2.1 The wireless channel

Characterizing the wireless channel is extremely important. In this section, channel

impairments such as multipath propagation, the Doppler effect, small scale fading,
shadowing, and path loss will be discussed briefly.

2.1.1 Multipath fading and the Doppler effect

Due to multiple obstructions and scatterers in the wireless channel, the received
signal is the superposition of many signals with different time delays and phases
[20]. These multiple copies will cause Inter symbol Interference (ISI), and will
severely degrade the performance of the receiver.
The received signal may be represented as [20]
 
XN (t) 
j(2πfc (t−τn (t))+φDn )
r(t) = R αn (t)u(t − τn (t))e , (2.1)
 
n=0

where u(t) is the complex envelope of the transmitted signal, αn is the channel gain

9
for the n-th multipath component, N(t) is the number of resolvable multipath com-
ponents, τn (t) is the time delay for the n-th component, fc is the carrier frequency,
and φDn is the Doppler phase shift.
The power delay profile (PDP) represents the average power associated with a
given multipath delay (τ ) [20]. The average delay and the root mean square (r.m.s.)
delay are important statistics of a wireless channel. They are defined as
R∞ sR
∞
0
τ P τ dτ 0
(τ − µτ )2 Pτ dτ
µτ = R ∞ , and στ = R∞ , (2.2)
0
Pτ dτ 0
Pτ dτ

respectively. If the time period of a transmitted signal is defined to be Ts , frequency

flat fading occurs if Ts >> στ . Otherwise, the signal would experience frequency
selective fading which is not desirable. The term coherence bandwidth is usually
defined as
1
Bcoh ≈ , (2.3)
στ
which is roughly the frequency range in which a signal experiences frequency flat
fading.
The time variation of the channel is described by the Doppler effect, which is
caused by the relative frequency shift between the received signal and the transmit-
ted signal. If the transmitter and receiver are stationary, the Doppler shift is zero.
However, when the transmitter and/or receiver move/moves, the maximum Doppler
shift is given by
fc v
fd = , (2.4)
c
where fc is the signal frequency, v is the relative velocity between the transceivers,
and c is the speed of light. The Doppler spectrum of the channel represents the
power associated with a particular Doppler shift (between 0 and fd ). In a similar
manner to the PDP, the average and the r.m.s. Doppler spread can be calculated.
1
Furthermore, the coherence time Tcoh is defined to be approximately Bd
, where Bd
is the Doppler spread. If the signal period Ts << Tcoh , the signal is said to undergo
slow fading. Otherwise, the signal undergoes fast fading.

10
2.1.2 Small scale fading models

Small scale fading is the random fluctuation of signal amplitude over short dis-
tances, and occurs due to the effects of multipath propagation. Small scale fading
can be characterized by various mathematical models. The best model for a given
channel depends on its characteristics.

Rayleigh fading

Rayleigh fading is the most common model to represent wireless channels includ-
ing broadcast and mobile systems. Popular due to its mathematical tractability, this
model is valid when there is no line-of-sight path between the transmitter and re-
ceiver [20]. The probability density function (PDF) of the received signal power
under this model is given by

1 − xγ̄
fγ (x) = e , 0 ≤ x < ∞, (2.5)
γ̄

where γ̄ is the average received signal power. It should be noted that while the
Rayleigh distribution denotes the envelope amplitude, the power is specified by an
exponential distribution.

Rician fading

Rician fading occurs when there is a dominant line-of-sight component. This model
is especially useful for channels such as satellite links. The PDF of the received
signal is [21]
s !
(K + 1) −(K+ x(K+1) )I xK(K + 1)
fγ (x) = e γ̄
0 2 , 0 ≤ x < ∞, (2.6)
γ̄ γ̄

where γ̄ is the average received signal power, K is the ratio between the line-of-
sight component power and the power of the other scatterer components, and I0 (·)
is the modified Bessel function of the first kind.

11
Nakagami-m fading

Nakagami-m fading is a model proposed by [22] which fits the empirical measure-
ments of wireless channels. Its PDF is given by
 m
xm−1 m mx
fγ (x) = e− γ̄ , 0 ≤ x < ∞, m > 0.5, (2.7)
Γ(m) γ̄

where m is a parameter describing the severity of fading. The model is versatile;

for example, m = 1 yields Rayleigh fading, and m → ∞ yields the no-fading case.

2.1.3 Shadowing

Shadowing is the random variation of signal amplitude due to blockages from large
obstacles such as mountains and buildings in the transmission path. The distances
in which shadowing occurs depend on the dimensions of the obstacle causing the
shadowing effect [20].

Log-normal shadowing

The most common model for shadowing is the log-normal shadowing model. The
PDF of the ratio between transmit to receive power ψ is given by [20]
(10log 10 (ψ)−µψ )2
dB
ξ −
σ2
fΨ (ψ) = √ e ψdB
, 0 ≤ ψ < ∞, (2.8)
2πσψdB ψ
10
where ξ = ln10
, µψdB is the mean of ψdB , and σψdB is the standard deviation of ψdB .
Because (2.8) is not mathematically tractable readily, several approximations
have been proposed. One such approximation is the Gamma model [23, 24], where
the two distributions show a close match except in the lower tail region. Another
approximation is the mixture Gamma model developed in [25].

Composite shadowing and fading models

As shadowing and fading occurs simultaneously, it is convenient to have the com-

bined channel effect in a single PDF, rather than work with separate distributions.
Thus, several composite models incorporating shadowing and fading have been

12
proposed [21]. Rayleigh-lognormal, and Nakagami-lognormal models have been
proposed [26].
However, the log-normal model (2.8) does not lend itself easily for mathemat-
ical analysis. Therefore, by approximating (2.8) by a Gamma model, and incorpo-
rating Rayleigh fading, [27] has derived the Generalized-K distribution.
According to this distribution, the composite parameter characterizing Rayleigh
fading and shadowing has the form X = Xs Xf , where Xs and Xf respectively
denote shadowing and Rayleigh fading. Xf is modeled as a unit exponential random
variable, and thus the PDF of X can be written as [23]
1 − Xx
fX/Xs (x) = e s, 0 ≤ x < ∞. (2.9)
Xs
The shadowing component Xs can be modeled as a Gamma random variable, with
the PDF  λ
1 λ λ
fXs (y) = y λ−1e− Ωs y , 0 ≤ y < ∞, (2.10)
Γ(λ) Ωs
where Γ(x) is the Gamma function. It has been shown in [28] that λ = eσ21−1 and
q
λ+1
Ωs = λ
, where σ 2 is the variance of corresponding log-normal shadowing.
When expressed in the decibel scale, σdB = 8.686 σ.
The composite PDF of Rayleigh fading and Gamma shadowing is obtained by
averaging (2.9) over (2.10) as [28]
 1+λ
2 b 1+λ √

fX (x) = x 2 Kλ−1 b x , 0 ≤ x < ∞, (2.11)
Γ(λ) 2
q
which is the Generalized-K distribution, where b = 2 Ωλs . This composite model
will be subsequently used to characterize channels where both fading and shadow-
ing are present.

Gamma approximation to the Generalized-K distribution

However, because the Generalized-K distribution (2.11) may still be cumbersome

for analysis, it has been further approximated by a Gamma PDF [29]. The scale and
shape parameters of the Gamma distribution have been chosen by moment match-
ing. Therefore, (2.11) can be approximated as
1 x
fX (x) = xk−1 e− θ , 0 ≤ x < ∞. (2.12)
θk Γ(k)

13
 2 1


The scale and shape parameters θ and k are λ
+ 1 Ωs and 2
+1
, respectively.
λ

2.1.4 Path loss

Path loss is the reduction in signal amplitude over distance between the transmitter
and receiver. Path loss variations only occur over large distances [20].

Free space path loss model

The free space path loss (FSL) model is the most simple path loss model. It can be
written as  2
4πd
F SL = , (2.13)
λ
where d is the distance between the transmitter and receiver, and λ is the wavelength
of the transmitted signal.

Empirical path loss models

Because free space conditions do not hold for the wireless environment which en-
compasses many variable factors such as buildings, trees, hills, and houses, the path
loss modeling is difficult. Thus, several empirical models have been developed us-
ing real world experimental data. These include the Okumura model, Hata model,
COST 231 Hata model, and COST 231 Wolfisch-Ikegami model [20].
The COST 231 Hata model for path loss (PL) can be written as [20]

P LdB (d) = 46.3 + 33.9log10 (fc ) − 13.82log10 (ht ) − a(hr )

+ (44.9 − 6.55log10 (ht ))log10 (d) + CM , (2.14)

where ht is the transmit antenna height, hr is the receiver antenna height, fc is the
transmit signal frequency, and d is the distance between the transmitter and receiver.
For suburbs and small cities, a(hr ) is defined as

a(hr ) = (1.1log10 (fc ) − 0.7)hr − (1.56log10 (fc ) − 0.8). (2.15)

CM is 0 for small cities and suburbs, while it is 3 for large cities.

14
Simplified path loss model

The most common path loss model used in analysis is the simplified path loss
model, and will be used in the subsequent chapters. According to this model, the
power at a certain distance r from the transmitter is given by
 r α
0
P (r) = P0 , (2.16)
r

where P0 is the observed power at a distance r0 from the transmitter, and α is the
path loss exponent. This model includes the free space path loss model (2.13) as a
special case when α = 2.

2.1.5 Power control

Power control is the variation of transmit power according to the situation such as
distance from the receiver, other users’ activities, and channel conditions. Power
controlling schemes have a dual benefit of saving transmitter power as well as re-
ducing unwanted interference. They can be based on numerous factors such as
QoS feedback from the receiver, distance to the receiver, and locations of other
co-channel nodes. Several schemes have been proposed for CR networks [30, 31].
However, this thesis mostly ignores power control schemes, except when a simple
power control scheme based on distance to the receiver is considered in Chapter 4.

2.2 Spatial distribution models

Spatial distribution refers to the random locations of interferer nodes in a given

area. For interference analysis, modeling the spatial distribution of the interfering
nodes (Fig. 2.1) is essential. Several spatial distributions have been used in liter-
ature to model the random locations of nodes. The two most popular ones are the
Poisson point process [32], and the binomial point process [33]. A point process is
a random pattern of points in d-dimensional space (where usually d = 2 or d = 3

15
in applications) [34]. For modeling the distribution of interferers, a 2 dimensional
space is commonly used.

2.2.1 Poisson point process

The Poisson point process (PPP) is the most common spatial distribution used to
model the distribution of interfering nodes in a given area [35–40]. The average
number of spatial points in a given area is the intensity. A PPP is termed homo-
geneous if the intensity parameter is constant. For non-homogeneous PPPs, the
intensity parameter is a function of the location. For a 2-dimensional homogeneous
spatial PPP, the probability of having n nodes in a region B is given by [33]

(λv(B))n −λv(B)
P (N(B) = n) = e , n = 0, 1, 2, . . . (2.17)
n!

where λ is the intensity parameter, and v(B) is the area of B. The PPP model used
in the thesis will be explained in detail within the following section.

Thinning

Thinning refers to the process of removing certain points from the total set of
points [34]. Thinning comes in extremely useful while modeling interference where
the whole set of nodes may not be active (not transmitting). In such a scenario, the
thinned set of points will model the active nodes. The thinning process occurs by
marking the set of points by an indicator random variable IX (which can take the
values 0 or 1), and deleting the points marked with 0. Thinning has two forms: de-
pendent thinning and independent thinning [34]. When IX are independent random
variables, it is independent thinning. In dependent thinning, any point in the PPP
which has a neighboring node closer than a certain distance will be deleted.

Clustering

Clustering is the process of taking a set of spatial points A, and replacing the point
A by a random set of points ZA for each point A ∈ A [34]. Fig. 2.2 shows an ex-
ample for the clustered PPP. Clustering is important when modeling the interfering

16
 Figure 2.1: Random spatial distribution of interferer nodes.

nodes distributed around secondary base stations. In such a scenario, the secondary
base stations will be the original set of points, and the nodes will be the clustered
points after the transformation. References [35, 41–43] consider clustering in their
analysis.

2.2.2 Binomial point process

The Binomial point process (BPP) is useful to model interferers when the total
number of nodes is fixed. Reference [44] argues that the BPP is a better model
to represent the spatial distribution of nodes because the total number of nodes is
often fixed. For a given area of Z, the total number of nodes in B (B ∈ Z) is given
by [45],
 
N
P (N(B) = n) = pn (1 − p)N −n , n = 0, . . . , N (2.18)
n

17
Figure 2.2: Clustered PPP. The right image shows the original set of points, and the
image on the left shows the clustered points after the transformation.

v(B)
where N is the total number of nodes in Z, and p = v(G)
.

