EURASIP Book Series on Signal Processing and Communications

UWB
Communication Systems
A Comprehensive Overview
Edited by: Maria-Gabriella Di Benedetto, Thomas Kaiser,
Andreas F. Molisch, Ian Oppermann, Christian Politano,
and Domenico Porcino

UWB Communication Systems

A Comprehensive Overview

EURASIP Book Series on Signal Processing and Communications, Volume 5

UWB Communication Systems

A Comprehensive Overview
Edited by: Maria-Gabriella Di Benedetto, Thomas Kaiser, Andreas F. Molisch,
Ian Oppermann, Christian Politano, and Domenico Porcino

Hindawi Publishing Corporation

http://www.hindawi.com

EURASIP Book Series on Signal Processing and Communications

Editor-in-Chief: Alex Gershman
Editorial Board: Zhi Ding, Moncef Gabbouj, Peter Grant, Ferran Marques, Marc Moonen,
Hideaki Sakai, Giovanni Sicuranza, Bob Stewart, and Sergios Theodoridis
Hindawi Publishing Corporation
410 Park Avenue, 15th Floor, #287 pmb, New York, NY 10022, USA
Nasr City Free Zone, Cairo 11816, Egypt
Fax: +1-866-HINDAWI (USA Toll-Free)

2006 Hindawi Publishing Corporation

All rights reserved. No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical, including photocopying, recording,
or any information storage and retrieval system, without written permission from the publisher.
ISBN 977-5945-10-0

Contents

Preface

vii

1. Introduction
1.1. Introduction
1.2. UWB basics
1.3. Regulatory bodies
1.4. Applications of UWB
1.5. Impulse radio schemes
1.6. Multicarrier schemes
1.7. Conclusions

1
1
2
4
9
10
14
17

2. UWB propagation channels

2.1. Introduction
2.2. Measurement techniques
2.3. Propagation eects
2.4. Path loss and shadowing
2.5. Delay dispersion and small-scale fading
2.6. Standardized channel models
2.7. Body-area networks
2.8. Channel estimation techniques

21
21
26
40
57
67
87
94
118

3. Signal processing
3.1. Introduction
3.2. Impulse radio schemes
3.3. Multicarrier schemes
3.4. Pulse shapes
3.5. Data modulation
3.6. Spectrum randomisation and multiple access
3.7. Synchronisation
3.8. Impulse radio demonstrator for 4-PPM
3.9. Conclusion

143
143
144
147
150
157
167
175
181
200

4. Higher-layer issues: ad hoc and sensor networks

4.1. Introduction
4.2. Power-ecient UWB networks
4.3. Location-aware UWB networks
4.4. Power-ecient and location-aware medium access control design
4.5. Performance analysis in specific test cases

205
205
206
209
219
222

vi

Contents

5. Spatial aspects of UWB

5.1. Introduction
5.2. A model for the ultra-wideband space-variant indoor
multipath radio channel
5.3. UWB antenna arrays
5.4. UWB polarization diversity
5.5. Spatial diversity
5.6. UWB beamforming and DOA estimation
5.7. Performance analysis of multiantenna UWB
wireless communications
5.8. Channel capacity of MIMO UWB indoor wireless systems

253
253

6. UWB ranging
6.1. Introduction
6.2. UWB location system techniques, architectures, and analysis
6.3. Comparison of UWB and alternative radio-based systems
6.4. A typical RF link budget for UWB positioning systems
6.5. Characteristics of a fine-grained UWB positioning system
6.6. Positioning techniques in harsh environments
6.7. UWB precise ranging with an experimental antenna-array system
6.8. Systems integration and UWB positioning technology

411
411
412
418
421
423
426
429
440

7. Regulation and standardization

7.1. Introduction
7.2. Regulation
7.3. Standardization
7.4. Coexistence with radio systems

447
447
448
459
471

Index

493

254
269
281
302
330
353
376

Preface

Ultra-wideband (UWB) communication systems oer an unprecedented opportunity to impact the future communication world. The enormous available bandwidth, the wide scope of the data rate/range trade-o, and the potential for verylow-cost operation, which will lead to pervasive usage, all present a unique opportunity for UWB systems to impact the way people and intelligent machines communicate and interact with their environment. In particular, UWB is a promising
area oering enormous advantages for short-range communications. Nevertheless, the technology still requires much work from the research community as well
as solid proof of its viability in the commercial world before it can claim to be
successful.
The world of UWB is changing rapidly, and it may be argued that the information contained in any general text on the subject is obsolete before the ink has
dried. Even between the writing of the manuscript (Winter 2004) and the actual
production, we have seen a number of interesting developments in the field. Our
book attempts to provide an understanding of the (longer-term) fundamentals
of UWB, the major research and development challenges, as well as a snapshot
of the work in progress addressing these challenges. Due to the rapid progress of
multidisciplinary UWB research, such a comprehensive overview can generally be
achieved by combining the areas of expertise of several scientists in the field.
More than 30 leading UWB researchers and engineers have contributed to
this book, which covers the major topics relevant to UWB. These topics include
UWB signal processing, UWB channel measurements and modelling, higher-layer
protocol issues, spatial aspects of UWB signalling, UWB regulation and standardisation, implementation issues, UWB applications, and positioning with UWB systems.
The book is targeted at advanced academic researchers, wireless designers,
and graduate students wishing to greatly enhance their knowledge of all aspects
of UWB systems. The reader should be left with a high-level understanding of
the potential advantages of UWB in terms of high-data-rate communications, and
location and tracking capabilities.
Due to the sheer number of authors who have contributed to this book and
many others involved, it is dicult to equally thank them all, so we generally apologize for the absence of personal acknowledgements.
Introduction
This introductory first chapter by I. Oppermann is intended to give the reader
a high-level understanding of the scope and the role of UWB as well as some

viii

Preface

appreciation for the dicult birth UWB has experienced in the crowded world
of communication standards. The enormous bandwidth and very low power spectral density of UWB make it dicult to detect; therefore it is potentially dicult
to operate in such as way as to realise these benefits. The nature of UWB also leads
to very significant technical diculties, which are then compounded by regulatory
and commercial resistance to UWB as a technology. The introduction covers some
of the basic UWB signal generation techniques, working definitions of UWB, the
on-going regulatory situation broad application areas and current research focus
areas.
UWB propagation channels
Chapter 2 describes the propagation of UWB radiation from the transmitter to
the receiver, covering the physical processes of the propagation as well as the measurement and modelling of the channels. The first section by A. F. Molisch is an
introduction that outlines the basic properties of UWB propagation channels (as
compared to the well-known narrowband channels) and also gives a synopsis of
the subsequent sections. Section 2.2, Measurement techniques, by J. Kunisch describes how to measure the transfer function or impulse responses of UWB channels and pays particular attention to the impact of the antennas used during the
measurement. As it is desirable to isolate the impact of the channel (without the
antennas), the de-embedding of the antenna eects becomes of paramount importance. Next, Section 2.3 by R. Qiu describes the fundamental propagation effects for UWB radiation. In particular, the diraction of an ultra-wideband wave
(corresponding to a short pulse) by a half-plane is a canonic problem that is analysed. It is shown that the pulse is distorted during that process; this has important
consequences for the design of optimum receiver structures (matched filters). After those fundamental investigations, Section 2.4 to Section 2.6 describe statistical
channel models. Section 2.4 by D. Cassioli gives an overview of measurements and
models of path loss and shadowing, which describe the large-scale channel attenuation. Section 2.5 by A. F. Molisch and M. Buehrer then analyses the small-scale
fading and delay dispersion eects that are caused by the multipath propagation.
This section also investigates the frequency selectivity of the reflection coecients
of various materials and shows that it can lead to similar distortions of each multipath component as the diraction eects described in Section 2.3. Finally, Section 2.6 by M. Pendergrass describes a standardised channel model established by
the IEEE 802.15.3a group. This model, which is suitable for indoor environments
with distances of up to 10 m, has been used for the evaluation of various highdata-rate UWB systems. Another type of system is body area networks, where various devices mounted on a human body communicate via UWB radiation. This
rather unique environment gives rise to new challenges in the channel measurement and modelling. Section 2.7 by I. Z. Kovacs, G. F. Pedersen, and P. C. F. Eggers
describes such channels. Finally, Section 2.8 by S. Roy and I. Ramachandran gives
an overview of channel estimation techniques for both OFDM-based and impulseradio-based UWB systems.

Preface

ix

Signal processing
This chapter addresses signal processing issues in UWB. The first three sections by
I. Oppermann, M. Hamalainen, J. Iinatti, and A. Rabbachin, present both impulse
radio techniques and multiband techniques. UWB systems may be primarily divided into impulse radio (IR) systems and multiband systems. Multiband systems
oer the advantage of potentially ecient utilisation of spectrum, while IR systems have the advantage of simplicity, and thus have potentially lower cost. The
IR UWB concepts investigated support many modulation schemes including orthogonal and antipodal schemes. However, the basic modulation must also include
some form of spectrum randomisation techniques to limit the interference caused
by the transmitted pulse train. Both time-hopping (TH) and direct-sequence (DS)
randomisation techniques were examined. Deciding which modulation scheme to
use depends on the expected operating conditions and the desired system complexity. Section 3.4 by B. Allen, S. A. Ghorashi, and M. Ghavami introduces the
application of impulses to UWB wireless transmissions. A number of candidate
pulse waveforms are characterised in the time and frequency domains. The application of orthogonal pulse waveforms is introduced. These waveforms enable advanced modulation and multiple-access schemes to be implemented. The success
of these schemes, however, is determined by the extent of pulse distortion caused
by the transmitter and receiver circuitry and the propagation channel. Thus, distortion mitigation techniques are required. The issue of coexistence of impulse
radio with other spectrum users is also discussed. Section 3.7 by I. Oppermann,
M. Hamalainen and J. Iinatti discusses synchronisation in IR systems. After a brief
introduction of optimal synchronisation schemes, a more realistic approach is investigated in closer detail. Finally, the last Section by O. Albert and C. F. Mecklenbrauker presents a UWB radio testbed based on pulse position modulation
(PPM) for investigating the properties of short-range data communication. The
testbed is designed to realise data transmission at 6 Msymb/s over a distance of a
few meters in indoor oce environments. The focus of this eort is on the implementation of commercial-grade microwave circuitry and algorithms for ultrawideband data transmission, especially concerning mobile battery-driven devices.
The testbed hardware is described in detail as well as the two-stage approach used
for receiver synchronisation.
Higher-layer issues: ad hoc and sensor networks
In the last few years, the increasing interest in applications based on the deployment of ad hoc networks has triggered significant research eorts regarding the
introduction of the energy-awareness concept in the design of medium access
control (MAC) and routing protocols. Chapter 4, which was edited by M.-G. Di
Benedetto and includes contributions by L. De Nardis, S. Falco, and M.-G. Di
Benedetto, investigates this issue in the context of UWB, ad hoc, and sensor networks. Progressing from the analysis of the state of the art in energy-ecient MAC
and routing protocols, the chapter presents an innovative energy-aware MAC and

Preface

routing solution based on the position information provided by UWB by means

of a distributed positioning protocol.
Ad hoc networks are considered as a viable solution for scenarios in which
fixed infrastructure and, consequently, unlimited power sources are not available.
In such scenarios, an ecient management of the limited power supply available
in each terminal is a key element for achieving acceptable network lifetimes. This
is particularly true for sensor networks, for which long battery duration is one of
the basic requirements, given the typical size of such networks (up to thousands
of terminals), as will be analysed in Section 4.1 by L. De Nardis and M.-G. Di
Benedetto.
Location information is another valuable way of achieving energy awareness
in ad-hoc networks. In Section 4.2, by L. De Nardis, we will first review locationaware routing protocols with a focus on power eciency. We then address the
problem of information exchange through the network by means of specifically
designed protocols.
Next, we introduce in Section 4.3, by L. De Nardis, a MAC protocol that foresees a dedicated procedure for the acquisition of distance information and is tailored on UWB features.
Section 4.4, by L. De Nardis and S. Falco, analyses the eect of mobility on the
behaviour of the proposed MAC and routing strategies.
Spatial aspects of UWB
The aim of this chapter is to discuss the spatial aspects of the UWB radio channel from various perspectives. In the second section by J. Kunisch, a spatial model
is presented based on measurement data that was obtained in oce environments
with line-of-sight, non-line-of-sight, and intermediate conditions. The distinguishing feature of the ultra-wideband indoor radio channel is that certain individual paths are recognisable and resolvable in the measurements, which are adequately reflected by the proposed spatial model. Moreover, the model also covers
movements of the receiving or of the transmitting antennas on a small scale (several wavelengths). In Section 5.3 by W. Sorgel, C. Waldschmidt, and W. Wiesbeck,
the state-of-the-art UWB antenna array concepts and their applications are introduced. Simulation results for a linear exponentially tapered slot antenna array
(Vivaldi antennas) for time domain beam steering are given, and further theoretical modelling is substantiated by experimental results. F. Argenti, T. Bianchi, L.
Mucchi, and L. S. Ronga present in Section 5.4. a two-transmit-antenna scheme
with orthogonal polarisation in order to uniformly cover an indoor area. A single
antenna receiver results in a 2 1 MISO (multiple-input single-output) system
so that space-time coding becomes applicable. The goal is to employ the polarisation diversity in order to obtain quality constant symbol detection while the
receiver moves around within the covered area. Section 5.5 by A. Sibille addresses
several basic issues pertaining to spatial diversity in UWB systems. It mainly concentrates on impulse-based radio, but it also evokes eects concerning frequency
domain modulation schemes such as OFDM. The main subjects are the impact

Preface

xi

of electromagnetic coupling between sensors and the impact of channel properties (like fading and angular variance) on spatial diversity. Besides the two main
pillars of MIMO signal processing, namely, spatial multiplexing and space-time
coding, the third and most classical one is beamforming, which is revisited in Section 5.6 by S. Ries, C. Senger, and T. Kaiser. Since the pulse duration is shorter
than the travel time between two colocated antennas, beamforming for UWB signals has some special properties that are dierent from the narrowband case. For
instance, because of the absence of grating lobes in the beam pattern, the spacing of the array elements is not limited by half of the wavelength, so that high
resolution can be achieved with only a few array elements. At the end of this chapter the principal feasibility of direction-of-arrival (DoA) estimation is shown by
an illustrative example. In Section 5.7 by W. P. Siriwongpairat, M. Olfat, W. Su,
and K. J. Ray Liu, the performance of UWB-MIMO systems using dierent models for the wireless channels and dierent modulation schemes is presented. In
particular, the performance merits of UWB-ST-coded-systems employing various
modulation and multiple-access techniques, including time-hopping (TH) M-ary
pulse-position modulation (MPPM), TH binary phase-shift keying (BPSK), and
direct-sequence (DS) BPSK, are mentioned. At the end, the application of multiple transmit and/or receive antennas in a UWB-OFDM system is discussed. The
last section by F. Zheng and T. Kaiser presents an evaluation of ergodic capacity
and outage probability for UWB indoor wireless systems with multiple transmit
and receive antennas (multiple-input multiple-output, MIMO). For some special
cases, analytic closed-form expressions for the capacity of UWB wireless communication systems are given, while for other cases the capacity is obtained by Monte
Carlo simulation approach. The contribution reveals that the UWB MIMO communication rate supportable by the channel increases linearly with the number of
transmit or receive antennas for a given outage probability, which is reminiscent
of the significant data rate increase of MIMO narrowband fading channels.
UWB ranging
One of the most innovative features of ultra-wideband technology is the very-high
temporal resolution associated with the typical spread of UWB energy over large
frequency bands. This feature has inspired and is still inspiring a new generation of
technical developments and applications looking at the challenges of detection of
the location of people and objects (people and asset tracking, smart spaces, ambient intelligence). Chapter 6 of this book introduces the reader to the basics of location technology via UWB systems. An analysis of the main potential and practical
techniques with their accuracy is presented in Section 6.2 by A. Ward. In Section
6.3 by A. Ward and D. Porcino, a comparison with alternative radio-based systems
is closely considered, while the main implications on the physical layer of a radio
design and a sample RF link budget are discussed in Section 6.4 by D. Porcino. In
the next section, results and advantages are then presented from commercial finegrained UWB positioning systems by A. Ward, leading the discussion to operations
in very challenging propagation conditions (such as metal cargos) where the signal

xii

Preface

distortion is very significant and makes ranging calculations very dicult. These
issues are part of Section 6.6 by D. Porcino. An experimental antenna-array system
is then introduced in Section 6.7 by J. Sachs and R. Zetik, with test results showing the limits of what has been achieved today in terms of maximum positioning
accuracy in controlled situations. In the last section by A. Ward, the important
aspects of system integration of UWB positioning technology in the real world of
complex buildings, hospitals, and houses are presented with attention given to the
requirements that this technology will put onto system integrators and designers.
This chapter guides the reader with a language suitable to comprehend the basic
principles of the promising UWB ranging features, which are about to be widely
explored in the commercial world.
Regulation and standardization
Since typical UWB radio may use part of the radio frequency spectrum already
assigned to operative primary or secondary radio services, UWB radio devices
despite their extremely low transmission output powermay be a potential source
of interference for incumbent radio services and should provide the ability to
coexist with legacy radio services. In this chapter, we provide the reader with a
comprehensive description of the worldwide UWB regulation and standardization framework under completion in order to elaborate on novel and eective
means of spectrum management based on coexistence mechanisms instead of using conventional frequency sharing mechanisms. The first section by C. Politano
is dedicated to regulation aspects for UWB communication and positioning applications to operate without requirement for individual right to use radio spectrum
(license-exempt), and on a no protection, no harmful interference basis. This
section introduces the UWB regulation framework elaborated under ITU-R to allow UWB devices and provides a detailed overview on the European regulatory
approach in order to explain how prudent but constructive regulation rules are
elaborated in this region. The second section by W. Hirt presents UWB standardization overview in the USA and in Europe. It provides the global overview of IEEE
working groups mandated for wireless personal area network (WPAN) including
high-data-rate (IEEE802.15.3a) and low-data-rate (IEEE802.15.4a) communication applications, and introduces the most popular UWB technology candidates
identified during the year 2004. Then in this section, the European standardization process is also presented with an overview of the ETSI-TG3 working group
mandated by the ECC for the definition of harmonised standards for short-range
devices (SRD) using UWB technology. The last section by R. Guiliano describes
the methodology compatibility study for evaluating UWB interferences risk with
incumbent radio services and presents some evaluation results of compatibility
studies conducted for a few coexistence scenarios between UWB and UMTS, and
fixed services (PP, PMP). The particular interest of this section is the description
of the UWB characteristics that are impacting compatibility studies (activity factors, trac characteristics, radio access modes) and how interference risks may be
reduced by applying these features, allowing incumbent radio receivers to coexist

Preface

xiii

with UWB devices. The results of such compatibility studies have been used for
the elaboration of mitigation techniques (proposed for standardization) in order
to specify coexistence mechanisms to reduce UWB interferences with incumbent
radio services.
Maria-Gabriella Di Benedetto, Thomas Kaiser, Andreas F. Molisch,
Ian Oppermann, Christian Politano, and Domenico Porcino
March 2006

Introduction
Ian Oppermann

1.1. Introduction
Ultra-wideband (UWB) communication systems have an unprecedented opportunity to impact communication systems. The enormous bandwidths available,
the wide scope of the data rate/range tradeo, and the potential for very-low-cost
operation leading to pervasive usage, all present a unique opportunity for UWB to
impact the way people interact with communications systems.
The spark-gap transmission experiments of Marconi in 1901 represent some
of the first experiments in a crude form of impulse radio. Pioneering contributions
to modern UWB radio were made by Ross and Bennett [1] and Harmuth [2]. The
earliest radio communications patent was published by Ross (1973). In the past 20
years, UWB has been used for radar, sensing, military communications, and niche
applications.
A substantial change occurred in February 2002 when the US Federal Communications Commission (FCC) [3, 4] issued a ruling that UWB could be used
for data communications as well as radar and safety applications. This book will
focus almost exclusively on the radio communications aspects of UWB.
The band the FCC allocated to communications is 7.5 GHz between 3.1 and
10.6 GHz; by far the largest allocation of bandwidth to any commercial terrestrial
system. It was little wonder that eorts to bring UWB into the mainstream were
greeted with great hostility. First, the enormous bandwidth of the system meant
that UWB could potentially oer data rates of the order of Gbps. Second, the
bandwidth was overlaid on many existing allocations, causing concern from those
groups with the primary allocations. When the FCC proposed the UWB rulings,
they received almost 1000 submissions opposing the proposed rulings.
Fortunately, the FCC UWB rulings went ahead. The concession was however
that available power levels would be very low. At the time, the FCC made it clear
that they were being deliberately cautious with the setting of the maximum power
masks. If the entire 7.5 GHz band is optimally utilized, the maximum power available to a transmitter is approximately 0.5 mW. This is a tiny fraction of what is

Introduction

available to users of the 2.4 GHz and 5.8 GHz ISM (industrial, scientific, and medical) bands used by wireless LAN (WLAN) standards such as IEEE 802.11 a/b/g.
The power limitation eectively relegates UWB to indoor, short-range communications for high data rates, or very low data rates for longer link distances. Applications such as wireless USB and personal area networks have been proposed
oering hundreds of Mbps to several Gbps with distances of 1 to 4 metres. For
ranges of 20 metres or more, the achievable data rates are very low compared to
existing wireless local area network (WLAN) systems.
Part of the enormous potential of UWB is the ability to move between the
very high data rate, short-link distance applications and the very low data rate,
longer-link-distance applications. The tradeo is facilitated by the physical layer
signal structure. The very low transmit power available means that multiple, lowenergy UWB pulses typically must be combined to carry one bit of information.
In principle, trading data rate for link distance can be as simple as increasing the
number of pulses used to carry one bit. The more pulses per bit, the lower the data
rate, and the greater the achievable transmission distance.
1.1.1. Goals and scope of this book
This book explores the fundamentals of UWB technology with particular emphasis on the physical layer (PHY) and medium access (MAC) layer. Regulatory aspects and the state of the art will also be presented. The topics addressed include
UWB signal processing, UWB channel measurement and modeling, higher layer
issues, UWB spatial aspects, UWB standardization and regulation, implementation issues, and UWB applications including positioning.
The goals of the early parts of the book are to provide the essential aspects
of knowledge of UWB technology, especially in communications and in control
applications. A literature survey examining books, articles, and conference papers
presents the basic features of UWB technology and current systems.
1.2. UWB basics
Other terms associated with UWB include impulse, short-pulse, nonsinusoidal, carrier-less, time domain, super wideband, fast frequency chirp,
and monopulse [5].
Impulse radio communication systems and impulse radars both utilize very
short pulses in transmission that results in an UWB spectrum. For radio applications, this communication method is also classified as a pulse modulation technique because the data modulation is introduced by pulse-position modulation
(PPM). The UWB signal is noise-like which makes interception and detection
quite dicult. Due to the low power spectral density, UWB signals cause very little interference with existing narrowband radio systems, and according to current
FCC regulations, UWB signals are not necessarily regarded as radio transmissions.
Depending on the attitude of national and international regulatory bodies, this
should allow licence-free operation of radio systems.

Ian Oppermann

Time-modulated (TM) impulse radio signal is seen as a carrier-less baseband

transmission. The absence of carrier frequency is the fundamental characteristic
that dierentiates impulse radio and impulse radar transmissions from narrowband applications and from direct-sequence (DS) spread-spectrum (SS) multicarrier (MC) transmissions, which can also be characterized as an (ultra-) wideband
technique. Fast slewing chirps and exponentially damped sine waves are also possible methods to generate UWB signals.
1.2.1. Characteristics of UWB
UWB has a number of features which make it attractive for consumer communications applications. In particular, UWB systems
(i) have potentially low complexity and low cost;
(ii) have a noise-like signal spectrum;
(iii) are resistant to severe multipath and jamming;
(iv) have very good time-domain resolution allowing for location and tracking applications.
The low complexity and low cost of UWB systems arise from the baseband nature of the signal transmission. Unlike conventional systems, the UWB transmitter
produces a very short time-domain pulse which is able to propagate without the
need for an additional RF (radio frequency) mixing stage. The RF mixing stage
takes a baseband signal and injects a carrier frequency or translates the signal to
a frequency which has desirable propagation characteristics. The very wideband
nature of the UWB signal means that it spans frequencies commonly used as carrier frequencies. The signal will propagate well without the need for additional
up-conversion. The reverse process of down-conversion is also not required in the
UWB receiver. Again, this means the omission of a local oscillator in the receiver,
and the removal of associated complex delay and phase tracking loops.
Consequently, UWB systems can be implemented in low-cost, low-power integrated circuit processes [6]. UWB technique also oers grating lobe mitigation
in sparse antenna array systems without weakening of the angular resolution of
the array. Grating lobes are a significant problem in conventional narrowband antenna arrays.
Due to the low energy density and the pseudo-random (PR) characteristics
of the transmitted signal, the UWB signal is noise-like which makes unintended
detection dicult. Whilst there is some debate, it appears that the low-power,
noise-like UWB transmissions do not cause significant interference to existing radio systems. The interference phenomenon between impulse radio and existing
radio systems is one of the most important topics in current UWB research.
UWB systems oer high data rates for communication. Hundreds of Mbps
have been reported for communication links. In [6], it is estimated that the number of users in an impulse radio communication system is much larger than in
conventional systems. The estimation is claimed to be valid for both high- and
low-data rate communications.

Introduction

Because of the large bandwidth of the transmitted signal, very high multipath
resolution is achieved. The large bandwidth oers (and also requires) huge frequency diversity, which together with the discontinuous transmission makes the
UWB signal resistant to severe multipath propagation and jamming/interference.
UWB systems oer good LPI and LPD (low probability of interception/detection)
properties which make it ecient for secure and military applications. These properties, in turn, make commercial systems more challenging to develop.
The very narrow time-domain pulses mean that UWB radios are potentially
able to oer timing precision much better than GPS (global positioning system)
[6] and other radio systems. Together with good material penetration properties,
UWB signals oer opportunities for short-range radar applications such as rescue
and anti-crime operations, as well as in surveying, and in the mining industry. One
should however understand that UWB does not provide precise targeting and extreme penetration at the same time, but UWB waveforms present a better tradeo
than many conventional radio systems.
1.3. Regulatory bodies
One of the important issues in UWB communication is the frequency of operation. Due to the use of very wide spectrum range, UWB systems are not intended
to operate under any specific allocation. There are many systems operating under
allocated bands in the UWB signal band. Some companies in the USA are working towards removing the restrictions from the FCCs regulations for applications
utilizing UWB technology. These companies have established an UWB working
group (UWBWG) to negotiate with the FCC. Some other companies and organizations in the USA are examining solutions to allow existing narrowband allocated
services to be protected from possible interference generated in UWB systems.
Discussions on frequency allocation and protection of other radio services from
interference are also taking place in Europe. Currently, there are no dedicated frequency bands for UWB applications identified by ETSI (European Telecommunications Standards Institute), in the ECC (European Communication Committee)
decisions or in ITU (International Telecommunications Union) Radio Regulation
treaty.
1.3.1. UWB regulation in the USA
Before the FCCs First Report and Order [3, 4], there was significant eort by industrial parties to convince the FCC to release UWB technology under the FCC
Part 15 regulation limitations and allow license-free use of UWB products. The
FCC Part 15 rules permit the operation of classes of radio frequency devices without the need for a license or the need for frequency coordination (47 CFR 15.1).
The FCC Part 15 rules attempt to ensure unlicensed devices do not cause harmful interference to other users of the radio spectrum (47 CFR 15.5). Within the
FCC Part 15 rules, intentional radiators are permitted to operate within a set of
limits (47 CFR 15.209) that allow signal emissions in certain frequency bands.

Ian Oppermann

Intentional radiators are not permitted to operate in sensitive or safety-related frequency bands, which are designated as restricted bands (47 CFR 15.205). UWB
devices are intentional radiators under FCC Part 15 rules.
In 1998, the FCC issued a Notice of Inquiry (NOI) [7]. Despite the very low
transmission power levels anticipated, proponents of existing systems raised many
claims against the use of UWB for civilian communications. Most of the claims
related to the anticipated increase of interference level in the restricted frequency
bands (e.g., TV broadcast bands and frequency bands reserved for radio astronomy and GPS). The Federal Aviation Administration (FAA) expressed concern
about the interference to aeronautical safety systems. The FAA also raised concerns about the direction finding of UWB transmitters.
When UWB technology was proposed for civilian applications, there were
no definitions for the signal. The Defence Advanced Research Projects Agency
(DARPA) provided the first definition for UWB signal based on the fractional
bandwidth B f of the signal. The first definition provided that a signal can be classified as an UWB signal if B f is greater than 0.25. The fractional bandwidth can be
determined using the formula [5]

Bf = 2

fH fL
,
fH + fL

(1.1)

where fL is the lower and fH is the higher 3 dB point in the spectrum, respectively.
In February 2002, the FCC issued the FCC UWB rulings that provided the first
radiation limitations for UWB, and also permitted technology commercialization.
The final report of the FCC First Report and Order [3, 4] was publicly available
during April 2002. The document introduced four dierent categories for allowed
UWB applications, and set the radiation masks for them.
The prevailing definition has decreased the limit of B f to a minimum of 0.20
defined using the equation above. Also, according to the FCC UWB rulings, the
signal is characterized as UWB if the signal bandwidth is 500 MHz or more. In the
formula above, fH and fL are now the higher and lower 10 dB bandwidths, respectively. The radiation limits set by the FCC are presented in Table 1.1 for indoor
and outdoor data communication applications.
1.3.2. UWB regulation in Europe
At the time of writing, regulatory bodies in Europe are awaiting further technical input on the impact of UWB on existing systems. The process of regulation in Europe has been ongoing for a considerable time and has experienced repeated delays. The European approach is somewhat more cautious than that of the
USA, as Europe requires that a new technology must be shown to cause little or
no harm to existing radio services. Of course, the European organizations must
take into account the FCCs decision, in full awareness of the potential benefits to

Introduction
Table 1.1. FCC radiation limits for indoor and outdoor communication applications.
Indoor

Outdoor

EIRP (dBm/MHz)

EIRP (dBm/MHz)

9601610

75.3

75.3

16101990

53.3

63.3

19903100

51.3

61.3

310010600

41.3

41.3

Above 10600

51.3

61.3

Frequency (MHz)

consumers of achieving globally compatible conditions of radio spectrum use for

UWB.
Currently in Europe, the recommendations for short-range devices belong to
the CEPT (the European Conference of Postal and Telecommunications) working
group CEPT/ERC/SRD MG/REC 7003 [8].
In March 2003, the European Commission gave a mandate [9] to the European standardization organizations, with the purpose of establishing a set of
harmonized standards covering UWB applications. ETSI in turn established Task
Group ERM TG31A to develop a set of harmonized standards for short-range
devices using UWB technology by December 2004, following the completion of
spectrum compatibility studies by CEPT. The output of this group has been delayed. The standardization working groups for UWB include ERM/TG31A covering generic UWB, and ERM/TG31B which covers UWB for automotive applications at higher bands. These groups have developed two technical reports on
communication and on radio location UWB applications, respectively. In the related report [10] a preliminary slope mask is presented that allows UWB communication systems to operate. This mask referred to in Figure 1.1 as the provisional
CEPT/ETSI mask is supported by the UWB industry within ETSI as the starting
point for studies by CEPT. ETSI should further develop two standards on communication and on radio location UWB applications.
The European Commission has further given CEPT a mandate [11] related to
the development of harmonized conditions of use of UWB devices in the European Union. This mandate requires an in-depth analysis of the risks of harmful
interference of UWB on all potentially aected radio services, including via aggregation eects, as well as taking into account the existing regulatory environment.
In March 2004, the ECC (within the CEPT) established Task Group TG3 to
take on the task of responding to this mandate. TG3 has produced a first report
in mid-2004, which states that one key factor in the consideration of appropriate
frequency bands for use by UWB is the compatibility with radio communication
services using these bands in CEPT countries. The CEPT final report to the European Commission is to be approved by the ECC in April 2005.

UWB EIRP emission

level (dBm/MHz)

Ian Oppermann

40

1.99

50
60

3.1

70
80

0.96
0

1.61
2

6
8
Frequency (GHz)

Indoor handheld
Outdoor handheld
Part 15 limit

10

12

CEPT indoor mask (proposal)

CEPT outdoor mask (proposal)

UWB EIRP emission

level (dBm/MHz)

(a)
40
45
50
55
60
65
0.96
70
75

1.99

3.1

10.6

1.61
2

6
8
Frequency (GHz)

Through-wall radar
GPR ( fc > 3.1 GHz)

10

12

GPR ( fc < 960 MHz)

Part 15 limit

(b)
Figure 1.1. UWB radiation mask defined by FCC and one early CEPT proposal.

The CEPT has already undertaken a number of compatibility studies which

indicate that limits for emissions from UWB devices for communication applications would need to be lower than the FCC limits in order to avoid harmful
interference to some of the radio communication applications in CEPT countries.
It is generally expected that ETSI/CEPT will follow some of the FCCs recommendations but will not necessarily directly adopt the FCCs regulations, now out of
date!
Figure 1.1 shows an early proposal for CEPT/ETSI mask limits as well as the
agreed FCC masks. Figure 1.1(a) represents the masks for data communication
applications for indoor and outdoor uses. Figure 1.1(b) gives the FCC radiation
mask for radar and sensing applications. In all cases, the maximum average power
spectral density follows the limit of FCC Part15 regulations [12]. Therefore, the
limit will depend on compatibility studies results made by CEPT/TG3.
1.3.3. UWB regulation within ITU
The ITU-R Study Group 1 has established Task Group 1/8 [13] to carry out relevant studies related to the proposed introduction of UWB devices and the

Introduction

implications of compatibility with radio communication services. The mandate

of TG 1/8 is focused on the study of
(i) compatibility between UWB devices and radio communication services
(Question ITU-R 227/1);
(ii) spectrum management framework related to the introduction of UWB
devices (Question ITU-R 226/1);
(iii) appropriate measurement techniques for UWB devices.
TG1-8 is to complete its work in 2005 and will present to SG1 four new
recommendations on UWB characteristics, UWB compatibility with other radio
services, UWB spectrum management framework, and UWB measurement techniques, respectively.
1.3.4. UWB standardization in the IEEE
1.3.4.1. IEEE 802.15.3a
The IEEE established the 802.15.3a Study Group to define a new physical layer concept for short-range, high-data-rate applications. This alternative physical (ALT
PHY) layer is intended to serve the needs of groups wishing to deploy high-datarate applications. With a minimum data rate of 110 Mbps at 10 m, this study group
intends to develop a standard to address such applications as video or multimedia
links, or cable replacement. Whilst not specifically intended to be an UWB standards group, the technical requirements lend themselves very much to the use of
UWB technology. The study group has been the focus of significant attention recently as the debate over competing UWB physical layer technologies has raged.
The work of the study group also includes analyzing the radio channel model proposal to be used in the UWB system evaluation.
The purpose of the study group is to provide a higher-speed physical layer
(PHY) candidate for the existing, approved 802.15.3 medium access standard. The
targeted applications are those which involve imaging and multimedia [14]. The
main desired characteristics of the alternative PHY are
(i) coexistence with all existing IEEE 802 physical layer standards;
(ii) target data rate in excess of 100 Mbps for consumer applications;
(iii) robust multipath performance;
(iv) location awareness;
(v) use of additional unlicensed spectrum for high-rate WPANs (wireless
personal area network).
1.3.4.2. IEEE 802.15.4
The IEEE established the 802.15.4 Study Group to define a new physical layer concept for low-data-rate applications. The IEEE 802.15 TG4 is chartered to investigate low-data-rate solutions for very low power and very low complexity systems.
It is intended to operate in unlicensed, international frequency bands. Potential
applications are sensors, interactive toys, smart badges, remote controls, and home
automation.

Ian Oppermann

One option examined by the working group is utilizing UWB technology at

the air interface. The study group addresses new applications which require not
only moderate data throughput, but also require long battery life such as low-rate
wireless personal area networks, sensors, and small networks.
1.3.5. The multiband OFDM alliance (MBOA)
The MBOA is an alliance of industry and academic partners which aims to develop
physical layer and medium access layer technologies for UWB based on OFDM
techniques [15]. The stated mission of the MBOA is to develop the best overall
solution for UWB-based products in compliance with worldwide regulatory requirements as well as to ensure coexistence with current and future spectrum users
[15].
In practice, the MBOA is formed in response to the impasse which existed in
the IEEE 802.15.3a standardization attempt. The MBOA is still heavily involved
in promoting its particular standard within the IEEE, but is moving to develop
product for the market irrespective of the position of the IEEE standardization
process.
1.4. Applications of UWB
Even with the significant power restrictions, UWB holds enormous potential for
wireless ad-hoc and peer-to-peer networks. One of the major potential advantages
in impulse-radio-based systems is the ability to trade data rate for link distance by
simply using more or less concatenated pulses to define a bit. Without dramatically
changing the air interface, the data rate can be changed by orders of magnitude
depending on the system requirements. This means however that high-data-rate
(HDR) and low-data-rate (LDR) devices will need to coexist. The narrow timedomain pulse also means that UWB oers the possibility for very high positioning
accuracy. However, each device in the network must be heard by a number of
other devices in order to generate a position from a delay or signal angle-of-arrival
estimate. These potential benefits, coupled with the fact that an individual lowpower UWB pulse is dicult to detect, oer some significant challenges for the
medium access (MAC) design.
Figure 1.2 shows the area of application for UWB with respect to existing and
future networks. The low-power restricts UWB to very short-range high-data applications, or very low data rates for moderate-range applications. The power restrictions on transmission eectively prohibit UWB from most outdoor applications. The increasing trend of users to combine both cellular and ad hoc technologies (e.g., 802.11b WLAN and 2G cellular) is a strong indicator for the inclusion
of high-speed short-range wireless in a comprehensive picture of future wireless
networks.
UWB systems have been targeted at very HDR applications over short distances, such as USB replacement, as well as very LDR applications over longer
distances, such as sensors and RF tags. Classes of LDR devices are expected to

10

Introduction
Wide-area network (WAN)
-Large coverage
Personal area network (PAN)
-High cost
-Connectivity
-Cable replacement
-Low cost

Walk

4G

2G cellular

Outdoor

Walk
Stationary

Indoor

Mobility

Vehicle

3G
cellular

HiperLAN/2

Local area network access

-Hot speed
-High speed
-Moderate cost

Stationary
desktop
0.1

10

100

User bit rates

Mbps

Figure 1.2. Role of UWB in future systems.

have very low complexity and very low cost. The expected proliferation of lowcost UWB devices means that solutions must be devised for devices to coexist or
preferably interoperate with many dierent types of UWB devices with dierent
capabilities. The complexity limitations of LDR devices may mean that very simple
solutions are required. HDR devices, which are expected to have higher complexity, may have much more sophisticated solutions.
Figure 1.3 shows applications for UWB as perceived by members of the European Union project PULSERS [16]. These scenarios cover wireless local area
networks (WLANs), wireless personal area networks (WPANs), sensor networks,
and short-range peer-to-peer networks.
1.5. Impulse radio schemes
1.5.1. Impulse radio UWB
Time-modulated ultra-wideband (TM-UWB) communication is based on discontinuous emission of very short pulses or pulsed waveforms (monocycles), for example, one of those presented in Figure 1.4. Impulse-radio-based UWB has much
in common with spread-spectrum systems. A major dierence however is that the
spread-spectrum chip is replaced by a discontinuous UWB pulse or monocycle transmission. The UWB pulse employed determines the bandwidth occupied
by the signal.
Each pulse has the ultra-wide spectrum in the frequency domain (as presented
in Figure 1.5). This type of transmission does not require the use of additional
carrier modulation as the pulse will propagate well in the radio channel. The technique is therefore a baseband signaling technique. This radio concept is referred as
impulse radio (IR).
One transmitted symbol is spread over N pulses to achieve a processing gain
which may be used to combat noise and interference. This is similar to the approach used for spread-spectrum systems.

