STJOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS

1
INTRODUCTION P. 3
ART. 1
Wireless Sensor Networks devices: overview, issues, state-of-the-art and promising technologies
by Laurent Chalard, Didier Helal, Lucille Verbaere, Armin Wellig, Julien Zory (STMicroelectronics). P. 4
ART. 2
Ad Hoc Wireless Sensor Networking: Challenges and Issues
by Danilo Blasi, Vincenzo Cacace, Luca Casone, Marco Rizzello, Salvatore Rotolo (STMicroelectronics);
Luciano Bononi (Department of Computer Science, University of Bologna, Italy). P. 19
ART. 3
Pico Radio: From Vision To Reality
by M. Sheets, B. Otis, H. Qin, N. Pletcher, F. Burghardt, J. Ammer, T. Karalar, P. Monat, Y. Cao, D. Markovic,
A. Vladimirescu, J. Rabaey (Berkeley Wireless Research Center, University of California, Berkeley);
S. Cervini (STMicroelectronics). P. 33
ART. 4
Cupola-Sensing: an Indoor Climate Monitoring Test Case
by Max Cortiana, Andrea Labombarda, Laura Vanzago (STMicroelectronics). P. 52
ART. 5
ZigBee The JourneyToward Mass Market Adoption
by Gilles Thonet, Marc Bruel (Schneider Electric). P. 65
ART. 6
Localization in Sensor Networks
by K. Yao (University of California, Los Angeles);
F. Lorenzelli (STMicroelectronics). P. 80
ART. 7
Inertial Sensors for Wireless Body Area Networks: The WiMoCA Solution
by Elisabetta Farella, Augusto Pieracci, Luca Benini (DEIS - University of Bologna, Italy);
Andrea Acquaviva (ISTI - Urbino University, Italy). P. 97
ART. 8
Managing Impulsive Interference in Impulse Radio UWB Networks
by Manuel Flury, Ruben Merz, Jean-Yves Le Boudec (EPFL, School of Computer and Communication Sciences). P. 118
ART. 9
The Aloha access (UWB)
2
protocol revisited for IEEE 802.15.4a
by Maria-Gabriella Di Benedetto, Luca De Nardis, Guerino Giancola,
Daniele Domenicali (School of Engineering University of Rome La Sapienza). P. 131
ART. 10
An ultra-low energy asynchronous processor for Wireless Sensor Networks
by L. Necchi, L. Lavagno (Politecnico di Torino);
D. Pandini, L. Vanzago (STMicroelectronics). P. 142
ART. 11
Implementing ECC for 8-bit Systems and Power Consumption Considerations
by Guido Bertoni (STMicroelectronics); Luca Breveglieri, Matteo Venturi (Politecnico di Milano). P. 154
2 COPYRIGHT STMICROELECTRONICS, 2007
STJOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
ST JOURNAL OF RESEARCH
EDITOR IN CHIEF
Andrea Cuomo
MANAGING EDITORS
Osvaldo M. Colavin
Flavio Lorenzelli
Antoine Hue
SCIENTIFIC COMMITTEE
Giulio Casagrande
Bruno Murari
Robert Bob Kryziak
Guy Lauvergeon
Jefferson Owen
EDITORIAL COMMITTEE
Fabrizio Rovati
Benedetto Vigna
Friedbert Berens
Enrica Filippi
Erven Rohou
ASSOCIATE EDITOR
Clara Colombo
st.journal@st.com
LEGAL INFORMATION
DIRETTORE RESPONSABILE
Maria Teresa Gatti
REDAZIONE
STMicroelectronics Srl
Via C. Olivetti 2
20041 Agrate Brianza (Milano) - Italy
LAYOUT AND PRODUCTION
La Carta Stampata
Milano, Italy
PHOTOS
STMicroelectronics - Zerodue
Milano, Italy
STAMPA
C&M Print Sas
Pioltello (Milano), Italy, via Roma 4
Iscrizione presso il Tribunale di Monza
Nr. 1650/2003
EDITOR IN CHIEF
Andrea Cuomo
Executive Vice President, Chief Strategy
and Technology Officer, STMicroelectronics
ADVISORY BOARD
Igor Alexander -
Imperial College of Science Technology and Medicine
Giovanni De Micheli - Stanford University
Joseph A.Fisher - Hewlett Packard
Nick McKeown - Stanford University
Alberto Sangiovanni Vincentelli - Berkeley University
Robert M.White - Carnegie Mellon University
MANAGING EDITORS
Osvaldo Colavin
Flavio Lorenzelli
Antoine Hue
GUEST EDITORS
Laura Vanzago
Julien Zory
ASSOCIATE EDITOR
Clara Colombo
Volume 4, Number 1
Wireless Sensor Networks
STJOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
3
Ambient Intelligence is a vision where environment
becomes smart, friendly, context-aware and responsive to any
type of human needs. In this world, computing and networking
technology coexist with people in a ubiquitous and pervasive
way. In this world, numerous miniature and interconnected
smart devices create a new intelligence and interact with each
other seamlessly.
Wireless Sensor Networks (WSNs) are one of the first real
world examples enabling that vision of Ambient Intelligence.
Within a WSN, sensory data originate from multiple sensors
of different types spread over potentially wide areas. Virtually
any kind of physical quantities such as temperature, voltage,
pressure, acceleration, ultrasound, gas flow, etc. can be
acquired. Likewise, actuators can be used as a means to
influence or control the environment. For coordination and
information exchanges, the various nodes communicate over
a wireless mesh network in a self-organized manner.
Originally motivated by military applications such as battle-
field surveillance, the applications of WSNs are many and
varied. Industrial and home automation benefit from the
capability to wirelessly monitor data that would be difficult or
too expensive to monitor using wired sensors. Likewise,
health care or environmental control applications take advan-
tage of the miniaturization and self-organized nature of
WSNs. The many potential applications recently motivated
major players of the semiconductor industry to invest in this
promising field. STMicroelectronics has been among the first
companies to enter the WSN market with the introduction of
sensors, low-power microcontrollers, wireless radio devices,
etc., in its portfolio. STMicroelectronics is a promoter of the
ZigbeeTM alliance, an industry consortium working on the
definition of network, security and application layers towards
truly interoperable WSNs.
One major technical challenge of WSNs is to produce low-
cost, low-power and tiny sensor and actuator nodes.
Research work is thus needed and involves numerous fields
such as Radio Frequency communication, ad-hoc networks,
low-power microcontrollers, transduction principles, energy
management, middleware, etc. Another key aspect is the
coexistence and interoperability of WSNs with existing tech-
nologies such as the Internet, broadband wireless networks
or database management systems. Given this wide list of
topics, this issue of the ST Journal of Research includes sever-
al overview papers that help the reader capture the most
important challenges as well as a selection of contributions
covering specific design topics. Eleven papers have been
selected from worldwide recognized experts and ST
researchers in the field.
The first section of this special issue includes three tutorial
papers. A general introduction to the problems and chal-
lenges of developing Wireless Sensor Network devices is
given, in the first paper by Chalard et al. from
STMicroelectronics. A particular emphasis on ad-hoc net-
working technologies is proposed in the second paper by
Blasi et al. from STMicroelectronics and Bononi from
Universita di Bologna. Sheets et al. from the Univeristy of
California, Berkeley and Cervini from STMicroelectronics
describe in the third paper the main technical challenges
faced in the PicoRadio Project carried out at the Berkeley
Wireless Research Center.
Four contributions describing the application-specific nature
of WSN optimization are included in the second part of the
issue. An indoor climate monitoring prototype developed by
our STMicroelectronics colleagues illustrate the challenges of
integrating WSNs with the Internet via gateway and data-
base client-server tools. Thonet and Bruel from Schneider
Electric discuss the mass market adoption of ZigbeeTM solu-
tions in the segments of Building Automation, Home control,
Automated Meter Reading and Industrial Automation. The
third paper, authored by Yao from UCLA and Lorenzelli from
STMicroelectronics, introduces different techniques for source
and node localization. In the last paper of this section,
Farella et al. from University di Bologna and Urbino expose
the reader to the application of WSN and MEMS sensors in
the context of body area networks, more specifically posture
and activity recognition.
As mentioned above, a WSN device is a complex integrat-
ed system implementing a combination of functions such as
communication, security and signal or application process-
ing. The last part of the issue includes four papers dedicated
to specific design experiences carried out in those areas.
Two papers explore the promising use of low data rate
UltraWideBand radio communications in the context of
WSN. Specifically, Flury et al. from EPFL analyze the impact
of impulsive interference on network throughput and energy
consumption. Di Benedetto et al. from University La Sapienza
focus their work on medium access control schemes pro-
posed for the IEEE 802.15.4a standard. In the third paper
Necchi and Lavagno from Politecnico di Torino together with
Pandini and Vanzago from STMicroelectronics present their
work in trading off power consumption and die area for the
design of an asynchronous 8-bit processor. Finally, in the last
paper, performances of two Elliptic Curve Cryptography
coprocessors are explored by Bertoni from STMicroelectronics
together with Breveglieri and Venturi from Politecnico di
Milano. Wireless Sensor Networks have the potential to
open a new era. However, as illustrated by the broad range
of topics covered by this special issue, technical issues still
exist in the production of low-cost small devices operating for
years in a self-organized way and seamlessly integrated with
Internet access and appropriate human interfaces. We
believe that only companies having the capability to master
this broad spectrum of activities and to succeed in both
more Moore and more-than-Moore directions will provide
winning solutions for Ambient Intelligence.
The Guest Editors would like to thank all the Authors as well
as all those involved in the review and editorial process, with
a special mention to Antoine Hue for his huge contribution.
Julien Zory & Laura Vanzago
Guest Editors
Wireless Sensor Networks: a journey towards Ambient Intelligence
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
4 COPYRIGHT STMICROELECTRONICS, 2007.
Laurent Chalard, Didier Helal,
Lucille Verbaere, Armin Wellig, Julien Zory
STMicroelectronics
WIRELESS SENSOR
NETWORKS DEVICES:
OVERVIEW, ISSUES,
STATE-OF-THE-ART AND
PROMISING TECHNOLOGIES
Revolutionary networking concepts
and unprecedented mix of technical
challenges have made Wireless Sensor
Networks (WSN) one of the major
research trends of the 21th century.
However, only very recently did such
systems start to appear as products
on the market. In this new context, this
paper aims at providing a general
status on WSNs, with a special focus
on the device aspects, independent
of application peculiarities. We first
investigate why, despite years of
research and development and technical
maturity, WSN products and solutions
are yet neither fully adopted nor widely
deployed. We then review the state-of-the-
art technologies that build most of todays
available products and finally briefly
discuss some advanced techniques for
future generations of WSN systems.
1. INTRODUCTION
Wireless Sensor Network is a generic term for a system made
of several autonomous sensors and/or actuators, called nodes,
capable of exchanging information with others by means of
a dedicated wireless network. Beyond this basic definition,
additional specific characteristics are generally assumed de
facto. Among others, nodes are expected to be simple and low
cost, which implies, for instance, favoring their redundancy
rather than their performance; furthermore, they shall be low
power, which implies limited communication range and hence
use of multi-hop connections; additionally, they are assumed to
be autonomous with underlying features like self-configuration
and self-healing of the network.
This common understanding of WSN probably originates from
its initial research context, supported among others by numerous
funded projects in the US (e.g., SmartDust, NEST). The vision
was then a kind of self-organized, homogeneous, tiny, multi-
hop, and resource-constrained set of devices. Applications were
initially mostly military and slowly started to encompass other
monitoring fields, like agriculture or species tracking. More
recently, several technological advances made such systems a
promising alternative for numerous commercial applications as
well: home automation, healthcare, industrial plant monitoring,
WIRELESS SENSOR NETWORKS DEVICES: OVERVIEW, ISSUES, STATE-OF-THE-ART AND PROMISING TECHNOLOGIES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
5
or asset tracking, to name a few (see Fig. 1). Such potential mass
production markets put great expectations on WSN.
One of the most famous initiatives consolidating the possible
deployment of WSN systems was the IEEE802.15.4 Task
Group (TG), which specified in 2003 a physical (PHY) and a
Medium Access Control (MAC) layer dedicated to low data rate
communication devices [1]. The main motivation to develop a
dedicated standard, and not to rely on existing technologies like
Bluetooth or WLAN, was to ensure low complexity energy-
efficient implementations so as to enable multi-year battery
lifetime and, hence, to open the door to new application fields
(e.g., hard-to-access or non-accessible nodes) and to lower the
maintenance costs. Most of todays commercially available radio
solutions use the 2.4GHz unlicensed worldwide ISM band, whose
regulation allows for 1mW of transmitted power, thus spanning a
range between 30 to 100m as a function of the environment. Its
packet-based processing allows multiple users via Carrier Sense
Multiple Access and Collision Avoidance (CSMA/CA), and its
flexibility accommodates various application requirements (e.g.,
latency, number of devices, network topology, encryption, and
duty cycle). Future releases are already under development,
mainly via the IEEE802.15.4a TG, which focuses on an Ultra
Wide Band (UWB) alternative with location capabilities and the
IEEE802.15.4b TG, which enhances the current Narrow Band
(NB) standard.
In order to consolidate deployment of low data rate systems based
on IEEE802.15.4, as well as to speed up their development, the
Zigbee Alliance specified in 2004 a generic protocol stack that
provides all networking features up to the application interface
[2]. It supports various pre-defined application profiles, e.g.,
home automation, in order to allow designers to focus on the
application while lower layers are configured accordingly. Note
that even if the Zigbee Alliance is an initiative of several industry
leaders, the specification itself is open and unlicensed. Still, it
has not initially been developed for WSN specifically, and
numerous proprietary solutions claim to better fit specific needs;
consequently, there is a hard push to get shares of the potentially
huge WSN market.
Today, first generation WSN systems are commercially
available, based on either standard or proprietary technologies.
End products range from home automation (Control4, Eatons
FIGURE 1: POSSIBLE APPLICATION FIELDS OF
WIRELESS SENSOR NETWORKS
Building & Home
Automation
Energy management
HVAC, AMR
Lighting control
...
Industrial Control
Monitoring
& diagnosis
Security
Automation
...
Consumer
Automation
Remote control
Toys & games
Education
...
Healthcare
Monitoring & sensing
Alarm
Tracking
...
Security
Alarm systems
Monitoring & sensing
Access control & lock
...
Others...
Environment
monitoring
Structural monitoring
Agriculture monitoring
Species monitoring
...
FIGURE 1: POSSIBLE APPLICATION
FIELDS OF WIRELESS SENSOR
NETWORKS.
6 COPYRIGHT STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
HomeHeartBeat, Kalirels Cyclope), to security (RayMarines
LifeTag), industrial control (Siemens Apogee), automatic
meter reading (Wellspring) and others. The list increases
every day. Application designers can choose among various
technologies, hardware devices, software, tools, and all-
in-one solutions (STs SN250, TIs CC2430). In 2005, the
Zigbee chipset market had already registered 2.5 million
unit shipments and generated revenues of $11.3 million [3].
Moreover, an exponential growth is expected for the coming
years, as will be further discussed.
The remainder of this article is structured as follows. Section 2
addresses the ambiguous situation of WSN today by investigating
what key issues prevent or simply delay its true market ramp-up.
Section 3 then provides an insight of the WSN state-of-the-art
technology for some of the core hardware and software elements.
Section 4 focuses on some of the research fields expected to drive
tomorrows WSN technology. Finally, Section 5 summarizes the
key highlights of the paper and provides concluding remarks.
2. GREAT EXPECTATIONS,
YET MANY QUESTION MARKS
Today, the WSN market as a whole is rather confused. Despite
the already wide offerings in low data rate products, there are
many concerns that have prevented a massive adoption by the
market up to now. Although only time will tell what is going
to happen, it is worth highlighting the most important issues of
today. These are addressed below.
2.1 Market uncertainties
Pioneer research work on modern WSNs started back in the 80s
under the DARPA umbrella; the Zigbee alliance was formed
in October 2002 and the IEEE802.15.4 standard published in
October 2003. However, the time between technology concept
and real market adoption is always longer than expected. Recent
examples include Bluetooth or UMTS, and this will apply as
well to WSN in general and to Zigbee in particular. After a few
years of strong design activity, mostly by small companies, first
generation hardware and software solutions are being shipped,
and big silicon manufacturers recently have become very active
as well. In this ramp-up phase, home automation has the biggest
potential; however, consumer demands for low cost and plug-
and-play solutions might delay the true mass-market volume.
In the meantime, industrial applications lead the path to market
adoption. Once high volumes and technology maturity are
achieved, many other application segments will benefit from
WSN. Several market analysts report that the overall WSN
market could hit the 1 billion dollar target as soon as 2009-2010.
The setup of standard solutions will definitely help in reducing
device costs and creating true interoperability. However, up
to now, most WSN solutions on the market were built upon
proprietary solutions, and it is still unclear whether Zigbee
will succeed in all market segments or not. Some of Zigbees
competitors are trying to promote simpler solutions tailored to
specific market segments (e.g., Insteon in Home Automation)
or building upon other technological bricks (e.g., Zwave from
Zensys).
2.2 Spectrum and regulations
Today, most WSN solutions use NB modulations and operate
in license-free bands. The 433MHz band is available worldwide
(with minor modifications in Japan) with the major restriction
that applications such as audio or continuous data transmission
are typically not allowed in this band. The 868MHz (in Europe)
and 915MHz (in USA) bands are not application restrictive and
allow for low cost implementation but suffer from not being
worldwide. It is thus no surprise that most Zigbee vendors today
select the 2.4GHz band for worldwide operation. However, the
use of several air interfaces operating in different frequency
bands potentially limited to specific countries only adds up
to the WSN market confusion. Regulation for UWB is under
discussion.
Moreover, with a maximum of 80MHz of allocated bandwidth
(in the 2.4GHz NB band) to be used simultaneously by multiple
WIRELESS SENSOR NETWORKS DEVICES: OVERVIEW, ISSUES, STATE-OF-THE-ART AND PROMISING TECHNOLOGIES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
7
WSNs (e.g., HVAC and alarm system) and shared with other
types of devices operating in this band (Bluetooth, Wifi, micro-
wave ovens, etc.), the issue of coexistence is foreseen as one
major threat to WSN mass-market adoption.
2.3 Reliability
The Zigbee Alliance slogan is Wireless Control that Simply
Works. Reliability is probably the number one requirement
for adoption by the mass-market. Before being low cost, low
power, self-organized, or super-smart, wireless sensor networks
should simply work. Major progress has already been made
during the past years on this aspect, but quite a few challenges
might still strongly affect the image of (standard or proprietary)
WSN solutions. Transmit power variations upon manufacturing
and temperature conditions, sensitivity to Electro Magnetic
Interference, or limited battery lifetimes are typical factors
which may affect the node itself; however, those can be easily
anticipated. More troublesome are interference, shadowing,
or multi-path fading effects which affect the wireless link
reliability; those are fairly hard to systematically cope with as
they highly dependent upon channel conditions. Even more
application specific, hence difficult to anticipate, are all the
reliability problems related to the wireless network configuration
where orphanage procedures, route discoveries, or security
checks are needed.
2.4 Ease of use
Some equipment manufacturers may choose to purchase System-
on-chip (SoC) solutions, select the right antenna, and dig into
board design and complex programming tasks. However, for those
application developers who still are not familiar with wireless
connectivity, complete packaged modules start to be available on
the market. The work effort is then limited to supplying energy
(battery), connecting a few sensors, and programming the
application software, potentially on an external microcontroller.
Unfortunately, the end-user experience might be affected as well
by the wireless connectivity. It is nice to get rid of the cables as
long as all this remains transparent to the user. Nodes should be
easy to deploy, wireless links self-configured, and networks self-
healing. Likewise, upgrades and remote maintenance should be
enabled by the appropriate interfaces and tools.
2.5 Application requirements
How could pressure monitoring of pipelines in the desert and
tracking of precious medical equipment in a hospital share the
same requirements? The answer is probably that they cannot.
The WSN design space is very large and linked with constraints
from the device (size, cost, power, etc.), the air interface
(data rate, range, robustness, etc.), the network configuration
(topology, scalability, ad-hoc vs infrastructure, etc.), and even the
application features (localization, QoS, mobility, latency, etc.). In
the absence of a single killer application, next generation WSN
solutions will have to offer the appropriate tradeoff between
flexibility and efficiency. It is expected that current solutions
will evolve incrementally to optimize existing applications
(sensitivity, security, etc.) while disruptive changes will be
needed to enable new ones (energy scavenging, localization, etc.).
3. KEY COMPONENTS
SURVEY & STATE-OF-THE-ART
As illustrated on the left side of Fig. 2, any WSN mote is a
complex device that embeds several dedicated hardware and
software components, irrespective of the application. Several
motes are then combined to build either simple (e.g., star, tree)
or very complex ad-hoc mesh networks, which can potentially be
bridged to other types of information networks such as internet
(right side of Fig. 2).
The purpose of the following section is to briefly describe the
most relevant components and to comment on the status of their
key features and performance so as to get a better picture of
where WSN technology stands today.
3.1 Sensors & actuators
A sensor is a physical device that probes physical, biological,
or chemical properties of its environment and converts these
8 COPYRIGHT STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
IEEE1451 is pursuing an effort to develop a family of open,
network-independent communication interfaces for connecting
transducers to microprocessors, instrumentation systems, and
control/field networks [5]. It particularly defines the concept
of Transducer Electronic Data Sheets (TEDS), a memory
device attached to the transducer, which stores transducer
identification, calibration, correction data, measurement range,
and manufacturer-related information, among others. This
allows connecting virtually any Smart Transducer Interface
Module to any Network Capable Application Processor in a
Plug-and-Play manner.
3.2 Transceivers
Each node of a WSN system comprises a transceiver unit, which
is in charge of the wireless communication with peers. This
dedicated hardware usually supports the complete PHY layer,
as well as some time-critical operations of the MAC layer which
would not be efficiently performed in software. Regarding the
MAC, mechanisms such as CSMA fit WSN constraints well
because neither central coordination nor synchronization
between nodes is required; nevertheless, time synchronization is
a hot topic, for it allows the effective implementation of power
duty cycle solutions. Conversely, applications with very low
traffic may actually rely on a simpler Aloha medium access
mechanism.
properties into an electrical signal. Sensors for temperature,
light, oxygen, distance, blood pressure, moisture, and torque are
some of the many examples. An actuator typically accepts an
electrical signal and converts it into a physical action to act upon
the environment. Sensors and actuators belong to the broader
family of transducers.
Classical transducers such as temperature or pressure sensors
are available as off-the-shelf components and can be easily
integrated at the board or package level. More complex ones like
CMOS image sensors, inertial sensors, or micro-fluidic actuators
have recently emerged, thanks to technological advances. Those
smart sensors typically require dedicated logic for calibration,
signal processing, or analog-to-digital conversion and sometimes
include a micro-controller. The LM135 precision thermal sensor,
TDA0161 proximity detector, and LIS3LV02DQ three-axis
digital accelerometer from STMicroelectronics are typical
examples of sensor technology [4]. As recently demonstrated
with the In-Check platform (a biochip for amplifying tiny DNA
samples), it is expected that ongoing efforts on CMOS and Micro
Electro-Mechanical System (MEMS) technologies will further
enlarge transducer potential in the near future.
Transducer manufacturers serve a vast variety of applications
using diverse physical interfaces and control protocols. The
Sensor
Energy scavenging Energy Storage Server Clients
Power management
Transit Network
(TCP-IP Internet)
ADC
Sensor DIO
Actuator
Memory
Peripherals
microcontroller
application
networking
middleware
BB+MAC
+RF
transceiver
Interface
DAC
IP Gateway
FIGURE 2:
WSN MOTE
(LEFT) AND
WSN
COMPLETE
INFORMATION
PROCESSING
SOLUTION
(RIGHT).
WIRELESS SENSOR NETWORKS DEVICES: OVERVIEW, ISSUES, STATE-OF-THE-ART AND PROMISING TECHNOLOGIES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
9
PHY layers for WSN systems are typically based on Radio
Frequency (RF) technologies. Although the preference for
unlicensed bands has already been discussed for NB systems, an
additional advantage of such frequencies is that the corresponding
technology is mature, hence lower in cost. Moreover, transmitted
power and typical receiver sensitivity, around 1mW and
-100dBm respectively, fit well for indoor operation. Typical
NB modulations are Phase Shift Keying enhanced with Direct
Sequence Spread Spectrum to increase link reliability and ease
packet detection and synchronization. Data rates over-the-air
range up to tens of kbit/s for NB systems.
The communication device usually deserves careful design
because its power consumption is often far from negligible.
Currently, IEEE802.15.4-compliant transceivers at 2.4GHz
typically require 20-25mA in both transmit (at 0dBm) and
receive mode, at 1.8V. Future generations shall get down to
figures of 10-15mA [6]. Initial UWB devices are expected
to be more demanding, but this could be compensated for by
reduced duty cycles due to the higher peak rates over the air.
In the longer run, advanced research targets even lower figures.
For NB alternatives, a 2.4GHz transceiver shall require less than
1mA in both modes [7]. Research on UWB foresees down to
0.4mW for the transmitter and 0.6mW for the receiver at 1Mbit/s
[8]. Hence, probability is high for future WSN transceivers to
benefit from a significant decrease in power consumption as well
as size and cost.
3.3 Networking
Networking is a huge research topic in WSN; thus, it would be
quite ambitious to provide an exhaustive state-of-the-art discussion
in a few lines. Nevertheless, because it encompasses key features
of any WSN solution, it at least deserves a quick overview.
In its simplest form, a WSN may rely on a central base-station
collecting sensory data from all other devices in a synchronized
and (infra)structured way. Infrastructure-based systems typically
enable very efficient usage of the available bandwidth, though
with high complexity or limited scalability (number of nodes
and covered area). A more adequate approach to low cost and
low power wireless networks consists of implementing an
infrastructure-less network, also known as ad-hoc network, where
heterogeneous devices communicate in a self-organized manner.
Mesh topologies, self-organization, and multi-hopping are
commonly assumed in WSN. Compared to wired networks,
wireless networks benefit from the broadcast nature of wireless
communications but suffer from potentially severe and irregular
channel conditions. The Zigbee consortium has defined what
could prove to be the first large scale implementation of WSN,
although alternatives like Z-Wave or Insteon products exist. For
all those technologies, the networking layer is responsible for
many functions such as network creation procedures, address
assignment, packet routing, security, application management,
etc. Advanced features such as self-healing capability, energy
aware routing, or mobility support are being explored by the
research community and will be progressively introduced in
commercial products.
The networking functionality is commonly implemented all in
software and shall run on low-end microcontrollers. The footprint
of a Zigbee stack is in the range of 64kbytes of program memory
and requires a few kbytes of RAM. Much simpler network
protocols may be implemented for those applications which
require only basic features (fixed addressing, star topology,
etc.), albeit at the cost of reduced flexibility and interoperability.
3.4 Microcontroller
As discussed in the previous sections, increased networking
capabilities, data logging features, or self-healing capabilities of
WSN motivate flexible solutions. A microcontroller (MCU) is a
self-sufficient and cost-effective computer-on-a-chip used to
control the functionality and data flow of (embedded) electronic
devices and to store/process data. On top of the processing core
- a Central Processing Unit (CPU) sometimes enhanced with
Digital Signal Processing (DSP) hardware - todays MCU cells
include customized memory sub-systems, non-volatile memory
10 COPYRIGHT STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
(embedded Flash), and interfaces such as ADC, UART, SPI,
counters, and timers. In this way, it can interact with embedded
sensors and communication devices to compose a fully-
integrated sensor node. There are many types of microcontrollers
with respect to the micro-architecture (CISC versus RISC),
instruction set bit width (i.e., Atmels ATmega165P 8-bit [9],
TIs MSP430 16-bit [10], or ARMs Cortex-M3 32-bit [11]),
different low-power operating modes, and peripheral diversity
or energy-efficiency (Watt/MIPS) characteristics. Commercially
available state-of-the-art MCUs report power consumption
figures on the order of 250A/MIPS in active mode, 1A in
real-time clock (or idle) mode, and 0.1A during RAM retention
(standby mode). It should be noted that the exact figures depend
on the actual application, number, and type of active peripherals
and operating temperature. The ideal choice of the MCU is the
one that matches its performance level with application needs.
3.5 Energy management
In WSN, sophisticated energy management techniques are
essential since WSN devices (running on a single battery) are
expected to last for several years. Energy-aware computing
includes ultra-low power microcontrollers (see Section 3.4) and
Dynamic Power Management (DPM, [12]) hardware to provide
requested performance and service levels while minimizing
power consumption (by turning off inactive sensors/peripherals
and parts of the IC). Among the key characteristics of energy-
aware software (e.g., TinyOS) are power-aware scheduling,
exploitation of concurrency, and the support for (high-level)
programming routines to efficiently control the low power
operating modes of MCUs. Finally, power management
techniques of the radio transceiver include wake-up mechanisms
(e.g., based on carrier sensing) in conjunction with advanced
MAC protocols (e.g., WiseMAC [13]).
Voltage regulators and DC/DC converters are key building
blocks in the implementation of dynamic power management
techniques. While the voltage regulator provides a constant DC
voltage regardless of changes in the load current, the DC/DC
converters match the voltage level of the variable power supply to
the individual voltage levels required by the different sub-circuits
(so-called voltage islands). Whereas todays commercial
WSN products rely solely on battery power, energy-harvesting
techniques coupled with secondary (rechargeable) batteries may
soon replace conventional solutions in the WSN market [14].
As a case study, a WSN mote - composed of a MCU, embedded
memory, sensor(s), and a 2.4GHz IEEE802.15.4 radio
transceiver - is reported in [15] and is compatible with energy
harvesting techniques such as piezoelectric materials or solar
cells. For very low-duty cycle sensing applications ( 10Hz), the
power consumption was reduced from several tens of mWatts
below one mWatt by implementing DPM, where both the radio
and sensors sleep between the sensors sampling periods.
Note that, in this particular hardware configuration, the radio
transceiver is the biggest power drainer with ~45mWatts in a
continuous radio link setup. The power consumption was greatly
reduced (to ~5mWatts in continuous mode) by processing and/or
logging data, rather than streaming it over the air. Thus, there
is an important local computation/storage versus radio
transmission architectural trade-off to be optimized as a
function of the application profile.
3.6 Others
On top of the aforementioned components, a WSN mote includes
many others which serve the application purpose; interfaces of
various kinds (Analog-Digital Converter, Universal Synchronous-
Asynchronous Receiver/Transmitter, Serial Peripheral
Interface, etc.), general purpose Inputs/Outputs, support for
external crystal oscillator or integrated RC oscillator, peripherals
(timers, encryption hardware acceleration engine, etc.),
middleware (operating system, drivers, etc.), and memory
storage for program and data are typical.
Then, beyond the single mote, a WSN complete solution is a
very complex integrated system where sensory data collected
throughout a network of motes can be made accessible to the
WIRELESS SENSOR NETWORKS DEVICES: OVERVIEW, ISSUES, STATE-OF-THE-ART AND PROMISING TECHNOLOGIES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
11
end-user by different means. In its simplest form, data might be
gathered in a so-called sink mote device. Bridges to other kinds
of information networks are, however, very common. Gateways
can, for instance, be developed to create legacy industrial
networks, Ethernet TCP-IP, or powerline backbones. On the
other side of this transit network, sensory data management is
handled by traditional database client-server infrastructures.
4. FUTURE WSN TECHNOLOGIES
Addressing the state-of-the-art of current technologies allows
picturing where WSN systems stand today, but a closer look at
what may build next generations deserves some discussion as
well. Many trends are being investigated by STMicroelectronics,
some of which are related to communication aspects (transceivers
and networking) and are presented in the following section.
Specifically, improvements of current narrowband systems,
UWB technology including a dedicated prototype, and position-
enabled sensor networks are addressed, followed by a glimpse at
even longer term disruptive technologies.
4.1 Improved Narrowband systems
As previously discussed, due to gaps in technology maturity, first
generation WSN systems shall remain NB-based for some time.
While disruptive approaches like UWB could take the lead some
day, it is still worth investigating NB evolutions as well because
these are expected to be implemented and become profitable in
a shorter time.
Even if there are many different NB alternatives, some of their
weaknesses are common independently from their specific
characteristics. Among such key limitations, co-existence in
unlicensed ISM bands has already been pointed out. Another
well-known weakness of NB systems is their sensitivity
to multipath fading, which is particularly critical in indoor
environments. Given their typical carrier frequencies and
bandwidths, NB receivers suffer flat fading, which can result
in more than 30dB of signal attenuation [16]. The statistics
of such fades mostly depend on mobility, and static scenarios
might appear as the worst cases, since links suffering deep fades
would then remain weak, if not broken, forever. To overcome
such effects, a solution well adapted to ad-hoc networks
consists of using multi-hop capabilities by clever and dynamic
routing, depending on the reliability of each link of the network;
obviously, such an alternative is expensive but nevertheless
appealing since many applications may require similar features
for other purposes as well (e.g., routing).
Another way to deal with multipath fading consists of exploiting
space diversity by means of multiple antennas at the transmitter
and/or the receiver side. Lets for instance assume 2 antennas
per receiver. Provided they are at a distance of at least a fraction
of the wavelength apart, the correlation between both received
signals is usually low enough to justify some kind of selection
or combination [17]. Depending on the application and the
environment, various alternatives allow then a tradeoff of
performance for complexity. The simplest one consists of a
straightforward combination of the antenna signals, which is
rarely optimal from a performance perspective, but does not
require more processing or even energy than used by a single
antenna receiver. A more complex scheme consists of selection
diversity, which means choosing the best antenna at any time.
In a general WSN context, such a selection is not trivial because
the typical information exchange duration is too short to allow
reliable real-time evaluation of each alternative. However, such a
scheme may increase link reliability in static scenarios where a
dedicated setup phase is used for measurements, while selection
diversity is used afterwards with the assumption that the
channel conditions do not change in-between. A more efficient
use of space diversity consists of combining both signals
but in a weighted way so as to maximize the signal-to-noise
radio (SNR). This usually requires demodulating each signal
independently to get the weighting coefficients and only then
to combine them accordingly. The price to pay in complexity
and power consumption is high, given that two receivers are
necessary down to the combination process. Nevertheless, sub-
optimal alternatives that lower complexity at the price of some
12 COPYRIGHT STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
performance decrease do exist; for instance, [18] suggests using
a single receiver to demodulate both signals while allowing
their sub-optimal separation by artificially delaying one of them
before combination.
4.2 UWB technology
A plurality of UltraWideBand (UWB) systems is possible, each
with relative advantages and disadvantages. While an OFDM-
based modulation is adopted for high data rate applications [19],
an impulse-based radio signaling scheme, as illustrated in Fig. 3
(right side), is adopted for Low Data Rate (LDR) WSN systems.
More specifically, the MAC and PHY layers are defined within
the (on-going) IEEE802.15.4a standardization framework [20],
which proposes 3 device options operating respectively in
the DC-966MHz, 2.4GHz ISM, and 3.1-10.6GHz frequency
ranges. The UWB preamble is composed of single (ternary-)
coded pulses, whereas the payload data is modulated using a
combination of Burst Position Modulation (BPM) and Binary
Phase Shift Keying (BPSK), with each symbol being composed
of an active burst of UWB pulses.
in very good interference rejection and thus is the key enabler
for network co-existence. Another key feature is the fine time
resolution (< 1 nanosecond) proper to UWB signaling, resulting
in accurate Time-of-Flight (ToF) estimations which allow for
accurate ranging (or distance measurement) among peers.
Despite these promising features, the design of energy-efficient
and low-cost UWB transceivers is still a challenging task and is
addressed in the following.
4.2.1 UWB analog front-end
A standard-compliant transmitter should generate pulses of
500MHz bandwidth minimum and up to 1GHz (optional). The
possible center frequencies of the pulse are clustered in a low
band group (3494.4MHz, 3993.6MHz, or 4492.8MHz) and a
high band group (ranging from 6489.6MHz to 9484.8MHz).
The payload burst structure requires that the pulse generator
reaches a pulse repetition frequency of 499.2MHz together
with the capability of flipping the pulses polarity at the same
rate. An accurate control of the pulse envelope will also help
-10mW/MHz
75nW/MHz
10.6 GHz
UWB
Spectrum
3.1 GHz
"Part 15 Limit"
-41.3dBm/MHz
Emitted
Signal
Power
NB
Spectrum
'0' '1' '0'
FIGURE 3: SPECTRUM CHARACTERISTICS (LEFT) AND TEMPORAL SHAPE (RIGHT) OF UWB PMM (GREEN) AND NB FSK (BLUE) MODULATED SIGNALS.
Compared to NB systems, the UWB does not suffer from multi-
path fading effects since the individual time-delayed multi-path
replicas of a short pulse are resolvable at the receiver. Moreover,
the inherent low duty cycle of pulse-based signaling schemes
coupled with pseudo-random time-hopping techniques results
relax necessary filtering to fit inside the power emission masks
imposed by the regulation.
At the receiver, the architecture choice is not only driven by the
targeted data rate but also by the requirement for time-of-flight
WIRELESS SENSOR NETWORKS DEVICES: OVERVIEW, ISSUES, STATE-OF-THE-ART AND PROMISING TECHNOLOGIES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
13
4.2.2 UWB digital processing
The digital baseband (BB) receiver functions within an
IEEE802.15.4a-compliant (non-coherent) UWB-LDR system
consist of synchronization, channel and drift estimation,
Synchronization Frame Delimiter (SFD) detection and ranging, as
well as data demodulation and decoding. The task dependencies
are shown in Fig. 4 and infer sequential processing of most of
the tasks to allow energy-efficient (clock) gating of inactive
processing modules. Synchronization is the most processing-
intensive task due to the low-duty cycle of pulse-based preamble
waveforms - defined by the pulse repetition period (PRP) - and
the high Nyquist rate of UWB signals. The two-dimensional
search space is defined by the number of clock cycles per PRP
(time-alignment) and the code length of one synchronization
symbol (phase-alignment). To reduce this search space, the
analog front end implements a form of dynamic sub-sampling
motivated by the observation that UWB communications in
realistic propagation environments demonstrate dense diffuse
multipath effects with rms delay spreads of tens of nanoseconds.
Our sub-sampling scheme exploits this spreading of the pulse
energy by accumulating consecutive (Nyquist rate) samples to
form one combined detection statistic, which is then transferred
to the synchronization hardware composed of a matched filter
and a threshold-based periodicity detection module to guarantee
an optimal detection performance. It is up to the BB to control
the sub-sampling period, which is either set statically (based on
analytical channel models) or updated dynamically, based on
ongoing channel estimations and a-priori information. On top
of reducing the effective cycles per PRP, sub-sampling reduces
the acquisition time for a fixed number of hardware resources
and associated clock frequency requirements (design exploration
parameter).
After initial time and phase alignment, the channel delay profile
is estimated by comparing the time-aligned matched filter
statistics with an energy threshold to decide whether or not a
given sub-sampling phase contains sufficient signal energy to
be considered for further processing. To improve the SNR, the
estimation. A coherent receiver offers accurate synchronization,
which helps in both directions but imposes tighter requirements
on the frequency synthesis and digital baseband operations. The
use of a non-coherent receiver relaxes phase noise specifications
and overall power consumption at the expense of a performance
loss and has been the selected architecture for a first UWB LDR
prototype by STMicroelectronics.
The need for spatial resolution will lead to relatively high
sampling rates that the low data rate does not require. A quick
look to the literature shows that ADC technology has progressed
a lot during the last decade, and state-of-the-art efficiencies
on the order of tens of picoJoules per sample have been
demonstrated over a few bits [21]. Another means to reducing
the power consumption linked to the ADC is to take advantage
of the wide band characteristics of the UWB signals, which can
be strongly clipped with very limited impact on their correlation
properties [22]. It has also been shown that in low SNR and low
signal-to-interferer ratios, the resolution of the ADC has low
impact on performances [23]. This method offers the additional
advantage of not having to use a variable gain amplifier in the
receive chain.
The objective of the non-coherent receiver is to measure the
power of an incoming signal, usually averaging the measure
over a sufficient duration so as to detect variations of the mean
signal level compared to its variance. So-called correlation
detectors can be implemented through an analog correlator using
a multiplier to combine the incoming signal and its replica (or
signals coming from two antennas).
An alternative introduced by Faran and Hills [24] based on the
work of Van Vleck is to use a polarity coincidence correlator
which performs the same correlation after strongly clipping
the signals. Cheng [25] showed that the loss due to clipping
with respect to the analog multiplier is less than 4dB under
any condition of signal power, noise power, and shape of
auto-covariance. This tradeoff brings a much lower complexity
through the use of only 1 bit ADC.
14 COPYRIGHT STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
resulting binary channel mask is fed to the Trellis-based drift
estimation [26], SFD detection composed of thresholding, and
matched filtering, ranging based on threshold-based first-path
detection and demodulation implementing a rake-like energy
detection scheme adapted to a 2-BPM modulation.
4.3 Position-enabled sensor networks
Pervasive, ubiquitous, invisible, adaptive, or embedded are among
the many adjectives commonly used to describe the Ambient
Intelligence concept linked with WSN. Indeed, the ability to
create seamless communications between hundreds of intelligent
tiny devices anywhere and anytime is what enables a whole class
of new applications. Networking is crucial in WSN to perform
many actions such as remotely controlling the devices, querying
one particular sensor, or adding some level of coordination and
self-organization among the nodes. Unfortunately, networking
in general and more particularly wireless networking brings up
a series of issues. An ideal network should be self-configured
and should guarantee packet delivery and quality of service,
resist malicious eavesdropping, offer self-healing capability,
etc. Refer to [27] for an outlook of those network related topics.
In addition to the transducer data itself, it can be very useful to
transmit the context (location, time, etc.) and the destination (user
profile, history, etc.) as well. Knowing the location or position of a
particular node in a widespread network is particularly interesting
from several points of view. At the radio level, this information
could be useful to dynamically adapt the emission power so as
to save energy or to reduce the risk of packet collisions. At the
networking level, this information is useful to, for example,
optimize broadcast or routing procedures. At the security level,
the position of devices can be used to detect malicious device
intrusion or to limit the risk of packet interception. However,
the biggest benefit of position information goes to applications.
It can greatly simplify the maintenance tasks of existing WSN
applications; it can enhance already existing applications with
location-based query mechanisms, as an example; it can finally
contribute to the creation of brand new applications such as
structural health monitoring or indoor navigation systems, where
accurate position information is mandatory.
Military radars and GPS are two typical examples of outdoor
positioning techniques that have existed for many years. The
challenge of positioning in WSN is the need for a low cost,
low power, and scalable indoor positioning system with a
reasonable accuracy. One important part of a positioning
TABLE II. PERCENTAGE OF ERROR BETWEEN
THE MODEL PREDICTION AND THE ACTUAL
MEASURED ADDITIONS AND MULTIPLICATIONS
PER PIXEL. FOR EACH GOP OF EVERY
SEQUENCE, THE AVERAGE ERROR OVER A
NUMBER OF ADAPTATION POINTS IS PRESENTED.
FIGURE 4: FUNCTIONALITY OF AN IEEE 802.15.4A-COMPLIANT
DIGITAL BASEBAND RECEIVER; IN THE MANDATORY MODE, EACH
SYNCHRONIZATION SYMBOL IS COMPOSED OF A TERNARY CODE OF
LENGTH 31.
C
o
a
r
s
e

s
y
n
c
S
F
D

D
e
t
e
c
t
i
o
n
D
e
m
o
d
u
l
a
t
i
o
n
D
e
c
o
d
i
n
g
C
h
a
n
n
e
l

e
s
t
i
m
a
t
e
D
r
i
f
t

e
s
t
i
m
a
t
e
R
a
n
g
i
n
g
Task scheduling
PHR
Data
(< 128 bytes)
IEEE 802.15.4a
frame format
Signal waveform
Preamble
(16,64,1024 or 4096 symbols)
SFD
(8 symbols)
4 3 2 1
FIGURE 4: FUNCTIONALITY OF
AN IEEE 802.15.4A-COMPLIANT
DIGITAL BASEBAND RECEIVER;
IN THE MANDATORY MODE,
EACH SYNCHRONIZATION SYMBOL
IS COMPOSED OF A TERNARY
CODE OF LENGTH 31.
WIRELESS SENSOR NETWORKS DEVICES: OVERVIEW, ISSUES, STATE-OF-THE-ART AND PROMISING TECHNOLOGIES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
15
system is its capability to measure the distance between two
devices, called ranging. Several techniques exist that rely on a
given signal (infrared, electromagnetic, ultrasound, etc.) and a
measurement mean (angle of arrival, signal propagation loss,
time of flight, etc.). Since RF devices are already in use for data
communications in WSN, radiolocation-based techniques are
commonly used and offer the advantage of having good obstacle
penetration. Received Signal Strength techniques are preferred
for narrowband radio systems (such as the IEEE802.15.4)
but suffer from imperfect radio propagation loss models,
manufacturing variation, battery levels affecting the emission
level, and wireless channel effects varying in space and time.
Time of Arrival techniques offer a better accuracy for UWB
radios, thanks to the excellent signal resolution of nanosecond
long impulses. The effect of multipath propagation and potential
clock drift, however, usually degrades the achievable accuracy,
and a tradeoff must be found between the cost/power of the
solution and the ranging accuracy.
The 2-D position of a device can be easily inferred from the
known position of three reference (or anchor) devices by
performing one-to-one ranging measurements and using simple
triangulation methods. However, in widespread WSN, it is
usually not possible to have all nodes in the vicinity of at least
three (or four in 3-D scenarios) anchor nodes, due to limited radio
range and the relatively higher cost of anchor nodes. Distributed
techniques to compute an estimate of each (so called) blind node
position from a small subset (typically 5 to 20%) of anchors
have been the subject of a lot of research in the past years
[28]. The simplest techniques simply combine approximations
of the distance to anchor nodes over a multihop link with
triangulation-like methods. More advanced algorithms exist that
try to iteratively minimize a global error function that models
the discrepancies between latest distance and position estimates.
4.4 Disruptive technologies
Beyond the progressive enhancements of current technologies,
some more revolutionary ones are worth mentioning in the
context of WSN because they may open the door to even wider
applications and contexts of use. Even if the list is not exhaustive,
this section highlights some research fields that might sooner
or later impact the design of WSN devices, especially those
regarding power consumption and integration aspects.
4.4.1 Semiconductor processes
The market penetration of sensor networks is closely linked
to robustness of the wireless link while form-factor and cost
are strong but secondary constraints. Out-of-band noise and
interference rejection are key to robustness and cannot always
be realized in classical CMOS technology, due to limited quality
coefficient of passives. At the expense of higher costs, the
Silicon-on-Insulator technology can improve filter performances
due to excellent isolation to the substrate through higher quality
of buried oxide. If not sufficient, alternatives exist that requires
module assembly such as on-wafer mounted bulk acoustic
wave filters or multi-layer thin film technology and system-
in-package that both provide the capability of integrating
high quality passives and antennas. For some types of sensors
such as accelerometers or gyroscopes of digital micro mirrors,
surface-micro-machining is an alternative that is compatible
with classical CMOS process. Use of this technology can also
be extended to energy scavenging.
4.4.2 Energy scavenging
Thanks to technology and design advances, the power needed
by a sensor to operate is decreasing to levels as low as 100uW
on average. The use of a 1cm3 Li-ion battery would enable a
sensor lifetime of 3 years maximum. Alternative technologies
make it possible within the same volume to scavenge or harvest
the same amount of energy from sources in the environment of
the device, that is, without limiting the lifetime of the sensor.
The main sources that have been envisioned up to now are
kinetic, thermal, and electromagnetic. The first one is mainly the
transformation of vibrations to electricity through piezoelectric,
electrostatic, or magnetic induction effects and can provide up
to 300uW/cm3. The second one is the transformation of a heat
16 COPYRIGHT STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
flux into electricity and can provide up to 40uW/cm3. The third
one retrieves energy from light - at levels up to 15000uW/cm3
outdoor and down to 10uW/cm3 indoor - or from RF radiation
up to tens of milliWatts in the case of a strong source such as an
RFID tag reader. The best candidates for integration into CMOS
technology are the electrostatic through MEMs cantilevers and
heat generators.
4.4.3 Wake-up radio
Given the low duty cycles of WSN applications, the ability to
partly switch off the node during inactive periods can provide
important power consumption savings. Existing MAC strategies
combine communication scheduling mechanism and carrier
sense. In the case of UWB, the signal is subject to low signal-
to-interferer ratios that prevent the use of a simple peak detector.
Examples of passive detection of UWB pulse trains using
specific IDT techniques have been demonstrated in the literature
but suffer from the very high insertion losses of the Surface
Acoustic Wave devices (~60dB). An active detection has also
been proposed, taking advantage of the very high gain of super-
regenerative architecture [29]. Given the power consumption,
the budget of the wake-up system should be limited to a fraction
of that of the transceiver, a very attractive concept requiring
strong and innovative efforts.
4.4.4 Low power design
Besides the more classical techniques to reduce active and
standby power - including low-power memory subsystems,
individually enabled peripherals, pipelines, clock and power
gating, dynamic voltage scaling, or multi-threshold standard
CMOS cell libraries to trade-off speed versus leakage - the
authors would like to point out two promising research activities
that will further improve the SoCs energy-efficiency referred
to as asynchronous and subthreshold design. While the first
technique reduces the power consumption by omitting clocks for
sequencing, the latter one operates the circuit in the transistors
subthreshold region. In [30], an 80% improvement in energy-
efficiency was reported for an asynchronous 8-bit processor
implementing the Atmel AVR instruction set architecture in
STMicroelectronics 0.13-m CMOS process. In [31], an FFT
processor design in a 0.18-m CMOS process which operates
down to 0.18 Volts at the expense of computing bandwidth has
been realized.
5. CONCLUSIONS
This paper has presented an extended survey and status of
Wireless Sensor Networks. The purpose was both to describe
the state-of-the-art of current technologies and to address future
trends and expectations as well. Application issues have been
left aside, allowing us to focus on generic subsystems and
typical features of WSN systems like wireless communication,
networking aspects, or energy management.
The evolution of the WSN context from research concept to mass
market products has first been addressed to better understand
which are todays open issues other than technological
limitations. Specifically, market uncertainties, spectrum and
regulations, reliability, setup/configuration complexity, and a
lack of a killer application have been discussed. This pointed out
that technology is only one among many factors impacting the
possible future of WSN systems.
As a second step, the key features of such systems have been
addressed and analyzed one by one to get a picture of where
WSN technology stands today. On the hardware level, the state-
of-the-art subsystems such as transceivers and microcontrollers
have been described, and relevant metrics such as power
consumption have been assessed. On a more global level, energy
management architectures and techniques have been introduced.
On an even higher level, networking aspects and topologies have
been addressed as well. This overview has highlighted the fact
that a clever design of WSN nodes requires care of the global
system as much as that of each independent function.
Finally, research fields and promising technologies that could
shape the future of WSN have been investigated. Possible
WIRELESS SENSOR NETWORKS DEVICES: OVERVIEW, ISSUES, STATE-OF-THE-ART AND PROMISING TECHNOLOGIES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
17
enhancements of current narrowband systems, for instance, by
means of space diversity, were addressed. UWB technology has
been introduced, and a dedicated prototype has been described.
Innovative networking features such as positioning have also
been investigated. Possibilities of disruptive technologies such
as advanced processes, energy scavenging, or wake-up radio
have also been addressed.
In conclusion, it has been shown that, even if the WSN market
suffers many uncertainties, today all building blocks and several
integrated solutions are already available to fulfill requirements
of the first generations of applications. Research continues to be
very active in investigating revolutionary approaches to answer
to higher needs while providing new or better features. In other
words, the potential of improvement of WSN technologies is still
huge, and concepts such as energy scavenging may even open
the way to application fields that have not yet been thought of,
fields which would expand WSN deployment even further.
REFERENCES
[1] IEEE802.15.4 specications, October 1
st
2003, http:
//www.ieee802.org/15/pub/TG4.html
[2] Zigbee specications, version 1.0, December 14
th
2004,
http://www.zigbee.org
[3] Frost & Sullivan research, WORLD UWB & ZIGBEE
CHIPSET MARKET, May 16
th
2006, http://www.frost.com
[4] http://www.st.com/stonline/products/families/sensors/
sensors.htm
[5] IEEE1451 specications, http://ieee1451.nist.gov
[6] W. Kluge et al., A FULLY INTEGRATED 2.4GHZ IEEE
802.15.4 COMPLIANT TRANSCEIVER FOR ZIGBEE
APPLICATIONS, IEEE Int. Solid-State Circuit Conference
(ISSCC), pp. 372-373, 2006.
[7] B. W. Cook et al., AN ULTRA-LOW POWER 2.4GHZ RF
TRANSCEIVER FOR WIRELESS SENSOR NETWORKS IN 0.13UM
CMOS WITH 400MV SUPPLY AND AN INTEGRATED PASSIVE
RX FRONT-END, IEEE ISSCC, pp. 370-371, 2006.
[8] I. D. ODonnell, A BASEBAND, IMPULSE ULTRA-WIDEBAND
TRANSCEIVER FRONT-END FOR LOW POWER APPLICATIONS,
PhD Thesis, U.C. Berkeley, May 8
th
2006.
[9] A. M. Holberg and A. Saetre, INNOVATIVE TECHNIQUES
FOR EXTREMELY LOW POWER CONSUMPTION WITH 8-BIT
MICROCONTROLLERS, ATMEL White Paper
on the AVR-ATmega165P, February 2006.
[10] F. Forster, ULTRA-LOW POWER MICROCONTROLLERS FOR
EMBEDDED SECURITY APPLICATIONS, TI White Paper on
the MSP430F2xx, February 2006.
[11] P. Clarke, STMICROELECTRONICS LICENSES ARMS
SANDCAT CORE, EE Times, July 2005.
[12] L. Benini, A. Bogliolo and G. De Micheli, A SURVEY OF
DESIGN TECHNIQUES FOR SYSTEM-LEVEL DYNAMIC POWER
MANAGEMENT, IEEE Trans. on Very Large Scale Integration
(VLSI) Systems, vol. 8, no. 3, pp. 299-316, 2000.
[13] A. El-Hoiydi and J.-D. Decotignie, WISEMAC:
AN ULTRA LOW POWER MAC PROTOCOL FOR THE DOWNLINK
OF INFRASTRUCTURE WIRELESS SENSOR NETWORKS, 9
th
Int. Symposium on Computers and Communications
(ISCC), vol. 1, pp. 244-251, July 2004.
[14] B. H. Calhoun et al., DESIGN CONSIDERATIONS FOR
ULTRA-LOW ENERGY WIRELESS MICROSENSOR NODES,
IEEE Trans. on Computers, vol. 54, no.6, pp. 727-739,
June 2005.
[15] S. W. Arms et al., Power Management for Energy
Harvesting Wireless Sensors, SPIE Int. Symposium on
Smart Structures & Smart Materials, pp. 1-9, March 2005.
18 COPYRIGHT STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
[16] T. S. Rappaport, WIRELESS COMMUNICATIONS, PRINCIPLES
& PRACTICE, Prentice Hall PTR, 1996.
[17] G. E. Corazza et al., A CHARACTERIZATION OF INDOOR
SPACE AND FREQUENCY DIVERSITY BY RAY-TRACING
MODELING, IEEE Journal on Sel. Areas in
Communications, vol. 14, no. 3, pp. 411-419, April 1996.
[18] T. Weiguo and Y. Chenyang, A DELAY DIVERSITY SYSTEM
BASED ON IEEE 802.15.4 WITH RAKE-LIKE RECEIVING,
7
th
Int. Conference on Signal Processing (ISCP),
vol. 2, pp. 1686-1690, 2004.
[19] MB-OFDM PHY 1.1 Specication, http://wimedia.org
[20] IEEE Std 802.15.4a, PART 15.4: WIRELESS MEDIUM
ACCESS CONTROL (MAC) AND PHYSICAL LAYER (PHY)
SPECIFICATIONS FOR LOW-RATE WIRELESS PERSONAL AREA
NETWORKS (LR-WPANS), Draft P802.15.41, April 2006.
[21] G. Van der Plas, S. Decoutere and S. Donnay,
A 0.16PJ/CONVERSION-STEP 2.5MW 1.25GS/S 4B ADC
IN A 90NM DIGITAL CMOS PROCESS,
ISSC conference, February 2006.
[22] J.H. Van Vleck and D. Middleton,
THE SPECTRUM OF CLIPPED NOISE, IEEE proceedings,
vol. 54, pp2-19, January 1966.
[23] I.D. ODonnell and R.W. Brodersen, AN ULTRA-WIDE
BAND TRANSCEIVER ARCHITECTURE FOR LOW POWER,
LOW RATE, WIRELESS SYSTEMS, IEEE Trans. On Vehicular
Technologies, vol. 54, issue 5, pp.1609-1622,
September 2005.
[24] J. Faran and R. Hills, CORRELATORS FOR SIGNAL
RECEPTION, Acoustic Research Labs, Harvard University,
Cambridge, MA, Tech. Memo No 27, 1952.
[25] M.C. Cheng, THE CLIPPING LOSS IN CORRELATION DETECTORS
FOR ARBITRARY INPUT SIGNAL-TO-NOISE RATIOS, IEEE
Trans. on Information Theory, vol. 14, no 3, May 1968.
[26] A. Wellig and Y. Qiu, TRELLIS-BASED MAXIMUM-
LIKELIHOOD CRYSTAL DRIFT ESTIMATOR FOR RANGING
APPLICATIONS IN UWB-LDR, IEEE Int. Conference on
UltraWideBand (ICUWB), pp.539-544, September 2006.
[27] D. Blasi et al., AD-HOC WIRELESS SENSORS NETWORKS:
CHALLENGES AND ISSUES IMPACTING ON NETWORKING,
ST Journal of System Research, Special Issue on Wireless
Sensor Networks, 2006.
[28] Langendoen and Reijers, DISTRIBUTED LOCALIZATION
IN WIRELESS SENSOR NETWORKS; A QUANTITATIVE
COMPARISON, Int. Journal of Computer and
Telecommunications Networking, Elsevier Computer
Networks, vol. 43, issue 4, pp. 499 518, 2003.
[29] M. Pelissier, A NEW PULSE DETECTOR BASED ON SUPER-
REGENERATION FOR UWB LOW POWER APPLICATIONS,
IEEE ICUWB conference, pp.639-644, 2006.
[30] L. Necchi et al., AN ULTRA-LOW ENERGY ASYNCHRONOUS
PROCESSOR FOR WIRELESS SENSOR NETWORKS, 12th
IEEE Int. Symposium on Asynchronous Circuits and
Systems (ASYNC), pp. 78-85, 2006.
[31] A. Wang and A. Chandrakasan, A 180-MV SUBTHRESHOLD
FFT PROCESSOR USING A MINIMUM ENERGY DESIGN
METHODOLOGY, IEEE Journal of Solid-State Circuits,
vol. 40, no. 1, pp. 310-319, January 2005.
CONTACT: ST.JOURNAL@ST.COM
AD HOC WIRELESS SENSOR NETWORKING: CHALLENGES AND ISSUES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
19
AD HOC WIRELESS SENSOR
NETWORKING:
CHALLENGES AND ISSUES
Wireless Sensor Networks (WSN)
constitute an emerging application field
of microelectronics that promises wide
support of the interaction between people
and their surroundings.
Because of the extremely variable nature
of this interaction, the topic of WSN
is still in its early stages of development
and is giving rise to several challenges,
from the design of a smart device
that allows easy and reliable access
to the environmental characteristics,
to the creation of a wireless network
of devices that cooperate to perform
complex tasks. In this field,
the cooperation among nodes is the most
peculiar aspect reflecting directly
on the network operation.
This paper gives an overview of the topic
from the Ad Hoc networking concept
perspective. Although the seeds
of these concepts can be found
in some already existing implementations
(e.g., ZigBee), we believe many further
1. INTRODUCTION
It is easy to recognize the main changes of our daily life caused
in the past decade by the effect of the marriage of computation
and communication technologies.
The synergy resulting from such a combination is producing
a powerful technological push: the Wireless Sensor Networks
(WSN) is emerging as the new revolution that will make a reality
the vision of people like Gordon Bell and Mark Weiser [1], who
envisaged the ability of microelectronics to form new computer
classes and the capability of networking to support the Pervasive
or, Ubiquitous Computing, i.e., the idea of integrating
computation and communication into the environment, allowing
the seamless interaction of computing entities with people in a
natural and automatic way.
The potential of the WSN concept simply lies in the three
words composing the acronym itself: Wireless puts the focus
refinements could be introduced to fully exploit
the potential of the widely investigated Ad
Hoc approach and the strength of the mutual
interaction of networked elements.
Danilo Blasi
(1)
, Vincenzo Cacace
(1)
, Luca
Casone
(1)
, Marco Rizzello
(1)
,
Salvatore Rotolo
(1)
Luciano Bononi
(2)
(1) STMicroelectronics