2.3 Annular underlay networks

The importance of CR networks in improving spectral efficiency was discussed

in Chapter 1. However, although underlay CR nodes are intelligent, residual in-
terference to the primary system is unavoidable. A comprehensive understanding
of the aggregate interference is critical when designing interference management
schemes. Interference analysis in cognitive radio networks has thus been a hot re-
search topic in recent times [30,46–53]. One method which has been used to reduce
the interference in an underlay network is the use of an exclusion region (contention
control) [30, 54, 55] . In such a setup, nodes in an exclusion zone around the pri-
mary transmitter or receiver are not permitted to transmit, and nodes outside this

18
 RE

RG

Figure 2.3: Annular underlay network. Legend: black dot = interferer nodes, black
square = PR, yellow square = PT.

zone are permitted to transmit below a certain power level. As discussed in Chap-
ter 1, this process may be enabled by a beacon signal from the primary transmitter
or receiver, prior location knowledge, or Global Positioning Systems. In a beacon
setup, if the beacon is from the primary transmitter or from the primary receiver,
the exclusion region would be around the transmitter or receiver respectively. We
will be considering an exclusion region around the PR.
Although interferers can be distributed over arbitrary shaped regions, analysis
of such may be intractable. Furthermore, the interferers may be located in a polygon
shaped area (e.g. square or hexagonal). An annular region (Fig. 2.3) [49, 50, 56,
57] is a good approximation for arbitrary and polygon shaped distributions, and is
mathematically conducive for further analysis.
The interferer nodes form an underlay CR network which is intelligent enough
to adhere to the exclusion region around the PR. This region has an inner radius
(guard distance) of RG and an outer radius of RE . The guard distance ensures a
minimum performance on the PR. The interference from the nodes beyond RE is
assumed to be negligible due to path loss. An infinite region (RE → ∞) is a special
case of the model.
The primary transmitter (PT) to PR distance is denoted as R, and it doesn’t

19
depend on the guard distance. The number of interferer nodes N is random. The
nodes are distributed according to a homogeneous Poisson point process which has
2
the distribution of (2.17), where β is the interferer density, and AI = π(RE −
2
RG ) is the total area encompassing the interferer nodes (v(B)). Non-homogeneous
situations will be application or geography specific, and thus not very useful for a
general analysis.
Not all interferer nodes may be active at a given time, and some may be inhibited
from transmitting due to interference between the CR nodes themselves. However,
an ad-hoc network of interferer nodes where all the nodes concurrently engage in
transmission is assumed. Our results thus give a worst-case upper bound for the
aggregate interference and outage when interferer nodes use a medium access con-
trol protocol. Such cases however can be handled by introducing an activity factor.
Conversely, without the loss of generality, we can consider β as the density of all
the active nodes.
For the purposes of this thesis, we will only consider one PR, and that other PRs
are located a sufficient distance away. If the next closest PR is less than 2RG away
from the PR under consideration, the exclusion region will be of a complex shape.
Furthermore, if it is less than a distance of 2RE away, the interferer area will not be
annular. However, these considerations are out of the scope for this thesis.

2.3.1 Aggregate interference

The aggregate interference experienced at the primary receiver is the sum of in-
terference from all the active interferer nodes transmitting in the same frequency
spectrum block. Therefore, the aggregate interference I can be written as
N
X
I= Ii , (2.19)
i=1

where Ii is the interference caused by the i-th interferer, and N is the number of
interferers.
As the PDF of the aggregate interference is intractable, this thesis develops an
MGF based approach. The MGF can be obtained relatively easily because, for a

20
sum of independent interferers, the total MGF would be the multiplication of each
interferer’s individual MGF. The MGF MIi (s) of the interference from a single node
can be written as

MIi (s) = E[e−sIi ], (2.20)

where E[·] denotes the expectation, and s is the Laplace variable. The aggregate in-
terference is dependent upon various factors. They include the channel parameters
such as path loss, shadowing, and small scale fading. Furthermore, power control-
ling schemes and transmit powers of the nodes, the spatial distribution of nodes,
and the design of exclusion regions also play a vital role. The sensing procedures
of the beacon transmission/ frequency spectrum and missed detection of these by
the nodes also need to be considered.
Modeling of aggregate interference to fit well known distributions has been ex-
tremely popular due to the intractability of exact analysis. These include approxi-
mation by Gaussian distributions, log-normal distributions, tailed α-stable distribu-
tions, and as a sum of normal and log-normal distributions [58–61].
In the following chapters, several different models for the aggregate interference
will be developed.

2.4 Conclusion

This section provided background material for the thesis. Fading and shadowing
models, spatial distributions, and the annular underlay network model were de-
scribed.

21
Chapter 3

Aggregate Interference Analysis

Chapter 2 provided introductions and justifications to the system model, and ex-
plained the need to characterize aggregate interference. This chapter develops an
exact analysis and a Gamma approximation of the aggregate interference for the
annular underlay network model introduced in Section 2.3.

3.1 Introduction

In Chapter 2, the annular underlay model for investigating the aggregate interfer-
ence was presented. The random aggregate interference depends on several factors
such as channel parameters, spatial distribution of the interferer nodes, activity fac-
tors, and power control. These parameters were briefly described in Chapter 2.
As the modeling of aggregate interference at the primary receiver (PR) is criti-
cal to characterize performance degradation, it has received much attention recently.
Reference [56] analyzes the average aggregate interference when transmission con-
straints among CR nodes are considered, while [57] analyzes the capacity-outage
of a CR network due to aggregate interference. Reference [58] shows that under
certain conditions, the aggregate interference is not lognormal. In reference [59],
the authors suggest that the aggregate interference can be modeled as the sum of a
normal random variable and a lognormal random variable. Reference [52] consid-
ers different activity models for the CR nodes and obtains cumulants of the aggre-
gate interference by using Campbell’s theorem. Reference [49] derives the moment

22
generating function (MGF) of the combined interference without shadowing for a
number of different path loss exponent values.
The aggregate interference will be investigated further in the subsequent sec-
tions. First, the moment generating function of the aggregate interference will be
derived for the annular underlay model (Fig. 2.3). The exact and asymptotic outage
performance will be analyzed. Second, the aggregate interference will be modeled
with a Gamma model.

3.2 Aggregate interference of annular underlay net-

works

3.2.1 Introduction

This section provides a comprehensive analysis of the aggregate interference for

the annular underlay model (Fig. 2.3). The analysis considers all relevant channel
impairments. The interfering signals are assumed to undergo composite fading and
shadowing, and path loss (arbitrary exponents). The exact closed-form MGFs of
the aggregate interference for both the generalized-K distribution, and the Gamma
approximation to it are derived. Closed-form expressions for the outage and an
asymptotic performance analysis are provided.

3.2.2 System model

The system model of Fig. (2.3) is used for the analysis, with the same parameters.
From the simplified path loss model, the received power at distance r from the
transmitter is given by (2.16). For brevity, we define the quantity P0 r0α to be the
power level of the transmitter (which depends on the transmit power, gains, and
frequency). The total interference power received at the PR is given by (2.19). The

23
interference power Ii is given by

Ii = Ps ri −α Xi , (3.1)

where Ps is the power level of an interferer, and ri is the distance between the i-th
interferer and PR. Ps is a constant, and no power control occurs. Xi characterizes
the combined effects of small-scale fading and shadowing. The PDF of Xi follows
the Generalized-K distribution of (2.11).

3.2.3 Interference Statistics

In this subsection, the exact MGF of the aggregate interference and an approximate
MGF to the aggregate interference are derived.

MGF of the aggregate interference

Because ri and Xi are independent, the MGF MIi (s) of the i-th interferer (i =
1 . . . N) is given by (2.20),

MIi (s) = EXi ,ri [e−sIi ] = EXi [E[e−sIi ]]. (3.2)

For the homogeneous PPP considered, the CDF of ri when RG < ri < RE can be
written as

FRi (ri ) = P (Ri < ri )

ri2 − RG
2
= 2 2
. (3.3)
RE − RG
Therefore, the PDF of ri can be obtained by differentiating the CDF as
 πri
2 AI , RG < ri < RE
fR (ri ) = . (3.4)
0 , otherwise
By averaging e−sIi using the the PDF of interferer distance (3.4), we get
Z RE  
i −sPs r −α Xi πr
MI/Xi (s) = e 2 dr
RG AI
2π RE −(sPs Xi )r−α
Z
= e r dr
AI RG
    
2π 2 Ps sXi 2 Ps sXi
= R E 2+α −RG E 2+α , (3.5)
αAI E α RE α α RG α

24
where En (x) is the generalized exponential integral. Averaging further respect to
the composite fading model (2.11), the MGF of the interference from the i-th inter-
ferer is derived as
2 4
π (Ps s) α −2 b− α −2
MIi (s)= 2 (Q(RE )−Q(RG )) , (3.6)
(−1)λ+ α 4α2 (1 + α)Γ(λ)AI
where Q(R) is defined as
4 2
b4+ α (−1) α +λ α2 Γ(λ − 1)R2α+2 Rα b2
 
2 2
Q(R) = 2 2 F2 2, 2 + ; 3 + , 2 − λ;
(Ps s) α α α 4Ps s
4
25+ α (Ps s)2 (1 + α)π
+
sin(πλ)

2
× (2 + α)Γ(1 + λ + )
α
2 Rα b2 2 Rα b2
 
+ α λΓ(1 + λ + , − ) − Γ(2 + λ + , − ) . (3.7)
α 4Ps s α 4Ps s
The special functions 2 F2 (, ; ; ) and Γ(x, a) are the generalized Hypergeometric
function and the upper incomplete Gamma function respectively [2]. Due to the
complexity of (3.6), an accurate approximation is desirable. To this end, the Gamma
approximation to the generalized-K distribution (2.12) can be used.
i
Using this approximation, we get MI,approx (s) as
  αk+2  α  αk+2  α

i 1 2π RE RE RG RG
MI,approx (s) = I − I . (3.8)
2 + kα AI θk (Ps s)k Ps sθ (Ps s)k Ps sθ
where I(x) is given by

I(x) = 2 F1 (k, k + 2/α; 1 + k + 2/α; −x) , (3.9)

and 2 F1 (, ; ; ) is the Gauss’ Hypergeometric function [2].

Because each interferer is assumed to be independent, the MGF of I given N
can be written as
n
Y
n
MI/n (s) = MIi (s) = MIi (s) . (3.10)
i=1

By averaging MI/N (s) over the probability distribution (2.17), we find

∞
X (βAI )n −βAI
MI (s) = MI/n (s) e . (3.11)
n=0
n!
Substituting (3.10) into (3.11), we get
i
MI (s) = eβAI (MI (s)−1) . (3.12)

25
3.2.4 Performance Analysis

This section derives the outage probability, the asymptotic outage and the diver-
sity/coding gains.

Receiver SINR characteristics

Here, we derive the CDF and the PDF . The SINR γ at the PR can be written as

Pp R−α Y
γ= , (3.13)
I + σn2

where Pp is the power level of the PT, σn2 is the noise variance, and Y is the channel
gain between the primary transmitter and receiver. We only consider the case where
the primary signals undergo path loss and Rayleigh fading due to the mathematical
complexity of analyzing for other cases. Then, Y is a unit exponential PDF with
the distribution of (2.5). The variables Y and I are independent, and the CDF of γ
is

Pp R−α Y
 
Fγ/I (x) = P ≤x
I + σn2
x(I + σn2 )
 
= P Y ≤
Pp R−α
 
2)
x(I+σn
−
Pp R−α
= 1−e . (3.14)

Averaging with respect to I, we get


2
 "  #
xσn x
− −I
Pp R−α Pp R−α
Fγ (x) = 1 − e EI e
 
xσn2  
−
Pp R−α
x
= 1−e MI . (3.15)
Pp R−α

Substituting γT h instead of x gives the outage equation, where γT h is the threshold

SINR level for the PR.

26
 The PDF fγ (x) can be obtained by differentiation of (3.15) as

BV H k + α2

   
J Hk J
fγ (x) = A− −I + I
x (Ux + J)k Ux (Ux)k Ux
!!
F k + α2

   
G Gk G
− +I − I
(Ux + G)k Ux (Ux)k Ux
 

F G H J


−B−Ax+BV
(U x)k I Ux
− (U x)k I Ux
× e , (3.16)

Rαk+2
2
 
σn 1 1 2
where A = Pp R−α , B = βAI , U = Pp R−α , V = 2+kα AI θk , F = EP k , G =
s
−Rα Rαk+2 −Rα
E
Ps θ
, H= G
Psk
and J = G
Ps θ
.

Asymptotic Outage

Since the outage probability (3.15) is complicated, a mathematically tractable asymp-

totic expression is useful. An asymptotic outage probability expression is derived
Pp
for, when T
is significantly larger than Ps and σn2 . By expanding I(Cx), we can
obtain  
2 + αk k(2 + αk) 1
I(Cx) = + k+1 k+1 + O k+2 , (3.17)
2xk C k x C (α − 2) x
where C is a constant. Thus, by using (3.17), for the expression of the MGF (3.8) in
Pp
(3.15), and with some algebraic manipulations, the CDF for high γT h
with respect
to the noise and interference can be obtained as
 
xσn2
− 2βπθkRα Ps
FγAsy (x) = 1−e Pp R−α
e α−2 Pp (R2−α
E −R2−α
G )x . (3.18)

For small x, ex can be written as 1 + x. Therefore, (3.18) can be approximated by

xσn2
 
FγAsy (x) = 1 − 1−
Pp R−α
2βπθkRα Ps
 
2−α 2−α


× 1+ RE − RG x . (3.19)
α − 2 Pp
2 α
Defining A = Rσ−α and B = 2βπθkR 2−α 2−α


n
α−2
P s R E − RG , we get
  
x x
FγAsy (x) = 1 − 1 − A 1+B
Pp Pp
x
≈ (A − B) , (3.20)
Pp

27
which is the asymptotic CDF.