Audio

TV

.
.
Printer

PDA

UWB connection

Access box

PDA

Ad hoc
connection option

Wireless body area

network (WBAN)

Intelligent wireless
area network (IWAN)

Outdoor peer-to-peer
networking (OPPN)

Internet

Wide-area
cellular
network

Temperature
sensor

Home
controller

Position
sensor

Light
sensor

Position sensor

Sensor, positioning, and

identification network (SPIN)

Motion
sensor

Position
sensor

Wireless bridge

Wide-area cellular
network (alarms,
remote control)

Figure 1.3. Applications for UWB (source: PULSERS White Paper [16]).

Hight-data-rate
data exchange (gaming)

PDA

3G core
Opt. network
PDA

Internet Access
fix line

Opt.

Wireless
bridge

Laptop
Digital
computer camera

Desktop
computer

Hot-spot wireless personal

area network (WPAN)

PDA

DVD

Monitor
Camcorder

Gateway

Ad hoc application scenarios

Position
sensor

Position sensor

Ian Oppermann
11

12

Introduction

Normalized amplitude

1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
1

0.5

0.5

Time (ns)
Tp = 0.25 ns
Tp = 0.5 ns
Tp = 0.75 ns

Tp = 1 ns
Tp = 2 ns

Figure 1.4. Gaussian monocycle in time domain.

The UWB pulse waveform can be any function which satisfies the spectral
mask regulatory requirements. Common pulse shapes used in theoretical studies
include Gaussian, Laplacian, Rayleigh, or Hermitean pulses. In practical systems,
exact pulse shapes are extremely dicult to generate due to antenna and radio frequency matching problems. Only coarse approximations of the theoretical waveforms are available in practice.
Data modulation is typically based on simple modulation schemes such as
pulse position modulation (PPM) and/or pulse amplitude modulation (PAM)
[17]. A commonly proposed UWB receiver is a homodyne cross-correlator, that
is, based on the architecture that utilizes a direct RF-to-baseband conversion. Intermediate frequency conversion is not needed, which makes the implementation
simpler than in conventional (super-)heterodyne systems.
Unlike spread-spectrum systems, the pulse (chip) does not necessarily occupy
the entire chip period. This means that the duty cycle can be extremely low. The
receiver is only required to listen to the channel for a small fraction of the period
between pulses. The impact of any continuous source of interference is therefore
reduced so that it is only relevant when the receiver is attempting to detect a pulse.
This leads to processing gain in the sense of a spread-spectrum systems ability to
reduce the impact of interference.
The nominal centre frequency and the bandwidth of the monocycle (see
Figure 1.5) depends on the monocycles width. The 3 dB bandwidth is approximately 116% of the monocycles nominal centre frequency f0 = 1/ p [6]. Considering an example monocycle duration of 0.75 nanosecond (as shown in Figure
1.4), the nominal centre frequency is 1.33 GHz and the half-power bandwidth
is 1.55 GHz. The spectrum of a Gaussian monocycle is asymmetrical, as seen in
Figure 1.5.

Ian Oppermann

13
1

0.8

PSD

0.6

0.4

0.2

0
0

0.5

1.5

2.5

3.5

Frequency (GHZ)
Tp = 0.25 ns
Tp = 0.5 ns
Tp = 0.75 ns

Tp = 1 ns
Tp = 2 ns

Figure 1.5. Gaussian monocycle in frequency domain and spectra of the Gaussian monocycle.

Impulse radio transmitter

Modulation

Impulse radio receiver

Pulse
generator

Crossgenerator

Variable
delay

Variable
delay

Code
generator

Code
generator

Time
base

Time
base

LPF

BPF

DSP

Modulation
output

Figure 1.6. Block diagram of the TH-PPM UWB impulse radio concept by Time Domain Corporation.

Figure 1.6 presents a block diagram of a time-hopping UWB impulse radio

concept that utilizes pulse position modulation (concept by Time Domain Corporation, USA).
1.5.2. Fast stepped frequency chirps
Both the bandwidth and spectral mask limitations must be satisfied by the UWB
signal. There are many means of generating a very wideband signal. One technique is based on fast frequency chirps, which are commonly used in impulse

14

Introduction
C(1)

cos(2 f1 t)

cos(2 f1 t) g1

C(2)

cos(2 f2 t)

cos(2 f2 t) g2

C(n) ..
.

cos(2 fn t)

cos(2 fn t) gn

LPF
LPF
.
.
.
LPF

f1 f2

fn

Figure 1.7. Block diagram and spectrum for multicarrier CDMA system.

radar applications. It is possible to generate a wideband transmission by sweeping the transmitters oscillator in the frequency domain. A bandwidth of several
hundred MHz can be achieved with 10 nanosecond sweep time [18]. Wider
bandwidths can be achieved using this technique, for example, ground penetrating radar (GPR) systems with 501200 MHz bandwidth have been documented in
[19].
GPR systems based on UWB technology are suitable for object detection in,
for example, landmine sweeping and avalanche rescue operations, because of the
good signal penetration ability and fine spatial resolution.
1.6. Multicarrier schemes
Another approach to producing an UWB signal is to extend the techniques utilized
for direct-sequence spread-spectrum or code-division multiple-access (CDMA)
schemes that are used for third generation mobile systems. Wideband CDMA systems with optional multicarrier (MC) techniques can be used to fill the available
spectral mask.
There are three main techniques to generate an SS-MC transmission; multicarrier CDMA, multicarrier DS-CDMA, and multitone (MT) CDMA [20]. Each
of these techniques relies on, and benefits from, the properties of conventional
spread-spectrum signals. However, multicarrier systems are reasonably complex
to implement. In particular, multicarrier systems require several mixers or digital
fast Fourier transform (FFT) techniques to place the dierent signal components
in the required bands.
Figure 1.7 shows a block diagram for an MC-CDMA system. The original data
stream is spread over the dierent subcarriers fi with each chip of pseudo-random
(PR) code Ci . The spectrum spreading is done in the frequency domain [20]. The
signal is despread at the receiver using corresponding chips gi of the spreading
code C. In UWB applications, the individual modulated carrier needs to fulfil the
500 MHz bandwidth requirements.
Figure 1.8 shows a block diagram for a multicarrier DS-CDMA system. This
method involves spreading the original data in the time domain after serial-toparallel conversion of the data stream [20]. The system needs to obtain the 500
MHz minimum bandwidth requirement to be treated as UWB.

Ian Oppermann

15

C(t)
Serial-to- C(t)
parallel
.
converter C(t) ..

cos(2 f1 t) C(t)

cos(2 f1 t)
cos(2 f2 t)

LPF

cos(2 f2 t) C(t)

cos(2 fn t) C(t)

cos(2 fn t)

.
..

LPF

Parallel-toserial
converter

LPF

f1 f2

fn

Figure 1.8. Block diagram and spectrum for multicarrier DS-CDMA system.
C1
Serial-to- C2
parallel
converter Cn ..
.

cos(2 f1 t)

cos(2 f1 t)

cos(2 f2 t)

cos(2 f2 t)
cos(2 fn t)

cos(2 fn t) .
.
.

RAKE
combiner 1
RAKE
combiner 2
RAKE
combiner n

Parallel-toserial
converter

f1 f2 fn
f
Figure 1.9. Block diagram and spectrum for multitone CDMA system.

Figure 1.9 shows the block diagram of a multitone-CDMA system. The bandwidth of an MT-CDMA system is smaller than in the previous multicarrier systems because of the small subcarrier spacing. This kind of multicarrier approach
is the closest to the original UWB concept. However, it causes the highest selfinterference due to the overlapping spectra.
Multicarrier technology is currently used in high-data-rate applications, for
example, in WLAN systems such as Hiperlan2, digital audio or video broadcasting
(DAB and DVB, resp.) and in asymmetric digital subscriber line (ADSL).
The advantage of multicarrier technologies over single carrier systems is that
the data rate in each subcarrier is lower than for single carrier systems. This eases
synchronization with the spreading sequence at the receiver, and helps to avoid intersymbol interference (ISI). The disadvantage is the increased complexity of the
receiver requiring either multiple mixing stages or fast Fourier transform processing. MC-CDMA schemes spread the signal in the frequency domain. The signal
consists of overlapping, relatively narrow carriers which fill the available UWB
signal spectrum. MC-DS-CDMA and MT-CDMA schemes apply spreading in the
time domain. MT-CDMA has a similar bandwidth to a basic DS-CDMA scheme
[20] with relatively small separation of the tones used ( f1 to fn in Figure 1.9). Consequently, the spreading factor employed for MT-CDMA schemes are much higher
than for the MC schemes. The increased spreading factor increases the processing
speed required at the receiver.
The conventional DS-SS technique without multicarrier properties can also
be characterized as an UWB technique if the chip rate is high enough. This calls

16

Introduction
1

Power spectrum magnitude (dB)

0
1
2
3
4
5
6
7
8
9
10

7
Frequency

10

11

109

Figure 1.10. Multiband frequency plan and spectral planGroup A.

for extremely fast digital signal processing which may be impractical. Issues such
as synchronization will also be a significant challenge.
1.6.1. Multiband UWB
A recent proposal within the IEEE 802.15.3 Working Group for UWB signals [14]
utilizes overlapping groups of UWB signals which each has a bandwidth of approximately 500 MHz. Whilst not currently the leading proposal, this technique oers
an interesting insight into the potential use of the very large spectrum available.
This so-called multiband UWB ensures adherence to the FCC minimum bandwidth requirements and allows ecient utilization of the available spectrum.
Figure 1.10 shows the spectrum plan for the first group of UWB signals
(Group A). The subbands are spaced 470 MHz apart, and any number of 500
MHz signals may be utilized. This allows for flexible coexistence with existing
communications systems (such as WLAN systems). Each subband is generated by
a pulse with 10 dB bandwidth of 520 MHz. Figure 1.11 shows the second group
of UWB signals (Group B) which overlaps the first group of UWB signals by
235 MHz. This potentially enhances the systems flexibility with respect to coexistence, interference mitigation, and multiple access.
Each group of UWB signals is divided into lower sets (subbands 18) and
upper sets (subbands 915). Only 7 subbands are used in the lower set, which
means that one subband can be avoided for coexistence. The upper set is used in
conjunction with the lower set to increase the bit rate.
1.6.2. Multiband OFDM
A hot candidate for standardization in the IEEE 802.15.3a Group is multiband
OFDM as supported by the MBOA. The technique proposed by the MBOA is to

Ian Oppermann

17
Spectral plangroup A + group B

Power spectrum magnitude (dB)

0
1
2
3
4
5
6
7
8
9
10

Frequency

10

11

109

Figure 1.11. Multiband frequency plan and spectral plan of two overlapping frequency groups
Group A and Group B.

Channel 1 Channel 2 Channel 3

3168
MHz

3696
MHz

4452 . . .
4224
MHz
MHz
Frequency

Figure 1.12. MBOA channel allocation scheme.

utilize OFDM techniques to produce a signal (channel) which is approximately

528 MHz wide. The signal structure is shown in Figure 1.12. The OFDM signal
can be produced using standard FFT techniques [21].
1.7. Conclusions
The fact that UWB technology has been around for so many years and has been
used for a wide variety of applications is a strong evidence of the viability and flexibility of the technology. The simple transmitter and receiver structures which are
possible make this a powerful technology for low-complexity, low-cost communications. As will be discussed in later chapters, the physical characteristics of the

18

Introduction

signal also support location and tracking capabilities of UWB much more readily
than in existing narrowerband technologies.
The severe restrictions on transmit power, allowing less than 0.5 mW maximum power over the 7.5 GHz band in the FCC emission mask, have substantially
limited the range of applications of UWB to short-distance high-data-rate, or lowdata-rate, longer-distance applications. The great potential of UWB is to allow
flexible transition between these two extremes without the need for substantial
modifications to the transceiver. On the other hand, UWB technology emission
masks outside the USA are still under discussion pending the outcome of compatibility studies with other radio services.
Whilst UWB is still the subject of significant debate, there is no doubt that the
technology is capable of achieving very high data rates and is a viable alternative
to existing technology for WPAN; short-range, high-data-rate communications;
multimedia applications, and cable replacement. Much of the current debate centers around which PHY layer(s) to adopt, development of a standard, and issues of
coexistence and interference with other radio services.
At the time of writing, no standard exists for a MAC which has been specifically developed for UWB which will support all of the unique advantages of UWB.
The potential of UWB systems to oer flexible data rates, support coexistence, enable large numbers of users and positioning services are critically dependent on
the capabilities of the MAC.
Research is ongoing in industry and academic consortia including the IEEE,
MBOA, and in European Union Projects such as PULSERS [22]. There is work
ongoing in standardization bodies to adapt, optimize, and enhance existing MACs
for use over UWB. Part of the reason for the slow progress in developing a flexible,
comprehensive MAC specifically for UWB is due to the inherent complications of
using a physical layer signal which is dicult to detect and dicult to synchronize
for intended users. Eorts have been made to develop MACs which allow ecient
operation for large numbers of devices. However, this work is still at an early stage.
Acknowledgments
The author would like to thank colleagues from CWC for input to this chapter. In
particular, thanks to Matti Hamalainen and Jari Iinatti. The author would also like
to thank colleagues from the Wireless World Research Forum (WWRF) Working Group (Short Range Communications) for their input related to the regulatory status section. In particular,thanks to Frederic Lallemand and Guy Salingue.
Thanks also to Therese Oppermann for careful proofreading.
Bibliography
[1] C. L. Bennett and G. F. Ross, Time-domain electromagnetics and its applications, Proceedings
of IEEE, vol. 66, no. 3, pp. 299318, 1978.
[2] H. F. Harmuth, Nonsinusoidal Waves for Radar and Radio Communication, Academic Press, New
York, NY, USA, 1981.

Ian Oppermann

19

[3] FCC press release, Febrauary 2002, http://www.fcc.gov/Bureaus/Engineering Technology/News

Releases/2002/nret0203.html.
[4] Federal Communications Commission (FCC), First Report and Order in The Matter of Revision of Part 15 of the Commissions Rules Regarding Ultrawideband Transmission Systems,
ET-Docket 98-153, FCC 02-48, released April 2002.
[5] J. D. Taylor, Ed., Introduction to UWB Radar Systems, CRC Press, Boca Raton, Fla, USA, 1995.
[6] Time Domain Corporation, Comments of Time Domain Corporation, Docket 98-154. In the
Matter of Revision of Part 15 of the FCCs Rules Regarding Ultra wide-band Transmission Systems, 1998.
[7] Federal Communications Commission (FCC), FCC NOI: Rules Regarding UWB Transmission
Systems, ET Docket No. 98-153, September 1998.
[8] CEPT/ERC Recommendation 70-03, Relating to the use of Short Range Devices (SRD).
[9] EC Mandate M/329 to the European Standardisation Organisations, referenced in document
RSCOM03-07.
[10] ETSI TR 101 994-1 for communication and positioning applications.
[11] EC Mandate to CEPT, Mandate to CEPT to harmonise radio spectrum use for Ultra-wideband
Systems in the European Union, referenced in document RSCOM03-40 rev. 3.
[12] Federal Communications Commission (FCC), http://www.fcc.gov, (Many standards documents
may be found on this site).
[13] I. Oppermann, F. Lallemand, and G. Salingue, WWRF White Paper, chapter 5, UWB Spectrum
Landing Zones, 2004.
[14] http://www.ieee802.org/15/pub/.
[15] http://www.multibandofdm.org/about.html.
[16] PULSERS White Paper, WWRF Meeting, Eindhoven, The Netherlands, December 2002.
[17] J. T. Conroy, J. L. LoCicero, and D. R. Ucci, Communication techniques using monopulse waveforms, in Proceedings of IEEE Military Communications Conference (MILCOM 99), vol. 2, pp.
11811185, Atlantic City, NJ, USA, OctoberNovember 1999.
[18] G. F. Stickley, D. A. Noon, M. Chernlakov, and I. D. Longsta, Preliminary field results of an
ultra-wideband (10620 MHz) stepped-frequency ground penetrating radar, in Proceedings of
IEEE International Geoscience and Remote Sensing (IGARSS 97), vol. 3, pp. 12821284, Singapore,
August 1997.
[19] L. Carin and L. B. Felsen, Eds., Ultra-Wideband, Short-Pulse Electromagnetics 2, Conference Proceedings, Kluwer Academic/Plenum Press, New York, NY, USA, 1995, 605 pages.
[20] R. Prasad and S. Hara, An overview of multi-carrier CDMA, in Proceedings of IEEE 4th International Symposium on Spread Spectrum Techniques and Applications (ISSSTA 96), vol. 1, pp.
107114, Mainz, Germany, September 1996.
[21] MBOA White paper, http://www.multibandofdm.org/, September 2004.
[22] http://www.pulsers.net/main.shtml.

Ian Oppermann: Department of Electrical and Information Engineering, University of Oulu,

FIN-90014, Finland
Email: [email protected]

UWB propagation channels

Andreas F. Molisch, Jurgen Kunisch, Robert Qiu,
Dajana Cassioli, Michael Buehrer,
Marcus Pendergrass, Istvan Z. Kovacs,
Gert F. Pedersen, Patrick C. F. Eggers,
Sumit Roy, and Iyappan Ramachandran

2.1. Introduction
2.1.1. General aspects of channel modeling
As with any other communications system, it is the channel that determines the ultimate (information-theoretic) performance limits, as well as the practical performance limits of various transmission schemes and receiver algorithms. For UWB
systems, this channel is the ultra-wideband propagation channel. Understanding
this channel is thus a vital prerequisite for designing, testing, and comparing UWB
systems. Just like UWB communications itself, channel modeling for UWB is a relatively new area. And just as the interest for UWB systems has intensified in the
last years, so has the importance of modeling the UWB propagation channels. In
this chapter, we will give a comprehensive overview of the state of the art in this
exciting area.
Quite generally, wireless channel modeling is done for two dierent purposes.
(i) Deterministic channel modeling tries to predict the behavior of a wireless channel in a specific environment. If a complete description of the
geometry, as well as of the electromagnetic properties of the materials,
of the surrounding of transmitter and receiver is given,1 then Maxwells
equations can be solved exactly, and the channel impulse response (or
an equivalent quantity) can be predicted. This approach, which had long
been deemed too complicated, has become popular in the last 15 years.
Ray tracing and other high-frequency approximations, as well as the advent of more powerful computers, have made it possible to perform the
required computations within reasonable time.
(ii) Stochastic channel models try to model the typical or canonical
properties of a wireless channel, without relating those properties to a
specific location. As the most simple example, the probability density
1

The size of the surroundings depends on the environment, as well as on the desired dynamic

range.

22

UWB propagation channels

function of the (narrowband) amplitude of the received signal is modeled as Rayleigh-distributed within a small area (part of a room). In contrast, a deterministic model would try to predict the exact amplitude
values at each location in that part of a room.
There is an enormous amount of literature for narrowband wireless channels;
for textbooks and overview articles, see [110]. However, those models cannot be
easily generalized to UWB channels, due to some important basic dierences in
the propagation processes and the resulting models. Some of these dierences are
the following.
(i) Each multipath component (MPC) can lead to delay dispersion by itself,
due to the frequency-selective nature of reflection and diraction coefficients. This eect is especially important for systems with large relative
bandwidth.
(ii) The signals are received with excellent delay resolution. Therefore, it often happens that only a few multipath components make up one resolvable MPC, and the amplitude statistics of such a resolvable MPC is not
complex Gaussian anymore. Similarly, there is an appreciable probability that areas of no energy can exist, that is, (resolvable) delay intervals
during which no significant amount of energy is arriving at the receiver.
These issues are mostly important for systems with large absolute bandwidth.
By definition, a UWB channel is a more general description of nature than a
narrowband channel: it is always possible to obtain a narrowband channel from a
UWB channel by just filtering the UWB channel. Extrapolation from narrowband
to UWB is tempting, as it allows the reuse of existing measurements. However, it is
extremely dangerous, and has to be backed up by measurement data. Every aspect
of narrowband channel modeling has to be questioned before it can be accepted
for UWB modeling!
2.1.2. Regulatory and application-specific aspects
The frequency range of UWB systems is limited by transceiver design issues (transit frequency of semiconductor components, antenna bandwidth, etc.), as well as
rulings of the frequency regulators. Channel models should be mindful of those
limitations: after all, the main purpose of channel models is the design and testing
of UWB systems, and as the systems face practical and legal limitations, there is no
need to be overly general in the models either.
Up to now, only the regulator in the USA has issued rulings that allow the
use of UWB.2 The report and order of the FCC from 2002 [11] allows unlicensed
operation of UWB communications devices in indoor environments, mainly in
the frequency range between 3.1 and 10.6 GHz. The admissible power spectral
density is 41.3 dBm/MHz, and considerably lower outside that band (the exact
2

Singapore has recently established a UWB-friendly zone, and other countries are expected to
issue rulings within the next few years.

Andreas F. Molisch et al.

23

UWB EIRP emission level (dBm)

40
45
50
55
60
65
70
75

100

101
Frequency (GHz)
Indoor limit
Part 15 limit

Figure 2.1. FCC mask for indoor communications.

mask is shown in Figure 2.1). Communications is also allowed at frequencies below 960 MHz. There are certain restrictions in the applications, as well as in peakto-average ratios; however, those have no impact on the channel modeling. It is
important that outdoor devices must not be fixed, or have a connection to power
mains-only peer-to-peer communications of portable devices is allowed. These restrictions limit the frequency range in which the channel model needs to be valid,
as well as the possible distance between transmitter and receiver.
The envisioned applications give us guidelines about what propagation environments are of greatest interest for UWB applications. A main point of emphasis
is high-data-rate personal area networks (PANs). A typical application there is the
transmission of digital TV signals via a wireless link. It follows that such high data
rates, in conjunction with the limits on the power spectral density, can be performed only over short distances (on the order of 10 m), and are of interest mostly
for residential and oce environments. Another application of interest is sensor
networks. As the data rates are much lower in this context, longer ranges are possible. Furthermore, sensor networks can be of great interest in factory environments,
where indoor distances can be larger, and the abundance of metallic reflectors
can lead to quite dierent propagation conditions.
Yet another application for UWB is emergency communications, possibly
combined with geolocation, for example, from a building that collapsed during an
earthquake, or from an avalanche. Propagation conditions in such rubble or snow
environments will obviously dier considerably from normal environments; required distance can be up to 100 m. Outdoor UWB applications, as mentioned
above, are limited to peer-to-peer communications in the USA; however, since
other countries might allow fixed-location transmitters, it is worthwhile to also
study the base-station to mobile-station scenario. Finally, communications between cars is of interest.

24

UWB propagation channels

2.1.3. Synopsis of the chapter

A first step in understanding UWB propagation has to be measurements of UWB
channels. For this task, we have two dierent types of measurement techniques at
our disposal: frequency-domain measurements, employing a vector network analyzer, give the transfer function of the channel; alternatively, the channel can be
excited with a short pulse, and a time-domain measurement of the received signal
returns the convolution of the pulse with the channel impulse response. In either
of those cases, the impact of the antenna on the measurement results is crucial.
In fact, it is much more important than in the narrowband case, because the antenna characteristics vary significantly with the frequency, and exhibit a dierent
response in dierent directions. Section 2.2 by Kunisch gives a detailed mathematical model for these eects, and suggests a procedure for the deconvolution of the
antenna eects.
As mentioned above, a crucial property of UWB channels is the fact that
each multipath component can show delay dispersion by itself. That means that
a short pulse that, for example, undergoes only a single diraction will arrive
at the receiver with a larger support (extent in the delay domain) compared to
the originally transmitted pulse. The reason for this is that the diraction coefficient is frequency-dependent. Qiu describes in Section 2.3 the mathematics of
these processes. Exact and approximate formulations, based on the uniform theory of diraction, allow a modification of the classical Turin model, so that the
impulse response in a multipath environment is now the sum of distorted, scaled,
and delayed pulses. This has important implications for the design of correlation
receivers, which are also addressed in this section. From a more experimental point
of view, Molisch and Buehrer discuss the impact of frequency-selective reflection
coecients in Section 2.5.5. The results from these sections can be combined with
classical ray tracing approaches, resulting in a deterministic channel prediction
and modeling method.
For stochastic channel modeling, we have to investigate the path loss and
shadowing, as well as the delay and angular dispersion of the channel. These two
topics are treated in Sections 2.4 (by Cassioli) and 2.5 (by Molisch and Buehrer),
respectively. While the path loss is a well-known quantity in narrowband channel modeling, Section 2.4 shows that it actually has to be treated as frequencydependent when the relative bandwidth becomes significant. On the other hand,
many narrowband measurement campaigns in dierent frequency bands can be
reused for path loss predictions. The delay dispersion properties also show
a significant dependence on the bandwidth. While the rms delay spread shows
only a weak dependence on the bandwidth, the amplitude statistics and the statistics of the arrival times of the multipath properties strongly vary with the bandwidth, as well as with the center frequency of the considered signal. All the different aspects of stochastic channel modeling had an impact on the standardized
IEEE 802.15.3a channel model, which is described by Pendergrass in Section 2.6.
It was used in the downselection process for the standard for the physical layer
of high-data-rate UWB communications, and is therefore restricted to modeling

Andreas F. Molisch et al.

25

short-range communications in indoor oce environments. More general channel models are currently under development within the IEEE 802.15.4a group,
but not yet finalized at the time of this writing, and therefore not further treated
here.
All of the above considerations are for a channel without human presence
measurement campaigns (as well as the models based on them) usually try to avoid
the influence of human bodies on the measurement results as much as possible.
This assumption might be realistic, for example, for an antenna mounted on a
DVD-player, which establishes a video link. However, it becomes untenable for
body-area network, where the antenna is placed directly on a body of a user
antenna and body become inseparable. This case requires new measurement techniques, and also new channel models. Kovacs, Pedersen, and Eggers describe measurement setups that are especially suitable for this situation (see Section 2.7). A
main diculty is the impact of cables between antenna and network analyzer (or
oscilloscope); this problem is solved by using RF-over-fiber.
Finally, Section 2.8 by Roy deals with the topic of channel estimation. While
it is clearly similar to the problem of measuring channels, there are also important
conceptual dierences. Channel estimation always must be viewed in light of the
system it is being performed for, while channel measurement and modeling tries
to remain as system-independent as possible. Also, channel estimation has to use
a signaling structure that is similar to the signaling (modulation) of the payload
datacost considerations do not allow the building of an additional transceiver
with a dierent structure, just to obtain channel estimates. Due to these reasons,
the estimation techniques for OFDM-based UWB systems on one hand and for
impulse-radio and direct-sequence CDMA systems on the other hand are quite
dierent. Both are treated in Section 2.8.
The state of the art of UWB channel modeling is by now suciently advanced
that it merits a review such as the one given in this chapter. The main problems
and challenges for UWB channel models have been identified, and generic model
structures can be established. However, we have to recognize that UWB channel
models are still far less comprehensive than narrowband modelsthis is no wonder when we compare the duration of UWB channel research with the more than
30 years that have been spent on narrowband channels. A main gap in our knowledge is the parameterization of the models in dierent environments. Up to now,
almost all measurements have been performed in indoor oce and residential environments. Those measurements give a first indication of occurring parameters,
though important questions (angular dispersion, what is the best distribution to
describe fading statistics?, how do the results depend on the measurement bandwidth and center frequency?) need more thorough investigation. Parametrization
for other environments, like factory halls, disaster areas, and so forth, are completely unknown, and many measurements will be required to obtain statistically
reliable parametrizations. It is thus safe to say that the understanding of UWB
channels has made some important advances, but much more work is required
before we can claim to have a thorough understanding of this important communications medium.

26

UWB propagation channels

2.2. Measurement techniques

2.2.1. Introduction
In this section, we will give a review of the measurement techniques that can be
used to characterize UWB radio channels and discuss some of their properties.
We give analytical expressions for the transfer functions of some elementary UWB
radio channels. A particular problem for UWB measurements is the removal of antenna influences from radio channel measurements (de-embedding), which needs
to account for specific UWB properties of antennas. For that purpose, characteristic functions are defined that describe the UWB transmission and reception
properties of an antenna. A general relationship that exists for these functions is
derived, to finally arrive at a formulation of the de-embedding problem suited for
ultra-wide bandwidths.
What is a radio channel, in the first place? In this section, we will consider
any arrangement of two or more antennas, of which at least two are coupled by far
field radiation, a radio channel. If the total number of connectors of all antennas is
N, the corresponding radio channel can be viewed simply as an N-port. This definition covers the cases of SISO, SIMO, and MIMO channels. This definition also
includes the antennas into the channel, and consequently changing the antenna
characteristic may change the properties of the channel as well. Therefore, a radio
channel can be viewed as the result of combining antennas and propagation.
The inclusion of antennas into the radio channel definition follows practical
considerations regarding system design to some extent: all radio systems (including those for measuring radio channels) need to use some form of antenna, and in
many cases the intended application imposes constraints on the antennas that can
be used for a particular systemfor example, handset antenna designers have to
cope with strong restrictions related to the form factor of the handset.
This poses the problem that results obtained for a particular radio channel
using a certain measurement setup in the strict sense only apply to channels incorporating the antennas used during that measurement. To use such results for
systems with dierent antennas, some further steps are required. Consequently, it
is frequently required to consider a channel concept that is independent of the employed antennas. We will refer to this type of channel as a propagation channel;
its ports are not related to feed lines of antennas, rather they are directly related to
the propagating fields.
Deriving propagation channel characteristics from measured radio channel
data is a de-embedding problem which in principle can be solved if sucient
measurement data and a complete characterization of the antennas are available.
However, in many cases the eort to obtain such a complete characterization is
too high; that is, the eects of changing antennas cannot be easily assessed. In this
case, some approaches are in common use. The most obvious way is to use the
antennas of the target system for measurementssuch in vivo measurements save
the need for recalibration due to dierent antennas. Unfortunately, for various
reasons, such measurements are often not possible.

Andreas F. Molisch et al.

27

Another approach employs antennas that are neutral in the sense that radio
channel measurements with these antennas do not disturb the characteristics of
the underlying propagation channel significantlyso these may be recovered easily. Isotropic antennas would make a good choice for this type of measurement;
however, isotropic antennas exist only as a theoretical device. A frequent compromise is the use of omnidirectional antennas. Using this type of antenna at the
transmitter, for example, allows to determine unbiased azimuth-power spectra
at the receiver in the horizontal plane.
All of the above considerations are not specific to UWB radio channel measurements. However, what is specific to UWB conditions is the particular significance of frequency dependence. While it is often feasible to assume the radiation pattern to be constant over the band for (non-ultra-) wideband systems (e.g.,
WCDMA with a fractional bandwidth of approximately 1/400), the frequency dependence of antennas must not be neglected in UWB applications. Furthermore,
the UWB radio channel may exhibit certain frequency dependencies even for frequency-independent antennas, which is due to the fact that an antenna cannot
be frequency-independent in both transmission and reception; which in turn is
caused by constraints imposed by reciprocity. Before we look into this and the implications of these frequency dependencies for the formulation of the de-embedding problem, we first give a short review of some basic measurement approaches
that can be used for UWB radio channel measurements.
2.2.2. Basic measurement approaches
An important assumption for the characterization of mobile radio channels is linearity. An exhaustive treatment of time-variant linear systems has been given by
Bello [12, 13] in terms of various system and correlation functions that are related
by Fourier transforms. The basic measurement approaches can be directly related
to several of these system functions. For example, following [12], the response w(t)
of a time-variant linear system to a stimulus z(t) can be written as


w(t) =

z(t )g(t, )d =

z(t )h(t , )d,

(2.1)

where g(t, ) and h(t, ) are the input delay-spread function and output delayspread function, respectively. Using an input signal that is localized in time, for example, an impulse z(t) = (t t0 ), the output signal w(t) = g(t, t t0 ) = h(t0 , t t0 )
is directly related to these system functions. Occasionally, this type of measurement is referred to as a time-domain measurement. A common measurement
setup consists of an impulse transmitter and a digital sampling oscilloscope (DSO)
as receiver. This technique is intuitively appealing as a UWB measurement technique due to its conceptual simplicity in dealing with time-variant UWB channels.
The downside of this approach, however, is that calibration of such a measurement
setup requires additional eort that much aects its simplicity. Furthermore, care
has to be applied in considering possible rms averaging of oversampled signals

28

UWB propagation channels

performed inside the DSO, as this may aect the amplitude statistics of the signal
finally delivered by the DSO.
In a similar fashion, for an input signal Z( f ) that is localized in frequency
domain, for example, Z( f ) = ( f f0 ),3 the input-output relationships


W( f ) =

Z( f )H( f , )d =

Z( f )G( f , )d

(2.2)

allow to determine W( f ) = H( f , f f0 ) = G( f0 , f f0 ), where H and G are the

input Doppler-spread function and output Doppler-spread function. These equations are useful for channels that are discrete in Doppler domain, that is, channels
with quasiperiodic time variance, with only a few fundamental Doppler frequencies. For example, consider a single ray that passes through a fluorescent tube: the
oscillating plasma inside the tube introduces a Doppler shift of multiples of twice
the local power line frequency. Using a tunable CW source as transmitter and a
DSO as receiver, the Doppler behavior of this type of channel can be accurately
characterized over ultra-widebands.
The time-variant transfer function


T( f , t) =

g(t, ) exp( j2 f )d

(2.3)

and the associated input-output relationship



w(t) =

Z( f )T( f , t) exp( j2 f t)df

(2.4)

are useful for frequency-domain measurements. For Z( f ) = ( f f0 ), we have



w(t) = T f0 , t exp j2 f0 t ,

(2.5)

and for slowly time-varying channels, that is,



T( f , t) T f , t0

for t t0  <

1
,
f0

(2.6)

we get


w(t) T f0 , t0 exp j2 f0 t ,

(2.7)

3Note that to obtain a real signal z(t) R, Z( f ) has to exhibit complex conjugate symmetry with

respect to the origin; for example, Z( f ) = ( f + f0 ) + ( f f0 ).

Andreas F. Molisch et al.

29

leading to4


 

W( f ) T f0 , t0 f f0 = T f0 , t0 Z( f ).

(2.8)

Equation (2.8) is the quasistatic equation underlying vector network analyzer

(VNA) measurements (of S21 for a two-port radio channel). While this technique
has the major advantage that calibrated measurements of even ultra-wide bandwidths are more easily performed, this method has the serious drawback of being applicable only to quasistatic channels. At first glance, condition (2.6) seems
to be not too restrictive; however, for traditional VNA architectures that employ
IF filtering, f0 has to be replaced by the IF bandwidth which will result in much
stronger constraints for the permissible time variance of the channel. Furthermore, the channel must not change during a sweep over the entire band. For example, with an IF bandwidth of 300 Hz and N = 1601 frequency points, sweep
times in the order of several seconds result. A further constraint for the maximum
sweep rate is related to the delay introduced by the channel. Note that even for
time-invariant channels (T( f , t) = T( f )) and under ideal noise-free conditions,
(2.8) is only an approximation due to the finite-time interval TM available to obtain a transfer ratio for a frequency point. For continually swept measurements,
the finite sweep rate df /dt will contribute another error caused by changing the
stimulus frequency during the measurement of a frequency point, thus resulting
in T( f ) eectively getting averaged over f TM df /dt.
Both time-domain and frequency-domain measurements are subject to errors
caused by narrowband interference. Another measurement technique that is more
robust against narrowband interference employs
factorizing of a signal z(t) into a

correlation product z(t) = q() ,t q()  q()q( + t)d, where ,t denotes
the correlation operator. Assume that q has the finite support [T/2, T/2].5 For an
input signal centered around t0 , z(t t0 ), (2.1) can be rewritten as


(t) =

q() q + t t0 g(t, )d d.

(2.9)

Due to the assumed support of q, the inner integral can be nonzero only for |t
t0 | T. Then, if g(t, ) is slowly time-varying such that g(t, ) g(t0 , ) for
|t t0 | T,


w(t)
=

 

 

q() q + t t0 g t0 , d d


q( t) q t0 g t0 , d d

= q()

p , t0 ,

(2.10)

p(,t0 )

(2.11)

,t
4Assuming that an estimate of W( f ) can be determined from (2.7) during a reasonably short
interval of time such that the approximation (2.6) holds.
5Then the support of z is a subset of [T, T], that is, z is localized around t = 0.

30

UWB propagation channels

where p(, t0 ) is the response of the channel to the shifted correlation factor q(
t0 ). Equation (2.11) states that the channel response w(t) to the input signal z(t
t0 ) can approximately be determined by transmitting the shifted correlation factor
q( t0 ) instead of z(t t0 ) and then performing a correlation of the channel
response with the correlation factor q().
The correlation method has two major advantages. First, due to the correlation at the output of the channel, interfering signals can be suppressed to some
extent. Second, instead of using an impulse excitation (like it is the case for timedomain measurements), by proper selection of the correlation factor q, it is possible to significantly reduce the crest factor6 of the transmitted signal. This can
considerably relax the linearity requirements for the measurement setup (in particular, power amplifiers). Commonly used signals are, for example, chirps and
PN-sequences.
A drawback of the correlation method, however, is that the correlation factors can be much less localized than the original signal. Therefore, the maximum
achievable measurement rate is reduced, and the amount of permissible time variance is even lower.
2.2.3. The de-embedding problem
To assess the de-embedding problem associated with the transition from a radio
channel to a propagation channel more specifically, what is needed in the first
place is a clear definition of the propagation channel and its relation to the radio
channel. Unlike the radio channel, where, for example, the antenna connectors
provide convenient ports, no such obvious choice is available for the propagation
channel a priori. In this section, we approach the problem by giving a simple analytical expression for the transfer function of the UWB multipath radio channel
that provides a natural separation of antenna and propagation terms. A particular
aspect which is of paramount importance for UWB channels is that both antenna
and propagation terms have to include phases in order to preserve coherence over
the entire band.
We start by introducing a generic antenna model by defining two complex
characteristic functions A and h that describe the spatiotemporal transmission and
reception characteristic of an antenna.7 As will become clear below, these functions
are closely related to the familiar concepts of radiation pattern and eective length.
2.2.3.1. Antenna model
The antenna definition used here comprises two elements: (a) a material distribution (r, ), (r, ) which diers from free space only within a finite volume; (b)
a port defined by a reference plane R on a waveguide (the antenna feed line). One
part of the feed line up to the reference plane is considered part of the antenna. In
6Ratio of peak-to-average power.
7In the remainder of this section, a bold typeface will be used to denote vectors; a caret  on top

of a vector denotes a unit vector: r  r/ r.

Andreas F. Molisch et al.

31

the vicinity of the reference plane, the feed line is considered to be uniform such
that at the reference plane power wave amplitudes a and b (corresponding to waves
approaching and leaving the antenna, resp.) may be defined by
Etan =


n

Htan =

Zn en (x, y) an e jkn z + bn e+ jkn z ,

Yn hn (x, y) an e jkn z bn e+ jkn z ,

(2.12)

where en , hn , and Zn = 1/Yn R are the normalized electric and magnetic

fields tangential to the reference plane and feed-line
impedance for mode n, with

hn = z en and normalization such that en em = m,n , where the integration
extends over the cross-section of the feed-line. The positive z-direction points towards the antenna and z = 0 on R. In the following, a single mode n is considered
and for simplicity the index n is dropped. If necessary, corresponding results for
several modes may be superimposed, as the modes on the considered feed-line are
orthogonal.
At the reference plane, the antenna acts as a source according to b = bi + a
with the source strength bi evoked by an incident field and the output reflection
coecient due to the antenna input impedance ZA , which in general is dierent
from the feed-line impedance Z; a is the wave amplitude (at the reference plane)
of the input signal used to excite transmission.
When transmitting, the electric field at a point r2 in the far-field of the transmitting antenna located at r1 , excited at its port by a wave a, may be given as

  e jk0 r12 

A r12 , a().
E1 r2 , = Z0
4r12

(2.13)

Equation (2.13) defines the spatiotemporal transmit characteristic A(r12 , ) as

function of the transmission direction r12 = (r2 r1 )/ r2 r1  (direction of departure) and the angular frequency ; Z0 is the free-space impedance. The dependence of A(r12 , ) on r12 is related to the radiation pattern and the dependence on
describes the (temporal) filter characteristics of the transmitting antenna.
 Under
far-field conditions, A(r12 , )r12 . Note that the normalization factor ( Z0 /4)
may be chosen arbitrarily because (2.13) is a definition (for A).
The power flux density S corresponding to the electric field of (2.13) is given
by8
 

S r2 =

8Using rms phasors.