(2) Department of Computer Science,
University of Bologna, Italy
COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF BOLOGNA, 2007.
20 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF BOLOGNA, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
on the freedom that the elimination of wires gives in terms
of mobility support and ease of system deployment; Sensor
reflects the capability of micro-/nano-technology to provide
the means to perceive and interact in a wide sense with the
world; Networks gives emphasis to the possibility of building
systems whose functional capabilities are given by a plurality of
communicating devices, possibly distributed over large areas.
In this paper, we concentrate directly on the keyword Network,
presenting a general overview of the topic and identifying the main
challenges that still require solutions for the effective introduction
of some real innovation. Given the WSNs peculiarities, which
make them quite different from traditional (wired and wireless)
networks, real, effective solutions may only be designed by
exploiting the cooperation among many elements, i.e., the
potential resources of each single networks member can be
aggregated and organized to implement multifaceted features and
to perform complex operations. In the network perspective, this
design approach leads to the Ad Hoc networking approach [2].
Ad Hoc networks are commonly defined as a kind of general,
infrastructure-less, cooperation-based, opportunistic network,
possibly customized for specific scenarios and applications, as
the Latin expression Ad Hoc indicates something which is
tailored to a given matter. This networking approach has to face
frequent and random variations of many factors (radio channel,
topology, data traffic, etc.), implying the dynamic management
of a large number of parameters in the most efficient, effective,
and reactive way.
To this end, a number of key research problems have been
studied (and solutions have been proposed) by Ad Hoc
networking researchers:
self-configuration and self-organization in infrastructure-less
systems;
support for cooperative operations in systems with
heterogeneous members;
multi-hop peer-to-peer communications among network nodes;
networks self-healing behavior providing a sufficient degree
of robustness and reliability; and
seamless mobility management and support of dynamic
network topologies.
In the following sections, we will discuss more details of the Ad
Hoc networking concepts applied to WSNs. In Section 2, we
sketch the architecture of a generic WSN node, just to highlight
some peculiarities quite different from typical elements of a
data network that directly impact the networking operation.
In Section 3, we investigate the most important requirements
emerging on the basis of the communication features of WSN
networks, and we show how an Ad Hoc approach could apply
to specific applications, in terms of communication model and
traffic characterization. Section 4 identifies some added values
of the WSN networking features. Section 5 illustrates some of
the state-of-the-art solutions and offers a short perspective on
what the proposed standards have already captured. Conclusions
are reported in Section 6.
2. DEVICE FEATURES
A possible simplified functional architecture of a generic WSN
device can be found in [2] and is shown in Fig. 1.
Altogether, the illustrated components form a complex system
which can be configured according to a wide set of application
Environment
Device
Sensor
Actuator
Actions
Samples
ADC
DAC
XDCR
Power
MCU RADIO
Application
Networking protocols
Operating system
Power scavenging
External energy
Memory
S
e
r
v
i
c
e
s
FIGURE 1: THE FUNCTIONAL ARCHITECTURE OF A WSN DEVICE.
AD HOC WIRELESS SENSOR NETWORKING: CHALLENGES AND ISSUES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
21
requirements. By means of a transducing unit (XDCR), which
may host multiple kinds of sensors and/or actuators [3], the
device interacts with the surrounding environment by collecting
samples of environmental characteristics (e.g., temperature,
humidity, pressure, etc.) and by causing actions (e.g., air
conditioning, light level adjusting, fire alarming, etc.). A
processing unit (MCU) controls the overall system and manages
the procedures that make the device collaborating with the other
network members to carry out the assigned tasks.
A memory unit stores all processed data, networking control
information, and the code to be executed by the MCU (e.g.,
application, networking protocols, operating system, services),
thus providing the device with direct access to the information it
needs. A radio interface unit connects the device to the network
through the wireless channel and makes it able to share information
with its peers via packets exchange. The whole system is
usually powered by a battery pack that might be recharged by
additional ambient power scavenging units such as solar cells.
Besides flexibility, common WSN applications demand low cost
devices, imposing severe constraints on the system design and
configuration.
Typical WSN devices could be as large as coins, possibly
resulting in limitations of processor power and memory.
Moreover, they are often expected to be untethered, which
emphasizes their capability of saving energy while allowing
the completion of network operations. Furthermore, given the
need for such devices to be used in hostile environments, the
robustness of these devices becomes a central issue.
The kind of operations WSN devices may participate in when
networked depends on the solutions to the above issues, as will
be illustrated in the following sections. In this paper, we point out
the two key factors that most differentiate WSNs from other data
networks: the application dependency, which mainly impacts the
communication features, and the Ad Hoc networking approach,
which specializes in the monitoring and control contexts.
3. COMMUNICATION FEATURES
WSNs are operated for instrumenting their surroundings [4].
They typically count on a large number of devices that are able
to detect and/or react to external stimuli associated with events
or objects of interest. As already mentioned when we introduced
the concept of cooperation, these devices, when networked, may
work together to perform complex monitoring and control tasks,
depending on the intended application (e.g., the devices may be
demanded to report simple detections of the presence of a moving
object or the number of such kind of detections performed in a
predefined time interval). The application dependency, combined
with other peculiarities of WSNs (wireless channel usage, device
heterogeneity, device redundancy, to name a few), impacts on
the network communication model (i.e., the roles assumed by
devices in the communication context, the rules adopted for
information gathering and distribution) and on the characteristics
of the data/control traffic injected into the network.
3.1 Communication model
Users of a WSN are likely interested in gathering information
about environmental conditions, current events, objects, or
living beings as well as in remotely enabling a conditioning
system located inside some possibly hazardous place to regulate
its environments characteristics. Monitoring and control
tasks can be data centric, that is, they could aim to react to
the existence of a given set of information rather than being
triggered by who produces some portion of the information set.
This consideration becomes more relevant when taking into
account that device redundancy is normally exploited to improve
accuracy and reliability of the network operations; for instance,
the correctness of a metering procedure can benefit from the
collection of correlated data coming from multiple devices, in
terms of both data availability (e.g., active originators may be
replacing failed ones in producing data) and measure settling
(e.g., mean of different samples). On the other hand, deploying a
large number of transducing devices in a region for the purposes
previously described would make the reference to each single
networks member (e.g., address) less relevant. This is because it
22 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF BOLOGNA, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
contributes to operating the network without being essential for
that; in other words, the single device may become anonymous
in the communication context.
This line of reasoning distinguishes WSNs from traditional
networks in the characterization of the communication paradigm;
the address centric approach used in end-to-end transmissions
between specific devices, with explicit indication of both source
and destination addresses in each packet, could be replaced with
an alternative (and somewhat new) data centric approach [5],
[6]. Perhaps, a straightforward effect of such address blindness
would be related to the choice of the most suitable data diffusion
strategy for data centric networks. As an example, in a WSN,
a great amount of similar data (i.e., highly correlated samples)
might be produced by multiple sensor devices that are reporting
the occurrence of a common phenomenon. In a well-known data
centric technique called in-network aggregation [5], [6], rather
than separately propagating these data items to data consumers,
redundant data items may be aggregated, depending on the target
application, as they are flowing through the network, so that
considerable energy and bandwidth savings can be achieved.
These considerations may lead designers to model this kind of
network as a sort of distributed data repository where anonymous
devices may behave as data originators (also denoted as sources)
and/or data recipients (also denoted as sinks), depending on
many factors (e.g., application requirements, device capabilities,
and device resources). The typical network deployment would
consist of the sources placed around the areas to be monitored
and the sinks near easily accessible places, the sinks provided
with adequate storage capacity to hold the data from the sources.
Sources may send information to sinks in accordance with
different scheduling policies: periodic (i.e., time-driven mode),
event specific (i.e., event-driven mode), a reply in response to
requests coming from sinks (i.e., query-driven mode), or some
combination thereof [7]. It is worth noting that, if the sources
are unaware of identities of the sinks (as would be the case with
the data centric communication paradigm), their data will be
flooded over the network, eventually reaching their recipients.
Conversely, if the sources know the sinks addresses, data
packets can include such information to be forwarded only to
their intended destinations.
Finally, a mixed communication paradigm, which is data-
centric (i.e., one based on anonymity) at the source side and
address-centric at the sink side, can be defined for WSNs. To
the best of our knowledge, such a mixed communication model
has not been defined or investigated in the literature. A sink
might publish its interest in data by issuing a query packet and
by attaching its address so that every receiving source would
be able to refer to it. Queries might also bring time or event
indications along to give instructions about the data transmission
scheduling to the receiving sources; in this way, time-driven and
event-driven scheduling policies would be specializations of the
query-driven one and thus allowed.
In the mixed model, the efficiency of data delivery obtained
by limiting it only to the interested collectors would not be
exploitable in any way for the query distribution, as it is
not supposed to differentiate among anonymous sources.
Nevertheless, for most applications, the sinks interest should be
directed towards sources located inside specified target regions,
for instance, lamp switches in a building room [6], [8]. To allow
this to occur, queries might somehow specify their target regions
so that every receiver of a query packet might determine whether
or not it belongs to the reported target region, or, in other terms,
whether or not it is a candidate source with respect to that query.
Obviously, with these assumptions, the adoption of a localization
system that enables devices to know their current position (see
Sec. 3.3) becomes mandatory.
3.2 Traffic characterization
One of the key aspects of WSNs is the possibility of easily
tasking sensor devices to produce information about a certain
phenomenon by giving them the necessary instructions about the
type of information to be produced, the type of actions to be taken
AD HOC WIRELESS SENSOR NETWORKING: CHALLENGES AND ISSUES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
23
3.3 A key service: the Localization system
The location information management is an essential feature in
most WSN applications [2]. In most cases, the data gathered by
devices are really useful if and only if they are stamped with
location and time information; for example, a temperature
value without indication of where and when it was detected is
considered meaningless.
The location information could be exploited for both application
and networking scopes, for example, to gather data from a specified
geographic zone, as already described in Sec. 3.1. According to
the mixed communication model, an efficient routing protocol
might introduce some smart rules for delivering queries only to
nodes inside target zones, and corresponding data only to queries
originators, respectively [6]. Discovery and tracking services
(e.g., people, assets) are possible examples of location-enabled
application functionalities required in industrial scenarios.
The location information should be reliable and available, no
matter where the device is placed. Technologies such as the
Global Positioning System (GPS) [9] could be used to obtain
the location information needed, but they may be sensitive to
external disruptive factors or to physical obstacles, regardless
of any considerations about costs and power consumption. A
similar reasoning can be extended to the use of a centralized
location information database, which may not be fault-tolerant
enough when implemented by one single location server. Hence,
the aim of achieving robustness may suggest building up a
distributed or replicated localization system [10].
Besides reliability and availability, another main feature of the
localization system is its accuracy. Again, a GPS without signal
corruption provides quite accurate absolute location information.
Alternatively, under less favorable conditions, the accuracy
should be evaluated in relation to the ranging capability of
devices. This consists of making devices able to estimate the
distance from their neighbors (i.e., devices able to exchange
data packets directly) on the basis of measurements of the
for getting that information, and the time-plan for the reports
production. In addition, a general agreement among all network
members on the coding of the above instructions/information
would ease the system control and increase the efficiency; in
other words, adopting the terminology used in [6], data and/or
control messages in WSNs should be named, that is, structured
by the aggregation of attribute-value pairs that all network
members can interpret. To this end, all devices should refer to the
same code-book, locally stored as a reference, defining a list of
attributes (e.g., type of event, position, time, intensity, accuracy,
etc.) and their respective set of valid values. Besides defining a
common language, the naming techniques could be exploited
for allowing in-network aggregation (introduced in the Sec. 2.1)
of multiple data with respect to some common attributes.
The benefits of the in-network processing, the need for saving
limited network resources, as well as the utility-based filtering
of information to be transmitted would make preferable the
extraction of essential features from detected phenomena,
instead of detailed reproduction. For example, a video coming
from a camera (e.g., for detecting intruders or fire) could be
as helpful as a simple signal sent to the control system, but it
would require more resources and a more complex transmission
management.
Consequently, WSNs are not commonly expected to carry
multimedia data flows or, in other words, sequences of time-
correlated information that cannot be handled one at a time.
Nonetheless, some kind of quality of service (QoS) support
should be provided for managing time critical data (e.g.,
fire alarms) whose delivery may require upper bounded time
interval (i.e., latency). To meet such a requirement, transmission
scheduling solutions for preventing the network congestion which
would cause unacceptable delivery delays should be addressed,
even by considering that congestion may be frequent in WSNs,
due to several factors (e.g., use of a shared wireless medium,
limited bandwidth and buffer size, high level of local contention,
and the multi-hop nature of WSN) that are discussed in Sec. 4.
24 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF BOLOGNA, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
received signal strength and/or the propagation time (e.g., time
of arrival, ToA) of predictable control signals. Ultra Wide Band
(UWB) [11] the impulse-radio version [12] is an example of
technology that can provide high ranging accuracy by estimating
the round trip time between transmitter and receiver. A number
of such technologies and solutions are presented in literature
[10]; methods based on triangulation allow blind devices to
infer their position from the collection of measurements obtained
by assuming the knowledge of reference points, called anchors,
even for devices without ranging capability [13].
4. NETWORKING FEATURES
WSNs are a class of wireless networks that may be generally
conceived in two different topologies, i.e., star and mesh [14];
the former is basically a one-hop topology, where all terminals
communicate with a central node, while the latter refers to
the most general multi-hop topology, where multiple nodes
implement a distribution system for data communication among
wireless terminals.
The focus of this paper is on mesh-configured networks. Such
kinds of network can be self-contained when information/
commands generated by nodes/terminals are directed towards
resident nodes/terminals; nevertheless, gateway nodes may be
provided to allow connections with the external world e.g.,
Internet. Mesh-configured WSNs are supposed to have no need of
any fixed, centralized management infrastructure, while they are
instead to rely on the adoption of the cooperation-based Ad Hoc
networking approach [2, 15]. Each node, besides being a sensing
or actuating device, may act as router, at which point it is able to
receive packets generated by other devices and forward them to
the next hop along a multi-hop path towards their final destination.
Therefore, each device has to multiplex (share) the available
link bandwidth with all active neighbors. Because the WSN
density and utilization needs may be high, the bandwidth
may be negatively affected. On the other hand, WSNs are not
expected to be communication intensive at the application layer,
as their devices are expected to have low communication duty
cycles, reducing the average long-term per-device bandwidth
requirement [16].
While the traffic bandwidth requirement is not the main WSN
networking issue, the reliability is strongly expected to be
fulfilled [2]. Any WSN is deeply involved in and related to
the monitored environment, and any change occurring to the
surroundings will significantly influence its performance;
nevertheless, the network must be able to tolerate and survive
any change by implementing proper reactions and adaptation
mechanisms sustaining communications for both sensed data and
commands. In order to comply with the self-working paradigm,
a WSN should implement a set of viable WSN management
guidelines, including:
self-configuration, i.e., the ability to automatically and
autonomously set relevant parameters to operate according to
some given specifications;
self-organization, i.e., ability to detect the presence of the
other network devices and to modify the working behavior,
accordingly;
self-adaptation, i.e., ability to automatically interpret feedback
information and to adjust optimal settings to well operate in
the environment; and
self-healing, i.e., ability to detect devices and links failures,
by providing autonomous reactions to restore operating
conditions without human intervention.
To obey the self-working paradigm, WSN protocols should be
designed with strong attention to both device coordination and
redundancy exploitation issues, both of which might have to
cope with the network member resource heterogeneity.
4.1 Heterogeneity exploitation
Network node heterogeneity will involve both hardware and
software.
Possible hardware differences include the transducing
capabilities, computational power, storage room, power supply
AD HOC WIRELESS SENSOR NETWORKING: CHALLENGES AND ISSUES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
25
(e.g., power line, battery), and many others. Being somehow
complementary to the redundancy, the hardware heterogeneity
can be exploited to improve the way the whole networks
resources are used, thus helping to improve the networks
resilience to failures and prolonging the whole system lifetime.
The resource awareness can be a good guideline to follow in the
protocol design process. Basically, it may govern the assignment
of the most important and critical network management roles
(e.g., router, gateway, coordinator, etc.) to the devices with the
greatest number of resources; further, it may improve the energy
consumption optimization by allowing dynamic transmission
power adjustments or packet exchange rate minimization.
Greater availability of hardware resources on any device may
be translated into the possibility of putting on that device more
computation intensive codes (e.g., protocol stack, application
programs) i.e., the hardware heterogeneity may turn into
software heterogeneity. For instance, similarly to what has been
already proposed by the IEEE 802.15.4 group [14], WSN devices
may be classified into terminals and nodes: terminals can be
identified as the networks users, or end points, so that they
can act as source and/or sink only and participate in a restricted
set of network operations; conversely, nodes are meant to run the
full set of networking functionalities (e.g., coordination, address
assignment, packet relaying, etc.). Altogether, the nodes form
a distribution system for packets coming from or directed to
terminals, while simultaneously giving them the possibility to
access the network services. Software heterogeneity should be
considered while dealing with the devices coordination, since it
puts some ties on the way nodes can share their functionalities; for
instance, any failed node can be replaced only by another one having
similar resources. Moreover, it could require smart capabilities
sharing policies, since the network may potentially provide
several functions, though widely scattered throughout itself.
4.2 Device coordination
Device coordination is the way the network may achieve both
self-configuration and self-organization. Basically, it means that
devices have to work together to make the network effective.
Even if coordination may be achieved through either centralized
or distributed protocols, the latter are usually preferable; in fact,
distributed approaches, despite being less efficient, are likely
more fault-tolerant and, therefore, more suited to the typical
WSN scenarios.
Distributed protocols could be partially supported by pseudo-
centralized management and coordination schemes even in
the absence of any fixed infrastructure, by making devices
able to self-organize into some form of stable hierarchy. This
is the case of the clustering algorithms (see, for instance, the
ZigBee Cluster-Tree organization [17], or Ref. [18, 19]), where
some nodes may be elected to play a leading role such as the
cluster-head. The clustering concept is differently applied to
WSNs with respect to address-centric networks, according to
the used communication model (see Sec. 3.1). For instance,
cluster-heads may be their clusters representative (e.g., the
cluster identifier is set to the cluster-heads one) [3] in networks
where member addressing is needed, while they are often
used to perform data in-network aggregation to decrease the
number of transmitted messages in WSNs [20]. The advantage
of clustering can be better exploited under the distributed
management and coordination viewpoint when such algorithms
are smart enough to assign leading roles to devices having more
capabilities and resources than others (see Sec. 4.1). In case of
node failures and/or topology changes, clustering algorithms
must be able to dynamically perform cluster re-organization
procedures by adapting the role assignment. For example, [20]
suggests a randomized rotation of the role of the cluster-head
so that uniform energy dissipation due to data collection and
merging is obtained. In addition, the clustering scheme should
be properly assisted and considered in the cross-layered design
of network communication protocols to cope with resources,
stability, and performance issues [19]. Hence, WSN coordination
functionalities must not rely exclusively on a single device (or a
static group of devices) so that the effectiveness of the operations
can be preserved also under conditions of failure.
e
26 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF BOLOGNA, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
Two factors have to be taken into account while dealing with
the distributed protocol design process, namely the network
dimension and the network topology dynamics. The former,
easily appreciable through different system parameters (e.g.,
number of nodes, network extension, node density, etc.), is very
important while assessing the protocols scalability, i.e., their
ability to tolerate performance degradation smoothly, and thus
acceptably, when the networks dimension grows [21]. The latter
tells us how fast the network topology is prone to change because
of the occurrence of disruptive events (e.g., failures of nodes
running out of energy, mobility, variable channel conditions,
etc.), which impacts the ability of protocols to maintain their
correct working behavior.
4.3 Redundancy exploitation
Redundancy exploitation means designing protocols able to take
effective advantage of the existence of multiple information
sources, network devices, routing paths, and so on. Networking
protocols, to be properly deemed as self-healing and self-
adapting, should be capable of smartly and dynamically choosing
among several available resources; in case some resources
become unavailable, the protocol must be able to select and to
start using the remaining ones.
Many examples of redundancy exploitation can be found among
routing solutions for WSNs. Reference [22] proposes an energy-
aware routing protocol that maintains a set of paths alternately
used to convey data packets so that the energy budget of any
single path will not be depleted quickly. In [6], the Directed
Diffusion strategy performs in-network aggregation of redundant
data coming from different sources, building up a robust routing
mesh that gets data from multiple sources to multiple sinks.
Reference [23] proposes a variant of Directed Diffusion called
Gradient-Based Routing. Its key idea is to allow network devices
to know the hop distance from a common sink and to force data
to pass through nodes with decreasing distances until the data
is delivered to the sink; exploiting the typical high density of
WSNs and the likely presence of multiple relayers with the same
distance from the sink, the forwarding process becomes highly
reliable. Device redundancy is exploited also by some MAC
schemes. GeRaF [24] is a technique that integrates routing,
MAC, location awareness, and topology management to forward
packets just towards their intended destinations. In this scheme,
a node calculates its priority in acting as a relayer of a received
broadcast message on the basis of its position towards the final
destination; thus, a set of best positioned nodes may volunteer
to relay the message, possibly originating multiple paths to be
followed.
As the resource unavailability is more likely to occur as a
consequence of node failures e.g., node breakages or energy
depletion the management of such failures is a main issue in
WSNs. A failure classification is thus helpful here, as it allows
for a better understanding of which management policies are
required to handle them. A straightforward categorization can
split them between predictable or unpredictable ones. The
former require an explicit signaling approach to anticipate correct
actions before the failure occurrence; the latter can be handled
only through reaction-based (adaptive) mechanisms, as failures
occur in an unpredictable way. Given that the unpredictable
events in WSN scenarios are likely to occur more often than the
predictable ones, the self-adapting and self-healing capabilities
are expected to be more effectively supported by a reaction-
based (i.e., adaptive) mechanism. This policy may then realize
an acceptable trade-off between simplicity and efficiency issues
with both kinds of events.
5. SOME AD HOC CONCEPTS
IN THE EXISTING STANDARDS
Todays users, industrial, professional, or consumer, are
maturing in their awareness of and practice in the adoption of
pervasive computation and communication systems like WSNs.
The sensor and control arena has been enriched by a wide range
of possible choices, which unfortunately are mostly composed
of many proprietary solutions that are almost incompatible.
Standardization efforts are generally a positive factor in moving
AD HOC WIRELESS SENSOR NETWORKING: CHALLENGES AND ISSUES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
27
producers and vendors to concrete actions for developing largely
reusable and interoperable components and/or devices, with
benefits in terms of time to market, reliability, and cost reduction
for users. Without standards, which require us to put progress
ahead of self-interest, there would be no mass production or mass
communication and, as such, no modern economy and progress.
In the WSN field, several standard efforts have been started
recently; most of them have been in evolution for several
years and are now beginning to have market impact that will
grow over the next few years. There are standards related to
wireless communications and sensor management promoted
by the Institute of Electrical and Electronics Engineers (IEEE)
and those focusing on the application scenarios promoted by
Industrial Alliances. We will consider only the networking-
related standards, starting from a short description of IEEE
802.15.4 that addresses the lower layers of the ISO/OSI
networking reference model for wireless personal area networks
(WPAN). Then, we will introduce ZigBee, based upon IEEE
802.15.4, as an industrial standard technology for providing both
network and application management with enlarged capabilities
obtained by taking up some of the solutions to the typical
challenges of the Ad Hoc networks; these will be highlighted
along with a short description of ZigBees main peculiarities.
5.1 IEEE 802.15.4
The IEEE 802.15 working group defines the physical layer
(PHY) and the medium access sub-layer (MAC) for low-
complexity, low-power consumption, low bit-rate WPAN
connectivity; currently, among the four IEEE 802.15 different
frameworks, the IEEE 802.15.4 is considered the most relevant
for WSN scenarios.
A global RF standard for WSN is fundamental because it would
accelerate the technology evolution by identifying leading
directions, and it could yield benefits such as the lowering of
design costs and the interoperability at the communication level.
Moreover, at the raw physical level, standards have to cope with
the issue of frequency band allocation. Different regulations
in different regions of the world should not be a problem for
most applications, whereas the most used wireless technologies,
including all those suitable for WSNs, fall in the ISM bands;
this could bring the advantage of interoperability but also a
concentration of technologies, with possible problems for the
coexistence of different wireless systems.
Nevertheless, the available frequencies are only the starting
point for operating wireless links. In order to make equipment,
software, protocols, and applications, all made by different
manufactures, interoperable, the industry needs to set standards.
The IEEE 802.15.4 Standard [25], approved in 2003 and
amended in 2006 with a b version, is contributing to all of
these aims, and several compliant products are already available
on the market, even if more as development kits only than real
end-products. The standard provides for a low bit-rate (i.e., up to
250kbps, or 1 Mbps in b version) connectivity in the Personal
Operating Space (POS), typically 10/100 meters, through 27
RF channels in the ISM RF bands, in order to guarantee wide
adoption in several market segments worldwide.
The most innovative IEEE 802.15.4 feature is the full support
of mesh networks for battery powered nodes, through the
classification of devices into two different types i.e., Full
Function Device (FFD) and Reduced Function Device (RFD).
An IEEE 802.15.4 network should include at least one FFD
operating as the PAN coordinator for special (but not centralized)
functions, whereas all the other FFDs would contribute to
realize the WSN backbone; RFDs, which are usually intended
as the leaf nodes of the WSN spanning tree, perform simple
tasks more related to sensing than networking. An RFD can
communicate only to one FFD, while an FFD can communicate
to both RFDs and FFDs.
This asymmetry of roles arises from the fact that RFDs have
minimal resources (energy and memory) and basically act as
parasites of FFDs functions by limiting working activity to
(typical) 1% duty-cycle.
FIGURE 2 THE ZI
28 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF BOLOGNA, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
of messages, as described in Sec. 4, and adds the logical
network, security, and applications management on top of the
referenced IEEE 802.15.4 standard by defining the upper layers
of the protocol stack, from network to application (see Fig. 2).
In addition, ZigBee defines application profiles [17], which refer
to a set of template-based description of device configurations,
each one specialized for working to a common cooperative
and distributed application. Aside from its technical aspects,
one of the main tasks of the ZigBee Alliance is to certificate
interoperability among devices made by different manufacturers,
thus expanding their potential adoption.
Technically speaking, ZigBee is a fairly good standardization
effort that is gaining wide acceptance of big players, but it is
currently subject to several refinement attempts, which would
risk reducing its market momentum. The first official release
of the standard, known as ZigBee v.1.0 and dated December
As anticipated, the Standard has recently come out of revision,
introducing simplifications to the overall architecture and
resolving some design ambiguities and inconsistencies, while
improving interoperability worldwide. Concurrently, under
development is IEEE 802.15.4a, which essentially regulates
the use of UWB physical medium with enhanced technical and
protocol specifications, thus allowing high bit-rate transmissions
and new management features like ranging/localization
(described in Sec. 3.3).
5.2 ZigBee
On the basis of the current IEEE 802.15.4 specifications, a
consortium of more than 200 companies is negotiating and
working on the adoption of an industrial standard called ZigBee,
whose name and working principle is inspired by the social
behavior of bees that work together to tackle complex tasks.
ZigBee exploits cooperation to allow for the multi-hop exchange
Application
Framework
...
ZigBee Device Object
(ZDO)
Z
D
O

M
a
n
a
g
e
m
e
n
t

P
l
a
n
e
Application (APL) Layer
Application
Object 1
Application
Object 240
Application Support Sublayer (APS)
APS Security
Management
APS Message
Broker
Reflector
Management
Network (NWK) Layer
Medium Accross Control (MAC) Layer
NWK
Security
Management
NWK
Message
Broker
Network
Management
Routing
Management
Physical (PHY) Layer
2,4 GHz
Radio
868,915 MHz
End manufactured
defined
ZigBee
TM
Alliance
defined
IEEE 802.15.4
Layer function
Layer Interface
Endpoint 1
APSDE-SAP
Endpoint 240
APSDE-SAP
Endpoint 0
APSDE-SAP
MLDE-SAP MLME-SAP
PD-SAP PLME-SAP
Z
D
O

P
u
b
b
l
i
c
I
n
t
e
r
f
a
c
e
s
A
P
S
M
E
-
S
A
P
N
L
M
E
-
S
A
P
NLDE-SAP
Security
Service
Provider
FIGURE 2: THE ZIGBEE PROTOCOL
STACK.
AD HOC WIRELESS SENSOR NETWORKING: CHALLENGES AND ISSUES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
29
2004, is currently shipped by many platform providers. It is
typically referred to as the Home Control version of the
stack. Unfortunately, its lifetime is marked, and it is changing
in a non-backward-compatible way, while maintaining the
overall skeleton. For the time being, the Alliance has decided
to propose ZigBee in two flavors: a base version featuring the
Home Automation market only, named ZigBee, which is an
enhancement of the old v.1.0, and a ZigBee Pro version,
most likely available starting in 2007, which will incorporate
several features we listed in the previous sections.
The new releases improve ZigBee networks scalability
and performance through full support of mesh networking,
abandoning the hierarchical organization of the old v.1.0
known as Cluster-Tree, and bringing in the so-called many-
to-one routing feature, which allows a concentrator (a gateway
or more generally a sink) to establish routes from all nodes in the
network without large increases in route table size and without
creating a broadcast storm within the network.
This is a first step in the direction of a data centric paradigm (see
mixed mode in Sec. 3.1), which, together with the support for
multicast, will fill the gap with the 802.15.4b added support.
Even more relevant is the new alternate addressing scheme,
which will replace the tree organization by stochastically
assigning the network level addresses and resolving conflicts
where originated. This will result in a lighter management of
addressing issues as the address of the node will no more
depend on its physical position on the tree enhancing the
scalability and thus allowing pure mesh topologies as well as
node mobility.
The concept of reliability is hence strengthened by mesh
networking because its many different and dynamic paths could
be exploited for routing. The frequency agility feature enhances
reliability as well. Starting from the assumption that the channel
and the entire RF band conditions can vary dynamically in both
time and space, this feature will allow the operation to continue
reliably and in an unattended way in the presence of well-known
interference sources, like WiFi or Bluetooth, or in the presence
of an unusable operating channel as well. This is a big step
towards the self-working paradigm (see Sec. 4).
The heterogeneity exploitation is also being considered for
the 2007-later on versions, by adding features like end-
device management and battery powered routers, allowing,
respectively, the employment of battery-powered devices as
leaf nodes - by designating the relevant parent router as a proxy
- and even permitting the deployment of battery-powered router
nodes, given that in the current version they are supposed to be
mains (i.e. line) powered.
If all these are networking add-ons that will differentiate ZigBee
Pro from previous versions, two important enhancements will
have been produced at the application level, starting from the
ZigBee stack. The old mechanism used to describe and
identify the application profiles has been updated with the new
ZigBee Cluster Library (ZCL), which can be imagined as
a code-book (see Sec. 3) where all devices, classified on the
basis of functional domains, have been described in terms of
their attributes/properties in an application-independent fashion;
as an example, the same description for a temperature sensor
can be adopted even if it will be used in a Home or Industrial
Automation application.
All these are only a part of the discussed features in the
ZigBee Alliance. This can be expected, given that such an
organization counts several members with different strategies
and requirements, such as OEMs as well as application-oriented
and semiconductor companies. Now momentum has come, and
2007 will be the year expected for launching ZigBee to the mass
market. At the same time, ZigBee will continue to enhance its
capabilities; mobility management, location awareness, QoS,
power awareness, network-wide and time synchronization,
and new routing strategies will be the ZigBee response to the
classical Ad Hoc paradigm challenges.
30 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF BOLOGNA, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
6. CONCLUSIONS
WSNs are an emerging application field that deserves the great
attention of the microelectronics industry and research. This paper
presents an overview of WSNs with special focus on networking
challenges and issues due to the peculiarities of such networks,
mainly identified as application-dependency and natural demand
for the Ad Hoc approach. ZigBee / IEEE802.15.4 promises this
kind of network availability in a fairly short time, but, in our
opinion, further steps still need to be taken before making WSNs
truly user-friendly.
readily reacting to customer needs, while the standards often
adapt to the market-based process. The companies with the
products that solve real customer needs and desires in a timely
manner will likely become the standard in the home market.
The recent partnership between ST and Ember Corporation
[27] will enable ST to become one of the leading providers of
the ZigBee networking systems. However, ST has also started
thinking about the next generations of WSN devices in order
to preserve its competitiveness while gaining a competitive
advantage by adding improvements at the device and software
levels. In this sense, STMicroelectronics contributes to the
growing knowledge of WSN solutions, continuously fed by
relationships with the academic world. Distributed localization
algorithms, geographical routing, and fairness and congestion
control at MAC layer are only some of R&D areas under
investigation, areas whose importance has been extensively
pointed out in the paper.
REFERENCES
[1] M. Weiser, THE COMPUTER FOR THE 21ST CENTURY,
Scientific American, 265(3), pp. 94-104, 1991.
[2] I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, E. Cayirci,
A SURVEY ON SENSOR NETWORKS, IEEE Communications
Magazine, 40(8), pp. 102-114, 2002.
[3] D.J. Baker, A. Ephremides, and J.A. Flynn, THE DESIGN
AND SIMULATION OF A MOBILE RADIO NETWORK
WITH DISTRIBUTED CONTROL, IEEE J. Sel. Areas
Commun., Sac-2(1), 226, 1984.
[4] D. Estrin, L. Girod, G. Pottie, and M. Srivastava,
INSTRUMENTING THE WORLD WITH WIRELESS SENSOR
NETWORKS, Proc. of ICASSP, 2001.
[5] B. Krishnamachari, D. Estrin, and S. Wicker,
MODELING DATA-CENTRIC ROUTING IN WIRELESS SENSOR
NETWORKS, Proc. of IEEE Infocom, 2002.
e
v
o
l
u
t
i
o
n
Miniaturization
FIGURE 3: CHALLENGES AND ISSUES OF WSNS.
The bubble diagram in Fig. 3 provides a list of main issues that the
natural technology evolution and the market raise. Some of them
have already been tackled by ZigBee / IEEE802.15.4, while some
others are still representing challenges for academic and industrial
researchers, in both the fields of general and Ad Hoc WSNs.
According to recent marketing reports such as those found
in [26], todays wireless sensor industry is being driven
by customer demand for reliability, simplicity, price and
availability, independent of the standardization processes,
whose advancement is often limited by external influences and
political or strategical issues. This can be seen today in the home
automation market, where leading companies are simply and
AD HOC WIRELESS SENSOR NETWORKING: CHALLENGES AND ISSUES
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
31
[6] C. Intanagonwiwat, R. Govindan, D. Estrin, J. Heidemann,
and F. Silva, DIRECTED DIFFUSION FOR WIRELESS
SENSOR NETWORKING, Proc. of IEEE/ACM
Transactions on Networking, 11(1), 2003.
[7] Y. Yao and J. Gehrke, THE COUGAR APPROACH TO
IN-NETWORK QUERY PROCESSING IN SENSOR NETWORKS,
Proc. of SIGMOD, 31(3), 2002.
[8] Y. Yu, D. Estrin, and R. Govindan, GEOGRAPHICAL AND
ENERGY-AWARE ROUTING: A RECURSIVE DATA
DISSEMINATION PROTOCOL FOR WIRELESS SENSOR
NETWORKS, Technical report UCLA-CSD
TR-01-0023, 2001.
[9] NAVSTAR GLOBAL POSITIONING SYSTEM
SURVEYING, ASCE Publications, Technology
& Industrial Arts, 2000.
[10] N. Patwari, J.N. Ash, S. Kyperountas, A.O. Hero III,
R.L. Moses, and N.S. Correal, LOCATING THE NODES
- COOPERATIVE LOCALIZATION IN WIRELESS SENSOR
NETWORKS, IEEE Signal Processing Magazine,
pp. 54-69, 2005.
[11] M.G. Di Benedetto and G. Giancola,
UNDERSTANDING ULTRA WIDE BAND RADIO
FUNDAMENTALS, Prentice Hall Communications
Engineering and Emerging Technologies Series, 2004.
[12] IEEE 802.15 WPAN Low Rate Alternative PHY
Task Group 4a (TG4a)
[13] D. Niculescu and B. Nath, DV BASED POSITIONING IN
AD HOC NETWORKS, KLUWEIR Journal, 2003.
[14] http://standards.ieee.org/reading/ieee/std/lanman
/restricted/802.15.4-2003.pdf
[15] S. Rotolo, D. Blasi, V. Cacace and L. Casone, FROM
WLANS TO AD HOC NETWORKS, A NEW CHALLENGE
IN WIRELESS COMMUNICATIONS: PECULIARITIES, ISSUES
AND OPPORTUNITIES, in Handbook Of Wireless Local
Area Networks: Applications, Technology, Security,
and Standards, CRC Press, pp. 2.15-154, 2005.
[16] E.H. Callaway, Wireless Sensor Networks
Architectures and Protocols,
Auerbach Publications (CRC Press), 2004.
[17] ZigBee Alliance internal documents
(http://www.zigbee.org)
[18] D. Simplot-Ryl, I. Stojmenovic and J. Wu, ENERGY-
EFFICIENT BACKBONE CONSTRUCTION, BROADCASTING
AND AREA COVERAGE IN SENSOR NETWORKS,
in Handbook of Sensor Networks: Algorithms
and Architectures, John Wiley & Sons Inc.,
pp. 343-380, 2005.
[19] L. Bononi, M. Di Felice, L. Donatiello, D. Blasi,
V. Cacace, L. Casone, S. Rotolo, DESIGN AND
PERFORMANCE EVALUATION OF CROSS LAYERED MAC
AND CLUSTERING SOLUTIONS FOR WIRELESS AD HOC
NETWORKS, Performance Evaluation 63 (2006),
Elsevier, pp. 1051-1073
[20] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan,
ENERGY-EFFICIENT COMMUNICATION PROTOCOL
FOR WIRELESS MICRO SENSOR NETWORKS, Proceedings
of the 33rd Hawaii International Conference on System
Sciences (HICSS 00), 2000.
[21] F. Martincic and L. Schwiebert, INTRODUCTION
TO WIRELESS SENSOR NETWORKING, in Handbook
of Sensor Networks: Algorithms and Architectures,
John Wiley & Sons Inc., pp. 1-40, 2005.
32 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF BOLOGNA, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
[22] R. C. Shah and J. Rabaey, ENERGY AWARE ROUTING
FOR LOW ENERGY AD HOC SENSOR NETWORKS,
IEEE Wireless Communications and Networking
Conference (WCNC), 2002.
[23] C. Schurgers and M.B. Srivastava, ENERGY EFFICIENT
ROUTING IN WIRELESS SENSOR NETWORKS,
MILCOM Proceedings on Communications for Network-
Centric Operations: Creating the Information Force, 2001.
[24] M. Zorzi and R. R. Rao, GEOGRAPHIC RANDOM
FORWARDING (GERAF) FOR AD HOC AND SENSOR NETWORKS:
ENERGY AND LATENCY PERFORMANCE,
IEEE Transactions on Mobile Computing, 2(4), 2003.
[25] IEEE Std 802.15.4-2003 Standard: Wireless
Medium Access Control (MAC) and Physical Layer (PHY)
Specifications for Low-Rate Wireless Personal
Area Networks (LR-WPANs).
[26] K. West, WIRELESS SENSOR TECHNOLOGY AND MARKET
TRACKING SERVICE: ZIGBEE, ZWAVE, INSTEON, RFID,
IEEE 802.15.4 AND THEIR COMPETITION,
Report abstract WTRS, 2005.
[27] http://www.ember.com/
CONTACT: ST.JOURNAL@ST.COM
PICO RADIO: FROM VISION TO REALITY
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
33
M. Sheets
(1)
, B. Otis
(1)
, H. Qin
(1)
, N. Pletcher
(1)
,
F. Burghardt
(1)
, J. Ammer
(1)
, T. Karalar
(1)
, P.
Monat
(1)
, Y. Cao
(1)
, D. Markovic
(1)
,
A. Vladimirescu
(1)
, J. Rabaey
(1)
, S. Cervini
(2)

(1) Berkeley Wireless Research Center,
University of California, Berkeley
(2) STMicroelectronics
PICO RADIO:
FROM VISION TO REALITY
The vision of ubiquitous, dense,
heterogeneous, ad-hoc Wireless Sensor
Networks (WSNs) can only become
a reality through the combined efforts
of system miniaturization and power/cost
minimization.
In turn, these efforts open up a variety
of interrelated issues, including protocol
stack design, energy scavenging,
cost/precision tradeoff of components,
and low-power CMOS design.
At BWRC all of these issues have been
addressed in a harmonized way, and
a WSN node (Quark), incorporating
the outcome of multidisciplinary research,
has been designed. This article presents
an overview of the current state of
the Quark node. STMicroelectronics
has manufactured many of the chips
developed at BWRC under the Pico
Radio project.
1. INTRODUCTION
This paper offers an overview of the efforts undertaken at BWRC
(Berkeley Wireless Research Center) in the field of Wireless
Sensor Networks (WSNs) by a research group led by Prof. Jan
Rabaey.
Novel techniques in the areas of protocol stack algorithms,
energy trains, node locationing, digital processing, and RF
communication circuitry have contributed to making the
implementation of a WSN node more energy efficient. At BWRC
a WSN node that integrates these technological advances into a
self-contained energy-scavenging sensor node (Quark Node) has
been developed.
The overall energy consumption of the node is attained by means
of a system level codesign of the protocol stack and circuit-level
low-power design of two ASICs. The size of the node has
been minimized through the design of a custom PCB with an
integrated planar inverted-F (PIFA) antenna and the design of
the energy train so as to allow the node to be self-contained with
batteries recharged by solar energy.
The Quark node is implemented as a printed circuit board that
contains two custom chips (the digital Charm and the RF Strange),
a PIFA antenna, and a power train, described respectively in
COPYRIGHT UNIVERSITY OF CALIFORNIA, BERKELEY, STMICROELECTRONICS, 2007.
34 COPYRIGHT UNIVERSITY OF CALIFORNIA, BERKELEY, STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
subsections 3, 4, 6 and 7. Off-the-shelf analog components for
the baseband complete the system. The block diagram of the
Quark node is shown in Fig. 2. This paper is organized as follows:
section 2 summarizes the functionalities and algorithms used in
the protocol stack; section 3 presents the digital processor and
how SRAM power leakage can be minimized by setting the VDD
at the minimum value as predicted by a DRV (Data Retention
Voltage) Model; section 4 describes the design of the transceiver;
section 5 presents a wake-up receiver architecture that can be
used to reduce power consumption further by allowing the node
to remain in sleep mode as long as possible, waking up only
when another node is transmitting to it; section 6 presents the
main features of the planar inverted-F antenna (PIFA); section 7
describes the energy scavenging and storage subsystems; section
8 shortly discusses optimizations of the clock subsystem; and
section 9 presents the power consumption breakdown among the
various system components.
2. PROTOCOL STACK
The node implements a wireless protocol stack that is tailored
for sensor network applications. Subsystems generally follow
the OSI reference model [1] and include application, network,
data link, and physical layers. The protocol stack is augmented to
include a locationing subsystem because a location is required to
associate the sensed data with the source or destination.
2.1 Application Layer
Node application functions fall into three categories: controllers,
sensors, and actuators. A sensor application waits passively until
it receives instructions from a controller application. Controllers
can request a one-shot measurement or schedule measurements
to be performed periodically. Upon receiving a request, the
sensor application queries the appropriate external sensor
and returns the data. Controller applications can also instruct
actuators to perform an action, such as toggling an external
FIGURE 1: FOUR CHIPS DESIGNED AT BWRC
AND MANUFACTURED BY STMICROELECTRONICS.
A) CHARM: DIGITAL PROTOCOL PROCESSOR, 2004 IEEE;
B) STRANGE: ANALOG TWO-CHANNEL TRANSCEIVER,
2004 IEEE;
C) 1.9 GHZ DIGITALLY CONTROLLED OSCILLATOR,
2005 IEEE;
D) SRAM LEAKAGE-CONTROL TEST CHIP, 2005 ELSEVIER.
PICO RADIO: FROM VISION TO REALITY
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
35
relay. Flexibility in the application protocol layer allows any
combination of these three operations to be performed through
programmable parameters and detachable user interface, such as
a laptop or PDA.
2.2 Locationing
In a self-configuring sensor network, most sensor nodes are
deployed without any presumed position. Only a small number
of anchor nodes are given a priori information about their
positions with respect to a global coordinate system. The rest
of the nodes then calculate their positions using the position and
distance to these anchors.
The locationing subsystem performs 3-D triangulation using at
least four anchor node reference points. If more anchor nodes
are present in the network, the error in the calculation is reduced
[4]. During triangulation, inter-nodal distances are expressed in
terms of the reference and unknown position coordinates. These
equations are then cast into an over-determined linear system,
which is solved using a least squares algorithm. The hop count
between an anchor and the node is used as a substitute for the real
Euclidean distance. This algorithm, known as Hop-TERRAIN,
was initially proposed by Savarese, et.al. [5]. The sensor network
is assumed to be mostly static, so only slow position changes are
expected and tracked. The location update rate is programmable
and usually operates on an order of minutes.
2.3 Network Layer
The network protocol layer is responsible for implementing
the multi-hop routing protocol. The protocol in use is a form
of the directed diffusion algorithm in [2]. In order to make the
network truly ad-hoc, there are no absolute node addresses.
Packets are routed to a 3-D box-shaped region defined by the
two locations of its opposite corners. The nearest neighbor
links are distinguished by dynamically assigned node
identifiers (IDs) exported from the data link layer (DLL).
In this implementation, the controller can initiate an interest
packet, which is then broadcast to all nearest neighbors.
Neighbors that are spatially closer to the destination repeat
the broadcast. Each intermediate node that sees an interest
packet keeps a record of its controller source location and the
ID of the node from which it came. After the interest packet
reaches its destination, a data packet may be returned to the
controller location in response. At each intermediate node, one
of the saved paths back to the controller location is selected,
and the packet is forwarded down that link. Although our
current selection metric selects a path with the lowest hop
count to the destination, recent results [3] indicate that a
stochastic selection based upon energy metrics can increase
network survivability. Since the network layer is implemented
in software, this change can be made in future revisions.
A flexible platform for further exploration of such alternative
routing protocols is one of the main benefits of this node
implementation.
2.4 Data Link Layer
The data link layer performs ID assignment and maintains
link-to-link connectivity. Network layer routing depends on
the local uniqueness of IDs within every neighborhood. When
a node is activated, it requests the list of used IDs from each
nearest neighbor to create a list of used IDs within two degrees
of separation (two-hop neighbors). It then selects an unused ID
and notifies its neighbors. In certain situations, such as those
involving a lost packet or hidden nodes, the assumption of
Charm
Chip
Discrete
Logic
Analog Rx
Baseband
Strange
Chip
Switch
Array
charge
pump
12 C
hdr
SPI
hdr
JTAG
hdr
GPIO
hdr
UART
hdr
Sensor
W/A/D
16 MHz
osc
buck reg
EEPROM
power monitor
1.8v
reset
battery
from
solar
FIGURE 2: QUARK BLOCK DIAGRAM, 2004 IEEE.
36 COPYRIGHT UNIVERSITY OF CALIFORNIA, BERKELEY, STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
local ID uniqueness may be violated. A detection mechanism is
used to identify and correct this situation.
2.5 Physical Layer
The node physical layer consists of a 1.9GHz two-channel
transceiver and a mostly analog baseband. The air interface is
a 50kbps OOK modulation scheme, and the symbol pulse shape
is determined by the oscillator startup and dampening. Received
symbols are matched and filtered in the analog domain and
then sliced using an adaptive threshold determined using the
packet header. The synch header is 23 bits and consists of a
16-bit string of alternating ones and zeroes, used for bit timing
synchronization and threshold-setting, followed by a 7-bit PN
sequence used for packet synchronization. The synchronization
and detection algorithm is designed to achieve 10
-4
BER at 13dB
input SNR. The maximum packet length is 1024 bits, which is
dictated by the 50ppm clock accuracy and the Doppler effects
in the channel.
3. PROTOCOL PROCESSOR
(CHARM CHIP)
The digital processing chip (Fig. 1a) contains the logic for the
upper levels of the protocol stack, including the digital portion
of the baseband, and the locationing subsystem. The application
and network layers run on an integrated 8051-compatible
microcontroller and can be programmed using C code. The DLL,
locationing, and digital baseband subsystems are implemented
in direct-mapped hardware. The chip is clocked at a relatively
slow 16MHz. This relaxed timing requirement was exploited to
save power and hardware area. For example, in the case of the
locationing subsystem, area savings was achieved by performing
the least squares computation in a time serial fashion using a
single set of computational units along with a register file.
An aggressive power controller selectively gates the power rails
to the eight internal subsystems. An internal voltage converter
generates a 300mV data retention voltage used to reduce leakage
during idle modes (more on this on 3.1). The Charm chip is
implemented in a 0.13m process with six metal layers. The
chip is (2.7x2.7) mm
2

and integrates 3.2M transistors.
3.1 SRAM Leakage Power Minimization
Continuous technology scaling over the past four decades has
enabled higher speed and higher integration capacity in VLSI
designs. However, the chip leakage power increases about five
times each technology generation and becomes one of the main
challenges in future CMOS design [22]. In battery-supported
applications with low duty-cycles, such as the Pico-Radio
wireless sensor nodes [23], cellular phones, or PDAs, in most
situations, active power only accounts for a small portion of
the system power consumption, with leakage power ultimately
determining the battery life. On the other hand, microprocessor
designs of today incorporate large memory components, which
consume a significant portion of system power budget. For
instance, 30% of Alpha 21264 and 60% of StrongARM power
are dissipated in cache and memory structures [23].
While activity factor is usually small in memory structures,
leakage causes a major part of the memory power consumption.
In 70 nm technology, it has been projected that 70% of the cache
power budget will be the leakage power [24]. As a result of both the
technology scaling and large leakage power dissipation in memory
structure, memory leakage suppression is critically important
for the success of power-efficient designs, especially ultra-low
power (ULP) applications. While the leakage of logic modules
in a chip can be effectively controlled by gating off these paths at
standby mode, the leakage suppression of memories is especially
difficult due to the data retention requirement in such structures.
An effective scheme for SRAM leakage suppression in battery
powered applications such as wireless sensor nodes is needed.
Since leakage power of the peripheral circuits during an idle
period can be eliminated by turning off these leakage paths
with switched source impedance (SSI) [25], only the leakage
suppression of SRAM cells is considered. In the proposed
approach, the standby supply voltage (VDD) of the whole
memory is minimized with the memory states preserved. As
a result of reduced voltage, all the leakage components in an
PICO RADIO: FROM VISION TO REALITY
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
37
To achieve optimal power-performance tradeoffs, compiler-level
cache activity analysis is employed to balance the potential for
saving leakage energy against the loss incurred in extra cache
misses. As an example, the cache decay technique applied
adaptive timing policies in cache line gating, achieving 70%
leakage savings at a performance penalty of less than 1% [30].
To exploit leakage control in caches with large utilization ratio
even further, the approach of drowsy caches allocates inactive
cache lines to a low-power mode, where VDD has been lowered
while preserving memory data [24].
With a conservatively chosen standby VDD in the drowsy
caches approach, leakage energy savings of over 70% in a data
cache can be achieved [24]. However, the question still remains
on the lower bound of standby VDD that still preserves the
data, namely DRV. Knowledge of DRV allows a designer to
exploit the maximum achievable leakage reduction for a given
technology. Furthermore, in the sub-1V low power VLSI designs
of today, the reliability requirement on memories has become
the bottleneck in further reducing the system VDD. To enable
even more aggressive memory supply voltage minimization, an
understanding of low voltage SRAM data preservation behavior
is required to quantitatively evaluate the SRAM data retention
reliability under low VDD and optimize the future SRAM
designs for ULV operation conditions.
To obtain silicon verification of the DRV model and explore the
potential of SRAM leakage suppression with ultra low standby
VDD, a 4 KB SRAM test chip with dual rail standby control
was implemented in a 0.13mm technology. Designed for ultra
low-power applications, this scheme puts the entire SRAM into a
deep sleep mode during the system standby period. The test chip
consists of a standard 4 KB SRAM module and a custom on-
chip switch-capacitor (SC) converter that generates the standby
VDD with 85% conversion efficiency. Compared to the existing
SRAM leakage control techniques, the simplicity of this scheme
leads to minimized design effort and, therefore, the minimum
extra power necessary to support control circuitry.
SRAM cell are effectively minimized. This analysis of ultra low
voltage reliable data retention and its results can also be used
for future SRAM design in ULP applications with aggressively
scaled operational VDD.
Presently, at the circuit level, the most effective low power
design methods in minimizing SRAM cell leakage power
involve lowering supply voltage and increasing transistor
threshold voltage (Vth), both of which are detrimental to the
speed of memory read/write operations. For this reason, leakage
reduction techniques at this level typically exploit dynamic
control of transistor gate-source and substrate-source bias to
enhance driving strength in active mode and low leakage paths
during standby periods [26].
For example, the driving sourceline (DSL) scheme connects the
source line of the cross-coupled inverters in an SRAM cell to
negative voltage VBB during read cycle and leaves the source
line floating during write cycle. As a result, the cell read access
time is improved with boosted gate-source voltage and forward
bias of the sourcesubstrate junction of the transistors. The write
cycle is also improved since the NMOS transistors in the cross-
coupled inverter pair are inactive [27]. Another technique is the
negative word-line driving (NWD) scheme. It uses low Vth
access transistors with negative cut-off gate voltage and high Vth
cross-coupled inverter pairs with boosted gate voltage to achieve
both improved access time and reduced standby leakage [28].
The dynamic leakage cut-off (DLC) scheme biases the substrate
voltages of non-selected SRAM cells ~2VDD for VNWELL and
-VDD for VPWELL [29]. A concern in some of these schemes is
that the gate voltages of transistors far exceed VDD, which raises
reliability issues [29]. All of these techniques achieve enhanced
memory operation speed and suppressed standby leakage
current at sub-1V supply voltage compared to those achieved in
conventional cell implementation.
At the architectural level, leakage reduction techniques include
gating-off the supply voltage (VDD) of idle memory sections or
putting less frequently used sections into drowsy standby mode.
38 COPYRIGHT UNIVERSITY OF CALIFORNIA, BERKELEY, STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
When designing for an ultra low standby VDD, reliability of the
SRAM data retention at idle mode is the top design concern.
Besides process variations, the other factors that may disturb
the memory state preservation are mainly noises on the standby
supply rail and radiation particles. In this scheme, power supply
noise is mostly caused by the output voltage ripple of the
SC converter. Thus, an appropriate noise margin needs to be
provided in order to achieve the desired reliability. Assigning
a guard band of 100 mV above DRV for standby VDD gives
about 50 mV SNM in an SRAM cell. With the 20 mV peak-
to-peak ripple on the SC converter output, a guard band of 100
mV provides worst case SNM of 45 mV, which is sufficient
for state preservation. In comparison to power supply noise,
the radiation particle events pose a more serious hazard. With
parasitic capacitance at the data storage node of about 1fF in
a 0.13 um technology SRAMcell, the critical charge (Qcritical)
for a 1 V VDD is simulated to be approximately 3 fC. This is
the minimum amount of charge injection on the storage node
needed to disrupt the state preserved in this cell. For a reduced
VDD at 100 mV above the DRV, Qcritical is reduced to 0.5 fC.
Considering the danger of data loss (i.e., soft error), a larger
guard band is needed. Other options to ensure reliable state
preservation include additional storage capacitance [34] or
implemenation of error-correction schemes.
For a dual supply scheme, other design considerations
include the operation delay overhead due to the power
switch resistance, memory wake up delay, and the power
penalty during mode transition. Targeted for ultra low power
applications, the system requirements of this design are much
more stringent with respect to power than performance [22].
In this context, the concern of the operation delay overhead
is not crucial. A 200-um-wide PMOS power switch with 30
conducting resistance is used to connect the memory module
to a 1 V active mode supply voltage. With the same switch, the
memory wake up time is simulated to be within 10 ns, which
is typically a small fraction of the system cycle time in battery-
operated applications [35].
The wake up power penalty incurred during switching from the
standby mode to the full-VDD mode determines the minimum
standby time for the scheme if net power savings over one standby
period is to be achieved. This break-even time is an important
system-design parameter, as it helps the power control algorithm
to decide when a power-down would be beneficial. With the
parasitic capacitance information attained from the process
model, the minimum standby time in this design is estimated
to be several tens of microseconds, which is much shorter than
the typical system idle time in a battery-supported system.
Layout of the 0.13 um SRAM test chip is shown in Fig. 1d.
The two main components are a 4 KB SRAM module and SC
converter. This memory is an IP module with no modifications
from its original design.
0
100 200
DRV (mV)
H
i
s
t
o
g
r
a
m

o
f

3
2
K

S
R
A
M

c
e
l
l
s
300 400
1000
2000
3000
4000
5000
6000
FIGURE 3: MEASURED DRV DISTRIBUTION OF A 4 KB SRAM CHIP,
2005 ELSEVIER.
Using automatic measurement with a logic analyzer, the DRV
of all 32 K SRAM cells on one test chip was measured. Fig.
3 shows the distribution of the 32 K measurement results.
The DRV values range from 60 to 390 mV with the mean
value around 122 mV. Such a wide range of DRV uncertainty
reflects the existence of considerable process variations during
fabrication. Due to global variations, the lower end of measured
DRV is slightly lower than the 78 mV ideal DRV, assuming
PICO RADIO: FROM VISION TO REALITY
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
39
perfect process matching. As a result of large process variation,
the long DRV tail at the higher end reduces the leakage reduction
achievable by minimizing the SRAM standby VDD.
To improve the gains in leakage power, more advanced
techniques, such as error tolerant schemes, are required to
cope with this situation. Temperature dependency of DRV was
investigated experimentally. When the test chip was heated up
to 100C, a 10 mV increase in DRV is observed. The proposed
analytical DRV model not only predicts the ideal DRV values
but also fully captures the impact of process and temperature
variations.
Thus, it can serve as a convenient base for further design
optimizations. Furthermore, Fig. 4 shows the first presented
spatial distribution plot of DRV on the measured SRAM
chip. From the plot, it can be observed that the on-chip DRV
distribution is the combination of random within-die mismatches
and systematic deviations on the boundaries of SRAM sub-array
blocks. The pattern of SRAM DRV spatial distribution can be
exploited in the future work of designing effective error tolerant
scheme for even more aggressive SRAM voltage scaling.
Leakage measurement result of the 4 KB SRAM is shown in Fig.
14. The leakage current increases substantially when VDD is
high. This phenomenon reflects the impact of process variations
on SRAM leakage, more specifically the fluctuations in channel
length and Vth. For short channel transistors, a drain-induced
barrier-lowering (DIBL) effect causes Vth degradation, resulting
in even higher leakage in high VDD conditions. The shaded area
in Fig. 5 indicates the range of measured DRV (60390 mV).
Although the memory states can be preserved at sub-400 mV
VDD, adding an extra guard band of 100 mV to the standby VDD
enhances the noise robustness of state preservation as discussed in
Section 3.1. With the resulting 490 mV standby VDD, the SRAM
leakage current can still be reduced by over 70%. Subsequently,
the leakage power, as the product of VDD and leakage
current, is reduced by about 85% compared to 1V operation.
The dual rail scheme is shown to be fully functional through the
DRV measurements. With a 10 MHz switch control signal, the
SC converter generates the standby VDD with less than 20 mV
peak-to-peak ripple. A wake up time of 10 ns is observed during
mode transition, while the sleep time spans around 10 us. The
delay overhead in the SRAM read operation is measured to be
about 2X, which is reasonable for an ultra low-power application
where the system clock period is typically 10 times the operation
cycle of a low leakage SRAM.
Transistor sizing is an important factor that determines the
DRV of an SRAM cell. While sizing has long been an effective
technique in conventional power and speed optimization, taking
FIGURE 4: DRV SPATIAL DISTRIBUTION OF A 4-KBYTE SRAM CHIP,
2005 ELSEVIER.
Supply Voltage (V)
DRV + GB
max
Measuerd
DRV range
4

K
B

S
R
A
M
L
e
a
k
a
g
e

C
u
r
r
e
n
t

(


A
)
0
10
20
30
40
50
60
1 0,8 0,6 0,4 0,2 0
FIGURE 5: MEASURED SRAM LEAKAGE CURRENT, 2005 ELSEVIER.
40 COPYRIGHT UNIVERSITY OF CALIFORNIA, BERKELEY, STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
DRV into account is important for future ULP SRAM designs.
In conventional performance-optimized SRAM cells, the pull-
down NMOS devices are sized about 2 times larger than the
PMOS devices. These NMOS transistors are also typically
designed with a smaller L to minimize the cell area. Although
providing good stability at high-VDD, this imbalance in the
pull-up and pull-down leakage paths leads to exacerbated VTC
deterioration at low VDD and degrades DRV. The minimum L of
NMOS transistors is also highly sensitive to process variations,
which leads to an increase of DRV. Therefore, it is of interest to
investigate the impact of each of the sizing variables on DRV.
area), which is sensitive to the process variations. Lp has a much
smaller impact on DRV due to its larger nominal value.
In summary, the following are the guidelines for DRV-aware
ULV SRAM cell optimization for applications in which the
read/write performance is not a major concern:
Increasing Ln SRAM cell reduces DRV most effectively,
followed by increasing p.
Reducing both n and Lp improves DRV, but the improvement
space is very limited.
The sizing of access transistors has a negligible effect
on DRV.
As an example of power and area tradeoff, Fig. 7a plots the
leakage power and SRAM cell area when tuning Ln and p for
the minimized DRV. In this analysis, the SRAM cell transistor
area is simply modeled as the sum of the transistor gate areas.
A 30% increase in SRAM cell transistor area brings about 30
mV reduction in DRV and almost 70% additional leakage power
savings. In Fig. 7b, Ln first exploits the increase of area budget
due to its effectiveness in reducing DRV. The effectiveness of
increasing Ln is utilized until Ln is about 25% larger than the
nominal value, where its impact on DRV drops and, from this
point on, p can be used to further reduce DRV under given area
constraint. Although an increase of p continuously reduces DRV,
no more savings in leakage power is attained due to the positive
correlation between PMOS leakage and its sizing ratio p.
4. TRANSCEIVER
The transceiver specifications for wireless sensor networks
demand extremely low power consumption, very high levels of
integration, and a fast turn-on time for both the transmitter and
receiver [9]. These requirements ensure that the transceiver and
its energy source will be small and cheap, enabling deployment
in quantities sufficient to create a truly ambient intelligent
environment. Due to the heavily duty cycled nature of these
systems, long transceiver turn-on times dramatically degrade the
global efficiency of the system. The level of radio integration
Sizing Ratio
D
R
V