Diversity Gain and Coding Gain

The diversity gain and coding gain fully characterize the asymptotic performance
of a system. The outage probability at high SINR can be written as

Pout (γ) ≈ (Gc γ)−Gd , (3.21)

where Gc is the coding gain and Gd is the diversity gain. Increasing the SINR
is analogous to increasing Pp while having a constant σ and Ps . Therefore, Pout
becomes  −1
1 Pp
Pout ≈ . (3.22)
(A − B) x
1
From (3.22), we observe that Gd = 1 and Gc = (A−B)T
.

3.3 Gamma model approximation

3.3.1 Introduction

In the previous section, the exact and approximate MGFs were obtained for the
annular underlay model. However, the complexity of these results is high, and even
the approximate MGF (3.8) is complicated. Although the CDF (3.15) and PDF
(4.6) of the SINR when the primary signals undergo Rayleigh fading were derived,
obtaining the CDF and PDF of the aggregate interference (I) itself is extremely
difficult. A simple approximation to the aggregate interference PDF will help in the
analysis of systems such as multiple antenna (MIMO) systems and relay networks.
Therefore, an accurate but simple approximation for the aggregate interference is
useful. A proper approximation should be valid under varying inner and outer radii,
path loss exponent values, shadowing variances, and node densities. In the rest of
this section, the Gamma approximation for I will be derived. The accuracy of the
approximation will be described later in the numerical results section.

28
3.3.2 The Gamma Approximation

The Gamma approximation would use the approximate MGF (3.8) for analytical
purposes. Due to the use of Gamma shadowing, the aggregate interference may
also follow a Gamma model, rather than a Gaussian model which can be obtained
via the use of the Central Limit Theorem. The aggregate interference (2.19) is thus
assumed to be modeled by a Gamma random variable. Therefore, the PDF of the
aggregate interference I, is given by
1 x
fI (x) = ka
xka −1 e− θa , (3.23)
θa Γ(ka )
where ka is the shape parameter and θa is the scale parameter. Suitable values of ka
and θa to approximate the distribution of I are needed.
These parameters can be obtained by employing a moment matching method
[29]. For the Gamma approximation, the first and second order matching of mo-
ments are sufficient to find the shape and scale parameters. Higher order moment
matching is not needed.
The first- and second-order statistics of the aggregate interference are important
performance means by themselves. E[I], the expected value of the interference is
βAI E[Ii ], where the interference from a single interferer, Ii is given by

Ii = Ps ri −α Xi . (3.24)

Because ri and Xi are independent, E[Ii ] can be written as

E[Ii ] = Ps E[ri ]E[Xi ]. (3.25)

After performing the expectations, it can be shown that

√ 2  R2−α − R2−α 
E G
E[I] = 2πPs β eσ . (3.26)
2−α
But, there will be a singularity under α = 2. Therefore, by applying the L’Hospital’s
rule, we can obtain E[I] when α = 2 as follows;
" #
√ 2 d (RE 2−α
− RG
2−α
)
E[I]α=2 = 2πPs β eσ dα d
dα
(2 − α)
α=2
√ 2
= 2πPs β eσ (log (RE ) − log (RG )) . (3.27)

29
The variance is defined as,

V ar[I] = E[I 2 ] − E[I]2 . (3.28)

After obtaining an expression for E[I 2 ], V ar[I] is found out to be

 2−2α 2−2α 
2 2 RE − RG
V ar[I] = πβPs kθ (1 + k) . (3.29)
1−α

The expected value and variance can be matched with those of a Gamma dis-
tribution. The expected value and variance of the Gamma distribution are ka θa and
ka θa2 respectively. Therefore, by matching these moments, the shape parameter and
(E[I])2
the scale parameter of the Gamma approximation can be found out to be V ar[I]
, and
V ar[I]
E[I]
respectively. By substituting E[I] and V ar[I], we can find that

2−α 2−α 2
2πβ (1 − α) (RE − RG )
ka = σ 2 2 2−2α 2−2α , (3.30)
e (2 − α) (RE − RG )

and
 2  23 (2 − α) (R2−2α − R2−2α )
θa = Ps eσ E
2−α
G
2−α , (3.31)
(1 − α) (RE − RG )
respectively. The accuracy of the approximation, and distribution will be discussed
in detail in the numerical results section.

3.4 Numerical Results

This section provides the numerical results, simulations and comparisons of the
aggregate interference for the annular underlay model, and the Gamma model of
the aggregate interference.

3.4.1 Numerical results of Aggregate interference for annular

underlay networks

Here, we show the exact and asymptotic characteristics of the outage probability
with the variation of the primary power level, interfering CR power level, and node

30
 0

−0.5

−1
Outage Probability

−1.5

−2

−2.5

−3

−3.5
20
40 60
60 20 30 40 50
10
P (dBm) Ps (dBm)
p

Figure 3.1: The outage probability vs Pp and Ps , under γT h = 1, α = 2, σ = 2,

σn2 = 0.001, β = 0.0001, R = 30, RG = 15 , RE = 100.

density under differing conditions. A Gamma shadowing environment has been

i
considered for the simulation, and MI,approx has been used for our theoretical cal-
culations. When σ → 0, we have the situation where shadowing is negligible.
In Fig. 3.1, the outage probability is plotted with respect to both the primary
power level Pp and interferer power level Ps for fixed values of R, RG , RE and a
noise variance σn2 = 0.001. It uses free space propagation (α = 2), shadowing index
σ = 2, and the interferer density β = 0.0001. The outage probability decreases
slowly at high Ps and, approaches the noise limited scenario for low Ps values. Even
if Pp is increased, if Ps increases correspondingly, the outage probability remains
unchanged.
Fig. 3.2 depicts the same scenario of Fig. 3.1 except for the fact that in this
case, the shadowing variance σ = 0. It is interesting to observe that at higher Ps ,

31
 0

−0.5

−1
Outage Probability

−1.5

−2

−2.5

−3

−3.5
20
40 60
60 20 30 40 50
10
Pp (dBm) P (dBm)
s

Figure 3.2: The outage probability vs Pp and Ps , under γT h = 1, α = 2, σ = 0,

σn2 = 0.001, β = 0.0001, R = 30, RG = 15, RE = 100.

the σ = 0 plot has worse performance and vice versa at lower Ps .

Fig. 3.3 shows the exact and the asymptotic outage probability with respect to
Pp for differing levels of shadowing and path loss exponents. At high Pp , the asymp-
totic curves are a perfect match to the exact outage plots. An important observation
is that at this interferer node density, when the path loss exponent increases, shad-
owing has little effect on the outage. When the path loss exponent is 4, the plots
for both shadowing index values show little difference; however, under free space
propagation, the plots vary significantly for the two different σ values. This result
is consistent with the derivation obtained earlier in (3.22). For the values for RG ,
RE and R that we selected, the outage probability increases with α. If R << RG ,
a higher α will ensure a lower outage probability.
Fig. 3.4 compares the outage probability when the interferer density β varies,

32
 0
10

α=4
−1
10
Outage Probability

10
−2
α=3

−3
10 no shadowing
σ=2
Asymptotic
α=2
−4
Simulation
10

−5
10
30 35 40 45 50 55 60 65 70 75 80
P (dBm)
p

Figure 3.3: The exact and asymptotic outage probability vs the normalized transmit
power Pp , for different values of σ and α under σn2 = 0.001, β = 0.0001, Ps =
30dBm, R = 30, RG = 15, RE = 100, and γT h = 1.

for different values of path loss exponents and shadowing. This figure shows that
the effect of the shadowing index on environments with different path loss expo-
nents depends on the interferer node density. When the number of interferer nodes
increase to very high values, Pout approaches 1, while at low interferer node densi-
ties, Pout is governed primarily by noise. It is interesting to note that at high node
densities, the effect of α on the outage probability is minimal.
Fig. 3.5 compares the outage probability when the interferer density β varies,
for different values of path loss exponent α and shadowing index σ for RG = 25
and RE = 500. Unlike Fig. 3.4, this figure has a constant primary signal received
power, e.g. when power controlling is enabled on the primary system. It can be
seen that the effect of the shadowing index on environments with different path loss

33
 0
10

−1
10
Outage Probability

−2
10

α = 2, no shadowing
α = 2, σ = 1
10
−3 α = 2, σ = 2
α = 3, no shadowing
α = 3, σ = 1
α = 3, σ = 2
−4 α = 4, no shadowing
10 α = 4, σ = 1
α = 4, σ = 2

−50 −45 −40 −35 −30 −25 −20 −15 −10

β (dB)

Figure 3.4: The outage probability vs interferer density (β) for different values of
σ and α under Ps = 30 dBm, Pp = 70 dBm, σn2 = 0.001, R = 30, RG = 15,
RE = 100, and γT h = 1.

exponents depend on the value of interferer node density. For low node densities,
the effect of shadowing when the path los exponent α = 4 is lower than that when
α equals 2 or 3. When the number of interferer nodes increase to very high values
Pout approaches 1, while at low node densities, Pout is governed primarily by noise.
Fig. 3.6 compares the outage probability with respect to the primary transceiver
distance R. In this simulation, we vary R from 10m to 30m such that it lies within
the exclusion region, and also within the CR transmitting region. As we can see, the
outage probability increases with R, and also increases when the path loss exponent
α increases. It can also be observed that at lower α values, the outage for the
shadowing index σ = 0 and σ = 2 differs considerably. This is not the case at
higher path loss exponents. Also, at lower R values, the outage for α = 2 and

34
 0
10

α=2 α=3 α=4

−1
10
Outage Probability

σ=0
−2
10 σ=1
σ=2

−60 −50 −40 −30 −20 −10 0 10 20

β (dB)

Figure 3.5: The outage probability vs interferer density (β) for different values of
σ and α under Ps = 30 dBm, Pp R−α = 25 dBm, σn2 = 0.001, RG = 25, RE = 500,
and γT h = 1.

σ = 2 almost coincides with α = 3 and σ = 0. Therefore, the effect of shadowing

is extremely high at lower path loss exponent values.

3.4.2 Numerical results for the Gamma approximation

Fig. 3.7 shows the CDF of the simulated aggregate interference, and the CDF of
the Gamma equivalent aggregate interference. This figure reveals that the aggre-
gate interference roughly follows a skewed alpha-stable distribution [61], where the
skewness parameter reduces as β is increased. The two curves show a tight fit for
both large and small node density (β) values. Therefore, the Gamma approximation
of the aggregate interference is accurate.
Fig. 3.8 depicts the PDF of the approximated aggregate interference for dif-
ferent node densities. As expected, when the node density increases, the average
value of the aggregate interference increases and the tails become heavier due to the

35
 −1
10

−2
α=4
10
Outage Probability

−3
α=3
10

α=2
−4
10

σ=0
σ=2
−5
10
10 15 20 25 30
R

Figure 3.6: The outage probability vs primary transceiver distance (R) for different
values of σ and α under Ps = 30 dBm, γT h = 1, Pp = 70 dBm, RG = 15, RE =
100, and σn2 = 0.001.

increased variance. It is interesting to note that as the node density increases, the
PDF becomes more symmetric, and shows a similarity to the Gaussian distribution
suggesting that the central limit theorem would be appropriate for use in high node
density environments.
In Fig. 3.9, the PDF is plotted when RG , RE , and the shadowing variance σ
are varied. As σ is increased while keeping the other parameters constant, the PDF
flattens out and shows a heavier tail. As the guard distance RG is increased for
fixed shadowing, the interference power reduces as expected, and the PDF is much
tighter. Conversely, when RE is increased while keeping the shadowing index σ
constant, the PDF shows a similarity with the Gaussian distribution because many
interferer nodes are in the network.

36
 1

0.9 β = 0.0001

0.8 β = 0.0005

0.7
β = 0.001
0.6
CDF

0.5
β = 0.002
0.4

0.3

0.2
Exact
0.1
Gamma Approximation
0
0 5 10 15 20 25 30
Aggregate Interference Power

Figure 3.7: The CDF of the aggregate interference and the Gamma approximation
for different values of β under Ps = 30 dBm, RG = 15, RE = 100, α = 2, and no
shadowing.