  2
E1 r2 

Z0

  2 2
A r12  |a|
2
4r12
.

(2.14)

32

UWB propagation channels

From the power |a|2 approaching the antenna, only the amount


Pt = |a|2 1 ||2

(2.15)

actually enters the antenna. Isotropic radiation of Pt would produce a power flux
density which on comparison with (2.14) gives the relation between A and the gain
over an isotropic antenna as
 
2

 A r12 , 
Giso r12 , =
2 .

1 ()

(2.16)

For the case of a receiving antenna at position r1 , let



 exp jk0 kr
 = Ei k,

Ei r, k,

(2.17)

be the electric field at position r of an incident plane wave travelling into direction
 where k0 = /c0 is the free-space wave number for angular frequency . The
k,
corresponding source strength of the receiving antenna bi may be given as


bi () = 4h k,




 Ei r1 , k,


Z0 .

(2.18)

 ) of the anEquation (2.18) defines the spatiotemporal receive characteristic h(k,

tenna under far-field (incident locally plane wave) conditions as function of the
propagation direction k of the incident field (direction of arrival9) and the an for plane waves, we additionally require
 )k
gular frequency . Because E(r, k,
 to obtain a unique definition of h. It should be noted that h is defined
 )k
h(k,
with respect to the incident field, that is, the field without the field scattered at the
antenna. Because h linearly maps the electric field strength to the source strength
bi , h acts like an eective length operator.10
The definitions (2.13) and (2.18) of the characteristic functions involve nominal antenna positions r. For a given physical antenna, there is some degree of arbitrariness regarding the relation of the nominal antenna position r and the actual
material distribution that constitutes the antenna. Moving the nominal antenna
position with respect to the antenna body will introduce a corresponding change
in A and h.

9Note that by direction of arrival we designate the direction into which the arriving wave propa-

gates.
10Some more relations of A and h to commonly used antenna parameters are given in Appendix

2.2.B.

Andreas F. Molisch et al.

33

2.2.3.2. Transmit and receive characteristic relation

Using the reciprocity principle, it can be shown that the transmit and receive characteristics are related by (cf. Appendix 2.2.A)


 = c0 A k,
 ,
2 jh k,

(2.19)

where c0 = 1/ 0 0 . Note the minus sign in front of k in the left-hand side of

(2.19) which is due to the definition of h with respect to directions of arrival, while
A is defined with respect to directions of departure.
If (2.19) holds for all , the corresponding time-domain expression11
2

 
  
 t
 k,
h k, t = c0 A
t

(2.20)

illustrates the well-known temporal derivative relationship between transmit and

receive impulse responses [1421]. From (2.19) we immediately get the even
better-known property that, for a given frequency, the (spatial) transmission and
reception patterns are equal:




A k,
h k,
 
 
 =
 .
 
 
maxk h k,
maxk A k,

(2.21)

A commonly encountered assumption is that UWB antennas, on transmission, dierentiate the input signal. Note, however, that (2.20) only supports the
statement that the impulse responses in transmission and reception are related by
a temporal derivative, without further constraint regarding their particular form
(cf. Appendix 2.2.C).
2.2.3.3. The ideal free-space far-field UWB radio channel
In this section, we proceed by giving expressions for the transfer function of the
most simple UWB radio channel, that is, the ideal free-space far-field UWB radio
channel made up by two antennas in free space and mutually in far field. Single
scattering is assumed, that is, contributions that bounce back and forth between
the two antennas are neglected.
For two antennas (designated by indices 1 and 2), and under the above given
assumptions the transfer function


S21 () =

11We use

b2 
b2i

=
a1 a2 =0 a1

to denote time-dependent quantities.

(2.22)

34

UWB propagation channels

of the corresponding ideal free-space far-field UEB radio channel may be given in
various forms by combining (2.13), (2.18), and (2.19) as


S21 () = A1 (r12 , )h2 r12 ,

 e jk0 r12

(2.23)

r12



 e jk0 r12
= h1 r21 , )A2 r21 ,

(2.24)

r12

 
 e jk0 r12
1 c0 
A1 r12 , A2 r21 ,
j 2
r12

jk
 
 e 0 r12
2 
= j h1 r21 , h2 r12 ,
.
c0
r12
=

(2.25)
(2.26)

It is important to stress that the various vectors in this equation refer to the same
reference frame. In particular, if the orientation of an antenna changes with respect
to that frame, the corresponding A and h will change. In the likely case that A and
h are given with respect to an antenna fixed frame, the orientation of the antenna
has to be accounted for by a proper transformation of A and h.
The above equation already has the desired form: the propagation aspects (delay, attenuation) are described by the exp( jk0 r12 )/r12 term while the other terms
(including 2 j/c0 factors) are related to the antennas.
In particular, (2.25) illustrates the fact that for ideal impulse radiating antennas (where A is not a function of , cf. (2.13)) on both ends, already the most
fundamental UWB channel exhibits a 1/ frequency dependency, while for impulse receiving antennas (where h is not a function of , cf. (2.18)) on both ends,
a frequency trend is imposed. In all cases, this behavior is due to the reciprocity
relation (2.19). While for (non-ultra-) wideband conditions the influence of these
dependencies is small and may even be neglected (consider again the WCDMA
case with a fractional bandwidth of 1/400), this is not the case for the UWB
channel. The presence of these particular frequency trends can be considered an
essential property of the UWB channel.
2.2.3.4. Extension to multipath conditions
If the transmission via a given path involves some intermediate interaction with
the environment, the transfer function of that path may be modified to


 

S21 , r1 , r2 , kD , kA = A1 kD , h2 kA , G , r1 , r2 , kD , kA ,

(2.27)

where kD and kA are the directions of departure and arrival, respectively. The function


G , r1 , r2 , kD , kA

(2.28)

is analogous to a scalar Greens function and describes the propagation channel for reception at r2 (from direction kA ) when transmitting into direction kD

Andreas F. Molisch et al.

35

at r1 . For example, for the ideal free-space far-field channel, there is a nonzero
G(, r1 , r2 , kD , kA ) = exp( jk0 r12 )/r12 only for kD = kA = r12 ; and for a path containing an ideal specular reflection, G(, r1 , r2 , kD , kA ) = exp( jk0 r)/r is nonzero
if kD and kA are the directions to and from the mirror point, and r is the total path
length.
Summing over all paths, and designating the lth path with a superscript l, a
general expression for the ideal UWB multipath channel transfer function is


r1 , 
r2 =
S21 , 

 

A1 kDl , h2 kAl , G , r1 , r2 , kDl , kAl .

(2.29)

Like (2.23)(2.26), both (2.27) and (2.29) can be cast into various forms.
Equation (2.29) describes a multipath channel with a discrete spectrum of
paths. In general, however, a continuum of paths may exist and the summation
has to be replaced by an integration.
We will refer to G as the propagation channel transfer function. The above
discussion suggests to specify the ports of the (far-field) propagation channel by
 of a position and a direction, and assuming locally plane waves. As there
pairs (r, k)
is potentially an infinite number of paths between two given antenna positions r1
and r2 , even a SISO radio channel corresponds to a propagation channel with a
potentially infinite number of ports.
For a MIMO channel with M transmit antennas at positions rm
1 , m = 1 . . . M,
and N receive antennas at positions rn2 , n = 1 . . . N, the elements of the MIMO
radio channel transfer matrix are obtained from (2.29) by replacing r1 , r2 , A1 , and
n
m
n
h2 by rm
1 , r2 , A1 , and h2 , respectively. Depending on whether antennas are colocated or not, the directions of departure and arrival may have to be adapted as
well.
2.2.3.5. Formulation of the de-embedding problem
The de-embedding problem can now be posed as the problem of determination of
G from S21 , that is, solving (2.29) for G. The algebraic structure of (2.29) is similar
to that of a quadratic form; we may write (2.29) as


A k 1 ,
1 D 

 



2 ,
A1 k

  
D

r1 , 
r2 = h2 kA1 , , h2 kA2 , , . . . , h2 kAM , G
S21 , 



M

A1 k ,

(2.30)
with G = diag(. . . , G(, r1 , r2 , kDl , kAl ), . . . ).12 It is obvious that a single measurement of S21 is not sucient to determine G.
12A more general form that includes path splitting (the phenomenon that a single direction of

departure results in several directions of arrival, or vice versa) results in a nondiagonal G.

36

UWB propagation channels

A straightforward way is the use of double-directional measurements (cf.

[22]), which essentially means that individual path transfer functions S21 (, r1 ,
r2 , kD , kA ) of (2.27) are measured or estimated, from which G(, r1 , r2 , kDl , kAl ) can
be determined easily.
2.2.4. Summary
In this section, we have discussed some basic channel measurement techniques.
Furthermore, the relation between radio channels, which include antennas, and
propagation channels, which do not, has been given in terms of characteristic
functions that describe the UWB spatiotemporal transmission and reception properties of antennas. Both transmit and receive characteristics are related due to the
reciprocity principle. While the familiar fact that the spatial aspects of an antenna
are the same for transmission and reception still holds under UWB conditions,
this is not the case for the temporal behavior. In particular, the temporal filter
responses in transmission and reception are related by a temporal derivative operator. It is essential to take this into accountfor example, when de-embedding
the propagation channel from radio channel measurements.
Appendices
2.2.A. Derivation of transmit-receive relation
If two solutions (Ea , Ha , Ja , Ka ) and (Eb , Hb , Jb , Kb ) of Maxwells curl equations
E + jH = K,

(2.31)

H + jE = J

pertain to the same material distribution within a volume V (and possibly dierent materials outside), then these solutions are subject to the Lorentz reciprocity
constraint [18]

V

Ea Hb Eb Ha n
=


V

Ja Eb Ka Hb Jb Ea Kb Ha .

(2.32)

Two such solutions are constructed as follows.

Solution (a). We consider an antenna at position r1 embedded in free space.
In the reference plane R of the antenna port, electric and magnetic surface current
densities

JA = z hn Y n an ,

KA = en z Z n an

(2.33)

are impressed to excite the field of mode n travelling into the antenna with amplitude a. z is normal to R and points towards the antenna. Due to this source

Andreas F. Molisch et al.

37

arrangement, no fields are excited on the field line on the other side of the reference plane. Eectively, these surface current densities constitute an ideal wave
source for mode n of amplitude a connected to the input port of the antenna. The
field at a point r2 is then given by (2.13).
Solution (b). No current densities are impressed at the antenna port. An elementary electric dipole, that is, an electric current density


J2 = I2 l 3 r r2 v

(2.34)

is impressed at position r2 (the same as for solution (a)) in the presence of the
antenna at r1 , leading to a total field


E2 (r) = Ei2 r, r21 + Es2 r, r21 , Ei2 ,

(2.35)

where Es2 (r, r21 , Ei2 ) is the field scattered at the antenna.
The result (2.19) is obtained by inserting the above constructed solutions into
the reciprocity relation (2.32), letting V R3 , assuming vanishing fields at infinity, and considering that according to the definition of h in (2.18) the antenna
source strength bi depends only on the incident field, that is, the field the elementary electric dipole would produce in absence of the antenna,
Ei2

r1 , r21




exp jk0 r21
= j0
I2 l v r21 r21 v ,
4r21

(2.36)

and furthermore considering that the orientation v of the elementary dipole can
be chosen arbitrarily.
For V R3 , (2.32) gives


Ja Eb Ka Hb =

Jb Ea Kb Ha .

(2.37)

For the left-hand side, we have



Ja Eb Ka Hb





JA Eb KA Hb = an


R


 



z hn Y n Eb en z Z n Hb



Y n en Eb + Z n hn Hb
= an
R



  Ei2 r1 , r21

= 2an bn = 2an 4h r21
Z0




jk
r
0
21

  v
z
 r21 r21 
e

= 2an 4 j0
I2 lh r21
= 2an

4r21
Z0
jk0 r21
  v

e
I2 lh r21  .
4 j0
4r21
Z0

(2.38)

38

UWB propagation channels

For the right-hand side, we have



Jb Ea Kb Ha =

J2 E1 =


  e jk0 r12  
I2 l 3 r r2 v Z0
A r12 an
4r12

 e jk0 r12
 
= I 2 l Z0
an vA r12 .

(2.39)

4r12

Combining both sides gives

2an () 4 j0

 e jk0 r12

 v



e jk0 r21
I2 lh r21 ,  = I2 l Z0
an ()vA r12 , ,
4r21
4r12
Z0
(2.40)

and after cancelling terms



2 j0 h r21 , v = Z0 vA r12 , .

(2.41)

This equation
is valid for arbitrary v. Therefore, with k  r12 = r21 , r12 = r21 ,


Z0 = 0 /0 , and c0 = 1/ 0 0 , we get


 = c0 A k,
 .
2 jh k,

(2.42)

2.2.B. Relation to standard antenna parameters

In this section, we give a few common antenna parameters in terms of the transmit
and receive characteristic functions A and h.
The power available from the receiving antenna is given by
 i 2
b 

 2  
2
4 h k  Ei r1 , k 
Pav =
=
,
1 ||2
1 ||2
Z0

(2.43)

where matching polarization states are assumed. Hence, the well-known relation
between the eective aperture (for reception of a plane wave incident from direc
 and the gain over an isotropic antenna (for radiation into direction k)
tion k)
follows as (using (2.16) and (2.19))
  2
 
2
  Pav


4 h k 
2 A k 
2

=
=
=
Ae k =
Giso k .
2
2
S
1 ||
4 1 ||
4

(2.44)

If the antenna port is terminated with an ideal open circuit (reflection coecient 1), the open circuit voltage is given by

U O.C. = Z

2
1

bi = Z

4 Ei
h   he Ei .
1 Z0

(2.45)

Andreas F. Molisch et al.

39

From that, the eective length follows as



he

4
=
1

Z
h.
Z0

(2.46)

Finally, combining (2.43), (2.15), (2.22), (2.26), and (2.44) (in the form Ae =
4 h2 /(1 ||2 )) and setting cos = |h1 h2 |/ h1 h1 , Friis equation for the
ratio of the power available from the receiving antenna and the power actually
entering the transmit antenna [23] is obtained:
 2

 2

bi 
b i 
Ae,1 Ae,2
Pav
1
1
2
2
=
=
cos2 .
 2 
 =  2
2
2
a1  (1 ||2 )2
Pt
1 || a1  1 ||2
2 r12
(2.47)

2.2.C. Other formulations

i
The received wave
ib corresponds to a voltage wave leaving the antenna with am
plitude U = Zb . Introducing

  4h k,
 ,
hN k,

(2.48)

the receive relationship (2.18) may be written in the form





 i
U ()
 E2 r1 , k,

= hN k,
Z0
Z

(2.49)

and, in time domain,



 t
 i r1 , k,
E
U (t)    

= hN k, t t 2 
,
Z
Z
0

(2.50)

where t denotes convolution with respect to t.

Similarly, considering
that an input wave a corresponds to a voltage wave with

amplitude U + = Za approaching the antenna, from (2.13), (2.19), and (2.48),

we get



 U + ()
E1 r2 ,
e jk0 r12

=
jhN r12 ,
2c0 r12
Z0
Z

(2.51)

with various forms of the time-domain equivalent



 1 r2 , t + r12 /c0

E
1 1  
U + (t)

=
hN r12 , t t
2c0 r12 t
Z0
Z




1 1 
U + (t)
=
hN r12 , t t
2c0 r12 t
Z

1 1  
U + (t)
.
=
hN r12 , t t
2c0 r12
t Z

(2.52)

40

UWB propagation channels

Equations (2.50) and (2.52) express both the receive and transmit operations by
means of a single eective length operator hN . These equations correspond to
those given in [16]. Note that the last form of the transmission equation (2.52)
suggests that the input signal is always dierentiated before being transmitted.
Alternatively, however, instead of using a single eective length operator hN ,
we may introduce a renormalized transmit characteristic




A r12 ,

AN r12 , 
.

(2.53)

Then, from (2.13), (2.18), (2.19), and (2.53),

U (t)

= 2c0
Z

 t t
 N k,
A

N
= 2c0 A

 t
Ei2 r1 , k,

dt
Z0

 i   
r1 , k, t
 t t E2 
k,
dt,


Z0



 1 r2 , t + r12 /c0

E
1  
U + (t)

=
AN r12 , t t .

Z0

(2.54)

r12

(2.55)

Transmission equation (2.55) is equivalent to (2.52) and suggests that the input
signal is never dierentiated before being transmitted.
This is only an apparent contradiction because the actual output signal depends on two factors: the input signal and the impulse response (represented by
hN or AN ). In general, hN or AN can introduce any filter characteristic (within
constraint (2.19)). However, a few general statements regarding operations that
are always or never performed on the input signal by an antenna can be made. For
example, despite the apparent proportionality of radiated field and input voltage
in (2.55), radiation at = 0 is always impeded because A( = 0) = 0 due to (2.19)
for finite h( = 0).
2.3. Propagation effects
2.3.1. Introduction
The theoretical and mathematical foundation of UWB signal processing and system modeling is based on the interaction of a transmitted pulse with the transmission medium, for example, free space, buildings, furniture, trees, hills, rivers,
stones, and so forth. The historical logical development of the UWB propagation
processes can be conveniently divided into four stages: (1) fundamental concepts
and mechanisms; (2) high-frequency approximation techniques (GTD/UTD); (3)
transient and impulse-response approximations; (4) per-path pulse waveform distortion. The first physical cornerstone to understand the medium-interaction
mechanisms of UWB pulse propagation and shape distortion was the work of
Sommerfeld [24] on the diraction of pulse by a perfectly conducting half-plane
[24]. His solution was the first exact solution of Maxwells equations. Lamb [25]

Andreas F. Molisch et al.

41

and Friedlander [26] also studied the pulse distortion caused by diraction by
a half-plane. These solutions provided us with good pictures of understanding
the physical mechanisms of pulse diraction [2730]; however, these solutions
were impractical in engineering and applied science. Secondly, another conceptual breakthrough was the birth of the geometric theory of diraction (GTD) developed by Keller [31], based on the Lunberg-Kline series in 1949 [30]. Rather than
seeking the exact solution for Maxwells equations, GTD uses a generalization of
Fermats principle to find high-frequency harmonic Maxwell equation solutions
for many complicated objects. The uniform theory of diraction (UTD) was proposed in 1974 to complement GTD in situations where GTD was invalid [32]. The
GTD/UTD framework has become the practical framework in narrowband radio
propagation [2, 33] and will be also the foundation of our proposed framework.
Thirdly, the impulse-response waveform of the finite three-dimensional target was
approximated by a linear system theory in UWB radar in 1965 [34]. The chief value
of the impulse response concept was to sum up the radar signature of an object in
a unique manner which adds insight into the complex relationship between the
target shape and radar reflectivity. A linear system is fully characterized by its impulse response. Thus, in general, the impulse response fully describes the properties of the radar target for any frequency. Finally, this concept of impulse response
and scattering center widely used for a UWB radar was borrowed in 1995 [35], to
describe the impulse response of a generalized multipath path in UWB wireless
communications [3640]. The resultant generalized model [4143, 35, 4451] extended the widely used Turins model [52] that was first suggested in 1958. This
model already finds some applications [5360]. This section is aimed at a tutorial logical development of a unified theory based on this generalized model (also
called physics-based generalized multipath model). The starting point of this theory will be Maxwells equations. We will address all the aspects that are important
from the theory and engineering point of view.
2.3.2. UWB pulse propagation physics and mechanisms
The starting point of our unified theory that takes into account pulse distortion
is Maxwells equations. Maxwells equations are linear, as is the UWB propagation
medium. Therefore, the linear system theory and Maxwells equations are valid everywhere. We will approach our problem using exact solutions and high-frequency
GTD/UTD approximations. Although we will give exact solutions for two special
cases (half-plane and wedge), we will focus on developing the GTD/UTD framework that is practical in UWB communications.
GTD/UTD may be viewed as an asymptotic (wavelength 0 or )
theory of the solutions of Maxwells equations. The algorithms of GTD/UTD are
rather simple and lead to explicit expressions for the desired quantities. The shortwave (or high-frequency) behavior of the known exact solutions (diraction by
a wedge, cylinder, and sphere) reveals that the laws of geometric optics break
down only in narrow transition regions where diraction fields are described by
GTD/UTD coecients. Fortunately the narrow transition regions are local, that

42

UWB propagation channels

is, the diracted fields only depend on the local properties of the incident waves
and bodies in these regions. Focks localization principle (1946) [61, 33] is a cornerstone of our framework. With Focks principle, we postulate that the complex
structures can be decomposed into separated scattering centers where each scattering center is described by an impulse response hl ()of the lth path. When a ray
is distorted in pulse shape, we call this ray a generalized path to dierentiate it
from the conventional one used in Turins model [52]. Therefore, all ray fields can
be described by the ordinary geometric optics (GO) and GTD (UTD). All UWB
signals energy is transferred along these ray tubes. Let us consider an example
to illustrate how a UWB pulse travels along a generalized path. The amplitude
of a diracted ray (ud ) is proportional to the amplitude of the inducing primary
ray (ui ) at the point of incidence (r0 ). Mathematically the diraction coecient
D( j; ti ; td ) connects ui with ud [61]:




1
ud =  e jks D j; ti ; td ui ,
J

(2.56)

where J is the Jacobian of the transformation to ray coordinates, s the eikonal

defining the phase structure of the field, ti and td unit vectors of the incident and
diracted rays, k the wavenumber, and the angular frequency. In (2.56) the coecient of diraction depends on the local geometry of the body in the vicinity
of the incident ray in the case of corners and edges. Below we will use half-plane
and wedge as two examples to show the method of obtaining D( j; ti ; td ) using
GTD/UTD. Other complicated geometric structures in engineering applications
can be also worked out. For these structures, often we need trace all the significant
rays to capture the major energy. Before we show the examples, let us consider the
impact of the D(; ti ; td ) on the incident pulse waveform.
UWB pulse waveform distortion. We can model the localized diraction phenomenon as a linear system. The transfer function of this linear system is determined
by the local diraction coecient H( j) = D( j; ti ; td ) located at the point of
emanation of the ray (say r0 ) on the edge or apex. When a pulse travels through a
locally located diraction point, the pulse is modeled as passing through an equivalent linear system of H( j). This point acts as a linear filter. In reality this mathematical point is a localized region (called scattering center) [62]. To apply the
linear system theory, our observation must be far enough from the scattering
center, for the approximation conditions of the high-frequency GTD/UTD to be
valid. For a given pulse of incident waveform p(t), the pulse waveform right after r0 is r0 (t) = p(t) h(t), where h(t) is the inverse Laplace transform of H( j)
[63]. Along the tube of this diracted ray, the pulse waveform is in the form of
r0 ( 0 ) = r0 () ( 0 ), where the delay is 0 = s/c with c being the speed of
light. In other words, it follows that
r(t) = p(t)

h(t)

t 0 .

(2.57)

Andreas F. Molisch et al.

43

A linear system is distortionless if h(t) = (t). This condition is valid for Turins
model but generally invalid for the physics-based model. Turins model only takes
into account the eect of delays through ( 0 ), as included in (2.57). For example, for diraction by a perfectly conducting (PEC) half-plane, it follows from

the GTD that h() = A0 / U() [64, 65], where A0 is real and U(x) is the unit
step function. The mathematical property of this diraction phenomenon is very
far-reaching. The role of the impulse response of the PEC half-plane is the semiintegral of the waveform of the incident primary ray field, with a notation of
(d/d)1/2 or ( j)1/2 . Fractional calculus has proven to be a very good tool to
handle these problems [49, 50], see Section 2.3.A for details.
Pulse diraction by a perfectly conducting half-plane. As an example let us consider
the canonical problem of diraction by a half-plane. This problem, first solved by
Sommerfeld, is both theoretically and practically important. For monochromatic
incident field ui = A0 exp( jt), the complex-valued amplitude u of the scattered
field at the distance r is given by [33]
u( j) = u1 u2 = A0 e jkr cos(0 ) F



A0 e jkr cos(+0 ) F

2kr cos 0







(2.58)

2kr cos + 0 ,

where F(x) = (e j(/4) / ) x e j d and the plus and minus signs denote the
H and E polarization incident fields. 0 and are, respectively, the incident and
observation angles; k = /c is the wave number. The Fresnel function F(x) can
be expressed in terms of the error function erfc(x) as F(z) = (1/2)erfc(e j/4 z)

where erfc(z) = (2/ ) z ex dx. Using the Laplace transform of an error function, we can derive the closed form of the impulse response defined by h() 
L1 {u( j)/ui }:

 "

cos (1/2) 0
cos (1/2) + 0
2r/c




hhalf-plane () =
2
+ (r/c) cos 0
+ (r/c) cos + 0


1
r

U ,
c
r/c

(2.59)

where U(t) is the unit step function of t. The transient exact solution is remarkably
simple, compared with its time-harmonic frequency-domain solution (2.58). To
compare our results with others, consider the normal incidence where = 2
and 0 = . Equation (2.59) reduces to

1
1 2r sin(/2)

hnormal () =
,
c r cos c r

(2.60)

which is identical to the derivative of the unit step response results obtained by
[30]. Reference [66] also obtained similar expressions.
Equation (2.59) has two singularities: (1) when + (r/c) cos( 0 ) and +
(r/c) cos( + 0 ) approach zeros; (2) when approaches r/c. We only consider the

44

UWB propagation channels

second case where the wave front c r assumes a singularity of (1/ )U() if
the singularity of case 1 is absent. We only have one scatter center occurring at the
edge of the half-plane. If a pulse p(t) propagates through this scattering center, the
pulse waveform will be distorted since the impulse response of the half-plane is
not a Dirac function ().
In Kellers GTD model for the diraction at the edge of a half-plane [64, 65],
the monochromatic diraction wave is given as


1
ud1 = e jkr D , 0 , k ui ,
r

(2.61)

where


Ds,h , 0 , k =

exp( j/4)
1
1

 

 

cos 0 /2
cos + 0 /2
2 2k

(2.62)

is a diraction coecient obtained through the first-order asymptotic approximation of the exact solution of (2.58). This first-order solution breaks down within
the transition zone, knownas semishadow, which is defined by the region q q0
where q is defined as q = 2kr cos(( 0 )/2) and q0 is the minimum value of
q such that the approximation (2.62) is satisfactory within q q0 . With the aid
of a Laplace pair, from (2.62) the time-domain diraction coecient predicted by
Kellers GTD is derived as
hs,h () =


c



 

2 2 cos 0 /2

1
1
r
 
U .
c
r/c
cos + 0 /2
(2.63)


Pulse diraction by a perfectly conducting wedge. As another example, we consider

the wedge. The exact transient solution to a perfectly conducting wedge is available
[27] as follows (for > and  < ):
hwedge (r, r  ; )


 |r r  |/c
|r r  |/c 
U |  | +
U(  )

4 |r r |
4 |r r  |

c Q(,  , ) + Q(,  , ) 

U c D0 ,
2

4
sinh()
(2.64)

where
#

Q(,  , ) =

sin (/)(  )



2 cosh(/) cos (/)(  )


sin (/)(  + )

 ,

cosh(/) cos (/)(  + )

(2.65)

Andreas F. Molisch et al.

45

is the angle of the wedge. Polar coordinates (, , z) and ( ,  , z ) are used for
observation point and source point, respectively. Equation (2.64) assumes Neumann boundary condition on the wedge faces. Parameters , T, D0 are defined
by
= cosh1


cT = c2 2 (z z )2

%

c2 T 2 2

1/2

2

&

(2 )1 ,

D0 = ( +  )2 + (z z )2

1/2

(2.66)
.

There is only one scattering center in this case. The first two terms represent the
geometric-optics ray fields. The third term represents the diracted field caused
by an impulse plane wave. Obviously (2.64) will predict some pulse waveform distortion if a pulse signal is diracted by a perfectly conducting wedge. The transient
responses given in (2.59) and (2.64) are valid at all observation times, thereby providing an insight into the field behavior at observation times immediately (earlytime) and long (late-time) after the passage of a wavefront. The time-domain GTD
(Kellers) coecient has been derived as follows [64, 65]:


r
1
U ,
hs,h () = A
c
r/c

(2.67)

where
'

c sin /n
A=

1
1

 

  .
cos /n cos 0 /n
cos /n cos + 0 /2
(2.68)

The two faces of wedge are = 0 and = n. The impulse response obtained
from UTD has been derived as [65, 64]
hs,h () =

1
 U(),
Am 
c

+ Xm /c
m=1

r
=t ,
c

(2.69)

where Am and Xm are time-independent parameters. For practical UWB pulses,

we find that the methods (GTD, UTD, and exact) in (2.64) give us calculated pulse
waveforms that are not dierentiable [46]. Equations (2.67) and (2.69) can readily
be shown to have a singularity of type 1/2 U() in the wavefront = t r/c.
More generalized singularities will be given in (2.73) and (2.74) as well as (2.81)
and (2.82).
2.3.3. Early-time response and asymptotic high-frequency
time-harmonic solutions
Let us now investigate the relationship between the early-time response and asymptotic high-frequency time-harmonic solutions, due to its fundamental nature

46

UWB propagation channels

in our unified theory. A basic diculty of the GTD/UTD-based Laplace approach

is that we have to justify the validity of the GTD/UTD-based high-frequency approximation for each individual scattering center [64, 63, 65]. The communications medium can be regarded as the composite interactions of separated scattering centers [54, 44, 60]. Due to the high-resolution capability of a UWB pulse signal, these scattering centers are well resolved in the time-domain. This is one of the
most fundamental features of a UWB system from a signal analysis point of view.
The property of each scattering center can be treated as dierent (or composite)
mechanism corresponding to reflection, refraction, and diraction [62]. Under
the GTD/UTD framework, each mechanism is fully described by GTD/UTD coefficients. The accuracy of these time-domain GTD/UTD coecients is not well established. Can we use the time-domain GTD/UTD coecients to model the modern UWB communications/sensing systems? Fortunately we have already made a
decisive step [46] to answer this. Compared with the exact solutions of wedge and
half-plane, the time-domain GTD/UTD coecients are shown to be suciently
accurate for the modern UWB communications/sensing system modeling. This
has paved the way to a more systematic use of the time-domain GTD/UTD coefficients as building blocks in UWB signals modeling. These closed-form solutions
based on UTD/GTD provide big advantages over the conventional methodology.
The mathematical foundation of the GTD/UTD is the Kline-Luneberg expansion of the field components. The main result of this subsection is to tell us how to
obtain the asymptotic series representation of the steady-state or time-harmonic
solution of Maxwells equations if we know the behavior of the corresponding
pulse solution in the neighborhood of its singularities [30]. We will give the form
of asymptotic series that is valid in terms of the singularities of the pulse solution.
The asymptotic series is determined only by the singularities of the pulse solutions.
The mathematical result of this expansion is intimately related to an Abelian theorem (Watson lemma) in Laplace transform theory [67]. This theorem states [30]
that if the original function h(t) has the convergent or asymptotic form

h(t)

Cn t n

for t near 0 + and n > 1,

(2.70)

n=0

and h(t) < 0 for t < 0 (causal), then the image function H(s) of h(t) has the
asymptotic form

1 + n
H(s)
Cn
sn
n=0

as s  .

(2.71)

In this theorem s = j + may be complex, where is greater than zero. This theorem connects the early-time response (the initial transient fields on and near the
wavefront) and the high-frequency time-harmonic solutions. Let us assume the

Andreas F. Molisch et al.

47

field near the wavefront t = r/c behaves as h(r, t) A(t r/c) , where A is timeindependent and > 1. This is the case for (2.59) and (2.63). The asymptotic
behavior of the field behaves as
H( j) exp( jkr)A

( + 1)
.
2( j)+1

(2.72)

This is true for (2.62). Recall that GTD/UTD solutions are asymptotic solutions
of the model problem solutions that are exact. The field behavior near t = r/c is
related to the time-harmonic behavior for .
Equation (2.72) is inadequate to describe the phenomena at times t r/c corresponding to the arrival of the wavefront. As described above, the wavefront fields
are synthesized by the high-frequency time-harmonic wave constituents. The asymptotic expansions of the time-harmonic fields (such as obtained by GTD/UTD)
can be employed to develop corresponding expansions of the transient fields for
t r/c (early-time response), improving the leading term expressions of (2.72). A
generalized version of the time-harmonic field is given by [29]
!

"

1 jk(r) 
An (r)
1
A (r) +
( + n + 1) .
e
H( j)
+1
2
( j) n=0 ( j)n

(2.73)

Then, from (2.70) and (2.71), the transient field near t = (r)/c is
h() A (r)() +

n=0

An (r) n ,

=t

(r)
0.
c

(2.74)

The appearance of the phase (r) instead of r generalizes the applicability of

(2.72). An (r) and A (r) are time-independent amplitudes. In (2.73) and (2.74), the
first term represents the geometric-optics fields and the second term the diracted
fields.
According to (2.70) and (2.71), the time-domain ray solution (GTD/UTD)
is only valid for early to intermediate times since these high-frequency solutions are asymptotic in the frequency domain and remain accurate for moderate
to high frequencies. Again this is not a problem for UWB signals within FCC
band (3.110.6 GHz). The transient responses in intervals close to ray arrivals
times of the TD-UTD/GTD are generally strongest and are directly associated
with the waves arriving from the local features of the radiating (scattering) object.
The TD-UTD/GTD is based on the asymptotic (high-frequency) formulas of the
time-harmonic UTD and therefore the late-time behavior of the TD-UTD/GTD
impulse-response is expected to be erroneous [64]. But, on the other hand, if the
TD-UTD/GTD impulse-response is convolved with an excitation pulse waveform
that has a frequency spectrum (say FCC band) dominated by frequencies for which
the time-harmonic UTD solution is expected to be accurate, then the resultant
transient waveforms (per-path impulse response) predicted by the TD GTD/UTD

48

UWB propagation channels

should be accurate also. Another reason for paying special attention to the TD
GTD/UTD is that the current state of the art cannot allow us to sound the UWB
channel accurately. In practical analysis, an impulse response of the UWB channel
is needed. When the sounding pulses are much shorter than the practical signals to
which the measured channel response signals will be applied, we can regard these
measured signals as a good approximation of channel impulse response. However,
this condition cannot be met for a UWB channel. In the near future, it is infeasible
to use measurement equipment that has a bandwidth much larger than the practical UWB signal bandwidth (up to 7.5 GHz). This is one fundamental reason for
using the TD UTD/GTD, which gives us impulse responses of each ray.
Let us now deal with some practical issues. Although in theory we assume we
have the pulse solutions available, then the asymptotic time-harmonic solutions
can be determined. In practice, the case is on the contrary since there are rich asymptotic time-harmonic solutions and techniques available in the literature. One
question will naturally arise. Are these solutions (such as obtained by GTD/UTD)
suciently accurate for the current UWB communications? In other words, do
the early-time response of the scattering center suciently approximate its true
time-domain response? As an example consider the time-domain GTD solution
to the half-plane edge. Equation (2.63) gives the closed-form solution. If two such
edges are put together to form a strip, the two edges will interact with each other.
We assume that the frequency is high enough, so that the two edges can be treated
as two independent scattering centers. The wavefronts from each scattering center arrive at an observation point at dierent times. Let the first arrival be t = t1 .
The solution is asymptotic as 1 = t t1 0, although it is valid only within the
time interval t, where t = t t1 is small. However, when the next waveform
(caused by the other edge) arrives at t2 , t2 t1 is significantly larger than t. The
total response after t2 will not be accurate and must be found by other methods.
Therefore, one must be careful when using the time-domain GTD in a problem
involving more than one scattering center or multiple scatterings. Unfortunately
we always deal with multiple scatterings in modeling multipath. This problem is
essential to practical applications.
Based on the comparisons between the time-domain GTD and the exact solution [46, 65], it is discovered that the useful period t = t t1 may be varied depending upon factors such as scattering structure, source point, observation
point, and so forth. For example, the useful period of t 10x/c for the rightangle (90 degree) wedge, where x is the distance from the edge (scattering center).
At high frequencies, the field tends to be localized (Focks principle of locality).
The energy tends to be localized in the scattering center (a singularity). If we have
a UWB pulse of duration of T nanoseconds (ns), the condition to use the timedomain GTD to describe the pulse response of the scattering center is that the
UWB pulse duration T must be smaller than the useful period of t 10x/c,
that is, T < t 10x/c. For T = 0.1 1 nanosecond (110 GHz bandwidth),
x > cT/10 = 3T = 0.3 3 (cm), where c = 30 cm/nanosecond (speed of light).
In other words, the observation point must be 0.33 cm away from the scattering
center.

Andreas F. Molisch et al.

49
x

P(r)

r
S

Ei

L
y
Figure 2.2. Coordinates of impulse-response definitions.

2.3.4. Impulse-response for electrically large objects

Before generalizing the above results in a heuristic manner to model the UWB
channel (see Section 2.3.5), we must introduce the method in the literature for
modeling large radar objects. The same method will be still valid for UWB pulse
propagation in communications environment. Physically and mathematically the
two problems are equivalent. Radar interrogation (per-path pulse response modeling) is essentially a transient process [34]. In order to simplify the calculation of
radar reflectivity (per-path pulse response), one usually assumes that values obtained for a monochromatic source (time-harmonic solutions) are representative
of the target (per-path pulse response) and can be used with little or no modification, to predict the return for a pulsed signal with the same carrier frequency.
This assumption cannot be made for ultrashort pulses or any other UWB signaling
waveform. To obtain a target reflectivity model (per-path pulse response model)
applicable to all types of signal, time-domain concepts suggest use of an impulseresponse waveform to characterize each radar target (per-path pulse response).
The first such attempt to estimate an impulse-response waveform in the radar
scattering context was perhaps that of Kennaugh and Cosgri (1958). Using extremely simple approximations for high- and low-frequency backscatter from conducting sphere and spheroids, estimates of the impulse-response waveform were
obtained. The mathematical physical tools to attack the problems involving finite
three-dimensional targets are described in Section 2.3.2. As suggested in (2.56), a
linear system can be used to describe the object. We will follow [34] in the following derivations of this subsection.
To apply linear system analysis to scattering analysis problems [54, 44, 60],
we first define input and output quantities as electrical field intensities at a pair
of points in space and in fixed directions. As shown in Figure 2.2, the input x(t)
is the x-directed electric field intensities at the origin produced by a plane wave
Ei = xx(t z/c). If, now, this plane wave is incident upon an arbitrary scattering object (per-path obstacles), a scattered electromagnetic field Es is produced.
We may select an arbitrary component of this vector field at any point P(r) to be
the output of y(t). Thus, restricting the direction and polarization of the incident

50

UWB propagation channels

plane wave, the relationship between the x(t) and y(t) is that of a one-dimensional
linear system. If we assume that the scatterer does not move, so that the system is
time-invariant. The impulse-response waveform h0 (t) is that obtained when the
incident plane wave x(t) is impulsive, that is, x(t) = (t), the Dirac delta-function.
This approach is essentially the same as in Section 2.2.
Several features of the impulse-response waveform can be inferred immediately if the scattering body is of finite size:
(i) h0 (t) is identically zero for t < 0 if the scatterer lies to the right of the xy
plane, as in Figure 2.2. The linear system must be causal,
(ii) h0 (t) decays exponentially for large values of = t r/c, where r is the
distance to the observation point from the origin, as in Figure 2.1,
(iii) when r  L, where L is the maximum dimension of the scattering object, h0 (t) varies with r in a given scattering direction as r 1 f (t r/c)
for transverse components of the scattered field.
Properties of impulse-response waveform h(t). The normalized impulse-response
h(t) and the transfer function H( j) are a Laplace transform pair. Here H( j) is
related to a transverse component Es (t r/c) of the scattering field at large r by
)

Re H( j)e j(tr/c) =

 

2r s
r
E t ,
c
c

(2.75)

the incident plane wave being given by Ei = x cos (t z/c). We introduce the
notation
H(s) =


0

h(t)est dt,

H(s) = L h(t) ,

h(t) = L1 H(s) .