(
m
V
)
130
140
150
160
170
180
1 1,5 2,0 0,5
K

p
K

n
K

a
K
K
K
Lp
Ln
La
FIGURE 6: DRV AS A FUNCTION OF SIZING PARAMETERS AND L,
2005 ELSEVIER.
Fig. 6 shows simulated DRV over the sizing variables i and
Li for different transistors. For each curve, all the other sizing
variables are fixed at their nominal values from an industrial
SRAM cell design. These simulations assume 3 local process
variations in SRAM transistor channel length and Vth. The X-axis
of the plot is the sizing ratio that each variable (, L) is scaled by.
Parameter i represents (W/L) ratio of the transistor: the pull-up
PMOS (p), pull-down NMOS (n), and access transistors (a).
The range of sizing ratios in Fig. 6 is constrained by i >1 and
Li > Lmin. From the plot in Fig. 6, it can be observed that DRV
can be reduced only by increasing p or Ln, with the impact of
Ln being much stronger. This strong DRV dependence on Ln is
the result of its small nominal value (chosen for minimum cell
PICO RADIO: FROM VISION TO REALITY
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
41
220
240
0,2
1.5
Area / Area
nom
P
l
e
a
K

/

P
l
e
a
K
n
o
m
l
e
a
K

l
e
a
K
D
R
V
m
i
n

+
1
0
0
m
V

G
B
m
i
n

1
0,4
0,6
0,8
1
260
280 (a)
FIGURE 7: DRV-AWARE
SRAM OPTIMIZATION
WITH P AND LN.
(A) AREA TRADEOFF
WITH DRV AND LEAKAGE
POWER; (B) OPTIMIZED
P AND LN;
2005 ELSEVIER.
must be high, eliminating the possibility of some standard radio
components such as quartz crystals.
To address these stringent requirements, a new and emerging
technology was investigated: RF-MEMS (RF Micro-Electro
Mechanical Systems) resonators. Recently, much progress
has been made on GHz-range MEMS resonators, typically
for use in bandpass filters and duplexers [10][11]. Since these
resonators exhibit quality (Q) factors greater than 1000, these
devices have the potential to facilitate the design of low power
RF transceivers. The presence of an RF frequency reference
can eliminate the need for quartz crystals in the system, greatly
increasing the level of integration while decreasing the cost.
For this implementation, Agilent Technologies Thin-Film Bulk
Acoustic Wave (FBAR) resonators [11], which exhibit resonant
frequencies at approximately 1.9GHz, were used,. This relatively
high carrier frequency allows the use of small chip antennas,
further increasing system integration [12].
4.1 Transmitter
In a typical sensor network, the transmitter sends out sporadic
bursts of short data packets to neighboring sensor nodes (<
10m). For a receiver sensitivity of 60dBm and indoor multi-
path fading conditions, the application of the Friis wave
propagation equation [14] shows that a transmit power of about
0dBm is required. The transmitter must exhibit fast turn-on
time and high efficiency. The transmitter architecture shown in
Fig. 8 is well-suited for these requirements. The MEMS-based
oscillator is directly modulated by the baseband data, and the
power amplifier efficiently boosts the RF signal power. Direct
modulation eliminates power hungry mixers and PLLs. Multiple
channels can be implemented by tuning the oscillation frequency
or by adding oscillators/ transmit chains in parallel.
In this implementation, carrier generation is accomplished with
an oscillator co-designed with an FBAR resonator [8]. This
architecture provides a low power 1.9GHz frequency reference
without a PLL and quartz crystal reference. In addition, the
start-up time is approximately 1s, enabling the oscillator to
be power cycled between transmitted bits, thus increasing the
overall efficiency of the transmitter.
The schematic of the low-power amplifier is shown in Fig. 9.
Maximizing the drain efficiency requires the voltage swing Vds to
PA
Osc
Matching
Network
Baseband
Data
FIGURE 8: DIRECT MODULATION TRANSMITTER ARCHITECTURE,
2004 IEEE.
1
1,5
1.5
Area / Area
nom
O
p
t
i
m
i
z
e
d


p
,
L
n
1
2
p
Ln
(b)
42 COPYRIGHT UNIVERSITY OF CALIFORNIA, BERKELEY, STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
to demonstrate the effectiveness of RF-MEMS resonators in
low power transceivers. The antenna feeds a 50 impedance
presented by the LNA. The LNA drives a tuned load consisting
of two channel select amplifiers (CSAs). Although two channels
were used in this implementation, this architecture is scalable
to larger numbers of channels. Each CSA is tuned by an FBAR
resonator, which performs receiver channel selection. The
CSAs drive an envelope detector, which acts as a self-mixer to
perform signal downconversion. Baseband buffers are included
to drive test instrumentation. Since the frequency stabilization is
performed entirely by the MEMS resonators, no quartz crystals
are used in this receiver architecture. The absence of a phase-
locked loop (PLL) ensures a much faster receiver start-up time
than that of a traditional radio.
The two-channel embodiment displays flexibility in terms of
modulation schemes: the receiver can detect two unique OOK
data streams at two carrier frequencies or it can detect FSK.
For dense wireless sensor networks, it is anticipated that two
separate OOK channels will be used, with one reserved for
beaconing. Changing between these two modulation schemes
can be accomplished with no receiver modifications and can be
performed dynamically in either the analog or digital baseband
detection circuitry.
be maximized. For 1.2mW output power with a 1.2V supply and
100mV knee voltage, a 500 load resistance at the drain of M2
is required. To achieve this, the 50 antenna (RL) is transformed
to the required impedance using C1 and C2. Inductor L1 tunes out
the transformed capacitance and the parasitic drain capacitance.
Capacitive transformers are preferred over LC matching networks
or inductive transformers because on-chip capacitors have much
higher Q (>50) than on-chip inductors (Q of 510), resulting in
less loss. In the current prototype, the required L1 is 1.2nH and is
implemented using an off-chip inductor. However, a short bond
wire or an on-chip inductor can be used for a fully integrated
solution. A cascode transistor M2 ensures that the drain voltage
does not exceed the low gate breakdown voltage for deep sub-
micron CMOS. Cascoding increases the isolation between the
input and output and improves the efficiency by boosting the
drain resistance of M2.
4.2 Receiver
The receiver block diagram is shown in Fig. 10. The two-
channel tuned radio frequency (TRF) architecture was chosen
bias
V
in
V
1
M
2
M
2
C
L
R
1
L
0
V
dd
V
ds
V
ds
I
ds
V
1
C
FIGURE 9: SCHEMATIC OF THE LOW POWER AMPLIFIER, 2004 IEEE.
CSA BB
Envelope
Detector
CSA BB
Envelope
Detector
LNA Match
FIGURE 10: RECEIVER ARCHITECTURE, 2004 IEEE.
antenna
V
bias
V
out
V
d
L
g
L
p C m C
FIGURE 11: LNA SCHEMATIC, 2004 IEEE.
The low noise amplifier must achieve high gain and moderate
noise figure while simultaneously providing input impedance
matching. In order to satisfy these requirements, the architecture
PICO RADIO: FROM VISION TO REALITY
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
43
shown in Fig. 11 was implemented. The nonquasi static (NQS)
gate resistance of the input transistor was transformed via a
passive matching network to match the 50 source resistance.
The NQS gate resistance is inversely proportional to the device
transconductance and is especially pronounced at low current
levels [13]. An off-chip matching inductor (Lg) was used for
maximum flexibility in the prototyping phase, but it could easily
be integrated on-chip. The parasitic capacitance Cp results from
ESD diodes, bond pads, and board and layout traces. For a given
bias current and fixed Cp, an optimal device size exists that
maximizes the cut-off frequency of the transistor (ft).
The output tank inductor was implemented on chip by shunting
the two top-level metal layers in order to increase the quality
factor. It was characterized using a 3D electromagnetic solver
and experimentally verified.
The channel select amplifier (Fig. 12) provides high RF gain
to overcome the high noise figure of the detector as well as
to interface the electrical signal with the acoustic resonator in
order to perform high-Q filtering of the signal. The amplifier
must exhibit high RF gain with a low power consumption while
limiting the extent to which the resonator is de-tuned.
The amplifier consists of an NMOS cascode transconductance
stage and a tuned load. The amplifier input capacitance is
absorbed into the LNA load. A cascode transconductor structure
is used to increase reverse isolation, thereby ensuring amplifier
stability. The tuned load consists of an FBAR resonator to
perform channel selection and an inductive PMOS-RC structure,
which stabilizes the low-frequency bias point of the amplifier.
The loaded quality factor of the tuned load is approximately 600,
yielding an RF bandwidth of 3MHz.
The on-chip baseband circuitry downconverts the OOK or
FSK modulated data to DC, performs filtering, and buffers the
signal for driving subsequent instrumentation or circuitry. The
first baseband block, shown in Fig. 13, is a non-linear low pass
filter, or envelope detector. This circuit, composed of MOSFETs
biased in the deep subthreshold regime, performs a self-mixing
operation on the RF signal that has been filtered by the channel-
select amplifiers.
The low-pass filter has a 300kHz cutoff frequency that attenuates
all fundamental tones passing through the receive chain. A replica
envelope detector provides a reference DC level, producing a
pseudo-differential baseband output. The current consumption of
each detector is 200nA.
Each channel in the receive chain also contains a low power
buffer to drive off-chip instrumentation. The buffers are capable
of driving a 20pF off-chip load while consuming 50A. On-chip
threshold-referenced bias circuits provide a moderate level of
power-supply independence.
out
V
FBAR
LNA
V
( )
FIGURE 12: CHANNEL
SELECT AMPLIFIER
SCHEMATIC,
2004 IEEE.
rf
V
out
- +
( )
V
replica
V
FIGURE 13: ENVELOPE
DETECTOR SCHEMATIC,
2004 IEEE.
44 COPYRIGHT UNIVERSITY OF CALIFORNIA, BERKELEY, STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
5. WAKE-UP RECEIVER
In order to enable fully asynchronous communication between
nodes, even under heavy duty cycling, each transceiver should
be able to monitor the channel continuously and wake up only
when data is being received. To enable this behavior, a carrier
sense receiver is needed to monitor the channel for activity.
Since this receiver is always listening, its power consumption
should be on the order of 100W.
A digitally controlled oscillator to be used as the frequency
synthesizer of such a carrier sense receiver has been designed.
By employing digital tuning, the oscillator may be periodically
calibrated to an accurate reference such as a crystal or MEMS-
based oscillator [17]. The calibration aligns the oscillator with
the system reference, performing channel selection and tuning
for the carrier sense receiver. Between calibration cycles, the
accompanying tuning loop circuitry is turned off to save power,
and the oscillator runs free until the next tuning cycle. Re-
calibration could occur each time the main transceiver turns on
to send and receive data.
The frequency tuning specifications for the digitally controlled
oscillator (DCO) are derived from the intended application with
periodic calibration, where the MEMS reference tolerance is
500ppm [17]. Accordingly, the target frequency resolution is set
at 500kHz, which translates to approximately 1fF of switched
capacitance at 1.9GHz. In addition, the total tuning range is
chosen to be 200MHz (10%) in order to cover uncertainty in the
tank inductance value and other process variations. In order to
minimize the DCO power consumption, two major techniques
have been utilized. At the circuit level, the DCO core transistors
are designed to run in weak inversion, where more device
transconductance is available for a given bias current. The
circuit is also designed to run on a supply voltage well below
the technologys nominal value to further minimize the power
consumption. Recently, there has been interest in operating
synthesizer VCOs at low supply voltages. In [18], transformers
were used to enable the VCO internal nodes to swing above Vdd
and below ground to maximize voltage headroom. In the design
presented here, subthreshold device operation simultaneously
enables low current and low voltage operation while still
achieving the desired performance.
The use of subthreshold transistor biasing has been successfully
employed in low power analog circuits for many years in order to
take advantage of the high transconductance efficiency available
in this regime [19]. Unfortunately, weak inversion operation
usually results in large devices with substantial parasitics, so this
technique has traditionally not been attractive for RF designs.
10
-4
Inversion Coefficent (IC)
Center of Moderate Inversion
(IC=1)
10
-2
10
0
g
m
/
I
d
f
t
(
H
z
)
10
2
10
6
10
-1
10
0
10
1
10
2
10
7
10
8
10
9
10
10
10
11
FIGURE 14: TRANSCONDUCTANCE EFFICIENCY AND FT FOR 0.13 M CMOS,
2005 IEEE.
DD V
b
M
1
M
2
M
1-5
5 5
-
D
6-9
4
D
1-5
+
D
lsb2
C
lsb1
C
msb
C
L L
bias
V
FIGURE 15: SIMPLIFIED SCHEMATIC OF OSCILLATOR CORE, 2005 IEEE.
The conflicting trends of transconductance efficiency and
device fT are illustrated in Fig. 14, where simulated values of
PICO RADIO: FROM VISION TO REALITY
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
45
these parameters are plotted across the entire range of device
operation for a 0.13m technology. The parameter IC represents
normalized current density with IC=1 indicating the center of
moderate inversion. In this scaled technology, peak ft is close to
100GHz, meaning that transistors no longer need to be biased for
optimal ft when low power consumption is one of the chief design
goals. The weak inversion regime (IC 0.1) may then be used to
take advantage of high transconductance efficiency, which may
be more than a factor of two higher than in strong inversion.
The simplified schematic of the oscillator core is shown in
Fig. 15. In order to reduce power consumption, the oscillator
is designed to operate on a 0.5V supply, exploiting the trend
towards low supply voltages in scaled CMOS technologies. A
differential NMOS-only topology is adopted because it provides
higher output swing for low supply voltages. With only two
stacked transistors, the chosen oscillator topology is able to
operate on very low supply voltages. Although the low supply
reduces the maximum possible voltage swing (2Vdd), this is not
a large disadvantage because the ultra low power oscillator is
designed to run in the current-limited regime. Devices M1 and
M2 are sized with large W/L =300m/0.13m for a design bias
current of 100A in each device. At this bias point, the device
inversion coefficient is approximately 0.1, providing gm/Id 25,
while still yielding manageable parasitics. An additional benefit
of subthreshold operation for this design is that Vdsat and Vgs
are both much lower than they would be in strong inversion,
which conserves valuable voltage headroom. The current source
PMOS transistor Mb provides bias current to the core and is sized
to operate in saturation for supply voltages of 0.5V and below.
6. PLANAR INVERTED-F
ANTENNA (PIFA)
The antenna of the Quark node is a planar inverted-F antenna
(PIFA). The PIFA structure is low-profile, reliable, cheap, and
has a nearly omni-directional radiation pattern. This type of
antenna is printed directly on the circuit board during the copper
etching process. The circuit board area for the antenna is small
and therefore adds minimal cost to the node. Like most single
ended antennas, the PIFA needs a ground plane to which it can
couple, in this case, the PCB used for the rest of the node).
In a wireless network, a transceiver must be able to receive and
transmit a signal in any direction because it is not aware of the
location of other nodes. Because the PIFA is a type of a monopole
antenna, it has a relatively omni-directional (donut shaped)
radiation pattern that enables it to communicate with other PIFA.

Additionally, the PIFA avoids the use of inefficient matching
networks. In low-voltage IC design, the efficiency of a power
amplifier increases with output resistance typically above
50. Most of-the-shelf antennas have a radiation resistance of
50, which requires the use of a matching network. Matching
networks comprised of discrete parts are expensive, and on-chip
matching networks are inefficient. A PIFA, however, allows the
designer to choose almost any radiation resistance (preferably
between 50 to 500) by simply changing the geometrical
dimensions. Optimization of the geometrical dimensions is
performed using electromagnetic optimization software.
Although the PIFA is a narrowband antenna, it still provides
more than 250MHz of bandwidth at 1.9GHz with a Voltage
Standing Wave Ratio (VSWR) < 2. This is more than sufficient
for the low data rate, narrowband communication of a PicoRadio
wireless network.
7. POWER TRAIN
The PicoRadio node must be self-powered since physical access
to nodes in a deployed network may be difficult and the sheer
number of nodes makes changing batteries or manual refueling
unrealistic. At the present level of integration for Quark, the best
option for powering the board is solar scavenging augmented by
rechargeable batteries. Solar scavenging provides more energy
per unit area than most alternatives, and the cells are readily
available commercially. For the Quark node, there are three
main challenges: utilizing solar power with strict array size
46 COPYRIGHT UNIVERSITY OF CALIFORNIA, BERKELEY, STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
restrictions, storing energy, and regulating voltage to the Quark
circuitry. Since minimizing board size is a primary goal, any
viable solution must be small in terms of area or volume.
Obtaining sufficient power from solar energy is the first
challenge. Solar panels are characterized by open circuit
voltage and short circuit current in full sunlight. However,
most nodes will be deployed indoors where the cell
efficiency drops off dramatically under fluorescent lighting.
For instance, a (2x2.5)cm
2

single crystal cell has an open circuit
voltage of 530mV and a short circuit current of 80mA when
measured outside on a typical day. Under standard fluorescent
lighting, however, this same cell yields half the voltage and a
small fraction of the current 240mV, 360A, respectively. To
determine how much power is available between open and
short circuit conditions, a series of measurements were taken
under fluorescent lighting conditions. Peak power for a single
crystal cell is 32W (125mV, 250A), but, when three additional
cells are wired in series, power does not quadruple as might be
expected. Instead, peak power is only 2.3 times the single cell
case (74W). For this test, the combined open circuit voltage is
700mV, and the short circuit current is 290A.
Due to the use of several off-the-shelf components, the power
budget of the current Quark version is in the single-digit
milliwatt range. An array sufficient to power a milliwatt device
or to produce supply voltages directly usable by the Quark
circuitry would be unacceptably large. Thus, solar power is
insufficient to power the node in typical indoor conditions.
The second challenge is the efficient storage of energy in a
rechargeable battery. While there are multiple efforts to develop
thin film battery technologies, only commercially available parts
were considered for this design. Coin cells are desirable from
a size standpoint but are typically primary (non-rechargeable)
technologies. Lithium-ion coin cells that are attractive for their
nominal voltage (3.6V)are available but have complex charging
requirements. Nickel Metal Hydride (NiMH) is also available
in a variety of capacities. These batteries can be trickle-charged
at about 1/10 capacity without damage but only produce 1.2V.
Therefore, the choice of NiMH requires a series connection or
boost regulation to reach the voltage requirement.
The third power train challenge is the conversion of the battery
voltage into DC levels appropriate for the Quark circuitry. The
original Quark design called for three supply voltages: 1.8V for the
Charm I/O interface and most support chips, 1.2V for the Strange
chip, and 1.0V for the core Charm logic. There are a number of
tradeoffs between nominal battery voltage, battery charge/discharge
characteristics, minimum input voltage for a regulator chip,
efficiency of regulation, and the complexity and size of the circuit.
There are three generally available methods for regulation:
switched capacitor charge pump, linear regulation, and switching
DC-DC regulation. The charge pump is the simplest in terms of
external components but is limited mainly to unregulated voltage
inverters or doublers. Linear regulators are also relatively simple
but efficiency is related to V
out
/V
in
. Switching regulators (buck
converters) are most efficient within their design output current
range, but the circuit board layout is relatively large, and the
efficiency drops sharply at low current levels. Unfortunately,
there are no existing off-the-shelf power supply components
specifically designed for a low milliwatt assembly with severely
limited available power.
One way to overcome input voltage minimums (enabling use
of a single 1.2V battery) is first to increase the input voltage
using a switching boost regulator. However, efficiency losses
are multiplicative, so if the boost stage and step-down stages
are each 80% efficient, the combined efficiency will be 64%.
In practice, these numbers are optimistic, as a test circuit with a
boost-buck topology that converted 1V to 2.75V to 1.2V was less
than 50% efficient at 1mA.
Another option is to tolerate an unregulated supply and source
power directly from the battery. This is reasonable for some
PICO RADIO: FROM VISION TO REALITY
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
47
devices; for instance, the custom Strange and Charm chips are
tolerant to supply variations. Variation on the battery voltage,
however, can violate the V
in
specifications for some of the
commercial parts driven by the Charm, which usually specify
a supply voltage above the 1.2V nominal of a single NiMH
cell. Two NiMH cells in series raise the battery voltage to 2.4V,
which is higher than nominal for devices on the Quark board but
generally within operating range.
Given the above analysis, a solar module with five single-crystal
cells in series and two 15mAh NiMH coin batteries for power
collection and storage was selected. The module produces
sufficient voltage in full or partial sun to charge the batteries at
an acceptable rate, and, with the addition of a voltage doubling
charge pump, can charge at lower light intensities (cloudy days
or indoors) or at a higher rate. The Charm core and the Strange
radio are powered directly from a single NiMH battery. The result
is a power train that will perform with reasonable efficiency and
recharge the batteries when there is enough available light.
8. CLOCK
The Quark node uses three primary clocks: a digital logic
system clock (16MHz), a sampling baseband clock (1MHz),
and an idle mode clock (160kHz). Currently, the three clocks
are derived from a single external crystal oscillator running at
16MHz. The accuracy of the idle and system clocks are not
tightly constrained, but the accuracy of the baseband clock must
be tight enough to hold synchronization during the reception of
a complete packet (currently 50ppm). Derivation of the baseband
clock from the main clock thus overconstrains the accuracy of
the other two clocks. If this accuracy requirement is relaxed,
alternative lower-power clock generation solutions such as an
integrated micro electro-mechanical systems (MEMS) oscillator
or a simple free-running inverter ring can be considered.
The baseband clock may still require an external crystal to
achieve the accuracy requirements, but it could be turned off
entirely when the radio is not in use. This same idea can be
extended to disable the system clock, since only the low speed
idle clock is required during idle mode. Although we currently
save significant power in the Charm by clock gating, separating
the clocks into lower-power sources and controlling them
individually will reduce total power.
9. POWER CONSUMPTION
BREAKDOWN
Clearly, further reduction of the nodes power requirements will
increase achievable duty cycles. Once the power requirements
are reduced below the amount provided by the solar cells, the
number or size of the cells can be reduced.
In Fig. 16, the power consumption breakdown of the entire node
during reception (a) and idle mode (b) is shown. An interesting
observation is that the custom chips together consume only about
25% of total power when receiving (transmit is comparable) and
only 3% with the radio off. The four largest power sinks are
the inefficient voltage regulators, the external clock oscillator,
miscellaneous external logic, and the external analog components
in the baseband block.
Oscillator
15%
Analog
BB
29%
Supplies
16%
Strange
20%
Charm
6%
Misc
logic
14%
(a)
FIGURE 16: NODE POWER CONSUMPTION BREAKDOWN
IN RX (A) AND WITH RADIO OFF (B), 2004 IEEE.
Oscillator
42%
Supplies
17%
Charm
3%
Misc
logic
38%
(b)
48 COPYRIGHT UNIVERSITY OF CALIFORNIA, BERKELEY, STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
The commercial voltage regulators are inefficient because they
are not designed for low-voltage, low-current operation. Custom
power regulation circuitry would address this problem.
All of the miscellaneous logic can be integrated into either the
Charm or Strange chip, where power consumption would be
reduced by perhaps two orders of magnitude.
Lastly, almost 30% of the node power in Rx mode is dissipated
in the analog baseband circuitry. Although disabled when the
node is not receiving, it is still a significant drain on the power
train when operating. In addition, it consumes nearly 10% of
the board area. Preliminary work shows that integrating this
circuitry into the RF chip will reduce power consumption by a
factor of twenty.
10. CONCLUSION
This article summarizes the research and design efforts of
the BWRC towards the realization of the PicoRadio vision.
The confluence of progress made in the fields of protocol
stack algorithm design, energy train, node locationing, digital
processing, and RF communication circuitry has been crucial
to the results obtained. These technologies were integrated into
a self-contained energy scavenging sensor node (Quark Node).
Two custom CMOS chips (manufactured by STMicroelectronics)
have been designed and implemented for the Quark node, and a
custom PCB with integrated PIFA antenna has been developed
to reduce the size of the node.
The main contributions that the system-level integration of the
Quark node has given to the field of WSNs are a custom protocol
stack using energy-efficient algorithms implemented mostly in
low-power custom ASICs and an energy train designed to allow
the node to be self-contained with batteries recharged by solar
energy. Future versions of the node would decrease the node
size and increase the duty cycle by integrating custom baseband
circuitry, custom circuits for the power train, and MEMS-based
clock references.
A third chip in 0.13 m CMOS (STMicroelectronics) implements
a 100W digitally controlled oscillator suitable for reactive
receivers. Bondwire inductors are employed for high Q, and no
external components are used. Subthreshold device operation is
shown to be a feasible method of increasing transconductance
efficiency at RF.
The ever increasing problem of SRAM power leakage has
been extensively addressed by exploring the limit of SRAM
data preservation under ultra-low standby VDD. A commercial
SRAM module (from STMicroelectronics) with high-Vth
process is shown to be capable of sub-400 mV standby data
preservation. With additional 100 mV guard band to account for
power supply ripple and cosmic particles, leakage power savings
of more than 85% can be achieved with an SRAM module under
490 mV standby VDD, compared to 1V active mode. The DRV
is observed to be a strong function of process variation and
also SRAM cell sizing. With proper sizing optimization, an
additional 70% leakage power savings can be achieved with
only a 30% SRAM cell transistor area increase. Besides sizing,
more variables are being investigated for their impacts on ULP
SRAM cell design. Such variables include the transistor Vth and
body bias voltages. With the control of body bias voltages, the
SRAM DRV and leakage current can be dynamically adjusted in
different operation modes.
REFERENCES
[1] Reprinted from Huifang Qin, Yu Cao, Dejan Markovic,
Andrei Vladimirescu, Jan Rabaey, STANDBY SUPPLY
VOLTAGE MINIMIZATION FOR DEEP SUB-MICRON SRAM,
Microelectronics Journal, Vol. 36, 2005, pp. 789-800,
Elsevier, with permission from Elsevier.
[2] Nathan M. Pletcher and Jan M. Rabaey , A 100 W, 1.9
GHZ OSCILLATOR WITH FULLY DIGITAL FREQUENCY
TUNING, IEEE Proceedings of ESSCIRC, Grenoble,
France, 2005.
PICO RADIO: FROM VISION TO REALITY
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
49
[3] M. Sheets, B. Otis, F. Burgardt, J. Ammer, T. Karalar,
P. Monat, and J. Rabaey, A (6X3)CM2 SELF-CONTAINED
ENERGY-SCAVENGING WIRELESS SENSOR NETWORK NODE,
IEEE Wireless Personal Multimedia Communications
(WPMC), 2004, Abano Terme, Italy
[4] INFORMATION TECHNOLOGY OPEN SYSTEMS
INTERCONNECTION BASIC REFERENCE MODEL,
International Organization for Standardization (ISO).
Standard number ISO/IEC 7498-1:1994, 1994.
[5] C. Intanagonwiwat, R. Govindan and D. Estrin,
DIRECTED DIFFUSION: A SCALABLE AND ROBUST
COMMUNICATION PARADIGM FOR SENSOR NETWORKS,
IEEE/ACM Mobicom, 2000, pp. 56-67.
[6] R. Shah, J. Rabaey. ENERGY AWARE ROUTING FOR LOW
ENERGY AD HOC SENSOR NETWORKS. 2002 IEEE Wireless
Communications and Networking Conference Record.
WCNC 2002 (Cat. No.02TH8609). IEEE. Part vol.1, 2002,
pp.350-5 vol.1. Piscataway, NJ, USA.
[7] C. Savarese, K. Langendoen, J. Rabaey, ROBUST
POSITIONING ALGORITHMS FOR DISTRIBUTED AD-HOC
WIRELESS SENSOR NETWORKS, USENIX Technical Annual
Conference, Monterey, CA, 2002.
[8] C. Savarese, J. Rabaey, J. Beutel, LOCATIONING IN
DISTRIBUTED AD-HOC WIRELESS SENSOR NETWORKS,
Int. Conf. on Acoustics, Speech, and Signal Proc.
(ICASSP), pp 2037-2040, Salt Lake City, UT, May 2001.
[9] H. Qin, Y. Cao, D. Markovic, A. Vladimirescu, J. Rabaey,
SRAM LEAKAGE SUPPRESSION BY MINIMIZING STANDBY
SUPPLY VOLTAGE, ISQED 2004.
[10] B. Otis, Y.H. Chee, R. Lu, N. Pletcher, J. Rabaey, AN
ULTRA-LOW POWER MEMS-BASED TWO-CHANNEL
TRANSCEIVER FOR WIRELESS SENSOR NETWORKS,
IEEE VLSI Circuits Symposium, 2004.
[11] B. Otis, J. Rabaey. A 300W 1.9GHZ CMOS
OSCILLATOR UTILIZING MICROMACHINED RESONATORS,
IEEE J. of Solid State Circuits, Vol 38, p. 1271-1274,
July, 2003.
[12] J. Rabaey, J. Ammer, T. Karalar, S. Li, B. Otis, M. Sheets,
T.Tuan, PICORADIOS FOR WIRELESS SENSOR NETWORKS:
THE NEXT CHALLENGE IN ULTRA-LOW POWER DESIGN,
IEEE ISSCC, pp. 200-1, Feb 2002.
[13] B. Bircumshaw et. al, THE RADIAL BULK ANNULAR
RESONATOR:TOWARDS A 50OHM RF MEMS FILTER, Tech.
Dig., 12th Int. Conf. on Solid State Sensors, Actuators
and Microsystems, Boston, pp. 875-878, June 8-12, 2003.
[14] R. Ruby et. al, ULTRA-MINIATURE HIGH-Q FILTERS AND
DUPLEXERS USING FBAR TECHNOLOGY, IEEE ISSCC,
pp.120-1, Feb. 2001.
[15] S. Roundy, B. Otis, Y.H. Chee, J. Rabaey, P. Wright,
A 1.9GHZ RF TRANSMIT BEACON USING ENVIRONMENTALLY
SCAVENGED ENERGY, Dig. IEEE Int. Symp. on Low Power
Elec. And Devices, Seoul, Korea, 2003.
[16] Y. Tsividis. Operation and Modeling of the MOS
Transistor. McGraw-Hill, Boston, 1998.
[17] C. A. Balanis, Antenna Theory Analysis and Design,
2nd ed. New York, Wiley, 1997.
[18] P. Choi, et. Al, AN EXPERIMENTAL COIN-SIZED RADIO FOR
EXTREMELY LOW POWER WPAN (802.15.4) APPLICATION
AT 2.4 GHZ, IEEE ISSCC, pp. 92-3, Feb 2003.
[19] T. Melly, et. al, AN ULTRALOW POWER UHF TRANSCEIVER
INTEGRATED IN A STANDARD DIGITAL CMOS PROCESS:
TRANSMITTER, IEEE J. Solid State Circuits, vol. 36, no. 3,
Mar 2001.
[20] B. Otis, Y. Chee, and J. Rabaey, A 400W-RX,
1.6MW-TX SUPER-REGENERATIVE TRANSCEIVER
50 COPYRIGHT UNIVERSITY OF CALIFORNIA, BERKELEY, STMICROELECTRONICS, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
FOR WIRELESS SENSOR NETWORKS, in IEEE ISSCC
Digest of Technical Papers, Feb. 2005, pp. 396397.
[21] K. Kwok and H. Luong, ULTRA-LOW-VOLTAGE HIGH
PERFORMANCE CMOS VCOS USING TRANSFORMER
FEEDBACK, IEEE J. Solid-State Circuits,
vol. 40, no. 3, Mar. 2005.
[22] E. Vittoz and J. Fellrath, CMOS ANALOG INTEGRATED
CIRCUITS BASED ON WEAK INVERSION OPERATION, IEEE
J.Solid-State Circuits, vol. SC-12, pp. 224231, June 1977.
[23] R. Staszewski, C.-M. Hung, D. Leipold, and P. Balsara,
A FIRST MULTIGIGAHERTZ DIGITALLY CONTROLLED
OSCILLATOR FOR WIRELESS APPLICATIONS, IEEE Trans.
Microwave Theory Tech., vol. 51, no. 11, Nov. 2003.
[24] J. Craninckx and M. Steyaert, A 1.8-GHZ CMOS
LOW-PHASE-NOISE VOLTAGE-CONTROLLED OSCILLATOR WITH
PRESCALER, IEEE J. Solid-State Circuits, vol. 30, no. 12,
Dec. 1995.
[25] S. Borkar, Design challenges of technology scaling,
IEEE Micro 19 (4) (1999) 2329.
[26] S. Manne, A. Klauser, D. Grunwald, PIPELINE GATING:
SPECULATION CONTROL FOR ENERGY REDUCTION,
International Symposium Computer Architecture
(1998) 132141.
[27] N. Kim, DROWSY INSTRUCTION CACHES: LEAKAGE POWER
REDUCTION USING DYNAMIC VOLTAGE SCALING AND CACHE
SUB-BANK PREDICTION, Proceedings of the 35th Annual
Intl Symposium Microarchitecture (MICRO-35),
IEEE CS Press, 2002. pp. 219230.
[28] M. Horiguchi, T. Sakata, K. Itoh, SWITCHED-SOURCE-
IMPEDANCE CMOS CIRCUIT FOR LOW STANDBY
SUBTHRESHOLD CURRENT GIGA-SCALE LSIS,
IEEE Journal of Solid-State
Circuits 28 (11) (1993) 11311135.
[29] K. Itoh, LOW VOLTAGE MEMORIES FOR POWER-AWARE
SYSTEMS, Proceedings of the ISLPED (2002) 16.
[30] H. Mizuno, T. Nagano, DRIVING SOURCE-LINE (DSL)
CELL ARCHITECTURE FOR SUB-1-V HIGH-SPEED LOW-POWER
APPLICATIONS, Digest of technical papers. Symposium
on VLSI circuits (1995) 2526.
[31] K. Itoh, A.R. Fridi, A. Bellaouar, M.I. Elmasry,
A DEEP SUB-V, SINGLE POWER-SUPPLY. SRAM CELL
WITH MULTI-VT, BOOSTED STORAGE NODE
AND DYNAMIC LOAD, Digest of technical papers.
Symposium on VLSI circuits (1996) 132133.
[32] H. Kawaguchi, et al., DYNAMIC LEAKAGE CUT-OFF
SCHEME FOR LOW VOLTAGE SRAMS,
Digest of technical papers. Symposium on VLSI circuits
(1998) 140141.
[33] S. Kaxiras, Z. Hu, M. Martonosi, CACHE DECAY:
EXPLOITING GENERATIONAL BEHAVIOR TO REDUCE CACHE
LEAKAGE POWER, Proceedings of the ISCA (2001)
240251.
[34] E. Seevinck, F.J. List, J. Lohstroh, STATIC-NOISE MARGIN
ANALYSIS OF MOS SRAM CELLS, IEEE Journal
of Solid-State Circuits SC-22 (5) (1987) 748754.
[35] J. Rabaey, A. Chandrakasan, B. Nikolic, DIGITAL
INTEGRATED CIRCUITS: A DESIGN PERSPECTIVE,
second ed., Prentice-Hall, 2002.
[36] J. Lohstroh, E. Seevinck, J.D. Groot, WORST-CASE
STATIC NOISE MARGIN CRITERIA FOR LOGIC CIRCUITS
AND THEIR MATHEMATICAL EQUIVALENCE,
IEEE Journal of Solid-State Circuits SC-18 (6)
(1983) 803807.
[37] C. Lage, et al., SOFT ERROR RATE AND STORED CHARGE
REQUIREMENTS IN ADVANCED HIGH-DENSITY SRAMS,
Proceedings of IEDM (1993) 821824.
PICO RADIO: FROM VISION TO REALITY
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
51
[38] M.J. Ammer, et al., A LOW-ENERGY CHIP-SET
FOR WIRELESS INTERCOM, Proceedings of DAC (2003).
[39] K.D.T. Ngo, R. Webster, STEADY-STATE ANALYSIS
AND DESIGN OF A SWITCHED-CAPACITOR DC-DC
CONVERTER, Proceedings of PESC (1992) 378385.
[40] Y. Cao, T. Sato, D. Sylvester,M. Orchansky, C. Hu,
NEW PARADIGM OF PREDICTIVE MOSFET AND
INTERCONNECT MODELING FOR EARLY CIRCUIT DESIGN,
Proceedings of CICC (2000) 201204.
CONTACT: ST.JOURNAL@ST.COM
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
52 COPYRIGHT STMICROELECTRONICS, 2006.
CUPOLA-SENSING:
AN INDOOR CLIMATE
MONITORING TEST CASE
In this paper, we describe a test case
for indoor climate monitoring deployed
in our laboratory. The test case represents
a significant class of Wireless Sensor
Network (WSN) monitoring applications.
It was developed to practice with data
gathering and in-network reprogramming
and to identify the basic requirements for
interfacing a WSN within Internet to enable
network analysis. The system architecture
includes a WSN physically implemented
with COTS (Components Off The Shelf),
a gateway device connecting the network
to Internet, and client tools running on PCs
connected via Ethernet to the gateway.
The client tools allow a user to observe
the WSN behavior at run time
for network debugging and to analyze
the sensing data stored in a database
integrated in the gateway. The SW platform
running on sensor nodes is built on top
of TinyOS, a widely used open source
operating system specifically designed
for sensor network applications.
1. INTRODUCTION
A large number of Wireless Sensor Networks have been
proposed in many different application scenarios, from
environmental monitoring to asset tracking to health monitoring.
WSN applications can be grouped into three main categories:
monitoring, tracking, and control, differing in terms of system
requirements such as lifetime, cost, ease of deployment, and
response time constraints. For example, in most control scenarios,
real-timeness is a fundamental constraint, while, in general, it is
not a significant requirement for the monitoring applications.
Therefore, different protocols and architecture solutions can be
defined to optimize the behavior of a WSN in each category.
Environmental monitoring applications are widely accepted as
a common driver for WSN research. A significant number of
deployments have been documented in the literature, describing
specifications and constraints for several indoor and outdoor
scenarios [1], [2], [3].
In general, environmental monitoring applications are
characterized by long lifetime requirements, low data rates,
and relatively static topology features. Environmental data
collection applications typically use hierarchical - often tree-
based - routing topologies, where each data packet is rooted at
Max Cortiana,
Andrea Labombarda,
Laura Vanzago
STMicroelectronics

CUPOLA-SENSING: AN INDOOR CLIMATE MONITORING TEST CASE
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
53
high-capability nodes storing sensed data (sink). With a tree-
based data collection, each node is responsible for forwarding
the data to all its descendants. Once the network is configured,
data is periodically transmitted from child nodes to parent nodes
up the tree structure until it reaches the sink. For many scenarios,
the interval between these transmissions can be on the order of
minutes. In addition to large sample intervals, environmental
monitoring does not have strict latency requirements, as,
in general, data is collected for future analysis and not for
real-time operation. In order to meet lifetime requirements,
each communication event is often required to be precisely
scheduled, allowing the sensor nodes to remain in sleep mode
most of the time, waking up only to transmit or receive data.
If the precise schedule is not met, the communication event will
fail, highlighting the role of distributed time synchronization
as an important service to be implemented in this type of
network [19].
Besides having the capability to collect data, an additional
advantage of a WSN is its ability to operate for extended
periods of time without human intervention. In addition, many
environmental monitoring applications make physical interaction
with the nodes for maintenance purposes unacceptable because
nodes can only be accessed with difficulty or because, within the
network, the nodes can be very numerous. End users must be
able to add or change the functionalities of a deployed network
remotely to fully utilize its capabilities and to dynamically adapt
to different environmental conditions or application requirements.
Therefore, in-network reprogramming is a key feature for the
success of a WSN and, due to limited node resources, a very
active research field spanning innovative middleware and
operating system capabilities, high-level programming, and
abstraction models for heterogeneous distributed systems. Some
approaches that have been investigated consist of injecting the
network with queries that change the working state of the node
[4], with a binary image of the entire program [5], or with a
script requiring a virtual machine within the software platform
of the node [6].
Implementing both data gathering and in-network system
functionalities for indoor climate observation, the test case
we designed represents the class of monitoring applications
introduced above. The high-level application is indoor climate
observation. The main objective of the test case was to
experiment with those system functionalities by means of a real
deployment. We also outlined problems of WSN integration
within Internet and WSN network analysis by means of client
utilities running on any PC connected on the Internet network.
Therefore, the test case includes:
a WSN that uses COTS, namely TelosSky [7] motes platforms,
while running TinyOS [8] operating system; we used TelosSky
because of its solid representation of sensor nodes capabilities
as well as the support offered by the TinyOS community at the
time we started the test case development;
a set of SW components conceptually referred to as gateway
services. They include a database which stores sensed data,
a web server, and Java applications which interact with the
database and create the packets used to reprogram the
operational conditions of the nodes remotely; and
client tools which analyze the data stored within the database
and which interact with the network by observing its behavior
and topology at run time or which select the parameters of
in-network reprogramming commands.
Our design choices have been strongly influenced by the
additional objective of using the test case to identify a fast
prototyping framework to quickly evaluate performances of
new HW platforms or components in WSN applications. To
reach this goal, we decided to base our developments on open
source libraries and free design environments. The advantages of
open source solutions are well known. The most notable of them
fitting our needs is the fact that an open source enables reuse,
improvements, and unlimited tuning. This makes it possible to
port the code to new hardware, adapt it to changing conditions,
and reach a detailed understanding of how the system works,
taking full advantage of previously developed solutions.
54 COPYRIGHT STMICROELECTRONICS, 2006.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
Therefore, our test case is based on the adoption of:
a TinyOS library environment;
a Linux machine working as gateway server;
a database built on top of MySQL [10]; and
an Apache web server [11].
For implementation, we used NesC [9] for node applications,
PHP [12] for dynamic web pages, and java for gateway and
graphical utilities.
The paper is organized as follows: in section 2, we introduce
the TinyOS framework which constitutes the heart of our SW
platform. In section 3, we describe the test case architecture by
giving implementation details of its main components: the sensor
network, gateway services, and client utilities. In section 4, we
discuss some limits of the test case and we introduce recent
developments. Section 5 is dedicated to conclusions.
2. TINYOS
An extended class of nodes typically used in many WSN
applications is characterized by limited HW resources [13].
Therefore, an operating system framework is needed to manage
those resources efficiently while supporting concurrency-
intensive operation in a manner that achieves efficient modularity
and SW robustness [14].
Popular embedded operating systems do not meet the size,
power, or efficiency requirements. Therefore, several operating
systems have been developed specifically for tiny networked
devices [15],[16],[17], to be used within WSN applications.
We decided to select TinyOS, the most adopted open source
operating system in the WSN research community, an operating
system which is very well supported through a mailing list and
a working group which has joined development efforts from
both industry and academia. In our implementation, we used
version 1.1.15 of the TinyOS package including an extended
SW library for sensor nodes, utilities implementing gateway,
and client services. Addressing the goal of a fast prototyping
framework design, we were confident the popularity of TinyOS
would enable us to reuse a large number of components already
available. Moreover, the involvement of leading research groups
in the field of WSN, would also allow our reusing of components
representing the state of the art in MAC and networking protocols,
components we were not interested in developing further.
TinyOS was designed by targeting Wireless Sensor Network
requirements. It implements an event-driven execution
paradigm where every execution is triggered by some external
event representing hardware interrupts coming from the radio,
timers, or sensor interfaces. TinyOS provides physical device
abstractions such as a conventional operating system. To be
portable on architectures with very limited resources, the TinyOS
approach defines a simple component model and develops a set
of components that are statically wired together to support a
specific application.
Therefore, a TinyOS application consists of a scheduler and a
graph of components, each described by interfaces and internal
implementation. An interface can declare commands and events.
The first ones are implemented by the components that use
that interface, while the second ones are implemented by the
components providing it. Then, a wiring specification describes
how the interfaces are wired together to form the overall
application.
TinyOS takes a language-based application development
approach, implemented in the NesC language, an extension of C
language. The NesC compiler consists of a preprocessor translating
a NesC application into a C module where only the necessary
parts of the operating system are compiled with the application.
The programming model exposed by NesC incorporates the
event-driven execution, the flexible concurrency model, and the
component-oriented application design of TinyOS. Restrictions
on the programming model allow the NesC compiler to perform
whole-program analyses, including data-race detection, which
CUPOLA-SENSING: AN INDOOR CLIMATE MONITORING TEST CASE
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
55
improves the reliability of the SW by safely checking the
concurrency introduced by the programmer.
3. CUPOLA-SENSING TEST CASE
In this section, we describe the main features of Cupola-
Sensing, which we used to monitor the indoor climate within our
laboratory for about three months. During this time, we validated
and refined our design, and we collected a significant amount of
data, which is now accessible through our web site. The test case
name derives from the shape of the roof of the building where
the laboratory is located.
At the system level, the test case consists of the multi-tiered
architecture represented in Figure 1.

The first tier consists of a WSN built with COTS platforms
running TinyOS and integrating a 802.15.4 radio, a processor, and
sensors for climate (humidity, temperature and light) monitoring.
The WSN implements two high level functionalities: data
gathering and command broadcasting. Data gathering means that
motes periodically send packets containing the values collected
by the sensors and information about their working status (i.e.,
voltage level and parent node address). Data are routed through
a multihop network to a node, called a basestation, identified
with a 0 network address, and physically connected to the
gateway with a USB port. Commands Broadcasting is a kind of
in-network reprogramming based on the usage of commands sent
through the gateway to modify some variables characterizing the
node working conditions.
The second tier is a Linux PC implementing gateway
functionalities. The gateway allows us to interface the WSN with
Internet, and it is responsible for collecting the data coming from
sensors, parsing and storing them in a database. It is also responsible
for creating the command packets broadcasted to the network.
The third tier includes WSN client tools that allow an end user
or a network developer to interact with the WSN with a network
visualizer, a web browser (Explorer), and DB clients. Client tools
run on PCs linked with the gateway through a TCP/IP connection.
3.1 WSN
The WSN was constituted by 7 nodes deployed in the laboratory,
as showed in Fig. 1. The rationale behind selecting the node
position was to cover the lab with a reasonable granularity and
to monitor significant areas such as entrance and meeting
rooms to maximize the differences among values sensed by
different nodes. Two nodes were always powered with batteries:
one located at the main entrance door and one in a closet where
Data Gathering
WSN Gateway
Client Data and
Network Browsing
Command
Broadcasting
Internet
SQL
Data
Base
Fwd
Client
(Java)
HTTP
Server
php S
e
r
i
a
l
F
o
r
w
a
r
d
e
r
R&I
HQ
Lab
HQ
AST
HQ
I/O
1
2
3
4
5
6
0
tcp
FIGURE 1:
CUPOLA SENSING TESTBED SYSTEM
ARCHITECTURE.
56 COPYRIGHT STMICROELECTRONICS, 2006.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
The architecture integrates a ChipCon radio implementing the
IEEE 802.15.4 standard adopted by the Zigbee Alliance and
extensively promoted in many WSN scenarios requiring low data
rate communications. TelosSky also integrates both temperature
and light sensors that we wanted to use for climate monitoring. The
reprogramming of the on-board processor is efficiently done through
a USB port that allows powering the motes, as an alternative to a
battery-based solution.
3.1.2 Data Gathering
Data gathering implies an unidirectional data flow towards the
basestation and includes the operations of periodic sampling of the
power line sockets were not available. The other network nodes
were connected to PCs in the lab through their USB ports,
allowing them to be powered with batteries only when the
PCs were switched off during the night or when the owner was
absent. Therefore, our network replicated mixed power-supply
conditions for a realistic indoor scenario by which to analyze
distributed power management network requirements.
Fig. 2 outlines the main components of Cupola Sensing and
their interfaces, where components are grouped in three layers.
While we reused most of the Driver and Networking modules
from the TinyOS package, the Cupola Sensing Component
Cupola
Sensing
ChangeParam
Send
Msg
Clock Get Voltage
APPLICATION
NETWORKING
DRIVERS
HARDWARE
Timer
Bcast
Forward
Queued Send
Send
Send
Receive
Generic
Comm
Send/Receive
CC2420
RadioC
Leds
MultiHop
Photo Temp
Leds ADC ADC
Internal
Voltage
HWClock
Timer
FIGURE 2:
SIMPLIFIED VIEW OF CUPOLA
SENSING. BOXES REPRESENT
COMPONENTS, AND EDGES
REPRESENT INTERFACE WIRING.
and global wiring have been specifically developed for this
test case. Cupola Sensing includes all the payload composition
and command execution functions. The binary image of the
full application (including drivers, networking and application
components) was about 22 kBytes.
3.1.1 Nodes
To implement the WSN, we decided to use TelosSky. The
following reasons motivated our choice.TelosSky well
represents the popular MICA family originally dened
by UC Berkeley [18] for WSN prototyping and it is well
supported within the TinyOS development community.
node sensors, dispatching of packets to the gateway, and tree-based
routing maintenance. The network protocol implements multihop
routing policies, meaning that not every node directly communicates
with the basestation, thus requiring the implementation of a routing
mechanism on each node of the network.
For the implementation of the routing protocol, we reused a
module (LQIMultiHopRouter) already integrated in the TinyOS
library package. It allows for the building of the routing tree
and the routing of messages towards the basestation connected
to the gateway, which represents the root of the tree. The tree is
built and maintained by means of beacon messages periodically
CUPOLA-SENSING: AN INDOOR CLIMATE MONITORING TEST CASE
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
57
broadcasted by each node within the network. The beacon contains
information related to its depth within the tree. When a node is
powered on, it selects an initial parent by listening to messages
and choosing the node corresponding to the smallest depth
within the tree; to seed the creation of the tree, the basestation
periodically broadcasts beacon messages with depth 0. Multiple
networks can be active in the same physical spaces by setting
different group IDs in a group field of radio Message Header.
Beacon packets fields are represented in Fig. 3. The cost field
indicates the total cost to reach the root node and is calculated
by adding the cost indicated within the beacon of the parent with
the parent link quality derived from the channel estimation hook
provided by the radio transceiver. The hopcount field represents
the total number of hops needed to reach the basestation. The
cost and hopcount fields are used by the routing algorithms to
dynamically determine the best way to reach the root node and
consequently the parent to be used. Thus, by allowing each node
to change the current parent, the routing protocol network is
robust against events that can create bad reception on the channel
link or against events caused by possible parent failures.
The CupolaSensing application module on each node samples
the sensors by using a programmable rate, builds the payload by
using those data, and then sends the packet to the network layer
that appends its multihop header. The MAC modules complete
the packet before sending it to the transceiver by also adding the
corresponding message header. The resulting packets and field
details are represented in Fig. 3.
The radio module within the node board consumes 20mA
in reception mode and 17mA in transmit mode. By keeping
the radio always on, the batteries lasted only three days.
Thus, to augment the lifetime of the node, we implemented
a power management policy to reduce the radio duty cycle.
Key consideration is that the radio can be switched off when
communication is not occurring. By considering that potentially
incoming packets were a) beacons, b) sensing data from
other nodes, c) commands coming from the basestation, we
empirically evaluated a minimum active timing window of about
two minutes. For example, by considering a duty cycle of 13%
(sampling data every 15 minutes) the lifetime of battery-supplied
nodes was extended to 15 days. Of course, our policy implies
8 BITS 16 BITS 8 BITS 16 BITS 16 BITS 8 BITS 8 BITS

Length Frame Dsn Destpan Addr AM type Group
Control
FIGURE 3: DATA AND BEACON PACKETS.

MSGHEADER
16 BITS 16 BITS 16 BITS 16 BITS 16 BITS
Source Origin Seqno Origin HopCount
Addr Addr SeqNumber

MULTIHOP
16 BITS 16 BITS 16 BITS

Parent Cost HopCount

BEACON PAYLOAD
16 BITS 16 BITS 16 BITS 16 BITS 16 BITS 16 BITS 32 BITS
Type Light Temperature Humidity Parent Voltage Sequence
Address Number

DATA PAYLOAD
MSGHEADER MULTIHOP DATA PAYLOAD DATA PACKET
MSGHEADER MULTIHOP BEACON BEACON PACKET
58 COPYRIGHT STMICROELECTRONICS, 2006.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
that commands are sent from the gateway in a synchronous way.
While this was not a limitation for the application, the need to
align commands broadcasting with the network sampling rate
was annoying for the WSN user. Hence, we implemented a
function CommandSync in the gateway that simply delays the
delivery of a command asynchronously requested by a user to
meet the radio active time slot of the nodes.
and their payload fields are logically correlated with the state
defined for each node established by:
the LED status (RED, Green, Blue on/off);
the sensing rate, representing the time interval between two
successive sensor readings;
the radio transmit power, ranging in the interval -25 dB and 0
dB as allowed by the radio component; and
16 BITS 16 BITS 32 BITS 16 BITS 16 BITS

CommandID Type NewRate Address Flag
NewPower
FIGURE 4: COMMAND AND ACK PACKETS.

COMMAND
PAYLOAD
16 BITS
SeqNumber

BROADCAST
HEADER
16 BITS

CommandID

ACK PAYLOAD
COMMAND PACKET MSGHEADER BROADCAST HEADER COMMAND PAYLOAD
ACK PACKET
MSGHEADER MULTIHOP ACK PAYLOAD
3.1.3 In-Network Reprogramming
To enable the post-deployment working condition update
of sensing nodes, we implemented a policy for remote
reprogramming based on diffused commands. The commands,
unicast or broadcast, are composed on the client side, injected
within the network through the gateway, and flooded by each
node. For protocol implementation, we relied on components
already integrated within TinyOS library package (BCastM).
The command packet format is represented in Fig. 4. The
sequence number is the only information contained in the
broadcast header and is used by the protocol to establish if the
packet received was new or previously received. The sequence
number is filled by the basestation node before sending the
message. Five different kinds of commands have been defined,
the power save (on/off) that activates the radio duty cycle
policy.
In the data gathering operation for this application scenario,
message loss has no significant consequence on the global
application results. Instead, a command message delivery
requires reliable communication, as the user must be notified
if the command has been correctly received by the target
node. Therefore, the CupolaSensing component on each node
implements an acknowledgement policy by sending the Ack
packets represented in Fig. 4 after the command has been
executed. The Ack packets are sent to the basestation using the
multihop protocol described in the previous paragraph. The only
payload field is the command sequence number, which allows
the gateway to associate the Ack packet to the right command.
CUPOLA-SENSING: AN INDOOR CLIMATE MONITORING TEST CASE
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
59
3.2 Gateway
Within a WSN system, the gateway plays the fundamental
role of interfacing a WSN with other networks. In our case
study, the gateway has been mapped on a PC running Linux
allowing us to interface the WSN with Internet to use client
tools for data analysis and network maintenance. The main
features implemented enable us (a) to read the packets received
by the basestation through the USB port, (b) to interpret their
content, (c) to store data in a database, (d) to compose command
packets used to reprogram the nodes, and (e) to provide basic
services to the clients connected, for example, via Web or via
a TCP socket. The overall gateway functional architecture is
represented in Fig. 5.

The Serial Forwarder module in Fig. 5 is an application
belonging to the TinyOS package. Its main function is to read,
without computation, ack and data packets coming from the
USB port and, vice versa, to deliver command packets generated
on the gateway side to the USB port. The Serial Forwarder opens
a TCP connection to which every client can connect. While the
U
S
B

t
o

S
e
r
i
a
l
S
e
r
i
a
l

F
o
r
w
a
r
d
e
r
Packets Access
methods
(MIG generated)
Read&Store
Packet
(Java)
Send Commands
(Java)
HTTP
Server
php
Socket
Internet
My SQL
DB
R&I
HQ
Lab
HQ
AST
HQ
I/O
1
2
3
4
5
6
0
FIGURE 5:
GATEWAY ARCHITECTURE
AND INTERCONNECTIONS.
Readings
Date
Time
OriginAddress
HopCount
ParentAddress
SequenceNumber
Light
Temperature
Humidity
Voltage
Commands
Date
Time
CommandID
Type
NewRate
Address
Flag
Nodes
Node
Name
X
Y
Flag
Acks
CommandID
Node
Command Type
CommandID
Node
Gateway measures
Packets Content
Metadata
FIGURE 6: MYSQL DATABASE TABLES.
physical connection between the PC and Telos uses the USB
port, the communication protocol transfers the packets serially
with a micro dedicated to the conversion on the Telos and of a
SW driver on the PC side. The gateway implements a database
service based on MySQL. The database is used by the gateway
applications and by the web server integrated within the gateway.
Moreover, the database can be accessed by other tools on the
client side. In the database, we implemented five tables, which
are represented in Fig. 6.

Readings contains data collected by the network. Its fields
correspond to the data payload, multihop header, and additional
information identifying the time where the packet is received by
the gateway.
Commands contains the list of all the commands injected in the
WSN with related parameters.
60 COPYRIGHT STMICROELECTRONICS, 2006.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
Acks contains the list of commands for which an ack signal has
been received.
Nodes and CommandType are additional tables with some
descriptive info about command types and node attributes, such
as their position within the network.
Table content is used by most of the gateway functions providing
services to the clients. For example, information stored in Acks
and Commands are composed to derive the network status
corresponding to the last acknowledged commands for each node
or to identify the beginning of the radio active timing window
of nodes in power saving. This allows for the synchronization of
the command injections within the network.
The gateway modules have all been written in Java. To parse or
generate the bit streams going through the serial forwarder, all the
applications use the methods generated by the Message Interface
Generator (MIG) tool, included in the TinyOS package. MIG is
a tool which automatically generates Java classes corresponding
to the message format used in a given application. MIG reads the
NesC file where the structure of application packet is specified
as a struct definition and generates a Java class that takes care
of the details of packing and unpacking fields in the message
byte format. Therefore, the use of MIG is valuable in saving
a programmer from parsing message formats for each given
application.
The last module implemented on the gateway is an Apache
web server providing web pages for client use. The web pages
have been written in PHP and can execute simple queries on the
MySQL or run the Java applications described above.
3.3 Clients
In the overall WSN system structure, a significant role is played
by client tools that are fundamental for network topology
visualization, data analysis, network debugging, and network
programming. Therefore, we believe the availability of suitable
client tools can greatly facilitate the investigation and even
the adoption of WSN solutions. In our test case we relied on
graphical utilities to interact with the database, to compose
commands for network reprogramming, and to analyze at run
time the topology and behavior of the sensor network. Most of
those utilities are accessible through a WEB browser interacting
with the gateway web server or are available as Java services.
The choice of a web browser as user interface has clearly been
motivated by the popularity within the PC community.