3.5 Conclusion

This chapter investigated the aggregate interference on the primary receiver from
interferers for the annular underlay model. A Poisson point process of interferer
nodes, a composite fading model with the Generalized-K distribution, and arbitrary
path loss exponent values were considered. The exact and approximate MGFs of
the aggregate interference, and the exact and asymptotic outage probabilities of the
PR were derived. Our numerical results confirmed the analysis and showed that the
effect of shadowing is significantly lower with higher path loss exponent values.
The aggregate interference was further modeled as a Gamma random variable.
This was necessitated by the mathematical complexity of using the MGF for fur-
ther analysis. The first two moments of the aggregate interference were derived
and matched to the respective moments of a Gamma distribution. This resulted in
a simple yet highly accurate approximation of the aggregate interference which in-

37
 0.25
−4
β = 1´ 10
−4
β = 5´ 10
0.2 β = 1´ 10−3
β = 2´ 10−3

0.15
PDF

0.1

0.05

0
0 5 10 15 20 25 30 35 40
Aggregate Interference Power

Figure 3.8: The PDF of the Gamma approximated aggregate interference for dif-
ferent values of β under Ps = 30 dBm, RG = 15, RE = 100, α = 2, and no
shadowing.

corporated all the variable system and channel parameters. The approximation is
highly accurate in comparison to the actual distribution for varying node density
values.

xr

38
 0.25

RG = 15, RE = 100, σ = 0
0.2
RG = 15, RE = 100, σ = 1
RG = 40, RE = 100, σ = 0
0.15 RG = 15, RE = 500, σ = 0
PDF

0.1

0.05

0
0 10 20 30 40 50
Aggregate Interference Power

Figure 3.9: The PDF of the Gamma approximated aggregate interference for differ-
ent values of RG , RE , and σ under Ps = 30 dBm, β = 1 × 10−3 , and α = 2.

39
Chapter 4

Approximation models for the

aggregate interference

In Chapter 3, an exact analysis and a Gamma approximation to the aggregate in-

terference were developed. This chapter will introduce two more models for the
aggregate interference: the multiple-ring model, and the nearest interferer model.

4.1 Introduction
Previously, Chapter 2 introduced the annular underlay model, and Chapter 3 pro-
vided an exact analysis and a Gamma approximation to the aggregate interference.
This chapter will develop several different models to characterize the aggregate in-
terference which can be used to approximate the annular underlay model, and which
are derivatives of that model.
Many approximations and statistical models for the aggregate interference have
been widely investigated in literature. For example, [59] proposes a model for the
aggregate interference as the sum of normal and log-normal random variables. Ref-
erence [59] also obtains an upper bound for the complementary cumulative distri-
bution function (CCDF) of the total interference. In [30], the authors model the
aggregate interference when the CR nodes employ power control, contention con-
trol, and hybrid power and contention control schemes. The authors fit some of the
interference PDFs with log-normal PDFs to reduce complexity. Reference [52] pro-
poses a new statistical model for aggregate interference using cumulants obtained
via the application of Campbell’s theorem, while [61] shows that when the CR ex-

40
clusion region around the primary receiver (PR) is zero, the aggregate interference
can be accurately modeled as a heavy tailed α-stable distribution. Reference [61]
further says that when a significant exclusion region is defined, the tails of the ag-
gregate interference PDFs shorten considerably. Reference [60] investigates the
aggregate interference distribution via the central limit theorem. It is shown that
interferers in the near field cause the aggregate interference to be a heavy tailed dis-
tribution, and in the far field to be a Gaussian distribution. Furthermore, it is shown
that the presence of fading reduces the convergence of the aggregate interference
to a Gaussian distribution. Moreover, reference [50] presents a statistical model for
cognitive radio networks.
This chapter makes two main contributions. First, a new multiple-ring model
for the aggregate interference from the annular underlay network is developed. This
model is accurate, versatile, and mathematically tractable. Second, a nearest inter-
ferer model to approximate the aggregate interference is developed. This model is
useful when one particular interfering CR node is more dominant. In some practical
situations, this is indeed the case and this analysis will help in developing better in-
terference mitigation procedures. The accuracy of this approximation is evaluated
for different parameters.

4.2 A multiple-ring model for underlay interference

4.2.1 Introduction

The annular underlay model in Section 2.3 consists of a Poisson field of cognitive
radio (CR) nodes over an annular area with a guard region to reduce interference
(Fig. 4.1.a). However, the analysis developed in Chapter 3 is fairly complicated.
Therefore, we propose a new model that yields simpler analytical results, and which
provides flexibility and versatility to handle different parameters. This model is
developed as an approximation of the annular underlay network configuration (Fig.
4.1.a), repeated here for covenience from Section 2.3.

41
 RE
R2

RG R

RM

(a) Spatially distributed interferer nodes in (b) Proposed multiple-ring model with the
the shaded region. interferer nodes on the rings.

Figure 4.1: System model. Legend: Black circle = interferer nodes, black square =
PR, R2 = 2R, and RM = MR.

This new model consists of a set of multiple rings around the PR (Fig. 4.1.b)
with a set of interferer nodes on each ring. The system parameters of the new
model are chosen to match those of the annular underlay model. The MGF of the
aggregate interference for this model will be analyzed under both Rayleigh fading,
and composite fading and shadowing. The probability density function (PDF) of the
aggregate interference will be obtained for deterministic interferer node numbers,
and the exact and asymptotic outage probabilities are derived.
The approximation error of the proposed model is small, and can be further
reduced by fine tuning the parameters.

4.2.2 Proposed System Model

The common model for an interferer system is the annular underlay model (Fig.
2.3) discussed in the preceding chapter, with a guard region of radius RG , and an
RE
outer radius of RE . Usually, RG
is taken to be between 3 to 20 [49]. For this system,
we will develop a new multiple-ring model (Fig. 4.1).
In the proposed model, the interferer nodes are distributed in multiple rings
around the PR (Fig. 4.1.b). The radius of the t-th ring is tR, where t = 1 . . . M.

42
For M rings in total, MR is taken for the distance RE . The radius of the first ring
R can be treated as the guard distance RG . If a smaller value than RG is used, the
guard distance can be taken as the distance to the ν-th ring, νR.
The total number of interferer nodes in a certain ring is modeled as a Poisson
random variable arising from a linear Poisson point process. It is independent of the
number of interferer nodes in other rings, and the interferer nodes are distributed
uniformly on the ring. Let Nk be the number of interferer nodes located on the t-th
ring. P (Nt = n) can be written from (2.17) as

(βt 2πtR)n −βt 2πtR

P (Nt = n) = e , n = 0, . . . (4.1)
n!

where t is any integer from 1 to M, and βt is the interferer density of the t-th ring.
The density βt is chosen such that the average number of interferer nodes in the
multiple ring model is equal to that of the annular underlay model:
M
X
2 2
β(RE − RG )= βt 2tR. (4.2)
t=ν

For the considered homogeneous scenario, βt will not depend on the ring, and will
be a constant βl .

4.2.3 Interference Statistics

To demonstrate the validity of the multiple-ring model, we will next derive the MGF
of its aggregate interference. This derivation will consider Rayleigh fading, and
combined Rayleigh fading and Gamma shadowing. These MGF expressions are
simple, exact, and can be compared against the exact MGF of the annular underlay
model (Fig. 4.1.a).
As observed, the average number of interferer nodes in a given ring increases
with t. The worst case where all the interferer nodes in the rings are transmitting is
assumed. Even if activity factors of the interferer nodes are considered, the distri-
bution can be considered to be that of the active nodes without loss of generality.
The path loss will be modeled to follow the simplified path loss model of (2.16).
In our system model, we assume that all the interferer nodes are transmitting at the

43
same transmit power, and thus, the power level (defined as P0 r0α ) of each interferer
is Ps . The total interference power received at the PR may be written as (2.19)
M
X
I= It , (4.3)
t=1

where It is the total interference generated from the interferer nodes of the t-th ring.
It can be written as
Nt
X
It = Ps Xt,l [tR]−α , (4.4)
l=1
where Xl is the channel gain corresponding to fading or combined shadowing and
fading, for the l-th interferer in the t-th ring. We assume that all the fading and
shadowing of the interferer signals are independent of each other, even when they
are from the same ring.
If only Rayleigh fading is considered, Xt,l can be modeled to follow the expo-
nential distribution (2.5). Similarly, if both Rayleigh fading and Gamma shadow-
ing are considered, Xt,l was shown to be modeled as a generalized-K distribution
(2.11). In Chapter 2, it was stated that the generalized-K distribution can be ap-
proximated by a Gamma distribution as proposed under [29].
Using (4.4) in (4.3), the total interference can be written as
M X
X Nt
I= Ps Xt,l [tR]−α . (4.5)
t=1 l=1

Let MI (s) be the MGF of the aggregate interference given by (2.20). Using (4.5),
the MGF can be written as
PM PNt
Ps Xt,l [tR]−α
MI (s) = E[e−s t=1 l=1 ]
PNt
−sPs R−α M
P
t=1 t l=1 Xt,l
−α
= E[e ]. (4.6)

Rayleigh fading

Under Rayleigh fading, all the Xt,l values will be independent exponential random
variables. Therefore, we can write (4.6) after expanding as

MI (s) = E [EX1,1 [e−sPs 1

−α R−α X
] . . . EX1,N1 [e−sPs 1
−α R−α X
1,1 1,N1
]...

44
 × EXM,1 [e−sPsM
−α R−α X
] . . . EXM,NM [e−sPs M
−α R−α X
M,1 M,NM
]], (4.7)

where the first expectation is with respect to N1 . . . NM . After performing the inner
expectations, the MGF becomes
 
1 1
MI (s) = E ...
(1 + sPs R−α 1−α )N1 (1 + sPs R−α M −α )NM
"M #
Y 1
= E . (4.8)
−α t−α )Nt
t=1 (1 + sPs R

Because the number of interferer nodes in the t-th ring is independent of the number
of interferer nodes in any other ring, we can write (4.8) as
   
1 1
MI (s) = EN1 . . . ENM . (4.9)
(1 + sPs R−α 1−α )N1 (1 + sPs R−α M −α )NM

Using (4.1), the expectation with respect to the number of interferer nodes in the
t-th ring Nt can be written as
∞
(βl 2πtR)Nt −βl 2πtR
  X
1 1
ENt Nt
= e
(1 + sPs R−α t−α ) Nt =0
(1 + sPs R−α t−α )Nt Nt !
 Nt
∞ βl 2πtR
X 1+sPs R−α t−α
= e−βl 2πtR
N =0
N t !
t
 
1
βl 2πtR −1
= e 1+sPs R−α t−α . (4.10)

Using this result (4.10), (4.9) can be written as

     
1 1 1
βl 2πR −1 βl 2π2R −1 βl 2πM R −1
MI (s) = e 1+sPs R−α 1−α e 1+sPs R−α 2−α ...e 1+sPs R−α M −α

M  
1
βl 2πtR −1
Y
= e 1+sPs R−α t−α , (4.11)
t=1

which is the MGF of the aggregate interference.

Composite Rayleigh fading and shadowing

When composite Gamma shadowing and Rayleigh fading are considered, it was
shown earlier in the section that Xt,l values can be represented by a Gamma PDF.

45
It is assumed that all the Xt,l coefficients are independent. Therefore, similar to
Rayleigh fading, we can write MI (s) as (4.7). After performing the expectations
with respect to the Xt,l values, MI (s) becomes
" #
1 1
MI (s) = E ...
(1 + θsPs R−α 1−α )kN1 (1 + θsPs R−α M −α )kNM
"M #
Y 1
= E . (4.12)
(1 + θsP R −α t−α )kNt
t=1 s

Similar to the case with Rayleigh fading, the number of interferer nodes in the t-th
ring is independent of the number of interferer nodes in any other ring. Therefore,
the MGF can be written as
" # " #
1 1
MI (s) = EN1 . . . ENM .
(4.13)
(1 + θsPs R−α 1−α )kN1 (1 + θsPs R−α M −α )kNM

Similar to the derivation of (4.10), the expectations with respect to the number of
interferer nodes in the rings can be performed. Finally, the MGF of the aggregate
interference for Rayleigh fading and Gamma shadowing becomes
 
M 1
Y βl 2πtR −1
MI (s) = e (1+θsPs R−α t−α )k . (4.14)
t=1

Extended guard region

When the guard region is extended, the MGF of the aggregate interference is de-
pendent on the rings beyond the ν-th ring (1 < ν < M). Thus, the MGF of the
Rayleigh fading scenario becomes
M  
1
βl 2πtR −1
Y
MI (s) = e 1+sPs R−α t−α . (4.15)
t=ν

The MGF of the combined fading and shadowing case is similar.

46
4.2.4 Suitability as a stand-alone system model

This section will discuss the suitability of the multiple ring model as a stand-alone
system model, and the special cases which may occur.
The analysis of the annular underlay network model (Fig. 4.1.a) presents several
challenges.

• The incorporation of different transmit powers, path loss exponents, and shad-
owing variances is complicated.

• Non-homogeneous setups and areas with different CR node densities are dif-
ficult to analyze.

The proposed multiple ring model can be used to incorporate these conditions.
Moreover, the multiple-ring model may be used to approximate non-annular shapes.
For example, interfering nodes over a different polygon region may be approxi-
mated.

Differing parameters for interferer nodes

Substituting βt instead of βl , αt instead of α, and Pt instead of Ps , in equations

(4.11) and (4.14) will give the desired MGF when the node densities, path loss
exponents, and the transmit powers of the interferer nodes are unique. The symbols
βt and Pt denote the node density and transmit power of an interferer in the t-th ring,
respectively. Therefore, this multiple-ring system model is extremely versatile to
handle general cases. Comparatively, the MGF obtained in the previous chapter for
the standard system model is fixed for a particular interferer transmit power and a
node density.