(2.76)

In practice, we usually first obtain H( j) using the GTD/UTD asymptotic solutions, then we can replace s = j in H( j) to obtain H(s). Secondly, we apply the
Laplace transform pair to obtain the h(t). Not all the functions have their inverse
Laplace transform in closed forms. Due to the essential property of causality, a
Laplace transform rather than a Fourier transform must be used, in order to guarantee that the so obtained impulse-response is physically sound. Such obtained
impulse-response must be a real function. This can serve as sanity check for the
analysis.
High-frequency impulse-response based on scattering centers. Let us present such a
model based on the concept of scattering centers. Extensive development of highfrequency scattering theory (developed in Sections 2.3.2 and 2.3.3) traces back at
least to 1940s1960s. Experimental conditions and analytic tools were mature until 1960s, mainly driven by the study of short pulses caused by nuclear explosions.
Two categories of work were performed. One was to model the impulse-response
of simple shapes and composite shapes (finite three-dimensional objects). Another
category was to study the radio propagation of pulse in a dispersive medium like
plasma at a distance of kilometers. Here we are only interested in methods developed for the first category, especially based on experimental work [68, 69].

Andreas F. Molisch et al.

51

It is postulated that for a suciently short pulse, the solution can be represented as a sequence of pulses scattered from the resolved scattering centers.
(Section 2.3.2 has discussed the conditions of validity of the time-domain GTD
to model the resolved scattering centers.) This postulated solution has an approximate equivalent in the frequency domain. The back scattering amplitude at the
frequency will be of the form
H( j) 

An Hn ( j)e j2kLn =

An Hn ( j)e jn ,

n =

Ln
,
c

(2.77)

where An is the scattering amplitude, Hn ( j) is the frequency response for the

nth center, and Ln is the distance along the ray path from the first to the nth
scattering center. The so-called per-path transfer function (heuristically defined
in Section 2.3.5), Hn ( j), plays the central role in characterizing the pulse waveform. The per-path delay, n = 2Ln /c, describes the intrapath delay that connects
the localized responses. When an inverse Laplace transform is applied in (2.77)
(see (2.70) and (2.71)), the early-time approximation to the impulse-response is
obtained as
h() =

An hn () n .

(2.78)

Both Hn ( j) and hn () have a unit energy for convenience. In general, (2.77) and
(2.78) are connected by the Laplace transform. We assume that Hn ( j) is centered
about f0 with a normal bandwidth B. If Hn ( j) is slowly varying for B/2 + f0
f B/2 + f , the term e j2kLn varies rapidly in the same band. For a special case, a
simple relation can be obtained as
h() 

An Hn 0 p

2Ln
,
c

(2.79)

where p(t) is the incident pulse. Thus the nth scattering center returns a pulse of
amplitude An Hn (0 ), delayed by n = 2Ln /c from the first term. From a physical
point of view, (2.78) and (2.79) imply that the dimensions of the scattering centers
are of the order one wavelength or less. Thus, if the incident pulse contains more
than several wavelengths (at the center frequency), the distortion of the scattered
pulses is negligible. Refinements of (2.79) could be included by accounting for the
frequency dependency of Hn ( j) near the center frequency, as in (2.78). It should
be noted that this representation in (2.78) and (2.79) may also include terms due
to multiple scattering.
For high frequencies, one may approximate H(s) by an expansion called
Luneberg-Kline series in negative powers of s (as discussed in Section 2.3.2 and
used in Section 2.3.5). Individual terms in such an expansion are related to impulse, step, ramp, and low-order discontinuities in h(t). Thus, if the asymptotic

52

UWB propagation channels

()

1
A1

A2

h1 ()

AN

h2 ()

hN ()

h() =

+N

n=1 An hn () ( n )

Figure 2.3. The structure of the generalized multipath model.

expansion of H(s) contains a term An sn es , the impulse-response waveform undergoes a discontinuous jump of magnitude An in its (n 1)th-order derivative at
t = . Terms of the form A0 es in the asymptotic expansions are contributions
predicted by geometric optics, giving impulses in h(t) at t = , of magnitude A0 .
It is clear that low-frequency approximation serves to determine waveform shape
and size, whereas high-frequency information relates to fine structure and detail in
the waveform. Both are thus interpreted in a single concept that serves to predict
the scattering behavior in the intermediate or resonance range of frequencies. The
intuitive hypothesis that the ramp response is most closely associated with the volume and size of a scatterer suggests that, as our knowledge of response waveforms
for special objects increases, we will ultimately be able to sketch, by inspection,
a reasonably accurate estimate of the ramp response of any target (per-path response).
2.3.5. Generalized multipath model based on per-path pulse waveform
The methods and mathematical physical results are historically developed for
UWB radar (see Section 2.3.4), but can be generalized to a UWB radio propagation channel if we treat the objects encountered by tracing a path in multipath
channel modeling. This will be the philosophy adopted in this subsection. Admittedly, the state of the art in predicting or approximating impulse-response waveforms for per-path is rather primitive. It has been applied to perfectly conducting
objects of simple shape [41, 42, 35, 4451]. With a combination of theoretical and
experimental techniques, the associated waveforms can be determined for other,
more complex, shapes. No technique or concept can make an intrinsically dicult
task simple, and the long history of channel modeling investigations indicates that
this is, indeed, a dicult task. Perhaps the chief value of the per-path impulseresponse concept is that it sums up the signal signature of a path (like the radar
signature of an object) in a unique manner which adds insight into the complex
relationship between each individual path response and the propagation environment. The relationship to statistical channel modeling, and some discussions of
measurement results, will be present in Section 2.5.

Andreas F. Molisch et al.

53

Based on the GTD framework and principle of locality [31], we postulate that
the total response is modeled by the sum of the impulse-responses of local scattering centers [35, 41, 42, 4451]:
h(, , ) =

N1

an (, ) n +

n=1

N2

An (, )hn (, , ) n ,

(2.80)

n=1

where N1 and N2 represent the number of rays without per-path pulse distortion
and rays with per-path pulse distortion, respectively. The first term is of the form
identical to Turins model. The second term is more general; as a matter of fact, the
first term is a special form of the second term when hn () = ().13 The function
hn () is defined to have unit energy. When an arbitrary pulse, p(t), passes through
the channel defined by (2.80), some pulses will be undistorted and some pulses
will be distorted. When the duration of this pulse, T p , is shorter than any interpath
interval of (2.80), that is, T p < |i j | for i = j, all pulses will be resolvable in
time. The hn () can be obtained through exact, experimental, numerical, or/and
asymptotic methods. As an example, let us take a look at the diracted rays. The
geometric optics rays can be treated as a special form of a generalized diracted
ray [30]. If in the neighborhood of the singularity = any field component has
the behavior



n
Cn 

n!
n=0

< ,

hn () =



n
Dn 

n=0

n!

(2.81)
> ,

where () and () are rather general, then the corresponding asymptotic frequency response has the form [30]
Hn ( j) e

jk

m=0

Dm 1
m! ( jw)m

 
0

1
Cm

m! ( jw)m

t m t
t e dt
jw

 
0

t m t

t e dt ,
jw

(2.82)

where (x, y, z) determines the wavefront at the (x, y, z) in question, that is,
(x, y, z) ct = 0 is the discontinuity hypersurface on which hn () has a finite
jump. As shown in Section 2.3.3, the behavior of the impulse-response solution in
the neighborhood of its singularities does determine the asymptotic series for the
corresponding time-harmonic problem. Further we learn the form of the asymptotic series in terms of the form of the singularity of the impulse-response solution.
The function of (2.81) can be infinite at the singularities. The results of (2.81) and
13Note the properties of Diracs function: ()

() = () and ()

( n ) = ( n ).

54

UWB propagation channels

(2.82) extend a known result in Laplace transform theorythe Abelian theorem

((2.70) and (2.71)).
Equation (2.82) is a special form of (2.80). Equations (2.80) and (2.82) are
insuciently generic to represent all the signals in the set of the actual physical field
signals that satisfy Maxwells equations. But they are generic enough to describe
almost all the practical series obtained by GTD/GTD and other high-frequency
asymptotic techniques. For a large category of physical signals, both () and ()
are of a form 1/2 , the asymptotic series contains fractional powers of (1/), for
example, 1/( j)m+1/2 , where m is an integer. For diracted signals () and ()
can take a more general form of (log )m , exp(1/), and , where is real
[30]. () and () may also have nonalgebraic singularities provided that these
functions are continuous at t = 0+. For , each term of the series vanishes.
This is as expected, for there is no classical diracted geometric-optics field [30].
The leading term for large may well serve as the geometric optics diracted field.
For a special case, using a Laplace transform pair, we obtain

m
Dm 

1
0 ,

hn () =

0,
Hn () =

m=0

m!

> 0 ,
<0 ,

(2.83)

Dm
3
m+
.
m+1/2
(
j)
2
m=0

The primary term in (2.83) is sucient in practice (e.g., GTD/UTD series). Now
consider the physics-based generalized multipath model of a form in the following:
H() =

N

n=1

An ( j)n e jn ,

h() =

 

An
 (1+n ) U()
n ,

n
n=1
(2.84)


where n is a real value, Hn ( j) = ( j)n , and hn () = (1+n ) U(). U(x) is Heavisides unit function. This model is asymptotically valid for incident, reflected,
singly diracted, and multiply reflected/diracted ray path field. The motivation
of using the function in (2.84) is its simplicity and feasibility for a large category of
problems. This form can allow us to use the powerful mathematical tool of fractional calculus [70, 71] to conveniently describe our problems. Equation (2.80)
gives a formal generic mathematical model for h(). A large category of problems
can be described by per-path impulse-response hn () given by (2.81). In (2.84), we
can model n as the random variables that can be experimentally measured. The
amplitudes and delays can be statistically described similarly to the Turins model.
As an example, in Figure 2.4, we obtain the pulse waveform distortion as
a function of for one generalized multipath defined by its impulse-response
U(). The incident pulse waveform is the second-order derivative of a Gaussian function (corresponding to the curve of = 0). Usually, a matched filter is
required for an optimum reception (see next subsection), so the new changed

Andreas F. Molisch et al.

55

1.5

d(t) and v(t)

1
0.5
0
0.5
1
1.5

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Time t (ns)

Figure 2.4. Pulse waveform distortion as a function of for one generalized multipath.

waveform after the incident pulse passing through the sole generalized multipath
is needed and calculated in Figure 2.4. For a pulse position modulation, a template pulse consisting of the incident pulse and its negative delayed version is used
for detection. This template pulse is only proper when the pulse waveform is not
changed by the channel. When a pulse is distorted by passing through a generalized multipath, then the pulse waveform is changed to a new waveform. Thus a
new template formed by the new changed waveform and its delayed version should
be used [47, 48].
2.3.6. Impact of propagation processes on optimum receiver structures
We will incorporate the per-path impulse-response into the optimal receiver when
intersymbol interference (ISI) is present. For pulse amplitude modulation (which
encompasses
PSK, QAM, etc.), the transmitted signals have the same form [43]
+
as s(t) =
n=0 an x(t nT), where an represents the discrete information symbol sequence with symbol duration of T, and x(t) is a pulse that is bandlimited.
The channel impulse-response h() is illustrated in Figure 2.3 for the generalized
multipath (see Section 2.3.5). The receiver signals are represented as
r(t) =

an y(t nT) + n(t),

(2.85)

n=0

where n(t) is AWGN. The received symbol signal waveform is y(t) = h(t) x(t),
consisting of a sequence of distorted pulses described by h(t) defined by (2.80).
It has been shown that the optimum receiver structure is illustrated in Figure 2.5.
The received signal r(t) first passes the matched filter followed by a sampler of
sampling rate of 1/T. The sampled sequence is further processed by a maximum
likelihood sequential estimation (MLSE) detector. This receiver structure is optimal in terms of minimizing the probability of transmission error in detecting the
information sequence. If we accept the condition that y(t) is finite in the UWB

56

UWB propagation channels

Received
signal
r(t)

Matched filter
y (t)

Sampler

MSLE
(Viterbi algorithm)

Data
output

Clock
t kT

Figure 2.5. Optimal receiver structure in presence of ISI in an AWGN channel.

receiver, this structure can immediately be applied to our problem. The output of
the matched filter is expressed as
q(t) =

an R y y (t nT) + v(t),

(2.86)

n=0

where R y y (t) = y(t) y (t) = Rxx (t) h(t) h (t) is the autocorrelation of
y(t) and v(t) is the response of the matched filter to AWGN noise n(t). Rxx (t) is the
autocorrelation of x(t). Pulse distortion will aect the optimal energy collection in
matched filter. The sampled values at t = nT can be expressed as
qk = ak +

an Rkn + vk ,

(2.87)

n=0, n=k

where ak term represents the expected information symbol of the kth sampling
period, the second term is the ISI, and vk is the additive Gaussian variable at the
kth sampling point. The sequence of qk will be further processed by MSLE (Viterbi
algorithm) before being sent to decision circuits. After pulse distortion is taken
into account in the ISI through (2.86) and (2.87), all the rest of analysis follows
the conventional case [43] when Turins model rather than (2.80) is used for h(t).
The suboptimum implementation of the matched filter is the Rake receiver
structure. Suboptimum linear equalizers such as zero-forcing, decision-feedback,
adaptive equalization can be used to replace the optimum MSLE. Since the perpath pulse waveform has been included in the channel impulse-response (illustrated in Figure 2.3), the resultant matched filter is sometimes called generalized
Rake receiver structure [45], which is a generalization of Altes receiver structure
[72].
Appendix
2.3.A. Representation of pulse distortion using fractional calculus
Following Sommerfeld [24], Lamb [25] and Friedlander [26] analytically and numerically studied the pulse distortion caused by diraction at a half-plane for
sound pulses. Friedlander transformed Lambs result into Abels integral equation,
which is the classical result in fractional calculus, and is thus a good starting point

Andreas F. Molisch et al.

57

for us to apply fractional calculus in UWB communications.14 We can readily represent their formulation in terms of the Rieman-Liouville fractional dierintegral
(d/dt) , which can be defined as
0

(d/dt) f (t) 0 Dt f (t)

1
()

t (1+) U(t)

f (t),

< 0,

(2.88)

where is negative and real. For fractional > 0, the above definition can be
applied with additional step 0 Dt f (t) = (d/dt)m 0 Dtm f (t), where m is a positive integer large enough so that ( m) < 0. According to (2.88), we find that
( j) F( j) is the Fourier transform of 0 Dt f (t). When assumes an integer value,
(d/dt) or 0 Dt reduces to the ordinary Riemann integral = n, where n is an (positive or negative) integer. This new tool provides a powerful means to describe the
pulse distortion. For example, for the PEC wedge and half-plane (a special case of
wedge), we have = = 1/2. For the generalized multipath signals in (2.84), we
have = n for the nth path. So (2.84) can be expressed in a new form as
h() =

An

n=1

d
d

n

()

n ,

(2.89)

where n is an arbitrary real number. When a pulse p(t) passes through the channel, the new pulse is
r(t) =

An

n=1

d
dt

n

p(t)

t n ,

(2.90)

According to (2.89) and (2.90), after a pulse travels through a generalized multipath, its waveform will be changed and represented by the fractional dierintegral
(d/dt)n of this pulse. The delay of this pulse is similarly modeled as in Turins
model [52] which ignores the eect of the fractional dierintegral (d/dt)n of this
pulse on the incident pulse. The fractional dierintegral has a lot of desirable properties that greatly simplify the analytical and numerical analysis of the signals in
(2.90) [45].
2.4. Path loss and shadowing
2.4.1. Introduction
Path loss models give an accurate estimate of the power level as a function of the
transmitter-receiver separation distance. Therefore, path loss models constitute
the basis for link budget calculations, since they allow the calculation of the SNR
14In electromagnetics, Engheta appeared to be the first to use fractional calculus [73].

58

UWB propagation channels

or the C/I in dierent locations in the area in which the assumed path loss model
holds. Analysis and simulations based on link budget calculations are usually required to evaluate the coverage and the capacity of the system under test. Indeed,
mobile radio systems tend to be interference limited (due to other users sharing
the same/adjacent channels) rather than noise limited, and this kind of studies
permits to predict the eects of the interfering signals that a user will experience
from other RF sources.
This issue becomes more critical for UWB systems, since they are committed to coexist with other systems sharing the RF spectrum without significantly
degrading their performance. The total interference to the radio systems can be
limited reducing the number of active UWB devices in the area, that is, limiting
the capacity of the UWB systems, or reducing the power emitted by each single
UWB device, as the current regulation establishes.
In this context, the propagation losses may have two counteracting eects:
(i) a reduction of the interference power detected by the victim receiver,
which facilitates the coexistence of UWB with other RF systems;
(ii) a degradation of the UWB signal that may be significant since the UWB
transmitted power cannot exceed the threshold imposed by the current
regulation; thus, the UWB communication link may be seriously compromised.
Link budget analyses based on accurate path loss models for UWB channels
give deeper insights about the coexistence/capacity issues for UWB systems. Furthermore, accurate path loss models permit to design a reliable radio system with
a reduced link margin.
Propagation models that predict the signal strength can be divided into two
categories:
(i) large-scale models which predict the signal power over large transmitterreceiver separation distances (several hundreds or thousands of meters
in outdoorstens of meters in indoors);
(ii) Small-scale models (also called fast fading), which characterize the actual signal power over very short distances (up to 15 or 30 meters in
outdoorsup to a few meters in indoors).
At the small scale, variations are very rapid and the signal may have deep fades
(e.g., 30 or 40 dB) when the receiver is moved just a fraction of a wavelength of the
radiated signal. Small-scale fading is due to the multipath propagation, that is,
many reflected replicas of the transmitted signal combine at the receiver with a
dierent phase (related to the dierent travelling distance). When the phases are
close, the waves add up and the result is a gain in power, but when the phases are
opposite (180 apart), then the waves cancel each other and a deep fade occurs. The
small-scale models will be treated in more detail in the subsequent Section 2.5.
Large-scale propagation models, instead, predict the local average power, that
is, the average power of the small-scale variations. Indeed, the power measurements used to develop these models are averaged over short distances. Becoming
more specific, the mean value of the power level as a function of distance is provided by path loss models. In practice, the signal level statistically fluctuates around

Andreas F. Molisch et al.

59

the deterministic value given by the path loss laws, due to the presence of obstacles
and objects around the transmitter and the receiver. This statistical behavior is
referred to as the shadowing.
The path loss law and the shadowing behavior vary depending on the carrier frequency and the signal bandwidth. Furthermore, they strongly depend on
the propagation environment. At the moment, the only available analysis which
statistically models the variability of path loss and shadowing laws with the propagation environment applies to residential environments [74, 75].
In this chapter, we summarize the most recent contributions to measurements
and path loss models for UWB channels in typical propagation scenarios [7485].
Measurements have been conducted also within some of the projects of the IST
program of the European Community, Fifth Framework [86, 87]. Furthermore,
the standardization groups SG3a and SG4a, established within the IEEE 802.15
to propose PHY layers potentially based on UWB technology, are working to set
up standard models of the UWB channel in order to fairly evaluate and optimize
the system proposals. The propagation model for the IEEE 802.15.3a standard has
been already selected and holds in residential/oce environments over areas of the
order of 10 m (see also Section 2.6). The model for the IEEE 802.15.SG4a, instead,
is currently in progress.15
Most of the path loss and shadowing models for UWB channels are based on
extensions of narrowband models. However, since the bandwidth of UWB signals
may extend for several GHz, the frequency dependence of diraction/reflection
coecients, dielectric constants, and so forth, can be significant in the band of interest. Some measurement campaigns over very wide bandwidths have permitted
to investigate on the frequency dependence of the UWB path loss laws and some
models have been proposed [76, 78, 84].
This chapter section is organized as follows. In Section 2.4.2, we describe the
typical UWB scenarios based on the envisaged applications for this technology.
In Sections 2.4.3 and 2.4.4, we review the measurements and models for the path
loss and shadowing presented in the literature for the UWB channel. Finally, in
Section 2.4.5, we discuss about the standard models for path loss and shadowing
already adopted for the high data-rate standard (IEEE 802.15.3a) and currently in
progress for the low-data-rate standard (IEEE 802.15.4a).
2.4.2. Typical propagation environments for UWB applications
The UWB propagation environments are various, since the envisioned UWB applications are numerous and each of them addresses a typical operation scenario.
(i) The use of a very wide bandwidth makes the UWB technology very attractive for high-data-rate PANs, including data and high-quality video
transmissions. The regulation limits on the power spectral density and
the quality-of-service requirements limit the coverage area of these UWB
15 The selection of dierent models for these standards is necessary because they address dierent

UWB applications which have their own scenarios and environments.

60

UWB propagation channels

devices. Therefore, a typical propagation environment for such applications is a small area with a radius on the order of 10 m within residential
or oce buildings.
(ii) Low power consumption and locationing capability, instead, make UWB
technology suitable for application in sensor networks. As the data rates
are much lower in this context, larger coverage areas are served. Typical
environments for sensor networks can be home/oce as well as factories, grocery stores, and so forth, where indoor distances can be larger.
However, sensor networks find applications also in the WBANs having
typical ranges of 12 m, or as cable replacement in the vehicles, whose
typical ranges are on the order of 510 m.
(iii) Another attractive application for UWB is emergency communications,
possibly combined with geolocation, for example, from a building that
collapsed during an earthquake, or from an avalanche. Required distance can be up to 100 m in outdoor environments.
At the moment, however, only measurement data and models for indoor office/residential environments are available; for the other environments, some campaigns are in progress, but not yet finalized.
2.4.3. Path loss laws
The path loss (PL) denotes the local average received signal power relative to the
transmit power. From electromagnetic field theory, the free-space path loss, PLFS ,
is expressed as


PLFS (, d) =

4d

2

(2.91)

which is a function of the distance d and the wavelength of the radiated signal.
The free-space path loss model, which is a square law with the transmitter-receiver
distance, applies to an ideal situation in which the electromagnetic field propagates in the absence of ground/obstacles reflections, multipath propagation, and
so forth.
A more general path loss model uses a parameter, n, to denote the power
law relationship between the transmitter-receiver separation distance d and the
received power:


PL(d)

d
d0

n

(2.92)

where PL is the mean path loss, n is the mean path loss exponent, which indicates
how fast path loss increases with distance (n = 2 for free-space propagation) and
d0 is a reference distance.
Absolute mean path loss, in decibels, is defined as the path loss from the transmitter to the reference distance d0 , plus the additional path loss described by (2.92)

Andreas F. Molisch et al.

61

Table 2.1. Path loss exponents and standard deviations measured in dierent buildings.
Building

Frequency (MHz)

(dB)

914
914
1500
900
1900
1300
1400
1300
1300
900
4000
1300

2.2
1.8
3.0
2.4
2.6
2.0
2.1
1.8
1.6
3.0
2.1
3.3

8.7
5.2
7.0
9.6
14.1
3.0
7.0
6.0
5.8
7.0
9.7
6.8

Retail stores
Grocery store
Oce, hard partition
Oce, soft partition
Oce, soft partition
Textile/chemical
Textile/chemical
Paper/cereals
Metalworking
Indoor to street
Textile/chemical
Metalworking

in decibels:
 

PL(d) (dB) = PL d0 (dB) + 10 n log10

d
.
d0

(2.93)

Generalizing (2.91), the frequency dependence of path loss is also included [88,
86]:
PL( f , d) dn f m .

(2.94)

Note that the distance dependence and the frequency dependence can have dierent exponents, that is, n = m.
In general, the exponent n is extracted by performing a linear regression of
the path loss values expressed in dB versus the logarithm of the distance. The path
loss values are computed from the experimental data by integrating the total signal power received at a given location at a distance d and then normalizing it to
the transmitted power, when available, or to the signal power received at the reference distance d0 .16 More accurate models average these values over a number of
realizations to remove the eects of the small-scale statistics (see Section 2.4.3.2).
Typical values of path loss exponents and standard deviations (shadowing) for
dierent classes of buildings are given in the literature. In general, values of n < 2
are given for indoor LOS propagation and n > 2 for indoor NLOS conditions.
Table 2.1 provides a range of typical values that have been measured in the past
using narrowband probing signals [1].
The presented models apply to narrowband signals and are largely employed
to the analysis of narrowband systems. However, some of those models are based
on measurements with probing signals having a bandwidth ranging from 200 MHz
to 1 GHz and can work as starting points for UWB studies, since they characterize
environments of interest for the UWB envisaged scenarios.
16In that case PL(d ) = 0 in (2.93).
0

62

UWB propagation channels

Measurements over a bandwidth of 200 MHz around 1 GHz over an area of

about 50 m radius have been carried out in a high-rise oce building and a threestory building with oces and laboratories, leading to an average path loss exponent of nav = 2.4 and a lognormal shadowing with a standard deviation from 2.82
to 4.56 dB [89, 90].
Experiments have been also performed in factory buildings in LOS and OBS
over distances from 25 to 75 m; the probing signal had a bandwidth of 400 MHz
around 1.3 GHz [9193]. The path loss exponents vary from 1.8 to 2.8, with an
average of nav = 2.2.
A measurement campaign at 1.5 GHz over a bandwidth of 200 MHz in a
medium-size two-story building at distances up to 50 m has reported a path loss
exponent of n = 3 or n = 4 [94].
In [95] a path loss model with n between 1.8 and 2.1 is provided for oce
environments for LOS, OBS, and NLOS for the band 4.5-5.5 GHz; furthermore,
the penetration losses for walls of 10 cm and 20 cm thickness are estimated to be
16 dB and 29 dB, respectively.
Measurements in LOS conditions over a band of 400 MHz centered at 1 GHz,
5.5 GHz, and 10 GHz in an oce environment (rooms are delimited mostly by
smooth metallic walls) have shown path loss exponents of 1.27, 1.49, and 1.26, respectively, which seems to be quite independent from the carrier frequency [96].
Similar conclusions have been drawn in [97] for LOS conditions in a laboratory/
oce environment over distances up to 30 m, while the OBS measurements have
shown that the path loss exponent increases with the increasing frequency. Those
measurements were made at 2.4 GHz and 4.75 GHz with a probing signal of
500 MHz bandwidth and at 11.5 GHz with a bandwidth of 1 GHz, obtaining path
loss exponents of, respectively, 1.86, 1.98, and 1.94 for LOS and 3.33, 3.75, and 4.46
for OBS.
Path loss exponents of 1.5, 1.7, and 1.6 have been measured at 2 GHz, 5 GHz,
and 17 GHz with a 500 MHz bandwidth in LOS conditions in a laboratory and the
adjoining corridor at the University of Swansea (UK); in OBS conditions, instead,
the attenuation increases with frequency [98].
Many experimental and theoretical activities (currently in progress) are devoted to understand if a simple extension of those models by just changing the
parameters can accurately models the UWB path loss and shadowing.
As explained in the following sections, many contributions confirmed the
power law with the distance, with exponents comparable with the narrowband
propagation. However, the frequency dependence of path loss becomes a critical
issue in UWB channels.
2.4.3.1. Distance dependence
The distance dependence of path loss for UWB signals is usually modelled by
(2.93), in which the parameter n is extracted by UWB experimental data. As in
narrowband cases, the path loss exponent depends on the environment and the

Andreas F. Molisch et al.

63

Table 2.2. Parameters of the model in [74].

LOS

NLOS

Mean

Std. dev.

Mean

PL0 (dB)

47

NA

50.5

Std. dev.
NA

n
(dB)

1.7
1.6

0.3
0.5

3.5
2.7

0.97
0.98

propagation conditions. Some models dierentiate between LOS, soft NLOS,

also known as obstructed LOS, and hard NLOS.
LOS exponents in indoor environments range from 1.0 in a narrow corridor
[86] to about 2 in an oce environment [77, 81, 87].
NLOS exponents typically range from 3 to 4 for soft NLOS [77, 81, 87] and
from 4 to 7 for hard NLOS [79, 86].
In [86] the path loss exponents for the three LOS, soft NLOS, and hard NLOS
conditions are reported for more than sixteen scenarios in typical European oce
buildings for distances up to 20 m.
In [82, 74, 75] the path loss exponent is modelled as a random variable that
varies from building to building in residential environments in the US for distances up to 10 m. They also found that the empirical distribution of this random
variable is well approximated by a Gaussian distribution whose parameters (mean
values and standard deviations) are summarized in Table 2.2.
The model in (2.93), which assumes a linear dependence between the path loss
(dB) and the logarithm of distance, is usually referred to as single-slope model. In
[80] the single-slope model has not given a good agreement with the experimental
values, yielding a too large value for the exponent n. A more accurate model than
the single-slope has been proposed taking into account separately the eect of the
distance and the attenuations introduced by the walls intercepted by the direct
path, that is,
PLMw (d) (dB) = PL(d) (dB) + Mw (dB),

(2.95)

where Mw is the multiwall component [80].

Such a multiwall model requires to evaluate the number and the nature of the
obstacles traversed by the direct path for each measurement. The model distinguishes between NLOS and hard NLOS, defining the multiwall component Mw
as

k l1

Mw (dB) =

lc

for NLOS,
for hard NLOS,

(2.96)

where k = 1, . . . , 4 is the number of penetrated walls having attenuation l1 =

6.25 dB and lc = 26.25 dB is a constant accounting for the cumulative eect of
the extremely heterogenous series of obstacles present in the hard NLOS link. The

64

UWB propagation channels

number k of walls intercepted by the direct path when the distance is d can be defined looking at the considered environment [80]. The dependence on distance is
still modelled by (2.93) with an exponent n = 7.33.
Finally, a multislope path loss model has been proposed in [99], in which the
breakpoint(s) is obtained by theoretical derivations. In particular, assuming that
the transmitted signal reaches the receiver through a multitude of paths, L, having
dierent path lengths {d + di }, i = 1, 2, . . . , L, equation (2.93) can be expressed
as
PL(d) 20 log10 (d) 10 log10

L 

1+

i=1

di
d

2

(2.97)

Dierentiating (2.97), with some algebra, the breakpoint dbp is easily derived
to be
dbp

3 2/20

1 /20

+L

i=1 di

(2.98)

The breakpoint dbp separates the region where the path loss increases of dB/dec
(with distance) from the region where a slope of 20 dB/dec (free space) is assumed.
A dual-slope model has also been proposed in [79], obtained by fitting experimental path loss values.
2.4.3.2. Frequency dependence
In narrowband channels, the path loss at a distance d is defined as in (2.94) and is
calculated from experimental data by
)

PL(d) =

2 
2 3
E PRX (d)
= E H f c , d  ,
PTX

(2.99)

where PTX and PRX are the transmitted and received power, respectively, fc is the
carrier frequency and H( f , d) is the frequency response of the radio channel [100].
The values of {PRXi (d)}, or the values of the {Hi ( fc , d)}, measured over a small area
at a distance d from the transmitter, are averaged in order to remove the eects
of small-scale variations of the signal level.
In wideband channels [100], the path loss is obtained by integrating the channel frequency response over the frequency band of interest [ f1 , f2 ]:
4

PL(d) = E

f2
f1

5


H( f , d)2 df .

(2.100)

Indeed, when the band of interest is such that the frequency dependence of the
propagation parameters is negligible, this integration is equivalent to averaging
over the multipath fading.

Andreas F. Molisch et al.

65

In a UWB channel, instead, the frequency dependence of the propagation parameters can be significant. It is well known that also the antennas gain (or, equivalently, their eective areas) is a function of frequency, and often it is not possible
in the experimental channel responses to isolate the channel behavior from the
antennas responses.
Reference [100] defines a frequency-dependent path loss as
#

PL( f , d) = E

$

H(, d)2 d ,

f + f /2 
f f /2

(2.101)

where f is small enough that diraction/reflection coecients can be considered constant in f . The total (frequency-averaged) path loss may be obtained by
integrating over the whole bandwidth of interest [100].
Many measurement campaigns have observed very large bands, such as 2
8 GHz [83, 101], 111 GHz [84], 113 GHz [76], and 212 GHz [78]. Some of the
UWB models that have been developed based on these campaigns assume that the
path loss dependence on frequency and distance may be expressed, similarly to the
narrowband case (2.94), by the product of two independent functions of frequency
and distance:
PL( f , d) = PL f ( f ) PLd (d).

(2.102)

Reference [84] models the frequency dependence of path loss as



PL f ( f ) f m ,

(2.103)

where m varies between 0.8 and 1.4.

Reference [76] instead proposes


log10 PL f ( f ) exp( f ),

(2.104)

with varying between 1.0 (LOS) and 1.4 (NLOS).

In [78] the frequency dependence of path loss is modelled implicitly through
the frequency dependence of the path loss exponent that gives the dependence on
distance. This analysis distinguishes between LOS and NLOS path loss: in LOS
conditions in the corridor, the exponent is about 1.5 and increases with the increasing frequency, while within the lab the exponent is about 1.0 and still increases with the increasing frequency; in NLOS conditions, instead, the exponent
varies from 2 to 6 diminishing with the increasing frequency with a decay that is
well approximated by a third-order polynomial.
2.4.4. Shadowing
Shadowing, or large-scale fading, is defined as the variation of the local attenuation
around the deterministic mean value given by the path loss law.

66

UWB propagation channels

Some measurement campaigns have shown that the UWB shadowing is fairly
similar to the narrowband case. Many contributions in the literature are consistent
in assuming the narrowband shadowing to be lognormally distributed, with dierent values of the standard deviation, since it strongly depends on the measurement
environment.
To be more specific, once that the variations of the mean of the power measurements (in dB) have been fitted to a straight line, which represents the path loss
model as a power of distance, there remains the task of characterizing the distribution of the samples Pi around their mean Pi at a given radial distance. In terms
of the residual errors, ei , we can write, in general,
Pi = Pi + ei .

(2.105)

At a given radial distance, it is found that the ei s obey the normal distribution with
mean = 0 dB and a standard deviation , that is, N(0, ), leading to
 

PL(d) (dB) = PL(d) + X = PL d0 + 10n log10

d
+ X ,
d0

(2.106)

where the term X N(0, ) represents the shadowing.

Some UWB models also assume a lognormal distribution for shadowing [79].
For indoor conditions, the standard deviation of shadowing is easily derived from
the scatter plot of the measured path loss data.
More exhaustively, [74] suggests to model the shadowing variance as a random variable; 2 is modeled as normally distributed with realizations that change
from house to house. The total attenuation due to shadowing and path loss is then
expressed as:


PL0 +10n log10 (d) + 10n1 n log10 (d) + n2 + n2 n3 ,

(2.107)

where n1 , n2 , and n3 are zero-mean, unit-variance Gaussian variables, but with

a recommendation to limit n1 to the range [0.75, 0.75] and n2 , n3 to the range
[2, 2] to avoid nonphysical values of the attenuation. The other parameters (i.e.,
the mean values of n and , n , and , resp., and their standard deviations, n ,
and , resp.) can be found in Table 2.2.
2.4.5. Standard models for UWB path loss and shadowing
Standard models are established to fairly evaluate dierent system architectures, as
well as to assess the benefits of dierent transceiver structures, alternative multiple
access techniques, and so forth.
The only UWB standard model presently available has been selected by the
IEEE 802.15.SG3a. The ith realization of the channel impulse-response, hi (t), is

Andreas F. Molisch et al.

67

defined as
hi (t) = Xi

L
K

i
aik,l t Tli k,l
,

(2.108)

l=0 k=0

where aik,l are the tap weights, Xi represents the lognormal shadowing, K is the
number of rays within the l-th cluster, and L is the number of ray clusters. A detailed description of the model can be found in Section 2.6 as well as [102], while
here we concentrate on the UWB standard models for path loss and shadowing.
The lognormal shadowing of the total multipath energy is captured by the
term Xi characterized by
 

20 log10 Xi N 0, x2 ,

(2.109)

where the standard deviation x is 3 dB.

The recommended path loss model is the one presented in [74, 103] and described above, even if a free-space path loss model has also been specified and has
mostly been used within the standardization.
In September 2003, standardization activities started for low-data-rate UWB
communications, in the IEEE 802.15.4a group, which established a channel modelling subgroup. This subgroup is currently working on formulating an appropriate UWB channel model for this standard, which presents many dierences with
the 802.15.3a. Specifically, the considered ranges are much larger (up to 100 m),
and the environments will include factories, storage halls, and so forth. The path
loss model will follow the general principles outlined above, though the exact parameters still need to be specified.
2.5. Delay dispersion and small-scale fading
2.5.1. Introduction
Delay dispersion and small-scale fading are those eects of the channel where the
UWB propagation channel diers most significantly from narrowband channels.
Delay dispersion is defined to occur when the channel impulse-response has a finite support, that is, lasts for a finite amount of time. Equivalently, the channel
can be said to be frequency-selective, that is, the transfer function depends on the
considered frequency. The impact of the delay dispersion on a specific system is
given by the product of the delay spread with the system bandwidth. If that product is much less than unity, then delay dispersion has little eect on the system
design [104]. The larger this product, the stronger the impact of the delay dispersion on the system performance. In UWB systems with large absolute bandwidth,
this impact is consequently very high. We find the following.
(1) The available delay diversity order is very high. A Rake receiver (or equivalent structure) with a large number of fingers can eectively eliminate

68

UWB propagation channels

fading [105]. It is well known that appropriately combining the signal copies from many independently fading processes approximates an
AWGN signal [106].
(2) When Rake receivers are used, the number of fingers required to collect
all multipath energy is considerable. Reference [107] observed that the
number of MPCs that are within 10 dB of the maximum is 34 in LOS,
but 15100 in NLOS situations. Thus, practical receivers, which have a
finite number of Rake fingers, suer from suboptimal energy collection.
Thus, one of the greatest strengths of UWB, namely, its fine delay resolution, can become a weakness. Solutions to this problem include the use
selective Rake receivers that collect only the instantaneously strongest
paths [108, 109], transmitted-reference schemes [108, 110], orthogonal
frequency division multiplexing [111], and Rake receivers with a very
large number of fingers that are made practical by using single-bit A/D
converters [112].
In narrowband channels, delay dispersion stems only from the multipath
propagation, that is, the arriving signal can be viewed as a sum of weighted and
delayed echoes of the transmit signal. In UWB channels, delay dispersion can also
stem from the fact that the interaction between the waves and a single object (e.g.,
diraction at a corner) is frequency-selective, see Section 2.3. We also find two
other important dierences to narrowband channels:
(1) the impulse-response of the channel can become sparse, in other words,
there are empty delay bins (bins containing no energy) between the
arriving multipath components,
(2) only few multipath components fall into each delay bin; that implies that
the central limit theorem is not fulfilled anymore, and the amplitude
fading statistics for each resolvable bin are no longer Rayleigh.
2.5.2. Arrival statistics of multipath components
2.5.2.1. Poisson-based models
We first turn our attention to the statistics of the arrival times of the MPCs, while
the path amplitudes and the distortions of each MPC by propagation eects will be
treated later. The delays of the MPCs are determined by the runtime through the
propagation environment. In the most simple case, where only single-interaction
processes occur, those runtimes can be uniquely associated with the locations of
the interacting objects (IOs).17 For multiple interactions, the relationship becomes
more complicated.
For a random model, the simplest assumption to make is that IOs are distributed in the environment uniformly according to a (multidimensional) Poisson
process. This assumption leads to a distribution of the MPC arrival times Tm at
17In many papers, the word scatterer is broadly used to describe the objects that interact with
the propagating waves. As the interaction processes can be specular reflection, diuse scattering, or
diraction, we use here the more general term interacting object (IO).