The MySQL database can be queried through MySQL clients,
as showed in Fig. 7b, or predefined queries can be submitted
through the Web browser, Fig. 7a.
An additional client utility we customized from an existing
application is a network visualizer that was a key debugging tool
during the experimentation of the test case. As represented in Fig.
7c, the GUI serves two purposes. First, it provides a graphical
interface for the user to explore, at run time, the various services
offered by the WSN such as temperature and light intensity
measures. Then, it ensures the validation of the proper network
operation via augmented visualization of its topology, routing
tree, and node health status. The network visualizer is accessible
from the browser as a java Web service. When activated, it
connects with the TCP/IP socket opened by the Serial Forwarder
on the gateway, thus receiving a copy of each packet coming from
the sensor network. By analyzing the payload and header with
MIG-generated methods, the network visualizer can graphically
show the network topology and the sensing values of each node.
4. RESULTS AND CURRENT WORK
In this section, we discuss the limits of the current implementation,
and we introduce new developments working towards the
definition of a general framework for fast prototyping, which we
are defining to evaluate STM platforms and sensor components
in WSN applications.
Many of the TinyOS modules used to build the Cupola Sensing
application and running on sensing nodes have been reused from
CUPOLA-SENSING: AN INDOOR CLIMATE MONITORING TEST CASE
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
61
FIGURE 7: (A) WEB BROWSER;
(B) MYSQL CLIENT;
(C) NETWORK VISUALIZER
the original TinyOS package. Besides fixing the bugs within
those components, specific developments were focused on
their integration and on the design of dedicated procedures for
command parsing, power management, and ack generations, as
described in the above sections. Concerning gateway and client
services, instead, most of the procedures have been developed
from scratch for the specific needs of the test case.
The entire development lasted about two months, during which
a significant effort was also carried out to practice with TinyOS
existing libraries. The validation of the test case was perfomed
for three more months, during which time we further refined and
fixed the system architecture and its components.
During the testing period, we stored a significant quantity of data
accessible through the database tables. Fig. 8 represents typical
graphs retrieved with the GNU plot tool activated by the PHP
applications running on the Web server. We precisely quantified
the climate conditions of the lab as expected by showing daily
and monthly cycles, as indicated in Fig. 8a and 8b, where
measures captured with temperature sensors are represented.
Furthermore, we roughly analyzed the network behavior by
verifying packet (data and ack) loss and some network dynamics
such as parent identity changes for each node due to channel
interference or node failures, as shown in Fig. 8c.
We believe client tools are a critical component of the WSN
system architecture from both a user and developer perspective.
The solutions implemented so far were characterized by low
flexibility, and current developments are directed towards
the integration of more powerful environments such as Excel
and Matlab. We believe the integration of these tools is well
(a) (c)
(b)
62 COPYRIGHT STMICROELECTRONICS, 2006.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
motivated by the advantages they provide in terms of data
analysis, data programming and processing, and run time system
management and support, all widely used among engineering
groups.
With respect to the gateway, we plan to increase its flexibility.
We are developing an XML translator that transforms raw data
coming from the Serial Forwarder into XML packets that are
made available on an additional connection socket. In this way,
the clients would not connect to the Serial Forwarder socket, thus
avoiding misalignments among MIG-generated libraries working
on different client machines. The implementation will be traded
off against the run time monitoring requirements of the client
tool. We are also defining an API library which will facilitate
the client access to other gateway services. Our next step will
be porting the developed gateway architecture on mobile HW
platforms such as a smartphone or a PDA.
On the network side, one of the main problems we experienced
was network debugging, as node and network failures may
happen in very unpredictable ways. Additionally, because
nodes do not provide a rich machine user interface, debugging
is made only more difficult. Wherever possible, a solution is
included within the data payload information about current
working conditions. For example, we found the use of in-node
voltage measurements (Fig. 8d) to verify when the battery was
approaching low charge thresholds very useful. When the battery
was under that threshold, we also observed unrealistic values of
the temperature sensor measurements, and we could easily
assess when this was related to the battery voltage. On the other
TABLE II. PERCENTAGE OF ERROR BETWEEN
THE MODEL PREDICTION AND THE ACTUAL
MEASURED ADDITIONS AND MULTIPLICATIONS
PER PIXEL. FOR EACH GOP OF EVERY
SEQUENCE, THE AVERAGE ERROR OVER A
NUMBER OF ADAPTATION POINTS IS PRESENTED.
20
22
24
26
28
30
18
t
e
m
p
e
r
a
t
u
r
e
FIGURE 8: EXAMPLES OF SAMPLED DATA: (A) TEMPERATURE DATA IN A DAY WINDOW; (B) TEMPERATURE DATA IN A WEEK WINDOW;
(C) PARENT CHANGES FOR EACH NODES DUE TO LQI EVALUATION; (D) BATTERY VOLTAGE MEASURES IN A MONTH TIME WINDOW.
16
18
20
22
24
26
28
30
32
14
t
e
m
p
e
r
a
t
u
r
e
400
600
200
800
0
edolo portolo eralo biolo
Node ID
P
a
r
e
n
t

P
a
r
t
i
t
i
o
n
Data From 2005-11-23 to 2005-11-27
labolo paolo
Base Station
edolo
portolo
eralo
biolo
labolo
paolo
1
1
/
0
5

0
0
:
0
0
1
1
/
0
5

0
0
:
0
0
1
1
/
0
5

0
0
:
0
0
1
2
/
0
5

0
0
:
0
0
1
2
/
0
5

0
0
:
0
0
1
2
/
0
5

0
0
:
0
0
1
2
/
0
5

0
0
:
0
0
1,8
2
2,2
2,4
2,6
2,8
3
1,6
V
o
l
t
a
g
e
(a) (b)
(c) (d)
CUPOLA-SENSING: AN INDOOR CLIMATE MONITORING TEST CASE
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
63
hand, this solution cannot always be applied, as many times
unexpected node behavior consists of the suspension of data
flow. We were not always able to understand the reason for these
failures. In the test case we did not develop a dedicated in-node
solution for auto-diagnosis and auto resetting by using watchdog
timers or by logging code to motes flash memory. Therefore, we
could only send a reset command remotely or push the reset key
of the Telos board.
The power management policy implemented to reduce the
radio duty cycle significantly increased the lifetime of the
batteried nodes, as described in section 3.1.2. Actually, our
current implementation is limited by the lack of distributed
synchronization algorithms integrated within the nodes. Hence,
we had to be very conservative in the evaluation of active
radio timing window (two minutes) to ensure the arrival of
beacons, commands, and data to their destination. In addition,
time stored in the database corresponds to arrival time and not
sampling time, as we do not stamp the data before sending
them through the radio. Given the limited network size and
our application target, the introduced approximations did
not significantly impact the accuracy results. Nevertheless, a
detailed analysis of synchronization within the network is one of
the on-going developments addressing applications with tighter
synchronization constraints (i.e., event localization).
5. CONCLUSION
In this paper, we have described a test case we deployed in our
laboratory for climate monitoring. We used this case study to
deepen our understanding of the engineering challenges faced by
WSN system designers. The functionalities for data gathering,
in-network reprogramming, and Web interfacing make this
test case highly representative of a large class of applications
for indoor monitoring. The implementation was based on open
source libraries and tools that allowed for rapid development;
it also improved our expertise with respect to state-of-the art
solutions in the area of OS for embedded systems and ad-hoc
networking protocols. Current developments are focused on the
generalization of the test case features which will identify a
framework for rapid WSN prototyping.
ACKNOWLEDGEMENTS
The authors want to thank Andrea Serlini for the important
contributions to the development and testing of the Cupola
Sensing test case.
REFERENCES
[1] A. Mainwaring, J. Polastre, R. Szewczyk, D. Culler,
J. Anderson, WIRELESS SENSOR NETWORKS FOR HABITAT
MONITORING, WSNA 2002.
[2] G. Werner-Allen, J. Johnson, M. Ruiz, J. lees and Matt
Welsh, MONITORING VOLCANIC ERUPTIONS WITH
A WIRELESS SENSOR NETWORK, EWSN 2005.
[3] G. Werner-Allen, Pat Swieskowski, and M. Welsh,
MOTELAB: A WIRELESS SENSOR NETWORK TESTBED,
IPSN 2005.
[4] Y.Yao, and J.E. Gehrke, THE COUGAR APPROACH TO
IN-NETWORK QUERY PROCESSING IN SENSOR NETWORKS,
ACM Sigmod Record, 2002.
[5] J. Hui and D. Culler, THE DYNAMIC BEHAVIOR OF A DATA
DISSEMINATION PROTOCOL FOR NETWORK PROGRAMMING
AT SCALE, International Conference on Embedded
networked Sensor Systems, ACM 2004.
[6] P. Levis and D. Culler, MATE: A TINY VIRTUAL MACHINE
FOR SENSOR NETWORKS, ACM ASPLOS 2002.
[7] Moteiv Corporation, telos (Rev B) datasheet
http://www.moteiv.com , Dec 2004.
[8] TinyOS Home Page http://www.tinyos.net.
64 COPYRIGHT STMICROELECTRONICS, 2006.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
[9] NesC Reference Manual http://nesc.sourceforge.net/papers.
[10] MySQL Home Page http://www.mysql.com.
[11] Apache Home Page http://www.apache.com.
[12] PhP Home Page http://www.php.net.
[13] J. Hill, M. Horton, R. Kling, L. Krishnamurthy,
THE PLATFORMS ENABLING WIRELESS SENSOR NETWORKS,
Communications of the ACM, 2004.
[14] J. Hill, R. Szewczyk, A. Woo, S. Hollar, D. Culler
and K. Pister, SYSTEM ARCHITECTURE DIRECTIONS
FOR NETWORKED SENSORS, ASPLOS, 2000.
[15] D. Culler, J. Hill, P. Buonadonna, R. Szewczyk,
and A Woo, A NETWORK-CENTRIC APPROACH
TO EMBEDDED SOFTWARE FOR TINY DEVICES,
Intel Research Report, 2002.
[16] C-C Han, R. Kumar, Roy Shea, E. Kohler,
and M. Srivastava, A DYNAMIC OPERATING SYSTEM
FOR SENSOR NODES, MobySys 2005.
[17] A. Dunksels, B. Gronvall, T. Voigt, CONTIKI A
LIGHTWEIGHT AND FLEXIBLE OPERATING SYSTEM FOT TINY
NETWORKED SENSORS, First IEEE Workshop on Embedded
Networked Sensors, 2004.
[18] J. Hill, D. Culler, MICA: A WIRELESS PLATFORM FOR
DEEPLY EMBEDDED NETWORKS, IEEE Micro 2002.
[19] J. Elson, K. Romer, WIRELESS SENSOR NETWORKS:
A NEW REGIME FOR TIME SYNCHRONIZATION,
HotNets-I 2002.
CONTACT: ST.JOURNAL@ST.COM
ZIGBEE: THE JOURNEY TOWARD DEPLOYMENT IN INDUSTRIAL APPLICATIONS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
65
Gilles Thonet, Marc Bruel
Schneider Electric
ZIGBEE THE JOURNEY
TOWARD MASS MARKET
ADOPTION
ZigBee has recently been introduced
as a promising technology for low-cost,
low-power wireless networks requiring
flexibility in node placement. Relying
on the use of application profiles,
ZigBee has been specifically devised
for such scenarios as home and building
automation as well as sensor networking
and automated metering. Although a
first specification was completed in
December 2004, few products are found
today on the market. One reason is
that the standard is still being amended
and application profiles are not all
finalized yet. Another reason is that large-
scale deployments in commercial and
industrial environments are perceived to
be hampered by potential hurdles. This
article revisits the path toward ZigBee
market introduction in various industries.
Objectives are twofold. The first one is to
identify technical issues that may constitute
an impediment to bringing ZigBee
products to life. In particular, the following
1. INTRODUCTION
Wireless sensor networks have gained much attention in the
last years. Responding at first to communication needs in
military operations, the concept of wireless sensor networking
has progressively been permeating most infrastructures and
environments we interact with daily. Homeland security and
surveillance have been strong catalysts for advancing such
a technology, especially since 9/11. Today, business drivers
are being found in many other markets. Home and building
concerns expressed by industry leaders
and end customers, namely, the potential
coexistence problems between ZigBee and
other wireless technologies such as WiFi,
are discussed and illustrated by experiments
carried out at Schneider Electric. The second
objective of the paper is to describe some
of the most promising application areas and
what needs to be acted upon to complete
the journey toward mass market adoption for
each.
COPYRIGHT SCHNEIDER ELECTRIC, 2007.
66 COPYRIGHT SCHNEIDER ELECTRIC, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
CMs to
lengths,
ng, and
automation, industrial control, environmental monitoring,
and automated meter reading (AMR) are among those most
often cited. The bottom line is deploying cheap, low-power,
easily-packageable radio-frequency nodes whose typical tasks
range from reporting relatively simple measurement data to
participating in more sophisticated control processes. The added
value of a wireless sensor network clearly does not reside in high
bandwidth but rather in cost efficiency, low power consumption,
and the ability to accommodate changing topologies.
Until recently, most wireless sensor network developments
have relied upon using proprietary technologies. In addition to
disallowing vendor and device interoperability, legacy systems
entail higher development costs and riskier sourcing for original
equipment manufacturers (OEMs). In 2005, the ZigBee Alliance,
an international consortium of major industrial and consumer
market leaders, publicly released the first version of an open
standard for wirelessly networking control and monitoring
devices. The cornerstones of ZigBee are flexibility in node
placement, low power consumption, and vendor interoperability
through the definition of public application profiles. A new,
enhanced version was announced in September 2006, and work
on specifying additional features is ongoing.
Manufacturers coming from a broad range of industries
have expressed high interest in ZigBees enabling some of
their products. Early introductions have occurred in markets
such as building automation, home control, and AMR. The
focus, however, has been more on making press releases and
investigating the respective markets than building up consistent
product ranges. According to many industry observers, ZigBee
clearly is on the rise, but adoption, and, to a further extent
deployment, are taking more time than initially expected.
Our article aims to revisit the journey toward ZigBee deployment
in residential, commercial, and industrial environments. To that
end, we have identified some potential obstacles. For instance,
a chief concern of building and home product manufacturers
seems to be possible interference with other wireless networks,
in particular WiFi (i.e., commodity IEEE 802.11b/g). Some
experimental results are presented here to support improvements
in the standard and motivate the adoption of deployment and
installation guidelines. Then, based on this discussion, we
further discuss application roadmaps, along with some of the
issues to be addressed before product commercialization can
begin. The conclusion summarizes the main steps that should
guide both the ZigBee Alliance and OEMs in speeding up market
introductions.
2. THE ZIGBEE STANDARD
ZigBee Alliance member companies come from a large
spectrum of industries ranging from consumer electronics to
industrial automation. They share one common goal: enabling
interoperability among wireless sensing and actuating devices
in various control environments. The ZigBee specification
defines a communication stack, several public application
profiles (including device and interface definitions pertaining to
a given application space, e.g., lighting), several stack profiles
(including stack configuration options relevant to a subset
of applications, e.g., tree addressing/routing for light home
automation), and means for ensuring standard compliance and
platform certification. Detailed information is available directly
from the ZigBee Alliance website [1]. The rest of this section
provides a brief overview of the standard.
2.1 Communication Stack
As shown in Fig. 1, the ZigBee stack architecture primarily
relies on three pillars: standard IEEE lower layers, ZigBee layers
that define networking and application-level communication
mechanisms, and a commissioning framework that will provide
reference procedures for initializing and configuring ZigBee
networks.
2.1.1 IEEE 802.15.4 Layers
Both physical (PHY) and medium access control (MAC) layers
of ZigBee correspond to the IEEE 802.15.4 standard [2]. Based
ZIGBEE: THE JOURNEY TOWARD DEPLOYMENT IN INDUSTRIAL APPLICATIONS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
67
on direct sequence spread spectrum (DSSS) modulation, the
PHY layer has been designed to accommodate low-cost and
high-integration constraints. It is actually a dual layer operating
in two sets of unlicensed frequency bands, as shown in Table 1.
Deployment, however, is expected to occur mainly in the 2.4
GHz band because of worldwide availability and higher data
rate. Transmission range typically covers between 10 and 50 m
indoor, depending on the surrounding environment.
The MAC design has been essentially driven by the capability of
supporting multiple network topologies with minimal complexity.
Based on carrier sense multiple access with collision avoidance
(CSMA/CA), it includes several features that facilitate power
management while reducing functionality whenever desirable.
Two principal IEEE 802.15.4 evolutions are already in the
pipeline:
IEEE 802.15.4a addresses amendments to IEEE 802.15.4
for an alternative PHY layer. Currently an IEEE task group, this
initiative aims at providing high throughput, ultra-low power,
and high-precision location capability to enable new application
and market opportunities. The baseline specification includes
two optional PHY layers consisting of an ultra-wideband (UWB)
impulse radio and a chirp spread spectrum radio. First
specification release had been so far planned for the end
of 2006.
IEEE 802.15.4b targets a refinement of the current IEEE
802.15.4 specification to clear up ambiguities, resolve
Application Layer
Application Support Sublayer
C
o
m
m
i
s
s
i
o
n
i
n
g

F
r
a
m
e
w
o
r
k
Network Layer
IEEE 802.15.4
ZigBee
Device
Object
Application
Framework
Application
Framework
Public
Application
Profiles
Private
Application
Profiles
IEEE Standard
ZigBee Standard (available today)
ZigBee Standard (under development)
Product/ Manufacturer Specific
- to be implemented
Implementation of device types
(coordinator, router, end device)
Application support functions
(binding, secure key management, ...)
Network management functions
(addressing, routing, route, discovery, ...)
MAC
PHY
FIGURE 1: ZIGBEE STACK ARCHITECTURE
Frequency Bands Industrial, Scientic, and Medical Sub-GHz
2.4 GHz (Worldwide) 868 MHz (Europe)
915 MHz (North America)
Channels 16 1 (Europe)
10 (North America)
Modulation Offset Quadrature Phase Shift Binary Phase Shift
Keying (O-QPSK) Keying (BPSK)
Maximum Raw Data Rate 250 kbps 20 kbps (Europe)
40 kbps (North America)
TABLE 1: DUAL IEEE
802.15.4 PHY LAYER
68 COPYRIGHT SCHNEIDER ELECTRIC, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
inconsistencies, and make specific extensions such as faster
sub-GHz data rates and time synchronization. The standard
has been approved in June 2006, and the first products are
expected by the end of 2007.
2.1.2 ZigBee Layers
The ZigBee specification actually takes place above the PHY
and MAC layers. The network layer as defined by the ZigBee
Alliance [3] provides primitives for managing star, cluster-tree,
and mesh topologies (Fig. 2). Star networks are the simplest ones
and provide for very long battery life. Mesh (or peer-to-peer)
networks form self-healing topologies, enabling higher levels of
reliability and scalability by offering more than one routing path.
Cluster-tree networks are based on a hybrid star/mesh topology
combining part of the benefits of both.
ZigBee nodes belong to one of the following three device types:
The coordinator is responsible for initializing the network
and discovering other nodes. There is one and only one
coordinator per ZigBee network. When using a star topology,
the coordinator routes all data from one node to another since
there are no peer-to-peer communications.
ZigBee coordinators are assumed to be mains-powered.
Routers participate to routing and node discovery tasks.
They generally are expected to be mains-powered to cope
with frequent data transmissions.
End devices typically are battery-powered sensing and
actuating nodes. They will be in sleep mode most of the time.
Although the three topologies were introduced in the standard
from its inception, support for full meshing has been quite
limited so far. This capability has until now been supported
by enhanced implementations of the specification, such as
Embers EmberZNet stack. Security services such as encryption
mechanisms are provided at network and application levels.
On top, the application layer supplies primitives to develop
interoperable ZigBee products through the use of public
application profiles. If interoperability is not a requirement,
private application profiles may be implemented and used instead.
2.2 Application Profiles and Stack Profiles
To ensure product and vendor interoperability, public application
profiles are being developed by the ZigBee Alliance for areas
such as home automation, commercial building automation,
and industrial process monitoring. The first one to have been
completed was Home Controls Lighting. Based on a rather rigid
stack configuration, this profile now serves illustration purposes.
Since then, newer application profiles have been developed
to address a much larger range of deployment environments
(buildings, homes, plants, sensor networks, etc.).
In order to bring further consistency into the choice of a
complete communication stack, the ZigBee Alliance has
FIGURE 2: ZIGBEE NETWORK
TOPOLOGIES
Coordinator
Star
Cluster
Tree
Mesh
Router
End Device
ZIGBEE: THE JOURNEY TOWARD DEPLOYMENT IN INDUSTRIAL APPLICATIONS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
69
Since these environments are likely to be the footstep of massive
wireless sensor network deployment, WiFi appears today as the
biggest obstacle when it comes to coexistence with ZigBee.
Although built-in mechanisms help mitigate the impact of WiFi
interference, ZigBee networks may be significantly affected by
tedious communication exchanges like downloads and media
streaming. It is therefore critical to characterize interference
levels accurately in order to devise appropriate mitigation
strategies and network installation recommendations.
3.1.1 Background
Coexistence performance is affected by a large number of
technological, environmental, and application-related factors.
Typical ones include:
distance of the desired transmitter to the desired receiver;
distance of the interfering transmitter to the desired
receiver;
frequency offset between desired and interfering devices;
transmitter output power of desired and interfering devices;
transmitter duty cycle; and
channel conditions such as path loss, multi-path, fading,
and polarization loss.
Channel conditions are usually not controllable, whereas
physical distances, carrier frequencies, and output power levels
are typical setup or installation parameters. In general, a smaller
distance between desired transmitter and desired receiver and a
greater distance between desired and interfering devices mean
higher coexistence performance. A similar reasoning applies to
output power levels. Regarding carrier frequencies, a greater
frequency offset from the desired carrier generally results in
better interference rejection.
One approach to characterizing ZigBee/WiFi coexistence is
superimposing spectrum usage of both technologies, as described
in [4]. For illustration purposes, Fig. 3 shows the alignment
between the 2.4 GHz European non-overlapping IEEE 802.11b
introduced the concept of a stack profile. Whereas an application
profile specifies device descriptions and standard interfaces, a
stack profile defines various stack configuration options (e.g.,
addressing mechanism, routing mode, security level, etc.) that
apply to a subset of application profiles. There currently are two
stack profiles:
ZigBee, previously called Home Controls (HC), which targets
moderate-size networks for simple residential applications.
This stack profile supports the Home Automation (HA)
application profile.
ZigBee Pro, previously called Commercial, Industrial, and
Institutional (CII), which focuses on larger-size networks or
applications requiring advanced functionality, such as
frequency agility, i.e., the possibility of changing channels in
case of persistent interference. This stack profile will be
working with all application profiles.
3. TECHNICAL HURDLES
As mentioned at the beginning of the paper, some claim that
ZigBee adoption is taking more time than originally expected.
Although a first version of the specification has been made
publicly available since 2005, there still are some uncertainties
and technical pitfalls that have slowed down mass adoption. This
section aims at discussing two of them and pinpointing possible
mitigation strategies.
3.1 Coexistence with Other Wireless Networks
While ZigBee can operate in three frequency ranges, 2.4 GHz
is expected to become the widest used band since it simplifies
worldwide deployment and provides higher data rate. The
increasing presence of other wireless technologies in the same
band (WiFi, Bluetooth, cordless phones in some regions of
the world) makes, however, network coexistence a source of
concern. In particular, WiFi concentrates the largest part of these
worries because of its wide spectrum spread, high output power
level, and growing presence in residential and office premises.
70 COPYRIGHT SCHNEIDER ELECTRIC, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
channels and IEEE 802.15.4 channels. A slightly different figure
applies to the North American situation.
There are exactly four IEEE 802.15.4 channels that fall in
the guard bands between or above the non-overlapping IEEE
802.11b channels:
channels 15, 16, 21, 22 in Europe, and
channels 15, 20, 25, 26 in North America.
Although energy is much lower within these channels, it will not
be zero, and interference with close WiFi channels is likely. This
is why this approach, while providing some guidance on how to
position ZigBee channels with regard to WiFi channels, requires
additional experiments to better assess overall coexistence
performance.
Some insights into the ZigBee/WiFi interference problem may
be found in the literature. Study [5], for example, reports the
following results:
In worst-case conditions with respect to frequency overlap,
local distance and high interferer traffic load, more than 90%
of the ZigBee frames can be lost. However, a few time slots
subsist for successful transmissions. This may not prove
sufficient for all applications, but makes a channel
switchover possible.
A minimum frequency offset of about two IEEE 802.11b
channels is required to obtain a negligible interference on
IEEE 802.15.4 transceivers.
Application note [6] provides a theoretical and experimental
investigation of the same issue for an off-the-shelf IEEE
802.15.4 transceiver. Frequency offset, channel bandwidth, duty
cycle, and transmitter output power of the desired and interfering
signals are all shown to impact coexistence performance. To
keep smooth coexistence, the frequency offset between ZigBee
and WiFi networks is recommended to be greater than 25 MHz.
3.1.2 Experimental Setup
To better characterize coexistence in real-world environments,
a MODBUS (widely spread industrial automation fieldbus)
serial line application exchanging data between IEEE 802.15.4
transceivers in the presence of IEEE 802.11b interference has
been run. Fig. 4 shows the corresponding test block diagram.
Two PCs act as FTP server and FTP client, respectively, to send
2480
26
2475
25
2470
24
2465
23
2460
22
2455
21
2450
20
2445
19
2440
18
2435
17
2430
16
2425
15
2420
14
2415
13
2410
12
2405
11
2472
13
2442
7
2412
2410 MHz
2 MHz
22 MHz
2483 MHz
IEEE 802.15.4
IEEE 802.11b
Ch:1
FIGURE 3: SPECTRUM USAGE IN EUROPE OF IEEE 802.15.4 AND IEEE 802.11B (NON-OVERLAPPING CHANNELS)
ZIGBEE: THE JOURNEY TOWARD DEPLOYMENT IN INDUSTRIAL APPLICATIONS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
71
and receive pseudo-continuous WiFi frames. The serial line
application consists of a Telemecanique TeSys U programmable
logic controller generating MODBUS frames that are sent
through an IEEE 802.15.4 transmitter (acting as ZigBee router) to
a remote IEEE 802.15.4 receiver (acting as ZigBee coordinator).
ZigBee devices in this setup are off-the-shelf Texas Instruments
CC2420 transceivers.
Experiments are carried out as follows:
1. Interfering transceivers are turned off. Desired signal carrier
frequency F
2
is set to a given ZigBee channel. N
0
, the number
of ZigBee frames received in one minute, is calculated.
2. Interfering transceivers are turned on with signal output power
of 20 dBm. Raw data rate is set to 11 Mbps. Interfering signal
carrier frequency F
1
is fixed at 2442 MHz, a predetermined
channel for all tests. WiFi transceivers start an FTP exchange.
3. Desired signal carrier frequency F
2
is set to a given ZigBee
channel. For a given distance d (ranging from 0.5 m to 4 m)
between the IEEE 802.11b access point and the IEEE 802.15.4
receiver, the number of ZigBee frames received in one minute,
N, is calculated.
4. Distance d is incremented by 0.5 m, and step 3 is run again.
5. Ratio N/N
0
is computed for each set of parameters (F
2
, d).
An overall ratio is finally determined as the average of 15
one-minute tests.
6. For all tests, a calibration phase using a spectrum analyzer
allows for the determination of the interfering received power
(P
r,1
) and the desired received power (P
r,2
).
Table 2 exhibits IEEE 802.11b received power levels (P
r,1
) with
respect to IEEE 802.11b access point/IEEE 802.15.4 receiver
distance (d), as estimated during calibration phase.
PC
IEEE 802.11b
Access Point
TeSys U
RJ45
PC
RJ45
FTP Server
FTP Client
D=8,5m
RS485
Crossover
Ethernet Cable
d=0,5-4m
(0,5m step)
d=6,5m
Crossower
Ethernet Cable
d
1
=4-8m
d
2
=11.2m
PC
IEEE 802.15.4
Receiver
(Coordinator)
IEEE 802.15.4
Transmitter
(Router)
RS232
RS485
RS232
RS232
F
1
F
2
F
2
F
1
IEEE 802.11b
Gateway
FIGURE 4: BLOCK DIAGRAM OF MODBUS SERIAL LINE EXPERIMENT
Table 3 lists IEEE 802.15.4 received power levels (P
r,2
) with
respect to IEEE 802.15.4 transmitter/IEEE 802.15.4 receiver
distance (D). The actual physical value is D = 8.5 m. Values D >
8.5 m have been simulated by varying the measured power output
levels P
r,2
at IEEE 802.15.4 receiver during calibration phase.
d [m] 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
P
r,1
[dBm] -16 -17 -21 -25 -26 -30 -32 -32
TABLE 2: IEEE 802.11B RECEIVED POWER LEVELS.
d [m] 8.5 20 30 90
P
r,2
[dBm] -61 -65 -71 -81
TABLE 3: IEEE 802.15.4 RECEIVED POWER LEVELS.
72 COPYRIGHT SCHNEIDER ELECTRIC, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
Using this experimental setup, the success rate of ZigBee frame
reception in the presence of WiFi interference can be calculated
with respect to the frequency offset and physical distance between
ZigBee and WiFi nodes. For illustration purposes, Fig. 5 presents
results obtained for a ZigBee received power level of 65 dBm
(corresponding to a physical distance of 20 m). Although a clear
performance decrease is observed whenever ZigBee and WiFi
channel frequencies are nearby (F < 10 MHz), coexistence may
still be an issue for larger frequency offsets when ZigBee and
WiFi nodes are physically close to each other.
Based on this set of results, physical distances and frequency
offsets leading to smooth coexistence have been determined in
Table 4, Table 5, and Table 6. Values are to be interpreted as
follows. For instance, to guarantee successful delivery of 80% of
the packets, two ZigBee nodes can be 30 m apart in free space if
the WiFi interferer is at least 2 m apart and the frequency offset
is greater or equal to 25 MHz.
FIGURE 5: RESULTS OF MODBUS SERIAL LINE APPLICATION FOR D = 20 M
D
max
[m] (P
r,2
[dBm]) d
min
[m] (P
r,1
[dBm]) F
min
[MHz]
90 (-81) 0.5 (-15) 0
TABLE 4: COEXISTENCE PARAMETERS FOR N/N
0
> 5%.
D
max
[m] (P
r,2
[dBm]) d
min
[m] (P
r,1
[dBm]) F
min
[MHz]
20 (-65) 2 (-25) 20
30 (-71) 2 (-25) 25
90 (-81) 2 (-25) 30
TABLE 5: COEXISTENCE PARAMETERS FOR N/N
0
> 80%.
D
max
[m] (P
r,2
[dBm]) d
min
[m] (P
r,1
[dBm]) F
min
[MHz]
30 (-71) 2 (-25) 30
90 (-81) 3 (-30) 30
TABLE 6: COEXISTENCE PARAMETERS FOR N/N
0
= 100%.
100%
95%
90%
85%
80%
75%
70%
65%
60%
55%
50%
45%
40%
35%
30%
25%
20%
15%
10%
5%
0%
-40 -40 -35 -35 -30 -30 -25 -25 -20 -20 -15 -15 -10 -10 -5 -5 00 55 10 10 15 15 20 20 25 25 30 30 35 35 40 40
d = Distance between WiFi
transmitter and ZigBee receiver
Main WiFi lobe
d = 0,5m
d = 1m
d = 1,5m
d = 2m
d = 2,5m
d = 3m
d = 3,5m
d = 4m
F (MHz) F (MHz)
ZIGBEE: THE JOURNEY TOWARD DEPLOYMENT IN INDUSTRIAL APPLICATIONS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
73
3.1.1 Mitigation and Recommendations
Analyzing previous experimental results leads to two basic
installation recommendations to guarantee an 80% packet
delivery rate:
Distance of the WiFi interferer to the ZigBee nodes should
be at least 2 m.
Frequency offset between both networks should be at least
30 MHz.
These thresholds are formulated as safe-side values, i.e., it is
likely that many situations and environments could afford more
relaxed recommendations. They should be considered as upper
bounds, ensuring smooth coexistence of both networks.
Experimental results also showed that, thanks to the CSMA/CA
mechanism implemented in the 802.15.4 MAC layer, ZigBee
transmissions are never completely interrupted, even in presence
of heavy WiFi traffic. Other research groups obtained similar
results [5]. This suggests that some mitigation mechanisms can
be introduced in the ZigBee specification to detect interference
and move to another channel distant enough to prevent
performance degradation due to WiFi proximity. This feature is
currently under investigation in the ZigBee Alliance. Devising
such a functionality is obviously a difficult task given the many
technical issues to be solved (e.g., how to accurately detect an
interference, what this means for very large networks, and how
to deal with sleeping devices).
3.2 Power Consumption Issues
Another important issue for wireless sensor network applications
is power consumption. Current ZigBee deployment models
assume that both the coordinator and routers are permanently
mains-powered, while end devices can be battery-powered.
Typical battery usage scenarios reach a 1-to-10 year lifetime,
depending on duty cycling. Whether power consumption is a
limiting factor highly depends on application requirements.
At least three scenarios will require improvements in power
consumption to come to maturity:
Residential applications such as lighting and temperature
control are asking for very low energy consumption. Although
sensors and actuators are expected to be in sleep mode most
of the time, changing batteries is always a cumbersome
operation for the end customer.
Batteryless devices like switches powered via piezoelectric or
electromagnetic means are extremely attractive solutions for
this market segment. The availability of such technologies in
the future could significantly boost ZigBee deployments.
Large-scale wireless sensor networks deployed for
environmental or structural health monitoring will obviously
not accommodate mains-powered supply.
This means that energy must be provided to all ZigBee nodes
(including routers) through batteries or external sources such
as solar cells. A Profile Task Group within the ZigBee Alliance,
Wireless Sensor Applications, aims at specifying modifications
or improvements in the current stack to fulfill these objectives.
New energy saving or routing strategies are to be worked out.
Industrial applications in which sensors are hardly reachable
or attached to moving or rotating parts are natural candidates
for alternative powering models.
Enhancements in battery technologies and power scavenging
techniques are two relevant research directions with respect
to that.
4. THE PATH TOWARD PRODUCT
DEPLOYMENT
Although some technical hurdles are still to be cleared, ZigBee
is on its way toward reaching mass market deployment. This
section browses through the most promising application
segments and, for each one, describes some of the steps required
to eventually achieve this goal.
4.1 Building Automation
4.1.1 Business Drivers
Todays commercial buildings are managed by large sets of
sensors and actuators which control such functions as lighting,
heating, air quality and security. Optimizing energy efficiency
74 COPYRIGHT SCHNEIDER ELECTRIC, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
TABLE II. PERCENTAGE OF ERROR BETWEEN
THE MODEL PREDICTION AND THE ACTUAL
MEASURED ADDITIONS AND MULTIPLICATIONS
PER PIXEL. FOR EACH GOP OF EVERY
SEQUENCE, THE AVERAGE ERROR OVER A
NUMBER OF ADAPTATION POINTS IS PRESENTED.
like controllers, switches, and sensors for light, temperature, air
quality, or presence detection, as illustrated in Fig. 6. With no
wiring required, these devices can be placed almost everywhere
and easily moved when the building organization or functionality
need to evolve. As underlined by research company Frost &
Sullivan in [7], Whether used in new constructions or retrofit,
flexibility is the ultimate benefit in deploying a wireless system
as opposed to a wired network. These sensors can be located or
relocated to optimize system performance, increase customer
comfort, and adapt to changing floor plans.
Usually, deploying more sensors and actuators means obtaining
more energy-efficient buildings. By enabling the collection
of a much larger amount of data, wireless systems pledge
even greater energy and cost savings associated with better
optimization of lighting and HVAC (Heating, Ventilation, Air
and increasing the comfort of occupants are two strong drivers
for using complex building management systems. Tertiary
and commercial premises typically rely on standardized
communication protocols (such as BACnet and LonWorks) to
transport command and sensing data across building networks.
To date, such communication networks have been almost
exclusively deployed on wired media.
Although the current cost of a wireless solution may not be always
lower for new constructions, it is definitely a strong driver for
retrofit installations or facilities in which it is highly expensive
to install a wired communication network (e.g., buildings with
concrete walls, museums, and architectural or historical sites that
cannot be disturbed). In these scenarios, adding wired controls
and sensors may end up being much more costly than deploying
a wireless solution. ZigBee chips can be embedded in devices
FIGURE 6: WIRELESS SENSORS AND ACTUATORS IN A TYPICAL BUILDING ROOM
ZIGBEE: THE JOURNEY TOWARD DEPLOYMENT IN INDUSTRIAL APPLICATIONS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
75
Conditioning) functions. The increasing pressure to improve
building efficiency by even a few percentage points makes such
solutions worthwhile.
One may argue that these benefits are not particular to ZigBee
but to wireless technologies in general. The added value of a
ZigBee solution resides in vendor and product interoperability.
An open international standard that fosters competitive sourcing
and the availability of multi-vendor platforms holds the promise
of bringing new control options that previously were only
available to high-end markets. The influential presence in the
ZigBee Alliance of leading building automation suppliers (such
as Honeywell, Mitsubishi Electric, Schneider Electric/TAC, and
Siemens) and the recent release of early ZigBee products (such
as TACs Andover Continuum Wireless and Siemens APOGEE
Wireless) put this market segment at the forefront of upcoming
deployments. Business drivers are summarized in the table below:
providers without additional expenses in reworking the
entire solution.
Integration with existing building automation protocols is also
desirable. For example, liaison with BACnet is being
investigated by the ZigBee Alliance. Appropriate gateways
also need to be developed to connect ZigBee networks to
existing infrastructures such as Ethernet, LonWorks and
KNX, for instance.
Coexistence issues as described in section 3.1 have frequently
been utilized by opponents to wireless technologies to
promote traditional wired or powerline carrier solutions. In
addition to frequency agility mechanisms to be implemented
in the ZigBee stack, building management applications require
installation guidelines to address concerns expressed by end
customers. One can imagine devising ZigBee/WiFi node
distance and frequency separation recommendations (based
on conclusions like the ones presented in section 3.1) that
would translate into specific rules for hotels, malls, or
other commercial premises.
Eventually, deployments in real and large-scale environments
are needed to assess architectural choices (e.g., single large
multi-hop network covering a significant part of the building
vs. several smaller ZigBee networks with a limited number of
hops and interconnected with each other).
4.2 Home Control
4.2.1 Business Drivers
Along with building automation, home control is expected to
become one of the top market segments in terms of product
deployment. The type of use foreseen for ZigBee in this area
encompasses everything from domestic TV remote controls and
central heating to lighting, rolling shutters, and alarm systems.
Developing such devices requires great OEM involvement,
which in turn needs large markets to generate adequate return
on investment.
Although vendor and product interoperability represents a
strong confidence factor in gaining access to large markets, this
Drivers for Adopting ZigBee in Building Automation
Vendor interoperability ensured by an open international
standard
Competitive sourcing market and availability
of multi-vendor platforms
Flexibility to adapt to changing building arrangement
or functionality
Strong enabler for increased energy efciency in large
facilities
Lower installation and maintenance costs
(rst in retrot installations, then in new constructions)
4.1.2 Next Steps
The following issues need to be engaged with in order to
accelerate the deployment of ZigBee for building automation:
From a standardization point of view, completion of the
appropriate application profile work (CBA, BACnet study
group) is a prerequisite to holding the promise of
interoperability. Common application bricks will deliver better
interfacing capabilities with other equipment manufacturers,
allowing customers to shift to other product or service
76 COPYRIGHT SCHNEIDER ELECTRIC, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
is not the principal driver for adopting a standard in the home
automation segment. Cost issues are much more sensitive than
those in building automation systems. For OEMs, having an open
standard like ZigBee means a dynamic sourcing market that drives
costs down and boosts technological innovation. Eventually,
these competitive pricing strategies will benefit end customers.
Beside cost issues, power consumption is also an important
factor, as mentioned in section 3.2. Allowing customers to easily
reposition switches and other low-end products while getting rid
of power mains is a strong driver for adopting a technology like
ZigBee. Although batteries are not foreseen to disappear soon,
scavenging energy to make control devices fully autonomous
brings great prospects to both new construction and renovation
markets. Principal business drivers for residential control
applications are summed up below:
Wireless coexistence is also an important concern for home
control product manufacturers. In contrast with building
automation environments, enforcing network installation rules
and recommendations might be more problematic among end
customers and installers.
While carefully choosing WiFi channels or constraining WiFi
throughput during heavy operations like streaming is feasible
in a commercial environment, this is clearly not achievable at
home. Addressing interference problems will need to rely mainly
on built-in mechanisms, such as frequency agility.
4.3 Automated Meter Reading
4.3.1 Business Drivers
Automated meter reading (AMR) was first deployed 40 years
ago by AT&T and a group of utilities. At the time, phone-based
services were provided at about four times more than the cost of
a person to read the meter. After these successful but expensive
experiments, AMR enjoyed constantly decreasing operating costs
and became a widely accepted technology. Today key drivers for
installing an AMR system are lowering costs, avoiding reading
errors, and collecting relevant data as quickly as possible.
The use of wireless technologies is mainly propelled by the
prospect of further cost reductions. According to research
house ON World, wireless-based AMR can cut water, gas, and
electricity costs by up to 25%. In a world where utility costs
are rising by as much as 10% per year, such figures are of great
appeal to this industry. ON World believes wireless networks
will grow the fastest of all AMR networking solutions over the
next years. New products are appearing at a fast pace for all
geographical needs (urban, suburban, rural) and across various
application requirements (industrial metering, utility metering,
and sub-metering in large commercial facilities).
Positioned on the low-cost edge of wireless technologies,
ZigBee can be considered as a strong contender in the race for
advanced AMR solutions. Aggressive pricing trends are likely
to be their best advocates. Although latency is not a critical
Drivers for Adopting ZigBee in Home Control
Competitive sourcing market and multi-vendor platforms
Low power consumption / Possibility for future batteryless
solutions
Flexibility to reposition control and sensing devices across
the home
Increased energy efciency and comfort
Vendor interoperability ensured by an open international
standard
Lower installation costs (rst in retrot installations,
then in new constructions)
4.2.1 Next Steps
A successful deployment roadmap for home control applications
has to deal mainly with the following provisions:
ZigBee standardization is more advanced on the residential
side than on the commercial building side. Although the HA
application profile has been completed, there are still some
uncertainties as to whether to go for ZigBee or ZigBee Pro as
the underlying stack profile. Many OEMs are willing to make
use of the forthcoming ZigBee Pro stack profile to benefit
from its advanced features.
ZIGBEE: THE JOURNEY TOWARD DEPLOYMENT IN INDUSTRIAL APPLICATIONS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
77
factor as it is for lighting, for instance, guaranteeing secure
data delivery is of prime importance. Against other low-rate
wireless technologies, ZigBee may also play the relatively high-
performance advantage. Principal business drivers for metering
applications are listed below:
Given the large and diverse range of communication requirements
in industrial environments, the deployment roadmap needs to be
split into two major application categories:
Open-loop applications refer to typical monitoring and data
collection tasks that are not part of the control loop itself.
Wireless communications are used in these cases for non-
critical operations such as commissioning, diagnostics,
condition monitoring, etc. ZigBee is here well-positioned in
terms of cost efficiency and flexibility to relocate industrial
devices across the plant.
Closed-loop applications are in contrast related to critical
tasks since they are part of the control loop. Examples include
real-time process control or safety-related operations like
commanding valves. Such scenarios will likely not be part
of wireless technology roadmaps for 3 to 10 more years.
In summary, one can pinpoint the following main business
drivers for adopting ZigBee in the industrial market segment:
4.3.1 Next Steps
In contrast to building and residential markets, interoperability is
not among the main drivers for AMR applications. Deployment
of AMR solutions is still highly fragmented across regional or
local markets, and utilities and OEMs enjoy the possibility of
providing legacy systems, if cost requirements are fulfilled.
From that perspective, having a non-ZigBee but IEEE 802.15.4-
based solution may be a viable option for some AMR players.
Nevertheless, the availability of a public ZigBee profile
addressing AMR needs may facilitate and accelerate product
development. Resolving current uncertainties around this public
profile will be key in adopting ZigBee in this industry.
4.4 Industrial Automation
4.4.1 Business Drivers
Industrial automation is a rather ambiguous market segment
for ZigBee. Although initially presented as one of its primary
target industries, industrial control now tends to be perceived
as a likely late adopter, given the specific constraints of
manufacturing and process environments. Robustness and real-
time communication requirements are often cited as potential
obstacles to using wireless technologies for industrial control.
An even bigger impediment is the conservative mindset that
characterizes the adoption of new technologies among industrial
automation customers.
Drivers for Adopting ZigBee in Automated Meter Reading
Cost reductions in deploying and monitoring utility
networks
High performance and built-in security
Competitive sourcing market and multi-vendor platforms
Drivers for Adopting ZigBee in Industrial Automation
Flexibility to adapt to changing plant conguration or
functionality
Lower installation and maintenance costs
Competitive sourcing market and availability of multi-
vendor platforms
Vendor interoperability ensured by an open international
standard
4.4.1 Next Steps
The first set of applications that will be ready for deployment
in manufacturing and process environments are open-loop
systems. Combined wired/wireless networks might be the first
instantiation, in which control data are transported through wires,
and a wireless link is used for commissioning and configuration
purposes. Other promising areas include the monitoring of
process variables, production equipment, or clamp-on types of
instruments for permanent or temporary installations. While
acquiring sensing data is fairly easy, conveying these data
78 COPYRIGHT SCHNEIDER ELECTRIC, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
to a central control system remains a significant challenge
since it requires tight integration with existing automation
infrastructures.
If a large number of devices is to be deployed, power
consumption may become a critical issue. Changing batteries
frequently is either not feasible or not wanted by plant operators.
This means that advanced energy optimization techniques need
to be devised to maintain an acceptable service level while
significantly decreasing duty cycles.
Further down the road are closed-loop applications involving
wireless communications. Several technical hurdles, such
as guaranteeing high quality of service and implementing
appropriate time synchronization mechanisms, need to be
addressed before these scenarios become reality. It is still
unclear at this stage whether ZigBee will be the most appropriate
technology to take up these challenges.
5. CONCLUSIONS
This paper has highlighted some important challenges on the
road toward deploying ZigBee in residential, commercial, and
industrial applications. Although much work has been carried
out since the completion of the first specification in December
2004, there still are some issues to be addressed before we
will see mass market adoption. In particular, this article has
pinpointed three important matters:
Finalizing public application profiles and elaborating
appropriate compliance testing and certification programs.
The most urgent actions are clearly required on the commercial
building side since interoperability is a less attractive feature for
metering, and to some extent industrial applications. The question
of choosing which stack profile for which application profile
also needs to be sorted out. Having several stack profiles could
also be questioned.
Addressing end customer concerns, in particular coexistence
with WiFi networks. This article has presented experimental
results showing that, although interference may have to be
dealt with, mitigation techniques can be implemented. ZigBee
traffic is never completely interrupted, and installation
rules can greatly help reduce the impact of WiFi perturbations.
Improving power usage of current and future energy sources.
Evolution of ZigBee deployment scenarios will have to align
with progress made on the power technology side. One can
expect an increasing demand for better power consumption
schemes involving either improved batteries or batteryless
systems relying on alternative scavenging methods. Such
enhancements will gain increasing momentum in residential
and industrial market segments.
The journey toward mass market deployment of ZigBee
technology is not over yet but is heading the right way. The
recent involvement of large OEMs and the increasing number of
ZigBee radio solutions indicate that the market is now ready for
end-product offerings. Adopting an open, international standard
that provides multi-source development platforms is today
the best strategy for OEMs to be successful in their markets.
From that perspective, ZigBee today appears as the only viable
alternative in the low-cost, low-power, wireless control segment.
The capability of the ZigBee Alliance to work out the remaining
technical and application profile issues will be instrumental in
accelerating mass market adoption.
REFERENCES
[1] ZigBee Alliance website: www.zigbee.org
[2] IEEE Std 802.15.4-2003. Part 15.4: WIRELESS
MEDIUM ACCESS CONTROL (MAC) AND PHYSICAL LAYER
(PHY) SPECIFICATIONS FOR LOW-RATE WIRELESS
PERSONAL AREA NETWORKS (LR-WPANS),
1 October 2003.
[3] ZIGBEE SPECIFICATION, Version 1.0.
14 December 2004.
ZIGBEE: THE JOURNEY TOWARD DEPLOYMENT IN INDUSTRIAL APPLICATIONS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
79
[4] A. Gutirrez, E. H. Callaway, R. L. Barrett,
LOW-RATE WIRELESS PERSONAL AREA
NETWORKS ENABLING WIRELESS SENSORS
WITH IEEE 802.15.4, IEEE Press. 2004.
[5] A. Sikora, COMPATIBILITY OF IEEE 802.15.4 (ZIGBEE)
WITH IEEE 802.11 (WLAN), BLUETOOTH, AND MICROWAVE
OVENS IN 2.4 GHZ ISM-BAND TEST REPORT,
Steinbeis-Transfer Center, University of Cooperative
Education, Lrrach. 12 September 2004.
[6] MC1319x Coexistence. Freescale Semiconductor
Application Note AN2935. July 2005.
[7] Wireless Sensors in Building Automation
Technical Insights, Frost & Sullivan. February 2005.
CONTACT: ST.JOURNAL@ST.COM
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
80 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF CALIFORNIA - LOS ANGELES, 2007.
LOCALIZATION
IN SENSOR NETWORKS
In this paper, we review the concepts
of source and node localization with
reference to a sensor network. We
introduce several algorithms that have
appeared in the literature as well as
some that have been developed and
used by the authors over the years. The
methods presented are based on different
measured signal quantities, such as
received signal strength, angle of arrival,
time or time-difference of arrival, all of
which are defined and described in the
paper.
1. INTRODUCTION
The progress made in communication, computer, networks,
and microelectronics in the last fty years made possible the
emergence of sensor networks (SNs). In recent years, SNs
constitute one of the most active academic research topics
in the system area. It is believed that sensor networking in
the 21st century will be enormously signicant by providing
measurement of the spatial and temporal physical phenomena
around us, leading to a better understanding and utilization of this
information in a wide range of applications. Sensor networking
will be able to bring a ner-grained and fuller measurement
(using acoustic, seismic, magnetic, infrared, imaging, etc., data)
and characterization of the world around us to be processed and
communicated, so the decision makers can use the information to
take actions in near-real-time.
Sensor Networks combine the technology of modern
microelectronic sensors, embedded computational processing
systems, and modern communication and computer networking
methodology. However, due to the limited power and energy from
the battery, hostile radio propagation environment, constrained
radio and processor capabilities, and random/unknown sensor
placements, the analysis, design, and realization of sensor
networks to perform meaningful tasks remain challenging.
K. Yao
(1)
, F. Lorenzelli
(2)

(1) University of California, Los Angeles
(2) STMicroelectronics
LOCALIZATION IN SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
81
1.1 Localization
Localization, the determination of the location of an object,
whether that of a source emitting some energy (e.g., RF, acoustic,
ultrasonic, optical, IR, seismic, thermal, etc.) of interest, or a
sensor node in the SN, is a major issue of interest. Two families of
algorithms can be identied under the heading of localization:
those used to locate the individual sensors in a SN with respect to
each other and possibly to a point of reference will be referred to as
node localization algorithms. This scenario arises when the nodes
are not (or cannot be) placed with desired accuracy, due perhaps
to a hostile environment or other reasons. Once the sensors
locations are known, either through a preliminary phase of node
localization, or by accurate placement and calibration, then it
becomes possible to locate external sources of energy. When this
is the case, we will be talking of source localization algorithms.
As an example of node localization familiar to many, in the U.S.,
the Federal Communications Commission (FCC) has mandated
the E911 regulation, (Europe has recommended the E112
regulation), requiring the cellular telephone provider to determine
the location of a cell phone user (in an emergency mode) to tens
of meters. Source localization may arise in various military
scenarios, such as the localization of a vehicle, or in automated
manufacturing, for instance the localization of a robotic device.
Localization of an object with an appropriate receiver can be
achieved by many means. Satellite-based localization based
on global positioning systems (GPS), and variations including
the Global Navigation Satellite System (GLONASS), and the
European Geostationary Overlay Service (EGNOS), have been
used for many years for military, commercial, and consumer
applications. In open-eld scenarios, these localization systems
may be adequate. However, localization based on these systems
may not be appropriate due to their limiting precisions caused
by severe propagation degradations inside buildings and in
non-line-of-sight (NLOS) situations, or due to the exceeding
complexity of the overall system, especially in low-cost SN
deployment situations.
In this tutorial paper, we introduce the basic operations, including
some equations, and their applications, to various localization
schemes. In Section 2, we deal with several methodologies
for sensor node localization. We explore the issue of statistical
characterization of the measurements and how it affects the
performance of the location estimators. We also describe
a Gauss-Newton iterative node localization method in its
centralized and distributed versions. In Section 3, we deal with
several methodologies for source localization. They include, the
trilateration/multilateration method, time-difference of arrival
(TDOA) method, the received-signal-strength (RSS) method,
and the direction-of-arrival (DOA) method (also called the angle-
of-arrival (AOA) method).
2. NODE LOCALIZATION
In sensor networks it is assumed that small, inexpensive,
collaborative, and relatively autonomous nodes are placed with
varying degrees of accuracy, with the purpose of monitoring
the environment. Depending on the applications, sensors are
required to respond to changes in the environment by sending
out the collected information to neighboring nodes or to a central
station. For these reasons, the nodes are equipped with sensing,
computation and wireless communication devices.
Applications may range from earthquake to agricultural (water
and fertilizer) monitoring, from automobile trafc control to
security alarm systems, etc. In many situations, it is paramount
that the sensors be aware of their relative, and sometimes
absolute, location. This is known as a node localization problem
in the sensor network literature. Only by knowing the position
of the sensor can the sensed information be placed in the context
and made relevant for the application at hand. It is assumed that
the individual devices are provided with small batteries that are
replaced relatively infrequently [18]. Because of this, localization
techniques based on global positioning systems may be considered
too power-hungry, and alternative techniques are then called for.

The different techniques for node localization can be generally
classied into two categories: localization based on anchors (or
82 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF CALIFORNIA - LOS ANGELES, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
beacons), and anchor-free localization [21]. Anchor nodes are
special nodes equipped with special positioning devices. Other
nodes try to determine their position relative to the anchors.
Once their positions have been estimated, regular nodes may
in turn become anchors, etc. Clearly, anchor-based localization
algorithms need another positioning system in place to provide
initialization information. This may be attained outdoors with
special nodes having global positioning system information,
or indoors by nodes whose positions are known by design. In
anchor-free localization algorithms, all nodes collaborate with
each other (usually with their neighbors) in order to determine
a relative map. A post-processing is subsequently performed to
convert the relative map into an absolute position information.
Another categorization that can be made of localization
algorithms is among centralized, localized and distributed
algorithms [11]. In centralized algorithms, all the information
collected by the sensors is transferred to a central node that
performs the localization algorithm and computes each nodes
location. The problem to solve typically is to search for a global
maximum of a likelihood function. This problem is in general
very hard, because of the multi-modal structure of likelihood
functions in most realistic situations. In addition, the function
to maximize is usually model-dependent, and any deviations
from the assumed model may signicantly degrade the estimates.
The problem may be formulated as a convex program, and
semidenite programming formulations have been proposed [1].
The localization algorithm may be sometimes distributed, so that
each node runs a subset of the operations required to obtain the
localization estimates, and passes information to its neighboring
nodes. In this way, the communication bottleneck due to the need
to forward all the information to a central processor is obviated.
More energy will probably be spent to exchange information
locally, as opposed to what is required to send all the information
to the central node, and it is conceivable that a trade-off can be
achieved. After a few iterations and a few information passes
between adjacent nodes, the localization information should
converge for all nodes. Ideally, the distributed algorithm will
nd the same global optimum as the centralized algorithm. In
making the algorithm distributed, convergence issues need to
be addressed to insure that this is indeed the case. In particular
circumstances, the position can be computed locally by the
individual nodes, for instance when each node receives a signal
from a set of beacon nodes, and estimates its own location
based on the received signals. Localized algorithms use the
estimated distance between the node and the anchor nodes in
much the same way as trilateration and multilateration in source
localization algorithms. It is generally assumed that wireless
sensor networks, in order to minimize installation expenses, are
relatively sparse, and anchor nodes may be separated by large
distances. Moreover, low-power sensor nodes are usually not
in the range of a sufcient number of anchors. In these cases,
localized algorithms are probably not the best choice.
Many localization algorithms rely on measurements of distance
between nodes [13]. These algorithms are known as range-based.
The types of distance estimation that is usually performed are
based on measurements of received signal strength (RSS), or
estimates of time of arrival (TOA) or time difference of arrival
(TDOA). As opposed to range-based algorithms, range-free
techniques do not rely on point-to-point distance estimation, but
try to estimate the nodes locations by computing quantities such
as the angle of arrival (AOA).
Not all localization algorithms make use of the characteristics
of the signals exchanged among sensors (such as amplitude,
phase or delay). Some algorithms are based on the network
connectivity information. A typical case in point is the family
of algorithms that estimate the distance between nodes by the
number of hops required for the packets to travel from source to
destination nodes. These types of algorithms will not be further
discussed in this paper.
2.1. Measurement Characteristics
The parameters to estimate are the coordinates of the sensor
nodes. The assumption is that in order to compute these estimates,
LOCALIZATION IN SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
83
physical quantities of the physical signals are measured, such as
RSS, AOA, TOA or TDOA. The signals involved can be seismic,
acoustic, RF, etc., or combinations thereof, depending on the
applications. Naturally, the measurements as well as the anchors
positions are affected by a degree of uncertainty. Range and angle
measurements are degraded by both time-varying errors (such as
noise or interference) and environment-dependent errors. In the
assumption that sensors are arranged in a mostly static network,
environment-dependent errors are usually static in nature and due
to the arrangements of objects (trees, buildings, furniture) in the
area of operation of the sensor network. Time-varying errors can
be averaged out by taking multiple measurements over time. The
statistical characterization of these measurements is important to
assess the estimation quality of the various algorithms. Usually
a number of simplifying assumptions are made. For instance,
measurements are typically assumed independent. In reality,
correlations among measurements exist due to obstructions, or
channel characteristics etc. Another assumption is the choice
of statistical distributions used to characterize the different
quantities. This choice is usually backed up by experiment, even
though often there simply is not enough data to give a complete
characterization.
In the following we consider the four quantities, RSS, AOA,
TOA and TDOA individually.
2.1.1 RSS
RSS is the strength (voltage) of the signal measured by the
receivers received signal strength indicator (RSSI) in arbitrary
units. Sometimes, it is reported as the squared magnitude of the
received signal strength. In the 802.15.4 PHY standard, RSS can
be measured by the link quality indication (LQI), used to report
the signal strength of the received packet to the higher layers.
RSS measurements are relatively inexpensive and simple to
implement, and for this reason they are widely used in localization
algorithms. Unfortunately they are also notoriously unpredictable.
The attenuation increases with distance (for instance, in free space
signal power decreases with the second power of the distance),
and is averaged over fast fading. Environment-dependent errors
are due to obstructions that the signal must go through or diffract
around. These effects are modeled as random. The ensemble
mean power at distance d is usually modeled as

P(d ) = P
0
10n
p
log
d
d
0
,
where P
0
is the received power in dBm measured at a reference
distance d
0
. The parameter n
p
is the path-loss exponent and varies
between two and four. The difference between the measured
power and its ensemble average is assumed to be log-normal
distributed (i.e., Gaussian, when expressed in decibels). This
model is justied by measurements and analysis [14,16]. The
standard deviation, expressed in dBm,
dB
, is relatively constant
with distance and varies between four and twelve. Because
the standard deviation is constant with range when measured
in decibels, the RSS-based range estimates have variance
proportional to the actual range. RSS errors are for this reason
multiplicative in nature, as opposed to the additive TOA errors.
RSS measurements depend on the manufacturing and calibration
of both the transmitter and the receiver, as well as transmitters
battery level, and this causes much of the unpredictability of
these measurements.
2.1.2 AOA
Information regarding the angle of arrival may be obtained by
using array processing techniques. Each sensor node has to be
equipped with at least two sensing devices (e.g., microphones or
antennas). The AOA is computed by calculating the difference
in times of arrival or, in case of a narrowband signal, in phase
difference. An alternative method makes use of difference in
RSS at two directional antennas placed on the sensor, for points
where the beam-patterns overlap. In either case, the need for
multiple sensing devices at each node may cause exceeding cost
and size. In other situations, such as environmental monitoring,
acoustic sensor arrays may be a very reasonable option [4]. AOA
is affected by noise and multipath. In particular, all the SNR of
the line-of-sight (LOS) path may be low due to multipath, to
(1)
84 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF CALIFORNIA - LOS ANGELES, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
TOA estimates rely on network synchronization. In case the
network cannot provide a level of synchronization accuracy
suited to the signal (good accuracy for acoustic signals may not
be sufcient for RF signals) [12], or in case of asynchronous
networks, two-way TOA measurements are made, assuming that
the internal delay between signal reception and re-transmission is
known or can be estimated. Alternatively, the estimation problem
can be augmented by including in the parameters to be estimated
the sensors clock biases.
2.1.4 TDOA
Time difference of arrival of the signal at two different antennas
can be used for localization purposes, with the advantage that
network synchronization information does not need to be
communicated or estimated.
2.2. Performance Bounds
The goodness of an estimator to provide a good estimate of
the desired parameters, in our case the node positions, can be
computed by evaluating the lower bound to the estimator mean
square error. Such lower bound takes the name of Cramr-Rao
bound (CRB) [15]. This bound can be a very useful tool to the
designer as well, not only because it sets a reference point to the
estimator accuracy, but also because it is function of the design
parameters, such as the network geometry, channel parameters,
number of sensors and number of anchors, etc.
For instance, in the case of TOA, the CRB takes the form
CRB(TOA) =
1
8
2
BT
s
f
c
2
SNR
,

where B

is the signal bandwidth, f
c
its central frequency and T
s
its
duration. The CRB shows that better accuracy is expected when
the bandwidth or the transmitted power is increased. The CRB
for the TOA-based location estimates does not change with a
scaling of the network geometry, while the CRB of RSS or AOA-
based estimates is actually proportional to the size of the system
[11]. The CRB is proportional to
T
for TOA and

for AOA,
the point that the LOS path may not be the strongest arriving
signal. In some cases, there may not even be a LOS path, due
to obstructions. In this case the AOA estimate will be incorrect.
Non-LOS (NLOS) situations may be detected by computing the
variance of the received signal strength, which will be much
higher in a NLOS situation. Even when a LOS path is present, its
peak may be smeared by early-arriving multipath signals. AOA
measurements are usually modeled as Gaussian, with ensemble
mean equal to the true value and standard deviation,

, on the
order of two to six degrees. Because the estimate depends on the
orientation of the sensors, it is sometimes required to add the
unknown orientation to the parameters to estimate.
2.1.3 TOA
TOA is the time of arrival due to the propagation of the signal from
transmitter to receiver and depends on the nature of the signal
being sent. For instance, RF signals travel at 10
6
times the speed
of sound. The travel time can be computed in a synchronized
network, or from the round-trip time from transmitter to receiver
and back. Time delays in the transmitter and receiver may also
affect the estimate. As for AOA, measurements are affected
by noise and multipath. In NLOS situations, the estimate will
have a positive bias that needs to be corrected. Denser networks
allow for more accurate TOA measurements between nearby
nodes because of the higher power of the LOS component.
Accuracy also increases with the bandwidth of the signals, even
though this requires in turn a higher speed signal processing
and potentially higher costs. The statistical model of choice for
short-range TOA measurements is Gaussian, with mean equal
to d
ij
/
p
+
T
and standard deviation equal to
T
, where
p
is
the propagation velocity and d
ij
is the inter-node distance. For
instance, measurements of RF wideband signals have reported
values for
T
in the order of units to tens of nanoseconds and