Non-random number of interferer nodes

Another special case that can arise is the rings having deterministic interferer node
numbers. Therefore, the number of nodes on the t-th ring Nt is not random. In such

47
 RG RE

Figure 4.2: Linear interferer network. Legend : black dot = interferer nodes, black
square = PR.

a scenario, the MGF for the Rayleigh fading case becomes (from 4.8)
M
Y 1
MI (s) = . (4.16)
t=1 (1 + sPs R−α t−α )Nt
The PDF of the aggregate interference is difficult to obtain when Nt is random. But,
it can be derived for this special case using the method of [62]. It can be clearly
seen that (4.16) is equivalent to the MGF for a sum of Gamma random variables.
Therefore, using (2.9) of [62], the PDF of the aggregate interference can be written
as
∞ x
−
X δt xρ+t−1 e Ps R−α M −α
fI (x) = C , x>0 (4.17)
Γ(ρ + t) (P R −α M −α )ρ+t
t=0 s

PM QM t
αNt 1
Pt+1
where ρ = t=1 Nt , C = t=1 M , δt+1 = t+1 i=1 iγi δt+1−i , δ0 = 1,
PM Nt α i
and γi = t=1 i 1 − Mt



. The PDF for the combined Rayleigh fading and
shadowing case can be derived similarly.

Approximating interferers in a linear network

To approximate a homogeneous linear interferer network (Fig. 4.2), the guard dis-
tance RG is taken as the distance to the ν-th ring νR, and MR is approximated as
RE . The interferers will be distributed on the rings according to the Poisson point
process of (4.1). The density βt is chosen such that:
M
X
βln (RE − RG ) = βt πtR, (4.18)
t=ν

where βln is the interferer density per unit length in the linear interferer network.
Unlike the approximation for the homogeneous annular underlay model, βt values
will not be equal. The average number of interferers in each ring will be a constant
carrying the form;
βt 2πtR = C, (4.19)

48
where C is a constant.

Approximating interferers in a square shaped area

The interfering nodes may be distributed homogeneously in polygon shaped areas.

As an example, lets consider a square shaped area (Fig. 4.3). Similar to the previous
cases, RG is taken as the distance to the ν-th ring νR. However, the distance to the
√
M-th ring MR is taken as 2RE . While tR < RE , the interferer density on the
√
t-th ring is a constant, βl . However, when RE < tR < 2RE , the interferer density
for the t-th ring will be proportional to the percentage of distance that it is within
the square. Therefore, the approximation will be as follows;
M
X
2 2
βsq (4RE − πRG )= βt 2πtR, (4.20)
t=ν

where βsq is the interferer density of the square region. The values for βt are,

RE
βt = βl , t < ,
R √
 
8ω RE 2RE
βt = 1− βl , <t< , (4.21)
2π R R

where ω = arccos RtRE in radians.

4.2.5 Performance analysis

CDF of the SINR

Here, we derive the CDF of the aggregate interference . The SINR γ at the PR can
be written as
−α
Pp Rpr Y
γ= , (4.22)
I + σn2
where Pp is the power level of the PT, Rpr is the distance between the primary trans-
mitter and receiver, σn2 is the noise variance, and Y is the channel gain between the
primary transmitter and receiver. Similar to the previous section, due to the math-
ematical complexity of analyzing for other cases, only the case where the primary

49
 RG

RE

Figure 4.3: Square shaped interferer network. The shaded area contains the inter-
ferers. Legend : black square = PR.

signals undergo path loss and Rayleigh fading is considered. Then, Y is a unit ex-
ponential PDF given by (2.5), with γ̄ = 1. The variables Y and I are independent.
Similar to the derivation of the previous chapter, the CDF of γ is
 
xσn2  
− x
Fγ (x) = 1 − e Pp R−α
pr MI −α
. (4.23)
Pp Rpr
For Rayleigh fading, substituting (4.11) for MI (s) in (4.23), the CDF of γ becomes,
  
1
 
xσn2 M β 2πtR −1
− Y l 1+ x Ps R−α t−α
Fγ (x) = 1 − e Pp R−α
 e Pp R−α
. (4.24)
pr pr

t=1

For Rayleigh fading and Gamma shadowing, substituting (4.14) for MI (s) in (4.23),
the CDF of γ becomes,
  
 
1
 M βl 2πtR k −1 
  
  
xσn2  
x
− Y 1+θ
−α Ps R−α t−α  
Pp Rpr
Fγ (x) = 1 − e Pp R−α
pr  e

.
 (4.25)
 t=1 
 

Substituting γT h instead of x gives the outage equation, where γT h is the threshold

SINR level of the PR.

50
 The PDF of the SINR can be found out by differentiating Fγ (x) for distinct
values of M. But, the PDF for a general M value is extremely complex to obtain.

Asymptotic analysis

The aim in this part of the subsection is to develop an asymptotic equation for the
outage probability under Rayleigh fading and both Rayleigh fading and Gamma
shadowing, ie. when Pp is considerably larger than Ps , γT h , and σn2 .
We first consider the Rayleigh fading case. By expanding ex = 1 + x + O (x2 )
for small x, we obtain
 
1
βl 2πtR −1
1+ x Ps R−α t−α Ps R−α t−α
x + O x2 .


e Pp R−α
pr = 1 − (βl 2πtR) −α
(4.26)
Pp Rpr
Therefore,
 
M βl 2πtR 1 M 
−1
Ps R−α t−α
x

Y 1+ Ps R−α t−α Y
e Pp R−α
pr ≈ 1 − βl 2πtR −α
x
t=1 t=1
P p R pr
M
!
Ps R−α X 1−α
≈ 1 − βl 2πR −α
t x. (4.27)
Pp Rpr t=1
 
xσn2
−
Substituting (4.27) in (4.24) and expanding e via ex ≈ 1 + x,Pp R−α
pr

M
! !
xσn2 Ps R−α X 1−α
 
FγAsy (x) ≈ 1 − 1 − −α
1 − βl 2πR −α
t x
Pp Rpr Pp Rpr t=1
M
!
2 −α X
σn Ps R 1−α 2


≈ −α
+ βl 2πR −α
t x + O x . (4.28)
Pp Rpr Pp Rpr t=1

σn2 PM
and B1 = βl 2πR PPs R t1−α we get
−α
Defining A1 = Pp R−α R−α t=1
pr p pr

FγAsy (x) = (A1 + B1 ) x + O x2 ,

(4.29)

which is the asymptotic CDF for the Rayleigh fading scenario.

Similarly, the asymptotic outage for composite Rayleigh fading and Gamma
shadowing is given by

FγAsy (x) = (A2 + B2 ) x + O x2 ,

(4.30)

51
 σn2 PM
and B2 = βl 2πR θkP sR
t1−α .
−α
where A2 = Pp R−α
, P R−α t=1
pr p pr

Moments

The first and second order moments of the aggregate interference are important
statistics. The, expected value is found as,

d
E[I] = (−1) MI (s)|s=0
ds
√ X M
E[I] = 2πβl R1−α Ps eσ2 t1−α . (4.31)
t=ν

The second moment is found as

2 d2
E[I ] = 2 MI (s)|s=0
ds
M
!2 M
√ 2X X
2 1−α 1−α 1−2α 2 2
E[I ] = 2πβl R Ps e σ t +2πβl R θ k(k + 1)Ps t1−2α (. 4.32)
t=ν t=ν

Error statistics

For any new model, it is important to investigate its error statistics. The root mean
squared error (RMSE) [63] will be investigated in this regard. If IAn and I are the
aggregate interference powers of the annular underlay model, and the multiple ring
√
model, the RMSE can be written as RMSE = MSE, where

MSE = E[(IAn − I)2 ]. (4.33)

Expanding this, and using (4.31), (4.32), (3.26), and (3.29),

2
MSE = E[IAn ] + E[I 2 ] − 2E[IAn ]E[I]
M
!2 M
√ 2X X
1−α 1−α 1−2α 2 2
= 2πβl R Ps e σ t +2πβl R θ k(k + 1)Ps t1−2α
t=ν t=ν
 2−2α 2−2α 
RE − RG
+ πβPs2 kθ2 (1 + k)
1−α

52
  √ 2  R2−α − R2−α 2
E G
+ 2πPs β eσ
2−α
M
!
√ 2X √ 2  R2−α − R2−α 
1−α σ 1−α E G
− 2 2πβl R Ps e t 2πPs β eσ .
(4.34)
t=ν
2 − α

Substituting βl using (4.2), MR for RE , and νR for RG , we obtain

2−2α 2−2α 2 2
PM 1−2α !
M − ν (M − ν ) t
MSE = πk(1 + k)θ2 Ps2 βR2−2α + PM t=ν
1−α t=ν t
 P 2
M
(M 2 − ν 2 )2 t=ν t
1−α
(M 2−α − ν 2−α )
2
2 2 2 2 2 4−2α 
+ 4π k θ Ps β R 2 +
(2 − α)2
 P
M
4 t=ν t
PM 1−α 2 !
2 2−α 2−α
t=ν t (M − ν ) (M − ν )
− . (4.35)
(2 − α) M
P
t=ν t

Coefficient of Variation of the Root Mean Squared Error (CVRMSE)

The coefficient of variation of the root mean squared error is given by

RMSE
CV RMSE = . (4.36)
E[IAn ]

Normalized error of the mean

The normalized error of the mean (NEM) value is defined as

|E[IAn ] − E[I]|
NEM = . (4.37)
E[IAn ]

The CV RMSE and NEM values for different parameter combinations have
been tabulated in Table 4.2.5 and Table 4.2.5. While the statistics for the R =
1, M = 100 pair are better, the NEM only depends on the path loss exponent value
for a given R, M pair. The multiple ring model shows better performance for higher
interferer densities, lower path loss exponents, and lower shadowing variances.

53
 CV RMSE NEM
β = 0.1, α = 2, σ = 0 0.0654 0.0318
β = 0.1, α = 2, σ = 1 0.0995 0.0318
β = 0.1, α = 3, σ = 0 0.1380 0.1061
β = 0.1, α = 3, σ = 1 0.1800 0.1061
β = 0.01, α = 2, σ = 0 0.1836 0.0318
β = 0.01, α = 2, σ = 1 0.2999 0.0318
β = 0.01, α = 3, σ = 0 0.2984 0.1061
β = 0.01, α = 3, σ = 1 0.4720 0.1061

Table 4.1: Error statistics for R = 5 and M = 20.

CV RMSE NEM
β = 0.1, α = 2, σ = 0 0.0552 0.0062
β = 0.1, α = 2, σ = 1 0.0907 0.0062
β = 0.1, α = 3, σ = 0 0.0831 0.0203
β = 0.1, α = 3, σ = 1 0.1343 0.0203
β = 0.01, α = 2, σ = 0 0.1736 0.0062
β = 0.01, α = 2, σ = 1 0.2861 0.0062
β = 0.01, α = 3, σ = 0 0.2555 0.0203
β = 0.01, α = 3, σ = 1 0.4205 0.0203

Table 4.2: Error statistics for R = 1 and M = 100.

54
4.3 Nearest Interferer approximation

4.3.1 Introduction

Until now, several models for the aggregate interference were developed. In Chap-
ter 3, a Gamma model was proposed, while in Section 4.2, a multiple-ring model
was developed. This section develops another approximation called the nearest in-
terferer approximation.
With exclusion regions and path loss included, the nearest interferer may dom-
inate the aggregate interference. Furthermore, when there are relatively few inter-
ferer nodes (low interferer density), Central Limit Theorem (CLT) based approxi-
mations are less accurate. Therefore, characterizing the interference from the near-
est node, and analyzing the conditions for its dominance are the focus points of this
section. As an intermediate step, the distance distribution of the nearest interferer
in the annular underlay network is derived. Moreover, another factor of interest is
the impact of power controlling on the aggregate interference.
Very few work has analyzed the spatial distribution of the nearest CR interferer,
and the approximation of the aggregate interference with the interference from the
nearest node. The works of [64] and [65] have approximated the aggregate interfer-
ence by that of the nearest node. Reference [64] investigates the trade-off between
outage probability and node density, and further studies the effects of interference
cancellation. Reference [65] checks the validity of the nearest node approxima-
tion and a Gaussian approximation to the aggregate interference. However, none
of the prior work do a performance analysis for the nearest node interference us-
ing a moment generating function (MGF) based approach. Furthermore, they do
not consider any power control schemes used by the interferer nodes. Also, none
have compared how the approximation will vary in accuracy as one or more system
parameters change.
This section’s main objective is to mathematically characterize the interference
caused by the nearest interferer, and approximate the aggregate interference with
it. In situations where the nearest interferer is more dominant, this analysis will

55
provide useful insights when designing CR networks. Furthermore, this will help
find the situations in which interference cancellation schemes where the dominant
interferer is cancelled provide the best results.
The following subsections will evaluate the probability density function (PDF)
of the nearest interferer distance, analyze its distribution, and approximate the ag-
gregate interference with the nearest CR interference. Using the approximation, the
MGF of the interference is obtained in closed-form for two cases. Namely, where
all interferer nodes transmit at a constant power, and where a basic distance based
power controlling scheme takes place. A closed-form solution is obtained for the
outage probability using the derived MGF. Finally, it will be shown that the nearest
CR interference is a lower bound to the aggregate interference which becomes tight
as either the interferer density or the distributed area of the interferers decreases, or
when the path loss exponent value increases.