Andreas F. Molisch et al.

69

the receiver as a nonhomogeneous one-dimensional Poisson process with an arrival

rate that increases with delay. The increasing arrival rate arises because reflections
with a large delay value arise from the many IOs located near an ellipse with large
circumference, while reflections with a small delay value arise from IOs located
near a smaller ellipse, which are fewer in number. In fact, in the two-dimensional
case with a uniform density of IOs, the average number m(t) of MPC arriving on
or before t seconds after the time of arrival of the line-of-sight path is given by
m(t) =

22
Dc t t t02 ,
4

(2.110)

where D is the spatial density of IOs in the environment, c is the speed of light,
and t0 is the line-of-sight propagation time from the transmitter to receiver. The
arrival rate v(t) is the derivative of (2.110), given for t > t0 by


v(t) =

Dc2 2t 2 t02
d

.
m(t) =
dt
4 t 2 t02

(2.111)

Note that v(t) is an increasing function of t, that is, the density of arrivals increases
(on average) with increasing delay.
However, this phenomenon of increasing MPC density with increasing delay
is counteracted by two other physical factors. First, the amplitudes of the MPCs decrease with delay, and second, the paths begin to overlap as their density increases.
Thus, many engineers, starting with [113], have found that the more familiar homogeneous Poisson process, characterized by a constant arrival rate , suces,


P k k1 > t = exp(t).

(2.112)

In a slight abuse of notation, we can write that the probability density function for
the arrival of path k (given a certain arrival time k1 for the previous time) is




p k |k1 = exp k k1 ,

k > 0.

(2.113)

For many environments, both indoor and outdoor, the assumption of uniformly distributed scatterers is not true. Rather, it has been observed frequently
that MPCs arrive in clustersthat is, several MPCs arrive shortly one after the
other, then there is a longer pause, and then another group of MPCs arrives.
One mathematical model that reflects this clustering is the K model, which
was first suggested for the outdoor case, and popularized for the indoor case by
[114]. It defines two states: state A, where the arrival rate of paths is , and state B,
where the rate is K. The model starts in state A. If a path arrives at time t, then a
transition is made to state B for a minimum duration . If no path arrives during
that time, the model reverts to state A; otherwise, it remains in state B. K
models were used for UWB channels in [115, 116]. Note that both the number of
clusters and the duration of the clusters become random variables in this model.

70

UWB propagation channels

Amplitude

Delay

1/
1/

Figure 2.6. Principle of the Saleh-Valenzuela model.

An alternative model for the clustering is the Saleh-Valenzuela (SV) model

[94], which uses the following discrete-time impulse-response:
h(t) =

L
K

ak,l t Tl k,l ,

(2.114)

l=0 k=0

where ak,l is the tap weight of the kth component in the lth cluster, Tl is the delay
of the lth cluster, k,l is the delay of the kth MPC relative to the lth cluster arrival
time Tl . By definition, we have 0,l = 0. The distributions of the cluster arrival
times and the ray arrival times are given by Poisson processes




p Tl |Tl1 = exp Tl Tl1 ,



p k,l |(k1),l

 

= exp k,l (k1),l ,

l > 0,
k > 0,

(2.115)

where is the cluster arrival rate, and is the ray arrival rate. Figure 2.6 shows the
dierent parameters and their meaning. The SV model has been used in several
papers, among them [117119].
When comparing the K model and the SV model, we find some important
dierences. In the K model, the number of clusters is a random variable, whose
realization is determined by how often system enters into state B; in the classical
SV model, the number of clusters is a constant that is a model parameter.18 In the
SV model, the clusters can overlap; if the product of cluster arrival rate and cluster
delay spread is much larger than unity, it even becomes dicult to see with the
naked eye where one cluster begins and another one ends.
Due to these reasons, the extraction of the number of clusters in an SV model
is a rather ambiguous procedure. It is often a somewhat arbitrary decision whether
to consider a (reasonably continuous) power delay profile as a superposition of
several closely spaced clusters, or as a single cluster. This has naturally a great impact on the number of clusters as well as the intercluster decay time constants.
18Modifications of the SV model exist that assume the number of clusters to be a Poisson-

distributed random variable.

Andreas F. Molisch et al.

71

Spatially resolved measurements, like those of [101], are helpful in resolving this
ambiguity, as the directions of arrival oer an additional domain from which to
determine whether MPCs are clustered or not.
The number of clusters is a function both of the measurement bandwidth,
and of the considered environments
(i) Reference [79] found only one cluster for 1 GHz bandwidth in baseband
in an indoor oce environment.
(ii) Reference [77] found between one and three clusters for 2.4 GHz bandwidth centered around 5 GHz in indoor oce environments.
(iii) Reference [75] measured mostly one cluster with a measurement bandwidth of 1.25 GHz around a center frequency of 5 GHz; for indoor residential and oce environments. Two clusters were found in some special scenarios, especially corridors.
(iv) Reference [101] measured the number of clusters in oces and residential environments. As mentioned above, those measurements also used
the spatial domain, which resulted in a finer resolution of clusters. They
found 12 clusters in oce LOS, up to 4 clusters in oce NLOS, 1 cluster in residential LOS, and 12 clusters in residential NLOS. It is also
interesting that the number of clusters slightly decreases for higher frequencies.
Measured cluster interarrival times are typically in the range of 1050
nanoseconds ([118]: 45 nanoseconds in oces at baseband frequences, [117]: 16
25 nanoseconds in NLOS and 39 nanoseconds in LOS at 5 GHz center frequency,
[119]: 1114 nanoseconds mean arrival time in NLOS and 60 nanoseconds in LOS
at 5 GHz center frequency.
Mean interpath arrival times, 1/, show a wide variation, from 0.18 nanoseconds in NLOS situations to more than 20 nanoseconds in LOS situations [117].
It is noteworthy, however, that these values depend on the dynamic range of the
measurement equipment, the SNR, and the evaluation procedure. Furthermore, it
is dicult even with high-resolution algorithms to clearly resolve paths that have
a delay dierence much smaller than the inverse measurement bandwidth; thus
measurements with lower bandwidth lead to higher interarrival times. Widely differing values (between 1 and 60 nanoseconds) have been reported for the intracluster decay constants .
Some authors have also analyzed the number of significant MPCs, by which
they mean the number of MPCs that is above a certain threshold, or the number
of components that carry a certain percentage (usually 90%) of the total energy.
One example for such evaluations is [115], where a mean number of 36 MPCs was
found for NLOS 04 m distance; the probability for a bin to be occupied by a significant MPC decreases from 0.6 for zero excess delay to 0.2 for 40-nanoseconds
excess delay for a 04 m distance, and falling from 0.8 at 0-nanoseconds to 0 at
100-nanoseconds excess delay for NLOS at 410 m. Note that this is a remarkable
dierence to modeling all components, where the occupancy probability increases
with increasing delay [84]. An alternative definition for the number of significant
MPCs counts the MPCs that are no more than 10 dB below the maximum: [107]

72

UWB propagation channels

measured 3-4 in LOS, but 1530 in residential NLOS, and 50100 in oce NLOS
situations.
2.5.2.2. Power delay profile
Besides the arrival times, we also have to define the power decay of the multipath components. The PDP is defined as the squared magnitude of the impulseresponse, averaged over the small-scale fading [12]. When assuming furthermore
that the channel is ergodic and any channel changes in the considered time scale
come only from the small-scale fading, the PDP becomes [104]19
2 *
)
P() = Et h(t, ) ,

(2.116)

where Et {} denotes expectation over time.

The most common model is an exponential decay, so that


P() =

exp
U(),
d
d

(2.117)

where U() is the Heaviside step function, and d is the decay time constant.
In an SV-type model (2.114), where multiple clusters are present, it is usually
assumed that each cluster is exponentially decaying, so that


) 2 *

T
E ak,l  = 0 exp l exp k,l ,

(2.118)

where 0 is the mean energy of the first path of the first cluster, is the intercluster
decay time constant, and is the intracluster decay time constant.
Several measurement campaigns give only the rms delay spread, which is defined as the second central moment of the PDP.20 Reference [115] shows a delay spread of 9 nanoseconds at 2 m for NLOS and 4 nanoseconds mean (delay
spread) for NLOS, and a slight increase of the delay spread with distance (about
1 nanoseconds/m). Reference [120] measured rms delay spreads of 10 nanoseconds in LOS, and 22.7 nanoseconds in NLOS; the delay spread being strongly correlated with the path loss, and thus with the distance. Reference [107] measures
an rms delay spread of 10.9 nanoseconds in LOS, 16.5 nanoseconds in NLOS, and
22.4 nanoseconds in hard NLOS in an oce, and 12.2 nanoseconds in LOS, 9.9
nanoseconds in soft NLOS, and 12.9 nanoseconds in hard NLOS in residential.
Reference [84] found a 14 nanoseconds decay time constant. All those measurements were done in the microwave range. Low-frequency (baseband) measurements in an indoor oce environment indicate delay spreads around 40 nanoseconds [79]. Low-frequency measurements in outdoor forest areas show delay
19note that the definition of power delay profile in UWB channels is still subject to debate.
20For the case of a single cluster with exponential decay, the rms delay spread is identical to the
decay time constant. However, for more general models (like SV), the rms delay spread is an insucient
description of the delay dispersion.

Andreas F. Molisch et al.

73
85

Single
dominant
paths

Mag(rms(S21 )) (dB(10 GHz))

90

Oce NLOS

95

Diuse
multipath

100
105
110

Shortest
path

115

50

100
Delay (ns)

150

Figure 2.7. Measured indoor multipath intensity profile in the microwave range, from [84].

spreads between 30 and 50 nanoseconds in [121]. It is also notable that the delay spread is itself a random variable that changes between dierent rooms as well
as dierent environments.
The delay spread and the PDP of ultra-wideband channels show little dierence to the narrowband channels. Thus, the extensive results available from those
narrowband measurements (for an overview, see, e.g., [7]) can be applied also in
this case.
2.5.2.3. Semideterministic modeling
References [84, 122] have independently suggested a somewhat dierent modeling
approach. In a first step, they establish deterministic components that are derived
from an (existing or virtual) geometry, by means of ray tracing. For outdoor LOS
environments, two deterministic components (direct wave and ground wave) are
often sucient [122], while for indoor environments, components that have undergone up to three reflections have to be taken into account [84], see Figure 2.7.
In addition to the deterministic components, a Rayleigh-distributed clutter describes the contributions that stem from diuse scattering and other propagation
paths that are not covered by the ray tracing.
The model of [123] thus proceeds in the following steps.
(i) Compute the contributions from ray tracing in a simplified model geometry (e.g., box), taking into account up to threefold reflections. Limit
the number of considered rays to K, and compute the attenuation for
each of those rays from a power law dn (the parameter n is a model
parameter).

74

UWB propagation channels

(ii) Each of the rays is associated with a cluster of Rayleigh-fading MPCs;
each cluster is exponentially decaying.

2.5.3. Fading of the multipath components

In the previous subsection, we have analyzed the mean behavior of the channel,
that is, when averaged over measurement points within a small area. In this subsection, we analyze the statistical behavior of the amplitudes within such a small
area. We can understand the occuring phenomena better by dividing the delay
axis into bins with a width that is the inverse bandwidth. Components of the
impulse-response that fall into one delay bin cannot be resolved by the system,
and are thus seen as a single (but fading) component.
In conventional narrowband, bandpass wireless systems, it is well established
that the amplitudes in delay bins are Rayleigh-fading


pdf(x) =

x
x2
exp 2 ,
2

(2.119)

or possibly Rician-fading for the LOS component. The Rayleigh fading is created
by the superposition of many multipath components within each delay bin, so
that the in-phase and quadrature-phase components each are Gaussian. In UWB
systems, the number of components falling within each delay bin is much smaller,
which leads to a change in the statistics. Various alternative distributions have been
used in the literature.
Nakagami distribution. This amplitude distribution was suggested in [79] (see
Figure 2.9), and was also confirmed in [77], and by [84] for bins with large delays. The pdf of the amplitudes is given as


pdf(x) =

2
m
(m)

m

x2m1 exp

m 2
x ,

(2.120)

where m 1/2 is the Nakagami m-factor, (m) is the Gamma function, and
is the mean-square value of the amplitude. The parameter is the mean-square
value = r 2 , and the parameter m is
m= 

2
r2

2 .

(2.121)

It is straightforward to extract those parameters from measured values. When the

amplitude is Nakagami-distributed, then the power is Gamma-distributed.
Rice distribution. This distribution has been suggested by [83], as well as by [84]
for bins with small delays in LOS situations


pdf(x) =

 

x
x2 + A2
xA
exp
I0 2 ,

2
2 2

(2.122)

Andreas F. Molisch et al.

75
30

Occurrences (%)

Occurrences (%)

30

20

10

0.01
Power (a.u.)

20

10

0.02

0.01
Power (a.u.)

(a)

0.02

(b)

Figure 2.8. (a) Experimental histogram of received powers, compared to (b) a histogram derived from
a theoretical Gamma distribution with m = 3.1.

where A is the amplitude of the dominant component, Kr = A2 /(2 2 ) is the Rice

factor, and I0 is the zero-order modified Bessel function of the first kind. Nakagami and Rice distributions have a quite similar shape, and one can be used to
approximate the other. For m > 1, the m-factor can be computed from Kr by

2
Kr + 1

,
m=

(2.123)

2Kr + 1

while

Kr =

m2 m

.
m m2 m

(2.124)

Note, however, that if those equations are used, the resulting pdfs have dierent
slopes close to r = 0, as well as for very large r.
Lognormal distribution. The lognormal distribution has the advantage that the
fading statistics of the small-scale statistics and the large-scale variations have the
same form; the superposition of lognormal variables can also be well approximated by a lognormal distribution, its pdf is given as
!

pdf F (F) =

10 log10 F 2 dB
20/ ln(10)

exp
FF 2
2 F2

2 "

(2.125)

where F is the variance of the local mean F, and dB is the mean of the values of
F in dB. The lognormal distribution was suggested by [119]; also Ghassemzadeh
et al. [75] use a similar modeling approach: they model the deviation of the
local MIP (multipath intensity profile, |h(t, )|2 around its expected value PDP)

76

UWB propagation channels

as lognormal. For the LOS case, the model uses

for i = 0,
 
=
C i + S i for i = i,

(2.126)

where C is a Gaussian-distributed random variable21 N(6.38, 1.98), = N(0.82,

0.25), S is a correlated Gaussian process with mean 0.41 dB, and the variance of
S is N(6.86, 0.92), all in dB ( in dB/ns). S is assumed independent from one delay
to another. For the NLOS case, the result is similar
 

Pr (t) = i + S i ,

(2.127)

where = N(0.91, 0.13), S is a correlated Gaussian process with mean 0.41 dB,
and the variance of S is N(7.2, 0.88).
POCA and NAZU distributions. POCA and NAZU distributions, which were theoretically derived as distributions for a small number of components without
(POCA) and with (NAZU) dominant component, were suggested in [124]. The
POCA distribution is defined as


pr (r) =

2K 2 (2n + 1)
r
a2
,

(n+1)/2 Pn 2
2

a b2
a2 b2

(2.128)

where Pn (x) is the Legendre function,

K(2n + 1) = 
a=1+

1
(n + 1)

,
(2n + 1) (2n + 1)/2

r2
2 (2n + 1)
b=

r4
8 4 (2n + 1)2

r4
8 4 (2n + 1)2

(2.129)

and n is the number of paths. The NAZU distribution adds one dominant component.
Weibull distribution. Reference [76] models the deviation of the local MIP
around its expected value by a Weibull distribution


pX (x) =

1

exp

 1

for x 0,

(2.130)

where the parameters and depend on the environment.

Several papers have found that the fading depth increases with increasing delay. This is also physically intuitive. For small delays, there is only a small number of possible propagation paths that have a delay within a considered delay bin.
21Dierent realizations of C should be used in dierent houses.

Andreas F. Molisch et al.

77

Mean of m

4.5
2.5
0.5

50

100
150
200
Excess delay (ns)

250

300

Variance of m

(a)
4
2
0

50

100
150
200
Excess delay (ns)

250

300

(b)
Figure 2.9. Mean and variance of the Nakagami m-factor as a function of the excess delay for a baseband (< 0.5 GHz) measurement (according to [79]).

For larger delays, more (but weaker) MPCs fall into one bin, so that the fading
is more Rayleigh-like. Reference [79], which modeled the amplitude statistics as
Nakagami, reflects this fact by treating the m-factor as a random variable whose
mean and variance decrease with increasing delay. Similarly, [83] found the Rice
factor to be exponentially decreasing with delay.
Apart from the amplitudes, also the phase (for passband channels) or polarity
(for baseband channels) of the MPCs has to be modeled. For passband channels,
a uniform phase distribution is commonly assumed. For baseband channels, the
polarity of the MPCs is assumed to be positive or negative with equal probability,
except for the LOS components that always have positive polarity. Another important point is whether the fading in adjacent delay bins is correlated. Reference [76]
found that components immediately following the LOS have a high probability of
having the same polarity as the LOS.
The above equations have described the fading of the MPCs. In most indoor
models, the MPCs are grouped in clusters. These clusters can also be fading, typically due to shadowing. An example of how this cluster fading can be combined
with the MPC fading is described in Section 2.6.2. As the sum of the cluster powers
is the total power (assuming that the fading of the clusters is statistically independent of each other), this fading is also related to the shadowing of the total signal
as described in Section 2.4.
2.5.4. Frequency-domain modeling
The previous subsections dealt with the modeling in the delay domain. Of course,
it is also possible to model the channel transfer function. This was done in a purely

78

UWB propagation channels

phenomenological approach by an autoregressive model in [125]:



1 H fi , t, d = ni 2 H fi1 , t, d 3 H fi2 , t, d ,

(2.131)

where ni are i.i.d. Gaussian variables with zero mean and unit standard deviation,
and the i are the parameters of the model that have to be extracted from the measurements. Autoregressive models have been suggested also for narrowband channels, but have not gained widespread popularity, due to diculties in interpreting
them physically.
2.5.5. The impact of the discrete-time impulse-response model
A fundamental assumption of the channel model given in (2.114), consistent with
nearly all channel models since Turins description of the tap-delay line model
[113], is that the received signal is the sum of a finite number of scaled, phaseshifted, and delayed versions of the transmitted signal. In other words, the observed signal is a superposition of a number of undistorted versions of the transmitted pulse. However, as was discussed in Section 2.3, a UWB pulse will often exhibit frequency-dependent distortion on a per-path basis.22 This distortion can result from propagation through materials, diraction, the antenna, or other mechanisms. This fact is discussed in detail for diraction mechanisms in Section 2.3;
in the current subsection, we concentrate on the reflection and transmission processes and the antennas.
In this section, we will investigate the impact of this frequency dependence on
the use of discrete tap channel models of the type given in (2.114). There are two
basic questions that should be addressed in this discussion. (1) What is the direct
impact of our assumption on the resulting model? (2) How does this distortion
influence our interpretation of the model? We will address question (1) first.
As stated, there are several potential sources of pulse distortion during propagation. One such source is the distortion caused to a pulse as it passes through
materials. A detailed study of the frequency-dependence of common building materials was carried out by Virginia Tech as part of the DARPA NETEX program
[126, 127]. We apply the results here to demonstrate the impact that materials
can have on UWB signals and the discrete-time channel model. For example, consider the impact that brick has on a UWB pulse as it passes through the material.
Figure 2.10 plots the received line-of-sight pulse using a TEM horn antenna with
and without the material present [126]. The original pulse has a pulse duration of
roughly 300 picoseconds and a 10 dB bandwidth of approximately 7 GHz. Clearly
the frequency-dependent nature of the material has caused pulse distortion as can
be seen in Figure 2.10.
22Note that we are not referring to frequency-dependent distortion of the entire received signal.
This is clearly present in even moderate bandwidth channels when the delay spread is large. We are
referring specifically to per-path frequency distortion.

Andreas F. Molisch et al.

79

0.4

0.8

0.3

0.6

0.2

0.4

0.1

0.2

0.1

0.2

0.2

0.4

0.3

0.6

0.4

0.8

0.2

0.4 0.6
t (s)
(a)

0.8

1
109

0.5

0.5

1
t (s)

1.5

2
109

(b)

Figure 2.10. (a) Original and (b) distorted UWB pulse after passing through brick.

The classic tapped-delay line channel model cannot explicitly account for this
distortion without modification. However, the model can account for the distortion implicitly. That is, physically, there is a single path that exhibits frequency
distortion. Thus, using a single path of the form given in (2.114), we cannot represent the distorted pulse. However, the model can accurately represent the pulse
if we relax the restriction that we use a single path. We will demonstrate this in
a moment. However, first we must discuss the creation of tap-delay line models
from measurement data.
Time-domain channel sounding techniques are discussed in detail in Section
2.2. A common method for obtaining the channel impulse-response from the
measurement data is deconvolution of the pulse shape from the received data.
This ignores angle-dependent antenna eects, but allows for a reasonable approximation of the channel impulse-response [126]. Typical frequency domain deconvolution techniques result in a continuous-time impulse-response due to the
band-limited nature of the original measurement data. The resulting impulseresponse must be binned in order to create a discrete impulse-response similar
to (2.114). An attractive alternative is to perform time-domain deconvolution using the CLEAN algorithm [118, 128, 129]. A full discussion of CLEAN is beyond
the scope of this work, but the algorithm is more amenable to the tap-delay line
model since it assumes that the channel impulse-response is a sum of scaled and
delayed impulses.
Thus, by deconvolving the undistorted pulse from the received signal, the
distorted pulse shown in Figure 2.10 can be represented by a sum of scaled and

80

UWB propagation channels

0.6
0.5
0.4
0.3
0.2
0.1
0
0.1

4
t (s)

5
1010

Figure 2.11. Discrete-tap channel model created by CLEAN algorithm using the distorted pulse after
passing through plywood.

delayed versions of the line-of-sight pulse. To demonstrate this, the CLEAN algorithm was used to deconvolve the LOS (i.e., undistorted) pulse in Figure 2.10(a)
from the received signal given in Figure 2.10(b). The resulting impulse-response is
plotted in Figure 2.11. This impulse-response was then convolved with the undistorted pulse and plotted in Figure 2.12(b) along with received signal. We can see
that the modeled signal very closely represents the true received signal (the mean
square-error between the received pulse and the reconstructed pulse is 4
10{4}). However, we require a 3-tap channel to do so, when the true physical
channel has only a single path. Thus, frequency distortion can be modeled using
the classic method, but one must take care in interpreting the results. There are not
3 true paths in the channel, despite the fact that the impulse-response contains 3
taps.
As stated, this is specifically due to the ultra-wideband nature of the pulse.
For moderate bandwidths, this is not a concern. For example, compare the previously described UWB pulse with a 10 MHz Gaussian pulse centered at 5 GHz
shown in Figure 2.13(a). The resulting pulse after passing through brick appears in
Figure 2.13(b). The pulse is not distorted which is clear from the impulse-response
generated by CLEAN shown in Figure 2.14. Thus, traditional carrier-based communication systems would typically not experience frequency-dependent distortion passing through brick. This analysis was repeated using the UWB pulse described previously for several common building materials whose frequency characteristics are presented in [126, 127]. Table 2.3 presents a synopsis of the discretetime CIR for several of these materials. Specifically, the table presents the number
of components in the impulse-response within 10 dB of the strongest component
and the percent of the CIR energy in the dominant component. A large number of components reflect the fact that the discrete tap model will exhibit more
paths than actually exist. We will refer to such paths as phantom paths since the
model includes them to represent the per-path frequency distortion rather than
independent temporal echoes. The percent energy measure describes the relative

Andreas F. Molisch et al.

81

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

0.1

0.1

0.2

0.2

0.3

0.3

0.4

0.4

0.5

0.5

1
t (s)
(a)

1.5

2
109

0.5

0.5

1
t (s)

1.5

2
109

(b)

Figure 2.12. (a) Distorted UWB pulse after passing through brick and (b) pulse produced from
CLEAN.

strength of the phantom paths. A large percent energy measure signifies that the
phantom paths are weak relative to the main path which represents the actual temporal echo.
The preceding discussion can be applied to several aspects of UWB propagation. For example, consider the impact of a UWB pulse reflecting from a metalic
cylinder. Such an experiment was conducted by the Virginia Tech Antenna Group
(VTAG) as part of the DARPA NETEX project as explained in detail in [126].
Measurements were taken in an anechoic chamber to characterize the distortion
caused by reflections from common shapes. An example of LOS pulse and the
pulse received after reflecting from a hemisphere with a 6-inch diameter is plotted
in Figure 2.15. The transmit and receive antennas both pointed at the hemisphere
to ensure the hemisphere was in the boresight of both antennas. Additionally, nonreflected (i.e., direct) paths were removed by taking a measurement without the
hemisphere present. The received pulse after reflection is essentially a single specular component (which is inverted due to the reflection) along with a resonant
component as can be seen in Figure 2.15. The classic model can easily represent
the specular component, but does not assume the existence of the resonant component. Again, using the CLEAN algorithm results in a discrete-time model which
has significantly more than one path as shown in Figure 2.16. While the additional
paths are well below the specular path (more than 12 dB), clearly the impulseresponse must contain a large number of additional phantom paths in order to
accurately represent the received signal. A third cause of frequency distortion was
explained in detail in Section 2.3.

UWB propagation channels

0.25

0.25

0.2

0.2

0.15

0.15

0.1

0.1

0.05

0.05

Amplitude

Amplitude

82

0
0.05

0
0.05

0.1

0.1

0.15

0.15

0.2

0.2

0.25

0.5

1.5
t (s)

0.25

2.5
107

0.5

(a)

1.5
t (s)

2.5
107

(b)

Figure 2.13. (a) 10 MHz pulse after passing through plywood and (b) pulse produced from CLEAN.

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

1 0.8 0.6 0.4 0.2

0 0.2 0.4 0.6 0.8

t (s)

Figure 2.14. Discrete tap channel model created by CLEAN algorithm using the distorted 10 MHz
pulse after passing through plywood.

As discussed, diraction on a single path can often be modeled in the frequency domain as


H() = e j

e j ,

(2.132)

Andreas F. Molisch et al.

83

Table 2.3. Discrete channel representation of UWB pulse after passing through common building
materials.
Energy in dominant component

Components in the IR

Brick wall
Cloth partition

87%
98%

3
2

Wooden door
Concrete block
Structure wood

98%
90%
100%

2
3
1

0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

Amplitude

Amplitude

Material

0
0.1

0
0.1

0.2

0.2

0.3

0.3

0.4

0.4

0.5

400 500 600 700 800 900

(a)

0.5

400 500 600 700 800 900

(b)

Figure 2.15. (a) Line-of-sight signal; (b) signal reflected from a 6-inch diameter hemisphere.

where c is the center frequency and depends on the diraction being modeled. For example, when the pulse experiences single edge diraction, it can be
shown that = 0.5. Consider the pulse shown in Figure 2.17(a) and the resulting
pulse from single edge diraction given in Figure 2.17(b). Again, the frequencydependent distortion can not be explicity accounted for in the model of (2.114).
One approach to include this eect is the GTD approach described in Section 2.3.
Alternatively, we can simply use multiple frequency independent paths in the
discrete-time model to represent the single frequency distorted path. Again using
the CLEAN algorithm, we can create a discrete-time impulse-response to represent the distorted pulse. The resulting impulse-response is given in Figure 2.18. In
this case, two additional paths are required each of which is 15 dB lower than the
main path. However, it should be noted that this is highly dependent on the path
and the diraction experienced as discussed in Section 2.3.

84

UWB propagation channels

0.02
0
0.02

12.7 dB threshold

0.04
0.06
0.08
0.1

50

100

150

200

250

300

Figure 2.16. Impulse-response resulting from the CLEAN algorithm when the received signal is reflected from a hemisphere.

0.2

0.2

0.15

0.15

0.1

0.1

0.05

0.05

0.05

0.05

0.1

0.1

0.15

0.15

0.2

0.5

1
t (s)
(a)

1.5
109

0.2

0.5

1.5
t (s)

2.5
109

(b)

Figure 2.17. Received pulse (a) with and (b) without single edge diraction using values given in
[130].

Finally, an additional potential source of pulse distortion in UWB systems

comes from the antenna itself. The impulse-response of the antenna can change
with angle in both elevation and azimuth. Clearly this is most prevalent with directional antennas such as the Vivaldi or TEM Horn antenna. However, pulse distortion will also occur with omnidirectional antennas such as the bicone. Specifically,
the transmitted pulse at dierent elevation angles will be distorted as compared
to the pulse observed at boresight. This can be observed in Figure 2.19, where we

Andreas F. Molisch et al.

1.8

85

105

1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

109
t (s)

Figure 2.18. Discrete-time impulse-response due to a single path experiencing single edge diraction.

have plotted the received pulse using bicone antennas measured at dierent elevation angles [126]. For details of the measurements taken by the Virginia Tech
Antenna Group please, see [126]. Note that as one would expect, the amplitude of
the received pulse amplitude is reduced at larger elevation angles due to less antenna gain. However, the pulse is also slightly distorted as compared to the pulse
transmitted (or received) at zero-degree elevation. This brings up a fundamental question about the pulse shape in UWB systems. Since the pulse shape produced by the antenna is dierent for dierent angles, the notion of a single pulse
shape is problematic. A typical assumption is that the line-of-sight pulse (i.e.,
the pulse received on boresight) is the transmitted pulse shape. This pulse shape
can be used for matched-filter correlation or to create the discrete tap impulseresponse channel model from measurement data as discussed above. However,
care must be taken when making this assumption. Matched filtering with the incorrect pulse shape will lead to a loss in correlation energy. Using a single pulse
shape for channel modeling will lead to the generation of additional multipath
components which simply represent the pulse distortion rather than actual paths.
This can be clearly seen in Figures 2.20 and 2.21. Figure 2.20 plots an example
transmitted pulse from a bicone antenna at elevation angles of (a) zero degrees
and (b) 60 degrees. If we assume that the zero degree elevation angle pulse is the
nominal pulse shape, the pulse at sixty degrees can be represented using the discrete channel impulse-response shown in Figure 2.21. In other words, if a pulse is
received at an elevation angle of sixty degrees, the channel model will account for
the resulting pulse distortion by again using additional paths (15 in this case). This
can also be demonstrated for pulses arriving (or departing) at other angles. This
also indicates that future statistical channel models should include directional information for the multipath components, so that it can be more easily combined
with arbitrary antennas.
In conclusion, it is clear that single path frequency distortion can arise from
several dierent mechanisms when UWB signals are employed. This frequency

86

UWB propagation channels

10

103

Amplitude

6
4
2
0
2
4

10

11

12
109

t (s)
0 deg
30 deg

60 deg
90 deg

Figure 2.19. Received pulse from various elevation angles using biconical antennas.

10

103

10

10
(a)

12
109

103

10

12
109

(b)

Figure 2.20. Received pulse from elevation angles of (a) zero degrees and (b) 60 degrees using biconical antennas.

distortion is not explicitly represented by the discrete-time channel model that is

commonly assumed. However, as we have shown, the model can clearly represent
this distortion by the inclusion of phantom paths. Of course, care must be taken

Andreas F. Molisch et al.

87

0.5
0.4

Amplitude

0.3
0.2
0.1
0
0.1
0.2

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

108
t (s)

Figure 2.21. Resulting impulse-response for pulse at 60 elevation angle assuming boresight pulse
was.

when interpreting the model since each path in the model does not necessarily represent a separate independent path.23 Thus, for UWB systems, we must understand
that the channel model is more of a black box than in wideband systems where
we can generally view each path in the model as a physical transmission path. This
places more burden on the researcher to understand what characteristics of the
channel are most important to the issue under study. If the receiver is impacted
only by the delay spread, temporal correlation, or other statistics of the received
signal, the discrete time model is suciently accurate. However, if the model is
being used as a guide for the number of necessary Rake fingers or physical interpretations of the model, it may not be as useful.
2.6. Standardized channel models
2.6.1. Introduction
In this chapter, we present a channel model for indoor short range ultra-wideband
(UWB) radio channels. The model was developed in 20022003 through an industrial/academic collaboration under the auspices of the IEEE 802.15.3a standards
task group, for the purpose of aiding in the development of a standard for UWB
wireless personal area networks (WPANs). The IEEE 802.15.3a solicited contributions from a broad spectrum of experts from industry and academia in the formulation of this channel model. This resulted in a wealth of data, both empirical
and theoretical, that was brought before the IEEE 802.15.3a committee. (see [131,
132, 119, 133, 103, 123, 115, 134136] for more details; these documents, as well as
other related materials, are available in the July 2002 through March 2003 online
archives of the IEEE 802.15 WPAN, at http://grouper.ieee.org/groups/802/15/pub/
23
Therefore, the fading of dierent delay bins need not be independent. However, the one existing
investigation [79] of this correlation found the maximum correlation between dierent delay bins to
be less than 0.2.

88

UWB propagation channels

download.html.) A subgroup of the 802.15.3a examined these data, and formulated a parameterized mathematical channel model that provided a reasonable fit
to the empirical data, while maintaining the simplicity and ease of use that is required of a model for use by a standards group. This model was accepted by the
full standardization group in early 2003. (The final report of the channel modeling
committee is contained in [137]; also, see [102].)
The model is based upon the well-known Saleh-Valenzuela model [94], see
also Section 2.5, but with fading statistics modified to more accurately represent
the phenomenology associated with ultra-wideband signals. In addition to the
general parameterized mathematical model, the IEEE 802.15.3a also specified four
particular sets of parameters for the model, intended to represent channel conditions that the committee deemed especially relevant to the UWB WPAN operating
space.
In the sections, below we describe the general model, focusing primarily on
the mathematical structure of the model, and the distributions of the random
quantities that underlie it. We then review the four parameter sets for the model
that were defined by the IEEE 802.15.3a committee to represent the specific channel conditions relevant to UWB WPANs, and briefly discuss the fit of the model to
a set empirical data measured under those same conditions.
2.6.2. The mathematical model
The mathematical channel model consists of a multipath model based on the work
of Saleh and Valenzuela [94], a lognormal shadowing model, and a path loss model
of power-law type. All of these aspects have been addressed in Sections 2.4 and 2.5.
Here we summarize these considerations (as far as they are relevant to the standard
model), and describe the specific parameter choices used in it.
2.6.2.1. Multipath model structure
Mathematically, the multipath model takes the form of a clustered point process, in which the multipath components (MPCs, also referred to as rays) are
unevenly distributed in time. The physical intuition behind these models is that
the high spatial resolution of UWB signals allows the receiver to resolve multiple
reflections from each interacting object (IO) in the environment. Thus, the receiver perceives clusters of MPCs, each cluster corresponding to some IO in the
environment. The individual MPCs in a cluster arise from the complex scattering
process that takes place as the ultra-wideband signal interacts with the structure of
the IO.
Thus, the multipath model naturally decomposes into two pieces. First, there
is the cluster arrival process that describes the times of arrival of clusters at the
receiver, as well as their amplitudes. Then, for each cluster, there is a ray arrival
process describing the times of arrival and amplitudes of individual rays within the
cluster. We consider the cluster arrival process first.

Andreas F. Molisch et al.

89

Each cluster is associated with a delay time T and an amplitude . Obviously,

T is determined by the total distance from transmitter to IO to receiver, and
models the overall fading associated with the cluster. Thus, an individual cluster is
modeled as


cl (t) = l t Tl ,

(2.133)

where l indexes the clusters, and the cluster arrival process is given by

c(t) =

cl (t) =

l=0

l t Tl ,

(2.134)

l=0

where L is the total number of clusters. The parameters L, Tl , and l , l = 0 . . . L, are

random variables whose distributions must be specified. Specifics on the distributions of all model parameters will be given in Sections 2.6.2.3 and 2.6.2.4 below.
The arrival of individual MPCs within a cluster is modeled by a ray arrival
process. An individual MPC within cluster l is specified by its delay relative to the
cluster delay, and by its amplitude relative to the overall cluster fading. We also
include a reflection coecient p, equiprobably 1, to account for signal inversions
due to reflections. Thus, an individual ray within cluster l is modeled as


rk,l (t) = pk,l k,l t k,l ,

(2.135)

where k indexes the rays within the cluster, and the ray arrival process for cluster l
is given by

rl (t) =

K(l)

rk,l (t) =

k=0

K(l)

pk,l k,l t k,l ,

(2.136)

k=0

where K(l) is the total number of rays associated with cluster l.

Putting together the pieces now, cluster l along with its associated rays is modeled as
cl rl (t) =

K(l)

l pk,l k,l t Tl l,k ,

(2.137)

k=0

(the denotes convolution), and the overall channel impulse-response is given by

h(t) =

L

l=0

cl rl (t) =

L K(l)


l=0 k=0

l pk,l k,l t Tl l,k .

(2.138)

90

UWB propagation channels

2.6.2.2. Shadowing and path loss models

Shadowing is an overall scaling factor X applied to the channel impulse-response
as a whole:
h(t) = X

L K(l)

l pk,l k,l t Tl l,k .

(2.139)

l=0 k=0

The shadowing factor X is of course random.

The path loss model adopted by the IEEE 802.15.3a for the evaluation of standards proposals was the simple free-space model, given by (see Section 2.4)


PL(d) =

4d

2

(2.140)

where PL is the path loss, d is the distance, and the wavelength is taken to be
the geometric mean of the upper and lower 10 dB cuto frequencies of the signal.
While this model provided a sound basis for the relative comparisons of proposals, it is not suitable for system performance modeling, because it does not take
into account the variation in path loss that occurs in the diverse environments in
which UWB WPANs are expected to operate. A more realistic model that nevertheless preserves much of the simplicity of (2.140) would replace the free-space
path loss exponent of 2 by a random path loss exponent , whose distribution is
chosen based on measurement data. This model would be further calibrated by
the amount of path loss experienced at a fixed reference distance (say one meter),
again determined by measurements. We will return to this issue in Section 2.6.2.4
below.
2.6.2.3. Distributions of the multipath model parameters
The cluster arrival process (2.134) is characterized by the cluster arrival times Tl
and the associated cluster fading terms l , both of which are random. The familiar
homogeneous Poisson process (see Section 2.5), characterized by a constant cluster
arrival rate , was adopted by the IEEE 802.15.3a channel modeling subcommittee. Thus, the cluster interarrival times are assumed to be independent identically
distributed (i.i.d.) exponential random variables, with mean 1/:


P Tl Tl1 > t = exp(t)

(2.141)

is called the cluster arrival rate; its value depends on the RF environment.
As for cluster amplitudes, we generally expect them to decrease as cluster delay increases, on average. The IEEE 802.15.3a model specifies that the conditional
expected value of the cluster amplitude should decay exponentially with cluster
delay:


E l2 | Tl = 0 eTl / .

(2.142)

Andreas F. Molisch et al.