T
in the order of units of nanoseconds. In order to account for
large observed errors, which point to a fatter-tailed distribution
than Gaussian, mixture distributions have been proposed. A two-
mode Gaussian mixture can be used to model the case of LOS
and NLOS situations [13].
(2)
LOCALIZATION IN SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
85
while it is proportional to d
dB
/n
p
for RSS. This result shows that
a higher path-loss exponent makes the estimates more accurate,
at the expense of higher transmitted power. From the CRB results
(which of course are a rst-order approximation because do not
take into account secondary effect, such as channel parameter
variation with path length), it appears that TOA-based estimates
are less sensitive than AOA or RSS to inter-node distances, and
are therefore preferable in sparser networks. AOA, TOA and
TDOA can achieve higher accuracy than RSS. On the other
hand, RSS-based estimates can be made more accurate in denser
networks, with shorter inter-node distances, and are considerably
cheaper than the alternatives. Whenever possible, it is advisable
to use the different estimation algorithms in combination,
especially when they seem to have complementary performance.
2.3 Estimators
Most estimators compute the sensors positions based on an
optimization criterion, such as the maximization of a likelihood
function, or the minimization of a least-squares problem. The
least-squares problem may in turn be weighted or regularized.
As examples of estimators, we present below an algorithm based
on a Gauss-Newton search, both in its centralized and distributed
formulation [8].
2.3.1 A Distributed Gauss-Newton Method
for Node Localization
Suppose the K anchor sensors at coordinates a
k
,

k = 1,, K, are
placed in known locations, while the x
i
,

i = 1,, N, sensors at
coordinates d
ij
, have unknown locations. Let d
ij
be the measured
distance between nodes i

and j and assume this distance has been
obtained through RSS or TOA measurements. The problem is
then to solve the following nonlinear least-squares problem with
respect to the node locations, x
i
,

i = 1,, N,

F ({x
i
}) = | r |
2
+ | s |
2
, min
{x
i
}
i, j i, k
ij ik
|

|
|
|
|
|
|

|
|
|
|
|
min
{x
i
}
where F({x
i
}) is the global cost function, r
ij
,

= ||x
i
x
j
||
2

d
2
ij

represents the residual error in the placement of pairs of sensors
i and j and s
ik
,

= ||x
i
a
k
|| d
2
i k
is the error in the placement of
sensor i with respect to anchor k. The calculation is restricted to
only the pairs of nodes within a given radius from each other (the
so-called radio range). This (centralized) problem can be solved
by means of a Gauss-Newton method. This algorithm can be
turned into a distributed algorithm by dening the cost function
at node i as
F
i
= | r |
2
+ | s |
2
,
j i i k
ij ik


where j i and k i represent the indices of the nodes and
anchors that are neighbors (in the radio range) of node i. It is
assumed that the nodes are ordered numerically, as indicated
by their indices. Each node will try to minimize the function
F
i
, assuming that all nodes adjacent to node i and index j < i
have been updated in the current iteration, and all the nodes
adjacent to node i and index j > i have been updated in the
previous iteration. In this way, the distributed, but sequential,
algorithm is performing the same operations as the centralized
algorithm above. Each local minimization can be performed by
approximating the problem with a linear least-squares problem
and then using the Gauss-Newton method.
The algorithm can once more be changed in such a way as to
become distributed and parallel. The problem being solved is
now a rst-order approximation of (4), and can be written as a
linear least-squares problem as follows
| r |
2
+ | s |
2 T T
ij ij
p s
i
. min
{p
i
}
ij ij
|

|
|
|
|
|
|

|
|
|
|
|
p
i k j i
r
i


If p
i
is the solution to the problem above, then the update of the
coordinates of sensor i is given by

x
i
: =x
i
+
i i p
,
for a properly chosen step size
i
. The calculations can be
performed in parallel in all nodes. In general, the parallel
algorithm requires a slightly higher number of iterations than
(3)
(4)
(5)
(6)
86 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF CALIFORNIA - LOS ANGELES, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
the sequential algorithm for the same accuracy, and thus the
computation complexity is slightly increased. Nonetheless the
parallel algorithm has much faster processing time because all
computations occur simultaneously. The step sizes can be chosen
so as to make the functions F
i
non increasing with the iteration
index. Although it is possible that some nodes will require at
times
i
= 0, in all the simulations performed to date the step
size has always been strictly positive and the functions F
i
strictly
decreasing from iteration to iteration.
Fig. 1 shows an example of N

= 100 sensors and K

= 10 anchors
placed randomly in a 1.6 X 1.6 region where the radio range is
0.35 and the noise factor is 0.1. The starting points are chosen
as random perturbations from the true sensor locations, where
the perturbation is given by a uniformly distributed random
variable in [0.1, 0.1]. The estimation results of the centralized,
sequential and parallel algorithms are given in Fig. 1, represented
by x. The true sensor locations are given by the o. Anchors
are denoted by . It can be seen from the gure that distributed
algorithms converge to almost the same results as the centralized
Gauss-Newton algorithm.
The parallel algorithm has been compared to the distributed
triangulation (or one-hop multilateration) method in [19] and
[20], where the node locations are calculated in two steps:
rstly an approximate distance is calculated by computing
the shortest hop distance and using knowledge of an average
hop size and secondly a more accurate distance estimate is
obtained by iteratively solving a local least squares problem.
Fig. 2 shows comparisons of the convergence behavior of the
parallel algorithm and the distributed triangulation method over
5 random realizations.
-0,8
-0,4
0
0,4
0,8
-0,8 -0,4 0 0,4 0,8
true location
estimation
anchor
FIGURE 1: ESTIMATION RESULTS OF THE CENTRALIZED, SEQUENTIAL
AND PARALLEL ALGORITHMS.
0 2 4 6 8 10
10
-2
10
0
10
2
10
4
10
6
Iteration
V
a
l
u
e

o
f

g
l
o
b
a
l

c
o
s
t

f
u
n
c
t
i
o
n
Triangulation
Parallel algorithm
FIGURE 2: ESTIMATION ERROR VERSUS ITERATION FOR DIFFERENT
ALGORITHMS. THE DIFFERENT CURVES REPRESENT 5 DIFFERENT
RANDOM REALIZATIONS.
40 80 120 160 200
0
5
10
15
20
network size (n)
a
v
e
r
a
g
e

n
u
m
b
e
r

o
f

i
t
e
r
a
t
i
o
n
s
Sequential
Parallel
FIGURE 3: AVERAGE NUMBER OF ITERATIONS.
We subsequently increase the size of the network, N, while
keeping the density of both sensors and anchors unchanged, that
is, N sensors and N/10 anchors located randomly in a 1.6N/100
LOCALIZATION IN SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
87
x1.6N/100 region. Fig. 3 shows the average numbers of iteration
over 20 realizations for both sequential and parallel algorithms to
reach a given accuracy (within 0.01 from the true location). The
radio range is 0.35 and the noise factor is set to zero in all cases.
It can be seen that the sequential algorithm requires a few more
iterations.
On the other hand, the total processing time in Fig. 4 is much
larger for the sequential algorithm. This is because the steps p
i
Fig. 5 shows the average energy consumption over 20 realizations
of different algorithms. It can be seen that when the size of the
network is large, distributed algorithms have a reduced energy
consumption.
3. SOURCE LOCALIZATION
Distributed sensor networks (DSNs) have been proposed for a
wide range of applications. The network may be used to monitor
an area for military, environmental, and manufacturing purposes.
In these and other applications, an important task is to perform
source localization by estimating the location of one or more
stationary or moving objects. The object(s) may be radiating
one more more form(s) of energy(ies). Sensor nodes in the
DSN, using one or more modality of reception can collect the
signature(s) of the object(s) and perform source localization.
3.1 Trilateration/Multilateration
Trilateration is conceptually the simplest method to perform
source localization of a single source based on range estimation.
Here we assume each sensor node can estimate the range from
the source to itself. There are various methods on how this range
estimate can be made in practical scenarios. In order to nd the
location of the source, we need to reference it with respect to
some coordinate system. Furthermore, one must assume that the
1
10
40 60 80 100 120 140 160 180 200
network size (n)
a
v
e
r
a
g
e

p
r
o
c
e
s
s
i
n
g
t
i
m
e

(
i
n

l
o
g

s
c
a
l
e
)
Sequential
Parallel
FIGURE 4: AVERAGE PROCESSING TIME (IN LOG SCALE).
in the sequential algorithm must be computed sequentially, while
in the parallel algorithm they can be computed simultaneously
(for simplicity, we ignore the processing time for nding the step
length).
Finally, we present some empirical results on the communication
energy cost. We use the model given by [17] that the propagation
loss is proportional to the fourth power of distance. For
centralized algorithms, each sensor only needs to transmit to the
central processor once.
However, as the size of the network increases or, equivalently, the
region increases, each sensor will need more energy to transmit
the information to a central processor (we assume the central
processor lies at location (0,0)). In distributed algorithms, each
sensor only communicates to the neighboring sensors. Thus the
total energy cost depends linearly on the total number of sensors
and number of iterations required.
c
o
m
m
u
n
i
c
a
t
i
o
n

e
n
e
r
g
y

c
o
n
s
u
m
p
t
i
o
n
Sequential
Parallel
Centralized
40
0
50
100
150
200
60 80 100 120 140 160 180 200
network size (n)
FIGURE 5: AVERAGE ENERGY CONSUMPTION DUE TO COMMUNICATION.
88 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF CALIFORNIA - LOS ANGELES, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
locations of the sensor nodes be known with respect to the same
coordinate system. Again, the issue of determining the locations
of the sensor nodes have been discussed in some details in
Sec. 2.
In trilateration, consider all the range-sensing nodes and the
source are situated on the xy plane. Range-sensing nodes A,
B, and C are located at (x
A
, y
A
), (x
B
, y
B
), (x
C
, y
C
) as shown
in Fig. 6. If node A estimates a source at range d
A
and B node
estimates that source at range d
B
,

then the intersection of these
two circles yields two possible ambiguous source locations
marked by S and S. Similarly, if node C estimates a range of
d
C
, then the intersection between node B and node C yields
two possible ambiguous source locations S and S, while the
intersection of node C and node A yields two possible ambiguous
source locations of S and S. However, the true location of the
source is at the intersection of all three circles at S. Of course, in
the presence of noisy estimations on d
A
, d
B
, d
C
,

the three circles
will not intersect at a single point to yield a unique location S.
In practice, denote the source coordinates (x, y), then d
A
, d
B
, d
C

satisfy the equations of
(x

x
A
)
2
+ (y

y
A
)
2
= d
2
A
,

(x

x
B
)
2
+ (y

y
B
)
2
= d
2
B
,

(x

x
C
)
2
+ (y

y
C
)
2
= d
2
C
.

The solution [x, y]
T
of (7) can be written as the least-squares
(LS) solution of
x
y
|
|
|
|
|
|
=
1
2 2
2(y
B
y
C
)
(y
A
y
C
)
2(x
B
x
C
)
(x
A
x
C
) x
x
B
2
x
C
2
y
B
2
y
C
2
+d
C
2
d
B
2
A
2
x
C
2
y
A
2
y
C
2
+d
C
2
d
A
2
.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|

In multilateration, N range-sensing nodes with known locations
(x
1
, y
1
), (x
2
, y
2
), (x
N
, y
N
) are used instead of three nodes
in trilateration. Their distances from the unknown source are
denoted by d
1
, d
2
, d
N
.

Then corresponding to (7), we have
( x x
1
)
2
+ (y y
1
)
2
= d
1
2
,
( x x
2
)
2
+ (y y
2
)
2
= d
2
2
,
%
(xx
N
)
2
+ (y y
N
)
2
= d
N
2
.
The LS solution of Ar
s
=

b corresponding to (8) is now given
by
r
s
= (A
T
A)
1
A
T
b ,

where
x
y
,
|
|
|
|
|
|
r
s
= A =
2( x
1
x
N
) 2( y
1
y
N
)
% %
2( x
N1
x
N
)
|
|
|
|
|
|
|
|
|
|
,
2 ( y
N1
y
N
)


b =
%
|
|
|
|
|
|
|
|
|
.
|
|
|
|
|
|
|
|
|
x
1
2
x
N
2
+ y
1
2
y
N
2
+ d
N
2
d
1
2
x
N1
2
x
N
2
+ y
N1
2
y
N
2
+ d
N
2
d
N1
2
2. LINEAR INTERSECTION
Consider now the situation where two range-sensing nodes,
A and B compute their distances from the source S as d
A
and
d
B
, and assume it is known that the points A, B and S are in
a clockwise sequence on the xy plane. Also assume that the
distance between A and B is known and equal to d. By simple
geometric considerations, one can dene the two quantities
(7)
(8)
d
d
B
S''
B
d
c
C
S'''
S'
A
S
A
FIGURE 6: TRILATERATION BASED ON THREE CIRCLES.
(9)
(10)
(11)
,
LOCALIZATION IN SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
89
=
dA
2
d B
2
+d
2
2d
2
, =
d A
2
d
2

2
,
and calculate the source location as x

= x
A

+ (x
B
x
A
)

+ (y
B
y
A
), y

= y
A
+ (y
B
y
A
)

+ (x
B
x
A
). Of course this formulation
assumes that the ranges are computed with no error. In case more
node pairs are available, all the computed source locations can
be combined in a weighted average, where the weights can be
chosen according to an assumed statistical distribution of the
errors [2].
3. MAXIMUM-LIKELIHOOD
SOURCE LOCALIZATION
The general situation where M sources radiate a broadband
signal and N synchronized sensors are used as in a sensor array
to compute the source location estimates can be described by the
following model
x
r
(n) =
m=1
M
a
r
( m)
s
0
( m)
( n t
r
( m)
) +w
r
(n) , n =1,, L , r = 1, , N,

where a
r
(m)
is the signal level of the mth source at the rth sensor,
s
o
(m)
is the source signal, t
r
(m)
is the time delay and w
r
(n)

is zero-
mean i.i.d. Gaussian noise with variance
2
. In the frequency
domain the array signal model is given by
X(k) = D(k)S
0
(k) + Z(k),
where D(k) is the steering matrix at frequency bin k and Z(k)
is the noise spectrum vector. Given certain assumptions on the
noise distribution, the source estimation problem can be cast into
a the problem of maximizing the likelihood function, which in
turn is equivalent to the following maximization

max J(r
s
) = max
k
P (k , r
s
) X(k)
2
,
r
s
r
s
where r
s
is the source location vector, P(k,r
s
) = D(k) D
+
(k)
is a projection operator, and
+
denotes pseudo-inverse. The
summation does not have to encompass all energy bins, but may
be limited to the frequencies that exhibit higher received signal
energy. In practice, due to the edge effect in the nite-length FFT
operation used to obtain the frequency-domain data, the above
described method is denoted as the Approximate Maximum-
Likelihood (AML) method. Fig. 7 shows the simulation result
of the near-eld normalized J(r
s
) metric of a vehicle source
having high values about its location inside the convex hull of an
array of ve sensors. Thus, for near-eld scenarios, this metric
is capable of estimating the location of the source. Fig. 8 shows
the simulation result of the far-eld normalized J(r
s
) metric
of a vehicle source having high values in the angular sector in
the direction of the source. Thus, for far-eld scenarios, this
metric is capable of estimating direction-of-arrival (DOA) of
the source. The details of this algorithm and these and other
simulation and eld-measured estimated results can be found in
[3]. Note that the set of parameters to estimate can be augmented
with the propagation velocity and the noise variance, if these are
unknown. This algorithm requires a fairly high computational
power, mostly due to the maximization of a function, J(r
s
),
that is usually non-convex and displays a large number of local
maxima. Moreover all the sensors that participate in the estimate
need to be synchronized and be fully collaborative.
The algorithm is centralized, i.e., the relevant information has to
be sent to a central node provided with sufcient computational
power. These are issues in networks of low-power sensors, but
may not be a problem in environment monitoring networks
where nodes are more complex and have higher available energy
supplies.

3.4 TDOA
Consider the waveform emitted by a source and then received
by a group of synchronized passive sensors with different time
delays due to different propagation paths. By using these sensor
data, time-difference of arrivals (TDOAs) can be used to perform
source localization.
Denote the source location in Cartesian coordinates by r
s
= [x
s
,
y
s
, z
s
]
T
and the rth sensor location by r
r
= [x
r
, y
r
, z
r
]
T
. Without
(12)
(13)
(14)
(15)
90 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF CALIFORNIA - LOS ANGELES, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
loss of generality, we choose r

= 1

as the reference sensor for
differential time-delays. Let the reference sensor be the origin of
the coordinate system for simplicity. The speed of propagation
v in this formulation can also be estimated from the data. In
some problems, v may be considered to be partially known (e.g.,
acoustic applications) while in others it may be considered to
be unknown (e.g., seismic applications). The differential time-
delays for N sensors satisfy

t
r1
= t
r
t
1
=
|| r
s
r
r
|| || r
s
r
1
||
,

v

for r = 2, , N.

This is a set of N 1 nonlinear equations which
makes nding its solution r
s
non-trivial.
However, the above equations can be reformulated as
|| r
s
r
r
||
2
|| r
s
||
2
= || r
r
||
2

2 (x
s
x
r

+ y
s
y
r
+ z
s
z
r
).
The left hand side of (17) is equivalent to
(|| r
s
r
r
|| || r
s
||) (|| r
s
r
r
|| + || r
s
||)

= vt
r1

(2 || r
s
||

+ vt
r1
).
in the case of r
1
= 0. Upon combining both expressions, we have
the following linear relation
r
T
r
r
s
+ vt
r1
|| r
s
|| + v
2
t
2
r1
/ 2 =

|| r
r
||
2

/ 2,
TABLE II. PERCENTAGE OF ERROR BETWEEN
THE MODEL PREDICTION AND THE ACTUAL
MEASURED ADDITIONS AND MULTIPLICATIONS
PER PIXEL. FOR EACH GOP OF EVERY
SEQUENCE, THE AVERAGE ERROR OVER A
NUMBER OF ADAPTATION POINTS IS PRESENTED.
0
5
10
0
5
-5
-5
10
0
0,2
0,4
0,6
0,8
1
N
o
r
m
a
l
i
z
e
d

J
(
r
s
)
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
X-axis (meter)
Y-axis (meter)

FIGURE 7:
PLOT OF J(r
s
) VALUES
OF A NEAR-FIELD SOURCE CAPABLE
OF SOURCE LOCALIZATION.
0 5 10 15
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
X-axis (meter)
Y
-
a
x
i
s

(
m
e
t
e
r
)
00
55
-5 -5
10 10
15 15
FIGURE 8: PLOT OF J(r
s
) VALUES OF A FAR-FIELD SOURCE CAPABLE
OF DOA SOURCE ESTIMATION.
LOCALIZATION IN SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
91
for the rth sensor. With N sensors, we formulate the least-squares
solution by putting N 1 linear equations into the following
matrix form
Ay = b,
where

A=
% % %
|
|
|
|
|
|
y = v
|
|
|
|
|
|
|
|
|
t
N1
2
/2 t
N 1
r
N
T
|
|
|
|
|
|
,
t
21
2
/2 t
21
r
2
T
r
3
T
t
31
t
31
2
/2
r
s
r
s
v
2
,
2
2
2
r
2
r
3
%
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
r
N
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
.
b =
1
2
|
|
|
|
|
|
|
|
|
For 3-D uncertainty, the dimension of A is (N 1) X 5. In the
case of six or more sensors, the pseudo-inverse of matrix A is
given by
A
+
= (A
T
A)
1

A
T
.

The LS solution for the unknown vector can be given by y =
A
+
b. The source location estimate is given by the rst three
elements of y and the speed of propagation estimate is given by
the square-root of the last element of y.
In the three dimensional case, there are ve unknowns in y.
To obtain an overdetermined solution, we need at least ve
independent equations, which can be derived from the data of
six sensors. However, placing sensors randomly does not provide
much assurance against ill-conditioned solutions. The preferred
approach would be to use seven or more sensors, yielding six or
more relative delays, and to perform a least squares tting of the
data. In the 2-D problem, the minimum number of sensors can
be reduced by one. If the propagation speed is known, then the
minimum number of sensors can be further reduced by one.
Notice in the unknown vector y of (29) that the speed of
propagation estimate can also be given by
v
,
= v

r
s
r

s
using the fourth and the rst three elements of y. To exploit this
relationship, we can add another nonlinear constraint to ensure
equivalence between the speed of propagation estimates from
the fourth and the fth elements. By moving the fth element
of y to the other side of the equation, we can re-write (20) to the
following
Ay = b + v
2
d
where
A =
r
2
T
t
21
r
3
T
t
31
% %
r
N
T
t
N 1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
, y =
r
s
v r
s
|
|
|
|
|
|
|
|
|
|
|
|
,
b =
1
2
2
r
2
2
r
3
%
2
r
N
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
, d =
1
2
t
21
2
t
31
2
%
t
N 1
2
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
.

In this case, the dimension of A is (N 1) X 4 for 3-D uncertainty.
The constrained least-squares (CLS) solution for the unknown
vector can be given by y = A
+
b + v
2
A
+
d. Dene p = A
+
b and
q = A
+
d. The source location and speed of propagation estimates
can be given by
x
s

= p
1
+ v
2
q
1
,

y
s

= p
2
+ v
2
q
2
,

z
s

= p
3
+ v
2
q
3
,

v || r
s
|| = p
4
+ v
2
q
4
,
(20)
(21)
(23)
(24)
(25)
(26)
(27)
(22)
92 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF CALIFORNIA - LOS ANGELES, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
where p
i
and q
i
are the ith entry of p and q, respectively. The
number of unknowns appears to be ve, but the ve unknowns
only contribute four degrees of freedom due to the following
nonlinear relationship
|| r
s
||
2
= x
2
s
+ y
2
s
+ z
2
s
.

By substituting (45) into (46), the following third order constraint
equation results
(v
2
)
3
+ (v
2
)
2
, (v
2
) +

= 0,
where
= q
2
1
+ q
2
2
+ q
2
3
,
= (p
1
q
1
+ p
2
q
2
+ p
3
q
3
) q
2
4
,
= q
2
1
+ q
2
2
+ q
2
3
2p
4
q
4
,
= p
2
4
.
In practice, there are various methods to estimate the TDOA
among the sensors. A blind beamforming method was proposed
by [5]. This method is based on the use of the maximum-power
collection criterion to obtain array weights from the dominant
singular vector or eigenvector associated with the largest singular
value or eigenvalue of the space-time sample correlation matrix.
This approach not only collects the maximum power of
the dominant source, but provides some rejection of other
interferences and noise. Theoretical justication of this approach
uses a generalization of Szegos theory of the asymptotic
distribution of eigenvalues of the Toeplitz form.
The relative phase information among the weights yields the
relative propagation time delays from the dominant source to
the array sensors. These TDOA estimates can be used to perform
source localization as well as detection, signal enhancement,
DOA estimation, and delay-steered beamforming. More details
on the TDOA estimation are given in [5] and [6].
A circular array with ten microphones located on a circle of
four-feet radius (at each hour of the clock except for the ones at
seven and ten oclocks) and one microphone at the center was
used to estimate the TDOA values of a moving vehicle from
measured data supplied by ARL (the Army Research Lab). The
data sampling rate was 1 kHz and approximately two minutes of
data of the vehicle at the far eld were used. The TDOAs of the
ten microphones on the circle relative to the center microphones
were estimated and shown in Fig. 9. Fig. 10a shows the estimated
DOA and Fig. 10b shows the estimated speed of propagation
using the TDOA-CLS method.
(28)
(29)
(30)
0 10 20 30 40 50 60 70 80 90 100
Time (sec)
-3
-2
-4
-1
0
1
2
3
4
R
e
l
a
t
i
v
e

t
i
m
e
-
d
e
l
a
y

(
m
s
e
c
)
FIGURE 9: RELATIVE TIME-DELAY ESTIMATIONS OF 10 MICROPHONES
RELATIVE TO THE CENTER MICROPHONE.
3.5. RSS
The Received Signal Strength (RSS) method is a conceptually
simple energy-based source localization procedure. An acoustic
or a RF source in free space radiating omni-directionally will
attenuate at a rate inversely proportional to the square of the
distance. In practice, we can assume an isotropic exponential
attenuation model of e
i
(t)

= s(t)/||r(t)

r
i
||

.

Here, e
i
(t) is the
energy value at the th sensor located at r
i
whose location is
assumed to be known, r(t) is the unknown coordinate of the
source, s(t)

is the unknown source waveform, and is the decay
exponent which is assumed to be known or can be estimated for a
given scenario. In ideal free space, =

2. In using this approach,
the ratio of e
i
(t)/e
j
(t)

is computed for all pairs of sensors.

LOCALIZATION IN SENSOR NETWORKS

ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
93
0 10 20 30 40 50 60 70 80 90 100
0
50
-50
-100
-200
-150
(a) Direction of arrival estimation
Time (sec)
D
O
A

(
d
e
g
)
FIGURE 10: (A) DOA AND (B) SPEED
OF PROPAGATION ESTIMATION USING
THE TDOA-CLS METHOD.
0 10 20 30 40 50 60 70 80 90 100
800
1000
1200
1400
1600
1800
(b) Speed of propagation estimation
Time (sec)
S
p
e
e
d

o
f

p
r
o
p
a
g
a
t
i
o
n

(
f
t
/
s
e
c
)
In this ratio, the unknown s(t)

waveform is eliminated and will
not be needed in further processing. Each of these ratios denes
a circle on which r(t)

may reside. In the absence of noise (or at
high SNR), with N sensors, only N 1 out of the total N(N 1)/2
of these ratios are independent. For noisy situations, more than N
1 of these ratios need to be used to determine r(t). Additional
details for acoustic sensors are given in [9] and [10] and for EM
sensors are given in [11].
3.6 DOA
In Sec. 3.1 and Sec. 3.2, two different methods can be used to
estimate the DOA in an array system. In this section, we describe
some experimental results on using the DOAs to perform source
localization in the far-eld. In [7], we built a wireless acoustical
testbed using the Compaq iPAQ 3760 Pocket PCs as the testbed
nodes. Each iPAQ has a microphone, an ADC, a codec, and a
802.11b wireless card. The combination of the COTS hardware
and open-source operating system was used to implement both
the TDOA-CLS and the AML DOA estimation algorithms. The
choice of these devices, which admittedly may be quite different
from the nodes of a sensor network, was motivated by the desire
to set up a working testbed that would allow the generation
of signicant data as well as the development and testing of
localization tracking algorithms. Fig. 11 shows a conguration
with three linear subarrays with each subarray having three iPAQ
microphones to obtain three DOA estimations. Fig. 12 shows the
results of six source localizations by using cross-bearings of the
three DOA values using both the TDOA-CLS and AML methods
for a music source. Next, consider a conguration with four square
subarrays, with each subarray having four iPAQs. Fig. 13 shows
94 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF CALIFORNIA - LOS ANGELES, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
the cross-bearings of the four DOAs to estimate the location
of a speaker playing a music sound. Then two speakers, one
playing the vehicle sound and another playing the music sound
simultaneously were used. Fig. 14 shows that the AML method
is capable of performing source localization of both speakers.
4. CONCLUSIONS
Sensor networks are being deployed in a wide range of
applications. In order for the sensed data to be relevant, it is
0
,
3
m
0
,
3
m
0
,
3
m
S1
S2
0
,
3
m
S3
0
,
3
m
S4
1
2
,
2
m
0
,
3
m
S5
0
,
3
m
S6
6,1m 6,1m
FIGURE 11: CONFIGURATION OF THREE LINEAR SUBARRAYS.
0 5 -5 10 15 20 25
0
5
10
15
20
25
X-axis (meter)
Y
-
a
x
i
s

(
m
e
t
e
r
)
Sensor locations
Source locations
AML location estimates
TDOA-CLS location estimates
Sensor locations
Source locations
AML location estimates
TDOA-CLS location estimates
FIGURE 12: LOCALIZATION BY CROSS-BEARING OF DOA ESTIMATES
OF A MUSIC SOURCE AT DIFFERENT LOCATIONS.
0 2 -2 4 6 8 10 12 14
0
2
4
6
8
10
12
14
X-axis (meter)
Y
-
a
x
i
s

(
m
e
t
e
r
)
Sensor locations
Source locations
AML location estimates
Sensor locations
Source locations
AML location estimates
FIGURE 13: LOCALIZATION BY CROSS-BEARING OF DOA ESTIMATES OF
A VEHICLE SOURCE.
0 2 -2 4 6 8 10 12 14
0
2
4
6
8
10
12
14
X-axis (meter)
Y
-
a
x
i
s

(
m
e
t
e
r
)
Sensor locations
Source 1 locations
Source 2 locations
Source 1 AML location estimates
Source 2 AML location estimates
Sensor locations
Source 1 locations
Source 2 locations
Source 1 AML location estimates
Source 2 AML location estimates
FIGURE 14: LOCALIZATION BY CROSS-BEARING OF DOA ESTIMATES OF
TWO SOURCES.
crucially important that the sensors geographical positions, i.e.,
their locations, be known or estimated. If the nodes cannot be
placed with certainty, or in case the nodes might be displaced
or moved during the course of their operation, then node
localization techniques must be employed. Location estimation
may be performed by one centralized processor, or by the
individual nodes in a distributed fashion. The function of sensors
is often to detect the occurrence of an event, which in most cases
is the presence of a source within their sensing range. The sensor
LOCALIZATION IN SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
95
network performs this task by applying techniques of source
localization.
In this paper we have described the problems of source and
node localization in sensor networks. The literature on both
these topics is very broad and by no means have we explored
all possible facets of the localization problem. We have covered
in some detail a few selected algorithms in which we have been
personally involved. We have also explored some issues related
to the statistical characterization of measurements and their
impact on the quality of the location estimates.
REFERENCES
[1] P. Biswas and Y. Ye, SEMIDEFINITE PROGRAMMING FOR
AD-HOC WIRELESS SENSOR NETWORK LOCALIZATION,
Info. Proc. in Sensor Networks, pp. 552553, 2004.
[2] M.S. Brandstein, J.E. Adcock, and H.F. Silverman, A
CLOSED-FORM LOCATION ESTIMATOR FOR USE WITH ROOM
ENVIRONMENT MICROPHONE ARRAYS, IEEE Tr. Speech and
Audio Proc., vol. 15, no. 1, pp. 4550, January 1997.
[3] J. Chen, R.E. Hudson, and K. Yao, MAXIMUM-LIKELIHOOD
SOURCE LOCALIZATION AND UNKNOWN SENSOR LOCATION
ESTIMATION FOR WIDEBAND SIGNALS IN THE NEAR-FIELD,
IEEE Trans. Signal Processing, vol. 50, pp. 18431854,
Aug. 2002.
[4] J.C. Chen, K. Yao, and R.E. Hudson, SOURCE
LOCALIZATION AND BEAMFORMING, IEEE S.P. Mag.,
vol. 19, no. 2, pp. 3039, March 2002.
[5] K. Yao, R.E. Hudson, C.W. Reed, D. Chen,
and F. Lorenzelli, BLIND BEAMFORMING ON A RANDOMLY
DISTRIBUTED SENSOR ARRAY SYSTEM, IEEE Jour. on Sel.
Areas in Communications, vol. 16, pp. 1555-1567,
Oct. 1998.
[6] J.C. Chen, K. Yao, T.L. Tung, C.W. Reed, and D. Chen,
SOURCE LOCALIZATION AND TRACKING OF A WIDEBAND
SOURCE USING A RANDOMLY DISTRIBUTED BEAMFORMING
SENSOR ARRAY, Inter. Jour. of High Performance
Computing Applications, vol. 16, pp. 259-272, Fall 2002.
[7] J.C. Chen, L. Yip, J. Elson, H. Wang, D. Maniezzo,
R.E. Hudson, K. Yao, and D. Estrin, COHERENT ACOUSTIC
ARRAY PROCESSING AND LOCALIZATION ON WIRELESS
SENSOR NETWORKS, Proc. of the IEEE, vol. 91,
pp. 1154-1185, Aug. 2003.
[8] B.H. Cheng, R.E. Hudson, F. Lorenzelli, L. Vandenberghe,
and K. Yao, DISTRIBUTED GAUSS-NEWTON METHOD FOR
NODE LOCALIZATION IN WIRELESS SENSOR NETWORKS,
Signal Proc. Advances in Wireless Comm.,
pp. 915919, 2005.
[9] D. Li, K.D. Wong, Y.H. Hu, and A.M. Sayeed,
DETECTION, CLASSIFICATION, AND TRACKING OF TARGETS,
IEEE Signal Processing Magazine, pp. 30-29, March 2002.
[10] D. Li and Y.H. Hu, ENERGY-BASED COLLABORATIVE
SOURCE LOCALIZATION USING ACOUSTIC MICROSENSOR
ARRAY, Eurasip Jour. on Applied Signal Processing,
pp. 321-337, March 2003.
[11] N. Patwari, J.N. Ash, S. Kyperountas, A.O. Hero III,
R.L. Moses, and N.S. Correal, LOCATING THE NODES,
IEEE Signal Processing Magazine, pp. 54-69, July 2005.
[12] L. Girod, V. Bychkovskiy, J. Elson, and D. Estrin,
LOCATING TINY SENSORS IN TIME AND SPACE: A CASE
STUDY, IEEE Intl. Conf. Comp. Design, pp. 214219, 2002.
[13] F. Gustafsson and F. Gunnarson, MOBILE POSITIONING
USING WIRELESS NETWORKS, IEEE S.P. Mag.,
vol. 22, no. 4, pp. 4153, July 2005.
96 COPYRIGHT STMICROELECTRONICS, UNIVERSITY OF CALIFORNIA - LOS ANGELES, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
[14] H. Hashemi, THE INDOOR RADIO PROPAGATION CHANNEL,
Proc. IEEE, vol. 81, no. 7, pp. 943968, July 1993.
[15] S.M. Kay, Fundamentals of Signal Processing
Estimation Theory, Englewood Cliffs, NJ,
Prentice Hall 1993.
[16] Y. Okumura, E. Ohmori, T. Kawano, and K. Fukuda,
FIELD STRENGTH AND ITS VARIABILITY IN VHF
AND UHF LAND-MOBILE RADIO SERVICE,
Rev. Elec. Commun. Lab., vol. 16, pp. 910, 1968.
[17] G. Pottie and W. Kaiser, WIRELESS INTEGRATED NETWORK
SENSORS, Comm. of the ACM,
vol. 43, no. 5, pp. 5158, May 2000.
[18] V. Raghunathan, C. Schurgers, S. Park, and M.B.
Srivastava, ENERGY-AWARE WIRELESS MICROSENSOR
NETWORKS, IEEE S.P. Mag., vol. 19, no. 2, pp. 4050,
March 2002.
[19] C. Savarese, J. Rabaey, and K. Langendoen, ROBUST
POSITIONING ALGORITHMS FOR DISTRIBUTED AD-HOC
NETWORKS OF SENSORS, USENIX Technical Annual
Conf., June 2002.
[20] A. Savvides, H. Park, and M.B. Srivastava,
THE BITS AND FLOPS OF THE -HOP MULTILATERATION
PRIMITIVE FOR NODE LOCALIZATION PROBLEMS, ACM
Mobile Net. and App., vol. 8, no. 4, pp. 443451, 2003.
[21] G. Sun, J. Chen, W. Guo, and K.J.R. Liu, SIGNAL
PROCESSING TECHNIQUES IN NETWORK-AIDED POSITIONING,
IEEE S.P. Mag., vol. 22, no. 4, pp. 1223, July 2005.
CONTACT: ST.JOURNAL@ST.COM
INERTIAL SENSORS FOR WIRELESS BODY AREA NETWORKS: THE WIMOCA SOLUTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
97
INERTIAL SENSORS
FOR WIRELESS BODY
AREA NETWORKS:
THE WIMOCA SOLUTION
Wireless Body Area Sensor Networks
(WBASN) are an emerging technology
enabling the design of natural Human
Computer Interfaces (HCI). Automatic
recognition of human motion, gestures,
and activities is studied in several contexts.
For example, mobile computing technology
is being considered as a replacement of
traditional input systems. Moreover, body
posture and activity monitoring can be used
for entertainment and health-care applications.
However, until now, little work has been
done to develop flexible and efficient
WBASN solutions suitable for a wide range
of applications. Their requirements pose new
challenges for sensor network designs, such
as optimizing traditional solutions for use
as environmental monitoring-like applications
and developing on-the-field stress tests.
In this paper, we demonstrate the flexibility
of a custom-designed WBASN called
WiMoCA with respect to a wide range of
posture and activity recognition applications
by means of practical implementation
and on-the-field testing. pproach taken
in the design of WiMoCA provides the
Elisabetta Farella
(1)
, Augusto Pieracci
(1)
,
Luca Benini
(1)
, Andrea Acquaviva
(2)

(1) DEIS - University of Bologna, Italy
(2) ISTI - Urbino University, Italy
necessary effectiveness. Software support,
communication protocols and hardware
architecture have been designed to match
requirements of WBASN applications. Nodes
of the network mounted on different parts of
the human body exploit tri-axial accelerometers
to detect its movements. The advanced digital
Micro-electro-mechanical system (MEMS)
based inertial sensor has been chosen
for WiMoCA because it demonstrated high
flexibility of use in many different situations,
providing the chance to exploit both static
and dynamic acceleration components for
different purposes. Furhermore, the sensibility
and accuracy of the sensing element
is perfectly adequate for monitoring human
movement, while keeping cost low and size
compact, thus meeting our requirements.
We implemented three types of applications,
stressing the WBASN in many aspects.
In fact, they are characterized by different
requirements in terms of accuracy, timeliness,
and computation distributed on sensing
nodes. For each application, we describe
its implementation, and we discuss results
about performance and power consumption.
COPYRIGHT UNIVERSIT DEGLI STUDI DI BOLOGNA, UNIVERSIT DEGLI STUDI DI URBINO, 2007.
98 COPYRIGHT UNIVERSIT DEGLI STUDI DI BOLOGNA, UNIVERSIT DEGLI STUDI DI URBINO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
1. INTRODUCTION
Technology advances in the design of low power digital
architectures and component miniaturization pushed the
development of wireless networks of sensors, which have
been extensively studied in the last decade in the context of
environmental monitoring, security of buildings and public
spaces, traffic, and, more recently, health monitoring. Body area
networks represent a recent evolution of this technology for the
development of a new generation of human-computer interfaces
(HCIs). Sensor technology enables the development of small
form-factor devices that can be mounted on wearable nodes and
communicate to each other and to a sensor fusion element, the
latter of which can be either a mobile device such as a cellular
phone or, a palmtop, or a personal computer.
Similar to traditional wireless sensor networks, body sensors
collect information about the environment (the human body),
that is subsequently correlated for monitoring and/or actuation
purposes. Current design and technology trends (such as Smart
Dust sensors [34] let us envision in the near future the availability
of dust-like non-intrusive and extremely low power sensor nodes
that can be easily worn by users.
In this scenario, body area sensor networks can be used to build
general purpose natural interfaces, enabling the user to interact
with the surrounding intelligent space by providing a new range
of services. For example, personal devices (such as mobile
phones or palmtop computers) can be enhanced by posture
recognition technologies [24], [20], [32], replacing traditional
input systems. Replacement of standard interfaces is also coming
up for user identification. In fact, personalized and secure access
will be ensured by body-based identification techniques [19],
[40], [16].Going beyond the simple replacement of traditional
interfaces, body area networks open the field to a new range
of services and applications that can be built on top of them.
For instance, health monitoring applications can exploit sensors
for real-time tracking of body movements for rehabilitation
purposes [7], [15]. Collaborative workplaces are another
example of innovative application. Here, the body network is
bridged to an external network that controls the surrounding
context. Workers can interact and collaborate with each other
through the company network. Finally, entertainment and
cultural applications can profit from the enhanced immersivity
of user interaction in virtual games, 3-D artificial worlds, and
virtual heritage [18], [17], [48].
Even if the research in this field is very active and fruitful, many
challenges have still not been faced. Wearability and autonomy
of the nodes require power consumption optimization. While, at
the same time, real-time sensing, processing, and communication
capability impose tight performance constraints. Recent work
has shown that wearability and low-power consumption
requirements can be handledthrough an efficient design of
sensor nodes [20], [32] in the context of a posture recognition
application. However, these requirements must be satisfied in
conjunction with ever increasing application demands for fast
response time and flexibility and must be tuned depending on
the application domain.
Moreover, compared to traditional sensor networks for
environmental monitoring that are commonly designed to handle
sporadic events, original requirements are imposed to achieve an
efficient body monitoring system implementation, both from the
node software design and network organization.
From a network perspective, by mounting the nodes on the
human body, they are all in the same coverage area and directly
connected to a base station (installed as an extension board on
a palmtop PC). Thus, the topology of the body network does
not require the implementation of routing and data aggregation
algorithms. The base station also represents a gateway that
connects the network to a personal computer (either a palmtop
or a desktop computer) where the user application is located.
On the other end, tight design constraints are imposed to match
fast response time requirements of interactive applications,
while providing configurability and flexibility, depending on
INERTIAL SENSORS FOR WIRELESS BODY AREA NETWORKS: THE WIMOCA SOLUTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
99
the application context. Moreover, to enable the development
of a general purpose body-centric natural interface, flexibility of
node architecture and network organization is a key issue.
In this paper, we present three applications exploiting natural
human movement detection interface based on our flexible
Wireless Body Area Network solution, called WiMoCA [20],
where sensors are represented by tri-axial integrated MEMS
accelerometers [27], [28].
Sensitive, compact, and inexpensive inertial sensors based
on MEMS technology enable the design of a system that is
easier to use, more functional, and more reliable [54]. Today,
communication, consumer, and industrial markets are the main
drivers for the development of commercial MEMS devices, and
their huge volumes, the tight power, and form factor constraints
strongly push toward lower price, size, and power consumption.
Clearly, many applications reap significant benefits from these
trends [10]. When a sensor has to be worn, accelerometers are
effective for recognition purposes because of their small size
and their immunity from electromagnetic interference and
obstacles [6], [23], [4]. All of the presented applications exploit
accelerometers to detect postures and gestures of the human
body, but they are different in terms of processing requirements,
real-time constraints, and distribution of the computation.
A posture detection system is presented as a first application that
implements a natural human-computer interface. Body sensors
collect inclination data from various parts of the human body
and perform a preliminary processing, then forward them to the
gateway node, which is connected to a palmtop PC. Software
running on the palmtop correlates data coming from various
sensors and compares the resulting position with a predefined
set of postures.
Body area sensor networks are a promising enabling technology
for health monitoring applications. In this paper, we show a
rehabilitation system where accelerometers in sensor nodes
are used to monitor user balance with respect to the center of
mass concerning the body part under control. A real-time data
processing algorithm implemented in a palmtop computer
detects the wrong position and sends a feedback signal through
a Bluetooth headset in order to stimulate the user to correct
its posture. In this kind of application, accuracy and real-time
constraints are critical for design guidelines.
WiMoCA has been exploited not only to detect static postures
(directly extracted from data by the firmware running on the
microcontroller), but also dynamic every-day life actions, like
different kind of gait. We designed a gait recognition system based
on a complex algorithm. Due to its complexity, the algorithm
runs on a desktop computer equipped with the gateway node.
The applications described in this paper have been successfully
implemented and tested in the field. They represent real-life test
cases for the design of body-area wireless sensor networks for
natural interfaces. Thanks to their heterogeneous characteristics,
they allow the implementation of various processing models,
where computation is either performed on the gateway node, or
partially performed locally on the nodes. This allows us to stress
the capabilities of the WiMoCA body area network solution.
Extensive experimental tests highlight that the power
consumption characteristics of the body area network are
suitable for human-centric applications, even in the worst case
of continuous activity.

2. BACKGROUND ON WIRELESS
SENSOR NETWORKS FOR HCI
To be considered wearable and unobtrusive interfaces,
accelerometers must be mounted on lightweight and tiny boards
without wires and cables for power supply and connectivity. To
this end, some research laboratories recently targeted low-power,
wireless, and wearable devices. Hand and head motion tracking
systems have been designed [21] for Virtual Reality applications.
This class of systems is dedicated and optimized for tracking
specific body parts and movements. WiMoCA targets general
100 COPYRIGHT UNIVERSIT DEGLI STUDI DI BOLOGNA, UNIVERSIT DEGLI STUDI DI URBINO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
purpose gesture recognition and for this reason is designed to be
installed on any body part. Researchers have done a consistent
amount of work to build absolute orientation detection devices
mainly targeted to robot navigation [4], [23]. In these kinds of
devices, accelerometers are used to compute orientation referred
to local vertical direction, while magnetometers are used to
compute the direction of gravity with respect to the Earth.
Recently, MARG modules equipped with triads of orthogonally
mounted accelerometers and magnetometer and angular rate
sensors have been designed [4] ; these modules target 3 DoF
(Degrees of Freedom) orientation measurements of static or
slowly moving rigid bodies. Since a module does not have
wireless communication capabilities, data must be transmitted
over a wiredlink for further processing. Accelerometers,
magnetometers, and gyroscopes are exploited in [23] to build
a wireless portable absolute orientation estimation device. The
device uses a Kalman filter to filter out accelerations due to user
movements. Designed to be part of a sensor network, the device
uses a Bluetooth module for network communications.
A new sensor node solution that has recently been developed
is the InterSense Wireless InertiaCube [26]. This solution is
composed of a set of sensors without processing capabilities
connected to a receiver, which acts as a gateway. Processing must
take place at the host side to which the gateway is connected.
Compared to our solution, it features a higher accuracy, but
it has a bigger form-factor and lacks processing capabilities
on the nodes, thus resulting in less flexibility. WiMoCA, on
the other hand, is a network of sensor nodes that enables the
implementation of distributed computation, thanks to nodes
processing capability. For example, in the posture recognition
application we describe in this paper, this capability is exploited
to perform partial recognition of postures inside the nodes.
Moreover, the cost of InertiaCube3 is one order of magnitude
higher than that of the WiMoCA nodes. Moreover, because
WiMoCA is fully wireless and equipped with standard interfaces,
it makes the system suitable for a wide range of applications for
wearable and mobile devices, like the ones presented in this
paper.
Comparing this class of sensors to WiMoCA, the main difference
is that they are designed to perform tracking of absolute
orientation. In our design, we are interested in gesture recognition
instead of tracking, and for this reason we do not need complex
filtering algorithms to compensate for the error introduced by
the integration steps. Instead, we use a geometric algorithm
which uses accelerometers as inclinometers. Moreover, because
they are not designed to detect absolute orientation, we do not
use gyroscopes and magnetometers. For this reason, WiMoCA
shows a smaller form-factor, making it more suitable to be
mounted on various body parts without impacting normal user
movements.
Accelerometers and gyroscopes have also been used to build
human-computer interfaces [44], [42], [12]. This class of HCI
interfaces differs from WiMoCA either because they target a
specific movement or because they are not designed to build
a distributed general purpose recognition system. In fact,
WiMoCA provides advanced wireless networking hardware and
software support to communicate with other WiMoCA modules
to build a body area network.
Wearable inertial sensors are a low-cost, low-power solution
to track gestures and, more generally, movements of a person.
The implementation of a body-centric network mounting inertial
sensors has been explored in many fields. Examples are in context-
aware applications [35], [45] and monitoring of patient activity
in the medical domain [2]. A combination of accelerometers,
magnetometers, and temperature and light sensors to be worn by
users has been applied to help indoor navigation [22] and to infer
a users location. Many research studies have focused on hybrid
sensor networks or sensor fusion techniques. The techniques
whose sensing elements are distributed along the body and
data processed both off-line or in real-time [56], [33]. Even if
these solutions are suitable to specific applications, they are not
INERTIAL SENSORS FOR WIRELESS BODY AREA NETWORKS: THE WIMOCA SOLUTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
101
tailored to applications requiring high wearability and very-low
power consumption.
Commercial solutions like Mica Motes [30] are designed to
handle sporadic or slowly changing events (such as temperature
and pressure variations) and to interface with web applications
for environmental monitoring. As such, they are equipped with
embedded operating systems that support a complete network
stack. Compared to these solutions, our system has been
designed for real-time interactive applications with low-power
requirements, and, for this reason, we focused on minimizing
software overheadby implementing our own component drivers
and communication layer. Moreover, Mica are equipped with
analog bi-axial accelerometers, which require additional
ADC conversion and are less tailored to gesture recognition
applications compared to digital integrated tri-axial devices
equipped by WiMoCA.
The software support we developed to drive the module exploits
the low power consumption of the sensing unit to improve
battery lifetime and provide a network stack, (physical and
MAC layer) to drive the wireless transceiver, to interface with
other modules, and to organize accelerometer data in packets to
be sent to a remote machine for further processing.

3. WBASN ARCHITECTURE
WiMoCA [20] sensing node is designed to be wearable and low-
power. It has a modular architecture to ease fast replacement
and update of each component. It is composed by three sections
(Fig. 1) namely MCU/sensors, RF, and power supply. The module
sensibility obtained is 2mg, and the maximum throughput is
36Kbps. The core of the WiMoCA node is the low-cost, low-
power ATmega8 8-bit microcontroller based on the AVR RISC
architecture. The sensing is based on a MEMS tri-axial digital
accelerometer by STMicroelectronics (LIS3LV02DQ [29]). This
one-chip three-axis MEMS-based accelerometers with digital
interface chipsets, combines small size, high resolution, and
low power consumption. In fact, maximum power consumption
is 5mW in active mode and 1 W in sleep mode (at supply
voltage of 3.3V). The RF section is based on the RFM TR1001
transceiver operating at 868 MHz and reaching a maximum bit
rate of 100kbps.
Among other nodes, the gateway is a special one which mounts
two layers dedicated to communication interface. Similarly to
other nodes, it has an RF section for exchanging data with the
body network, but it also interfaces to the external world, e.g.,
a PC or workstation, through an interchangeable link such as
RS232, Bluetooth, or Ethernet.
The WBASN is organized with a star topology and can be easily
worn. WiMoCA nodes are end-points of the star, and they are
responsible for sensing and acquiring data from the environment
before sending them to the gateway node (Fig. 2). The gateway
performs bridging functions but does not have a sensing device.
It is used to form the backbone of the system by connecting the
WBASN to a host machine or an external network. Coordination
among transmissions of various nodes to the gateway is
performed at MAC-level.
WiMoCA software implements a collision free MAC protocol,
inspired by real-time MAC presented in [9]. Compared to
collision avoidance protocols developed for sensor networks
such as TMAC [52], collision free protocols are suitable for
handling real-time traffic, since they avoid the overhead imposed
by collision detection packets by allocating a time frame to each
message [50]. In fact, the typical usage of the body area sensor
network imposes a periodic sampling of acceleration values from
a subset of nodes. For instance, a typical application of WiMoCA
FIGURE 1:
WIMOCA
SYSTEM
STRUCTURE.
RF section:
transceiver + antenna
Sensor + microcontroller
Power supply
102 COPYRIGHT UNIVERSIT DEGLI STUDI DI BOLOGNA, UNIVERSIT DEGLI STUDI DI URBINO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
is to recognize if the user is walking or sitting down through
sensors placed on arms and legs. This is obtained through real-
time processing of acceleration values. Moreover, this protocol
is suitable for networks with a constant number of nodes, such
as the one we target in this work.
The CFRT (Collision Free Real-Time) protocol basically divides
time into frames in which only one node is allowed to transmit.
The scheduling order is derived by a Message Table stored in each
node and is identical for all the nodes so that each of them knows
when it has the right to transmit. The table contains an entry
for each node allowed to transmit or receive in a frame. Fields
in the entry specify source, destination nodes, message length,
and message period. Compared with [9], where the table is built
by applying earliest deadline first scheduling protocol (EDF)
to a queue of waiting messages; here, the table is application
dependent. The scheduling order is programmed depending on
the job that must be performed by the body network. Moreover,
we do not allocate frames for inter-cell messages since, in the
body network, all the nodes belong to the same cell.
For this reason, a cross-layer approach has been followed
where the application is responsible for performing
message scheduling. The Message Table at each node can
be dynamically re-programmed via wireless channel at the
beginning of each frame. At this time, the gateway can notify
nodes that a table-update is required. As a consequence,
nodes prepare to receive the new table from the gateway. This
solution allows applications to program the Message Table,
depending on the sensing pattern, which in turn depends on
the movement or gesture to be recognized. In general, the
table keeps constant during an acquisition session, limiting
time/energy spent for re-programming.
During table-updating phase, a time synchronization message
is also sent by the gateway to each node, since all the nodes must
be synchronized to share the bandwidth. The synchronization
message is sent by the gateway at a predefined time interval,
as explained in the next section. In general, in each round of
the table, called table period, a time slot (management slot)
is allocated for synchronization and table updates. When it
is not used for synchronization or table updates, transceivers
of all the nodes are switched off for its entire duration. The
timeline of MAC protocol operations is shown in Fig. 3.
4. APPLICATIONS
Given the architecture described in the previous section, we
show in the remainder of this paper a set of three applications
FIGURE 2: EXAMPLE OF BODY NETWORK.
Bluetooth
WiFi
Bluetooth
WiFi
Local biofeedback Local biofeedback
Bluetooth Bluetooth
= wireless sensor nodes = wireless sensor nodes
I
n
t
e
r
n
e
t
I
n
t
e
r
n
e
t
= gateway = gateway
Remote biomonitoring Remote biomonitoring
GPRS GPRS
INERTIAL SENSORS FOR WIRELESS BODY AREA NETWORKS: THE WIMOCA SOLUTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
103
enabled by the WiMoCA wireless body area sensor network. We
implemented these applications to test the flexibility of the body
area network software and hardware.
They can be distinguished based on i) the distribution of
computation among nodes and gateway; ii) the type of data
transmitted on the network; iii) the level of accuracy required
in data sampling and processing and iv) the type of processing
performed at the host side and real-time constraints. Results of
characterization of each application in terms of performance and
power consumption are also described.

4.1 Posture Recognition Application
Posture recognition is aimed at detecting user postures among a
set of different possibilities. The size of this set depends on the
application, which in turns affects the number of sensing nodes
to be used.
Each sensing module monitors the inclination of a certain part
of the body, acquiring acceleration samples along three axes.
Acceleration values are averaged first, then the module tilt
with respect to the gravity is computed and encoded. Thus,
in this application, nodes are heavily involved in the overall
computation. After processing, data are sent to the gateway
according to the schedule imposed by the contention-free MAC
protocol implemented on each node. Each module inclination
is then collected at the gateway side frame by frame and
subsequently combined to interpret body posture. Finally, the
gateway communicates the result of detection to a Java application
running on the host machine, where a Graphical User Interface
(GUI) is updated to display the current user posture. Being only
interested in a predetermined set of postures, accuracy is not
critical in this kind of application. After averaging, inclination
data are only transmitted when they overcome a threshold that
determines the cross-point between two different postures.
In the present implementation, communication between the
gateway and the host machine can be performed both through
RS232 and Bluetooth protocol. The host machine can be a Desktop
PC but also a palmtop computer. In this latter case, the Bluetooth
solutionis preferred. The choice of a Java implementation
for the application provides the necessary portability.
4.1.1 Setting the context
Even though the idea of exploiting user postures to implement
alternative input systems for cell-phones and personal digital
assistants (PDA) has been explored in the last few years [24],
[3], many challenges have not yet been faced. From an
implementation viewpoint, recent work showed that wearability
and low-power consumption requirements can be faced through
an efficient design of sensor nodes [20], [32]. However, these
e

FIGURE 3: TIMELINE OF MAC
PROTOCOL.
RX1 S RX2 Gateway
Node 1
Node 2
Node 3
Synchronization (S)
T sync T frame
New
table
sent
New
configuration
Tr
T
RX3 off RX1 ... ... RX3 S RX1 RX3 S T RX3
... ...
... ...
Receiving (RX)
Idle (off)
Transmitting (TX)
... ...
104 COPYRIGHT UNIVERSIT DEGLI STUDI DI BOLOGNA, UNIVERSIT DEGLI STUDI DI URBINO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
requirements must be satisfied in conjunctionwith ever increasing
application demands for fast response time and flexibility. As a
consequence, a complete implementation has not been presented
up to now, to the best of our knowledge.
Posture recognition is often a primary goal of activity recognition
system for elderly care [15]. The distinction between static
postures (standing, sitting, lying supine, on a side, and prone)
is investigated in [53], where the user is monitored with two
accelerometers, one mounted tangentially on the thigh, and
the second mounted radially on the sternum. The signals were
acquired for subsequent off-line assessment. In a recent study
[8], more than 20 (sub)postures and motions are distinguished,
using four bi-axial accelerometers, worn by the user on different
parts of the body and wired by cables among them and with a data
recorder worn in a belt around the waist. After collection, data are
processed off-line, and feature extraction and pattern recognition
techniques are applied. This solution is suitable for motion and
posture analysis, but the system is obtrusive and cannot be used
for real-time interaction and posture recognition.
4.1.2. Posture recognition system
In the accelerometer-based posture monitoring application,
the user is equipped with 3 sensing modules placed along the
body (Fig. 4), precisely on the trunk (node A), the thigh-bone
(node B), and the shinbone (node C). If we refer to the plane
defined by the direction of gravity and the ideal line between
the shoulders, the projection onto this plane of axes relative to
each sensor module is shown in the right part of Fig. 4. The
application implemented aims atdetecting user posture among
seven different possibilities, as shown in Fig. 6 (sitting, standing,
and lying in four different manners).
Fig. 5 shows a sample of tilt data collected by the three end nodes.
The stream corresponds to a sequence of postures: 1) standing;
2) seated; 3) standing and 4) seated with legs extended (as lying
on a table). A different sequence can indicate, for example, that
the user has fallen down and then used to generate an alarm or
to contact assistance. The plot reports computed angle degrees
versus time (expressed as number of samples).
Programming end nodes detect posture by acquiring sensing
data and computing average acceleration in a predefined time
Trunk
yA
zA
xA
g
B
C
A
A
Thigh-bone
yB
zB
xB
B
Shinbone
yc
zc
xc
C
FIGURE 4: SCHEMA OF THE USER AND THE MODULE SETUP AND ORIENTATION
ALONG THE BODY.
FIGURE 5: SEQUENCE OF MOVEMENTS.
20
1 15 29 43 57 71 85 99 113 127 141 155 169 183 197 211
40
60
80
100
120
140
0
1 15 29 43 57 71 85 99 113 127 141 155 169 183 197 211
20
-20
40
60
80
100
120
20
1 14 27 40 53 66 79 92 105 118 131 144 157 170 183 196 209 222
60
0
100
140
Shinbone
Time in samples
Thigh-bone
Trunk
D
e
g
r
e
e
D
e
g
r
e
e
D
e
g
r
e
e
4 3 2 1
4 3 2 1
4 3 2 1
x
y
z
x
y
z
x
y
z
INERTIAL SENSORS FOR WIRELESS BODY AREA NETWORKS: THE WIMOCA SOLUTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
105
window to filter out temporary variations of user position or
noise. The choice of a time window of 8 samples simplifies
the averaging process, which is performed through a right shift
of 3 bit. The acceleration value is 16 bit for each axis in twos
complement encode (9 byte of memory are needed). In order to
reduce the amount of information to be sent over the wireless
channel, nodes identify one configuration among a known set.
For each node, in fact, there is a set of valid configurations that is
stored in the nodes memory. These configurations are the result
of tilt positions that are computed based on the sensor orientation
w.r.t. the gravity component [6]. Accelerometers provide data in
the range [-2g, 2g], though when performing tilt detection, the
range is restricted to values within [-1g, 1g]. We are interested
in detecting three different situations for each axis, directed as
orthogonal to or opposed to the gravitational acceleration. The
range of acceleration values has been quantized in classes, and
the actual acceleration value is thus classified in one of three
sub-ranges.
Combining the discretized value for each axis, the set of three
accelerations is associated with a specific 4-bit configuration
identifying the tilt of the module (see Table 1). The number of
possible tilts recognized can be varied according to application
need, introducing more than three sub-ranges for acceleration
quantization.
The sub-ranges dont necessarily cover all the possible
acceleration values as in our case. Thus, the process of module
tilt identification also considers the possibility of providing a
code corresponding to unclassified (last row in the table) as
output. At the end of the tilt identification process, the module
tilt code is sent to the gateway.
FIGURE 6: SEVEN STATIC POSTURE DETECTED.
0000 0000 0000 1 (Standing)
0000 0001 0000 2 (Seated)
0000 0001 0001 3 (Seated with leg up)
1000 1000 1000 4 (Prone)
0001 0001 0001 5 (Supine)
0010 0010 0010 6 (Lying on right side)
0100 0100 0100 7 (Lying on left side)
TABLE 2: POSTURE TABLE IN A PRACTICAL CASE.
TRUNK THIGH-BONE SHINBONE USER POSTURE (ASCII)
p o o 0000
o o g 0001
o p o 0010
o g o 0100
o o p 1000
g o o 1010
none none none 0101
TABLE 1: END-NODE POSSIBLE INCLINATION: P = PARALLEL BUT OPPOSED TO g, g
= DIRECTED AS g, O = ORTHOGONAL TO g.
X Y Z Module Tilt (4 bit)
On the gateway side, the firmware acquires and accumulates tilt
data from each end-module for a given frame and combines all
module tilts to identify a known body configuration. Finally, it
sends the identified position to the application. The gateway is
also responsible for handling the set-up of nodes and network
(e.g., Message Table transmission, association of each node with
a body part).
According to the Message Table, which is identical on end-nodes
and gateway, the latter waits for the reception of tilt data from
each end-node. On the gateway, a table, called Body Location
Table (BLT), containing correspondence among node ID and
position on the body is stored. A default BLT is initialized
each time the gateway is reset; nevertheless, the table can be
programmed by the application running on the general purpose
system at any time.
Supported by the BLT, data coming from different nodes are
combined to determine user posture among a set of possible
postures contained in another table, the Posture Table (see Table 2
106 COPYRIGHT UNIVERSIT DEGLI STUDI DI BOLOGNA, UNIVERSIT DEGLI STUDI DI URBINO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
picturedisplayed (Fig. 7). The GUI offers the user the chance to
enter a new BLT through specific text fields and send buttons. This
can be useful if a certain module is placed in a different location
from the default one at run-time. When the button is clicked, a
control character is sent to the gateway and subsequently the new
BLT with correspondence among nodes and parts of the body.
4.1.3. Performance and Power Consumption
To assess power consumption of the node while running
the posture recognition algorithm, we performed additional
characterization tests, whose results are shown in Fig. 8. With 3
nodes, we can achieve a maximum sample rate of 651 positions
per second. In practical cases, the maximum frequency of human
movement is 30Hz so that 60 positions per second is a sufficient
rate for detecting postures without losing information.
Battery lifetime depends on the frequency of samples. We note
that the strong dependency of the transmission power on the type
of bit also affects battery lifetime in a considerable way at full
node activity.
and Fig. 6). For clearness, we refer to the actual application where
the end-nodes are three, displaced on the trunk, on the thigh-bone,
and on the shinbone and, all oriented as shown in Fig. 4.
After encoding, the user posture is sent through the serial (or
Bluetooth) interface to the Java application.
The Java application on the host side accomplishes a set of
simple tasks: (i) continuous data acquisition from the serial or
Bluetooth port exploiting the Java COMM API; (ii) user posture
encode extraction from data stream; (iii) GUI implementation for
real-time visualization of an image corresponding to actual user
posture; (iv) initialization control commands, set-up of all tables
(BLT, Message Table, etc.) resident on the network.
Most of the time, the application listens to the COM port
(Bluetooth or RS232) for acquiring data coming from the
gateway. The extraction of the encoded posture is committed to a
specific thread that first identifies the start byte of the packet sent
by the gateway, then the time-frame through the counter byte, and
finally the posture code. These steps are continuously repeated
as in a state-machine. The code extracted is communicated to
the Graphical User Interface (GUI) thread, which updates the
FIGURE 7: GRAPHICAL USER INTERFACE SNAPSHOT.
Transmission of a data packet PTX-DP (M = 11) = 138,5 mW
PTX-DP (M = 0) = 21,60 mW
Total power consumption PNH (M = 11) = 46,00 mW
PNH (M = 0) = 16,85 mW
Battery life at latency = 1,536msec TBLH (M = 11) = 7h
(651 positions per second) TBLH (M = 0) = 19h
Battery life at latency = 16,7msec TBL (M = 11) = 18h
(60 positions per second) TBL (M = 0) = 21h
FIGURE 8: POWER CONSUMPTION AND BATTERY LIFE-TIME OF THE NODE DURING
POSTURE RECOGNITION.
Phase Our case (Vcc= 3.3V)
In addition, it must be noted that battery lifetime decreases
significantly much lower (from 19h to 7h) as the position sample
rate increases (from 60 pos/sec to 651 pos/sec). This is because
of the higher power efficiency of the transceiver at the higher
bit rate.