4.3.2 System model

The annular underlay network model defined in Fig. 2.3 will be considered. The
interferer nodes are assumed to be distributed uniformly in a ring shaped area (Fig.
4.4) with the PR at the center, which has an inner radius RG , and an outer radius
RE .
The aggregate interference at the PR can be written as (2.19). From the general
path loss model (2.16), Ii is expressed as

Ii = Pi ri −α Xi , (4.38)

where Pi is the power level (defined as P0 r0α ) of the i-th interferer, ri is the distance
from the i-th interferer to the PR. When Rayleigh fading is considered, Xi can be
written as a unit exponential random variable of the form (2.5). In our model, we
consider independent fading of the interfering signals.

56
 RE

rmin R
G

Figure 4.4: System model. Legend: black circle = interferer nodes, black square =
PR.

4.3.3 Distribution of the nearest interferer node

Our aim is to approximate the aggregate interference with the interference from the
nearest interferer. Therefore, the aggregate interference can be written as

−α
I ≈ Pmin rmin X, (4.39)

where Pmin , rmin and Xmin respectively denote the power level, distance to the PR
and the fading coefficient, of the nearest interferer node.
The MGF of the aggregate interference can be obtained from (2.20). In order to
evaluate this equation, we need to obtain the PDF of rmin , which is defined as

rmin = min ((r1 , r2 , . . . rN )) .

Using the fact that Pr[rmin < x] implies at least one element of (r1 , r2 , . . . rN ) is
less than x, the CDF of rmin given N can be obtained as

Frmin /N (x) = Pr[rmin < x] = 1 − [Pr(ri > x)]N . (4.40)

The PDF of ri can be expressed as

 πri
2 AI , RG < ri < RE
fR (ri ) = . (4.41)
0 , otherwise

57
 π 2
From (4.41), P rob(ri>x) can be found out to be AI
(RE − x2 ). Substituting this in
(4.40) and differentiating, we get
 N
π 2

N −1
frmin /N (x) = 2N x RE − x2 . (4.42)
AI
Averaging (4.42) with respect to (2.17) for N 6= 0, we get
∞  N
X π 2

N −1 (βAI )N e−βAI
frmin (x) = 2x N RE − x2
N =1
AI N!
∞ 2 N −1
X (πβ (RE − x2 ))
= 2πβe−βAI x . (4.43)
N =1
(N − 1)!

With some mathematical manipulations, and substituting for AI , we can obtain

frmin (x) for N 6= 0 as

frmin (x) = 2πβxeπβ (RG −x ) , RG < x < RE .

2 2
(4.44)

When N = 0, frmin (x) is non-existent with probability e−AI β .

Fig. 4.5 shows the distribution of frmin (x) under differing β for two RG values.
It is observed that when β is higher, the PDF of frmin (x) shows a sharp drop-off.
Conversely, at lower β values, the curves show a more gradual drop with respect to
rmin . Therefore, it is concluded that at higher β, we can expect a significantly lower
value for rmin , and thus a higher aggregate interference. At higher RG , the curves
for all β values show increased skewness towards RG . It is interesting to note that
frmin (x) is not dependent on RE .

It is more insightful to obtain the PDF of rRmin

G
. f rmin (x) can be obtained through
RG

a simple variable change of (4.44) as

RE
f rmin (y) = 2πηyeπη(1−y ) , 1 < y <
2
, (4.45)
RG RG
2
where η = βRG . Fig. 4.6 shows the distribution of f rmin (y) when N 6= 0 under
RG
RE
differing η for RG
= 4. It is observed that when η is higher, the PDF of frmin (x)
shows a sharp drop-off. Conversely, at lower η values, the curves show a more
rmin
gradual drop with respect to RG
. When η gets significantly low, f rmin (y) does not
RG
RE
drop off with respect to RG
, and increases. Therefore, it is concluded that at higher

58
 0.1

β = 0.0001
0.08 β = 0.0005
β = 0.001
0.06
PDF

0.04

0.02

0
20 40 60 80 100
rmin (m)
(a) RG = 15m

0.16
β = 0.0001
0.14 β = 0.0005
β = 0.001
0.12

0.1
PDF

0.08

0.06

0.04

0.02

0
30 40 50 60 70 80 90 100
rmin (m)
(b) RG = 25m

Figure 4.5: frmin (x) for varying β values when N 6= 0

59
 3.5
η = 0.01
3 η = 0.05
η = 0.1
η= 0.5
2.5

2
PDF

1.5

0.5

0
1 1.5 2 2.5 3 3.5 4
rmin/RG

RE
Figure 4.6: f rmin (y) for varying η values when N 6= 0, RG
= 4.
RG
.
rmin
η, we can expect a significantly lower value for RG
, and thus a higher aggregate
interference. It is interesting to note that the shapes of the curves are not dependent
RE
on RG
.
We can observe that the nearest interferer terminal would have a much higher
effect on the PR’s aggregate interference when η is lower. Moreover, in practice,
when designing a CR system at higher η values, increasing the inner radius instead
of reducing the node density is a more effective technique to ensure a guaranteed
PR performance while maximizing the amount of CR nodes employed.

4.3.4 Derivation of the MGF

In the following subsections, we will derive the MGF of the aggregate interference
when the CR nodes employ constant transmit power and when the CR nodes employ
a distance based power control scheme.

60
Constant power interferer nodes

When the power levels of each interferer is constant, and not dependent on distance,
(2.20) can be written as

MI (s) = EX,rmin [e−sI ] = Ermin [EX [e−sPs rmin X ]],

−α
(4.46)

where Ps is the constant power of a interferer. Averaging with respect to X, we find

 
1
MI (s) = Ermin −α . (4.47)
1 + Ps srmin

When Ps sr −α < 1, we can expand (4.47) as an infinite series, and get

"∞ #
t
X
−α


MI (s) = Ermin −Ps srmin . (4.48)
t=0

For most practical values of system parameters, the summation will converge. Us-
ing (4.44) and averaging (4.48), we obtain the MGF as,

2
MI (s) = e−AI β + πβeπβRG
∞ 2 2
!!
X E tα(πβR )
G E tα(πβR )
E
× (−Ps s)t 2 tα−2 − 2 tα−2 . (4.49)
t=0
R G R E

The exact MGF can be obtained similar to the derivation of (3.12) as

i
MIexact (s) = eβAI (MI (s)−1) , (4.50)

where MIi (s) is the MGF of the interference from a single interferer, given by

MIi (s) = E[e−sPs ri Xi ]. (4.51)

The PDF of ri is given by (4.41), and the PDF of Xi is given by (2.5). Averaging
with respect to these gives the result
α α
 
i π 2 RG 2 RE
MI (s) = RG (W( ) −1) − RE (W( )− 1) , (4.52)
AI Ps s Ps s

where W(x) = 2 F1 (1, 2/α; 1 + 2/α; −x).

61
 rmin
θ
RCR

Figure 4.7: System model for a CR setup employing power control. Legend: black
circle = nearest interferer node, black square = PR, red square = CR receiver.

interferer nodes employing power control

In this subsection, we consider a distance dependent power control scheme [30] [66]
where the interferer nodes control their power in order to ensure a constant average
received power level to a single CR receiver. A practical example for such a scenario
would be wireless sensor nodes transmitting information to their base station.
Without loss of generality, we can take the PR to be located at the origin and the
CR receiver be located a distance RCR from the PR along the x-axis (Fig. 4.7). Let
θ be the angle from the positive x-axis to the nearest CR transmitter. We assume θ
to be uniformly distributed between 0 and 2π. Let Prec be the average power level
ensured at the CR receiver. RCR is assumed to be a constant in the analysis. But,
if required, it can be modeled as a random variable. The average transmit power of
the nearest interferer can be written as

2 2

α2
Pmin = Prec rmin + RCR − 2rmin RCR cos θ . (4.53)

Let IP C be the interference from the nearest interferer node when the power control
scheme takes place. The the MGF MIP C (s) of this can be written as

MIP C (s) = Eθ [Ermin [EX [e−sPmin rmin X ]]].

−α
(4.54)

62
Substituting for Pmin and carrying ou the expectation with respect to X, (4.54)
becomes,
  
1
MIP C (s) = Eθ Ermin −α . (4.55)
1 + sPmin rmin
After performing a series expansion, (4.55) becomes,
∞
X
MIP C (s) = (−sPrec )t
t=0
h h
−αt 2 2

αt2 ii
× Eθ Ermin rmin rmin+RCR− 2rmin RCR cosθ . (4.56)

When α is an even number, we can simplify (4.56) using the binomial expansion
and obtain the following equation for MIPapp
C
(s) after averaging with respect to θ and
rmin ;

MIP C (s) = e−AI β

2 c
αt k
∞ 2
 αt bX 
πβR2G
X X k
+ 2π 2 βe (−Prec s)t 2

t=0 k=0 l=0

2l k
2 2
 
2k−2l (4l − 1)!! Ek−l(πβRG ) Ek−l(πβRE )
× RCR 2k−2l−2
− 2k−2l−2
, (4.57)
(2l)! RG RE
where bxc denotes the largest integer less than x. The convergence of (4.56) is
relatively lower compared to (4.49). In the case that RCR is a random variable,
further averaging needs to be done on (4.57).

4.3.5 Moments

Moments of the nearest nodes’ interference may be important when cconducting

further research on the approximation. Specially, the first and second moments are
highly important.

Constant power interferer nodes

The general nth order moment of the interference from the nearest node can be
obtained from (4.49) as
dn
E[I n ] = (−1)n MI (s)|s=0
dsn
63
 !
2 2
2 E α2(πβRG ) E α2(πβRE )
= n!πβPsn (−1)n eπβRG α−2 − α−2 . (4.58)
RG RE

interferer nodes employing power control

Similar to above, the general nth order moment for the power controlling scenario
is obtained from (4.57) as

2 c
αt k
∞ 2  αt b 
πβR2G
X X X k
E[IPn C ] = 2n!π 2 βPrec
n
(−1)n e (−Prec s)t 2

t=0
k 2l
k=0
 2 2
 l=0
2k−2l (4l − 1)!! Ek−l(πβR G ) Ek−l(πβR E )
× RCR 2k−2l−2
− 2k−2l−2
. (4.59)
(2l)! RG RE

4.3.6 Outage Analysis

In this section, the outage probability at the PR will be derived. The signal to
interference and noise ratio (γ) at the PR is denoted similar to the previous section.
The CDF of γ can be obtained as
 
xσn2  
−
Pp R−α
x
Fγ (x) = 1 − e MI . (4.60)
Pp R−α
For the constant transmit power interferer nodes, Fγ (x) becomes
   
xσn2 xσn2
− AI β+ πβR2G −
Fγ (x) = 1 −e − πβe
Pp R−α Pp R−α

!!
∞   t E tα(πβR2 ) 2
E tα(πβRE )
X Ps x 2 G
× − −α tα−2 − 2 tα−2 . (4.61)
t=0
P p R RG RE

For interferer nodes employing power control (4.3.4), Fγ (x) becomes

 
xσn2
− AI β+
FγP C (x) = 1 −e Pp R−α

2 c
αt k
∞ 
 
xσn2 t X 2
 αt bX 
2 πβR2G −X
Pp R−α
Prec x 2
k
− 2π βe − −α
t=0
P p R k=0
k l=0
2l
2 2
 
2k−2l (4l − 1)!! Ek−l(πβRG ) Ek−l(πβRE )
× RCR 2k−2l−2
− 2k−2l−2
. (4.62)
(2l)! RG RE
Substituting the threshold level γT h for x gives us the outage probability.

64
4.4 Numerical Results

In this section a comparison is done between simulation results and the theoretical
equations in order to assess their accuracy. Furthermore, performance trends due to
parameter changes will be examined.

4.4.1 Numerical results of the multiple-ring model

This section shows the accuracy and performance of the multiple-ring model. The
exact and asymptotic variation of the outage probability with Pp is shown. Sim-
ulations are performed for both shadowing and non-shadowing environments to
confirm our theoretical results. Moreover, comparisons between the multiple-ring
model and the annular underlay model are performed. The following parame-
ter values will be used: noise variance σn2 = 0.001, interferer node power level
Ps = 30 dBm, γT h = 1 and Rpr = 30 for all the plots for comparison purposes.
It is necessary to establish the accuracy of the multiple-ring model for varying
path loss exponent (α) and shadowing levels (σ). In Fig. 4.8, the theoretical results
match perfectly with the simulations, and as Pp increases, the asymptotic curves co-
incide with the exact curves. When the path loss exponent increases, shadowing has
little affect on the outage. But, for free space propagation (α = 2), the shadowing
effects are substantial, and the curve for σ = 2 under α = 2 has worse performance
than the curve for Rayleigh fading under α = 3. For the chosen Rpr , the outage
increases with α, but this may not be the case if Rpr had been significantly smaller
than R.
We now compare our proposed model (Fig. 4.1.b) with the annular underlay
model (Fig. 4.1.a) in Fig. 4.9. We simply consider all rings with equal density such
that βl = 0.01. Then, the equivalent node density in the annular underlay model
is 6.25 × 10−4 . For these curves, the outage probabilities are virtually identical for
higher path loss exponent values, and extremely close under free space propagation,
while shadowing doesn’t seem to affect the accuracy of the outage significantly.