91

Here, 0 is a normalization factor, and is the cluster decay factor. It was

determined that the lognormal distribution (see Section 2.5) provided a good fit
to the measurement data available to the 802.15.3a group. Specifically, we have


l2 = exp ln 0

Tl 1 2
+ N,l ,

(2.143)

where {N,l : l = 0 . . . L} are i.i.d. standard normal random variables, and is

the lognormal standard deviation of the cluster fading. It is easily verified that l
defined by (2.143) satisfies the exponential decay condition (2.142). Moreover, the
cumulative distribution function (CDF) for the cluster amplitudes (2.143) is given
by
'

ln b2 /0 + Tl / + (1/2)2
,
P l < b | Tl =

(2.144)

where is the CDF of the standard normal distribution. In summary, the cluster
arrival process consists of a homogeneous Poisson process describing the cluster
arrival times, with associated cluster amplitudes that are lognormally distributed
and exponentially decaying with the cluster delay.
The ray arrival process (2.136) takes exactly the same formonly the distribution parameters change. Thus, the ray interarrival times are again exponential,
this time with ray arrival rate :


P k,l k,l1 > t = exp(t).

(2.145)

Again, conditional on the ray delay, the ray amplitudes are lognormally distributed,
with ray decay factor ,

2
k,l
= exp

k,l 1 2
+ N,k,l ,

(2.146)

implying that


2
| k,l = ek,l / ,
E k,l

(2.147)

where again {N,k,l : l = 0 . . . L, k = 0 . . . K(l)} are i.i.d. standard normal random

variables, and is the lognormal standard deviation of the ray fading. Relative ray
arrivals and amplitudes are completely independent of cluster arrivals and amplitudes in this model. It should be noted that the ray decay factor is the same for all
clusters in this model.
Finally, we need to specify the distributions of the number of clusters L, and
the number of rays K(l) for the lth cluster. Given the exponential decay of the cluster and ray amplitudes, the mathematically convenient choice would be simply to
let each of these parameters be infinite. Obviously this is not a satisfactory solution
for simulation purposes. A simple remedy in this case is to simply simulate clusters

92

UWB propagation channels

Table 2.4. Parameters for the multipath model.
Parameter

Symbol

Cluster arrival rate

Cluster decay factor
Cluster fading depth

Ray arrival rate

Ray decay factor
Ray fading depth

and rays until the exponential decay of the amplitudes renders them insignificant.
This was the approach taken in the reference simulation code presented in [137].
There, cluster arrival times Tl were limited to Tl < 10, and relative ray arrival
times were limited to k,l < 10.
In summary, the parameters describing the multipath model are given in Table
2.4.
2.6.2.4. Distributions of the shadowing and path loss parameters
The standard deviation of the logarithm of the shadowing term X in (2.139) is
referred to as the shadowing depth. The IEEE 802.15.3a committee adopted a lognormal shadowing model, with a shadowing depth of 3 dB. Thus the shadowing
term X in (2.139) would be given by
20 log10 (X) = 3Z,

(2.148)

where Z is a standard normal random variable. However, shadowing depth is itself

a random quantity that can vary from realization to realization, even in the same
environment [82, 74]. This insight can be incorporated into the path loss model
by making the shadowing depth a Gaussian random variable, with a prescribed
mean and standard deviation (see [102]). Expressed in dB, the shadowing term X
would thus be given by
20 log10 (X) = x Z,

(2.149)

where x is normal. The shadowing depth (in dB) itself is now random, given by


x2 = Var 20 log10 (X) | x .

(2.150)

In Section 2.4, we discussed the fact that for system performance modeling the
free-space path loss model adopted by the IEEE 802.15.3a committee (for proposal
evaluation only) should be modified to take into account the dependence of path
loss parameters on the environment. Instead of using a path loss exponent of 2,
a random path loss exponent should be used, whose distribution is dependent on
the environment. In addition, the value of the path loss at one meter, denoted by

Andreas F. Molisch et al.

93

Table 2.5. Parameters for the shadowing and path loss models.
Parameter

Symbol

Mean shadowing depth

Ex

Variance of shadowing depth

Mean path loss exponent
Variance of path loss exponent

Var x
E
Var

Path loss at one meter

PL0

Table 2.6. Channel characteristics of the data groups.

Data group

Environment

LOS, 0 to 4 meters

2
3
4

NLOS, 0 to 4 meters
NLOS, 4 to 10 meters
Very high multipath

m (ns)

rms (ns)

N10 dB

Ne

5.05

5.28

24

10.38
14.18

8.03
14.28
25

35

36.1
61.54

PL0 , should also be determined from measurements. Ghassemzadeh et al. in [82,

74] report on an extensive measurement campaign comprising 300 000 frequency
responses taken in a variety of residential settings, in both line-of-sight and nonline-of-sight scenarios. These measurements provide a good basis for modeling the
path loss as a function of environment. The parameters describing the shadowing
and path loss models are given in Table 2.5.
2.6.3. Operational scenarios of interest
The mathematical channel model just presented was formulated in light of several
sets of channel measurement data. These measurements encompassed a wide variety of operational scenarios, from residential to oce-type environments, in both
line-of-sight (LOS) and non-line-of-sight (NLOS) conditions. The IEEE 802.15.3a
committee categorized these measurements into four broad groups, representing
distinct UWB WPAN operational environments. Measurements representative of
each group were analyzed on the basis of four channel characteristics: mean excess
delay (l ), rms delay (rms ), mean number of rays within 10 dB of the strongest ray
(K10 dB ), and mean number of paths that capture eighty-five percent of the energy
in the channel (Ke ). The results are summarized in Table 2.6.
The 802.15.3a committee then attempted to fit the mathematical channel
model to each of these groups. The resulting channel models, respectively, designated CM1 to CM4, are described in Table 2.7, and the corresponding channel
characteristics produced by the model are described in Table 2.8. Comparing Tables 2.7 and 2.8 shows a relatively good fit between the model characteristics and
the corresponding characteristics of the measurement data.

94

UWB propagation channels

Table 2.7. Model parameters values corresponding to data groups 14.

Parameter

Symbol

CM1

CM2

CM3

CM4

Cluster arrival rate

Cluster decay factor
Cluster fading depth

(1/ns)

(dB)

0.0233
7.1
3.4

0.4
5.5
3.4

0.0667
14.0
3.4

0.0667
24.0
3.4

Ray arrival rate

Ray decay factor
Ray fading depth

(1/ns)

(dB)

2.5
4.3
3.4

0.5
6.7
3.4

2.1
7.9
3.4

2.1
12.0
3.4

Ex (dB)
Var x (dB)
E

1.6
0.5
1.7

2.7
0.98
3.5

2.7
0.98
3.5

2.7
0.98
3.5

Var
PL0 (dB)

0.3
47

0.97
50.5

0.97
50.5

0.97
50.5

Mean shadowing depth

Variance of shadowing depth
Mean path loss exponent
Variance of path loss exponent
Path loss at one meter

Table 2.8. Channel characteristics of the channel models.

Environment

m (ns)

CM1
CM2

LOS, 0 to 4 meters
NLOS, 0 to 4 meters

5.0
9.9

CM3
CM4

NLOS, 4 to 10 meters
Very high multipath

15.9
30.1

Channel model

rms (ns)

N10 dB

Ne

5.0
8.0

12.5
15.3

20.8
33.9

15.0
25.0

24.9
41.2

64.7
123.3

2.7. Body-area networks

2.7.1. Introduction
Wireless body area networking (WBAN) is an emerging 4 G technological concept
for short (15 m) and very short (below 1 m) range communication systems with
a wide range of potential applications including health-care, surveillance, emergency, entertainment, shopping, and so forth, [138140]. Figure 2.22 shows the
envisioned WBAN and its possible components.
The target characteristics of the devices operating within WBANs are: data
rates scalable from low (110 kbps) to very high (100400 Mbps), low-cost, lowpower, and small-size. These user devices, operation in the close proximity of the
human body, are mainly small body-worn, for example, sensors and tags, however
medium-size devices, such as PDA, mobile phone, and so forth, are also envisioned
to be part of a typical WBAN.
Due to the specifics of WBAN contrasting with the traditional wireless networks WLAN, GSM, UMTS, several technical challenges remain to be solved for
a successful deployment and integration of these systems into future 4 G wireless
heterogeneous networks.
The ultra-wideband (UWB) technology is one of the main candidates for
noninvasive WBAN due to its potential in fulfilling most of the requirements listed

Andreas F. Molisch et al.

95
Display

Microphone

Headset
Sensors #1

Main unit

Legacy devices

Sensors #2

Figure 2.22. Envisioned WBAN and its possible components.

above. The network-level issues for WBAN are strongly dependent on the wide variety and on the specifics of the user scenarios in which the devices operate. The
underlying reliable and optimized air-interfaces and adaptive medium access control schemes for WBAN cannot be successfully implemented without a thorough
analysis of the specific around-the-body radio propagation channels.
In typical WBAN scenarios, for example, for small-size, body-worn, and
hand-held devices, the radio propagation channel most likely will exhibit some
interesting characteristics such as highly scattered paths and antenna near-field
eects especially due to body-proximity conditions. The user proximity eects
for handheld devices have been extensively investigated for the GSM/ UMTS frequency bands, for example, [141144]. For WBAN, some of these aspects have
been addressed in UWB propagation investigations presented in [145148].
Due to the antenna near-field eects and its dynamics, the traditional channel measurement techniques are not directly applicable to the investigation of the
radio propagation within the WBAN and new methodologies have to be adopted
[145, 146].
Traditionally, the channel sounding measurements have a twofold purpose.
(i) Firstly, a radio channel characterization or model can be obtained for the
investigated environments. Traditionally these results are obtained from
measurements using antennas with known and close to ideal characteristics. Thus the antenna influence can be easily separated (de-embedded)
from the obtained results.
(ii) The second general purpose of radio channel investigations is the study
of the antenna eciency in a given operating scenario and environment.

96

UWB propagation channels

In this case, the antennas which are under test are the ones envisioned
to be used in the actual communication devices.
In the case of WBAN, the clear separation between these two types of radio
channel investigations is not possible due to the specifics of the propagation conditions. The de-embedding problem in UWB was discussed in Section 2.2. Ideal
UWB antennas are electrically large and are not suitable for on-body scenario
investigations. Angle-of-arrival measurements using these large antennas would
considerably distort the propagation conditions and yield unrealistic results for
WBAN due to the high spatial resolution10 cm down to 1 cmpossible to obtain with UWB signals. Thus the eects of the body-worn and electrically small
antennas can not be separated and have to be part of the radio channel characterization and model. The dynamics in the scenarios with human users make this
separation even more complicated than in static scenarios.
The main conclusion from this observations is that the eects of the smallsize antennas in realistic user devices have to be included in the developed radio
channel model [145, 146].
In the following sections, we will introduce and analyze the most important
aspects of radio propagation and antenna measurement techniques for WBAN.
The focus here will be on user scenarios with small- and medium-size devices, and
two main topics are discussed:
(A) antenna design and analysis for small-size devices,
(B) measurement methodologies required for an appropriate characterization of the WBAN radio propagation channel including body-proximity
and antenna eects.
2.7.2. Measurement techniques
2.7.2.1. Antenna designs
In this section, we address the common issues which arise in the design of UWB
antennas for WBAN applications. In practice, the WBAN UWB antennas have to
share some of the characteristics with the integrated antennas types known from
the existing cellular phones such as simplicity, small volume, high total eciency,
and stable radiation characteristics with respect to the body proximity. For sensor
networks, where the devices are potentially 34 times smaller in volume than a
cellular phone, the antenna design is even a more challenge task.
Directional antenna radiation patterns are especially important for bodyworn devices where the aim is to minimize the human-body-proximity eects,
thus increasing the power eciency of the device (battery lifetime). Furthermore,
in order to minimize the signal distortion, the UWB antenna (transmitter or receiver) must have a flat frequency response and linear phase over the frequency
band used. These nondispersive properties are important especially when impulseradio (IR-) based systems are considered, with fractional bandwidths typically
higher than 25% [149]. For any UWB antenna, the flat frequency response requirement cannot be obtained for all spatial directions. Generally, there is only one

Andreas F. Molisch et al.

97

Monopole

Feed line

Ground
plane

Figure 2.23. Example of planar UWB monopole antenna printed on one substrate. Frequency band:
3 GHz to 6 GHz; monopole size: 11 mm 17 mm; ground plane size: 30 mm 60 mm.

radiation lobe (direction) in which the nondispersive conditions are well approximated. This is particularly true for directive antennas, for example, the Vivaldi
antenna.
Due to their inherent structure, the planar monopole antennas are not directly suited to WBAN applications. The planar monopole design can, however, be
further modified to be low-profile. Figure 2.23 shows an example of a simple planar UWB printed monopole antenna for the frequency range of 3 GHz to 6 GHz.
Other examples of relatively small-size antennas are the diamond dipole antenna
[150] and the folded radiating element, a few millimeters thick [151]. Several other
UWB antenna designs are available, which also meet some of the imposed requirements. Main drawbacks of these antennas, from the point of view of commercial
WBAN applications, are their size and/or manufacturing cost.
All the antenna designs presented so far present radiation patterns similar
to the one of a traditional monopole/dipole antenna. As mentioned earlier, a directional antenna radiation pattern would be desirable for body-worn devices in
order to minimize the eects of the human body proximity. To obtain this, the
radiation towards the body needs either not to take place or to be reflected by
a reflecting antenna part. Promising structures for this are the microstrip patch
[152, 153] or slot [154, 155] antennas. Although more complex than the monopole designs, the patch structure can provide low profile geometries and relatively
small-volume antennas. Figure 2.24 shows the UWB design for a microstrip antenna [152].
Directional antenna radiation patterns are possible to realize using patch designs or other types with reflectors such as dipole above ground plane. Directivity
can be obtained if the antenna is large in the direction of the radiation, such as
horn antennas, and so forth. Antennas for WBAN, however, need to be low-profile,
therefore only antenna types including a reflector are of practical interest. Research
in this area is ongoing.

98

UWB propagation channels

Patch #2
Substrate #2
Foam #2
Patch #1
Substrate #1
Foam #1
Aperture
Ground plane
Feed substrate
Tuning stubs
Feed line

Figure 2.24. Example of aperture-stacked patch antenna for UWB WBAN (copyright Maciej Klemm
and Troster [152]).

2.7.2.2. Antenna analysis

As presented previously, the choice for a specific antenna design has to be based
on their main implementation requirements in the WBAN devices: small size in
terms of volume, total eciency, and stable radiation characteristics. The low-cost
requirement considered herein is only from the geometrical structure complexity
point of view and not necessary the type of materials used.
Electrically small antennas have generally low total eciency, when matching
is included in the analysis. Furthermore, the proximity of human body can significantly modify the impedance bandwidth and radiation characteristics. These main
observations lead to the natural conclusion that for an ecient design the antennas used in WBAN devices have to be designed and tested in their normal operating scenarios. Reliable, numerical methods, for example, based on finite-dierence
time domain (FDTD), need to be employed in order to obtain simulation results.
These methods permit modeling of geometrically complex device structures (i.e.,
not only the antenna element itself) including the nearby antenna surroundings,
such as the human body proximity, for example, [156, 157].
The situation is not simple when it comes to the decision on the human body
model to be used. Certainly, complex models of the human tissue exist, with or
without considering the dispersive eects, for example, [158, 159]. Moreover, full
body-phantom models, such as the one presented in [157], for throughout simulations of a WBAN could be used. The precision of the results, but also the complexity, increases proportional to the complexity of the chosen body/tissue model.
Thus one has to make an appropriate choice on the body/tissue model, depending
on the actual antenna characteristics and operating scenarios. More detailed body
models also limit the results to smaller groups of persons.

Andreas F. Molisch et al.

99

It is important to distinguish between body/tissue models for internal field

and external field analysis. If the internal fields are of interest, a rather accurate
and detailed phantom is needed whereas a simple (homogeneous) phantom can
be accurate model for the external fields [160].
For example, in the design of antennas for body-worn devices operating in
the near-field of each other, the simulation of a multilayered human tissue may be
required to correctly obtain the radiation/propagation processes. When only the
antennas far-field radiation characteristics are to be optimized the simulation of a
complex human tissue is not needed anymore. While in the former case a full body
model is needed at least for the area where the devices (antennas) are located, in
the later case one can use a truncated model only covering the body part where the
device is mounted.
A significant influence of the actual UWB antenna radiation characteristics
when used or mounted close to the human body is to be expected. In this respect
two important questions arise.
(i) How close to the human body these antennas are located?
(ii) What are the dynamics of the human-body-to-antenna configuration?
The first issue is relevant when optimizing antenna design for operation in the
close proximity or on the human body. This leads to a static analysis and optimization of a particular antenna type when mounted close to a dielectric structure.
The second aspect is related to the changing near-field eects on the radiation
characteristics of the antennas. The geographical dynamics of the human body
(or body parts: hands, fingers) relative to the antenna of a body-worn device will
reflect in the particular dynamics of the radio propagation conditions (channel),
thus influencing the performance of the communication system. The statistical
evaluation of these dynamics and variations from user to user is possible with
specially design measurement setups, for example, [160].
The small-size, sensor-like devices can be mounted directly in contact with the
human skin (e.g., for health-care monitoring systems) thus constant separation
distances of maximum 23 mm are expected [138, 139]. Another case is that of
antennas mounted in clothing (textile antennas) where the separation distance
can be larger, in the order of 16 mm, and possibly varying up to 1015 mm due
the movements of the user. Headset and similar devices have relatively larger and
constant separation distance from the human body of 510 mm. The scenario is
similar for devices located in pockets. For handheld devices, the situation is similar
to scenarios investigated for the GSM/UMTS cellular phones, for example, [143,
144, 160, 161], and the users hands or fingers were found to be located typically
between 5 mm and 40 mm distance from the antenna(s) in the terminal. In such
cases, the interaction distance variations are generally large, in the order of 10
30 mm [160].
Regardless of the scenario, the influence of the human body proximity must
be analysed and considered when optimizing the antenna design for specific applications. This has further impact on the design of multielement antenna systems,
for example, [141, 142, 161], with potential improvement of the throughput and
reliability of the communication devices used in the WBAN.

100

UWB propagation channels

2.7.2.3. The RF antenna connection

The simulations for antenna design presented in Section 2.7.2.1 can be performed
using complex or simplified models of the antenna structures and human body
tissue. Details such as ground plane and other components of the communication
device (housing, PCB, etc.) can be included to obtain even more realistic radiation
results. The next step in the antenna investigations is building and testing the actual antennas. Similar to the simulations, the actual communication device can be
emulated if necessary for a correct testing (casing, etc.). Experimental radio channel investigations (channel sounding) can then be carried out in typical scenarios
envisioned for the WBAN.
Section 2.7.1 presented the argumentation for the need of new measurement
methodologies, with the conclusion that the channel characterization for WBAN
has to include the influence of the small-size antennas typically used. In traditional radio channel sounding, the antennas used are connected to the measurement equipment with radio frequency (RF) cables. This setup is perfectly suited
for electrically large antennas, but not for electrically small antennas, as the radiation characteristics are significantly influenced by the size of the ground plane or
any conductive cables acting as large ground planes. Therefore, the length and position of the connecting RF feed cables has a significant influence on the antenna
radiation [162].
The ground plane on which the antenna element is mounted can be designed
and simulated to fulfill the purpose. Changes in the antenna radiation characteristics due to the size of the ground plane are deterministic and can be accounted for.
This is, however, not the case with the RF feed cable connecting the antenna element to the measurement system. The position of the cable is dicult to control
during the actual measurements involving test users and can change the radiation significantly, which then lead to unrealistic radiation patterns [147]. Currents
running on the cable in the close proximity of the human body can significantly
change the signal absorption and signals radiation main direction, level, and so
forth. Investigating the radio propagation channel without considering these aspects can lead to unrealistic results in terms of radio channel models and associated
performance of dierent communication systems.
The observations above lead to one natural conclusion: the influence of the
connecting RF feed cables or any other conductive cables connected to the device,
which does not belong to the final application, ought to be minimized if not completely eliminated. A practical setup was proposed to achieve this, namely, the use
of an RF on fiber optic (FO) link instead of the usual conductive cable feed connections [162]. Given the large bandwidth of the used optic fiber (above 10 GHz),
a similar setup can be successfully used for UWB radio channel investigations as
well.
When the antenna under test is used as transmitter antenna, the proposed
setup is relatively simple based on a commercial FO transmitter and receiver pair.
The optical receiver is a small detector diode which allows the emulation of smallsize devices including the antenna. The optic receiver can be battery powered, thus

Andreas F. Molisch et al.

101
Small-size UWB device
UWB antenna

FORX

Optic receiver (battery powered)

Fiber optic link

FOTX

Optic transmitter
Input RF signal

Figure 2.25. Fiber optic RF feed for small-size antennas. UWB frequency ranges can be easily accomodated due to the bandwidth of the fiber optic.

no conductive cables are needed to be connected to this side of the measurement

chain. For short-range propagation investigations (13 m), this solution is good.
For longer range investigations, that is, with higher path loss, additional signal
amplifiers are required at the output of the optic receiver, which increases the size
of the actual emulated device and power consumption of the setup. Figure 2.25
shows the FO setup with the antenna under test connected as radio transmitter.
Of course, the antenna under test can be used as receiving antenna and still be
connected with FO. In this case, however, the setup requires significantly more
hardware components along with a careful design as the optic transmitter (laser)
needs to be next to the antenna. The laser is significantly larger than the detector
diode used in the optic receiver.
To illustrate the importance of this practical aspect and of the novel RF on
FO setup, simulations have been performed for the radiation patterns of a small
UWB antenna (folded radiating element) [163] with and without the conductive
RF feed cable, respectively. Figures 2.26 and 2.27 show the obtained simulation
results in one planar cut in the 3D far-field total (power) radiation pattern of the
antenna. In all simulations, the antenna ground plane size was 25 mm 55 mm.
Figure 2.26 shows the result obtained without any conductive RF feed cable. This
setup, practically, corresponds to the described RF on FO antenna connection
setup. Figure 2.27 shows the dierent radiation patterns corresponding to dierent orientations of a 50 mm length and 1.2 mm diameter conductive RF feed cable.
Naturally, similar results are obtained for increasing the ground plane size of the
antenna in the corresponding direction.
As can be seen, the radiation patterns are not only dierent, but also the main
radiation direction changes significantly when the conductive RF feed cable is connected to the antenna element. Changes in the electrical matching of the small-size
antenna element were also disclosed by the simulations.

102

UWB propagation channels

90
120

60
0 dBi
10 dBi

150

30

20 dBi

180

330

210
240

300
270

Figure 2.26. Simulated far-field antenna radiation pattern at 4.5 GHz for a small-size antenna without
the RF feed cable. This setup, practically, corresponds to the described RF on FO antenna connection
setup.

An important aspect, not illustrated in Figures 2.26 and 2.27, is related to the
changes in the polarisation of the small-size antenna when conductive RF feed cables are used. The simulations showed that the polarisation in the direction of the
connecting cable becomes dominant. With a moving cable connection such as in
dynamic WBAN scenarios, the polarisation state will be signifcantly modulated
by the actual position of the RF feed cable. This eect is also considerably minimised by using the RF on FO setup.
The presented results can be generalised for any type of small-size, narrowband, and ultra-wideband antenna. The important parameter is the electrical size
of the conductive cable and/or ground plane. Due to the frequencies ranges in
UWB, the electrical size of practical devices considered in WBAN becomes a factor which cannot be neglected. In the UWB case, the additional changes of the
radiation pattern in the frequency domain have to be also considered.
2.7.2.4. On-body measurement setup
In order to highlight the particularities of the required WBAN radio channel measurement setup, as opposed to traditional propagation experiments, we introduce
a practical configuration used in [145, 146]. We base our discussions on an emulated WBAN scenario, composed of two body-worn terminals: a portable base
station (PBS) and a body-worn device (BWD) [145, 146, 139]. Two small-size,
low profile UWB antennas [163] have been selected due to their commercial availability at the time of the experiment. In practice, any suitable small-size antenna
type as presented in Section 2.7.2.1 can be considered. The BWD was emulated
as a small device (e.g., the size of a sports watch) while the PBS was emulated
as a larger device (e.g., the size of a mobile phone). Accordingly, the antennas
for PBS and BWD have been used mounted on dierently sized ground planes,
50 mm 110 mm and 25 mm 55 mm, respectively.

Andreas F. Molisch et al.

103
90
120

60
0 dBi
10 dBi

150

30

20 dBi

180

330

210
240

300
270
(a)

90
120

60
0 dBi
10 dBi

150

30

20 dBi

180

330

210
240

300
270
(b)

Figure 2.27. Simulated far-field antenna radiation pattern at 4.5 GHz for a small-size antenna with RF
feed cable in dierent positions: (a) vertical and (b) bent to the right. The position of the feed cable is
shown schematically.

Figure 2.28 shows the BWD and PBS antennas setups. At the PBS, one antenna element, from the original mount, was used in the measurements. At the
BWD side, the RF cable connecting the BWD antenna was replaced by an RF FO
link. At the PBS side of the measured link, the larger electrical size of the ground
plane allowed the use of simple RF cable connection without a significant radiation pattern distortion. Furthermore, the cable connecting to the receive antenna
was always kept vertical.
The investigations were carried out using several adult test persons with different body structures (weight and height). These data can provide initial statistics over dierent body structures for the investigated radio channel parameters
[145, 146]. The PBS was mounted on the left side of the test person at hip height.
For the BWD typical, static positions have been investigated on the left arm, right

104

UWB propagation channels

cm

PBS

BWD

Figure 2.28. Example of antenna setup for on-body measurements [145, 146].

arm, and torso. Three dierent positions of the arms were used. More details can
be found in references [145, 146].
Dynamicthus more realisticend-user scenarios can be investigated with
similar antenna configurations, provided the measuring equipment is capable of
capturing the required Doppler spread in the radio channel. A swept time-delay
correlation channel sounder with a sucient flat bandwidth (above 1 GHz) is suitable for these dynamic UWB radio channel measurements [138].
2.7.2.5. Measurements with body phantom
In general, body phantoms (or body-part phantoms) can be used in both antenna
radiation investigations and radio channel measurements. The discussion presented in Section 2.7.2.2 on the choice of the type and complexity of body phantom/tissue for simulations is relevant also when considering antenna and channel
measurements. For the case of radiation investigation of cellular phones (i.e., relatively narrow band compared to UWB) body phantoms have been successfully
designed and analyses, for example, [157, 160, 164, 161]. Further development of
these body (or body-part) phantoms has to be performed for the ultra-wide frequency band.
In the on-body UWB radio channel investigations described briefly in Section
2.7.2.4, several test users have been used in order to obtain statistical description
of the channel parameters. The larger the number of test users, the better channel
statistics can one obtain. Additionally, several types of environments should be
investigated to yield geographical statistics as well. These kind of investigations
have been carried out successfully for the GSM and UMTS frequency bands, for
example, [143, 161]. In the case of the WBAN, the selection of the test users from
body structure point of view is also a relevant factor.

Andreas F. Molisch et al.

105

Figure 2.29. Human torso phantom used for radiation measurements in the GSM and UMTS frequency bands [164].

With these considerations in mind, one can see the high complexity of the
measurement campaigns required for throughout UWB WBAN radio channel
characterization and modeling.
A method, which possibly can simplify and in the same time complement
the methodology presented above, is the use of full body phantoms. Full body
phantoms with realistic tissue dielectric properties are expensive and generally
they are tuned for narrow frequency bands (GSM, UMTS, etc.) [159]. Simplified
body phantom structures are available and described in the literature, for example, [156, 144, 160, 164]. Figure 2.29 shows an example of body torso phantom.
These body phantoms can be used in UWB as long as appropriate validity tests
are made in the investigated frequency band. The obtained results generally limit
the set radio channel parameters which can be correctly evaluated using a specific
body phantom structure.
As an example, in the work carried out in [146], body phantom investigations
were conducted in order to make an initial assessment on the applicability of such
human body models in UWB WBAN radio propagation analysis. The body phantom used consists of two concentric acrylic cylinders, with the gap in-between
filled with saline solution conform the ETSI standard [158]. The total height of
the body phantom was 1.9 m with an external diameter of 0.3 m. In this configuration, the body phantom had no arms, legs, or head. Thus, from the
scenarios described in Section 2.7.2.4, only the torso position of the BWD could
be replicated with good geometrical approximations; the right arm position was
possible with the BWD mounted close to the body phantom, on the diagonally
opposite side from the PBS position.
The inherent cylindrical symmetry of the body phantom has a potential influence on the propagation mechanism in the around the body area. While for
test users the radio waves are highly scattered from the arms and legs, this phenomenon cannot be captured with the simple body phantom. Therefore, a natural
extension of this body phantom is the use of two cylindrical arms at a certain

106

UWB propagation channels

distance from the body. Another observation which has to made is related to the
frequency dependency of the dielectric properties of the body phantom. Given the
simple modelno explicit tissue modelone can expect a dierent behavior of
the induced currents contributing to the wave propagation along the surface of
the body phantom. This in turn leads to the conclusion that the presented body
phantom is not suitable for on-body sensor network measurements.
2.7.3. Radio propagation channel
2.7.3.1. Introduction
The network-level design for WBANs strongly depends on the wide variety and
specifics of the user scenarios in which the devices are expected to operate. The
underlying air-interfaces and adaptive medium access control schemes have to be
reliable and optimized. A successful implementation of these communication systems is not possible without a thorough analysis of the specific around-the-body
radio propagation channels. Traditionally, the radio propagation channel for wireless systems is described in terms of a standard set of parameters: link budget and
scattering functions (time-delay versus Doppler spectra). The evaluation of these
parameters in typical UWB propagation scenarios has been presented in Sections
2.4, 2.5, and 2.6.
For WBAN UWB systems, it was anticipated in Section 2.7.2 that the radio
propagation channel will exhibit some new and interesting characteristics such as
antenna near-field and body-proximity eects. All these phenomena are very characteristic to the around-the-body propagation channel, thus a suitable characterization and modeling is required for them. The far-field propagation equations
derived in Section 2.2 can be applied under certain conditions with respect to the
positions of the transmitter and receiver antennas on the body and antenna types.
Nevertheless, the solution for the de-embeding problem combined with doubledirectional measurements is not practical in typical WBAN scenarios where small
devices are used. Similar, the scattering amplitudes in the theoretical expression
for the UWB channel presented in Section 2.3 become very dicult to evaluate
for the dynamic (nonstationary) around-the-body propagation channel. Furthermore, the dynamic near-field eects due to the body proximity cannot be derived
and included analytically. In order to identify and separate the influence of the
environment on the measured radio channel, dierent user and body-phantom
positions can be measured. The data can then be analyzed in order to determine
the channel impulse-response variability caused by the body phantom or human
user only.
These observations lead us to the conclusion that, as opposed to more traditional channel modeling, the radio wave scattering in the around-the-body area is
very complex to be described completely in analytical way and the WBAN radio
channel has to be characterised and modelled based on empirical results and possibly complex simulations. In this Section, we present the basic principles of the
WBAN radio channel characterisation and modeling. For this, we start with a brief

Andreas F. Molisch et al.

107

review of the UWB WBAN radio channel measurement results available in the
literature. We make a distinction between the models obtained from traditional
channel measurements and the new methodology presented in Section 2.7.2. Finally, we present the relevant results from a set of novel WBAN investigations described in references [145, 146].
2.7.3.2. Traditional channel measurement setups
The work presented in [147] is focused on the eects of human body interaction
with a close proximity UWB antenna. Antenna performance measurements have
been performed in the 1 GHz to 11 GHz frequency band, in the anechoic chamber
and various indoor multipath environments. The investigated antenna was a commercially available diamond dipole UWB antenna [150]. For the on-body experiments, the test antenna (transmitter) was belt (waist height) and head mounted,
using a cell phone makeup; the 1 GHz to 3 GHz frequency band was measured
using a time modulated UWB system from Time Domain Corporation. The receiving antenna was a horn type located at 5 m distance from the antenna under
test.
The investigation results in [147] showed a clear antenna interaction with the
human body. In light multipath scenarios, this interaction resulted in a 23 dB null
in the angular distribution of the power radiated from the antenna + human
body system in the direction of the human body. In dense multipath scenarios,
only a variation of 6.8 dB was noticed, confirming the robustness of the UWB systems in such environments. More realistic WBAN scenarios have been investigated
in [148], with both antennas (transmitter and receiver) placed on the body. Two
types of scenarios have been considered with antennas positioned on the front side
of the body and around the upper torso. Anechoic chamber and an oce room environments were used. The antennas used were commercially available small-size,
low-profile type [163]. Cable RF feed connections have been used for both antennas but no details on the ground plane size of the antennas and antenna position
is given in the publication. Although the accuracy of the results depends on it,
the influence of the feed cables on the antenna radiation patterns is not addressed
either in [148].
The 3 GHz to 6 GHz frequency band has been measured with a vector network
analyzer and the on-body UWB radio channel was described in terms of path loss
index and signal mean and rms delay spread. Further results are presented in terms
of required Rake receiver fingers and the energy considerations for single- and
multihop links. Under the considerations made in Section 2.7.2, it is evident that
radio channel results (parameter values and range) are very sensitive to the used
measurement setup. Naturally the numerical values for the path loss and timedelay spread strongly depend on the antenna system (antenna element, ground
plane, and feed cable) radiation characteristics when located on the body. It has to
noted that the results presented in [148] are obtained based on the assumption of
ideal, omnidirectional antenna radiation patterns. Furthermore, no analysis of the
influence of the feed cable(s) and of the human body proximity on the antenna

108

UWB propagation channels

radiation characteristics is made, thus the free-space specifications given by the

antenna manufacturer are implicitly assumed.
Making one step further, the WBAN investigations presented in [145, 146] address these issues and use improved accuracy of the measurement equipment and
antenna setup. The most important remark here is that the extracted radio channel parameters still remain characteristic for the antenna types used and cannot be
directly generalized as proposed in [148].
2.7.3.3. Channel measurements with improved accuracy
In the following, we present the main results from investigations carried out following the methodology described in Section 2.7.2. These novel experiments and
simulations allow us to disclose the main aspects of the WBAN radio propagation
phenomena in terms of
(1) antenna radiation characteristics when operating in the close proximity
of a simulated arm phantom,
(2) radio channel characteristics and parameters relevant to the design of a
UWB based WBAN communications system.
Similar to the channel measurement in [148], two small-size, low-profile type
antennas specified for the 3 GHz to 6 GHz frequency band have been used [163].
The reasoning for the choice of our antenna mount, ground plane size, RF optic fiber feed connection, and user scenarios was presented in Section 2.7.2. The
first on-body UWB radio channel measurements using this setup were presented
in more details in references [145, 146]. The WBAN was emulated with two bodyworn terminals: a portable base station (PBS) and a body-worn device
(BWD), similar but not limited to the scenarios considered in [139]. The detailed
description of the antennas and the body phantom used in the investigations can
be found in Section 2.7.2 and references [145, 146].
For disclosing the eects of the body proximity on the antenna radiation characteristics simulations can be carried out, as described in Section 2.7.2.2. A practical example of antenna analysis results is given in the following. The details of the
study can be found in [145, 146].
2.7.3.4. Human body influence on the antenna radiation characteristics
The influence of the arm where the BWD is mounted was modeled as a dielectric
brick of 60 mm 114 mm 40 mm (width length thickness) of muscle tissue
with the conductivity = 2.22 S/m and permittivity r = 41.5 [159]. Similar for
the PBS, the body was modeled as a brick of 145 mm 155 mm 30 mm and
compared to a much larger part of the body without significant dierences in the
results. These dimensions for the body models were found after a few iterations
aimed to simplify the simulations without introducing significant changes in the
obtained results.
The influence of the human body proximity is highlighted in Figures 2.30 and
2.31. The simulated arm and body is located in the right half of the corresponding

Andreas F. Molisch et al.

109
90
120

60
0 dBi
10 dBi

150

30

20 dBi

180

210

330
240

300
270

Figure 2.30. Simulated BWD, far-field antenna radiation pattern a 4.5 GHz (yz-plane) with and without the proximity of arm. The ground plane size was 25 mm 55 mm. The simulated arm with dimensions 60 mm 114 mm 40 mm is located in the right half of the plot (shaded area).

90
120

60
0 dBi

150

30

10 dBi
20 dBi

180

210

330
240

300
270

Figure 2.31. Simulated PBS, far-field antenna radiation pattern at 4.5 GHz (yz-plane) with and without the proximity of body. The ground plane size was 50 mm 110 mm. The simulated body with
dimensions 145 mm 155 mm 30 mm is located in the right half of the plot (shaded area).

radiation pattern plots. Most of the radiation towards the human arm/body disappears due to absorption and reflections. The absorption is largest for the lowest
frequencies, some 80% for 3 GHz and some 65% for 6 GHz.
2.7.3.5. Antenna radiation pattern versus frequency
The antenna element used in our previous examples is not radiating in the same
way for all frequencies of interest. This is the general case for simple UWB antenna
structures orginally designed for operation in free-space conditions.To investigate
the change in radiation pattern over frequency, simulations can be performed for

110

UWB propagation channels

90
120

60
0 dBi

150

30

10 dBi
20 dBi

180

210

330
240

300
270

Figure 2.32. Simulated antenna radiation (xz-plane) versus frequency for the BWD antenna with
dimensions as used in the measurements [145, 146].

several frequencies, with and without the human body present. As an example,
Figure 2.32 shows the simulation results at 3 GHz, 4.5 GHz, and 6 GHz for the
BWD. It is clear that the radiation changes significantly over the frequency band
and this influences the channel characterization methodology which has to be applied. Due to this nonstationarity in the frequency domain, the analysis of crossand autocorrelation properties can yield more correct results if it is performed on
several subbands [145, 146].
2.7.3.6. Radio channel parameters
Three main radio channel parameters have been extracted and analyzed from the
investigation presented in Sections 2.7.3.2 and 2.7.3.3.
(1) Link budget: wideband power levels can be used for the description of
the shadowing eects which occur around-the-body propagation environment [145, 146, 148].
(2) Radio channel frequency transfer function: average slope and autocorrelation properties in consecutive subbands [145, 146]. The frequencydomain analysis is particulary important in UWB systems due to the
signal distortion which needs to be analysed, for example, [165, 166].
(3) Signal time-delay dispersion: rms delay spread [145, 146, 148] and delay
window length that contains 99% of the total power [145, 146]. These
parameters are important for system-level design especially for optimising the performances of low-cost, low-power devices.
Although a more complex description of the radio channel is possible (see
Sections 2.4 and 2.5) in typical WBAN scenarios, the above listed parameters were
found to be the essential ones. Furthermore, for the time-domain signal statistics
analysis, a suitable measurement equipment is required, such as a UWB swept
time-delay correlation channel sounder. There were no dynamic WBAN channel
investigation results available in the literature at the time of writing this book.

Andreas F. Molisch et al.

111

Table 2.9. Main channel parameter results for WBAN presented in [148].
PL0 (dB)

Min rms (ns)

Max rms (ns)

Front side
Anechoic chamber

82.2

3.3

1.4

2.2

Oce

75.8

2.7

2.1

7.5

1.4
2.1

7.3
9.9

Around the torso

Anechoic chamber
Oce

109.2

4.1

The relevant results from the available radio channel investigation [145, 146,
148] are analyzed in the following sections with focus on their impact on the operation of low-cost, low-power devices.
2.7.3.7. Link budget
The proposed path loss model in [148] is given by the expression:


PL(d) = PL0 +10 log10 d[m]/1m , [dB]

(2.151)

and Table 2.9 summarizes the main results for the channel parameters in dierent
scenarios [148]. Path loss exponents up to 4.1 were obtained for the oce propagation scenarios, with additional possible fluctuations of the received power levels
up to 20 dB due to the position of the arms. These results have been further utilised
to determine energy levels for single- and multihop links.
From the studies in references [145, 146], Figure 2.33 shows the global statistics obtained in terms of wideband power levels in typical positions of a bodyworn device relative to the human body. Five test users have been used in up to
four dierent positions in the same indoor environment. The reference power levels of 0 dB correspond to a free-space (far-field) path loss, assuming 0 dBi antenna
gain and a 4.5 GHz carrier and at the same average distances as in the human body
measurement scenarios. The averaged cumulative distribution curves over three
dierent cases of positioning the BWD on the body can be well approximated
with a linear curve with respect to the power levels expressed in logarithmic units.
Nevertheless, significant deviations from these average CDFs are to be expected.
2.7.3.8. Frequency transfer function
The analysis of the frequency-domain data presented in [145, 146] starts with a
global characterization, that is, over the whole frequency band which was measured.
Figures 2.34 and 2.35 show as an example the measured frequency transfer
function, H( f ), for one test person in the four rotated positions, with the BWD
in the left arm and right arm scenarios, respectively. The average and the standard
deviations of the H( f ) curves (over the 128 measurements) are also plotted to
illustrate the measured variability of the channel.