4.2 Bio-Feedback Application
In this application, the sensor network is distributed on the
users body for balance monitoring and correction. It represents
a relevant feasibility study for the implementation of a mobile
INERTIAL SENSORS FOR WIRELESS BODY AREA NETWORKS: THE WIMOCA SOLUTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
107
recognition system using feedback actuation. Software
architecture is characterized by nodes performing basic
sensing functions with almost no on-board processing. All the
computation is performed on the palmtop computer (we refer to
it as PDA - Personal Digital Assistant) that interfaces with the
wireless body network by means of the gateway node. The PDA
is responsible for data processing and is in charge of activating
the actuator that provides feedback to the user about the correct
user posture. In the implemented set-up, we adopted audio
feedback through a headset.
This application differs from the posture recognition system
because of the software organization but also because of the
use of accelerometers for fine position tracking. Accuracy
requirements are much more stringent w.r.t. posture recognition.
Data collected and transmitted are acceleration values that are
then processed at the host side. The hardware and software
architecture (communication protocols, power management
policies, and application-level control) have been tuned to
optimize cost, battery autonomy, and real-time performance
required for this application.
4.2.1 Setting the context
Inertial sensors (accelerometers in particular) and biofeedback
have proved to be essential elements in applications for balance
control because of their small size, portability, and supply of
useful kinematic information [46], [51]. Postural unbalance is a
crucial aspect in many musculoskeletal, neurological, and age-
related diseases [31], [47], [38] and in several sport disciplines
[49]. Recent studies have shown that improvement of balance
may be gained by use of biofeedback, based on the principles of
physiological adaptations and brain plasticity [11]. Nevertheless,
biofeedback systems for balance control are usually cumbersome
and expensive; furthermore, they are mainly for ambulatory use
and require specific expertise (e.g. [1]). Hence, a goal in the
development of biofeedback systems for balance and motor
rehabilitation is easy accessibility, specifically with respect to
unobtrusiveness, portability, and cost.
4.2.2 Bio-WWS system
Bio-WWS is composed of four kinds of nodes: one or more sensor
nodes; the audio actuator node; the PDA, which acts as a general
purpose node; and the gateway node. At present, the system has
been tested with three sensor nodes, one located on the trunk for
sway monitoring and the other two on a leg (thigh and calf) for
determining whether the user is walking or standing (Fig. 9).

FIGURE 9: BIO-WWS SET-UP.
A star topology is used as in posture application. With respect
to previous applications, here nodes do not perform preliminary
processing. They instead send acceleration data to the gateway,
which performs signal pre-processing and data normalization. It
also organizes data retrieved from the sensor nodes in packets to
forward them to the PDA over a Bluetooth link. Finally, the PDA
processes data from the body sensors in order to resolve a humans
posture condition and to give adequate feedback to the user.
The host PDA device we used for testing the biofeedback
application is a HP iPAQ 5550 with Bluetooth and WiFi
capabilities. To reach a fully wearable system, we used a wireless
link through a PDA and a headphone representing the auditory
feedback actuator.
The system considers trunk planar accelerations (in the forward-
backward and left-right directions), both gravitation and
108 COPYRIGHT UNIVERSIT DEGLI STUDI DI BOLOGNA, UNIVERSIT DEGLI STUDI DI URBINO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
ASES THE
dynamic, and determines an estimate of sway of the users center
of mass. Accelerations collected by the gateway are coupled
with a timestamp in order to have a screenshot of all sensors
placed in the human body at the same time. Then, the integrated
accelerations are sent to the PDA using a Bluetooth link.
To generate the feedback signal, accelerations are coded into
a stereo sound using a suitable algorithm, aimed at providing
high comfort level to the user while improving his/her stability
while standing [16], [17], [19]. In particular, accelerations in
the forward-backward direction are coded by frequency and
amplitude (volume) modulation of the sound (Fig. 10.a). The
frequency codes the value of the instantaneous acceleration
while the volume increases with the distance from the Target
Region (TR). The TR is a range of acceleration values which are
considered to be safe for the user. When in this region, the user
perceives a low but constant tone.
This has been shown to be more effective for the user than
suddenly turning on the audio signal. Thanks to the feedback
signal, the user notices when he/she goes outside the TR. When
the users accelerations are inside the TR range, the sound volume
and frequency are fixed [13]. The users accelerations in the left-
right direction are coded by Left/Right balance modulation of
the sound (Fig. 10.b). Besides TR, another parameter of the
biofeedback algorithm is the Limit Region (LR). This region
defines the operating range for the algorithm. If theuser crosses
the LT, a constant tone is played, meaning that the user is in
a dangerous position. A GUI was implemented on the PDA to
support easy setting of parameters of the biofeedback algorithm.
4.2.3 Power and performance characterization
Measurement of power consumption was performed with an
acceleration sampling frequency of 60Hz (to obtain a useful
bandwidth of 30Hz, adequate for capturing human movements).
Communication speed was set at 32kbps and at 230kbps from
gateway to PDA, transmitting via Bluetooth at 2,4GHz. Power
consumption when all components are active and the transceiver
is continuously sending data corresponds to 200mW for the
gateway and 45 mW for a single end-node. In an idle state
(where only the reception mode is active), the power lowers
respectively at 25mW and 20 mW. When active, the gateway
battery (500mAh) and the single node battery (100mAh) have
a lifetime of 8 hours. The sleep mode corresponds to having all
devices in the minor consumption state (1.5mW for gateway and
10 W for end-nodes). For a 20% duty cycle, the lifetime reaches
40 hours for the gateway and 38 hours for the node.
In terms of power consumption, tests revealed that the bottleneck
for a system lifetime is the palmtop computer (with a 3.7 V

FIGURE 10: AUDIO BIOFEEDBACK
SOUND DYNAMICS.
A. FORWARD-BACKWARD DIRECTION
B. LEFT-RIGHT DIRECTION
right
LR
TR
left
A B
sensor
node
f
r
e
q
u
e
n
c
y
V
o
l
u
m
e
LR
TR TR
F min
V min
F max
V max
INERTIAL SENSORS FOR WIRELESS BODY AREA NETWORKS: THE WIMOCA SOLUTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
109
battery of 1000mAh), which consumes around 350mW. As a
consequence, the power budget of the WBASN is suitable for
this kind of application.
Accelerometer accuracy and data transfer rate of the nodes
abundantly matches the requirements of the specific application
of Bio-WWS since human movements are below 50 Hz (even
below 10 Hz in quiet standing), and trunk accelerations range
from 5-10 mg [39] to 40-50 mg.
excursion with the use of BioWWS, then without it, in the same
sensory-altered conditions. Future development of the system
will include further investigation and experimental sessions
to complete the system validation and to optimize the sound
modulation parameters.
4.3 Gait
WBASN can be effective for implementing activity recognition
systems. WiMoCA is used to implement a Gait Analyzer, as we
called it, aimed at classifying human walking behavior, thus
demonstrating the flexibility of our WBASN. In this application,
we use asingle wireless end-node equipped with the triaxial
MEMS accelerometer and placed on the ankle for the purpose of
sampling accelerations to perform step recognition. Acceleration
values are packed and transmitted. Hence, compared to posture
detection system, accelerometers are not used as inclinometers.
Moreover, after calibration (as explained later), only the
accelerations on a single axis are transmitted. Similar to a
bio-feedback system, computation is offloaded from the end-
node. The gateway node isconnected to a personal computer
that processes acceleration data coming from the node. Signal
processing techniques are applied to data to isolate each step and
classify the data among many possibilities. We considered four
kinds of steps in the currentimplementation: i) step in place; ii)
staircase step; iii) step forward slow and iv) step forward fast.

4.4 Setting the context
Detection and classification of walking behaviors through inertial
measurements can be employed for activity recognition [14], gait
monitoring for healthcare purposes [15], or for indoor navigation
systems [36]. In [37], walking behavior detection is aimed at
supporting an indoor localization system counting the number of
steps and estimating a persons current location referenced on a
known starting location. Many of the techniques applied for this
purpose are based on a training phase on data collected during
daily activity and afterward processed off-line. In [41], gait
analysis targets clinical systems. Gait is continuously monitored
though an instrumented shoe, which sends data coming from a
FIGURE 11: RESULTS ON A REPRESENTATIVE SUBJECT: BIOWWS DECREASES
THE COP EXCURSION.
Forward-Backward CoP
5mm
2sec
Target
Region
Tests in the field have been also performed to assess the
effectiveness of the Bio-WWS application. Several experimental
trials were performed on 5 young healthy users as preliminary
validation procedure. Users were required to stand with eyes
closed on afoam-rubber surface that simulated sensory alterations
at foot level. Sway during quiet standing, which indicates the level
of achieved balance, was measured by means of a force platform
(Bertec 4060-08), which quantified movement of the center of
pressure (CoP) [55], i.e., the application point of the ground
reaction force. Subjects performed three trials first without
Bio-WWS and then three trials with Bio-WWS. Preliminary
results are exemplified in Fig. 11, where CoP displacement
for a representative subject is shown, without BioWWS and
with BioWWS (after the calibration time). Effectiveness of the
system for this representative trial is proved by the minor CoP
110 COPYRIGHT UNIVERSIT DEGLI STUDI DI BOLOGNA, UNIVERSIT DEGLI STUDI DI URBINO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
tri-axial accelerometer and a pressure sensor wirelessly and in
real-time. The shoe has been more recently enhanced with other
sensors obtaining a multimodal sensor interface used also to
provide an interactive music device for dance performance [43].
Gait monitoring has also been studied collecting accelerometer
data referred to the head and pelvis in young and older subjects
as they are walking on a level and an irregular walking surface.
The aim of this investigation is to understand how ageing affects
walking behavior and posture. In [14], a gyroscope attached to the
shank is used to detect walking upstairs, toe-off, heel-strike, and
foot-flat. Wavelet transformation in conjunction with a simple
kinematics model have been exploited and the system tested in
many different conditions. In [53], the dynamic contribution to
acceleration of a monoaxial accelerometer fixed on the thigh
has been used to distinguish among walking, stair ascent, stair
descent, and cycling. Compared to previous work, our solution
provides: (i) greater flexibility, since a single WBASN can be
used for many applications; (ii) processing is performed run-time
processing without the need of a pre-characterization phase and
(iii) it is a completely wireless system.
4.5 Gait recognition application
To classify movements correctly, the Gait Analyzer [25] needs to
know the exact orientation of the node as it has been placed on
the leg. For this reason, a system calibration must be performed
first. The user must place the sensor, which can be identified as a
dice, with one of the faces orthogonal with respect to the gravity
direction. The processing flow has four main tasks: (i) payload
extraction; (ii) system calibration; (iii) step count; and (iv) gait
identification.
The first task is similar to other applications discussed and does
not need further elaboration. Instead, focusing on calibration is
more critical. The calibration strategy we choose is one among
several possible alternatives. In our case, it is performed with
the user standing still. The maximum value of acceleration
measurable for any components is close to one, since it is
normalized with the gravity acceleration. During calibration,
data are captured for a fixed time window to obtain an average
value for each axis.
By comparing the computed average with thresholds shown in
Table 3, the orientation of the node is identified. This helps the
user set the node in the correct position. Note that the user is
requested to set the node so that acceleration values are close
to either zero or one (depending on which axis is orthogonal
or parallel to the ground). However, the user is not asked to
set a particular axis orthogonal to the ground plane; the node
orientation need not be carefully set either. The averages
computed during the calibration phase are then stored to be later
used later as a reference during the recognition.
FIGURE 12: DICE FOR EXTRACTING ORIENTATION.
Y
X
5 2
1
6
Z
4
3
Next, absolute position needs to be then found out. In general, to
identify the position of the tri-dimensional node in the space, it is
sufficient to consider it as a dice (Fig. 12). To achieve this target, a
number is associated with each surface of the dice. In Table 3, six
different positions are identified depending on the acceleration
values along the X, Y and Z axes. A position 1 means that the
surface 1 is parallel to the ground plane. In Fig. 12, the surface
1 is identified by acceleration values along the Z axis as -1.
The reference position in which the calibration takes place in our
case is with the node having a surface in contact with the leg.
INERTIAL SENSORS FOR WIRELESS BODY AREA NETWORKS: THE WIMOCA SOLUTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
111
This surface lies on a plane vertical w.r.t. to the ground. If the
surface of the dice lying on the leg is the one identified by the
number 1, as shown in Fig. 13, the average acceleration values
of the three axes are zero along X and Z, while they provide
an average close to -1 along Y axis because it is parallel to
the ground plane. Absolute no deposition is then computed by
combining the information obtained for each axis (Table 4).
After calibration, by means of the knowledge of node position
over time, it is possible to understand when the subject is walking.
The algorithm implemented to recognize single steps is based
on the observation of acceleration values along the component
orthogonal to the ground. This is the acceleration component
which, during calibration phase, was parallel to the gravity
acceleration. If the node is fixed on the ankle (see Fig. 13), the
axis involved is Y.
The overall step recognition is composed of two distinct phases.
First, the length of the step (i.e. the number of acceleration
values) is determined in a first row recognition phase. Then,
the algorithm performs a fine detection based on the step type.
Let us focus on the first phase. To perform row recognition, the
human step is considered as a combination of a raising phase
and a leaning phase. Based on this observation, when the foot is
raised, the acceleration collected along the Y axis increases due
to the dynamic component solicited by the upward movement.
Axis direction w.r.t ground orthogonal, direction down parallel to ground orthogonal, direction up
Symbol 1 0 -1
TABLE 3: POSITION CODES.
AVERAGE VALUE 0,5 -0,5 AVERAGE VALUE 0,5 AVERAGE VALUE -0,5
1 0 0 -1
2 0 -1 0
3 -1 0 0
4 1 0 0
5 0 1 0
6 0 0 1
POSITION CODE X Y Z
TABLE 4: POSITION CODES.
FIGURE 13: SETUP ON THE USER FOR THE SENSOR NODE.
Y
X
Z
Alternatively, as the foot touches the ground, acceleration along
Y decreases. At the end of the leaning phase, there is a bouncing
effect that is used to recognize the end of the step.
In fact, to perform step recognition, two thresholds have been
set (as shown in Fig. 14) that are used to compare positive and
negative acceleration values respectively. More precisely, first
the negative and the positive peaks are determined; then these
values are compared against the thresholds. These threshold
values have been obtained through extensive characterization
in the field.
A step is identified by three consecutive events (Fig. 14):
i) the positive peak is larger than the positive threshold;
ii) the negative peak is lower than the negative threshold; and
iii) the acceleration values stay under the thresholds for at least
three consecutive samples. All data referring to a single step are
stored in a static array.
112 COPYRIGHT UNIVERSIT DEGLI STUDI DI BOLOGNA, UNIVERSIT DEGLI STUDI DI URBINO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
Note that this procedure allows us to identify the length of the
step in real-time. Other approaches in literature perform an off-
line characterization phase that is then used to correlate sampled
data with stored data [5].
Once a single step has been stored in all three vectors (one for
each axis), data are processed to distinguish in real-time among
four kinds of steps (more can be added):
1. going up on a staircase;
2. step in place
3. step forward fast
4. step forward slow
To identify the action of going up step by step on a staircase, an
autocorrelation function is applied to the vertical axis vector. This
vector contains acceleration values for increasing time instants.
Autocorrelation values computed are stored in an output vector,
which is then processed to verify its monotonic behavior (see
Fig. 16). If the function is not monotonic, the user has mounted a
step of a staircase. If yes it is, the system verifies one of the other
possibilities (i.e., step in place or step forward).
To recognize if the user is performing a step in place, the result
coming from the auto-correlation function is processed to find
its higher value, verifying whether is greater or not that a
empirically defined threshold. If both results have a negative
outcome, the user has probably performed a step forward. To
identify the speed of the step, the standard deviation referred
to the axis parallel to the movement direction is used. Standard
deviation is compared with another empirically defined constant
acting as a threshold. This threshold can also be set by the user
depending on the speed that has to be identified.
5. EXPERIMENTS
In this subsection, we show experimental results about the
characterization of human steps. First of all, we show in Fig. 15
acceleration values corresponding to a sequence of steps. It can
be noted that each step has at least a positive acceleration peak,
a negative peak, and an intermediate phase, where acceleration
values stay under the thresholds.
Another set of experiments has been performed to tune the
algorithm for recognition of various types of steps, as explained
in the previous subsection. An example of such experiments is
shown in Fig.16.
From Fig. 16, it can be noted that different types of steps are
characterized by the different behaviors of the autocorrelation
function. Steps performed in place lead to autocorrelation values
FIGURE 14: ACCELERATION PLOT CORRESPONDING TO A STEP FOR Y AXIS.
0
0,2
0,4
0,6
0,8
1
1,2
1,4
1,6
1,8
2
Y Axis
Packets
Bouncing Leaning phase Raising phase
Positive
threshold
A
c
c
e
l
e
r
a
z
i
o
n
i
Negative
threshold
5 10 15 20 25
4
FIGURE 15: STEP ACCELERATION WAVEFORM.
-1,5
-1
-0,5
0
0,5
1
1,5
2
2,5
Packets
A
c
c
e
l
e
r
a
t
i
o
n
s
17 33 49 65 81 97 113 129 145 161 177 193 209 225
X Axis
3st Step 4st Step 2st Step 1st Step
Y Axis Z Axis
INERTIAL SENSORS FOR WIRELESS BODY AREA NETWORKS: THE WIMOCA SOLUTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
113
under a certain threshold. Staircase steps are identified by the
non-monotonic behavior of the autocorrelation function. Steps
forward are identified by the autocorrelation function overcoming
a predefined threshold, as shown in Fig. As mentioned before,
standard deviation can be used as an indication of the step speed.
A threshold helps distinguish between steps considered slow and
steps considered fast, as shown in Fig. 17.
5. CONCLUSIONS
In this paper, we have presented three applications exploiting
a natural human computer interfacing system implemented by
means of a WiMoCA wireless body area sensor network. We
FIGURE 17: STANDARD DEVIATION FOR 14 DIFFERENT STEPS FORWARD,
7 STEPS PERFORMED FAST, 7 STEPS SLOW.
1
0,1
0
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
2 3 4 5 6 7 8 9 10 11
Fast Slow
S
t
a
n
d
a
r
d

D
e
v
i
a
t
i
o
n
Threshold
12 13 14
FIGURE 16: AUTOCORRELATION OF TWICE THE VERTICAL ACCELERATION
(VECTOR CORRESPONDING TO Y AXIS).
0
5
10
15
20
25
30
35
A
u
t
o
c
o
r
r
e
l
a
t
i
o
n
Step forward Step up (staircase) Step in place
Threshold
Important
detail:
non-
monotonic
function
1 9 17 25 33 41 49 57 65 73 81 89 97 105 113 121 129 137 145 153 161
0
showed how WiMoCA is able to handle diverse application
requirements, thanks to its flexibility. We performed tests in-
the-field to stress the capability of the network. We proved that
our custom-designed, vertically integrated solution for body area
sensor networks allows for the implementation of a wide range
of HCI applications. Future work addresses the integration of
other kinds of sensors to collect more physiological data from
the user, and we plan to add vibro-tactile actuators for bio-
feedback provisioning. Moreover, we are already exploring
energy scavenging techniques to apply on the basic node and
evaluating the use of the Zigbee standard protocol.
REFERENCES
[1] www.onbalance.com. Ref Type: Internet Communication,
2005.
[2] K. Aminian, E. D. Andres, K. Rezakhanlou, C. Fritsch,
Y. Schutz, M. Depairon, P.-F. Leyvraz, and P. Robert,
MOTION ANALYSIS IN CLINICAL PRACTICE USING
AMBULATORY ACCELEROMETRY, Lecture Notes
i Computer Science, 1537(-):111, 1998.
[3] J. Angesleva, I. Oakley, S. Hughes, and O. S., BODY
MNEMONICS: PORTABLE DEVICE INTERACTION DESIGN
CONCEPT, Proceedings of the 16th annual ACM
symposium on User interface software and technology, 2003.
[4] E. Bachmann, X. Yun, D. McKinney, R. McGhee,
and M. Zyda. DESIGN AND IMPLEMENTATION OF MARG
SENSORS FOR 3-DOF ORIENTATION MEASUREMENT OF
RIGID BODIES, Proceedings. ICRA 03.
IEEE International Conference on Robotics
and Automation 2003, September 2003, pages 1419
[5] L. Bao and S. S. Intille. ACTIVITY RECOGNITION
FROM USER ANNOTATED ACCELERATION DATA, In Proc.
of Pervasive 2004, the Second International Conference
on Pervasive Computing, Springer, pages 117, 2004.
114 COPYRIGHT UNIVERSIT DEGLI STUDI DI BOLOGNA, UNIVERSIT DEGLI STUDI DI URBINO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
[6] R. Barbieri, E. Farella, A. Acquaviva, L. Benini,
and B. Ricc, A LOW-POWER MOTION CAPTURE
SYSTEM WITH INTEGRATED ACCELEROMETER,
First IEEE Consumer Communications and Networking
Conference, 2004. CCNC 2004., pages 418423,
January 2004.
[7] D. Brunelli, E. Farella, L. Rocchi, M. Dozza, L. Chiari,
and L. Benini, BIO-FEEDBACK SYSTEM FOR REHABILITATION
BASED ON WIRELESS BODY AREA NETWORK,
Proceedings of Fourth IEEE international Conference
on Pervasive Computing (Percom06) - Workshop Ubicare
2006, pages 527 531.
[8] J. Bussmann, W. L. J. Martens, J. H. M. Tulen,
F. C. Schasfoort, H. J. G. V. D. Berg-Emons, and H. J.
Stam, MEASURING DAILY BEHAVIOUR USING AMBULATORY
ACCELEROMETRY: THE ACTIVITY MONITOR, Behavior
Research Methods, Instruments and Computers,
2001, 33(3):349356.
[9] M. Caccamo, L. Y. Zhang, L. Sha, and G. C. Buttazzo,
AN IMPLICIT PRIORITIZED ACCESS PROTOCOL FOR WIRELESS
SENSOR NETWORKS, IEEE Real-Time Systems Symposium,
December 2002, pages 39 48
[10] M. Canina, A. Rovetta, F. Pasolini, M. Tronconi,
and E. Chiesa, INNOVATIVE SYSTEM FOR THE
ACCUMULATION OF ENERGY OF THE STEP IN A LIMB
PROSTHESIS, Proceedings of the 11th World Congress
in Mechanism and Machine Science, April 2004.
[11] D. Cattaneo, M. Ferrarin, W. Frasson,
and A. Casiraghi, HEAD CONTROL: VOLITIONAL
ASPECTS OF REHABILITATION TRAINING IN PATIENTS
WITH MULTIPLE SCLEROSIS COMPARED
WITH HEALTHY SUBJECTS, Arch. Phys. Med. Rehabil.,
86:13811388, 2005.
[12] A. D. Cheok, K. G. Kumar, and S. Prince, in Proceedings
of the Sixth International Symposium
on InWearable Computers, 2002. (ISWC 2002).
[13] L. Chiari, M. Dozza, A. Cappello, F. B. Horak,
V. Macellari, and D. Giansanti, AUDIO-BIOFEEDBACK
FOR BALANCE IMPROVEMENT: AN ACCELEROMETRY-BASED
SYSTEM, Biomedical Engineering, IEEE Transactions,
52:21082111, 2005.
[14] B. Coley, B. Najafi, A. Paraschiv-Ionescu,
and K. Aminian, STAIR CLIMBING DETECTION
DURING DAILY PHYSICAL ACTIVITY USING A MINIATURE
GYROSCOPE, Gait and Posture, 22(4):287294,
December 2005.
[15] K. M. Culhane, M. OConnor, D. Lyons, and G. M. Lyons,
ACCELEROMETERS IN REHABILITATION MEDICINE
FOR OLDER ADULTS, Age Ageing, 34(6):556560, 2005.
[16] K. Delac and M. Grgic, A SURVEY OF BIOMETRIC
RECOGNITION METHODS, In, Proceedings of the 46th
International Symposium on Electronics in Marine,
2004 (Elmar 2004), pages 184 193, 16-18 June 2004.
[17] G. Faidley, J. Hero, K. Lee, B. Lwakabamba,
R. Walstrom, F. Chen, J. Dickerson, D. Rover, R. Weber,
and C. Cruz-Neira, DEVELOPING AN INTEGRATED WIRELESS
SYSTEM FOR FULLY IMMERSIVE VIRTUAL REALITY
ENVIRONMENTS, In Proceedings of the Eighth
International Symposium on Wearable Computers
(ISWC 2004), 2004, volume 1, pages 178 179,
31 Oct.-3 Nov. 2004.
[18] E. Farella, D. Brunelli, L. Benini, B. Ricc,
and M. E. Bonfigli, PERVASIVE COMPUTING
FOR INTERACTIVE VIRTUAL HERITAGE, IEEE Multimedia,
12(3):46 58, July-Sept 2005.
INERTIAL SENSORS FOR WIRELESS BODY AREA NETWORKS: THE WIMOCA SOLUTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
115
[19] E. Farella, S. O Modhrain, L. Benini, and B. Ricc,
GESTURE SIGNATURE FOR AMBIENT INTELLIGENCE
APPLICATIONS: A FEASIBILITY STUDY In Proceedings
of the, 4th International Conference on Pervasive
Computing (PERVASIVE 2006), Dublin, Ireland,
pages 288304, May 7-10 2006.
[20] E. Farella, A. Pieracci, D. Brunelli, L. Benini,
and B. Ricc, DESIGN AND IMPLEMENTATION OF WIMOCA
NODE FOR A BODY AREA WIRELESS SENSOR NETWORK,
Proc. of Systems Communications 2005 IEEE SENET,
pages 342347, August 2005.
[21] E. Foxlin and M. Harrington, WEARTRACK:
A SELF-REFERENCED HEAD AND HAND TRACKER
FOR WEARABLE COMPUTERS AND PORTABLE VR.,
proceedings of the Fourth International Symposium
on Wearable Computers, pages 155162, October 2000.
[22] A. R. Golding and N. Lesh, INDOOR NAVIGATION USING
A DIVERSE SET OF CHEAP, WEARABLE SENSORS,
Proceedings of the 3rd IEEE International Symposium
on Wearable Computers (ISWC 99), 1999.
[23] T. Harada, H. Uchino, T. Mori, and T. Sato, PORTABLE
ORIENTATION ESTIMATION DEVICE BASED ON ACCELEROMETERS,
MAGNETOMETERS AND GYROSCOPE SENSORS FOR SENSOR
NETWORK, Proceedings of IEEE International Conference
on Multisensor Fusion and Integration for Intelligent
Systems (MFI2003), pages 191 196, July 2003.
[24] R. Headon and G. Coulouris. SUPPORTING GESTURAL
INPUT FOR USERS ON THE MOVE, in Proceedings
Eurowearable, pages 107112, September 2003.
[25] www-micrel.deis.unibo.it/wsn/myvideo/walking0001.wmv.
[26] http://www.isense.com/products/prec/ic3/wirelessic3.htm.
[27] http://www.st.com/stonline/prodpres/dedicate/mems/
mems.htm.
[28] http://www.st.com/stonline/prodpres/dedicate/mems/
products/products.htm.
[29] http://www.st.com/stonline/products/literature/ds/11115/
lis3lv02dq.pdf.
[30] http://www.xbow.com/Products/Wireless Sensor
Networks.htm.
[31] G. B. Jarnlo and K. G. Thorngren, STANDING BALANCE
IN HIP FRACTURE PATIENTS. 20 MIDDLE-AGED PATIENTS
COMPARED WITH 20 HEALTHY SUBJECTS, Acta Orthop.
Scand., 62:427434, 1991.
[32] E. Jovanov, A. Milenkovic, C. Otto, and P. C. de Groen,
A WIRELESS BODY AREA NETWORK OF INTELLIGENT MOTION
SENSORS FOR COMPUTER ASSISTED PHYSICAL
REHABILITATION, Journal of Neuro-Engineering
and Rehabilitation, 2:616, 2005.
[33] H. Junker, P. Lukowicz, and G. Troster, PADNET:
WEARABLE PHYSICAL ACTIVITY DETECTION NETWORK,
Proceedings of the 7th IEEE International Symposium
on Wearable Computers, 2003 (ISWC 03).
[34] J. M. Kahn, R. H. Katz, and K. S. J. Pister,
EMERGING CHALLENGES: MOBILE NETWORKING FOR SMART
DUST, Journal of Commun., 2(3):188, 2000.
[35] N. Kern, B. Schiele, and A. Schmidt, MULTI-SENSOR
ACTIVITY CONTEXT DETECTION FOR WEARABLE
COMPUTING, Proceedings of European Symposium
on Ambient Intelligence, pages 220232, November 2003.
[36] S.-W.Lee and K. Mase, RECOGNITION OF WALKING
BEHAVIORS FOR PEDESTRIAN NAVIGATION,
116 COPYRIGHT UNIVERSIT DEGLI STUDI DI BOLOGNA, UNIVERSIT DEGLI STUDI DI URBINO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
Proceedings of the 2001 IEEE International Conference
on Control Applications, 2001 (CCA 01),
pages 1152 1155.
[37] S.-W. Lee and K. Mase, ACTIVITY AND LOCATION
RECOGNITION USING WEARABLE SENSORS, IEEE
Pervasive Computing, 1(3):2432, Jul. Sept 2002.
[38] B. E. Maki and W. E. McIlroy, POSTURAL CONTROL
IN THE OLDER ADULT, Clin. Geriatr. Med.,
12:635658, 1996.
[39] M. J. Mathie, A. C. F. Coster, N. H. Lovell,
and B. G. Celler, ACCELEROMETRY: PROVIDING
AN INTEGRATED, PRACTICAL METHOD FOR LONG-TERM,
AMBULATORY MONITORING OF HUMAN MOVEMENT,
Physiol Meas., 25:R120, 2004.
[40] J. Mntyjrvi, M. Lindholm, E. Vildjiounaite,
S.-M.Mkel, and H. Ailisto, IDENTIFYING USERS
OF PORTABLE DEVICES FROM GAIT PATTERN
WITH ACCELEROMETERS, Proceedings of IEEE
International Conference on Acoustics, Speech,
and Signal Processing 2005 (ICASSP 05), volume 2,
pages ii/973 ii/976, 2005.
[41] S. Morris and J. Paradiso, SHOE-INTEGRATED SENSOR
SYSTEM FOR WIRELESS GAIT ANALYSIS AND REAL-TIME
FEEDBACK, Proceedings of the Second Joint
Conference on EMBS/BMES, 2002, 3:24682469,
23-26 Oct 2002.
[42] I. Oakley, J. Angesleva, S. Hughes, and S. OModhrain,
TILT AND FILL: SCROLLING WITH VIBROTACTILE DISPLAY,
In Proc. of EuroHaptics, June 2004, pages 313323.
[43] J. Paradiso, FOOTNOTES: PERSONAL REFLECTIONS ON
THE DEVELOPMENT OF INSTRUMENTED DANCE SHOES
AND THEIR MUSICAL APPLICATIONS, Quinz, E., ed.,
Digital Performance, Anomalie, 2:3449, 2002.
[44] J. K. Perng, B. Fisher, S. Hollar, and K. S. Pister,
ACCELERATION SENSING GLOVE (ASG), In IEEE
Symposium on Wearable Computers, pages 178180,
October 1999.
[45] C. Randell and H. Muller, CONTEXT AWARENESS
BY ANALYZING ACCELEROMETER DATA, Proceedings
of the 4th IEEE International Symposium on Wearable
Computers (ISWC 00), 2000.
[46] R.Moe-Nilssen, A NEW METHOD FOR EVALUATING
MOTOR CONTROL IN GAIT UNDER REAL-LIFE ENVIRONMENTAL
CONDITIONS, part 1: The instrument. Clin Biomech
(Bristol, Avon)., 13(4-5):320327., Jun 1998.
[47] L. Rocchi, L. Chiari, and F. B. Horak, EFFECTS
OF DEEP BRAIN STIMULATION AND LEVODOPA
ON POSTURAL SWAY IN PARKINSONS DISEASE, J. Neurol.
Neurosurg. Psychiatry, 73:267274, 2002.
[48] M. Sama, V. Pacella, E. Farella, L. Benini,
and B. Ricc, 3DID: A LOW-POWER, LOW-COST HAND
MOTION CAPTURE DEVICE, Proceedings of IEEE Design,
Automation and Test in Europe Conference
and Exhibition (DATE-06), 2006, 6-10 March 2006.
[49] J. M. Schmit, D. I. Regis, and M. A. Riley,
DYNAMIC PATTERNS OF POSTURAL SWAY IN BALLET
DANCERS AND TRACK ATHLETES, Exp. Brain Res., 163:
370378, 2005.
[50] J. Stankovic, T. Abdelzaher, C. Lu, L. Sha, and J. Hou,
REALTIME COMMUNICATION AND COORDINATION IN
EMBEDDED SENSOR NETWORKS, Proceedings of the IEEE,
volume 91, pages 10021022, July 2003.
INERTIAL SENSORS FOR WIRELESS BODY AREA NETWORKS: THE WIMOCA SOLUTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
117
[51] G. Uswatte, W. H. R. Miltner, B. Foo, M. Varma,
S. Moran, and E. Taub, OBJECTIVE MEASUREMENT
OF FUNCTIONAL UPPEREXTREMITY MOVEMENT USING
ACCELEROMETER RECORDINGS TRANSFORMED WITH A
THRESHOLD FILTER, Stroke, 31(3):6627, Mar 2000.
[52] T. van Dam and K. Langendoen, AN ADAPTIVE ENERGY
EFFICIENT MAC PROTOCOL FOR WIRELESS SENSOR
NETWORKS, Proceedings of the 1st international
conference on Embedded networked sensor systems
(SenSys 03), ACM Press, 2003, pages 171180.
[53] P. H. Veltink, H. B. Bussmann, W. de Vries, W. L.
Martens, and R. C. V. Lummel, DETECTION OF STATIC AND
DYNAMIC ACTIVITIES USING UNIAXIAL ACCELEROMETERS,
IEEE Trans Rehabil. Eng., 4(4):37585, Dec 1996.
[54] B. Vigna. MORE THAN MOORE: MICRO-MACHINED
PRODUCTS ENABLE NEW APPLICATIONS AND OPEN
NEW MARKETS, Technical Digest of IEEE International
Electron Devices Meeting 2005, (IEDM 2005),
pages 1 8, Dec. 5 2005.
[55] D. A.Winter, F. Prince, J. S. Frank, C. Powell,
and K. F. Zabjek, UNIFIED THEORY REGARDING
A/P AND M/L BALANCE IN QUIET STANCE, J. Neurophysiol.,
75:23342343, 1996.
[56] R. Zhu and Z. Zhou, A REAL-TIME ARTICULATED
HUMAN MOTION TRACKING USING TRI-AXIS
INERTIAL/MAGNETIC SENSORS PACKAGE,
IEEE Transactions on Neural Systems and Rehabilitation
Engineering, 12(2):295302, 2004.
CONTACT: ST.JOURNAL@ST.COM
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
118 COPYRIGHT ECOLE POLYTECHNIQUE FDRALE DE LAUSANNE, 2007.
Manuel Flury, Ruben Merz,
Jean-Yves Le Boudec

EPFL, School of Computer
and Communication Sciences
MANAGING IMPULSIVE
INTERFERENCE IN IMPULSE
RADIO UWB NETWORKS
Wireless sensor networks are ideally
built on low-cost, low-complexity nodes
that have a low power consumption to
guarantee a long network lifetime. These
are all properties that can potentially be
achieved with impulse radio ultra-wide
band (IR-UWB). In addition, IR-UWB has
a fine timing resolution enabling accurate
ranging and localization capabilities.
For all these reasons, IR-UWB
is an extremely interesting physical
layer technology for wireless sensor
networks. In this article, we consider the
management of impulsive interference
in IR-UWB networks. Impulsive
interference is due to uncoordinated
concurrent transmissions. It occurs, for
instance, when several independent
piconets operate in close vicinity
and is also present in some MAC
layer proposals that allow concurrent
transmissions. If not properly addressed,
impulsive interference can severely affect
the throughput and energy consumption of
1. INTRODUCTION
For the design of wireless networks, there are two choices with
respect to interference: we can design a system that tries to
control or even prevent interference, or we can intentionally allow
interference. Systems that let interference happen use some form
of adaptability to deal with the constantly changing environment.
Systems to control or prevent interference use mechanisms such
as tight power control, orthogonal communication channels, or
mutual exclusion [1].
However, even in systems designed to control interference, there
are always numerous external factors that are beyond the control
an IR-UWB network; as such, it already needs
to be taken into account in the design phase.
First, we show that impulsive interference is a
serious concern for IR-UWB networks. Second,
we present techniques at the physical layer
and at the link layer to cope with and combat
such interference efficiently. Finally, we present
DCC-MAC as an example of an interference-
aware design.
MANAGING IMPULSIVE INTERFERENCE IN IMPULSE RADIO UWB NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
119
of the system designer. For instance, there might be coexisting,
non-coordinated piconets that interfere with each other. This
external interference is difficult to foresee, and adaptive
mechanisms to cope with it are required.
We consider non-coordinated systems based on impulse
radio UWB (IR-UWB) physical layers that allow concurrent
transmissions without power control [2], [3]. Data transmission
at the physical layer occurs in sequences of very short pulses
1

with a large pulse repetition time (PRT). The most frequently
used physical layer model [4] is illustrated in Fig. 1 and briefly
introduced in the following. Time is divided into frames of length
T
f
. Each user transmits one pulse of length T
p
per frame. To provide
some multi-access capability, a frame is further subdivided into
non-overlapping chips of length T
c
, where T
c
T
p
. Each
user chooses the chip in which to transmit its pulse randomly
according to a (pseudo-random) time-hopping sequence (THS).
Such systems are subject to impulsive, non-Gaussian interference
created by the system itself, or by other, similar systems. On
Fig. 2, we can clearly observe the detrimental effect of impulsive
interference on an IR-UWB physical layer. Further, like any other
UWB system [5], they have to coexist with existing narrowband
technologies like 802.11. Managing interference to and from
such coexisting technologies has been extensively studied and is
out of the scope of this article. In this paper, we concentrate on
impulsive interference. The main source of impulsive interference
in IR-UWB systems are pulse collisions between concurrently
transmitting sources. Pulse collisions occur even though nodes
from different piconets generally use different THSs. This is
due to the fact that THSs in IR-UWB are usually not orthogonal 1. Or short bursts of short pulses
FIGURE 2: IN UNCOORDINATED IR-UWB NETWORKS, SOME FORM OF
INTERFERENCE MITIGATION AT THE PHYSICAL LAYER IS NEEDED. WE SHOW THE
BIT ERROR RATE (BER) VERSUS SIGNAL-TO-NOISE RATIO AT THE RECEIVER FOR
A SYSTEM WITH AND WITHOUT INTERFERENCE MITIGATION. THE MITIGATION
SCHEME USED HERE IS THE ONE USING INTERFERENCE MODELING (FURTHER
DESCRIBED IN SECTION 2.3.2); THE SCENARIO FOR THE SIMULATION IS THE SAME
AS IN FIG. 4. IT CAN CLEARLY BE SEEN THAT THE PERFORMANCE DEGRADATION
IS HUGE WHEN NOT MITIGATING THE EFFECT OF INTERFERENCE.
INTERFERENCE MITIGATION AT THE PHYSICAL LAYER IS DISCUSSED IN SECTION 2.2,
AND A POSSIBLE SOLUTION IS GIVEN IN SECTION 2.3.
0 2 4 6 8 10 12 14 16 18
10
0
10
-1
10
-2
10
-3
10
-4
Eb/N0 dB
B
E
R
No mitigation
Interference Mitigation
FIGURE 1: ILLUSTRATION
OF THE DEFINITIONS.
c
j
DENOTES THE TIME-
HOPPING SEQUENCE
OF USER i AND
i
IS THE DELAY BETWEEN
USER i AND USER 1.
THE DASHED CURVE
FOLLOWING THE PULSES
REPRESENTS THE MULTIPATH
PROPAGATION.
j=n
j=n
j=n-1
User 1
Interferer i
j=n+1
i
Tf
Tc
c Tc
.
(i)
j
(i)
120 COPYRIGHT ECOLE POLYTECHNIQUE FDRALE DE LAUSANNE, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
and therefore do not completely prevent collisions. Furthermore,
even if they were perfectly orthogonal, a tight synchronization
between all the nodes in different piconets would be
needed to prevent interference caused by misaligned THSs.
We focus on techniques and schemes that are used to react and
adapt to interference. We do not discuss protocols or techniques
that try to prevent or control interference (see [1] and the
references therein). A multipath propagation channel at the
physical layer further worsens the situation. The larger the delay
spread of the channel, the more a pulse is spread in time. This
increases the probability of pulse collisions. As IR-UWB systems
are likely to be used in environments exhibiting severe multipath
(indoor, factories, etc.), this is a serious issue. Another factor
that increases the probability of pulse collisions is the number of
users trying to transmit simultaneously. Even in systems with a
generally low duty-cycle, it can happen that a lot of users access
the channel at the same time. An example is a sensor network
detecting a fire outbreak. In this case, a specific event triggers
simultaneous transmissions from a large number of nodes.
Finally, one additional important factor concerning interference
is the near-far effect. As the systems under consideration do not
make use of power control, interferers close to the receiver might
not have a signal of much higher strength than that of the user of
interest. To ensure that small portions of these high power signals
do not predominate the received signal, they have to be mitigated
to prevent a huge performance loss.
Note that in a mobile ad hoc network, not only interference, but
also the variable environment calls for adaptability of the system.
Additionally, systems that try to prevent interference usually
need tighter control than systems that let interference happen.
This is often undesirable in an uncoordinated ad hoc network.

Impulsive interference in IR-UWB systems reduces the signal-
to-interference-and-noise ratio (SINR) at the receiver. It affects
the quality of the radio link, producing more packet losses,
which result in an overall rate reduction and an increased energy
consumption. Interference has a large impact on the system
performance and needs to be taken into account as early as in
the design phase. As we further show in this paper, interference
management is a cross-layer issue. It has to be dealt with at the
physical layer level as well as at the link layer level.
On the physical layer, some form of interference mitigation
(Section 2.2) is needed to deal with the near-far effect. The benefits
of an interference mitigation scheme are depicted in Fig. 2.

On the link layer, adaptive retransmission techniques must be
used. Also, the overall rate of a source has to be variable in order
to be adapted to the current level of interference at the receiver.
Systems with a fixed rate must be designed in order to sustain
the worst possible operating conditions, typically a poor channel
between a source and its destination. This in turn imposes a low
overall rate. Systems with an adaptive rate can take advantage
of good channel conditions to transmit with a higher rate. In the
case of degraded channel conditions, their adaptability prevents
complete communication outages.
We do not discuss the effect of these schemes on energy
consumption. There is, of course, a trade-off. A better system
performance reduces the number of retransmissions and hence
decreases the energy consumption. On the other hand, more
complex transceiver designs increase the energy-consumption.

The organization of this paper is as follows. In Section 2, we
present techniques to combat interference on the physical layer.
In Section 3, we discuss link layer techniques to cope with
interference. In Section 4, the DCC-MAC protocol [2] is presented
as a concrete example of an interference aware design which is
a rate-adaptive medium access control (MAC) protocol for IR-
UWB networks. Finally, we conclude the paper in Section 5.
2. COMBATING INTERFERENCE
AT THE PHYSICAL LAYER
Combating interference matters to all functions provided
by the physical layer, be it decoding, channel estimation, or
timing acquisition and detection. We will present some possible
MANAGING IMPULSIVE INTERFERENCE IN IMPULSE RADIO UWB NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
121
solutions for all of these functions in Section 2.3. Currently used
techniques to combat interference on the physical layer can be
divided into two classes, both of which are shortly discussed in
the following.
2.1. Techniques based on joint decoding
These are extensions or adaptations to UWB of classical, well-
established techniques that are also used in other systems like
CDMA [6]. They aim at cancelling or suppressing interference
by jointly estimating and decoding the signals of a large
number of users. For example, a near-far interferer would
be jointly received instead of being treated as interference.
This annihilates the near-far effect and makes joint decoding
potentially attractive. However, an optimal joint processing of all
users [7] is mostly not possible due to its very high complexity.
Therefore, suboptimal methods like minimum mean-square error
(MMSE) multi-user detectors (MUD) or receivers employing
successive interference cancellation (SIC) are used [8]. All of
these methods share the common factor that the receiver has to
acquire and actively decode each of the users. This might be
perfectly suited for a centrally coordinated and synchronized
system, where a base station communicates with a large number
of users at the same time. However, with a distributed IR-UWB
system, synchronizing the receiver with all the users is extremely
complex and impractical. In addition, the complexity of the
decoding operation is excessively high.
2.2. Techniques based on interference mitigation
In contrast to joint decoding, signals from interfering users
are treated as a common interference term. Techniques based
on interference mitigation try to reduce this interference term
and to mitigate and reduce its effect on the performance of the
physical layer. We distinguish two possible options, interference
modeling and thresholding:
2.2.1. Interference modeling
The interference term is assumed to follow an underlying
statistical model. The background noise is often directly
incorporated in the interference model. A receiver using
interference modeling proceeds in two steps. It first tries to
estimate the model parameters. In a second step, this model is
exploited to mitigate the effect of the interference. Modeling
interference is important as it has been shown that simply
assuming it to be Gaussian is not accurate [9].
In order to estimate the parameters of the model, techniques
based on interference modeling can either follow a data-aided
[10], [11], [12] or a blind approach [10]. In the data-aided
approach, a training sequence known to the receiver is used.
The receiver estimates the statistics of the interference model
exploiting the knowledge about the training sequence. In the
blind (non-data-aided) approach, the receiver jointly estimates
the model parameters as well as the unknown data sequence.

2.2.2. Thresholding
A simple thresholding mechanism can be applied. Samples
of the received signal that have an amplitude exceeding a
certain threshold are assumed to have a large interference
contribution[2], [13], [14], [8]. Although thresholding is easy to
implement, an issue common to all thresholding schemes is the
determination of the optimal threshold. This is often left as an
open issue, or it boils down to assuming an AWGN multi-user
interference (MUI) model and setting the threshold based on the
estimated average received noise power [13].
2.3 MUI-aware physical layer system design proposals
We now present some proposals for physical layer core
functionalities that were specifically designed for asynchronous
IR-UWB systems subject to impulsive interference and that use
the above mentioned techniques.
2.3.1. Timing acquisition and detection
In conventional detection methods, the transmitter prepends
each packet with an acquisition preamble known to the receiver.
The receiver correlates the received signal with this acquisition
preamble and performs a threshold check. If the output of the
correlator exceeds a certain threshold, a good match between
122 COPYRIGHT ECOLE POLYTECHNIQUE FDRALE DE LAUSANNE, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
the channel, this is not necessarily true, as samples with a high
interference level are likely to occur in bursts. Therefore, we
propose [12] to introduce correlation by modeling the sequence
of mixture components with a homogeneous Markov chain. The
resulting MUI model is a hidden Markov model (HMM), where
each state is associated with a Gaussian output distribution. The
GMM is just a special case of the more general HMM, where the
choice of the next state is independent of the current state
2
. We
find that the HMM is effectively better in modeling MUI than the
GMM. However, the performance difference is not that huge and
comes at the cost of increased complexity.
In [12], we also propose a coherent RAKE receiver that makes
use of a combination of thresholding and interference modeling
to mitigate interference in the decoding process and accounts for
the multipath nature of UWB channels. Interference modeling is
done using a data-aided approach. Let us assume for simplicity
that the interference model (GMM or HMM) only has two states,
s
1
and s
2
, where s
1
corresponds to a low interference level and
s
2
to a high interference level. The receiver proceeds in two
steps. In a training step, the channel coefficients as well as the
variances,
s
1
and
s
2
, associated with each of the two states,
are estimated based on the known training sequence. In the
subsequent data reception step, the receiver estimates for each
sample y
n
the probability
s
1
(n)

that it has an interference term
generated by state s
1
3
. Before passing the received samples to the
decoder, the receiver multiplies each sample with the following
weighting vector:
the received signal and the acquisition preamble is assumed,
and detection of the packet is declared. These methods have a
severe drawback when MUI is present. If one of the pulses of the
acquisition preamble at the receiver is aligned with a pulse of a
near-by interferer, this interfering pulse can affect the correlation
significantly. Consequently, a small number of aligned
interfering pulses can dominate the output of the correlator and
lead to a wrong detection. In [14], a power independent detection
(PID) method that addresses this problem was proposed. As for
the conventional methods, the PID uses thresholding. However,
it splits the correlation with the whole acquisition preamble into
a set of elementary correlations. Each elementary correlation
corresponds to only one pulse of the acquisition preamble.
A first threshold is applied at the output of each elementary
correlation. If the energy captured by the elementary correlation
exceeds the first threshold, detection of the corresponding pulse
is declared. A second threshold is then applied to the number
of detected pulses. If the number of detected pulses exceeds
the second threshold, detection of the packet is declared. This
procedure makes sure that all the pulses of the acquisition
preamble contribute equally to the final decision. This procedure
is therefore resistant to near-far interference.
2.3.2. Channel estimation and decoding
As already mentioned, a Gaussian model is not well suited to
model MUI in an IR-UWB system. A popular non-Gaussian
model is the Gaussian Mixture Model (GMM). The GMM
assumes that the interference has an underlying probability
distribution formed by a mixture of Gaussians with different
variances. Each interference term is then assumed to be
generated by one of these mixture components. The GMM seeks
to classify each sample and typically attributes samples with
high interference to mixture components with high variances.
In [11], the GMM is proposed as a MUI model for IR-UWB.
We will show how to perform channel and interference statistics
estimation based on this model using a data-aided approach.
The GMM assumes that the mixture components are
independently chosen. However, due to the multipath nature of
w(n) =

s
1
(n)

s
1
2
+
s
2
(n)

s
2
2
2. The reader is invited to read [12] for a more mathematically rigorous definition of the
respective interference models.
3. Here we only consider a two state model, so the probability that the interference term of
y
n
was generated by s
2
is of course
s
2
(n)=1-
s
1
(n).
2 2
MANAGING IMPULSIVE INTERFERENCE IN IMPULSE RADIO UWB NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
123
Consequently, samples with an interference term that stems
with high probability from s
2
get penalized through the factor
w(n). This ensures that samples with high interference do not
contribute excessively to the decision made by the decoder.
(Note the similarity to the power independent detection method
described in Section 2.3.1). The effect of applying this weighting
factor is shown in Fig. 3. Our coherent RAKE receiver also
employs thresholding in addition to the interference modeling
procedure described before. This is necessary in a data-aided
approach since we are facing three different interference
scenarios, only two of which are resolved by interference
modeling. If interference occurs during packet reception, it must
fall into one of the following three categories:
1. interference is present during both training and data
reception (type 1)
2. interference is present during training only (type 2)
3. interference is present during data reception only (type 3)
Interference of type 1 is taken care of by interference modeling.
Ideally, we estimate the interference during the training phase
and then deal with it during data reception as explained
above. Interference of type 2 should do even less harm; we
have estimated it, but it is not present during data reception.
Interference of type 3 is more difficult to tackle since it is not
present during training. Therefore, the estimated variances of
the interference term will be rather small (on the order of the
background noise variance). Samples with a lot of interference
will then still get a relatively high weight.
Hence, we propose the following thresholding mechanism. After
the training phase, we determine the largest of the estimated
variances. In the case of the two-state model discussed here,
this is
s
2
. We then determine a threshold , such that P(X)
, where XN(0,
s
2
), and is some predetermined small
probability. We then erase the samples with an estimated
interference and noise term exceeding the threshold by setting

s
1
(n)=
s
2
(n)=0 for these samples. This ensures that the samples
that cannot be explained by the estimated interference model
with high probability do not contribute to the decision made by
the decoder. Interference of type 3 is thus mitigated by detecting
a deviation from the estimated model. A similar thresholding
approach, rejecting samples suffering from high interference, has
FIGURE 3: HERE WE SHOW HOW AN ALGORITHM BASED ON INTERFERENCE MODELING PERFORMS INTERFERENCE MITIGATION. A TWO-STATE HIDDEN MARKOV MODEL
IS ASSUMED FOR THE MUI. IN (A), ONE PULSE OF THE RECEIVED SIGNAL AND ITS COMPONENT CORRESPONDING TO THE USER OF INTEREST IS SHOWN.
FOR EACH SAMPLE, THE RECEIVER ESTIMATES THE PROBABILITY THAT IT HAS A LOW CONTRIBUTION FROM INTERFERING USERS (LOW INTERFERENCE STATE) OR THAT IT IS
POLLUTED WITH A HIGH INTERFERENCE TERM (HIGH INTERFERENCE STATE). THE ESTIMATED PROBABILITY OF BEING IN THE LOW INTERFERENCE STATE IS ALSO SHOWN IN THE LEFT
FIGURE. WE CAN SEE THAT THE ALGORITHM NICELY IDENTIFIES THE PART THAT SUFFERS FROM A HIGH INTERFERENCE TERM. BASED ON THIS ESTIMATION, THE RECEIVER DESIGNS
A WEIGHT VECTOR THAT IS APPLIED TO THE RECEIVED SIGNAL. DIAGRAM (B) ADDITIONALLY SHOWS THE RECEIVED SIGNAL AFTER IT HAS BEEN MULTIPLIED WITH THE WEIGHTING
VECTOR, AND WE CAN SEE THAT THE MUI HAS BEEN SUCCESSFULLY REMOVED.
0
0.2
0.4
0.6
0.8
1
P
r
o
b
a
b
i
l
i
t
y

s
1
rx signal user of interest
rx signal
probability of low interference state
1.573 1.5737 1.5744
x 10
0
1
-1
2
-2
E
n
e
r
g
y
Time (ns)
5 (a)
1.573 1.5737 1.5744
x 10
0
1
-1
2
-2
E
n
e
r
g
y
Time (ns)
5
0
60
-60
121
-121
E
n
e
r
g
y

(
s
c
a
l
e
d

b
y

w
e
i
g
h
t
i
n
g

v
e
c
t
o
r
)
rx signal user of interest
rx signal
weighted rx signal
(b)
124 COPYRIGHT ECOLE POLYTECHNIQUE FDRALE DE LAUSANNE, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
been proposed in [8] for MUD in a synchronous UWB system,
and in [13], [2] as a stand-alone method without interference
modeling and without neglecting the multipath nature of the
UWB channel. The performance gain of our proposal over these
simple thresholding schemes can be seen in Fig. 4.
3. MANAGING INTERFERENCE
AT THE LINK LAYER
In this section, we discuss several link-layer techniques that can
be used to react and adapt to interference. Link layer techniques
control transmission parameters and the retransmission behavior
at the sender. Their goal is to adapt to the level of interference
experienced at the receiver. When the interference at the receiver
is low, adaptive transmission techniques allow for the increase
of the throughput. On the other hand, when interference at the
receiver is high, adaptive techniques avoid communication
outages and ensure a minimum throughput.
In the case of IR-UWB communications, the transmission
parameters to adapt can be the modulation order (number of
bits per symbol), the power, the rate of the channel code, or the
processing gain.
In order to adapt to these parameters, the transmitter must have
an estimate of the level of interference at its intended receiver.
In the context of uncoordinated networks, most techniques make
use of feedback information from the receiver to the sender.
Feedback information from the receiver can take various forms.
It is often a function of the SINR. Several other examples can be
found in [15]. However, with a UWB physical layer, measuring
the SINR is difficult in practice due to the very low transmit
power of UWB signals. For instance, the DCC-MAC protocol
discussed in Section 4 relies on information produced by the
channel decoder rather than on physical layer measurements.

3.1 Adaptation of the transmission parameters to the level
of interference
Adaptive modulation [16] allows for the efficient adaption of
the spectral efficiency to the level of interference. Adaptive
modulation essentially varies the number of bits per symbols.
In the case of IR-UWB, a simple and efficient technique is
to use M-ary PPM [17]. A further technique is the use of
joint modulation and coding, such as bit-interleaved coded
modulation (BICM) [18], [19]. Even in the presence of multi-
user interference (MUI), such a technique can considerably
increase the throughput of an IR-UWB link [20].
The amount of redundancy of the channel code, and hence the
rate, can be adapted to the level of interference. Practical schemes
such as rate-compatible punctured convolutional (RCPC) codes
[21] can be used. One of their main advantages is that only one
decoder is necessary at the receiver for a given family of RCPC
codes. Another interesting feature is that they can be used with
incremental redundancy techniques (see Section 3.2). An issue
that arises with IR-UWB physical layers and channel coding
is the detrimental effect that impulsive interference can cause.
FIGURE 4: WE COMPARE OUR INTERFERENCE MITIGATION TECHNIQUE WITH
A RECEIVER THAT NEGLECTS MULTI-USER INTERFERENCE (MUI) COMPLETELY
AND WITH A RECEIVER PERFORMING ONLY SIMPLE THRESHOLDING.
PHYSICAL LAYER PACKETS ARE GENERATED ACCORDING TO A POISSON PROCESS
AT HALF THE PEAK DATA RATE.
THE CHANNEL MODEL WE SIMULATE IS THE 802.15.4A INDOOR NLOS MODEL.
FURTHER, WE HAVE FOUR NEAR INTERFERERS WITH POWER LEVELS OF 10DB,
13DB, 16DB AND 20DB WITH RESPECT TO THE USER OF INTEREST. IT CAN
BE SEEN, THAT THE PERFORMANCE GAIN FROM MODELING THE INTERFERENCE
IS SIGNIFICANT. USING THE MORE SOPHISTICATED HMM TO CHARACTERIZE MUI
GIVES AN ADDITIONAL GAIN COMPARED TO THAT GIVEN THE GMM MODEL
0 4 8 18
10
0
10
-1
10
-2
10
-3
10
-4
10
-5
Eb/N0 dB
14 2 6 10 16 12
No mitigation
Threshold based
GMM
HMM
MANAGING IMPULSIVE INTERFERENCE IN IMPULSE RADIO UWB NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
125
If soft-decision decoding is used, a large interference sample
(for instance, in the case of near-far interference) can propagate
through the trellis of the decoder and result in several decoding
errors. However, if hard-decision decoding is used, this effect
is prevented. But then, the performance when only regular
Gaussian noise is present is impacted. Intuitively, the optimal
decoding policy should consist of an adaptive combination
of hard-decision when strong interferers are present and soft-
decision otherwise [2]. Hence, an interference mitigation scheme
(Section 2.2) should be used.
Adapting the processing gain for IR-UWB has been suggested
in [17], [22]. It is possible to either change the average pulse
repetition frequency or to change the number of pulses per
symbol
4
. The issues of near-far interference also apply in this
case. Changing the processing gain also has an impact on the
average emitted power.
Note that adaptive modulation and adaptive channel coding are
rate adaptation techniques. They are also procedures that are local
to a single sender and receiver pair, that is, only communication
between the source and the destination is required to perform them.
With power control [23], [24], a transmitter ensures that the
received SINR at its destination remains higher than a given
threshold. This threshold depends on the current level of
interference at the receiver.
Contrary to adaptive modulation and adaptive channel coding,
performing power control is a global procedure. Coordination
is required not only between the source and its destination, but
also with the neighbors of the transmitter. The transmitter should
make sure that it does not destroy any ongoing transmission by
reducing the SINR at receiving nodes in its vicinity. This requires
the estimation of the channel gain between the transmitter and
each node in the range of the transmitter.
The choice of rate adaptation and/or power control for IR-UWB
networks is analyzed in [25]. When the objective is to maximize
the overall throughput of the network, it turns out that the
optimum is to use rate adaptation and no power control. If the
primary objective is to minimize the energy consumption, this
is still an open issue. Still, some results in [1] suggest that rate
adaptation with no power control is not far from being optimal.