65
 0
10

−1
10 α=4
Outage probability

−2
10

−3 α=2
10

Rayleigh fading only

Fading and shadowing with σ = 2
−4
10
Simulation
Asymptotic
−5
10
40 45 50 55 60 65 70 75 80
P (dBm)
p

Figure 4.8: The exact and asymptotic outage probability vs the primary system
power level Pp , for different values of σ and α under βl = 0.01, R = 20, and
M = 5.

However, as was shown before, the error statistics of the model deteriorates when
shadowing is present. The perceived increase in accuracy of the outage at higher
path loss exponents is mainly due to the affects of noise.
It is important to calculate the error of the proposed model, and to observe
how the error changes as a function of ring radii. In Fig. 4.10, the percentage
error of the multiple-ring model is compared for different R and M values. The
parameters for the annular underlay model are β = 0.001, a guard distance of 20,
and the outer distance of 100. R and M pairs of (20,5), (10,10), and (5,20) are used
for comparisons. The node density βl for each case will be 0.016, 0.008889, and
0.004706 respectively. For R = 10, t is taken from 2 to M, and for R = 5, t is
taken from 4 to M. The percentage error reduces with R. With respect to Pp , the

66
 0
10

−1 α=4
10
Outage probability

−2 α=2
10

−3
10

Multiple ring model, σ = 0

−4
10 Annular model, σ = 0
Multiple ring model, σ = 2
Annular model, σ = 2
−5
10
40 45 50 55 60 65 70 75 80
Pp (dBm)

Figure 4.9: Comparison of the outage probability vs primary system power level
Pp with different values of σ and α, for the multiple-ring model and the annular
underlay model under βl = 0.01, R = 20, and M = 5. The values for β, RG and
RE have been chosen accordingly.

percentage errors for all 3 cases rise up to a certain level, and then keeps constant.
Furthermore, it is necessary to compare the performance of the proposed model
under varying CR node densities. In Fig. 4.11, the outage for the proposed system
model is almost identical to the annular underlay model. Moreover, there is no
increase of error as the interferer node density β increases. The approximation
error probability under Pp = 60 dBm is 2.9% for β = 0.001, 2.9% for β = 0.01,
and 1.9% for β = 0.1.

67
 14

Approximation error (%) 12

10
R=20, M=5
R=10,M=10
8 R=5,M=20

0
40 45 50 55 60 65 70 75 80
P (dBm)
p

Figure 4.10: The error percentage of the outage of the multiple-ring model vs the
primary system power level Pp , for different values of M and R under β = 0.001,
α = 2, and σ = 0. The value of βl has been chosen appropriately for each case.

4.4.2 Numerical results for the nearest interferer approximation

In this subsection, we show the behaviour of the exact outage probability, and the
nearest interferer approximation with respect to changing the primary power level
Pp and the guard distance RG , for different conditions, and obtain insights into the
situations where the nearest interferer approximation is applicable.
The outage with respect to the primary transmit power level Pp for the scenario
when interferer nodes employ basic power control is plotted in Fig. 4.12. We con-
clude that the approximation is very tight for low Prec values. When RCR increases,
the outage probability increases correspondingly because of the need to transmit at a
higher power, and the approximation diverges from the exact value. Thus, although

68
 0
10

β = 0.1
−1
10
Outage probability

10
−2 β = 0.01

10
−3 β = 0.001

−4
10 Annular model
Multiple ring model

−5
10
30 40 50 60 70 80 90
Pp (dBm)

Figure 4.11: The outage probability vs the primary system power level Pp , under
different values of β, for the annular underlay model and the multiple-ring model.
α = 2, σ = 0, R = 5, and M = 20.

the approximation shows a tight match when RCR = 0, it shows a significant vari-
ation at RCR = 275. The rest of this section will compare the performance of the
constant powered CR case.
Fig. 4.13 compares the outage probability of constant powered interferer nodes
under different primary transceiver distance R values. As R increases, the outage
probability increases significantly. This is because, when R increases, the primary
signal power diminishes and the interference power has a more significant effect on
the receiver. But, it is interesting to note that R does not have any bearing on the
error between the exact outage probability and the approximation. In addition, Fig.
4.13 compares the theoretical values of the outage probability to the ones obtained
through Monte-Carlo simulations. We see that the simulations and theory match

69
 0
10

Outage Probability

R = 0 exact
CR
R = 0 approximation
CR
−1
10 R = 275 exact
CR
R = 275 approximation
CR

25 30 35 40
P (dBm)
p

Figure 4.12: Outage probability vs Ps for power controlled interferer nodes under
RCR = 0 and RCR = 275 for β = 0.0001, α = 2, RE = 200, RG = 25, R = 30,
Prec = −30 dBm, γT h = 1, and σn2 = 0.001.

well.
Fig. 4.14 shows the variation of the outage probability with respect to the guard
distance RG , under three different node density β values. It is observed that at
higher RG , the exact outage probability, and the outage given by the nearest in-
terferer approximation converge. This is because the average number of interferer
nodes reduces as RG is increased and therefore, the nearest interferer is more dom-
inant. In addition, we see that the accuracy of the approximation depends on the
interferer density. At lower β values, the approximation is a close to the outage
probability, while at higher β, the two curves are more divergent. AS RG is in-
creased further, the performance of the PR is mainly inhibited by noise due to the
interference power reducing due to path loss. Therefore, the curves flatten out.
In Fig. 4.15, we plot the variation of the outage probability with respect to the
interferer power level Ps . It is seen that when the path loss exponent α is high,
the approximation is extremely accurate to the actual aggregate interference. This

70
 0
10

R = 20 R = 50 R = 100

Outage probability

−1
10

Theory (exact)
Theory (approximation)
Simulation
−2
10
20 25 30 35 40 45
P (dBm)
p

Figure 4.13: Outage probability vs Pp for different values of R, with RE = 500,

RG = 50, Ps = 20 dBm, β = 0.0001, α = 2, γT h = 1, and σn2 = 0.001.

is because at higher path loss factors, the interference is dominated by the nearest
interferer. In many practical wireless channels such as in dense urban environments
or hilly terrain, α is particularly high and the nearest interferer approximation is
extremely valid. For a path loss exponent value of 2 (free space propagation), when
Ps is low, the aggregate interference can be validly approximated by the interference
from the nearest interferer. Even when Ps is increased for free space propagation,
the error is minimal and when we reduce β, the error gets further reduced.
Therefore, the nearest interferer approximation is reasonably accurate under
several conditions. These include high path loss exponent values, lower interferer
node densities, and higher guard distances (RG ).

4.5 Conclusion

This chapter included 1) a new multiple-ring model to analyze the aggregate inter-
ference, and 2) the nearest interferer approximation to the aggregate interference.

71
 0.06

0.055 β = 2 ´ 10
−4

0.05 actual
approximate
0.045
Outage Probability

0.04
−4
0.035 β = 1 ´ 10

0.03
−5
β = 5 ´ 10
0.025

0.02

0.015
20 40 60 80 100 120 140 160
R
G

Figure 4.14: Outage probability vs RG for different values of β, with RE = 200,

R = 40, Pp = 50 dBm, Ps = 30 dBm, α = 2, γT h = 1, and σn2 = 0.0005.

For the proposed multiple-ring model, channel effects such as path loss, fading, and
shadowing were incorporated while shadowing was not considered for the nearest
interferer approximation due to mathematical complexity.
For the multiple-ring model, the number of interfering CR nodes in a partic-
ular ring was modeled as a Poisson process, and two fading models were consid-
ered; namely, a Rayleigh fading channel, and a composite fading model with the
Generalized-K distribution. The exact MGF of the aggregate interference was ob-
tained for both the cases. For the special case of a fixed number (deterministic) of
interferer nodes in the rings, the PDF of the aggregate interference was also derived.
The exact and asymptotic outage probabilities were derived, and it was shown that
the multiple-ring model can be used to accurately approximate the annular underlay
model.
In the nearest interferer approximation, the interference from the closest inter-
ferer node was approximated to the aggregate interference. The distribution of the
distance from the PR to the nearest interferer was derived and analyzed. The ex-

72
 0
10

−4
β = 10 , α = 6
−1
10
−4
β = 10 , α = 2
Outage Probability
−2
10
−4
β = 10 , α = 4
β = 5 ´ 10−5, α = 2
−3
10

−4 exact
10
approximation

−5
10
0 10 20 30 40 50 60 70 80
P (dBm)
s

Figure 4.15: Outage probability vs Ps for different values of β and α, with RE =

200, RG = 20, R = 10, Pp = 60 dBm, γT h = 1, and σn2 = 0.001.

act MGF of the interference for the approximation was derived for two scenarios;
namely, when the interferer nodes employ no power control, and when they employ
a distance based power control scheme. The outage probability was analyzed using
the derived MGFs when the primary signals also undergo Rayleigh fading. Simu-
lations confirmed the analysis and established that the approximation is extremely
tight under certain conditions. These conditions are higher path loss exponents,
lower node densities, and larger exclusion regions. Furthermore, the transmitter re-
ceiver distance of the primary network has minimal bearing on the accuracy of the
approximation.

73
Chapter 5

Conclusions and Future Research

Directions

5.1 Conclusions

This thesis investigated the aggregate interference on the primary receiver from a
random number of cognitive nodes accessing the licensed spectrum concurrently
with the primary system. The annular underlay network of cognitive radio nodes,
which was modeled as a Poisson point process, was considered.
Chapter 2 analyzed the aggregate interference under composite Rayleigh fading
and Gamma shadowing from interferer nodes in the annular model. The MGF
of the aggregate interference, moments, outage, and asymptotic expressions were
derived. The variation of outage under different shadowing conditions was lower at
high path loss exponent values. Furthermore, the aggregate interference was shown
to be reasonably approximated by a Gamma distribution.
In Chapter 3, a new Multiple-ring model for the interference was proposed. The
proposed model was shown to be highly versatile and mathematically simple while
being extremely accurate. The accuracy vs complexity was shown to be adjustable
by varying system parameters. Moreover, the ability of the proposed model to
accurately model different systems was discussed. Furthermore, the interference
from the nearest interferer node was analyzed. Under certain system conditions,
the nearest interferer dominates the aggregate interference, and yields a reasonable

74
approximation.

5.2 Future Research Directions

Several improvements and extensions can be performed in the future.

• Secondary medium access control

The analysis did not consider mutual interference among CR nodes. For ex-
ample, if a particular CR is receiving, another nearby CR transmission will
cause interference. Thus, medium access in the secondary network can be
considered for a more realistic analysis.

• Non-homogeneous cases
While interferer nodes were assumed to be a homogeneous Poisson process,
this may not always be a valid assumption, and non-homogeneous analysis
could prove useful. For example, in a city, node density decreases as the
distance from the city center increases. Moreover, interferer nodes distributed
in clusters can also be considered where the aggregate interference is due to
signals from multiple clusters.

• Mobility of nodes
The interferer nodes have been assumed to be static. In practice, these nodes
can move around randomly. Such mobility leads to time varying interfer-
ence, and analytical results may be derived for those cases. Moreover, the
power level of the interferer nodes has been assumed to be static in most
cases. However, due to various dynamic power control schemes and system
configurations, the power levels of the different interferer nodes vary in time.

• Correlated fading/shadowing
The work has assumed independent fading/shadowing on the links between
the different interferer nodes and the primary receiver. A possible extension
is to consider correlations in these variables.

75
• Analysis of different channel models
This extension to the work includes carrying out the interference analysis for
different fading/shadowing and path loss models.

76
Bibliography

[1] Qualcomm. (2011) The visible light communications mo-

tivation. [Online]. Available: http://visiblelightcomm.com/
the-visible-light-communications-motivation/

[2] I. Gradshteyn and I. Ryzhik, Table of Integrals, Series, and Products, 7th ed.
Academic Press, 2007.

[3] A. Molisch, Wireless Communications. Wiley-IEEE Press, 2011.

[4] Cisco. (2012) Cisco visual networking index: Global mo-

bile data traffic forecast update: 2011-2016. [Online]. Avail-
able: http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/
ns705/ns827/white paper c11-520862.html

[5] ITU. (2012) Key statistical highlights: ITU data release june 2012.
[Online]. Available: http://www.itu.int/ITU-D/ict/statistics/material/pdf/2011
%20Statistical%20highlights June 2012.pdf

[6] S. Haykin, “Cognitive radio: brain-empowered wireless communications,”

IEEE J. Sel. Areas Commun., vol. 23, no. 2, pp. 201–220, Feb. 2005.

[7] J. Mitola, “Cognitive radio: An integrated agent architecture for software de-
fined radio,” Ph.D. dissertation, KTH Royal Inst. of Technol., Stockholm,
Sweden, 2000.