112

UWB propagation channels

Relative wideband power CDF (%)

100
90
80
70
60
50
40
30
20
10
0

35

30

25

20

15

10

Wideband power relative to free-space (dB)

Left hand
Right hand

Thorax
Cumulated

Figure 2.33. Statistics calculated for the measured wideband power levels in typical body-area scenarios [145, 146].

Figures 2.34 and 2.35 also show the (average) power delay profiles calculated
from the H( f ) data. Similar results were obtained for the other WBAN scenarios
as presented in [145, 146].
The first remark has to be made here regarding the significantly dierent
frequency-domain behavior of the radio channel for the two scenarios with BWD
in the left arm and right arm scenarios. In the left arm scenario, the direct propagation path is very short in the order of 10 cm to 15 cm. This path dominates and
does not change significantly if the scenario is static as in our case, that is, the outside environment does not introduce significant changes. Certainly, scattered and
attenuated paths from the body and the ground exist, as determined also in [148],
and these paths give rise to the variations shown in Figures 2.34 and 2.35. In the
right arm scenario, the direct propagation path is generally more attenuated than
the scattered ones. The outside environment has a strong influence on the overall
propagation channel between the transmitter and the receiver. High power multipath components can be observed at relative delays up to 90 nanoseconds from
the direct path.
Another remark is that the average H( f ) displays generally both small-scale
and large-scale fluctuations in frequency domain. This observations is the basis of
the analysis of the correlation properties of H( f ), presented later in Section 2.7.3.9.
An autoregressive modelling approach as presented in Section 2.5.4 seems
possible to achieve. Nevertheless, due the high variability of the frequency transfer
function depending on the actual position of the user in the radio environment, a
large set of measurements is required. Furthermore, as exemplified in Figures 2.34
and 2.35, a radical change is to be expected when the device is moved to dierent
positions on the body and this eect also has to be introduced in the more general
model.

Andreas F. Molisch et al.

113
0

Norm. power (dB)

20
40

3.5

4.5

Pos. 1
5.5

3.5

4.5

Pos. 2
5.5

3.5

4.5

Pos. 3
5.5

3.5

0
20
40

0
20
40

0
20
40

4
4.5
5
Frequency (GHz)

Pos. 4
5.5

(a)

Pos. 1

20
40

Norm. power (dB)

10

30

50

70

90

0
20
40

110
Pos. 2

10

30

50

70

90

0
20
40

110
Pos. 3

10

30

50

70

90

0
20
40

110
Pos. 4

10

30

50
70
90
Excess delay (ns)

110

(b)
Figure 2.34. Example of (a) measured H( f ) and (b) the corresponding power delay profile for one
test person in the left arm Pos.V scenario in four dierent body positions.

In the first step, the slope of each average H( f ) was calculated via curve fitting
(least mean-square error) and expressed in dB/oct units. In the left arm setups, the
H( f ) slope values are generally high, in the range 25 dB/oct to 6 dB/oct, while
in the torso and right arm setups, the values are lower and range from 7 dB/oct

114

UWB propagation channels

0

Norm. power (dB)

20
40

3.5

4.5

Pos. 1
5.5

3.5

4.5

Pos. 2
5.5

3.5

4.5

Pos. 3
5.5

3.5

0
20
40

0
20
40

0
20
40

4
4.5
5
Frequency (GHz)

Pos. 4
5.5

(a)
0

Pos. 1

20
40

Norm. power (dB)

10

30

50

70

90

0
20
40

110
Pos. 2

10

30

50

70

90

0
20
40

110
Pos. 3

10

30

50

70

90

0
20
40

110
Pos. 4

10

30

50
70
90
Excess delay (ns)

110

(b)
Figure 2.35. Example of (a) measured H( f ) and (b) the corresponding power delay profile for one
test person in the right arm scenario in all four dierent body positions.

to +10 dB/oct. For dierent test persons in similar scenarios, large variations of
the H( f ) slope values were observed, up to 20 dB/oct. Thus, a clear dependency
of the H( f ) slope on the human body cannot be disclosed except for the above
observations. The range of the possible H( f ) slope values can be used, however,
when implementing adaptive (pulse shape, template, etc.) receiver structures.

Andreas F. Molisch et al.

115

2.7.3.9. Frequency correlation properties

Based on the discussions in Section 2.7.3.8, the correlation properties in the
frequency-domain are analyzed on consecutive subbands. The selected subbands
are 300 MHz wide, with an overlap factor of 50% (150 MHz). The investigated
3 GHz to 6 GHz band is analyzed with a total of 19 overlapping subbands
(300 MHz each) [145, 146]. The width of the subbands can also be chosen based on
visual inspection of representative experimental data. Frequency correlation bandwidths, BC0.5 and BC0.7 , can be calculated for the 0.5 and 0.7 frequency correlation
levels, respectively. Next, the global cummulative distribution function (CDF) of
the BC0.5 and BC0.7 bandwidths in each subband can be analysed using results from
several test users and their positions, in the same indoor environment.
Figure 2.36 shows the obtained CDF for the BC0.5 correlation level. Dierences can be observed between dierent frequency subbands in the left arm and
torso scenarios, due to the antenna radiation patterns in combination with the
frequency-selective absorption of the human body. In the right arm scenario, the
frequency autocorrelation function exhibits also significant fluctuations as a result
of the scattering in the total environment.
As expected, in the left arm setups (very short range), high average BC values
in the range of 140 MHz were obtained. For the torso setups, slightly lower, in
average 120 MHz, average BC values were determined. In the right arm setups, the
influence of the total radio environment is very significant, and low average BC , in
the range of 35 MHz, have been obtained.
The main observation which has to be highlighted here is that in consecutive frequency subbands the frequency correlation properties can be significantly
dierent in certain body-area scenarios (device position relative to the body). Significant variations were observed also between the results for dierent users, with
widely spread CDF curves around the results shown in Figure 2.36. Thus, for an accurate UWB WBAN channel description, a subband analysis is more appropriate
rather than considering the entire frequency band. For example, the autoregressive
modelling approach can be applied to frequency subbands.
2.7.3.10. Signal time-delay dispersion
In reference [148], rms delay spreads in order of 2 nanoseconds up to 10 nanoseconds have been calculated with the remark that the higher signal spread values in
the torso scenarios are possibly due to the echoes present in the measurement environment. From the studies in references [145, 146], Figure 2.37 shows the cumulative distribution curves of the calculated W99 parameter (delay window length
that contains 99 % of the total power) and rms delay spread values in all investigated body area scenarios.
The first important observation which can be made here is that the W99 and
the rms delay spread parameters do not have similar statistics. While there is a
clear separation in two distinct range clusters for the W99 parameter values
namely, below 40 nanoseconds and above 80 nanosecondsthis clustering cannot

116

UWB propagation channels

60

0.5 level correlation bandwidth (MHz)

75
150

Thorax

50

75

40

150

Right hand
30

75
20

150

Cumulated
10

75
150

3.3

3.6

3.9

4.2

4.5

4.8

5.1

5.4

0.5 level subband correlation bandwidth CDF (%)

Left hand

5.7

Frequency subband (GHz)

100
90
80
70
60
50
40
30
20
10
0

rms delay spread CDF (%)

W99 CDF (%)

Figure 2.36. Example of calculated global cummulative distribution function (CDF) of the BC0.5
bandwidths in each subband.

20

40

60

80

100 120

100
90
80
70
60
50
40
30
20
10
0

10

15

20

25

W99 delay window (ns)

rms delay spread (ns)

Left hand
Right hand

Left hand
Right hand

(a)

Thorax
Cumulated

30

Thorax
Cumulated

(b)

Figure 2.37. Statistics calculated for (a) the W99 parameter and (b) rms delay spread in typical body
area scenarios.

be observed in the rms delay spread values. Thus, a system performance evaluation
based only on the classical delay spread parameter is underestimating the particularities of the WBAN propagation channel.
Furthermore, the curves in Figure 2.37 show significant variation between the
results obtained for the dierent BWD positions.

Andreas F. Molisch et al.

117

2.7.4. Conclusions
In this Section, we introduced and analyzed the most important aspects of radio
propagation and antenna measurement techniques for WBAN. The focus was on
user scenarios with small- and medium-size devices.
Antenna design and analysis for small-size devices is certainly one of the key
issues when implementing WBAN. A significant influence of the actual UWB antenna radiation characteristics when used/mounted close to the human body was
disclosed [145148]. Numerical methods, such as FDTD, need to be employed in
order to analyze the radiation characteristics of small-size, body-worn UWB antennas. Nevertheless, further investigations are needed in UWB antenna design
for small-size, body-worn devices to optimize their radiation patterns and total
eciency when operating in the proximity of the human body.
The presented WBAN scenario measurements and simulations put in evidence the potential influence of the body proximity on the UWB radio propagation channel. The traditional measurement methodologies were shown not to be
optimal for an accurate characterization of the WBAN radio propagation channel including body-proximity and antenna eects. In WBAN radio channel measurements with electrically small antennas, the influence of the connecting cables
ought to be minimized if not completely eliminated. A practical setup was proposed in References [145, 146] to achieve this: the use of a RF on fiber optic link
instead of the usual conductive feed cable. Deriving a radio channel model without considering these aspects can lead to unrealistic characterization of the radio
channel and the associated performance of dierent UWB communication systems.
Body-phantom measurements and simulations can be used to obtain realistic
WBAN channel characterization. The use of a simple body phantom specified by
ETSI [158] can, however, lead to wrong estimation for parameters such as channel
frequency transfer function slope and autocorrelation. Our repeatability tests performed with the body phantom in a realistic but controlled environment showed
relatively high uncertainty levels for the estimated frequency and time parameters,
except for the wideband power. We believe that more complex-shaped phantoms
can be used to improve these results.
The wideband power-level fluctuations up to 20 dB obeserved in static WBAN
scenarios [145, 146] have to be taken in consideration as an additional slows(shadowing) fading margin when designing the communication link. For a proper
operation, the receiver input gain control loop has to be able to track relatively
slow signal-level variations with this order of magnitude. The frequency correlation properties of the radio channel change significantly over the investigated
frequency band of 3 GHz to 6 GHz, mainly due to the changing radiation characteristics of the antennas when operating in close proximity of the human body.
An analysis performed on consecutive and partly overlapped frequency subband
seems to give a more realistic characterisation of the channel transfer function
[145, 146].The influence of the channel frequency transfer function slope and autocorrelation characteristics on the communication link depends very much on

118

UWB propagation channels

the actual signal bandwidth. Thus the analyzed frequency-domain aspects have
to be taken in considerations when designing or optimizing UWB links with different pulse shapes (Gaussian, wavelet, etc.) [145, 146]. The signal energy spread
in time(-delay) domain was investigated and significant dierence were shown
for dierent body-worn device positions and the total scattering environment in
which the user is immersed can drastically change the radio propagation between two body-worn devices [145, 146, 148]. Maximising the captured energy,
while maintaining a high data rate, requires knowledge about the expected signal
spread, even for very short-range communications such as in WBAN.
The presented WBAN radio channel investigations have been only for static
scenarios. A natural extension will be to use dynamic and more realistic user scenarios in order to fully describe the associated UWB WBAN radio channels.
2.8. Channel estimation techniques
2.8.1. Introduction
Channel estimation in the UWB regime is of greater significance due to two basic
reasons: (i) the transmit power spectral density constraints due to (modified) Part
15 rules that seriously limit the symbol SNR available at the receiver and (ii) the severe multipath due to the ultra-wideband nature of the modulation, implying that
the per-path SNR is even lower. There is very little link margin left at the high data
rates (480 Mbps) over the more severe multipath channels (such as the CM3 and
CM4 [137]), implying that coherent demodulation is desirable so as not to further
compromise the link budget; this implies the need for accurate channel estimation
at the receiver. Typically, the transmission formats for high-rate link designs incorporate training or pilot symbols for appropriate channel estimation at the cost of
some system overhead. However, the choice of the pilot sequences should conform
to the choice of the modulation scheme, and the estimator designed accordingly.
In this chapter, we focus on channel estimation techniques for the two main
link-layer design approaches presented to IEEE 802.15.3a (high-rate ultra-wideband) Task Groupnamely, the multiband OFDM and the direct-sequence-UWB
(DS-UWB) based designs, respectively. The MB-OFDM system is based on orthogonal frequency division multiplexing (OFDM) as in 802.11; hence the transmitted pilot is a frequency-domain sequence leading naturally to frequencydomain channel estimation at the receiver. In contrast, DS-UWB is a time-domain
modulation scheme, leading to time-domain channel representations and estimation. The complexity and performance of channel estimation in these two cases is
then fundamentally determined by (i) Nc , the FFT size chosen for the MB-OFDM
implementation and (ii) BWc , where is the maximum delay spread and BWc
is the channel bandwidth for DS-UWB, which represent the parametrization of
the channel in the frequency and time domains, respectively.
In the following, the transceiver structures for the two designs including the
structure of training symbols are described. Subsequently, the choice of channel
estimators and their resulting performance is investigated. Finally, a representative

Andreas F. Molisch et al.

119

simulation result is presented that encapsulates the impact of the channel estimator on end-to-end link-layer bit error rates.
2.8.2. Multiband OFDM alliance proposal
2.8.2.1. Description of the MB-OFDM proposal
The MB-OFDM proposal to IEEE 802.15.3a [167] divides the UWB spectrum into
several subbands of 528 MHz width each. Within each band, information is transmitted using OFDM modulation. In order to introduce multiple accessing capability and to exploit the inherent frequency diversity, each OFDM symbol is transmitted on a dierent subband as dictated by a time-frequency code that leads to
band hopping.
The inherent ability of OFDM to eectively overcome intersymbol interference (ISI) for high-data-rate systems over frequency-selective multipath channels
is well known. A brief overview of OFDM is given here; for further details, the
reader is referred to [168]. Figure 2.38 shows a block diagrammatic overview of
the operations within the MB-OFDM transmitter and receiver implementations.
OFDM eectively converts a (wideband) frequency-selective channel to multiple flat-fading (nonselective) channels by demultiplexing a single high-rate serial
data stream onto several lower-rate parallel data streams which are then transmitted independently on the narrowerband subchannels using dierent carriers that
are carefully chosen to maintain mutual orthogonality. MB-OFDM uses Nc = 128
subcarriers separated by f = 4.125 MHz that span the 528 MHz width of each
subband. In order to ensure orthogonality among subcarriers, the symbol duration is fixed to be TFFT = 1/ f = 242.42 nanoseconds. Each subcarrier is modulated independently with the complex symbol Sn,k assumed to be QPSK for our
purposes; the subscript denotes the nth OFDM symbol carried on the kth subcarrier. Thus, within the symbol duration TFFT , the transmitted nth OFDM symbol is
given by
Nc 1
1
sn (t) = 
Sn,k e j2k f t ,
Nc k=0

0 t TFFT .

(2.152)

As is well known, the transmitted (baseband) OFDM symbol is created by a 128point IFFT operation on Sn,k .
Because of the multipath nature of the channel, the orthogonality among the
subcarriers is lost after passing through the channel. In order to preserve orthogonality during transmission, the MB-OFDM scheme appends 32 zero samples
spanning a 60.6-nanoseconds (TZP ) time duration to sn (t) in (2.152), which is
larger than the typical delay spread of UWB channels. To this a 9.5-nanoseconds
guard interval is added, thus making the OFDM symbol duration TSYM = 312.5
nanoseconds.

120

UWB propagation channels

Input
Scrambler
data

Convolutional
encoder

Puncturer

Bit interleaver

Constellation
mapping

IFFT
Insert pilots
Add prefix/Gl

DAC
exp( j2 fc t)

Time-frequency
code

(a)
cos( j2 fc t)

Q
LPF

VGA

ADC

Channel
estimation

sin( j2 fc t)

Descrambler

LNA

Viterbi
decoder

ADC

Deinterleaver

VGA

Freq.- domain
equalizer

LPF

Synchronization
FFT

I
Prefilter

Output
data

Decision
feedback

Time-frequency
code

(b)
Figure 2.38. Structures of the MB-OFDM (a) transmitter and (b) receiver.

At the receiver, orthogonality is restored by adding the last 60.6 nanoseconds

of the received signal to the first 60.6 nanoseconds duration of the OFDM symbol.
Correlation at the receiver is implemented as a 128-point FFT operation.
A PHY frame is constructed around the MAC payload for transmission whose
structure, shown in Figure 2.39, consists of a preamble, header, and the MAC payload. The PLCP preamble portion of the PHY frame is made up of 30 OFDM
symbols consisting of the following:
(1) packet synchronization sequence: 21 OFDM symbols constructed from a
time-domain sequence, which is used for packet detection and acquisition, coarse carrier-frequency estimation and coarse symbol timing,
(2) frame synchronization sequence: 3 OFDM symbols constructed from a
time-domain sequence, which is used to synchronize the receiver algorithm within the preamble (boundary detection),
(3) channel estimation sequence: 6 OFDM symbols constructed from
a frequency-domain sequence, which is used for channel estimation, fine
carrier frequency estimation and fine symbol timing.
2.8.2.2. Channel estimation and equalization
Availability of an accurate channel transfer function estimate is a prerequisite for
coherent symbol detection within an OFDM receiver. Channel estimation in
OFDM systems can be classified as frequency-domain- or time-domain-based;
each of this can be further divided as being pilot assisted, decision directed, or

Andreas F. Molisch et al.

121

Reserved Band extension Reserved Rate Reserved Length Reserved Scrambler init. Reserved
1 bit
3 bits
1 bit 4 bits 1 bit 12 bits 1 bit
2 bits
3 bits

PLCP preamble

Optional
Tail Pad
PHY Tail MAC
HCS
band
header bits header
bits bits extension

Frame payload
Variable length: 0-4095 bytes

PLCP header
55 Mb/s

FCS

Tail Pad
bits bits

55, 80, 110, 160, 200, 320, 480 Mb/s

(a)

0 . . . 0 C0 C1 . . . C127 0 0 0 0 0

0 . . . 0 C0 C1 . . . C127 0 0 0 0 0

PS0

PS1

PS20

FS0

Packet sync. sequence

21 OFDM symbols

FS1

FS2

Frame sync. sequence

3 OFDM symbols

CE0

CE1

CE5

Channel est. sequence

6 OFDM symbols

9.375 s
(b)
Figure 2.39. (a) Structure of the MB-OFDM PHY frame; (b) structure of the preamble (reproduced
from [167]).

blind startup. Since the MB-OFDM proposal includes a frequency domain channel estimation sequence transmitted in each PHY frame on all subcarriers, our
focus in this section will be on pilot-based frequency-domain (i.e., post-FFT in
the receive chain) channel estimation and equalization techniques as indicated in
Figure 2.38.
Since the OFDM symbol duration is much larger than the delay spread of typical UWB channels, each subcarrier experiences flat fading. As a result, the eect of
the radio channel in the frequency domain is characterized by a multiplicative distortion of the kth subcarrier signal via the (complex) channel response Hk . Thus
the received samples post-FFT is given by
Rk = Hk Sk + Nk ,

k = 1, 2, . . . , Nc ,

(2.153)

where Hk is the channel transfer function sampled at the kth subcarrier location
and Nk is the additive white noise of the channel. Equation (2.153) in matrix form
can be written as
R = HS + N,

(2.154)

122

UWB propagation channels

Hk

Nk

Sk

Rk

Hk+1

Nk+1

Sk+1

Rk+1

Figure 2.40. Eect of channel fading on subcarriers.

where H is a diagonal matrix with Hk as its diagonal entries and S, R, and N are the
transmitted symbol, received symbol, and the additive noise vectors, respectively,
as is shown in Figure 2.40.
The form of (2.153) shows that the least-squares (LS) estimate of the channel
coecients is simply given by
H k =

Rk
Sk

k = 1, 2, . . . , Nc ,

(2.155)

where the known frequency-domain sequence Sk is sent as part of the preamble in

every frame. This sequence is transmitted on all the subbands and done so twice in
each subband so as to improve the LS estimates obtained using (2.155). We discuss
approaches to enhancing the LS channel estimates more broadly in Section 2.8.2.2.
To obtain the (unknown) transmitted symbols, frequency-domain equalization is
needed to compensate for the channel based on the channel estimate obtained as
above. We consider the topic of equalization separately for uncoded and coded
systems.
To estimate the transmitted signal vector S from the received signal vector R,
consider the following linear estimator:
S = KT R.

(2.156)

The design of the estimator K could be accomplished based on two dierent approaches: (a) based on using the estimated channel or (b) using only the statistical
properties (and not estimates) of the channel. Typically, two design criteria are
usually considered: (i) zero forcing (ZF) which neglects the presence of noise and
compensates only for the channel distortion and (ii) minimum mean-square error
(MMSE) criterion, in which the channel distortion and noise are simultaneously
minimized.
Zero-forcing equalization. Zero-forcing (ZF) equalization attempts to invert the
channel transfer function to produce an estimate of the transmitted symbol vector.
In other words, the ZF equalizer has to satisfy the condition KTZF H = I, where I is

Andreas F. Molisch et al.

123

the N N identity matrix, whose solution is KTZF = H1 . Since the channel matrix
H is diagonal, the ZF estimate for the transmitted signal vector SZF = KTZF R can be
computed independently on a subchannel basis:
SZF,k =

1
Rk ,
Hk

k = 1, 2, . . . , N.

(2.157)

Zero-forcing equalization allows for optimum performance if little or no noise is

present. However, since 1/Hk takes on very large values for subcarriers in deep
fades, ZF equalization causes significant noise enhancement at those (near) spectral nulls and performance suers as a result.
MMSE equalization. MMSE equalization attempts to minimize the squared norm
of the error vector KTMMSE R S2 and can be obtained using the orthogonality
principle [169, 170]:
1
,
KTMMSE = RSR RRR

(2.158)

where RSR and RRR are the cross-covariance matrices between S and R and the
autocovariance matrix of R, respectively,


RSR = E SRH = E S(HS + N)H = Es HH ,



RRR = E RRH = E (HS + N)(HS + N)H = Es HHH + RNN ,

(2.159)

where Es is the average transmitted signal energy defined as Es = E(SSH ) and RNN
is the noise covariance matrix RNN = N2 I. Plugging (2.159) in (2.158), the MMSE
equalizer is given by


KTMMSE = HHH +

N2
I
Es

1

HH .

(2.160)

Since H is a diagonal matrix, equalization can again be done on a subchannel basis,

yielding transmitted symbol estimates given by
H
 Rk ,
SMMSE,k =  2 k
Hk  + 2 /Es
N

k = 1, 2, . . . , Nc .

(2.161)

Notice that when the noise level is significant, the MMSE solution avoids the noise
enhancement problem even when some Hk s are close to zero while at low noise
levels, the MMSE solution converges to the ZF solution.
Equalization in the presence of coding. If the receiver decisions are made independently on each subcarrier, symbols sent on subcarriers that have undergone
deep fades would be (with high probability) erroneous and would lead to an error
floor. Thus, uncoded OFDM is virtually unusable on multipath fading channels
with deep notches occurring in the frequency spectrum. Suitable channel coding

124

UWB propagation channels

with frequency-domain interleaving is thus essential in OFDM to recover symbols

lost in subcarriers in deep fades. MB-OFDM uses rate 1/3 convolutional encoding
with dierent puncturing patterns to achieve dierent code rates (for variable data
rates); the coded symbols are interleaved over three subbands.
With Viterbi decoding at the receiver, it is possible to implement an enhanced
equalizer for the received sequence based on (estimated) channel. The symbols
received on data subcarriers with a higher gain |Hk | are more reliable than those
received on subcarriers with smaller channel gains. We can expect to get better
performances if the Viterbi decoder is provided with a measure of this reliability,
as accomplished by
Sweighted,k = Hk Rk ,

k = 1, 2, . . . , Nc .

(2.162)

This equalizer essentially compensates for the phase distortion introduced by the
channel on each subcarrier and weights each symbol with |Hk |2 which enables
soft Viterbi decoding with performance equivalent to maximum ratio combining.
Since the subcarriers have dierent signal-to-noise ratios after equalization, the
above scheme is just a means of feeding the Viterbi decoder with modified metrics. This weighting avoids the noise enhancement inherent to equalized OFDM
systems.
Enhancing channel coecients estimates. The LS channel estimates obtained using (2.155) are unreliable in the presence of noise. While transmitting the channel
estimation sequence twice on each band could be used to reduce the eect of additive noise by 3 dB via averaging, this may not suce to yield estimates of desired
accuracy. Typically, the channel estimation sequences are only used to obtain the
initial estimates of the channel coecients which are then updated using decision feedback every OFDM symbol duration. Channel estimation with decision
feedback in context of OFDM systems is discussed in [171, 172] and can be described as follows. Initial estimates of the channel coecients H k(0) are obtained
using (2.155) using the frequency-domain pilot symbols in each frame. These estimates are used to demodulate the next OFDM symbol Rn+1,k after frequencydomain equalization. The vector of equalized symbols is now soft Viterbi decoded
to produce more reliable symbol decisions. The information is reencoded to give
Sn+1,k , which is used to update the channel coecients, Hk(1) = (Rn+1,k / Sn+1,k ), for
all k; these estimates are then used to demodulate the next symbol and so on.
The idea of decision-feedback is depicted pictorially in Figure 2.41. The decisionfeedback scheme has to be modified when applied to MB-OFDM since the interleaving spans 3 OFDM symbols; the channel estimates are updated every 3 OFDM
symbols instead of each symbol.
Statistical knowledge of the correlation of the channel frequency response at
dierent subcarriers can also be used to improve the quality of the channel estimates. This can be naturally accomodated within the MMSE framework. Channel
estimators that exploit such correlation information are described in [173, 174]
and the references therein.

Viterbi
decoder

Descrambler

1 OFDM
symbol
delay

Deinterleaver

Output of FFT, Rnk

125

Freq.-domain
equalizer

Andreas F. Molisch et al.

Output data

H kn1

Rnk 1

Snk 1

Channel
estimation

Create freq.domain TX
signal

No

Freq.-domain reference symbol

Known
channel
estimation
sequence?
Yes

Figure 2.41. Decision feedback for updating channel estimates in MB-OFDM proposal.

2.8.2.3. Simulations
The performances of the various channel estimation schemes described above are
depicted in Figure 2.42 for a UWB channel model numbered CM1 at 480 Mbps.
The UWB channel modeling subcommittee has identified and distinguished four
channel models, numbered CM1 to CM4 for baseline testing purposes, based on
the severity of the multipath channel [137]. It is clear from the figure that the zeroforcing equalizer has the worst performance owing to the enhancement of noise at
subcarriers in deep fades. The MMSE equalizer alleviates the problem of noise enhancement and would be near optimal in the absence of channel coding. However,
in the presence of channel coding, when the decoder is provided with the information about the reliability of the subcarriers, it is able to make a more informed
decision and thus the BER performance is significantly better in the presence of
channel-dependant weighting of the Viterbi decoder metrics.
2.8.3. UWB via impulse radio and direct-sequence spread spectrum
Ultra-wide bandwidth modulations can be achieved either via spreading in time or
in frequency domains. Traditionally, UWB has been associated with impulse radio
(IR) [175] that transmits information by modulating the amplitude or position
of very short (on the order of 100 picoseconds to 2 nanoseconds) pulses. In order
to permit multiple accessing, time-hopping IR was proposed [176]; the symbol
duration is divided into N f intervals called frames, each of which contains one
pulse. The position of the pulse within the frame is determined by a PN sequence
{ck } unique to the kth user of length Nc with elements ck {0, Nh } such that
Nh Tc < T f . The transmitted signal is thus given by
s(t) =

dk w t kNc T f ,

(2.163)

126

UWB propagation channels

100

Bit error rate

101
102
103
104

10

11

12

13

14

15

16

Eb /N0 (dB)
ZF equalization
MMSE equalization
Weighted soft-bit decoding
Figure 2.42. Channel estimation performance in MB-OFDM at 480 Mbps in CM1.

1 bit (Td = N T f )
TH-UWB

T p,TH

Tf
T p,DS

DS-UWB

Duty cycle = dc = T p /T f
Figure 2.43. Dierence between TH-IR and DS-UWB (reproduced from [177]).

where
w(t) =

N
c 1

p t jT f c j Tc ,

(2.164)

j =0

where p(t) is the UWB pulse with a duration of Tc . In the above expressions, the
modulation assumed is binary PAM. An alternative is to use PPM, where the data
bit time dithers the UWB pulse to achieve modulation. The data rate for TH-IR is
given by Rs = 1/Nc T f .
Alternatively, spreading can be achieved via familar direct-sequence UWB approach, where the pulses are transmitted continuously using a PN sequence of

Andreas F. Molisch et al.

127

9 10 11

9 10 11

Figure 2.44. DS-UWB spectrum (reproduced from [179]).

length Ns with consequent data rate 1/Nc Tc . A schematic of TH-IR and DS-UWB
is shown in Figure 2.43.
For a given pulse width and spreading code length, TH-IR would have a smaller pulse-repetition frequency (PRF) than DS-UWB and thus would be limited to
low data rates. For example, assuming Nh = 10 and Tc = 1 nanoseconds, the frame
duration, T f = 10 nanoseconds for the TH-IR systems or a PRF of 100 MHz.
On the other hand, DS-UWB uses a T f equal to the pulse duration and thus the
PRF is 1 GHz. Thus DS-UWB systems are potentially capable of accommodating
higher data rates than TH-IR, or equivalently more users for a given aggregate link
rate [178]. This conclusion extends from the AWGN scenario described above to
the realistic case of multipath channels, and is the primary reason why DS-UWB
approaches are preferred for higher rates. In either case, the receiver structures for
TH-IR and DS-UWB modulation are remarkably similar; hence the sequel just
concentrates on DS-UWB.
2.8.3.1. Description of DS-UWB proposal
In contrast to the MB-OFDM scheme that divides the UWB spectrum into several
528 MHz subbands, the DS-UWB [179] scheme uses single-carrier modulation
over a much wider bandwidth. Specifically, the UWB spectrum is divided into
the low and high bands as shown in Figure 2.44 and depending on the required data rate the transmission occupies one of the two bands or both simultaneously.
Transmission over the wide bandwidth is accomplished by the use of directsequence spread-spectrum technique (DS-SS); multiple accessing is achieved using
code-division multiple access (CDMA). In CDMA, a binary information sequence
at rate Rb is modulo-2 added with a higher-speed binary pseudorandom (PN) code
sequence {cn } (often called the chip sequence) producing a high-speed random
sequence. The inverse of the chip pulse duration Tc determines the approximate
transmission bandwidth and the ratio Tb /Tc (the number of chips transmitted per
information bit) is called the processing gain G p . The transmitted signal is thus

128

UWB propagation channels

Ts

r(t)

w(t 1 )

C0

Ts

Ts
CN 1

C1

.
.
.


b[n]

Channel
estimation

FM
w(t L )
Rake receiver

FM 1

F1

Ts

Ts

Ts

Equalizer

Figure 2.45. Structure of a Rake receiver.

given by

s(t) =

bk w t Tb ,

(2.165)

k=

where
w(t) =

N
c 1

ck p t Tc

(2.166)

k=0

is the spreading waveform, {bk } is the bit sequence to be transmitted, and p(t) is
the UWB pulse.
In DS-UWB proposal, one full length of the chip sequence is sent in each
bit interval; such spreading sequences should exhibit the following essential attributes.
(1) they should have favorable periodic autocorrelation properties, that is,
low autocorrelation at all nonzero shifts of the sequence;
(2) cross-correlation among dierent pairs of sequences at all lags should be
small to enable multiple access.
Because of the wide bandwidths employed in DS-UWB systems, a very large
number of resolved multipath components exist. For example, a multipath channel with maximum delay spread of 100 nanoseconds and signal bandwidth of
1 GHz, will result in 100 resolvable multipath components. Thus, the energy perpath at the receiver is typically very low and several paths must be combined for
more reliable data decisions. Typically a Rake receiver (an adaptive correlator with
chip-spaced taps) is used for the purpose of combining desired path components;
however, this must be tradedo with greater intersymbol interference (ISI) at Rake
output as the length of Rake is increased. The Rake combiner/equalizer constitutes
the most significant component complexity-wise in the baseband receive chain
of a DS-UWB receiver. Given complexity-induced constraints on the number of

Andreas F. Molisch et al.

Preamble

129

PHY header

MAC header

Acq. seq. (8 s) Training seq. (4 s)

HCS

Frame body & HCS SB & TS

Start frame delimiter

Figure 2.46. Structure of the DS-UWB frame.

allowed fingers, it follows that the Rake receiver must employ tracking of the significant multipath components as shown in Figure 2.45.
The code sequence used in the DS-UWB proposal has a basic chip rate of
1.368 Gcps for the low band and consists of a sequence of root-raised-cosine UWB
pulses with 30% excess bandwidth. The length of the code sequences and the number of UWB pulses or chips per data symbol are variable from 1 to 24 to support
various data rates sent using binary phase-shift keying (BPSK) modulation.
The structure of the PHY frame used in the DS-UWB proposal is shown in
Figure 2.46. The preamble, PHY header, and the MAC header are modulated using
a length 24 spreading code, while the rest of the frame uses a spreading code of
dierent length as dictated by the data rate. The preamble consists of
(1) an acquisition sequence, which is a random sequence of bits (one bit per
code word) and is used for initial symbol time acquisition,
(2) a training sequence, consisting of a fixed sequence of training bits which
the receiver uses to estimate the channel and to set the locations of taps
and their respective gains for the Rake receiver.
2.8.3.2. Channel estimation
To be able to eectively combine the multipath components, the Rake receiver of
Figure 2.45 needs to know the delays and attenuations of the multipath components. Since the DS-UWB proposal incorporates a training sequence to aid the receiver in channel estimation, we confine our discussion to data-aided approaches.
Since the acquisition sequence is transmitted ahead of the training sequence, it
will be assumed that the receiver has established symbol timing before attempting
channel estimation.
Let the impulse-response of the UWB channel be
h(t) =

k t k ,

(2.167)

where k and k are the delay and real gain of the kth path, respectively. For
a detailed discussion on the UWB channel model, refer to [137]. The channel

130

UWB propagation channels

k } and the associated gains {

k } using
estimator estimates the tap positions {
a limited number of L taps (due to complexity reasons) in such a manner that
+
the synthesized channel Lk=1 k (t k ) is well matched to the actual channel of
(2.167).
In the presence of additive white noise only, the maximum likelihood (ML) criterion is equivalent to the minimum mean-square error estimation. Thus, the delays
and channel gains may be obtained by minimizing the mean-squared error, that
is,
'

MSE = E h(t)

(2

.
k t k


(2.168)

k=1

Sliding window algorithm. If it is assumed that the delayed versions of the received
signal are orthogonal to each other, the sliding window algorithm provides ML estimates of the channel parameters. The sliding window algorithm cross-correlates
the received training sequence with the known sequence and uses the positions
of correlation peaks as the delays within the Rake receiver. The relative heights
of the correlation peaks at these delays provide an estimate of the tap weights
[180]. The eect of additive noise can be further alleviated by averaging the estimates over multiple symbols. We note that the assumption of the delayed versions
being orthogonal is only approximately valid in DS-UWB for any set of practical spreading codes. Furthermore, typical channel delays may exceed the symbol
duration, thereby leading to ISI. Under these conditions, the sliding window algorithm is suboptimal. Also, the complexity of the ML approach to channel estimation is prohibitive in such situations [181183] and simpler alternatives are
desirable.
Successive cancellation algorithm. As mentioned in the previous section, the joint
ML estimation of path gains and delays is computationally intensive. Successive
channel estimation algorithm [184] provides a more feasible approach to this
problem and oers improved performances over the sliding window algorithm
by using an iterative algorithm. The algorithm begins by using the sliding window
algorithm to find the parameters of the strongest path. The delayed version of the
transmitted signal corresponding to the estimated tap is then subtracted from the
received sequence and the algorithm is repeated to estimate the parameters of the
second strongest path and so on. The number of iterations required is equal to the
number of taps to be estimated. In [185], a comparison of the Rake performance
using sliding window algorithm and successive cancellation algorithm in context
of UWB is provided.
Number of Rake fingers. The number of fingers to be used in the Rake receiver
has a significant impact on receiver performance and complexity. Clearly, a good
choice of this parameter strongly depends on the channel power-delay profile. In
[186], the authors investigate the energy capture as a function of number of Rake
fingers in a typical UWB channel. A normalized energy capture metric was defined

Andreas F. Molisch et al.

131

as follows:
T 

2
+L

k s t k  dt
0 r(t) k=1
.
EC(L) = 1
T 
2


0 r(t) dt



(2.169)

normalized MSE
In the absence of any channel estimation errors, the normalized MSE defined in
(2.169) decreases as L increases and therefore the EC increases monotonically albeit with diminishing returns. In the presence of channel estimation errors, the EC
would start increasing with L after a certain point as described in the next section.
Eect of imperfect channel estimates. Any channel estimate will be imperfect due to
the presence of noise, mutual and multiple access interference leading to degradation in Rake performance as compared to the ideal case of perfect channel knowledge. For a fixed total bit energy, as the number of diversity or Rake branches increases, the energy per branch decreases. On the one hand, having more branches
enables greater diversity gain; on the other, more imperfect channel estimates leads
to decreasing the eective SNR available per branch. This fundamental tradeo between the diversity order and the channel estimation error is discussed for wideband CDMA in [187]. In [188], the eect of estimation errors and the number of
Rake fingers on BER performance is analyzed in a typical UWB channel. In [189],
an analytical framework is developed to quantify the eects of channel estimation
errors on the performance of UWB systems operating in dense multipath environments.
2.8.3.3. Equalization
The delay spread of typical UWB channels is typically larger than the symbol duration, which causes significant intersymbol interference (ISI). Even with perfect
estimation of the tap delays and coecients for Rake combining, ISI leads to high
bit error rates and results in irreducible error floor. This is compounded at higher
data rates (i.e., lower processing gains) where the performance approaches that of a
narrowband system without equalization. In [190], the authors analyze the detrimental eects of ISI on the performance of DS-CDMA systems with decreasing
processing gain. It is therefore essential that a symbol spaced equalizer following
Rake combining be incorporated to mitigate ISI as shown in Figure 2.45.
Considering the data rates that the UWB proposals attempt to achieve, and
the severity of UWB channels, nonlinear decision-feedback equalizers (DFEs) are
a natural choice for DS-UWB systems [191, 192].
Figure 2.47 shows plots of BER for 100 realizations of CM3 channel model
with and without decision-feedback equalization. The DF equalizer has 2 feedforward taps and 10 feedback taps [179] and is seen to improve performance considerably.

132

UWB propagation channels

0
1
2
3
4
5
6
7
8
9
10

6
8
Eb /N0 dB

10

12

14

10

12

14

(a)

1
2
3
4
5
6
7
8
9
10

6
8
Eb /N0 dB
(b)

Figure 2.47. Eect of ISI on DS-UWB at 57 Mbps in UWB CM3. Each plot shows 100 channel realizations. (a) No equalization; (b) DFE with feedback span of 10 symbols (reproduced from [179]).