3.2. Adaptive retransmission techniques: Hybrid-ARQ with
incremental redundancy
The techniques discussed in the previous section allows for
the adaption of the parameters of the transmitted signal to the
estimated level of interference at the receiver. However, there are
two issues associated with these techniques. First, the feedback
returned by the receiver is only an estimation of the level of
interference at the receiver. Second, the level of interference can
change significantly between the time the feedback is received
and the time where the transmission occurs. In the first case, a
solution is to include a safety margin. However, in the second case,
an increase of the level of interference can arise and induce an
error on the transmission between the source and the destination.
Hence, there is a need for an efficient retransmission procedure
in case of a transmission failure. Such schemes have been
extensively studied in the literature. They are denominated under
the general term of Automatic Retransmission reQuest (ARQ).
For an extensive overview of ARQ mechanism, the reader can
consult [26], [15] and the references therein. In the remainder of
this section, we will concentrate on adaptive mechanisms, the
so-called hybrid-ARQ schemes.
In its simplest form, an ARQ scheme retransmits the same
packet until successful reception occurs at the receiver. The
feedback is binary and indicates whether or not the packet was
properly received. However, this scheme will fail in the event of
a strong and lasting interference; indeed, the data transmission
will fail at each retransmission. Therefore, current ARQ
techniques are adaptive. In most cases, the ARQ mechanism
4. Pulse repetitions are a special case of channel coding. Indeed, it is nothing but a repe-
tition code.
126 COPYRIGHT ECOLE POLYTECHNIQUE FDRALE DE LAUSANNE, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
is combined with a variable-rate channel code. If the initial
data transmission fails, the retransmission occurs with the
data encoded with a more powerful code, i.e., at a lower rate
5
.

A further improvement of this scheme can be obtained by using
incremental redundancy. Instead of retransmitting the whole data
encoded at a lower rate, only the coded information necessary
to obtain the lower rate is sent. For instance, channel codes
such as the RCPC code in [21], [27] can provide incremental
redundancy. For a specific family of RCPC codes, a code of a
given rate is a subset of all the codes with a higher rate.
Note that these schemes do not specify how the rate should be
adapted. Hence, there is a large amount of freedom left for how
the overall retransmission mechanism can be designed.
The design of an adaptive ARQ scheme is largely dictated by
the flexibility of the channel code and by the type of feedback
available between a receiver and the sender.
4. DCC-MAC: AN UNCOORDINATED
MAC PROTOCOL FOR UWB
NETWORKS WITH RATE ADAPTATION
AND INTERFERENCE MITIGATION
In this section, we present the case study of a system designed
to be interference-aware. We consider the organization of non-
coordinated and asynchronous medium-access (MAC) protocol
for UWB networks. One proposal is the DCC-MAC protocol [2].
DCC-MAC is an interference-aware design that is conceived to
operate in a flawless manner in the presence of strong impulsive
interference.
In order to compare the performance of DCC-MAC against a
non interference-aware protocol, we compare DCC-MAC with
the (UWB)
2
protocol. (UWB)
2
is a more recent proposal that is
not interference-aware and does not support any mechanisms to
combat impulsive interference. The main characteristics of the
two protocols are summarized in Table 1.
In the following, we first describe the two main components
of DCC-MAC that permit to combat interference, namely rate-
adaptation and interference mitigation. In addition, we briefly
describe the main aspects of our protocol. Then, we present
simulation results that compare DCC-MAC with the (UWB)
2

protocol [3].
4.1. Rate-adaptation and interference mitigation
in the DCC-MAC protocol
The main ingredients of DCC-MAC to combat interference are
a rate-adaptation mechanism and an interference-mitigation
scheme. Rate-adaptation is obtained by using RCPC channel
codes. The modulation and the processing gain is fixed. The
family of RCPC codes is the one described in [27]. It offers a set
of twenty-five channel code rates that can be extended to thirty.
Only one pair of channel code encoder and decoder is necessary.
The rate-adaptation scheme is based on an additive-increase,
multiplicative-decrease (AIMD) policy similar to what is used by
TCP. Whenever a packet is successfully received, the destination
takes advantage of the decoding process to estimate the maximum
rate at which the data transmission could have occurred [2]. The
receiver subtracts a safety margin and returns this information
back to sender in the acknowledgment packet. Hence, in the case
of DCC-MAC, the feedback consists of the estimated rate at
which the next data transmission should take place.
Interference mitigation is used, albeit in the simple form of
thresholding. The mechanism is similar to what is explained
in Section 2.2. The transmission of data to a destination is
performed using a time-hopping sequence unique to the
destination. This time-hopping sequence can be created by
seeding a pseudo-random number generator with a unique
5. A different modulation could be used, but this is hardly done in practice.

TABLE 1: MAIN ASPECTS
AND DIFFERENCES
OF THE DCC-MAC
AND THE (UWB)
2

PROTOCOL FOR
NON-COORDINATED
IMPULSE-RADIO
UWB NETWORKS

DCC-MAC (UWB)
2

Interference-aware Non interference-aware
Interference mitigation n/a
Rate adaptation n/a
No control Common control
channel required, channel with
no RTS-CTS RTS-CTS
MANAGING IMPULSIVE INTERFERENCE IN IMPULSE RADIO UWB NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
127
network throughput of the two protocols is compared in a near-
far scenario. An example of the near-far topology for six nodes,
i.e., three concurrent communication links, is represented in
Fig. 6. Again, the throughput of the (UWB)
2
protocol decreases
with the number of concurrent links. On the other hand, the
DCC-MAC protocol can cope with the increasing number of
concurrent transmissions. Additional simulation results for
DCC-MAC can be found in [2] for various scenarios.
5. CONCLUSION
We have discussed the management of impulsive interference in
IR-UWB networks. We have shown that this kind of interference
is an issue in IR-UWB and has therefore to be taken care of.
identifier for the destination. Such an identifier can be, for
instance, the hardware address.
A typical transmission consists of a data packet transmission from
the source to the destination, an acknowledgment sent back by the
destination and the transmission of an IDLE packet from the source.
Hence, it has a simple design that does not require any common
control channel nor the use of any RTS-CTS type of handshake.
Along with a subtle control of timers and a careful use of
time-hopping sequences, the IDLE packet is necessary for the
protocol to operate properly in the absence of carrier-sensing as
well as in multi-hop environments [2].
4.2. Performance evaluation of the DCC-MAC and (UWB)
2

protocols
In order to emphasize the importance of an interference-aware
design, we compare the DCC-MAC protocol with the (UWB)
2

protocol [3]. Contrary to DCC-MAC, (UWB)
2
needs a common
control channel and uses an RTS-CTS handshake to arbitrate
access to a destination. (UWB)
2
uses neither power-control
nor rate-adaptation. Interference is not mitigated at the physical
layer. Table 1 summarizes the main aspects and differences of
the two protocols.
We use the ns-2 simulator [28] with an extension for UWB
physical layers. The code for the UWB extension is available
online at [29]. The parameters correspond to a typical 802.15.4a
scenario. The maximum rate of the physical layer is 1 Mbit/s.
For every scenario, the link distance is 10 meters. The transport
protocol is UDP. The throughput is the saturation throughput.
In Fig. 5, the network throughput of the two protocols is
compared in a multi-hop scenario. The topology is a line of
n nodes where one extremity of the line sends to the other
extremity. The throughput of (UWB)
2
drops dramatically as the
number of hops increases. On the contrary, the throughput of
DCC-MAC remains stable for more than three hops. In Fig. 7, the
FIGURE 5: PERFORMANCE OF THE DCC-MAC PROTOCOL AND THE (UWB)
2

PROTOCOL IN A MULTI-HOP SCENARIO. THE TOPOLOGY IS A LINE OF NODES
WITH A LINK DISTANCE OF 10 METERS. THE THROUGHPUT IS PLOTTED AGAINST
THE NUMBER OF NODES. THE TRANSMITTER AND RECEIVER ARE LOCATED AT EACH
EXTREMITY OF THE LINE. THE TRANSPORT PROTOCOL IS UDP.
THE DCC-MAC PROTOCOL CLEARLY OUTPERFORMS THE (UWB)
2
PROTOCOL.
2 4 6 8 10 12
0
50
100
150
200
250
300
350
400
450
Number of nodes
T
h
r
o
u
g
h
p
u
t

[
K
b
i
t
/
s
]
UWB
DCC-MAC
2
FIGURE 6:
NEAR-FAR TOPOLOGY
FOR SIX NODES, I.E.,
WITH THREE CONCURRENT
COMMUNICATION LINKS.
THE LINK DISTANCE
IS d. THE DIRECTION
OF THE COMMUNICATION
IS ALTERNATED FROM
ONE LINK TO ANOTHER.
n
0
n
1
distance
n
2
n
3
n
4
n
5
128 COPYRIGHT ECOLE POLYTECHNIQUE FDRALE DE LAUSANNE, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
We have further presented several techniques and proposals that
address impulsive interference on the physical as well as on the
link layer. Additionally, interference could also be managed
on the network layer. There is already some work on routing
protocols that try to route packets such that interference is limited.
There are also other aspects that we have left out. We have not
discussed the effect of these schemes on energy consumption.
There is of course a tradeoff. A better system performance
reduces the number of retransmissions and hence decreases
the energy consumption. On the other hand, more complex
transceiver designs increase the energy-consumption. Another
important aspect of IR-UWB is its ranging capability. As for
detection or channel estimation, interference will most probably
matter, and some ways to deal with it will have to be considered.
REFERENCES
[1] A. El Fawal, J.-Y. Le Boudec, R. Merz, B. Radunovic,
J. Widmer, G. M. Maggio, TRADEOFF ANALYSIS OF PHY-
AWARE MAC IN LOW-RATE, LOW-POWER UWB NETWORKS,
IEEE Communications Magazine, vol. 43, no. 12,
pp. 147155, December 2005.
[2] R. Merz, J. Widmer, J.-Y. Le Boudec, B. Radunovic,
A JOINT PHY/MAC ARCHITECTURE FOR LOW-RADIATED
POWER TH-UWB WIRELESS AD-HOC NETWORKS,
Wireless Communications and Mobile Computing Journal,
Special Issue on Ultrawideband (UWB) Communications,
vol. 5, no. 5, pp. 567580, August 2005.
[3] M.-G. Di Benedetto, L. Nardis, M. Junk, G. Giancola,
(UWB)
2
: UNCOORDINATED, WIRELESS, BASEBORN, MEDIUM
ACCESS CONTROL FOR UWB COMMUNICATION NETWORKS,
Mobile Networks and Applications,
vol. 10, no. 5, October 2005.
[4] M. Z. Win, R. A. Scholtz, ULTRA-WIDE BANDWIDTH
TIME-HOPPING SPREAD-SPECTRUM IMPULSE RADIO
FOR WIRELESS MULTIPLEACCESS COMMUNICATIONS,
IEEE Transactions on Communications,
vol. 48, no. 4, pp. 679691, April 2000.
[5] R. C. Qiu, H. Liu, X. Shen, ULTRA-WIDEBAND
FOR MULTIPLE ACCESS COMMUNICATIONS, Communications
Magazine, IEEE, vol. 43, no. 2, pp. 8087, 2005.
[6] S. Verd, Multiuser Detection. Cambridge University
Press, 1998.
[7] Y. C. Yoon, R. Kohno, OPTIMUM MULTI-USER DETECTION
IN ULTRA-WIDEBAND (UWB) MULTIPLE-ACCESS
COMMUNICATION SYSTEMS, in ICC 2002 - IEEE
International Conference on Communications, no. 1,
Apr. 2002, pp. 812816.
[8] E. Fishler, H. V. Poor, LOW-COMPLEXITY MULTIUSER
DETECTORS FOR TIME-HOPPING IMPULSE-RADIO SYSTEMS,
IEEE Transactions on Signal Processing, vol. 52, no. 9,
pp. 25612571, September 2004.
[9] B. Hu, N. Beaulieu, ACCURATE EVALUATION OF
MULTIPLE-ACCESS PERFORMANCE IN TH-PPM AND TH-
TABLE II. PERCENTAGE OF ERROR BETWEEN
THE MODEL PREDICTION AND THE ACTUAL
MEASURED ADDITIONS AND MULTIPLICATIONS
PER PIXEL. FOR EACH GOP OF EVERY
SEQUENCE, THE AVERAGE ERROR OVER A
NUMBER OF ADAPTATION POINTS IS PRESENTED.
FIGURE 7: PERFORMANCE OF THE DCC-MAC PROTOCOL AND THE (UWB)
2

PROTOCOL IN A NEAR-FAR SCENARIO. THE LINK DISTANCE IS 10 METERS.
THE THROUGHPUT IS PLOTTED AGAINST THE NUMBER OF CONCURRENT
COMMUNICATION LINKS. THE TRANSPORT PROTOCOL IS UDP. THE DCC-MAC
PROTOCOL CLEARLY OUTPERFORMS THE (UWB)
2
PROTOCOL.
THE (UWB)
2
PROTOCOL SUFFERS FROM THE CONTENTION ON THE COMMON
CHANNEL AS WELL AS THE RATE DROP DUE TO THE RTS-CTS EXCHANGE.
1.5 2 2.5 3 3.5 4 4.5
0
50
100
150
200
250
300
350
400
450
Number of concurrent links
T
h
r
o
u
g
h
p
u
t

[
M
b
i
t
/
s
]
UWB
DCC-MAC
2
MANAGING IMPULSIVE INTERFERENCE IN IMPULSE RADIO UWB NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
129
BPSK UWB SYSTEMS, IEEE Transactions on
Communications, vol. 52, no. 10, pp. 17581766, Oct. 2004.
[10] V. Lottici, A. DAndrea, U. Mengali, CHANNEL
ESTIMATION FOR ULTRA-WIDEBAND COMMUNICATIONS,
IEEE Journal on Selected Areas in Communications,
vol. 20, no. 9, pp. 16381645, Dec. 2002.
[11] V. Cellini, G. Dona, A NOVEL JOINT CHANNEL
AND MULTI-USER INTERFERENCE STATISTICS ESTIMATOR
FOR UWB-IR BASED ON GAUSSIAN MIXTURE MODEL,
in IEEE International Conference on Ultra-Wideband
(ICU 2005), Sept. 2005, pp. 655660.
[12] M. Flury, J.-Y. Le Boudec, INTERFERENCE MITIGATION
BY STATISTICAL INTERFERENCE MODELING IN AN IMPULSE
RADIO UWB RECEIVER, in International Conference
on Ultra-Wideband (ICUWB 2006),
Waltham, MA, September 2006.
[13] W. M. Lovelace and J. K. Townsend, THRESHOLD
DISCRIMINATION AND BLANKING FOR LARGE NEAR-FAR
POWER RATIOS IN UWB NETWORKS, IEEE Transactions on
Communications, vol. 53, no. 9, pp. 14471450, Sept. 2005.
[14] A. El Fawal, J.-Y. Le Boudec, A ROBUST SIGNAL
DETECTION METHOD FOR ULTRA WIDE BAND (UWB)
NETWORKS WITH UNCONTROLLED INTERFERENCE,
IEEE Transactions on Microwave Theory and Techniques,
2006, to appear.
[15] S. Nanda, K. Balachandran, and S. Kumar, ADAPTATION
TECHNIQUES IN WIRELESS PACKET DATA SERVICES, IEEE
Communications Magazine, vol. 38, no. 1,
pp. 5464, January 2000.
[16] S. T. Chung, A. Goldsmith, DEGREES OF FREEDOM
IN ADAPTIVE MODULATION: A UNIFIED VIEW, IEEE
Transactions on Communications, vol. 49, no. 9,
pp. 15611571, September 2001.
[17] N. August, R. Thirugnanam, and D. Ha, AN ADAPTIVE
UWB MODULATION SCHEME FOR OPTIMIZATION OF
ENERGY, BER, DATA RATE, in UWBST, 2004, pp. 182186.
[18] G. Caire, G. Taricco, E. Biglieri, BIT-INTERLEAVED
CODED MODULATION, IEEE Transactions on Information
Theory, vol. 44, no. 3, pp. 927946, May 1998.
[19] Y. Souilmi, R. Knopp, CODE CONSTRUCTIONS FOR
NON-COHERENT ON-OFF ULTRA-WIDEBAND SYSTEMS,
in IEEE international conference on Ultra Wide Band,
September 2005, pp. 2832.
[20] R. Merz, J.-Y. Le Boudec, EFFECT OF INTERFERING
USERS ON THE MODULATION ORDER AND CODE RATE
FOR UWB IMPULSE-RADIO BIT-INTERLEAVED CODED
M-ARY PPM, in IEEE UWBnets workshop,
October 2005.
[21] J. Hagenauer, RATE-COMPATIBLE PUNCTURED
CONVOLUTIONAL CODES (RCPC CODES) AND THEIR
APPLICATIONS, IEEE Transactions on Communications,
vol. 36, no. 4, pp. 389400, April 1988.
[22] I. Guvenc, H. Arslan, S. Gezici, H. Kobayashi,
ADAPTATION OF MULTIPLE ACCESS PARAMETERS IN TIME
HOPPING UWB CLUSTER BASED WIRELESS SENSOR
NETWORKS, in IEEE International Conference
on Mobile Ad-hoc and Sensor Systems (MASS),
October 2004, pp. 235244.
[23] M. Krunz, A. Muqattash, S.-J. Lee, TRANSMISSION
POWER CONTROL IN WIRELESS AD HOC NETWORKS:
CHALLENGES, SOLUTIONS AND OPEN ISSUES,
IEEE Network, vol. 18, no. 5, pp. 814, Sept.-Oct 2004.
130 COPYRIGHT ECOLE POLYTECHNIQUE FDRALE DE LAUSANNE, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
[24] F. Cuomo, C. Martello, A. Baiocchi, C. Fabrizio,
RADIO RESOURCE SHARING FOR AD HOC NETWORKING
WITH UWB, IEEE Journal on Selected Areas
in Communications, vol. 20, no. 9, pp. 17221732,
December 2002.
[25] B. Radunovic, J. Y. Le Boudec, OPTIMAL POWER
CONTROL, SCHEDULING AND ROUTING IN UWB
NETWORKS, IEEE Journal on Selected Areas
in Communications, vol. 22, no. 7, pp. 12521270,
September 2004.
[26] D. Costello, J. Hagenauer, H. Imai, S. Wicker,
APPLICATIONS OF ERROR-CONTROL CODING,
IEEE Transactions on Information Theory,
vol. 44, no. 6, pp. 25312560, October 1998.
[27] P. Frenger, P. Orten, T. Ottosson, A. Svensson,
RATE-COMPATIBLE CONVOLUTIONAL CODES FOR
MULTIRATE DS-CDMA SYSTEMS, IEEE Transactions
on Communications, vol. 47, no. 6, pp. 828836,
June 1999.
[28] THE NETWORK SIMULATOR NS-2,
http://www.isi.edu/nsnam/ns/.
[29] UWB RESEARCH AT EPFL-IC,
http://icawww1.epfl.ch/uwb/, 2006.
CONTACT: ST.JOURNAL@ST.COM
THE ALOHA ACCESS (UWB)
2
PROTOCOL REVISITED FOR IEEE 802.15.4A
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
131
THE ALOHA ACCESS
(UWB)
2
PROTOCOL
REVISITED FOR IEEE
802.15.4A
The IEEE 802.15.4a Task Group recently
proposed Impulse Radio Ultra Wide
Band (IR-UWB) for a physical layer that
can provide combined communication
and ranging in low data rate indoor/
outdoor networks. At present, it is
therefore particularly relevant to design
IEEE 802.15.4a MAC strategies
that are appropriately tailored on the
physical layer. Previously, we proposed
the Uncoordinated Baseborn Wireless
medium access control for UWB networks
(UWB)
2
, a UWB-tailored MAC based
on the low probability of pulse collision.
The (UWB)
2
adopted the Aloha principle
and enabled location-based network
optimization by providing and storing
estimates of distance between nodes.
This paper first revisits the (UWB)
2
MAC
protocol in view of its application to IEEE
802.15.4a. The structure of both control
and data MAC protocol data units is
defined based on the legacy 802.15.4
MAC in order to allow a seamless
Maria-Gabriella Di Benedetto, Luca De Nardis,
Guerino Giancola, Daniele Domenicali
School of Engineering
University of Rome La Sapienza
support, for both centralized and distributed
network topologies, as defined in the parent
standard. Secondly, this work extends and
completes the analysis of (UWB)
2
since it
takes into account multipath-prone channels.
Channel parameters, for both indoor and
outdoor propagation scenarios in Line Of
Sight (LOS) and Non-Line Of Sight (NLOS)
conditions, were derived from the channel
model defined within the 802.15.4a channel
sub-committee.
Results highlight that the (UWB)
2
protocol
is robust to multipath, and provides high
throughput and low delay, with performance
scaling gracefully as a function of the number
of users and the user bit rate. Results confirm
and support the adoption
of (UWB)
2
principles for low data rate UWB
communications.
Index Terms - Ultra Wide Band, MAC, Low
Data Rate
COPYRIGHT UNIVERSIT DEGLI STUDI DI ROMA LA SAPIENZA, 2007.
132 COPYRIGHT UNIVERSIT DEGLI STUDI DI ROMA LA SAPIENZA, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
1. INTRODUCTION
Low data rate and low cost networks for mixed indoor/outdoor
communications are of great interest in sensor and ad-hoc
networking. The interest towards low data rate networks led in
2003 to the definition of the IEEE 802.15.4 standard for low
rate, low complexity, and low power wireless networks [1]. The
802.15.4 standard also forms the basis of the ZigBee technology,
which provides a comprehensive solution for low data rate
networking from physical layer to applications [2].
Both IEEE 802.15.4 and ZigBee lack, however, an important
requirement of forthcoming low data rate systems, namely,
the capability of locating, with sufficiently high precision,
objects and individuals by means of distributed, infrastructure-
independent positioning algorithms.
The introduction of positioning in low data rate networks is a
top priority of the recently formed IEEE 802.15.4a Task Group
[3], which recently proposed Impulse Radio Ultra Wide Band
(IR-UWB) as an attractive transmission technique for indoor and
outdoor low data rate wireless networks [4], [5]. Thanks to its
ultra wide bandwidth that spans over several GHz, IR-UWB has
some interesting properties, in particular:
an inherently high temporal resolution that provides good
robustness in the presence of multipath, thereby allowing
communication despite obstacles and Non-Line-Of-Sight
(NLOS) propagation conditions.
the capability of providing accurate ranging, thanks to
its high temporal resolution. Distance information can
then be used for deriving the physical position of terminals
in the network.
The definition of the Uncoordinated Baseborn Wireless medium
access control for UWB networks (UWB)
2
protocol was
based on the above specific features of IR-UWB [6]. (UWB)
2

also evaluates and stores distances, which are then used by
positioning and routing algorithms. The (UWB)
2
approach for
propagation over AWGN channels was validated in [7].
In the present work, we revisit (UWB)
2
by redefining the
structure of both control and DATA MAC Protocol Data Units
(MACPDUs) based on the PDU structure of the original IEEE
802.15.4 MAC standard in order to guarantee compatibility of
the new MAC protocol with both distributed and centralized
network topologies defined in the 802.15.4 standard.
Next, we extend the analysis of the (UWB)
2
protocol to the case of
multipath-affected channels, for both indoor and outdoor channel
scenarios in Line Of Sight (LOS) and Non-Line Of Sight (NLOS)
conditions. Channel parameters were obtained from the channel
model proposed within the 802.15.4a Task Group, and a set of
channel realizations were considered for each selected scenario.
Finally, Multi User Interference (MUI) was also included in
the performance analysis based on an enhanced version of the
Pulse Collision model specific for IR-UWB [7], which takes
into account multipath. This MUI model is used to analyze
performance of (UWB)
2
by simulation, as a function of channel,
network size, and user bit rates.
The paper is organized as follows: Section 2 summarizes (UWB)
2

and the ranging scheme, and Section 3 defines the format of the
(UWB)
2
MAC Protocol Data Unit, based on the structure defined
within the preexisting 802.15.4 MAC. The Pulse Collision MUI
model is introduced in Section 4. Performance evaluation of
(UWB)
2
in presence of multipath and MUI for different number
of users and offered traffic is carried out in Section 5. Section 6
draws conclusions.
2. THE (UWB)
2
MAC PROTOCOL
The high temporal resolution of IR-UWB signals has the
beneficial side effect of reinforcing robustness to MUI, in
particular for low data rate applications [4]. As a consequence,
access to the medium in low data rate UWB networks can be
based on a most straightforward solution, Aloha ([8], [6]), by
which devices transmit in an uncoordinated fashion. Thanks to
the resilience to MUI offered by impulse radio, correct reception
THE ALOHA ACCESS (UWB)
2
PROTOCOL REVISITED FOR IEEE 802.15.4A
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
133
for multiple simultaneous links can be obtained. An Aloha-like
approach may also favor lowering costs, since it does not rely on
specific physical layer (PHY) functions, such as carrier sensing,
and may thus be adapted with little effort to different PHYs.
As for the duty cycle of emitted signals, low data rate scenarios
usually lead to an average Pulse Repetition Period (PRP), the
average time between two consecutive pulses emitted by a
device, on the order of 10
-4
10
-5
s, with an average duration of
emitted pulses typically on the order of 10
-10
s. Theoretically,
the duty cycle can thus be as low as 10
-6
. However, a detailed
analysis of this issue requires introducing the channel model in
order to take into account propagation effects on pulse duration.
When Time Hopping (TH) is the selected coding technique,
TH Code Division Multiple Access (TH-CDMA) is a natural
choice for multiple access. The adoption of TH-CDMA can
introduce an additional degree of freedom, since the effect of
pulse collisions is further reduced by the adoption of different
codes on different links. Two factors cooperate in determining
the robustness to MUI: low duty cycle of emitted signals, and
association of different TH-Codes with different links.
(UWB)
2
is a multi-channel MAC protocol that is based on the
combination of Aloha with TH-CDMA [6]. (UWB)
2
adopts the
combination of a common code for signaling, where terminals
share the same code, and code collisions are avoided thanks
to phase shifts between different links, and Transmitter codes
for data transfers, where each terminal has a unique code
for transmitting, and the receiver switches to the code of the
transmitter for receiving a packet.
The packet exchange between transmitter TX and receiver RX that
takes place during connection set-up may also serve for enabling
a simple ranging procedure, based on a three-way exchange.
During set-up, TX and RX prepare a DATA packet transmission
by exchanging a Link Establishment (LE) packet transmitted on
the Common Code, followed by a Link Confirm (LC) packet
transmitted on the Transmitter Code of RX, and finally by the
DATA packet on the Transmitter Code of TX. This handshake
allows storing distance between TX and RX at both TX and RX.
The protocol also foresees the presence of a procedure by which
each terminal i maintains a ranging database for all neighboring
terminals. Each entry of the database contains the ID j of the
neighbor, the estimated distance to j, and a timestamp indicating
the time at which the estimation was performed.
3. THE (UWB)
2
MACPDU FORMAT
The format of the MACPDU originally proposed in [6]
was revisited and modified in order to take into account the
characteristics of the future IEEE 802.15.4a PHY.
The MACPDU is composed of a header, a payload, and a trailer.
The standard header, shared by all PDUs and long up to 23
bytes, is derived from the 802.15.4 header and is organized as
follows:
frame control (2 bytes)
equence number (1 byte)
destination PAN identifier (2 bytes)
destination address (2/8 bytes)
source PAN identifier (2 bytes)
source address (2/8 bytes)
In the case of LE control packets (link set-up phase of
(UWB)
2
), the header includes the following additional fields:
Time Hopping flag (1 bit), used to inform destination
whether the standard Time Hopping code or a different
one is going to be adopted in the DATA transmission;
Time Hopping code (0/2 bytes), used for communicating
the TH code to the destination (e.g., by including the code
identifier, assuming that all nodes share a common
codebook).
In the case of DATA PDU, the header contains the 1 byte
additional field N
PDU
that indicates to the destination the
number of additional DATA PDUs that will be sent from the
source. If N
PDU
is different from 0, the destination will keep on
134 COPYRIGHT UNIVERSIT DEGLI STUDI DI ROMA LA SAPIENZA, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
listening on the DATA TH code and wait for additional DATA
PDUs. The length of the payload is set to 0 for LE and LC PDUs,
while the ACK PDU has a 2 byte payload containing the status
of the corresponding DATA PDU. Finally, DATA PDUs have a
payload length of up to 103 bytes.
All PDUs include a 2 byte trailer consisting of a CRC code
evaluated on the entire PDU.
The above PDU structure leads to a maximum PDU length of
129 bytes, corresponding to the case of a DATA PDU with full
header (24 bytes), full payload (103 bytes), and the 2 byte trailer.
4. BER EVALUATION UNDER
THE PULSE COLLISION MODEL
4.1. System model
We assume IR-UWB transmissions with Pulse Position
Modulation (PPM) and TH coding. Signals generated at TX are
described as follows:
s
TX
t ( ) = E
TX
p
0
t jT
S

j
b
j N
S
( )
j
,
[
[
where p
0
(t) is the energy-normalized waveform of the transmitted
pulses, E
TX
is the transmitted energy per pulse, T
S
is the average
pulse repetition period, 0
j
<T
S
is the TH time shift of the j-th
pulse, is the PPM shift, b
x
is the x-th bit of a binary source
sequence b, N
S
is the number of pulses transmitted for each bit,
and |x| is the inferior integer part of x.
Propagation for link m occurs over a multipath-affected channel
with impulse response given by:
h
m ( )
t ( ) = X
m ( )

k,l
m ( )
k=0
K
l=0
L
m ( )
t t
m ( )
T
l
m ( )

k,l
m ( )
( )
,
where X
(m)
is the amplitude gain, L
(m)
is the number of clusters,
K is the number of paths that are considered within each cluster,
(t) is the Dirac function, t
(m)
is the propagation delay, T
l
(m)

is the delay of the l-th cluster with respect to t
(m)
,
k,l
(m)
is the
delay of the k-th path relative to the l-th cluster arrival time, and

k,l
(m)
is the real-valued tap weight of the k-th path within the
l-th cluster. Tap weights are energy-normalized and thus verify:

k,l
(m)
( )
2
k=0
K
=1
l=0
L
m ( )
,
For all channel parameters in (2), we adopt the statistical
characterization that is suggested in [10] for 9 different
propagation environments, i.e., i) residential LOS, ii) residential
NLOS, iii) office LOS, iv) office NLOS, v) outdoor LOS, vi)
outdoor NLOS, vii) industrial LOS, viii) industrial NLOS, and
ix) open outdoor environment NLOS (farm, snow-covered open
area).
For link m, both channel gain X
(m)
and propagation delay t
(m)

depend on distance of propagation D
(m)
between TX and RX.
For X
(m)
, in particular, one has:
X
m ( )
=1 10
PL
m ( )
10
( )
,
where PL
(m)
is the path loss in dB, which can be modelled as
indicated in [10].
Reference TX and RX are assumed to be perfectly synchronized.
The channel output is corrupted by thermal noise and MUI
generated by N
i
interfering and asynchronous IR-UWB devices.
The received signal at RX input writes:
s
RX
t ( ) = r
u
t ( ) + r
mui
t ( ) + n t ( ),
where r
u
(t), r
mui
(t), and n(t) are the useful signal, MUI, and
thermal Gaussian noise with double-sided power spectral density
N
0
/2, respectively. By denoting as 0 the reference link between
TX and RX, the useful signal r
u
(t) writes as follows:
r
u
t ( ) = E
0

k,l
0 ( )
k=0
K
l=0
L
0 ( )

j
p
0
t jT
S

j
0 ( )
b
j/N
S
|
|
|
|
t
0 ( )
T
l
0 ( )

k,l
0 ( )
( )
,
where E
0
= (X
(0)
)
2
E
TX
is the total received energy per pulse.
(1)
(2)
(3)
(4)
(5)
(6)
THE ALOHA ACCESS (UWB)
2
PROTOCOL REVISITED FOR IEEE 802.15.4A
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
135
scenarios, for example, it might be better to have very cheap
devices with modest performance with respect to high-priced
terminals with better performance. In the examined scenario, we
adopt a basic IR receiver that analyzes a single component of the
received signal. This basic receiver is composed by a coherent
correlator followed by an ML detector [4]. In every bit period,
the correlator converts the received signal in (5) into a decision
variable Z that forms the input of the detector. Soft decision
detection is performed. For each pulse, we assume that the
correlator locks onto the multipath component with maximum
energy. By indicating with l
M
and k
M
the cluster and the path
of the maximum energy multipath component for the reference
user, the input of the detector Z for a generic bit b
x
is as follows:
Z = s
RX
t
( )
m
x
t T
0 ( )
( )
dt
xN
S
T
S
+ T
0 ( )
x+1 ( ) N
S
T
S
+ T
0 ( )
where
T
0 ( )
= t
0 ( )
+T
l
M
0 ( )
+
k
M
,l
M
0 ( )
and where
m
x
t ( ) = p
0
t jT
S

j ( ) (
p
0
t jT
S

j

( ))
j=xN
S
x+1 ( )N
S
By introducing (5) into (8), we obtain: Z=Z
u
+Z
mui
+Z
n
, where
Z
u
is the signal term, Z
mui
is the MUI contribution, and Z
n
is
the noise contribution, which is Gaussian with zero mean and
variance
n
2
= N
S
N
0
(), where ()=1-R
0
(), and where R
0
() is
the autocorrelation function of the pulse waveform p
0
(t) [4]. Bit
b
x
is estimated by comparing the Z term in (8) with a zero-valued
threshold according to the following rule: when Z is positive
decision is 0, when Z is negative decision is 1.
4.2. BER estimation under the Pulse Collision approach
According to Section 4.1, the average probability of error on the
bit at the output of the detector for independent and equiprobable
As for r
mui
(t), we assume all interfering signals to be
characterized by the same T
S
; thus:
r
mui
t ( ) = E
n

k,l
n ( )
k=0
K
l=0
L
n ( )
j
n=1
N
i

p
0
t jT
S

j
n ( )
b
j/N
S
n ( ) |
|
|
|
|
|
n ( )
t
n ( )
T
l
n ( )

k,l
n ( )
[
\
|

)
j
,
where index n represents the wireless link between the nth
interfering device and RX. In (7), E
n
= (X
(n)
)
2
E
TX
, and
(n)

are the received energy per pulse and the delay for link n. The
terms
j
(n)
, b
x
(n)
and N
S
(n)
in (7) are the time shift of the j-th
pulse for user n, the x-th bit, and the number of pulses per bit,
respectively for user n.
Both TH codes and data bit sequences are assumed to be
randomly generated and correspond to pseudonoise sequences,
that is,
j
(n)
terms are assumed to be independent random
variables uniformly distributed in the range [0,T
S
), and b
x
(n)

values are assumed to be independent random variables
with equal probability to be 0 or 1. Based on the above
assumptions, the N
i
relative delays
(0)

(n)
, with
n = 1,...,N
i
may be reasonably modelled as independent random
variables uniformly distributed between 0 and T
S
.
As well known, the optimum receiver structure for (6) consists
of a RAKE receiver composed of a parallel bank of correlators,
followed by a combiner that determines the variable to be used
for the decision on the transmitted symbol. Each correlator
of the RAKE is locked on one of the different replicas of the
transmitted waveform p
0
(t). The complexity of such a receiver
increases with the number of multipath components that are
analyzed and combined before decision, and can be reduced by
processing only a sub-set of the components that are available at
the receiver input [4].
Such a reduction, however, entails a decrease in the available
useful energy in the decision process, together with a consequent
decrease in receiver performance. As a result, system designers
have the possibility to trade the cost of the devices with the
performance of the physical layer. For some application
(8)
(9)
(10)
(7)
136 COPYRIGHT UNIVERSIT DEGLI STUDI DI ROMA LA SAPIENZA, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
transmitted bits is: BER = Prob{Z<0|b
x
=0} = Prob{Z
mui
<y},
where y = Z
u
+Z
n
is a Gaussian random variable with mean:
y
= N
S

( )

l
M
, k
M
0 ( )
( )
2
E
0
= N
S

( )
E
u
and variance
y
2
= N
S
N
0
(). The quantity E
u
in (11) indicates
the amount of useful energy conveyed by the maximum
multipath contribution. The average BER at the receiver output
can be evaluated by applying the Pulse Collision (PC) approach
[11]. First, we compute the conditional BER for a generic y
value, i.e., Prob{Z
mui
< y

|

y}, and we then average over all
possible y values, that is:
BER = Prob Z
mui
< y | y { } p
Y
y ( ) dy

+
Next, we expand the conditional BER in order to take into
account collisions between pulses of different transmissions. In
a bit period, the number of possible collisions at the input of the
reference receiver, denoted with c, is confined between 0 and
N
S
N
i
, with N
S
pulses per bit and N
i
interfering users. Thus:
BER = P
C
c ( )
c=0
N
S
N
i
Prob Z
mui
< y | y, c { } p
Y
y ( ) dy

+
where P
C
(c) is the probability of having c collisions at the
receiver input. For independent interferers, P
C
(c) can be
expressed through the binomial distribution:
P
C
c ( ) =
N
S
N
i
c
[
\
|

)
j
P
0
( )
c
1 P
0
( )
N
S
N
i
c
,
where P
0
is the basic collision probability, which is defined as
the probability that an interfering device produces a non-zero
contribution within a single T
S
. Given the receiver structure in
(8), we approximate P
0
as follows:
P
0
=
T
m
+ +
MAX
T
S
.
where T
m
is the time duration of the pulse waveform p
0
(t),
and
MAX
is the maximum among the values of the root mean
square delay spread for the N
i
channels between the interfering
devices and RX. Note that (15) provides acceptable P
0
values if
T
S
> T
m
+ +
MAX
, which is reasonable for LDR systems with
long pulse repetition periods. This condition guarantees that no
Inter Frame Interference (ISI) is present at the receiver, even in
the presence of multipath propagation.
As regards Prob(Z
mui
< y | y,c), we adopt the linear model
introduced in [11], that is:
Prob Z
mui
< y y, c
( )
=
1 for y n ( )
1
P
C
c ( )
2
1+
y
c ( )
[
\
|
|

)
j
j
for n ( ) < y 0
P
C
c ( )
2
1
y
c ( )
[
\
|
|

)
j
j
for 0 < y n ( )
0 for y > n ( ),
|

|
|
|
|
|
|
|
|
|
where (c) indicates the maximum interference contribution that
can be measured at the output of the correlator. Based on [11],
we propose here the following approximation for (c):
c ( ) =
c j +1
N
i
|
|
|
|
|
|
E
int
j ( )
T
m
+
rms
j ( )
[
\
|
|

)
j
j
j=1
N
i
where {E
int
(1)
, E
int
(2)
, ... , E
int
(Ni)
} are the interfering
energies {E
1
, E
2
, ... , E
Ni
} of (7), sorted in descending order
so that E
int
(j)
E
int
(j+1)
for j = 1,...,N
i
-1. The expression in
(17) indicates that the value of the maximum interference
contribution at the receiver output is computed privileging
dominating interferers, that is, those users with the highest
interfering energies.
Note that in (17) we multiply the value of jth interfering energy
E
int
(j)
by (T
m
+ )/
rms
(j)
.
This operation indicates that only part of the energy associated
with a colliding pulse contributes to Z in (8), corresponding to
the ratio between the correlator window T
m
+ and the length of
the pulse at the receiver, approximated by
rms
(j)
. By combining
(12)
(13)
(14)
(15)
(16)
(17)
(11)
THE ALOHA ACCESS (UWB)
2
PROTOCOL REVISITED FOR IEEE 802.15.4A
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
137
(16) into (13), one has:
BER
1
2
erfc
1
2
N
S
E
u
0
( )
[
\
|
|

)
j
j
+
P
C
c ( )
2
2
N
S
E
u
0
( ) ,
c ( )
2
N
S 0
( )
[
\
|
|

)
j
j
c=0
N
i
N
S
where
A, B ( ) =
1
2
erfc
A
2

B
2
[
\
|

)
j
+
1
2
erfc
A
2
+
B
2
[
\
|

)
j erfc
A
2
[
\
|

)
j
The first term in (18) only depends on the signal to thermal
noise ratio at the receiver input, while the second one accounts
for MUI. The proposed approach was demonstrated to guarantee
high accuracy in estimating receiver performance for impulse-
based transmissions, even in the presence of scarcely populated
systems, systems with dominating interferers, or low-rate
systems [11].
5. PERFORMANCE ANALYSIS
The (UWB)
2
protocol described in Section 2 was tested by
simulation. Simulation results were averaged over L different
simulation runs. In each simulation run, N nodes were randomly
located inside a square region with area A. Next, a realization of
the channel impulse response, path loss, and delay spread was
generated for each pair of nodes, with characteristics depending on
the considered propagation scenario. These quantities were used
by the interference module for introducing errors on the received
packets, according to the MUI model described in Section 4.2.
We considered the scenarios CM1, CM2, CM5, and CM6 defined
within IEEE 802.15.4a, corresponding to indoor propagation in
residential environments in LOS and NLOS conditions, and
outdoor propagation in LOS and NLOS [10]. CM1, CM2,
CM5, and CM6 channels will be indicated as Scenarios 1, 2, 3,
and 4, respectively. The settings for the path loss at a reference
distance and path loss exponent in the four channel scenarios are
presented in Table I.
(19)
1 43.9 dB 1.79
2 48.7 dB 4.58
3 43.29 dB 1.76
4 43.29 dB 2.5
TABLE 1: CHANNEL SCENARIOS CHARACTERISTICS.
Scenario Path Loss @ d=1m Path Loss exponent
L 10
Number of nodes From 10 to 20
Area 10 m 10 m (Indoor)
50 m 50 m (Outdoor)
Network topology Random node positions
Channel model See eq. (2) and [10]
User bit rate R From 10 kb/s to 30 kb/s
Transmission rate 966 kb/s
Power 36.5 W (FCC limit for Bandwidth
0.5 GHz)
Packet traffic model Poisson generation process,
uniform distribution for destination node
DATA packet length 1224 bits (+ 64 bits for Sync trailer)
Interference Model Pulse Collision (see section 4)
Physical layer settings N
s
= 4, T
s
= 258.8 ns
T
m
= 2 ns, Reed Solomon (43,51) FEC
TABLE 2: SIMULATION SETTING.
Parameter Setting
During all simulations, the maximum size of 1288 bits
was adopted for the PHYPDU. This value was obtained by
considering as PHY payload a full size MACPDU of 129 bytes
coded with a Reed Solomon (43,51) Forward Error Correction
code in compliance with the specifications for the UWB PHY
of the future 802.15.4a standard [12]. The 1224 coded bits were
then combined with a PHY synchronization trailer of length 64
bits, leading to a size of 1288 bits for each PHYPDU.
Table II presents the main simulation settings.
(18)
,
,
Performance of (UWB)
2
was analyzed as a function of:
channel characteristics (indoor vs outdoor)
number of terminals
user bit rate
access strategy (pure vs. slotted).
138 COPYRIGHT UNIVERSIT DEGLI STUDI DI ROMA LA SAPIENZA, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
The comparison between pure and slotted Aloha was motivated by
the fact that, as well known, in narrowband networks, slotted Aloha
guarantees a higher (up to two times) throughput with respect to
pure Aloha, thanks to a lower probability of packet collision. Our
goal was to verify if this large performance gap is also present in
low bit rate UWB networks, where the negative impact of packet
collisions is mitigated by the high processing gain.
Fig. 1 presents the throughput for the Pure Aloha strategy as a
function of the number of nodes, for a user bit rate R = 10 kb/s.
Fig. 1 shows that in all cases throughput is greater than around
98% for both indoor and outdoor LOS scenarios and stays above
85%, even in NLOS conditions, where the higher path loss
and the larger channel delay spread have a stronger impact on
network performance.
Fig. 2 shows the delay measured in the same simulations, taking
into account both the DATA PDU transmission time, equal to
1.33 ms, and the additional delay introduced by retransmissions
following PDU collisions. Fig. 2 shows that for LOS scenarios
the delay experienced by DATA PDUs is close to a minimum
possible value, given by the DATA PDU transmission time.
NLOS scenarios lead to a larger delay that is, however, below
2.2 ms in all cases.
Figs.3 and 4 show throughput and delay in the case of Slotted
Aloha.
Fig.3 shows that Slotted Aloha leads to throughputs comparable
to Pure Aloha, with values above 96% in LOS conditions and
above 85% in the NLOS case.
FIGURE 2: DELAY AS A FUNCTION OF NUMBER OF NODES FOR A USER DATA
RATE R = 10 KB/S AND PURE ALOHA ACCESS (FILLED LINE WITH WHITE CIRCLES:
SCENARIO 1 (INDOOR LOS); DASHED LINE WITH FULL CIRCLES:
SCENARIO 2 (INDOOR NLOS); FILLED LINE WITH WHITE SQUARES:
SCENARIO 3 (OUTDOOR LOS); DASHED LINE WITH FULL SQUARES:
SCENARIO 4 (OUTDOOR NLOS).)
1.4
1.5
1.6
1.7
1.8
1.9
2.0
2.1
2.2
1.2
Number of nodes
D
e
l
a
y

(
s
)
15 20 10
x 10
-3
FIGURE 1: THROUGHPUT AS A FUNCTION OF NUMBER OF NODES FOR A USER
DATA RATE R = 10 KB/S AND PURE ALOHA ACCESS (FILLED LINE WITH WHITE
CIRCLES: SCENARIO 1 (INDOOR LOS); DASHED LINE WITH FULL CIRCLES:
SCENARIO 2 (INDOOR NLOS); FILLED LINE WITH WHITE SQUARES:
SCENARIO 3 (OUTDOOR LOS); DASHED LINE WITH FULL SQUARES:
SCENARIO 4 (OUTDOOR NLOS).)
1
0.80
0.82
0.84
0.86
0.88
0.90
0.92
0.94
0.96
0.98
Number of nodes
T
h
r
o
u
g
h
p
u
t
15 20 10
FIGURE 3: THROUGHPUT AS A FUNCTION OF NUMBER OF NODES FOR A USER
DATA RATE R = 10 KB/S AND SLOTTED ALOHA ACCESS (FILLED LINE WITH
WHITE CIRCLES: SCENARIO 1 (INDOOR LOS); DASHED LINE WITH FULL CIRCLES:
SCENARIO 2 (INDOOR NLOS); FILLED LINE WITH WHITE SQUARES:
SCENARIO 3 (OUTDOOR LOS); DASHED LINE WITH FULL SQUARES:
SCENARIO 4 (OUTDOOR NLOS).)
1
x 10
-3
0.98
0.96
0.94
0.92
0.90
0.88
0.86
Number of nodes
T
h
r
o
u
g
h
p
u
t
15 20 10
THE ALOHA ACCESS (UWB)
2
PROTOCOL REVISITED FOR IEEE 802.15.4A
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
139
On the other hand, the additional delay in PDU transmission
introduced by the slotted time axis leads to higher delays, in
agreement with [7], as shown in Fig. 4.
Note, however, that the increase in the delay as a function of the
number of users is slower than that in the delay for Pure Aloha,
indicating that, for higher numbers of users, the Slotted Aloha
approach should eventually guarantee lower delays and thus
better performance.
In a second set of simulations, we evaluated the impact of the
user data rate R on performance. We considered a network of
10 nodes and measured throughput and delay for three different
data rates: 10, 20, and 30 kb/s, respectively.
The throughput obtained in the case of the Pure Aloha approach
is presented in Fig. 5.
Fig. 5 shows that performance of the (UWB)
2
degrades
gracefully as the offered traffic increases: the throughput is
above 90% in all considered cases, and is well above 95% for
the LOS scenarios. This behavior is confirmed by data on delay,
as shown in Fig. 6. LOS scenarios have very low delays, and
delays for NLOS scenarios are below 1.75 ms in all cases.
We measured throughput and delay for Slotted Aloha, presented
in Figs 7 and 8, respectively.
Fig. 7 shows that Slotted Aloha provides slightly better results
compared to Pure Aloha for high offered traffic loads, thanks
to a lower probability of packet collision. The difference is
more pronounced for NLOS scenarios, where Slotted Aloha
FIGURE 4: DELAY AS A FUNCTION OF NUMBER OF NODES FOR A USER DATA
RATE R = 10 KB/S AND SLOTTED ALOHA ACCESS (FILLED LINE WITH WHITE
CIRCLES: SCENARIO 1 (INDOOR LOS); DASHED LINE WITH FULL CIRCLES:
SCENARIO 2 (INDOOR NLOS); FILLED LINE WITH WHITE SQUARES:
SCENARIO 3 (OUTDOOR LOS); DASHED LINE WITH FULL SQUARES:
SCENARIO 4 (OUTDOOR NLOS).)
2.1
2
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
3
Number of nodes
D
e
l
a
y

(
s
)
15 20 10
x 10
-3
FIGURE 5: THROUGHPUT AS A FUNCTION OF USER DATA RATE FOR A NETWORK
OF 10 NODES AND PURE ALOHA ACCESS (FILLED LINE WITH WHITE CIRCLES:
SCENARIO 1 (INDOOR LOS); DASHED LINE WITH FULL CIRCLES:
SCENARIO 2 (INDOOR NLOS); FILLED LINE WITH WHITE SQUARES:
SCENARIO 3 (OUTDOOR LOS); DASHED LINE WITH FULL SQUARES:
SCENARIO 4 (OUTDOOR NLOS).)
0.91
0.9
0.92
0.93
0.94
0.95
0.96
0.97
0.98
0.99
1
User data rate (kb/s)
T
h
r
o
u
g
h
p
u
t
20 30 10
x 10
-3
FIGURE 6: DELAY AS A FUNCTION OF USER DATA RATE FOR A NETWORK
OF 10 NODES AND PURE ALOHA ACCESS (FILLED LINE WITH WHITE CIRCLES:
SCENARIO 1 (INDOOR LOS); DASHED LINE WITH FULL CIRCLES:
SCENARIO 2 (INDOOR NLOS); FILLED LINE WITH WHITE SQUARES:
SCENARIO 3 (OUTDOOR LOS); DASHED LINE WITH FULL SQUARES:
SCENARIO 4 (OUTDOOR NLOS).)
1.35
1.4
1.45
1.5
1.55
1.6
1.65
1.7
1.75
1.3
User data rate (kb/s)
D
e
l
a
y

(
s
)
20 30 10
x 10
-3
140 COPYRIGHT UNIVERSIT DEGLI STUDI DI ROMA LA SAPIENZA, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
TABLE II. PERCENTAGE OF ERROR BETWEEN
THE MODEL PREDICTION AND THE ACTUAL
MEASURED ADDITIONS AND MULTIPLICATIONS
PER PIXEL. FOR EACH GOP OF EVERY
SEQUENCE, THE AVERAGE ERROR OVER A
NUMBER OF ADAPTATION POINTS IS PRESENTED.
The performance of the (UWB)
2
MAC was evaluated in
the presence of multipath-affected propagation channels,
derived from the channel model proposed within the IEEE
802.15.4a Task Group. Performance in both pure and
slotted modes of operation was analyzed by simulation in
indoor and outdoor scenarios in both Line-Of-Sight and
Non-Line-Of-Sight conditions. In order to properly take
into account the impact of multipath and channel delay
spread on network performance, we also introduced an ad-
hoc MUI model based on the concept of Pulse Collision in
this work.
Simulation results show that the (UWB)
2
MAC guarantees
satisfactory network performance in both indoor and outdoor
scenarios, even in presence of NLOS propagation conditions.
Furthermore, network performance in case of high traffic loads
can be improved by adopting a Slotted Aloha approach. Results
suggest that, despite its simplicity, the (UWB)
2
MAC provides
high throughput and low delays for bit rates up to several tens
of kb/s and for networks composed of tens of terminals, thereby
making it a viable solution for future UWB low data rate
networks.
guarantees in all cases a throughput higher than 94%, vs. 91%
obtained by the Pure Aloha in the worst case.
The results are confirmed by Fig. 8, showing that the increase
in delay as the offered traffic increases is proportionally lower
than in the case of Pure Aloha, as a consequence of the higher
robustness of the Slotted Aloha approach in high traffic scenarios.
6. CONCLUSIONS
In this work, the (UWB)
2
MAC protocol, originally introduced
in [6], was revisited in view of its application to the future IEEE
802.15.4a standard. The structure of both control and DATA
MACPDUs of the (UWB)
2
protocol was derived from the PDU
structure of the existing 802.15.4 MAC, thus guaranteeing full
support for the network topologies defined within the original
standard. The (UWB)
2
protocol adopts Aloha for medium
access and CDMA for multiple access, based on the use of Time
Hopping codes. The protocol can operate in either a slot-free
(pure) or a slotted fashion, and can thus fit both centralized and
distributed network architectures. The protocol also includes a
ranging procedure in order to enable the operation of location-
based protocols at higher layers.
FIGURE 7: THROUGHPUT AS A FUNCTION OF USER DATA RATE FOR A NETWORK
OF 10 NODES AND SLOTTED ALOHA ACCESS (FILLED LINE WITH WHITE CIRCLES:
SCENARIO 1 (INDOOR LOS); DASHED LINE WITH FULL CIRCLES:
SCENARIO 2 (INDOOR NLOS); FILLED LINE WITH WHITE SQUARES:
SCENARIO 3 (OUTDOOR LOS); DASHED LINE WITH FULL SQUARES:
SCENARIO 4 (OUTDOOR NLOS).)
FIGURE 8: DELAY AS A FUNCTION OF USER DATA RATE FOR A NETWORK
OF 10 NODES AND SLOTTED ALOHA ACCESS (FILLED LINE WITH WHITE CIRCLES:
SCENARIO 1 (INDOOR LOS); DASHED LINE WITH FULL CIRCLES:
SCENARIO 2 (INDOOR NLOS); FILLED LINE WITH WHITE SQUARES:
SCENARIO 3 (OUTDOOR LOS); DASHED LINE WITH FULL SQUARES:
SCENARIO 4 (OUTDOOR NLOS).)
0.91
0.9
0.92
0.93
0.94
0.95
0.96
0.97
0.98
0.99
1
User data rate (kb/s)
T
h
r
o
u
g
h
p
u
t
20 30 10
2.4
x 10
-3
2.35
2.3
2.25
2.2
2.15
2.1
2.05
User data rate (kb/s)
D
e
l
a
y

(
s
)
20 30 10
THE ALOHA ACCESS (UWB)
2
PROTOCOL REVISITED FOR IEEE 802.15.4A
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
141
ACKNOWLEDGMENTS
This work was partially supported by the European Union
within the Integrated Projects n. 506897 - PULSERS and n.
511766 LIAISON and by STMicroelectronics Srl (Italy) within
the research contract UWB Ranging and Positioning in Radio
Communication Systems.
REFERENCES
[1] The IEEE 802.15.4 standard, available
at http://www.ieee.org.
[2] P. Kinney, ZIGBEE TECHNOLOGY: WIRELESS CONTROL
THAT SIMPLY WORKS, (2003), Available
at http://www.zigbee.org/en/resources/#WhitePapers.
[3] IEEE 802.15.TG4a page,
http://www.ieee802.org/15/pub/TG4a.html.
[4] M.-G. Di Benedetto and Giancola G., Understanding
Ultra Wide Band Radio fundamentals,
Prentice Hall, 2004.
[5] L. De Nardis and G. M. Maggio, LOW DATA RATE UWB
NETWORKS, in Ultra Wideband Wireless
Communications, John Wiley & Sons, Inc. (May 2006).
[6] M.-G. Di Benedetto, L. De Nardis, M. Junk
and G. Giancola, (UWB)
2
: UNCOORDINATED, WIRELESS,
BASEBORN MEDIUM ACCESS FOR UWB COMMUNICATION
NETWORKS, MONET: Special Issue on WLAN
optimization at the MAC and network levels, vol. 5, no. 10,
pp. 663-674, October 2005.
[7] L. De Nardis, G. Giancola and M.-G. Di Benedetto,
PERFORMANCE ANALYSIS OF UNCOORDINATED MEDIUM
ACCESS CONTROL IN LOW DATA RATE UWB NETWORKS,
in Proceedings of the 1
st
IEEE/CreateNet International
Workshop on ULTRAWIDEBAND WIRELESS NETWORKING,
within the 2
nd
International Conference on Broadband
Networks, (invited paper), pp. 206-212, October 2005,
Boston, Massachusetts, USA.
[8] L. De Nardis and M.-G. Di Benedetto, MEDIUM ACCESS
CONTROL DESIGN FOR UWB COMMUNICATION SYSTEMS:
REVIEW AND TRENDS, KICS Journal of Communications
and Networks, vol. 5, no. 4, pp. 386393, December 2003.
[9] E. S. Sousa and J. A. Silvester, SPREADING CODE
PROTOCOLS FOR DISTRIBUTED SPREAD-SPECTRUM PACKET
RADIO NETWORKS, IEEE Transactions on Communications,
vol. COM36, no. 3, pp. 272281, March 1988.
[10] IEEE 802.15.4a Channel Model Final Report,
Rev.1 (November 2004), available
at: ftp://ieee:wireless@ftp.802wirelessworld.com/15/04/15-
04-0662-00-004a-channel-model-final-report-r1.pdf.
[11] G. Giancola and M.-G. Di Benedetto, A NOVEL APPROACH
FOR ESTIMATING MULTI USER INTERFERENCE IN IMPULSE
RADIO UWB NETWORKS: THE PULSE COLLISION MODEL,
(invited paper) EURASIP Signal Processing Journal,
Special Issue on Signal Processing in UWB
Communications, vol. 86, no. 5, pp. 2172-2184, May 2006.
[12] G. M. Maggio, 802.15.4A UWB-PHY, IEEE 802.15.4a
Document # IEEE 15-05-0707-01-004a, November 2005.
CONTACT: ST.JOURNAL@ST.COM
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
142 COPYRIGHT IEEE, 2006. REPRINTED, WITH PERMISSION, FROM: AN ULTRA-LOW ENERGY ASYNCHRONOUS PROCESSOR FOR WIRELESS SENSOR
NETWORKS, BY L. NECCHI, L. LAVAGNO, D. PANDINI, L. VANZAGO, PROCEEDINGS OF ASYNC 2006 - 12TH IEEE INTERNATIONAL SYMPOSIUM
ON ASYNCHRONOUS CIRCUITS AND SYSTEMS, MARCH 13-15, 2006, GRENOBLE, FRANCE.
L. Necchi (1), L. Lavagno (1)
D. Pandini (2), L. Vanzago (2)
(1) Politecnico di Torino
(2) STMicroelectronics
AN ULTRA-LOW ENERGY
ASYNCHRONOUS
PROCESSOR FOR WIRELESS
SENSOR NETWORKS
This paper describes the design flow
used for an asynchronous 8-bit processor
implementing the Atmel AVR instruction set
architecture. The goal is to show dramatic
reductions in power and energy
with respect to the synchronous case
by exploiting aggressive dynamic voltage
scaling, while achieving almost the same
area, delay, and power at the same
voltage as normal synchronous operation,
and while retaining a traditional design
flow. The processor was implemented
in a 130nm technology using
desynchronization, starting from an initial
design downloaded from OpenCores.org.
It consumes 14pJ per instruction to
deliver 170 MIPS at 1.2 V and 2.7 pJ per
instruction to deliver 48 MIPS at 0.54 V.
It thus dramatically improves the energy
consumed per instruction with respect
to previous results from the literature.
1. INTRODUCTION
Wireless sensor networks (WSNs) promise to offer a variety of
low-cost measurement and actuation services for a very broad
range of applications, from home and factory automation, to
health care, to environment and building monitoring. Most
WSN architectures rely on a multitude of small devices
spread over the covered area in order to provide services in
an unobtrusive, reliable, and economical manner. Hence, node
cost, due to fabrication, deployment, and maintenance, is a key
aspect of WSN proliferation. Reducing both power and energy
requirements per computation activity (e.g., sensing operation,
packet reception or transmission, routing action) permits the
manufacturer to reduce the size of the battery and increase its
lifetime, thus impacting both unit and deployment costs.
While a majority of current demo applications for WSNs
spend most of the node power operating the radio, it is
clear that requiring more and more services from WSNs in
the future will mean increased power consumption for the
digital portion of the node as well. Examples of these power-
hungry services include network security [16], in-network
aggregation [22], and distributed data processing [20].
While adoption of these services in real WSN deployments
is still very preliminary, they are generally recognized as
AN ULTRA-LOW ENERGY ASYNCHRONOUS PROCESSOR FOR WIRELESS SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
143
very significant in the future, as they will improve WSN
robustness and global network lifetime.