[8] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty, “Next

generation/dynamic spectrum access/cognitive radio wireless networks:
A survey,” Computer Networks, vol. 50, no. 13, pp. 2127–2159,

77
 2006. [Online]. Available: http://www.sciencedirect.com/science/article/pii/
S1389128606001009

[9] R. Menon, R. Buehrer, and J. Reed, “Outage probability based comparison of

underlay and overlay spectrum sharing techniques,” in Proc. IEEE DYSPAN
2005, Nov. 2005, pp. 101–109.

[10] B. Wang and K. Liu, “Advances in cognitive radio networks: A survey,” IEEE
J. Sel.Topics Signal Process., vol. 5, no. 1, pp. 5–23, Feb. 2011.

[11] S. Srinivasa and S. Jafar, “Cognitive radios for dynamic spectrum access -
the throughput potential of cognitive radio: A theoretical perspective,” IEEE
Commun. Mag., vol. 45, no. 5, pp. 73–79, May 2007.

[12] E. Hossain, D. Niyato, and Z. Han, Dynamic Spectrum Access and Manage-
ment in Cognitive Radio Networks. Cambridge University Press, 2009.

[13] M. Vu, S. Ghassemzadeh, and V. Tarokh, “Interference in a cognitive network

with beacon,” in Proc. IEEE WCNC 2008, Apr. 2008, pp. 876–881.

[14] M. Sherman, A. Mody, R. Martinez, C. Rodriguez, and R. Reddy, “IEEE stan-

dards supporting cognitive radio and networks, dynamic spectrum access, and
coexistence,” IEEE Commun. Magazine, vol. 46, no. 7, pp. 72–79, Jul. 2008.

[15] C. Stevenson, G. Chouinard, Z. Lei, W. Hu, S. Shellhammer, and W. Cald-

well, “IEEE 802.22: The first cognitive radio wireless regional area network
standard,” IEEE Commun. Magazine, vol. 47, no. 1, pp. 130–138, Jan. 2009.

[16] K. Moessner, H. Harada, C. Sun, Y. Alemseged, H. N. Tran, D. Noguet,

R. Sawai, and N. Sato, “Spectrum sensing for cognitive radio systems: techni-
cal aspects and standardization activities of the IEEE P1900.6 working group,”
IEEE Wireless Commun., vol. 18, no. 1, pp. 30–37, Feb. 2011.

[17] A. Damnjanovic, J. Montojo, Y. Wei, T. Ji, T. Luo, M. Vajapeyam, T. Yoo,

O. Song, and D. Malladi, “A survey on 3GPP heterogeneous networks,” IEEE
Wireless Commun., vol. 18, no. 3, pp. 10–21, 2011.

78
[18] R. Research. (2011) Efficient use of spectrum. [Online]. Available: http://
www.rysavy.com/Articles/2011 05 Rysavy Efficient Use Spectrum.pdf

[19] J. Hoydis, M. Kobayashi, and M. Debbah, “Green small-cell networks,” Ve-

hicular Technology Magazine, IEEE, vol. 6, no. 1, pp. 37–43, 2011.

[20] A. Goldsmith, Wireless Communications. Cambridge University Press, 2005.

[21] M. Simon and M. S. Alouini, Digital Communications over Fading Channels,

2nd edition. Wiley, 2004.

[22] M. Nakagami, “The m-distribution, a general formula for intensity distribution

of rapid fading,” Statistical Methods in RadioWave Propagation, pp. 3–36,
1960.

[23] I. Trigui, A. Laourine, S. Affes, and A. Stephenne, “Outage analysis of wire-

less systems over composite fading/shadowing channels with co-channel in-
terference,” in Proc. IEEE WCNC, Apr. 2009, pp. 1–6.

[24] P. Bithas, N. Sagias, P. Mathiopoulos, G. Karagiannidis, and A. Rontogiannis,

“On the performance analysis of digital communications over generalized-K
fading channels,” IEEE Commun. Lett., vol. 10, no. 5, pp. 353–355, May 2006.

[25] S. Atapattu, C. Tellambura, and H. Jiang, “A mixture gamma distribution to

model the SNR of wireless channels,” vol. 10, no. 12, pp. 4193 –4203, Dec.
2011.

[26] G. L. Stuber, Principles of Mobile Communication. Kluwer Academic Pub-

lishers, Norwell, MA, 2001.

[27] D. Lewinski, “Nonstationary probabilistic target and clutter scattering mod-

els,” IEEE J. Antennas and Propagation, vol. 31, no. 3, pp. 490–498, May
1983.

[28] I. Kostic, “Analytical approach to performance analysis for channel subject to

shadowing and fading,” IEEE Proc. Commun., vol. 152, no. 6, pp. 821–827,
Dec. 2005.

79
[29] S. Al-Ahmadi and H. Yanikomeroglu, “On the approximation of the
generalized-K distribution by a gamma distribution for modeling composite
fading channels,” IEEE Trans. Wireless Commun., vol. 9, no. 2, pp. 706–713,
Feb. 2010.

[30] Z. Chen, C.-X. Wang, X. Hong, J. Thompson, S. Vorobyov, X. Ge, H. Xiao,

and F. Zhao, “Aggregate interference modeling in cognitive radio networks
with power and contention control,” IEEE Trans. Commun., vol. 60, no. 2, pp.
456–468, Feb. 2012.

[31] N. Hoven and A. Sahai, “Power scaling for cognitive radio,” in Proc. IEEE
WNCMC, vol. 1, 2005, pp. 250–255 vol.1.

[32] J. F. Kingman, Poisson Processes. Oxford University Press, 1993.

[33] I. Janine, A. Penttinen, H. Stoyan, and D. Stoyan, Statistical Analysis

and Modelling of Spatial Point Patterns (Statistics in Practice). Wiley-
Interscience, 2008.

[34] A. Baddeley, I. Barany, R. Schneider, and W. Weil, Spatial Point Processes

and their Applications. Springer, 2007.

[35] K. Gulati, B. Evans, J. Andrews, and K. Tinsley, “Statistics of co-channel

interference in a field of Poisson and Poisson-Poisson clustered interferers,”
IEEE Trans. Signal Process., vol. 58, no. 12, pp. 6207–6222, Dec. 2010.

[36] E. Salbaroli and A. Zanella, “Interference analysis in a Poisson field of nodes

of finite area,” IEEE Trans. Veh. Technol., vol. 58, no. 4, pp. 1776–1783, May
2009.

[37] E. Sousa and J. Silvester, “Optimum transmission ranges in a direct-sequence

spread-spectrum multihop packet radio network,” IEEE J. Sel. Areas Com-
mun., vol. 8, no. 5, pp. 762–771, Jun. 1990.

80
[38] E. Sousa, “Performance of a spread spectrum packet radio network link in a
Poisson field of interferers,” IEEE Trans. Signal Process., vol. 38, no. 6, pp.
1743–1754, Nov. 1992.

[39] J. Ilow and D. Hatzinakos, “Analytic alpha-stable noise modeling in a Poisson

field of interferers or scatterers,” IEEE Trans. Signal Process., vol. 46, no. 6,
pp. 1601–1611, Jun. 1998.

[40] P. C. Pinto and M. Z. Win, “Communication in a Poisson field of interferers–

part I: Interference distribution and error probability,” IEEE Trans. Wireless
Commun., vol. 9, no. 7, pp. 2176–2186, Jul. 2010.

[41] C.-H. Lee and M. Haenggi, “Interference and outage in Poisson cognitive net-
works,” IEEE Trans. Wireless Commun., vol. 11, no. 4, pp. 1392–1401, Apr.
2012.

[42] R. Ganti and M. Haenggi, “Interference and outage in clustered wireless ad

hoc networks,” IEEE Trans. Info. Theory, vol. 55, no. 9, pp. 4067–4086, 2009.

[43] G. Alfano, M. Garetto, and E. Leonardi, “Capacity scaling of wireless net-

works with inhomogeneous node density: upper bounds,” IEEE J. Sel. Areas
Commun., vol. 27, no. 7, pp. 1147–1157, 2009.

[44] J. Chen, M. Ding, and Q. Zhang, “Interference statistics and performance anal-
ysis of mimo ad hoc networks in binomial fields,” IEEE Trans. Veh. Technol.,
vol. 61, no. 5, pp. 2033–2043, Jun. 2012.

[45] D. Stoyan, W. S. Kendall, and J. Mecke, Stochastic Geometry and Its Appli-
cations. Wiley, 1996.

[46] M. Aljuaid and H. Yanikomeroglu, “A cumulant-based characterization of the

aggregate interference power in wireless networks,” in Proc. IEEE VTC, Ot-
tawa, Canada, May 2010, pp. 1–5.

[47] M. Timmers, S. Pollin, A. Dejonghe, A. Bahai, L. Van der Perre, and

F. Catthoor, “Accumulative interference modeling for cognitive radios with

81
 distributed channel access,” in Proc. IEEE CrownCom, Singapore, May 2008,
pp. 1–7.

[48] S. SINGH, N. Mehta, A. Molisch, and A. Mukhopadhyay, “Moment-matched

lognormal modeling of uplink interference with power control and cell selec-
tion,” IEEE Trans. Wireless Commun., vol. 9, no. 3, pp. 932–938, Mar. 2010.

[49] L. Vijayandran, P. Dharmawansa, T. Ekman, and C. Tellambura, “Analysis of

aggregate interference and primary system performance in finite area cogni-
tive radio networks,” IEEE Trans. Commun., vol. PP, no. 99, pp. 1–12, 2012.

[50] A. Ghasemi and E. Sousa, “Interference aggregation in spectrum-sensing cog-

nitive wireless networks,” IEEE J. Sel. Areas Commun., vol. 2, no. 1, pp. 41–
56, Feb. 2008.

[51] T. Velivasaki, T. Zahariadis, P. Trakadas, and C. Capsalis, “Interference anal-

ysis of cognitive radio networks in a digital broadcasting spectrum environ-
ment,” in Proc. IEEE IWSSIP, Jun. 2009, pp. 1–5.

[52] A. Rabbachin, T. Q. S. Quek, H. Shin, and M. Z. Win, “Cognitive network

interference,” IEEE J. Sel. Areas Commun., vol. 29, no. 2, pp. 480–493, Feb.
2011.

[53] M. Aljuaid and H. Yanikomeroglu, “Investigating the gaussian convergence

of the distribution of the aggregate interference power in large wireless net-
works,” IEEE Trans. Veh. Technol., vol. 59, no. 9, pp. 4418–4424, Nov. 2010.

[54] M. Vu, N. Devroye, and V. Tarokh, “On the primary exclusive region of cogni-
tive networks,” IEEE Trans. Wireless Commun., vol. 8, no. 7, pp. 3380–3385,
Jul. 2009.

[55] A. Hasan and J. Andrews, “The guard zone in wireless ad hoc networks,”
IEEE Trans. Commun., vol. 6, no. 3, pp. 897–906, Mar. 2007.

82
[56] M. J. Rahman and X. Wang, “Probabilistic analysis of mutual interference in
cognitive radio communications,” in Proc. IEEE GLOBECOM, Dec. 2011, pp.
1–5.

[57] M. Derakhshani and T. Le-Ngoc, “Aggregate interference and capacity-outage

analysis in a cognitive radio network,” IEEE Trans. Veh. Technol., vol. 61,
no. 1, pp. 196–207, Jan. 2012.

[58] M. Hanif, M. Shafi, P. Smith, and P. Dmochowski, “Interference and deploy-

ment issues for cognitive radio systems in shadowing environments,” in Proc.
IEEE ICC, Jun. 2009, pp. 1–6.

[59] A. Babaei and B. Jabbari, “Interference modeling and avoidance in spectrum

underlay cognitive wireless networks,” in Proc. IEEE ICC, May 2010, pp.
1–5.

[60] K. Woyach, P. Grover, and A. Sahai, “Near vs. far field: Interference aggrega-
tion in tv whitespaces,” in Proc. IEEE GLOBECOM, Dec. 2011, pp. 1–5.

[61] X. Hong, C.-X. Wang, and J. Thompson, “Interference modeling of cognitive

radio networks,” in Proc. IEEE VTC, May 2008, pp. 1851–1855.

[62] P. G. Moschopoulos, “The distribution of the sum of independent gamma ran-

dom variables,” Ann. Inst. Statist. Math., vol. 37, pp. 541–544, 1985.

[63] D. Wackerly and W. Scheaffer, Mathematical Statistics with Applications.

Thomson Higher Education, 2008.

[64] V. Mordachev and S. Loyka, “On node density - outage probability tradeoff
in wireless networks,” IEEE J. Sel. Areas Commun., vol. 27, no. 7, pp. 1120–
1131, 2009.

[65] Y. Wen, S. Loyka, and A. Yongacoglu, “The impact of fading on the outage
probability in cognitive radio networks,” in Proc. IEEE VTC 2010-Fall, Sep.
2010, pp. 1–5.

83
[66] M. Hoyhtya and A. Mammela, “Adaptive inverse power control using an
FxLMS algorithm,” in Proc. IEEE VTC, Apr. 2007, pp. 3021–3025.

84