This paper describes an implementation of a typical WSN
processor, the 8-bit Atmel AVR, using asynchronous techniques.
The resulting processor is called YUPPIE, which stands for Yet
another Ultra-low Power asynchronous Processor for wIreless
sensor nEtworks. The key contribution of this exercise is to show
that it is indeed possible to simultaneously achieve:
design re-use, since the synthesizable VHDL model of
the original microcontroller was downloaded from the
OpenCores.org web site.
implementation with a standard design flow and library,
based on a 130 nm technology from STMicroelectronics.
extremely short design time, since the complete design and
simulation was completed in about four months of work by
a person with no previous exposure to asynchronous design
techniques and limited exposure to synchronous ones. Most
of the time was actually spent learning to use the tools from
the industrial partners design flow.
aggressive power management by dynamic voltage scaling,
which in an asynchronous context naturally implies frequency
scaling as well.
virtually zero wake-up time, when returning from sleep
mode, and frequency change recovery time. In synchronous
circuits, both of these require from hundreds to thousands of
clock cycles, due to the long settling time of Phase Locked
Loops (e.g., the Atmel AVR requires 2 ms at 4MHz).
minimization of energy consumption by working with a
voltage supply very close to the process threshold.
reduction of (estimated) electro-magnetic emission, with a
promise of beneficial effects for the analog part of the circuit
as well.
Our work uses the desynchronization approach [2] because
it dramatically eases the transition from a synchronous to an
asynchronous design style. This was a key requirement from the
industrial partner of the project, STMicroelectronics.
It is well known that asynchronous circuits reduce electro-
magnetic emission with respect to equivalent synchronous
ones [11], [15], because they reduce the power consumption
peaks in the vicinity of clock edges. Hence, they produce a flatter
power spectrum and exhibit smaller voltage supply drops.
It is also well known that asynchronous circuits can be reliably
operated at very low voltages when device characteristics
exhibit second and third order effects, that is, when synchronous
operation becomes very problematic [13]. This is, of course,
especially true of quasi-delay insensitive or delay-insensitive
circuits, since the datapath signals its completion reliably,
no matter how slowly gates and wires switch. However, our
experiments with a fabricated circuit [5] have shown that even
cheaper bundled-data circuits can be operated at very low
voltages. This is due to the fact that the delay characteristics
of different gates, operating at the same low voltage, vary in a
reasonably similar manner (see Fig. 7).
Asynchronous circuits also offer the possibility of dramatic
increases in performance in the presence of extreme process and
operating condition variability. This advantage, which was not
a primary goal of this exercise due to the specific application
domain, is permitted by the reduced margins that asynchronous
design requires as compared to those required by synchronous.
In the synchronous case, all design, manufacturing, and
environment uncertainties must be handled by adding margins
to the clock cycles. It is widely reported that these margins are
currently in excess of 100% for a typical ASIC design flow
and manufacturing process. This means that a device whose
operating frequency is not determined a priori but is instead
chosen at run time (depending on both fabrication and operating
environment effects) can work twice as fast as an equivalent
synchronous one.
We are thus confident that desynchronization can easily be
used to perform initial experiments in the domain of aggressive
power, energy, and EMI minimization. The hope is that more
144 COPYRIGHT IEEE, 2006.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
radical asynchronous design techniques, which of course offer
advantages along all these dimensions, will be embraced, once
the advantages of asynchronicity have been shown.
This paper describes in detail the design flow that has been
used and the various choices that have been made. While our
design case study covers one specific architecture, we believe
that the techniques that we propose can be applied to any
digital architecture for low-power low-energy battery-powered
applications.
The desynchronization flow has the following characteristics:
It accepts as input a synchronous design, described with a
VHDL or Verilog RTL specification.
It uses standard EDA tools to synthesize, map, place, and
route the design. Standard tools are also used for timing
analysis, equivalence checking, local clock generation,
extraction, scan insertion, automated test pattern generation,
and so on. In this experiment, we used Synopsys Design
Compiler and Cadence SOC Encounter for logic synthesis
and physical design respectively, and Synopsys Nanosim for
performance and power simulation.
It fully automates the de-synchronization step [5], which
removes the clock tree and replaces it with a network of
asynchronous controllers.
2. PREVIOUS WORK
2.1 Wireless Sensor Network platforms and applications
Over the past few years, Wireless Sensor Networks have gained
more and more attention as a basic technology enabling the
vision of Ambient Intelligence. Nowadays, the advances of
silicon processing technologies permit the miniaturization of
low-cost devices that, by integrating sensing, computational,
and communication capabilities, constitute the basic interface
with the physical world. One of the most important constraints
in Wireless Sensor Networks, mainly built with battery-powered
nodes, is low power consumption, which has been driving
research in the definition of both system and node architecture
components [9]. Several device platforms, diversified in terms
of storage, communication, and computational capabilities,
have been identified as necessary for most Wireless Sensor
Networks, corresponding to the layers of multi-tiered network
architectures [8]. Berkeley Motes (i.e., the MICA Family) are a
notable example of a general-sensing-class device with limited on-
board capabilities. These motes, especially the MICA2 version,
are extensively used by a huge number of research groups for
Wireless Sensor Network prototyping in several typical ambient
intelligence application scenarios. These include, for example,
environmental monitoring [12], structural health monitoring
[21], personal health monitoring [19], and tracking [18].
The most recently commercially developed version of the MICA
family, i.e., the MICAZ, replaces the CC1000 radio used in
MICA2 with a CC2420 radio that is compatible with the standard
IEEE 802.15.4. This standard defines the physical and Medium
Access Control layer specifications for low data rate wireless
connectivity. Adopted by the Zigbee Industrial Consortium, the
use of this standard is aimed at promoting the usage of low data
rate networks in WSN application scenarios such as industrial
and home automation.
Currently, much of the research on WSN design at the system
level is based on reducing the quantity of raw sensing data
transmitted by the nodes by increasing the amount of local
computation performed on each node in order to reduce the
overall data traffic on the network. This increases the computing
capability requirements for each node and, as a consequence,
increases the importance of reducing the power consumption of
its digital components.
Our design example is thus oriented towards the definition of
an application-independent architecture because it addresses a
general purpose CPU design, namely, a desynchronized version
of the same processor used in the MICA family motes. We also
claim that desynchronization can also be profitably used to
derive asynchronous implementations of dedicated hardware
AN ULTRA-LOW ENERGY ASYNCHRONOUS PROCESSOR FOR WIRELESS SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
145
unit for signal processing, encryption, and so on, in order to
further reduce peripheral power and energy consumption.
2.2 Desynchronization
De-synchronization modifies the standard design flow for
synchronous circuits [4] in three key steps:
1. Conversion of the flip-flop-based synchronous circuit into a
latch-based one (M and S latches in Fig. 1(b)).
D-flip-flops are conceptually composed of master-slave
latches. To perform de-synchronization, this internal structure
is explicitly revealed (see Fig. 1(b)) to decouple local clocks
for master and slave latches (in a D-flip-flop, both derived
from the same clock), so as to separate concerns between
setup and hold time satisfaction. This step is essential in
avoiding the low clock skew constraints that are inherent in
the flip-flop-based design style.
2. Generation of matched delays for the combinational logic
(denoted by rounded rectangles in Fig. 1(b)). Each matched
delay must be greater than or equal to the delay of the critical
path of the corresponding combinational block.
Each matched delay serves as a completion detector for the
corresponding combinational block.
3. Implementation of the local controllers.
Each controller must open the latch after the following
one has been closed (to avoid double latching or hold time
violations) and close it after the previous one has been opened
and the combinational logic has settled (to avoid setup time
violations). Protocols and controller implementations are
described more in detail in [2].
Fig. 2 depicts a synchronous netlist after the conversion into
latch-based design. The shadowed boxes represent latches,
whereas the white boxes represent combinational logic. Latches
must alternate their phases.
Those with a label 0 (1) at the clock input represent the even
(odd) latches. All latches are transparent when the control signal
is high (CLK=0 for even, and CLK=1 for odd). Data transfers
must always occur from even (master) to odd (slave) latches and
vice-versa. Usually, this latch-based scheme is implemented with
two non-overlapping phases generated from the same clock.
The correctness of the desynchronization approach, i.e., the
fact that the asynchronous circuit implements exactly the
same I/O functionality as its synchronous counterpart has been
proved in [2].
FF FF
CLK
CL FF CL
S M S CL M
Cs Cm
CLK
gen
CLK
gen
CLK
gen
CLK
gen
CLK
gen
CLK
gen
S CL M
(a)
(b)

FIGURE 1:
(A) SYNCHRONOUS CIRCUIT,
(B) DE-SYNCHRONIZED
CIRCUIT.
146 COPYRIGHT IEEE, 2006.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
controller inputs (see Fig. 3(b)), which are assumed to respond
with an acknowledge faster than the clock. In the last case, only
the EMI advantages of desynchronization can be achieved.
Otherwise, peripherals which do not require a precise clock (see
Fig. 3(a)) or which may be adapted to provide an acknowledge
output (such as RAMs, see (Fig. 3(c)) can be driven by the clock
output produced by the desynchronized controllers.
2.3. Dynamic voltage and frequency scaling
It is well-known that, even under normal operating conditions,
the delay of a CMOS circuit scales almost linearly with its
voltage supply, while its power scales quadratically. Thus, the
normalized energy-per-cycle or energy-per-operation efficiency
measure scales linearly with voltage supply. However, it is very
difficult to use this optimization opportunity to the extreme, by
operating very close to the threshold voltage (as we do in this
paper), because:
1. library cells are seldom characterized by the manufacturer
at such extreme operating conditions. Hence, the normal
ASIC design flow is unsuitable for guaranteeing correct
operation under the lowest voltage, energy, and power
operating conditions.
2. the gate delay models deviate significantly from those
used under nominal conditions and make a straightforward
scaling of performance and power impossible, or at least
very risky.
3. the effects of various random or hard-to-predict phenomena,
such as threshold voltage variations, wire width variations,
and local voltage supply variations due to IR drop
are significantly magnified.
All this means that, even if one were able to use the traditional
design flow for circuits to be operated at a voltage supply close
to the transistor threshold voltage, the performance margins that
one would have to use to ensure correct operation would be huge
(they already exceed 100% under nominal conditions).
Two approaches have been proposed in the literature to tackle
this problem with purely synchronous means. Both are based
Interfacing with synchronous external peripherals and memories
can be done in different manners, depending on their performance
and clock precision requirements. Peripherals that require precise
clocks (e.g., UARTs, timers) can be driven from an external
clock, and then a synchronization interface can be added, or the
external clock can be used directly to drive the desynchronized
CLK
G 1
1
D
0
E
1
B
1
F
0
C
0
A
FIGURE 2: A SYNCHRONOUS CIRCUIT WITH A SINGLE GLOBAL CLOCK.
0
L
1
L CL
0
L
1
L CL
0
L
C C C
1
L CL
Fast
LOGIC
0
L
1
L CL
0
L
1
L CL
0
L
C C C
1
L CL
Slow
LOGIC
Extemal
CLK
0
L
1
L CL
0
L
1
L CL
Data path
Handshaking line
Contrloller
0
L
C
a)
b)
c) C C
1
L CL
C
Self
Timed
LOGIC
FIGURE 3: SYNCHRONOUS EXTERNAL LOGIC INTERFACING STRATEGY: (A) FAST
LOGIC; (B) SLOW LOGIC OR PRECISE CLOCKS REQUIREMENTS; (C) SELF TIMED LOGIC.
AN ULTRA-LOW ENERGY ASYNCHRONOUS PROCESSOR FOR WIRELESS SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
147
on sampling the output of a signal which is forced to make a
transition very close to the clock cycle and to slow down the clock
frequency or increase the voltage supply if this critical sampling
happens at the current voltage and frequency conditions.

The Razor CPU [7] is designed with double slave latches and an
ex-or in each master-slave pair (thus increasing by over 100%
the area of each converted latch). The second slave is clocked
half a clock cycle later than the first slave. A difference in values
between the slaves, as detected by the comparator, shows that
the input changed very close to the falling edge of the clock of
the first slave, consequently memorizing an incorrect value. The
Razor in that case skips a beat and restarts the pipeline with the
value of the second latch, which, assuming that environmental
conditions change slowly, is always latched correctly. An
external controller always keeps voltage and clock frequency
very close to this critical clocking condition in order to operate
the processor very close to the best Vdd point for the required
clock frequency under the current temperature conditions.
The approach, while very appealing for processors, has an
inherent problem that makes it inapplicable to ASICs. Due to
the near-critical clocking, it is always possible that the first latch
goes meta-stable. In that case, the whole detector and the clock
controller may suffer from meta-stability problems. That case
is detected with analogue mechanisms, and the whole pipeline
is flushed and restarted. This is easy to do in a processor for
which flushing and restarting is already part of a modern micro-
architecture. It is, however, very difficult, if not impossible, to
achieve automatically for a generic logic circuit, such as those
found in modern ASICs.
Another technique that has been proposed to dynamically monitor
the delay of the logic at the current voltage value and adjust the
clock frequency accordingly is the PowerWise technology
from National Semiconductors. It basically samples, with a high
frequency clock, the output of a digital delay line that toggles
once per system clock cycle. This is used, more or less as in
Razor, to measure the delay of the line in the current environment
conditions (temperature, V dd, etc.). The scheme is safer than that
of Razor because it allows one to insert enough synchronizers
after the delay line to reduce the meta-stability danger.
However, it is an indirect measure and requires a complicated
(patented) logic to monitor and control the clock speed.
Desynchronization achieves similar goals with much simpler
logic because the delay line output is directly used to generate the
clock period. Moreover, several controllers can be interspersed
with the logic so that local conditions (e.g., temperature, V dd,
transistor, and wire parameter variations) can be controlled much
more closely, thus requiring potentially smaller margins.
Moreover, desynchronization reduces electro-magnetic emission
(EMI) problems because the clock for every cluster of latches
driven by a single controller is out of phase with respect to the
others. This spreads the power spectrum, reducing both irradiated
and on-chip power noise. It also can improve performance
because it computes on-line the useful skew for each individual
group of latches. The performance of the system is thus not
given by the longest pipeline stage, as in the synchronous
case, but by the maximum cycle mean
1
. This means that even
unbalanced pipelines are automatically balanced at runtime by
the handshaking mechanism.
2.4. Low-power asynchronous processors
Several asynchronous micro-processors reported in the literature
exhibit very favorable power, energy, and EMI characteristics
with respect to their synchronous counterparts. For example,
the Lutonium from CalTech [14] implements the 8051 ISA and
reportedly achieves 500 pJ per instruction at 1.8 V and MIPS, and
140 pJ per instruction at 0,9 V and 66 MIPS, with 180 nm a process.
According to [3], scaling CMOS technology to the next
generation improves performance and reduces power
1. The maximum cycle mean is defined as the maximum overall cycles in the logic, which
may include one or more latches, of the total cycle delay divided by the number of latches
in the cycle.
148 COPYRIGHT IEEE, 2006.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
consumption, thus reducing overall energy consumption by
about 65% per generation. Based on this analysis, the Lutonium
should achieve 175 pJ per instruction at full voltage supply and
49 pJ per instruction at minimum voltage supply, using a 130 nm
process like our design.
The CalTech processors are based on Quasi-Delay-Insensitive
(QDI) circuits, and are thereby extremely robust with respect to all
sorts of variations, including data-dependent computation delays.
Desynchronization uses a bundled-data completion detection
mechanism, which results in smaller circuitry but requires
margins to ensure correct operation. As a result, it cannot exploit
as effectively the inherent characteristics of asynchronicity (low
power, modularity, etc.), but it is much simpler to implement,
due to its use of a fairly standard design flow.
The Philips 8051 is another 8-bit micro-controller [17], which
was synthesized starting from a Tangram specification. Unlike
the Lutonium, the Philips 8051 used a bundled-data datapath,
thus resulting in (presumably) lower area overheads. The main
advantages over its synchronous counterpart, which led to its
inclusion in some commercial products, were the low power
consumption and, more significantly, the low EMI, which
dramatically reduced the cost of implementing the analogue RF
portion of a pager. A second-generation implementation of the
Philips 8051, commercialized by Handshake Solutions using
a 180 nm process, consumes 150 pJ per instruction delivering
6.3 MIPS at 1.8 V. If implemented with a 130 nm process, this
micro-controller should achieve the energy consumption of 53
pJ per instruction.
The Tangram flow is very similar in spirit to the desynchronization
flow because it uses mostly standard tools for physical design,
design rule checking, and so on. However, it requires one to use
a new proprietary language based on Communicating Sequential
Processes, for specification and high-level simulation. Moreover,
synthesis and design for testability tools are Tangram-specific.
This is both an advantage, in that it improves the optimality
of the result, and a disadvantage, because it requires extensive
designer training and a partially new set of tools. In comparison,
desynchronization is much easier for an experienced synchronous
designer to pick up and use.
The SNAP/LE processor [6], [10] is an asynchronous low-power
16-bit processor that was designed specifically for Wireless
Sensor Network applications. Its Instruction Set Architecture
was optimized specifically for WSN applications because it
operates in event-driven mode: basically, all code is executed as
an interrupt service routine. It also features a set of instructions
that directly control the various peripherals (timer, message
handler) as co-processors rather than by using the normal
memory-mapped register interface.
SNAP/LE, implemented in a process like the Lutonium, requires
218 pJ per instruction to deliver 240 MIPS at 1.8 V and 23 pJ
per instruction to deliver 28 MIPS at 0.6 V. It was implemented,
like the Lutonium, using the QDI implementation style, which
is a significant departure from traditional synchronous design.
If implemented with a process, this micro-controller should
achieve the energy consumption of per instruction at full voltage
supply and per instruction at minimum voltage supply.
Our processor, to be described more in detail in Section 4, was
implemented in a 130 nm process (one generation after the other
processors described above). It requires 14 pJ per instruction to
deliver 170 MIPS at 1.2 V and 2.7 pJ per instruction to deliver
48 MIPS at 0.54 V. These results do not include program and
data RAM access power. We estimated, from the data sheets of
memory components for the process that we used, that using 4KB
of RAM for code and data, as could be required by a small WSN
application, would increase energy consumption by about 20%.
3. THE ASYNCHRONOUS
AVR PROCESSOR
3.1. The synchronous RTL design
The project started from an open-source implementation of
an 8-bit microcontroller that is compatible with the Atmel
AVR architecture, the AVR_CORE available from the
AN ULTRA-LOW ENERGY ASYNCHRONOUS PROCESSOR FOR WIRELESS SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
149
OpenCores.org web site. We chose this processor because
its Instruction Set Architecture is compatible with that of most
WSN nodes of the MICA family.
Our aim was to demonstrate the possibility of asynchronously
redesigning an existing standard CPU core for a WSN node in
a very short time at reasonable cost. As mentioned above, we
wanted to show reductions in terms of power consumption and
noise emission as well as potential increases in performance, while
keeping compatibility with the AVR instruction set. This permits
the direct reuse of the binary software developed for this ISA.
The VHDL design from which we started provides a full
implementation of the instruction set of the Atmel AVR
microcontroller and includes various extra component, such as
data memory, program memory, timers, and UART.
The design also includes a fully featured VHDL testbench that
allows one to test the behavior of the entire platform. This was
very important since it dramatically sped up all phases of design,
from debugging the implementation of the flow all the way to
measuring performance and power by simulation. (We used Mentor
Graphics Modelsim and Synopsys Nanosim for this purpose).
3.2. Logic synthesis
After making sure via simulation that the downloaded AVR_
COR indeed worked as advertised, we took the first step: the
logic synthesis of the initial synchronous design.
For this step, we used Synopsys Design Compiler. A short script
was written in order to:
load the VHDL core file;
define performance and area constraints;
synthesize the netlist;
map it on the chosen industrial standard cell library;
modify the project hierarchy to suit the needs of the
desynchronization step;
estimate critical path delay; and
save the resulting netlist on a Verilog file.
3.3. De-synchronization
The de-synchronization methodology [2] has been implemented,
as described in [1], in the standard design flow of our industrial
partner. The desync tool converts a technology-mapped
synchronous netlist into an asynchronous one, performing the
following steps:
loading and parsing the Verilog input file;
replacing registers with pairs of latches;
automatically grouping the combinational logic
into latch-separated islands;
estimating, using Synopsys PrimeTime, the logic delay
of each group;
implementing the delay chains and the controllers;
saving the de-synchronized output file in Verilog format.
In this case, we chose to use only one logic group (i.e., one
controller for the master and one for the slave latches) because
AVR_CORE only has 4500 gates organized as a 2-level pipeline
(fetch and execute). This obviously prunes some of the de-
synchronization advantages, principally, the electro-magnetic
emission reduction.
The total area of the desynchronized core is about 5% larger than
that of the synchronous one. Our results do not include the effect
of placement and routing, which are unlikely to be significant for
such a small design.
3.4. Logic simulations of the asynchronous core
The de-synchronized version of AVR_CORE was then simulated,
again using Modelsim, in order to verify the correct behavior
of the core. This required some minor modifications to the
simulation settings and to the environment model. For example,
we imported in Modelsim a new Standard Delay Format (SDF)
file that contains the detailed gate-level delay information and
library gate models.
Furthermore, we modified the VHDL testbench to properly
connect the handshake lines added by the de-synchronization
150 COPYRIGHT IEEE, 2006.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
tool instead of the externally generated clock. We connected
each peripheral to the falling edge of the Request Out
handshake signal of the desynchronized partition (we will call
this signal pseudo-clock). After this, logical simulation was
executed correctly.
3.5. Power estimations with Nanosim
Although Synopsys Design Compiler can perform power
estimations for both synchronous and asynchronous circuits,
using switching activity files (SAIF) derived from a simulator
such as Modelsim, does not allow us to use it beyond the
characterization corners of the standard cell library. Since we
wanted to verify how power changes with very low supply
voltages, something not usually considered among these corner
cases, we had to perform transistor-level simulations.
For this purpose, we used Nanosim, an advanced transistor-
level circuit simulation and analysis tool for analog, digital,
and mixed-signal design verification. Nanosim uses the same
electrical model parameters that simulators such as Spice use,
but it has more simplified transistor and wire models that permit
much faster simulation with a minimal loss of accuracy.
The inputs of the synchronous and desynchronized circuit were
driven using test vectors generated by Modelsim, and the output
vectors produced by both simulators under nominal conditions
were compared against each other to check that no errors had
occurred during the various synthesis, optimization, and translation
steps. The next section describes the results of such simulations.
4. SIMULATION RESULTS
AND ANALYSIS
The simulation results that we obtained indeed show the
expected behavior when lowering the voltage supply of both
the synchronous and the asynchronous circuit. The delay of each
circuit decreases with increasing Vdd as shown in Fig. 4, while
the power consumption increases with increasing Vdd, as shown
in Fig. 5. The energy per instruction, shown in 6, increases
almost linearly with the voltage supply. At the best power supply
level, around 0.5 V, the desynchronized circuit is about 5 times
more energy efficient than the synchronous one. Furthermore,
the power consumption is reduced by more than one order of
magnitude by scaling the voltage from the nominal level, 1.2 V,
to about 0.5 V.
Of course, one could also scale the voltage supply for the
synchronous circuit as well and obtain similar power reductions
by using either the Razor or the PowerWise approaches mentioned
above. However, this is much more difficult to do, as discussed
in Section 2.3, because it amounts to essentially measuring the
D
e
l
a
y

C
h
a
i
n

P
e
r
i
o
d

(
n
s
)

0
5
10
15
20
25
30
Voltage Supply (V)
1,4 1,3 1,2 1,1 1,0 0,9 0,8 0,7 0,6 0,5 0,4
Delay Chain Period
Synchronous clock (280MHz)
FIGURE 4: DESYNCHRONIZED PROCESSOR CYCLE PERIOD VERSUS SUPPLY VOLTAGE.
P
o
w
e
r

C
o
n
s
u
m
p
t
i
o
n

(
m
W
)

0
0.5
1
1.5
2
2.5
3
3.5
4
Voltage Supply (V)
1,4 1,3 1,2 1,1 1,0 0,9 0,8 0,7 0,6 0,5 0,4
Asynchronous Core
Synchronous Core (280MHz)
FIGURE 5: DESYNCHRONIZED PROCESSOR POWER VERSUS SUPPLY VOLTAGE.
AN ULTRA-LOW ENERGY ASYNCHRONOUS PROCESSOR FOR WIRELESS SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
151
access energy, which would add approximately 20% energy
consumption for 4 KB of RAM, the results are quite comforting
with respect to previous work in both the synchronous and the
asynchronous domain.
We can attribute them to the extremely low area and performance
overhead that desynchronization imposes over synchronicity.
Hence, the margins that are required due to the bundled-data
timing of a delay line at a given supply voltage and then tuning
the clock to match the circuit performance. Moreover, Razor
would require one to change the AVR architecture drastically
since it does not support pipeline squashing and restarting.
The desynchronized circuit offers the same advantages for
free. Moreover, it allows one to spread delay lines wherever
they are needed to measure the effects of spatially correlated
effects, for example, temperature and wire thickness
2
. It is also
not subject to the meta-stability problems that are inherent in the
near-critical sampling of a changing signal.
Fig. 7 shows the timing of the matched delay lines and of the
actual critical path of the combinational logic, when varying
Vdd. Remember that, in order to ensure correct operation of
a bundled-data circuit, the delay lines must always be slower
than the longest combinational path. The figure shows that
the matched delays track the actual logic delay very well,
and suggests that true measurement of logic delays, by dual-
rail encoding, e.g., may not be necessary after all, at least for
the technology under consideration, which is 130 nm. For
this example, we did not try to fine-tune the delay line since
performance was not the primary goal of the exercise, but we did
want to keep the design fairly robust. Hence, we chose a margin
of about 100%, which is close to what synchronous design would
require under nominal conditions which is much better than what
the same would require at low voltage supplies.
5. CONCLUSIONS
Our processor, which is ISA compatible with the Atmel AVR,
requires 14 pJ per instruction to deliver 170 MIPS at 1.2 V and
2.7 pJ per instruction to deliver 48 MIPS at 0.54 V. These results
seem to be much better than those of the other asynchronous
processors reported in Section 2.4 (which were implemented
with a 180 nm process), even when those are scaled to 130
nm. This means that, even after considering the improvement
in technology over one generation, which reduces according
to [3] energy consumption by 65%, and after considering RAM
2. Wire thickness, which affects both resistance and side capacitance, is spatially correla-
ted, due to the effects of the chemical-mechanical polishing step.
0
2
4
6
8
10
12
14
16
18
0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4
E
n
e
r
g
y

p
e
r

I
n
s
t
r
u
c
t
i
o
n

[
p
J
/
i
n
]
Voltage Supply (V)
FIGURE 6: DESYNCHRONIZED PROCESSOR ENERGY PER INSTRUCTION.
D
e
l
a
y

c
h
a
i
n

V
s

C
r
i
t
i
c
a
l

P
a
t
h

(
n
s
)

0
5
10
15
20
25
30
Voltage Supply (V)
1,4 1,3 1,2 1,1 1,0 0,9 0,8 0,7 0,6 0,5 0,4
Delay Chain Delay
Critical Path Delay
FIGURE 7: CRITICAL PATH VERSUS MATCHED DELAYS.
152 COPYRIGHT IEEE, 2006.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
approach (about 2X, as discussed in Section 4) are still much
less than the overhead due to dual-rail implementation of QDI
circuits and to the extremely low granularity controllers required
by the Tangram design flow.
Based on the results of this work, we can claim that the only
inherently robust manner of achieving extreme Dynamic Voltage
and Frequency Scaling is by using asynchronous techniques.
Such techniques measure reliably, rather than either estimate
or measure with inherent meta-stability risks, the actual delay
of combinational logic, thus taking into account manufacturing
process variations, environmental conditions, and so on.
The desynchronization technique is not perfectly and natively
asynchronous because all blocks still operate in lockstep
with respect to each other. However, it provides the means to
measure logic delays by using matched delay lines. Its very
easy adoption, which was also shown in this project by re-
implementing an existing legacy RTL core without any re-design
and almost without any knowledge of asynchronous design
techniques, makes it an appealing first step in the direction
of more widespread adoption by industry. We believe that
Wireless Sensor Networks provide a very promising application
domain for very low-power and low-energy digital electronics,
due to the increasing computational requirements and the
stringent constraints on battery life. In the future, we will
explore in more detail how the low-energy and (hopefully)
low-EMI characteristics of our processor can be exploited
in actual WSN applications. We will also verify how well delay
lines track the critical path delay in the presence of process
variability.
ACKNOWLEDGMENTS
The authors would like to thank Nikos Andrikos and Christos
Sotiriou from ICS-FORTH for the extensive support they
provided us while using their desynchronization tool and
Maurizio Tranchero from Politecnico di Torino for his help
during the first phase of this project.
REFERENCES
[1] N. Andrikos. Personal communication.
[2] I. Blunno, J. Cortadella, A. Kondratyev, L. Lavagno,
K. Lwin, and C. Sotiriou, HANDSHAKE PROTOCOLS FOR
DE-SYNCHRONIZATION, Proc. International Symposium on
Advanced Research in Asynchronous Circuits and Systems,
pages 149-158, 2004.
[3] Shekhar Borkar, DESIGN CHALLENGES OF TECHNOLOGY
SCALING, IEEE Micro, July-August 1999.
[4] D. Chinnery and K. Keutzer, editors, Closing the
Gap between ASIC and Custom: Tools and Techniques
for High-Performance ASIC design, Kluwer Academic
Publishers, 2002.
[5] J. Cortadella, A. Kondratyev, L. Lavagno, and C. Sotiriou,
DESYNCHRONIZATION: SYNTHESIS OF ASYNCHRONOUS
CIRCUITS FROM SYNCHRONOUS SPECIFICATIONS,
IEEE Transactions on CAD, 2005.
[6] V. Ekanayake, C. Kelly, and R. Manohar, AN ULTRA LOW-
POWER PROCESSOR FOR SENSOR NETWORKS. In Proceedings
of the international conference on Architectural
support for programming languages and operating
systems, (ASPLOS), 2004.
[7] D. Ernst, S. Das, S. Lee, D. Blaauw, T. Austin, T. Mudge,
N. S. Kim and K. Flautner, RAZOR: CIRCUIT-LEVEL
CORRECTION OF TIMING ERRORS FOR LOW-POWER OPERATION,
IEEE Micro, November 2004.
[8] J. Hill, M. Horton, R. Kling, and L. Krishnamurthy,
THE PLATFORMS ENABLING WIRELESS SENSOR NETWORKS,
Communications of the ACM, 47(6), 2004.
[9] J. Hill, R. Szewczyk, A. Woo, S. Hollar, D. Culler,
and K. Pister, SYSTEM ARCHITECTURE DIRECTIONS
FOR NETWORKED SENSORS, Architectural Support
TABLE II. PERCENTAGE OF ERROR BETWEEN
THE MODEL PREDICTION AND THE ACTUAL
MEASURED ADDITIONS AND MULTIPLICATIONS
PER PIXEL. FOR EACH GOP OF EVERY
SEQUENCE, THE AVERAGE ERROR OVER A
NUMBER OF ADAPTATION POINTS IS PRESENTED.
AN ULTRA-LOW ENERGY ASYNCHRONOUS PROCESSOR FOR WIRELESS SENSOR NETWORKS
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
153
for Programming Languages and Operating Systems,
pages 93-104, 2000.
[10] Clinton Kelly, Virantha Ekanayake, and Rajit Manohar,
SNAP: A SENSOR NETWORK ASYNCHRONOUS PROCESSOR,
Proc. International Symposium on Advanced Research
in Asynchronous Circuits and Systems, pages 24-33,
IEEE Computer Society Press, May 2003.
[11] Joep Kessels and Ad Peeters, THE TANGRAM FRAMEWORK:
ASYNCHRONOUS CIRCUITS FOR LOW POWER, Proc. of Asia
and South Pacific Design Automation Conference,
pages 255-260, February 2001.
[12] A. Mainwaring, J. Polastre, R. Szewczyk, D. Culler,
and J. Anderson, WIRELESS SENSOR NETWORKS
FOR HABITAT MONITORING, ACM International Workshop
on Wireless Sensor Networks and Applications (WSNA02),
September 2002.
[13] Alain J. Martin, Andrew Lines, Rajit Manohar, Mika
Nystrm, Paul Pnzes, Robert Southworth, and Uri
Cummings, THE DESIGN OF AN ASYNCHRONOUS MIPS
R3000 MICROPROCESSOR, Advanced Research in VLSI,
pages 164-181, September 1997.
[14] Alain J. Martin, Mika Nystrm, Karl Papadantonakis, Paul
I. Pnzes, Piyush Prakash, Catherine G. Wong, Jonathan
Chang, Kevin S. Ko, Benjamin Lee, Elaine Ou, James
Pugh, Eino-Ville Talvala, James T. Tong, and Ahmet Tura,
THE LUTONIUM: A SUB-NANOJOULE ASYNCHRONOUS 8051
MICROCONTROLLER, Proc. International Symposium on
Advanced Research in Asynchronous Circuits and Systems,
pages 14-23, IEEE Computer Society Press, May 2003.
[15] J. McCardle and D. Chester, MEASURING AN
ASYNCHRONOUS PROCESSORS POWER AND NOISE,
Synopsys User Group (SNUG), 2001.
[16] A. Perrig, J. Stankovic, and D. Wagner,
SECURITY IN WIRELESS SENSOR NETWORKS,
Communications of the ACM, 47(6), June 2004.
[17] Philips Semiconductors, P87CL888; 80C51 ULTRA LOW
POWER (ULP) TELEPHONY CONTROLLER
[18] C. Sharp, S. Schaffert, A. Woo, N. Sastry, C. Karlof,
S. Sastry, C. Karlof, S. Sastry, and D.Culler,
DESIGN AND IMPLEMENTATION OF A SENSOR NETWORK
SYSTEM FOR VEHICLE TRACKING AND AUTONOMOUS
INTERCEPTION, In European Workshop on Wireless
Sensor Networks, January 2005.
[19] V. Shnayder, B r. Chen, K. Lorincz, T. Fulford-Jones,
and M. Welsh, SENSOR NETWORKS FOR MEDICAL CARE,
Technical Report TR-08-05, Division of Engineering
and Applied Sciences, Harvard University, 2005.
[20] H. Wang, D. Estrin, and L. Girod, PREPROCESSING
IN A TIERED SENSOR NETWORK FOR HABITAT MONITORING,
EURASIP JASP, special issue of sensor networks, 2003.
[21] N. Xu, S. Rangwala, K.K. Chintalapudi, D. Ganesan,
A. Broad, R. Govindan, and D.Estrin, A WIRELESS SENSOR
NETWORK FOR STRUCTURAL MONITORING, Proceedings
of the international conference on Embedded networked
sensor systems, 2004.
[22] J. Zhao, R. Govindan, and D. Estrin,
COMPUTING AGGREGATES FOR MONITORING
WIRELESS SENSOR NETWORKS In International Workshop
on Sensor Network Protocols and Applications, May 2003.
CONTACT: ST.JOURNAL@ST.COM
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
154 COPYRIGHT STMICROELECTRONICS, POLITECNICO DI MILANO, 2007.
Guido Bertoni
(1)
,
Luca Breveglieri
(2)
, Matteo Venturi
(2)


(1) STMicroelectronics
(2) Politecnico di Milano
IMPLEMENTING ECC
FOR 8-BIT SYSTEMS
AND POWER CONSUMPTION
CONSIDERATIONS
Public key cryptography is widely
considered as the fundamental building
block for key exchange; different public
key algorithms are standardized and used
in many applications. Among them,
ECC (Elliptic Curve Cryptography)
is considered the best solution in terms
of security, computational requirements,
and storage needs for secret and public
keys. Energy consumption is one of the
main constraints to be considered in
wireless sensor networks. In the case
of sensor networks, the typical
approaches of minimizing latency
via a complete hardware coprocessor
and of reducing area overhead via
an efficient implementation of finite field
operations might not provide the best
solution. In this paper, two coprocessors
for minimizing both additional resources
and power consumption are presented.
Costs and performances of such
coprocessors are compared with
known results, showing that space
1. INTRODUCTION
The implementation of security services, especially those
related to secure data exchange, is becoming an indispensable
requirement for sensor network applications [1]. In high-end
systems, the cryptographic applications have already undergone
intensive development, while the security services implemented
in low-power low-cost systems are still in a primitive state. This
is a strong limitation which could prevent sensor networks from
covering a wider range of applications. A pervasive diffusion
of this technology will occur only if security requirements
are efficiently matched by the entire system. One of the major
barriers to deploying security on sensor networks is the limited
computation capabilities, particularly critical for public key
primitives.
Future technological advances will probably push the networks
towards lower cost and energy consumption rather than to higher
performances in computation [2]. Thus, it is essential to find a
smart solution to ensuring satisfactory security services which
will minimally impact performance and energy consumption
exists for the reduction of energy consumption
without degrading the other performance
figures.
IMPLEMENTING ECC FOR 8-BIT SYSTEMS AND POWER CONSUMPTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
155
while exploiting the few available system resources as much as
possible.
Public key cryptography is a building block that allows
secure communication in dynamic environments such as
sensor networks without the need of a permanently on-line
central authority [3]. The most efficient schemes that provide
cryptographic protocols are the Elliptic Curve Cryptosystems
(ECC) [4][5]. The most important operations performed by
those systems is the scalar multiplication, or kP. The execution
of one kP, for example, is necessary to compute a decryption
transformation, or a digital signature generation. The calculation
of two kP allows an encryption transformation, a signature
verification, or a key exchange. The security protocols based
on ECC are standardized by different institutions such as IEEE,
ANSI, NIST, and ISO [4].
Recent publications have evaluated the latency of software
implementations of ECC and the speed-up obtained with
dedicated instructions, or with hardware multipliers, but none
of them focused on the energy spent in ECC computations,
where traditional metrics have been used, as latency and silicon
area demand. Most of the previous scientific publications have
explored the field of high performance, where the only constraint
is the computational time.
We believe that, in applications such as sensor networks,
the latency of public-key operations is relatively important
compared to energy consumption because most sensor networks
have a limited bandwidth and relatively low communication
latency. The aim of this work is to evaluate the energetic cost
of different ECC implementations for low-end systems and to
propose a new hardware coprocessor architecture which not
only speeds up cryptographic operations but minimizes energy
consumption over all.
The paper is organized as follows: section 2 presents the main
topics of public key cryptography; section 3 briefly introduces
the ECC mathematical background; section 4 presents the
previous work in this field; section 5 evaluates the energy cost
of cryptographic operations; and section 6 presents the new
coprocessors and its performances. The conclusions of the work
can be found in section 7.
2. PUBLIC KEY CRYPTOGRAPHY
Cryptography offers techniques to protect communications
and to provide security services. These techniques consist of
applying an encryption scheme to transform data.
Encryption primitives are known in mathematics as one-way
functions with a trapdoor. The idea is that once the encryption
has transformed the information, it is computationally unfeasible
to recover the original content without the trapdoor information.
This piece of information, in cryptographic systems, is called a
key. The use of keys in cryptographic protocols is very useful.
When the security of the scheme relies on the secrecy of the
encryption and decryption functions, the entire scheme has to
be redesigned if these functions are revealed. On the contrary,
when security relies only on a secret key and the transformations
are publicly known, a simple key change can secure the whole
system once again.
When two entities want to communicate in a secure way, one
possibility is that they agree upon a unique secret key k and use
k to encrypt messages. The systems following this idea are called
symmetric-key cryptosystems, and the Advanced Encryption
Standard (AES) is an example thereof. The use of a common
secret key allows fast cryptographic operations, but the problems
of key agreement and management turn out to be unsolvable in
large and dynamic networks and in remote communications.
The problem of key management and agreement is solved by
the so called public-key systems (proposed for the first time in
[3]) at the expense of more complex operations. In public-key
schemes, each entity owns a pair of keys: a public key k
p
and a
secret key k
s
. The key k
p
is publicly known, and it can be used
by other entities when they want to send an encrypted message to
156 COPYRIGHT STMICROELECTRONICS, POLITECNICO DI MILANO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
its owner. The only possibility to recover the original message is
by using k
s
, only known by the legitimate owner of k
p
.
For obtaining the best performances from the two cryptosystems,
a hybrid approach is generally adopted. A session key is
exchanged using public-key cryptography and is then used in a
symmetric-key cryptosystem for encrypting data.
Public key cryptography is commonly used in Internet security
protocols such as SSL or IPSec, smartcard applications, and
Digital Right Managements. The dynamic composition of a
sensor network can benefit from the characteristics of public-
key cryptography.
A typical use of a public key algorithm is the so called Diffie-
Hellman key exchange. The protocol is a sequence of three steps:
private/public key generation, public key exchange, and shared
secret derivation. The result of the protocol is a function of the
parameters and of the two private keys, but no secret has been
exchanged over the insecure channel. A detailed explanation of
the DH is given in the next section, using ECC as the public key
primitives.
3. ELLIPTIC CURVE
CRYPTOSYSTEMS
The theoretical security of a cryptographic system is strictly
related to the hardness of the mathematical problem defining the
system itself. There are three mathematical problems exploited in
public key cryptography applications: the Integer Factorization
Problem (IFP), the Discrete Logarithm Problem (DLP), and the
Elliptic Curve Discrete Logarithm Problem (ECDLP). While
sub-exponential algorithms are known to solve the IFP and the
DLP, only fully-exponential algorithms have been proposed to
solve the ECDLP [4]. This means that ECDLP-based systems
can provide the same level of security as DLP- or IFP-based
systems, using smaller key sizes. As the computing time of the
basic operations is directly proportional to the operand size, the
implementation of ECDLP schemes represents a great advantage,
especially when calculations are executed by low-end processors.
Elliptic curves (EC) are defined by the Weirestrass equation:
E : y
2
+ a
1
x y + a
3
y = x
3
+ a
2
x
2
+ a
4
x + a
6.
One point P(x
P
; y
P
) belongs to the curve E if its coordinates
satisfy equation (1). In cryptographic applications, the coordinates
x
P
and y
P
are elements of a finite field GF(q), where q is the
number of field elements. Curve points can be represented as
well in projective coordinates P(X
P
: Y
P
: Z
P
). The curve point
operations are expressed in terms of field operations that involve
point coordinates. An exhaustive presentation of the EC theory
can be found in [4].
The most important operation in Elliptic Curve Cryptosystems
(ECC) is the scalar multiplication, briefly referred to as kP. This
operation is defined as:
Q = kP = P + P + + P (k times).
Solving the ECDLP means to find the integer k such that Q =
kP, knowing the starting point P, the resulting point Q, and the
parameters of the curve. The execution time of EC cryptographic
schemes is dominated by scalar multiplication. In this work,
scalar multiplication is the key issue leading comparisons
between different implementation strategies. Since scalar
multiplication is difficult to be inverted by an attacker, it is easier
to see that k is the secret key while Q is the public key.
The most efficient algorithm for calculating scalar multiplication
without precomputation but with a minimum use of inversion
in the finite field is the modification of the Montgomery
Scalar Multiplication proposed by Lopez and Dahab [12]. This
method is based on projective coordinates, executes only one
inversion during a kP, and requires the use of two points in
projective coordinates as temporary variables. In the case of
small architectures, the advantages provided by this algorithm
are remarkable. This explains why the Montgomery algorithm is
(2)
(1)
IMPLEMENTING ECC FOR 8-BIT SYSTEMS AND POWER CONSUMPTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
157
usually implemented on low-power devices [5][6][7] and why it
is used in our implementations as well.
In this paper, we use a binary finite field, GF(2
163
), for
the definition of the elliptic curve. We considered binary
fields because they have some arithmetical properties, such
as the absence of carry propagation, which provides good
implementation advantages. We selected a field size of 163
bits because it is widely considered the appropriate key size for
todays security requirements [10].
ECC can be modeled using architecture with 4 different layers,
an architecture which reflects the conceptual hierarchy of
implemented operations. The first layer is represented by the
hardware instructions available on the target device; the second
level defines the operations over binary fields GF(2
m
) (addition,
multiplication, etc.); the third level implements the operations
on curve points (P+Q, 2P, kP); and the fourth and upper level
defines the cryptographic protocols. The security protocols that
can be implemented (D-H, ElGamal, ECDSA, etc.) provide key
exchange, encryption/decryption, and digital signature services.
Diffie Hellman is one of the most used public key protocols; it
allows sharing a secret key through an insecure channel in order
to exchange data using symmetric algorithms, which are faster
and easier to be executed in software.
Lets consider two devices, A and B. Every device selects a
random number smaller than the group order. This is the secret
key, which we represent as k
a
and k
b
for the two devices. The
private key is kept secret and not shared with any other device in
order to protect the privacy of the communications.
Now every device computes its own public key, Q
a
and Q
b,

respectively. Up to this phase, all the computations could be done
off-line and possibly only once during the life of the device.
Once the online computation starts, the device exchanges the
public keys, and once public key is received, device A calculates
Q=k
a
Q
b
while B calculates Q=k
b
Q
a
.
The result of the calculation is equal for the two devices, as
shown by Q=k
a
Q
b
=k
b
Q
a
=k
a
k
b
P.
No secret is exchanged through the unsecured channel. The
only attack possible is the man-in-the-middle attack, where the
attacker substitutes its public key with the two exchanged over
the air. This problem is solved through the use of certificates or
certification authorities. This problem is well known in the field
of cryptography and is out of the scope of the paper. Reference
[4] provides a description of known solutions and problems.
4. RELATED WORK
The number of papers concerning the implementation of ECC
in low-end environments is increasing. Nevertheless, time
performances often remain the only evaluation parameter
throughout the research.
In [5], Gura et al. propose a software implementation that
executes one kP on a 160-bit prime field in 0.81 s on the
ATmega128 processor at 8 MHz. They propose a new multiply-
&-accumulate instruction that reduces the calculation time to
0.59 s. In [6], the kP is calculated on the 163-bit binary field
using the ATmega128. Its execution takes 4.14 s in software
implementation and 0.29 s with new instructions designed for
binary fields. In [7], Kumar and Paar present an FPGA hardware
multiplier such as an ISE for the ATmega128. They measured
0.113 s to execute one kP over GF(2
163
) at 4 MHz.
In [8], Savas et al. introduce a coprocessor containing a
multiplier and a divider for prime finite fields. The time required
for a scalar multiplication is 31.9 ms at 20 MHz for a prime field
GF(p) of 168 bits. They use the TSMC 0.13 m ASIC library for
synthesis and obtain a coprocessor size of 30 kgates and a power
consumption of 0.99 mW at 20 MHz and 1 V power supply. In [9],
Aigner et al. present a coprocessor for 8-bit architecture smart-
cards. The data path mainly consists of leaf cells which integrate
multiplication, addition, and inversion. The coprocessor has been
synthesized by a 0.13 m CMOS process, resulting in 25-kgate
size. The scalar multiplication over GF(2
191
) is executed in 34.1
ms at 10 MHz.
158 COPYRIGHT STMICROELECTRONICS, POLITECNICO DI MILANO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
that the size of the operands can be very large when they are
represented in small architectures. This results in a very large
number of memory accesses performed during one kP. Compared
to the average power consumption of a 12-15 mW 8-bit CPU, the
power consumption of one RAM access is remarkable. In fact,
we estimated the power consumption of an embedded 4-kbyte
RAM working at 8 MHz produced by ST Microelectronics,
and we determined that, during one memory access, the power
absorption is of 70 mW. For this reason, the number of memory
accesses performed during program execution can represent
an important parameter for evaluating energy consumption.
Having taken into account the figures reported in [5], we
calculated that over 4.15 million memory accesses are performed
during a software execution of the Montgomery algorithm
on 8-bit platforms and over 0.43 million with binary field
instructions.
It is possible to estimate the energy spent for executing one kP by
summing up the energy consumed by the CPU and all memory
accesses. It turns out that approximately 120 mJ are spent in the
software implementation and 11.1 mJ after the ISE.
6. DEDICATED COPROCESSOR:
DESIGN AND RESULTS
The performance of the systems can be largely increased if
the cryptographic calculations are committed to a dedicated
coprocessor capable of managing data blocks much larger
than the existing CPU architecture. In the specific case of our
implementation, the operands have a size of 163 bits, while
the existing data path is only 8 bits. We have designed three
different hardware devices in VHDL, and we have synthesized
them using the 0.18 um CMOS technology library designed by
STMicroelectronics. Using the Synopsys tools Design Compiler
and PrimePower, we obtained the area occupancy, the critical
path, and the power consumption for each implementation.
As the most time consuming operations during the kP are field
5. ENERGY CONSUMPTION
DURING A kP
In this work, we considered the implementation of public-key
ECC over the binary field GF(2
163
). This cryptosystem has been
implemented on the Atmel 8-bit AVR ATmega128. The choice of this
processor is due to its technical characteristics and to the previously
published work [5], which chooses it as the target processor.
ATmega128 has a RISC architecture and can work at a clock
frequency up to 16 MHz. For our implementation, we fixed the
clock frequency at 8 MHz. The CPU contains a Register File
of 32 temporary 8-bit registers; the memory access latency is
2 clock cycles. The processor can typically access a 4-kbyte
RAM. The limited resources of the processor yield a low-
power consumption, which makes such a processor suitable for
applications in scarce-energy environments. The average power
consumption of ATmega128 is 12-15 mW [13].
Considering the results reported in the related literature, it is
possible to make a rough evaluation of the energy consumed for
executing one kP operation. In [6], scalar multiplication takes
4.14 s in a software implementation and 0.29 s after an ISE of
the CPU core. The new instructions want to improve the field
multiplication efficiency because it is the most time-consuming
operation during a kP (89% of the total execution time for
software implementations). This happens because the integer
arithmetic instructions of standard processors support binary field
arithmetic rather inefficiently. These machine instructions are
obtained with few modifications to the integer multiplier usually
included in the processor ALU; the hardware overhead required
for the ISE, compared to the actual multiplier size, is estimated
at about 5%. It is reasonable to assume that the cost overhead
after the ISE is quite negligible compared to the cost of the entire
processor. Also the power consumption of the processor should
remain unchanged after the extension of the multiplier structure.
An important remark for energy consumption is that cryptographic
calculations normally require multiple precision operations and
IMPLEMENTING ECC FOR 8-BIT SYSTEMS AND POWER CONSUMPTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
159
multiplications, the first coprocessor we implemented was a
serial-digit hardware multiplier for the binary field GF(2
163
).
Its structure is shown in Fig. 1. The multiplication is carried
out in several iterations by multiplying a 163-bit multiplicand
(register A) by a few bits of the second operand (register B). The
partial result is accumulated into register C at each multiplication
iteration. Thanks to the hardware multiplier, it is possible to
reduce the number of memory accesses performed during a kP
to 255,000 I/O instructions. The time performances provided by
this multiplier depend also on the implemented digit size d: by
choosing d=4 bits, we are able to reach a good trade-off between
additional costs and performances.
After designing a hardware multiplier, we were able to improve
the efficiency of field operations considerably. At this point,
further improvements should be obtained by optimizing
memory operations, in particular by reducing the intensive data
flow between RAM and coprocessor. The expected advantages
concern the time performances and, more importantly, the
reduction of energy consumption.
The second coprocessor we designed targets this goal. In order
to reduce the memory operations, we added to the previous
multiplier three temporary registers of 163 bits each. These
registers can store the last three multiplication results so that it
is no longer necessary to write a result into the RAM if it is an
operand of the following multiplication.
In order to improve data flow efficiency, the coprocessor
executes a hardware logical XOR between two register outputs.
It is thus able to execute a time-costless field addition without
performing any memory operations. This 3-register coprocessor
autonomously executes one iteration of the Montgomery
algorithm, with an optimized memory access strategy. The number
of memory accesses performed during a kP drops to 47,628.
Further reductions of data flow are possible by adding four new
163-bit registers, thus obtaining a 7-register coprocessor. Two
of these registers store the x coordinate of P and the coefficient
b, both used throughout the calculation of kP (registers Rx and
Rb). The other two registers store the coordinates X
b
and Z
b
at
the beginning of each iteration (registers RX
b
and RZ
b
). This
coprocessor only has to access 84 RAM words, containing the x
and z coordinates of points A and B. Inverting the point addition
and doubling procedures, these coordinates are loaded and stored
from and into RAM only once at each iteration. The additional
FIGURE 1: STRUCTURE OF THE HARDWARE MULTIPLIER.
A register Shift & Module B reg.
And matrix
Accumulate
C register
Module
166 bits
4 bits
8 bits
166 bits
163 bits
163 bits
163 bits
2:1
166 bits
FIGURE 2: STRUCTURE OF THE COPROCESSOR WITH LOCAL REGISTERS.
Permanent Register
Result Registers
Multiplier
XOR
Mult.
Unit
Control
Unit
I/O
Unit
8 bits Data bus
A
d
d
r
e
s
s
b
u
s
16 bits
163 bits
160 COPYRIGHT STMICROELECTRONICS, POLITECNICO DI MILANO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
resources available to this coprocessor do not aim mainly at timing
improvements,; they rather intend to increase the efficiency of
data transport, reducing energy consumption as much as possible.
A detailed presentation of the sequence of operations performed
by the 7-register coprocessor can be found in appendix A.
The structure of the coprocessor is depicted in Fig. 2.
The silicon surface and the power consumption of the hardware
implementations are reported in Table 1. As a term of comparison,
we report the same parameters for an 8-bit processor. As it often
comes in the implementation of a cryptographic coprocessor, the
silicon area required by dedicated devices is comparable or larger
than the area of the processor. Nevertheless, it is interesting to
point out the good energetic properties of the new coprocessors,
which match the needs of low power environments.
the results reported in Tables 1 and 2, we calculated an energy
consumption of 5.95 mJ for calculating one kP on the hardware
multiplier and 1.09 mJ on the 3-register coprocessor, while the
consumption is reduced to 0.66 mJ for 7 registers. Table 3 shows
the trade off between area increase and execution time.
7. CONCLUSIONS
Elliptic Curve Cryptography is the best candidate for effective
and efficient implementation of public key primitives in a
wireless sensor network. Other research has been already
presented for the implementation of ECC in software and
hardware, but none of them has focused on the compromise of
low-energy consumption, reduced silicon area requirements, and
significant speed-up compared to software solutions.
Different hardware support has been proposed for improving
timing and power consumption. The coprocessor with 3 register
is probably the best compromise for low-end nodes like those
used in sensor networks; it computes a scalar multiplication in
17 ms @ 8 MHz with a request of 12 Kgates and an energy
consumption of 1.09 mJ. The solution seems very attractive
compared with previously known figures.
9. APPENDIX A
Point Doubling: B 2B
1. load Z
b
into register RZ
b
;
2. load X
b
into register RX
b
(and simultaneous calculation
of RZ
b
2
);
3. calculate RX
b
2
;
ONS, AND
In Table 2, we report the timing results and the memory
operations performed by the two coprocessors. Compared
with the previously analyzed implementations, the advantages
deriving from the use of a dedicated device are evident. The
performance gap of the two hardware devices shows how the
intensive data flow affects the cryptographic operations. Using
4-bit digit multiplier 5,425 172 W
3-reg coprocessor 11,957 305 W
7-reg coprocessor 18,550 503 W
Typical 8bit processor 6,500 12 mW
TABLE 1: SILICON AREA AND POWER CONSUMPTION OF HARDWARE
IMPLEMENTATIONS.
ARCHITECTURE AREA (GATES) CONSUMPTION
Software impl. 2.07 4,148,496 120 mJ
ISE implem. 0.29 435,618 11.1 mJ
4-bit digit mult 97.56 ms 255,550 5.95 mJ
3-reg coprocessor 17.05 ms 47,62 1.09 mJ
7-reg coprocessor 14.68 ms 27,216 0.66 mJ
TABLE 2: EXECUTION TIME, MEMORY OPERATIONS, AND ENERGY CONSUMPTION.
COPROCESSOR EXECUTION NUMBER OF ENERGY
TIME MEMORY CONSUMPTION
OPERATIONS
Software implementation - 2.07 s
ISE implementation 5,425 0.29 s
4-bit digit multiplier 11,957 97.56 ms
3-reg coprocessor 18,550 17.05 ms
7-reg coprocessor 6,500 14.68 ms
TABLE 3: COMPARISON OF AREA TIME TRADE OFF.
COPROCESSOR ADDITIONAL AREA EXECUTION
(GATES) TIME
IMPLEMENTING ECC FOR 8-BIT SYSTEMS AND POWER CONSUMPTION
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
161
4. calculate Z
b
RZ
b
2
RX
b
2
;
5. store Z
b
into memory (and simultaneous calculation
of RZ
b
2
RZ
b
2
);
6. calculate RX
b
2
RX
b
2
;
7. calculate Rb RZ
b
4
;
8. store into memory RX
b
4
(Rb RZ
b
4
).
Point Addition: A A + B
1. load X
a
into register B (and simultaneous calculation
of X
a
RZ
b
);
2. load Z
a
into register B (and simultaneous calculation
of Z
a
RX
b
);
3. calculate Z
a
[(X
a
RZ
b
) (Z
a
RX
b
)]
2
;
4. store Z
a
into memory (and simultaneous calculation
of (X
a
RZ
b
) (Z
a
RX
b
) );
5. calculate Px Z
a
;
6. store X
a
(Px Z
a
) [(X
a
RZ
b
) (Z
a
RX
b
)]
into memory.
REFERENCES
[1] A. Perring, H. Stankovic, and D.Wagner. SECURITY
IN WIRELESS SENSOR NETWORKS, in Communications
of the ACM, volume 47, pages 53-57, 2004.
[2] J. Hill, R. Szewczyk, A. Woo, S. Hollar, D. Culler,
and K. Pister. SYSTEM ARCHITECTURE DIRECTIONS
FOR NETWORKED SENSORS, in ACM, editor, ASPLOS IX,
pages 93-104, November 2000.
[3] W Diffie and M. Hellmann. NEW DIRECTIONS
IN CRIPTOGRAPHY, in Transaction in Information Theory,
IT-22, pages 644-654. IEEE Computer Society, 1976.
[4] D. Hankerson, A. J. Menezes, and S. A. Vanstone. GUIDE
TO ELLIPTIC CURVE CRYPTOGRAPHY, Springer-Verlag, 2004.
[5] N. Gura, A. Patel, A. Wander, H. Eberle, and S. Chang
Shantz. COMPARING ELLIPTIC CURVE CRYPTOGRAPHY
AND RSA ON 8-BIT CPUS, in M. Joye and J. J. Quisquater,
editors, Cryptographic Hardware and Embedded Systems
CHES 2004, volume LNCS 3156, pages 119132.
Springer-Verlag, 2004.
[6] H. Eberle, A. Wander, N. Gura, and S. Chang-Shantz.
ARCHITECTURAL EXTENSIONS FOR ELLIPTIC CURVE
CRYPTOGRAPHY OVER GF(2
m
), Sun Microsystems
Laboratories, 2005.
research.sun.com/sunlabsday/docs.2004/Micro.pdf.
[7] S. Kumar and C. Paar. RECONFIGURABLE INSTRUCTION
SET EXTENSION FOR ENABLING ECC ON AN 8-BIT
PROCESSOR, Chair for Communication Security, Ruhr-
Universitt Bochum, Germany, 2004. www.crypto.ruhr-uni-
bochum.de /imperia/md/../fpl04/final.pdf.
[8] E. ztrk and B. Sunar and E. Savas. LOW-POWER
ELLIPTIC CURVE CRYPTOGRAPHY USING SCALED MODULAR
ARITHMETIC, in M. Joye and J. J. Quisquater, editors,
Cryptographic Hardware and Embedded Systems
CHES 2004, volume LNCS 3156, pages 133147.
Springer-Verlag, 2004.
[9] H. Aigner and H. Bock and M. Htter and J.Wolkerstorfer.
LOW-COST ECC COPROCESSOR FOR SMART CARD,
in M. Joye and J. J. Quisquater, editors, Cryptographic
Hardware and Embedded Systems CHES 2004, volume
LNCS 3156, pages 107118. Springer-Verlag, 2004.
[10] A.K. Lenstra and E.R. Verheul. SELECTING
CRYPTOGRAPHIC KEY SIZES, in Journal of Cryptology, 2000.
[11] D. Hankerson, J. Lpez Hernandez, and A. Menezes.
SOFTWARE IMPLEMENTATION OF ELLIPTIC CURVE
CRYPTOGRAPHY OVER BINARY FIELDS, in . Ko
and C. Paar, editors, Second International Workshop on
Cryptographic Hardware and Embedded Systems CHES
2000, volume LNCS 1965, Berlin, 2000. Springer-Verlag.
162 COPYRIGHT STMICROELECTRONICS, POLITECNICO DI MILANO, 2007.
ST JOURNAL OF RESEARCH - VOLUME 4 - NUMBER 1 - WIRELESS SENSOR NETWORKS
[12] J. Lpez and R. Dahab. FAST MULTIPLICATION ON
ELLIPTIC CURVES OVER GF(2
m
) WITHOUT PRECOMPUTATION,
in Jr. . K. Ko and C. Paar, editors, Cryptographic
Hardware and Embedded Systems CHES 1999,
volume LNCS 1717, pages 316 327. Springer-Verlag, 1999.
CONTACT: ST.JOURNAL@ST.COM
[13] R. Venugopalan and A. Dean. IMPROVING ENERGY-
EFFICIENCY IN SENSOR NETWORKS BY RAISING
COMMUNICATION THROUGHPUT USING SOFTWARE THREAD
INTEGRATION, www.cesr.ncsu.edu/agdean/ODES 04/ODES
Venugopalan slides.pdf.
STMicroelectronics
Corporate Headquarters
39, Chemin du Champ-des-Filles
C.P. 21
CH-1228 Geneva
Plan-Les-Ouates - Switzerland
www.st.com