Acknowledgments
Part of the work of A. F. Molisch was financed by an INGVAR grant from the
Swedish Strategic Research Initiative. Useful discussions with Professor Moe Z.
Win, Professor Robert Qiu, and the members of the IEEE 802.15.3a and 802.15.4a
are gratefully acknowledged. The work of Istvan Z. Kovacs, Gert F. Pedersen, and
Patrick C. F. Eggers was supported by the European Power Aware Communications
for Wireless Optimised Personal Area Networks (PACWOMAN), IST Contract no.
IST-2001-34157, and the Danish Future Adaptive Communication Environment
(FACE), AAU, research projects.

Andreas F. Molisch et al.

133

Bibliography
[1] J. B. Andersen, T. S. Rappaport, and S. Yoshida, Propagation measurements and models for
wireless communications channels, IEEE Communications Magazine, vol. 33, no. 1, pp. 4249,
1995.
[2] H. L. Bertoni, Radio Propagation for Modern Wireless Systems, Prentice-Hall, Upper Saddle River,
NJ, USA, 2000.
[3] N. Blaunstein, Radio Propagation in Cellular Networks, Artech House, Boston, Mass, USA, 1999.
[4] G. D. Durgin, Space-Time Wireless Channels, Prentice-Hall, Upper Saddle River, NJ, USA, 2003.
[5] R. B. Ertel, P. Cardieri, K. W. Sowerby, T. S. Rappaport, and J. H. Reed, Overview of spatial channel models for antenna array communication systems, IEEE Personal Communications Magazine, vol. 5, no. 1, pp. 1022, 1998.
[6] A. F. Molisch, Modeling of directional mobile radio channels, Radio Science Bulletin, vol. 16,
no. 302, pp. 1626, 2002.
[7] A. F. Molisch and F. Tufvesson, Multipath propagation models for broadband wireless systems,
in CRC Handbook of Signal Processing for Wireless Commmunications, M. Ibnkahla, Ed., pp. 2.1
2.43, CRC Press, Boca Raton, Fla, USA, 2004.
[8] J. D. Parsons, The Mobile Radio Channel, John Wiley & Sons, New York, NY, USA, 1992.
[9] S. R. Saunders, Antennas and Propagation for Wireless Communications Systems, John Wiley &
Sons, New York, NY, USA, 2002.
[10] R. Vaughan and J. B. Andersen, Channels, Propagation and Antennas for Mobile Communications,
IEE Press, London, UK, 2002.
[11] Federal Communications Commission. First report and order 02-48, 2002.
[12] P. Bello, Characterization of randomly time-variant linear channels, IEEE Transactions on
Communications, vol. 11, no. 4, pp. 360393, 1963.
[13] P. Bello, Time-frequency duality, IEEE Transactions on Information Theory, vol. 10, no. 1, pp.
1833, 1964.
[14] C. Baum, General properties of antennas, Sensor and Simulation Note 330, July 1991.
[15] G. A. Deschamps, ILe principe de reciprocite en electromagnetisme; IIApplication du
principe de reciprocite aux antennas et aux guides dondes, Revue du Cethedec, no. 8, pp. 71
101, 1966.
[16] E. G. Farr and C. Baum, Time domain characterization of antennas with TEM feeds, Sensor and
Simulation Note 426, October 1998.
[17] M. Kanda, Time domain sensors for radiated impulsive measurements, IEEE Transactions on
Antennas and Propagation, vol. 31, no. 3, pp. 438444, 1983.
[18] Y. T. Lo and S. W. Lee, Eds., Antenna Handbook, chapter 2, Theorems and Formulas, Van
Nostrand Reinhold, New York, NY, USA, 1988.
[19] C. Roblin, S. Bories, and A. Sibille, Characterization tools of antennas in the time domain, in
Proceedings of International Workshop on Ultra Wideband Systems (IWUBS 03), Oulu, Finland,
June 2003.
[20] A. Shlivinski, E. Heyman, and R. Kastner, Antenna characterization in the time domain, IEEE
Transactions on Antennas and Propagation, vol. 45, no. 7, pp. 11401149, 1997.
[21] H. J. Schmitt, C. W. Harrison, and C. S. Williams, Calculated and experimental response of
thin cylindrical antennas to pulse excitation, IEEE Transactions on Antennas and Propagation,
vol. 14, no. 2, pp. 120127, 1966.
[22] M. Steinbauer, A. F. Molisch, and E. Bonek, The double-directional radio channel, IEEE Antennas and Propagation Magazine, vol. 43, no. 4, pp. 5163, 2001.
[23] H. T. Friis, A note on a simple transmission formula, Proceedings of IRE and Waves and Electrons, vol. 34, pp. 254256, May 1946.
[24] A. Sommerfeld, Theoretisches Uber die Beugung der Roentgenstrahlen, Zeitschrift fur Mathematik und Physik, vol. 46, pp. 1197, 1901.
[25] H. Lamb, The diraction of a solitary wave, Proceedings of the London Mathematical Society,
vol. 2, no. 8, pp. 422437, 1910.

134

UWB propagation channels

[26] F. G. Friedlander, The diraction of sound pulses I. Diraction by a semi-infinite plane, Proceedings of the Royal Society of London. Series A, Mathematical and Physical Sciences, vol. 186,
no. 1006, pp. 322344, 1946.
[27] L. B. Felsen, Ed., Transient Electromagnetics, Springer, New York, NY, USA, 1976.
[28] F. G. Friedlander, Sound Pulses, Cambridge University Press, Cambridge, Mass, USA, 1st edition,
1958.
[29] L. B. Felsen and N. Marcuvitz, Radiation and Scattering of Waves, Prentice-Hall, Upper Saddle
River, NJ, USA, 1973.
[30] M. Kline and I. W. Kay, Electromagnetic Theory and Geometric Optics, Robert E. Krieger Publishing Company, Melbourne, Fla, USA, 1979.
[31] J. B. Keller, Geometrical theory of diraction, Journal of the Optical Society of America, vol. 52,
pp. 116130, February 1962.
[32] R. G. Kouyoumjian and P. H. Pathak, A uniform geometrical theory of diraction for an edge
in a perfectly conducting surface, Proceedings of the IEEE, vol. 62, no. 11, pp. 14481461, 1974.
[33] L. Y. Atsanin, A. Kostylev, Y. Zinoviev, and A. Pasmurov, Radar Target Characteristics: Measurements and Applications, chapter 3, CRC press, Boca Raton, Fla, USA, 1994.
[34] E. M. Kenaugh and D. L. Moatt, Transient and impulse response approximation, Proceedings
of the IEEE, vol. 53, no. 8, pp. 893901, 1965.
[35] R. C. Qiu, Digital transmission media: UWB wireless channel, MMIC, and chiral fiber, Ph.D. thesis, Polytechnic University, Brooklyn, NY, USA, January 1996.
[36] A. Trindade, Q. H. Dang, and A.-J. van der Veen, Signal processing model for a transmitreference UWB wireless communication system, in Proceedings of IEEE Conference on Ultra
Wideband Systems and Technologies (UWBST 03), pp. 270274, Blacksburg, Va, USA, November 2003.
[37] R. T. Hoctor, et al., Delay-hopped transmitted reference UWBa tutorial, Special Issue on
UWB, January 2005.
[38] Pulson Communications Corp., PulsON Communications Corp request for pioneers preference in the personal communications services, Tech. Rep., GEN Docket 90-314, May 1992.
[39] H. Tomlinson and R. T. Hoctor, An overview of delay-hopped, transmitted-reference RF communications, Tech. Rep. 2001CRD198, GE Global Research, Niskayuna, NY, USA, January 2002.
[40] R. A. Scholtz, Multiple access with time-hopping impulse modulation, in Proceedings of IEEE
Military Communications Conference (MILCOM 93), vol. 2, pp. 447450, Boston, Mass, USA,
October 1993, invited paper.
[41] M. McClure, R. C. Qiu, and L. Carin, On the superresolution identification of observables from
swept-frequency scattering data, IEEE Transactions on Antennas and Propagation, vol. 45, no. 4,
pp. 631641, 1997.
[42] I.-T. Lu, R. C. Qiu, and J. Kwak, A high-resolution algorithm for complex spectrum search,
Journal of the Acoustical Society of America, vol. 104, no. 1, pp. 288299, 1998.
[43] J. G. Proakis, Digital Communications, McGraw-Hill, New York, NY, USA, 4th edition, 2000.
[44] R. C. Qiu, A study of the ultra-wideband wireless propagation channel and optimum UWB
receiver design (Part I), IEEE Journal on Selected Areas in Communications, vol. 20, no. 9, pp.
16281637, 2002, Special Issue on UWB Radio in Multiple Access Wireless Communications.
[45] R. C. Qiu, UWB Wireless communications, in Design and Analysis of Wireless Networks, Nova
Science Publishers, Hauppauge, NY, USA, 2004.
[46] R. C. Qiu, A generalized time domain multipath channel and its application in ultra-wideband
(UWB) wireless optimal receiver designpart II wave-based system analysis, IEEE Transactions
on Wireless Communications, vol. 3, no. 6, pp. 23122324, 2004.
[47] R. C. Qiu, A generalized time domain multipath channel and its application in ultra-wideband
(UWB) wireless optimal receiver design: Part III system performance analysis, to appear in IEEE
Transactions on Wireless Communications, February 2004.
[48] R. C. Qiu, A generalized time domain multipath channel and its application in ultra-wideband
(UWB) wireless optimal receiver design: system performance analysis, in Proceedings of IEEE
Wireless Communications and Networking Conference (WCNC 04), vol. 2, pp. 901907, Atlanta,
Ga, USA, March 2004.

Andreas F. Molisch et al.

135

[49] R. C. Qiu, Image Processing and Jump Regression Analysis, John Wiley & Sons, New York, NY,
USA, 2005.
[50] R. C. Qiu, UWB pulse propagationa generalized multipath model, January 2005.
[51] R. C. Qiu and I. T. Lu, Multipath resolving with frequency dependence for broadband wireless
channel modeling, IEEE Transransactions on Vehicular Technology, vol. 48, no. 1, pp. 273285,
1999.
[52] G. L. Turin, A statistical model of urban multipath propagation, IEEE Transransactions on
Vehicular Technology, vol. 21, no. 1, pp. 19, 1972.
[53] A. Taha and K. M. Chugg, On designing the optimal template waveform for UWB impulse radio
in the presence of multipath, in Proceedings of IEEE Conference on Ultra Wideband Systems and
Technologies (UWBST 02), pp. 4145, Baltimore, Md, USA, May 2002.
[54] R. M. Burher, UWB Channel Modeling, in DARPA NETEX Meeting, July 2002, Presentations
Slides.
[55] F. Sagnard, T. Quiniou, C. Vignat, and G. El Zein, High resolution algorithms for multipath
identification of construction materials: models and characterization for UWB applications,
IEEE Transransactions on Vehicular Technology, January 2004, submitted for publication.
[56] A. F. Molisch, Status Report of the Channel Modeling Subgroup, IEEE, Vancouver, British
Columbia, Canada, 2004.
[57] A. F. Molisch, Models for UWB Propagation Modelsa tutorial, Dynamics of Continuous,
Discrete and Impulsive SystemsAn International Journal for Theory and Applications, January
2005, Special issue on UWB.
[58] F. Ramirez-Mireles, On the performance of ultra-wide-band signals in Gaussian noise and
dense multipath, IEEE Transransactions on Vehicular Technology, vol. 50, no. 1, pp. 244249,
2001.
[59] F. Ramirez-Mireles, Signal design for ultra-wide-band communications in dense multipath,
IEEE Transransactions on Vehicular Technology, vol. 51, no. 6, pp. 15171521, 2002.
[60] W. Zhuang, X. Shen, and Q. Bi, Ultra-wideband wireless communicationsa tutorial, Wireless
Communications and Mobile Computing, vol. 3, no. 6, pp. 663685, 2003, special issue on UltraBroadband Wireless Communications for the Future.
[61] V. A. Borovikov and B. Y. Kinber, Geometrical Theory of Diraction, IEE Press, London, UK,
1994.
[62] R. Mittra and M. P. Hurst, Scattering center analysis via Pronys method, IEEE Transactions on
Antennas and Propagation, vol. 35, no. 8, pp. 986988, 1987.
[63] S. W. Lee, V. Jamnejad, and R. Mittra, An asymptotic series for early time response in transient
problems, IEEE Transactions on Antennas and Propagation, vol. 21, no. 6, pp. 895899, 1973.
[64] P. H. Pathak and P. R. Rousseau, Time domain uniform geometrical theory of diraction for
curved wedge, IEEE Transactions on Antennas and Propagation, vol. 43, no. 12, pp. 13751382,
1995.
[65] T. Veruttipong, Time domain version of the uniform GTD, IEEE Transactions on Antennas and
Propagation, vol. 38, no. 11, pp. 17571763, 1990.
[66] A. Mohsen and T. B. A. Senior, The impulse response of a half-plane, IEEE Transactions on
Antennas and Propagation, vol. 21, no. 3, pp. 254255, 1973.
[67] B. van der Pol and H. Bremmer, Operational Calculus Based on the Two-Sided Laplace Integral,
Cambridge University Press, Cambridge, UK, 2nd edition, 1955.
[68] S. M. Scarborough, Ultrawideband radar imaging and the diagnosis of scattering centers, M.S.
thesis, Ohio State University, Columbus, Ohio, USA, 1990.
[69] S. L. Borison and S. Hong, Short-pulse scattering by a conedirect and inverse, IEEE Transactions on Antennas and Propagation, vol. 16, no. 1, pp. 98102, 1968.
[70] K. B. Oldham and J. Spanier, The Fractional Calculus: Theory and Applications, Dierentiation
and Integration to Arbitrary Order, Academic Press, New York, NY, USA, 1974.
[71] K. S. Miller and B. Ross, An Introduction to the Fractional Calculus and Fractional Dierential
Equations, John Wiley & Sons, New York, NY, USA, 1993.
[72] R. A. Altes, Sonar for generalized target description and its similarity to animal echolocation
systems, Journal of the Acoustical Society of America, vol. 59, no. 1, pp. 97105, 1976.

136

UWB propagation channels

[73] N. Engheta, On fractional calculus and fractional multipoles in electromagnetism, IEEE Transactions on Antennas and Propagation, vol. 44, no. 4, pp. 554566, 1996.
[74] S. S. Ghassemzadeh, R. Jana, C. W. Rice, W. Turin, and V. Tarokh, A statistical path loss model
for in-home UWB channels, in Proceedings of IEEE Conference on Ultra Wideband Systems and
Technologies (UWBST 02), pp. 5964, Baltimore, Md, USA, May 2002.
[75] S. S. Ghassemzadeh, R. Jana, C. W. Rice, W. Turin, and V. Tarokh, Measurement and modeling
of an ultra-wide bandwidth indoor channel, IEEE Transactions on Communications, vol. 52,
no. 10, pp. 17861796, 2004.
[76] A. Alvarez, G. Valera, M. Lobeira, R. Torres, and J. L. Garcia, New channel impulse response
model for UWB indoor system simulations, in Proceedings of IEEE 57th Semiannual Vehicular
Technology Conference (VTC 03), vol. 1, pp. 15, Jeju, Korea, April 2003.
[77] D. Cassioli and A. Durantini, A time-domain propagation model of the UWB indoor channel
in the FCC-compliant band 3.6 - 6 GHz based on PN-sequence channel measurements, in
Proceedings of IEEE 59th Semiannual Vehicular Technology Conference (VTC 04), vol. 1, pp. 213
217, Milan, Italy, May 2004.
[78] D. Cassioli, A. Durantini, and W. Ciccognan, The role of path loss on the selection of the operating bands of UWB systems, in Proceedings of 15th IEEE International Symposium on Personal,
Indoor, and Mobile Radio Communications (PIMRC 04), vol. 4, pp. 27872791, Barcelona, Spain,
September 2004.
[79] D. Cassioli, M. Z. Win, and A. F. Molisch, The ultra-wide bandwidth indoor channel: from
statistical model to simulations, IEEE Journal on Selected Areas in Communications, vol. 20,
no. 6, pp. 12471257, 2002.
[80] A. Durantini and D. Cassioli, A multi-wall path loss model for indoor UWB propagation, in
Proceedings of IEEE Semiannual Vehicular Technology Conference (VTC 05), Stockholm, Sweden,
May 2005.
[81] A. Durantini, W. Ciccognani, and D. Cassioli, UWB propagation measurements by PNsequence channel sounding, in Proceedings of IEEE International Conference on Communications
(ICC 04), vol. 6, pp. 34143418, Paris, France, June 2004.
[82] S. S. Ghassemzadeh, L. J. Greenstein, T. Sveinsson, A. Kavcic, and V. Tarokh, UWB indoor path
loss model for residential and commercial environments, in Proceedings of IEEE 57th Semiannual Vehicular Technology Conference (VTC 03), Orlando, Fla, USA, October 2003.
[83] V. Hovinen, M. Hamalainen, and T. Patsi, Ultra wideband indoor radio channel models: preliminary results, in Proceedings of IEEE Conference on Ultra Wideband Systems and Technologies
(UWBST 02), pp. 7579, Baltimore, Md, USA, May 2002.
[84] J. Kunisch and J. Pamp, Measurement results and modeling aspects for the UWB radio channel, in Proceedings of IEEE Conference on Ultra Wideband Systems and Technologies (UWBST
02), pp. 1923, Baltimore, Md, USA, May 2002.
[85] M. Z. Win, R. A. Scholtz, and M. A. Barnes, Ultra-wide bandwidth signal propagation for
indoor wireless communications, in Proceedings of IEEE International Conference on Communications (ICC 97), pp. 5660, Montreal, Canada, June 1997.
[86] J. Keignart, J.-B. Pierrot, N. Daniele, et al., U.C.A.N. Report on UWB Basic Transmission Loss,
Tech. Rep. IST-2001-32710, Utility Consumers Action Network (UCAN), San Diego, Calif,
USA, March 2003.
[87] D. Cassioli, W. Ciccognani, and A. Durantini, D3.1: UWB channel model report, Tech. Rep.
IST-2001-35189 ULTRAWAVES, November 2003.
[88] A. Armogida, B. Allen, M. Ghavami, M. Porretta, G. Manara, and H. Aghvami, Path-loss modelling in short range UWB transmissions, in Proceedings of International Workshop on Ultra
Wideband Systems (IWUWBS 03), Oulu, Finland, June 2003.
[89] S. J. Howard and K. Pahlavan, Measurement and analysis of the indoor radio channel in the
frequency domain, IEEE Transactions on Instrumentation and Measurement, vol. 39, no. 5, pp.
751755, 1990.
[90] S. J. Howard and K. Pahlavan, Autoregressive modeling of wide-band indoor radio propagation, IEEE Transactions on Communications, vol. 40, no. 9, pp. 15401552, 1992.

Andreas F. Molisch et al.

137

[91] D. A. Hawbaker and T. S. Rappaport, Indoor wideband radiowave propagation measurements

at 1.3 GHz and 4.0 GHz, Electronics Letters, vol. 26, no. 21, pp. 18001802, 1990.
[92] T. S. Rappaport, Characterization of UHF multipath radio channels in factory buildings, IEEE
Transactions on Antennas and Propagation, vol. 37, no. 8, pp. 10581069, 1989.
[93] T. S. Rappaport, S. Y. Seidel, and K. Takamizawa, Statistical channel impulse response models
for factory an open plan building radio communication system design, IEEE Transactions on
Communications, vol. 39, no. 5, pp. 794807, 1991.
[94] A. A. M. Saleh and R. A. Valenzuela, A statistical model for indoor multipath propagation,
IEEE Journal on Selected Areas in Communications, vol. 5, no. 2, pp. 128137, 1987.
[95] S. Mangold, M. Lott, D. Evans, and R. Fifield, Indoor radio channel modeling-bridging from
propagation details to simulation, in Proceedings of 9th IEEE International Symposium on Personal, Indoor, and Mobile Radio Communications (PIMRC 98), vol. 2, pp. 625629, Boston,
Mass, USA, September 1998.
[96] G. Santella and E. Restuccia, Analysis of frequency domain wide-band measurements of the
indoor radio channel at 1, 5.5, 10 and 18 GHz, in Proceedings of IEEE Global Telecommunications
Conference (GLOBECOM 96), vol. 2, pp. 11621166, London, UK, November 1996.
[97] G. J. M. Janssen, P. A. Stigter, and R. Prasad, Wideband indoor channel measurements and BER
analysis of frequency selective multipath channels at 2.4, 4.75, and 11.5 GHz, IEEE Transactions
on Communications, vol. 44, no. 10, pp. 12721288, 1996.
[98] P. Nobles, D. Ashworth, and F. Halsall, Indoor radiowave propagation measurements at frequencies up to 20 GHz, in Proceedings of IEEE 44th Semiannual Vehicular Technology Conference
(VTC 94), vol. 2, pp. 873877, Stockholm, Sweden, June 1994.
[99] M. Dohler, B. Allen, A. Armogida, S. McGregor, M. Ghavami, and H. Aghvami, A new twist on
UWB pathloss modelling, in Proceedings of IEEE 59th Semiannual Vehicular Technology Conference (VTC 04), vol. 1, pp. 199203, Milan, Italy, May 2004.
[100] A. F. Molisch, Models for UWB propagation channelsa tutorial, Dynamics of Continuous,
Discrete and Impulsive SystemsAn International Journal for Theory and Applications, 2004.
[101] A. S. Y. Poon and M. Ho, Indoor multiple-antenna channel characterization from 2 to 8 GHz,
in Proceedings of IEEE International Conference on Communications (ICC 03), vol. 5, pp. 3519
3523, Anchorage, Alaska, USA, May 2003.
[102] A. F. Molisch, J. R. Foerster, and M. Pendergrass, Channel models for ultrawideband personal
area networks, IEEE Personal Communications, vol. 10, no. 6, pp. 1421, 2003.
[103] S. S. Ghassemzadeh and V. Tarokh, The ultra-wideband indoor path loss model, Tech. Rep.
P802.15 02/277r1SG3a, AT&T Labs, Florham Park, NJ, USA, 2002, IEEE P802.15 SG3a contribution.
[104] A. F. Molisch, Ed., Wideband Wireless Digital Communications, Prentice-Hall, New York, NY,
USA, 2000.
[105] M. Z. Win and R. A. Scholtz, On the energy capture of ultrawide bandwidth signals in dense
multipath environments, IEEE Communications Letters, vol. 2, no. 9, pp. 245247, 1998.
[106] J. G. Proakis, Digital Communications, McGraw Hill, New York, NY, USA, 3rd edition, 1995.
[107] J. Keignart and N. Daniele, Channel sounding and modelling for indoor UWB communications, in Proceedings of International Workshop on Ultra Wideband Systems (IWUWBS 03),
Oulu, Finland, June 2003.
[108] J. D. Choi and W. E. Stark, Performance of ultra-wideband communications with suboptimal
receivers in multipath channels, IEEE Journal on Selected Areas in Communications, vol. 20,
no. 9, pp. 17541766, 2002.
[109] D. Cassioli, M. Z. Win, F. Vatalaro, and A. F. Molisch, Performance of low-complexity RAKE
reception in a realistic UWB channel, in Proceedings of IEEE International Conference on Communications (ICC 02), vol. 2, pp. 763767, New York, NY, USA, 2002.
[110] R. Hoctor and H. Tomlinson, Delay-hopped transmitted-reference RF communications, in
Proceedings of IEEE Conference on Ultra Wideband Systems and Technologies (UWBST 02), pp.
265269, Baltimore, Md, USA, May 2002.
[111] A. Batra, et al., Multi-band OFDM physical layer proposal, Document IEEE 802.15-03/267r2,
2003.

138

UWB propagation channels

[112] D. Helal and P. Rouzet, Tg3a-stmicro-cfp-document, Tech. Rep. P802.15 03/140r0SG3a, 2003.
[113] G. L. Turin, F. D. Clapp, T. L. Johnston, S. B. Fine, and D. Lavry, A statistical model of urban
multipath propagation, IEEE Transransactions on Vehicular Technology, vol. 21, no. 1, pp. 19,
1972.
[114] H. Hashemi, Impulse response modeling of indoor radio propagation channels, IEEE Journal
on Selected Areas in Communications, vol. 11, no. 7, pp. 967978, 1993.
[115] M. Pendergrass and W. C. Beeler, Empirically based statistical ultra-wideband (UWB) channel
model, Tech. Rep. P802.15 02/240SG3a, Time-Domain Corporation, Huntsville, Ala, USA, June
2002, IEEE P802.15 SG3a contribution.
[116] F. Zhu, Z. Wu, and C. R. Nassar, Generalized fading channel model with application to UWB,
in Proceedings of IEEE Conference on Ultra Wideband Systems and Technologies (UWBST 02), pp.
1317, Baltimore, Md, USA, May 2002.
[117] D. Cassioli and A. Durantini, Measurements, Modeling and Simulations of the UWB Propagation Channel based on Direct-Sequence Channel Sounding, Wireless Communications and
Mobile Computing, vol. 5, no. 5, August 2005.
[118] R. J.-M. Cramer, R. A. Scholtz, and M. Z. Win, Evaluation of an ultra-wide-band propagation
channel, IEEE Transactions on Antennas and Propagation, vol. 50, no. 5, pp. 561570, 2002.
[119] J. R. Foerster and Q. Li, UWB channel modeling contribution from Intel, Tech. Rep. P802.15
02/279SG3a, Intel Corporation, Hillboro, Ore, USA, 2002, IEEE P802.15 SG3a contribution.
[120] P. Pagani, P. Pajusco, and S. Voinot, A study of the ultra-wideband indoor channel: propagation experiment and measurement results, in COST273 TD(030)060, Barcelona, Spain, January
2003.
[121] M. Z. Win, F. Ramrez-Mireles, R. A. Scholtz, and M. A. Barnes, Ultra-wide bandwidth (UWB)
signal propagation for outdoor wireless communications, in IEEE 47th Vehicular Technology
Conference (VTC 97), vol. 1, pp. 251255, Phoenix, Ariz, USA, May 1997.
[122] A. Domazetovic, L. J. Greenstein, N. B. Mandayam, and I. Seskar, A new modeling approach
for wireless channels with predictable path geometries, in IEEE 56th Vehicular Technology Conference (VTC 02), vol. 1, pp. 454458, Vancouver, British Columbia, Canada, September 2002.
[123] J. Kunisch and J. Pamp, Radio channel model for indoor UWB WPAN environments, presentation P802.15 02/281SG3a, IMST GmbH, June 2002, IEEE P802.15 SG3a contribution.
[124] H. Zhang, T. Udagawa, T. Arita, and M. Nakagawa, A statistical model for the small-scale multipath fading characteristics of ultra-wideband indoor channel, in Proceedings of IEEE Conference
on Ultra Wideband Systems and Technologies Digest of Technical Papers (UWBST 02), pp. 8185,
Baltimore, Md, USA, May 2002.
[125] W. Turin, R. Jana, S. S. Ghassemzadeh, C. W. Rice, and V. Tarokh, Autoregressive modeling
of an indoor UWB channel, in Proceedings of IEEE Conference on Ultra Wideband Systems and
Technologies (UWBST 02), pp. 7174, Baltimore, Md, USA, May 2002.
[126] R. M. Buehrer, W. A. Davis, A. Safaai-Jazi, and D. Sweeney, Ultra wideband propagation measurements and modeling - final report to darpa netex program, Tech. Rep., Virginia Tech,
Blacksburg, Va, USA, 2004, available for download at http://www.mprg.org/people/buehrer/
ultra/darpa netex.shtml.
[127] A. Muqaibel, A. Safaai-Jazi, A. Bayram, and S. M. Riad, Ultra wideband material characterization for indoor propagation, in Proceedings of IEEE Antennas and Propagation Society International Symposium (APS 03), vol. 4, pp. 623626, Columbus, Ohio, USA, June 2003.
[128] J. A. Hogbom, Aperture synthesis with a non-regular distribution of interferometer baselines,
Astronomy and Astrophysics Supplement Series, vol. 15, pp. 417426, 1974.
[129] S. M. Yano, Investigating the ultra-wideband indoor wireless channel, in IEEE 55th Vehicular
Technology Conference (VTC 02), vol. 3, pp. 12001204, Birmingham, Ala, USA, May 2002.
[130] R. C. Qiu, A generalized time domain multipath channel and its application in ultra-wideband
(uwb) wireless optimal receiver design: Wave-based system ananlysis, IEEE Transactions on
Wireless Communications, page in press, 2003.
Alvarez,

[131] A.
G. Valera, M. Lobeira, R. P. Torres, and J. L. Garca, UWB channel model contribution from university of cantabria and ACORDE, presentation P802.15 02/445, University of
Cantabria, ACORDE, Santander, Spain, 2002, IEEE P802.15 SG3a contribution.

Andreas F. Molisch et al.

139

[132] R. J.-M. Cramer, An evaluation of indoor ultra-wideband communication channels, presentation P802.15 02/325SG3a, TRW Space & Electronics, Redondo Beach, Calif, USA, June 2002,
IEEE P802.15 SG3a contribution.
[133] S. S. Ghassemzadeh, L. J. Greenstein, and V. Tarokh, The ultra-wideband indoor multipath
model, Tech. Rep. P802.15 02/282r1SG3a, AT&T Labs, Florham Park, NJ, USA, 2002, IEEE
P802.15 SG3a contribution.
[134] G. Shor, A proposal for a selection of indoor UWB path loss model, Tech. Rep. IEEE P802.1502/280-SG3a, Wisair, 2002.
[135] K. Siwiak, UWB propagation phenomena, Presentation P802.15 02/301SG3a, Time-Domain
Corporation, Huntsville, Ala, USA, 2002, IEEE P802.15 SG3a contribution.
[136] M. Z. Win, D. Cassioli, and A. F. Molisch, The ultra-wide bandwidth indoor channel: from
statistical model to simulation, Tech. Rep. P802.18 02/284SG3a, Mitsubishi, 2002, IEEE P802.15
SG3a contribution.
[137] J. R. FoersterEd., Channel modeling sub-committee report final, Tech. Rep. P802.15 02/490r1,
IEEE 802.15 SG3a, February 2003.
[138] My personal Adaptive Global NET (MAGNET), Technical report, European IST FP6 project,
IST-2004-507102, (http://www.ist-magnet.org), March 2004.
[139] Universal Remote Signal Acquisition for hEalth (URSAFE), Technical report, European IST FP5
project, Contract No IST-2001-33352, (http://ursafe.tesa.prd.fr), October 2003.
[140] M. A. Uusitalo, Wireless World Research Forum (WWRF), Technical report, general presentation (www.wireless-world-research.org), March 2004.
[141] W. A. Th. Kotterman, G. F. Pedersen, and K. Olesen, Capacity of the mobile MIMO channel for
a small wireless handset and user influence, in Proceedings of 13th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 02), vol. 4, pp. 19371941,
Lisbon, Portugal, September 2002.
[142] W. A. Th. Kotterman, G. F. Pedersen, and K. Olesen, Diversity properties of multi-antenna
small handheld terminals, EURASIP Journal on Applied Signal Processing, vol. 2004, no. 9, pp.
13401353.
[143] J. . Nielsen and G. F. Pedersen, On the influence of the liquid type on mobile phone measurements using body phantoms, in Proceedings of 4th International Symposium on Wireless Personal
Multimedia Communications (WPMC 01), vol. 2, pp. 755760, Aalborg, Denmark, September
2001.
[144] J. . Nielsen and G. F. Pedersen, In-network evaluation of body carried mobile terminal performance, in Proceedings of IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 01), vol. 1, pp. D109D113, San Diego, Calif, USA, September
October 2001.
[145] Future Adaptive Communication Environment (FACE), Technical report, Center For PersonKommunikation/Center for TeleInFrastruktur, Internal Research project, (http://www.cpk.auc.
dk/FACE), Aalborg University, Aalborg, Denmark, October 2003.
[146] Power Aware Communications for Wireless OptiMised personal Area Networks (PACWOMAN), Technical report, European IST FP5 project, Contract No IST-2001-34157, (http://
www.imec.be/pacwoman), October 2003.
[147] T. B. Welch, R. L. Musselman, B. A. Emessiene, et al., The eects of the human body on UWB
signal propagation in an indoor environment, IEEE Journal on Selected Areas in Communications, vol. 20, no. 9, pp. 17781782, 2002.
[148] T. Zasowski, F. Althaus, M. Stager, A. Wittneben, and G. Troster, UWB for non-invasive wireless body area networks: channel measurements and results, in Proceedings of IEEE Conference
on Ultra Wideband Systems and Technologies (UWBST 03), pp. 285289, Reston, Va, USA, November 2003.
[149] Revision of Part 15 of the Commissions Rules Regarding Ultra-Wideband Transmission Systems, First Report and Order, ET Docket 98-153, FCC 02-48. Technical report, Federal Communications Commission, April 2002.
[150] A. Petro and P. Withington, PulsON technology overview, Tech. Rep., Time Domain Inc.,
Huntsville, Ala, USA, July 2001.

140

UWB propagation channels

[151] 3.1-6.0 GHz Ultra-Wideband Antenna for Commercial UWB Applications, Tech. Rep., SkyCross
Inc., Melbourne, Fla, USA, January 2003.
[152] M. Klemm and G. Troster, Characterisation of an aperture-stacked patch antenna for ultrawideband wearable radio systems, in Proceedings of 15th International Conference on Microwaves, Radar, and Wireless Communications (MIKON 04), vol. 2, pp. 395398, Warsaw,
Poland, May 2004.
[153] M. Klemm and G. Troster, Small patch antennas for ultra-wideband wireless body area network, in Proceedings of 7th Ultra-Wideband, Short-Pulse Electromagnetics Conference, Magdeburg, Germany, July 2004.
[154] N. Behdad and K. Sarabandi, A multiresonant single-element wideband slot antenna, IEEE
Antennas and Wireless Propagation Letters, vol. 3, pp. 58, 2004.
[155] S. K. Sharma, L. Shafai, and N. Jacob, Investigation of wide-band microstrip slot antenna, IEEE
Transactions on Antennas and Propagation, vol. 52, no. 3, pp. 865872, 2004.
[156] S. C. Hagness, A. Taflove, and J. E. Bridges, Three-dimensional FDTD analysis of a pulsed
microwave confocal system for breast cancer detection: design of an antenna-array element,
IEEE Transactions on Antennas and Propagation, vol. 47, no. 5, pp. 783791, 1999.
[157] J. Wang, O. Fujiwara, S. Watanabe, and Y. Yamanaka, Computation with a parallel FDTD system of human-body eect on electromagnetic absorption for portable telephones, IEEE Transactions on Microwave Theory and Techniques, vol. 52, no. 1, pp. 5358, 2004.
[158] Technical Characteristics and Test Conditions for Radio Equipment Using an Integral Antenna
Transmitting Signals to Initiate a Specific Response in the Receiver, ETS 300 341, Technical report, European Telecommunications Standards Institute (ETSI), July 1995.
[159] C. Gabriel, Compilation of the dielectric properties of body tissues at RF and microwave frequencies, AL/OE-TR-1996-0037, Technical report, US Air Force Armstrong Laboratory, Brooks
AFB, Tex, USA, June 1996.
[160] G. F. Pedersen, Phantoms for radiation measurements of mobile phones, in Proceedings of IEEE
International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 01),
vol. 1, pp. C95C99, San Diego, Calif, USA, SeptemberOctober 2001.
[161] G. F. Pedersen, J. . Nielsen, K. Olesen, and I. Z. Kovacs, Antenna diversity on a UMTS handheld phone, in Proceedings of 10th IEEE International Symposium on Personal, Indoor and Mobile
Radio Communications (PIMRC 99), vol. 1, pp. 152156, Osaka, Japan, September 1999.
[162] W. A. Th. Kotterman, G. F. Pedersen, K. Olesen, and P. C. F. Eggers, Cable-less measurement
setup for wireless handheld terminals, in Proceedings of 12th IEEE International Symposium on
Personal, Indoor and Mobile Radio Communications (PIMRC 01), vol. 1, pp. B112B116, San
Diego, Calif, USA, SeptemberOctober 2001.
[163] 3.1-6.0 GHz Ultra-Wideband Antenna for Commercial UWB Applications, Tech. Rep., SkyCross
Inc., Melbourne, Fla, USA, November 2002.
[164] G. F. Pedersen and J. . Nielsen, Radiation pattern measurements of mobile phones next to
dierent head phantoms, in Proceedings of 13th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 02), vol. 4, pp. 18881892, Lisbon, Portugal,
September 2002.
[165] I. Z. Kovacs and P. C. F. Eggers, Short-range UWB radio propagation investigations using
small terminal antennas, in Proceedings of International Workshop on Ultra Wideband Systems
(IWUWBS 03), Oulu, Finland, June 2003.
[166] A. Taha and K. M. Chugg, On designing the optimal template waveform for UWB impulse radio
in the presence of multipath, in Proceedings of IEEE Conference on Ultra Wideband Systems and
Technologies (UWBST 02), pp. 4145, Baltimore, Md, USA, May 2002.
[167] A. Batra, et al., Multi-band OFDM physical layer proposal for IEEE 802.15 task group 3a,
www.multibandofdm.org.
[168] R. Van Nee and R. Prasad, OFDM for Wireless Multimedia Applications, Artech House Publishers,
London, UK, 2000.
[169] L. Scharf, Statistical Signal Processing, Addison-Wesley, New York, NY, USA, 1991.

Andreas F. Molisch et al.

141

[170] J.-J. van de Beek, O. Edfors, M. Sandell, S. K. Wilson, and P. O. Borjesson, On channel estimation in OFDM systems, in IEEE Vehicular Technology Conference (VTC 95), vol. 2, pp. 815819,
Chicago, Ill, USA, July 1995.
[171] M. Bossert, A. Donder, and V. Zyablov, Improved channel estimation with decision feedback
for OFDM systems, Electronics Letters, vol. 34, no. 11, pp. 10641065, 1998.
[172] V. Mignone and A. Morello, CD3-OFDM: a novel demodulation scheme for fixed and mobile
receivers, IEEE Transactions on Communications, vol. 44, no. 9, pp. 11441151, 1996.
[173] O. Edfors, M. Sandell, J.-J. van de Beek, S. K. Wilson, and P. O. Borjesson, OFDM channel
estimation by singular value decomposition, IEEE Transactions on Communications, vol. 46,
no. 7, pp. 931939, 1998.
[174] Y. Li, L. J. Cimini Jr., and N. R. Sollengerger, Robust channel estimation for OFDM systems
with rapid dispersive fading channels, IEEE Transactions on Communications, vol. 46, no. 7, pp.
902915, 1998.
[175] M. Z. Win and R. A. Scholtz, Impulse radio: how it works, IEEE Communications Letters, vol. 2,
no. 2, pp. 3638, 1998.
[176] M. Z. Win and R. A. Scholtz, Ultra-wide bandwidth time-hopping spread-spectrum impulse
radio for wireless multiple-access communications, IEEE Transactions on Communications,
vol. 48, no. 4, pp. 679689, 2000.
[177] M. Hamalainen, V. Hovinen, and J. Iinatti, Performance comparison between various UWB
signals in AWGN channel in presence of multitone interference at the GSM downlink band,
in Proceedings of 4th International Symposium on Wireless Personal Multimedia Communications
(WPMC 01), pp. 449453, Aalborg, Denmark, September 2001.
[178] V. S. Somayazulu, Multiple access performance in UWB systems using time hopping vs. direct
sequence spreading, in Proceedings of IEEE Wireless Communications and Networking Conference
(WCNC 02), vol. 2, pp. 522525, Orlando, Fla, USA, March 2002.
[179] R. Roberts, Xtremespectrum CFP document, http://grouper.ieee.org/groups/802/15/pub/2003/
Jul03/.
[180] M. Z. Win, Ultra-wide bandwidth spread-spectrum techniques for wireless multiple access communications, Ph.D. thesis, Department of Electrical Engineering, University of Southern California,
Los Angeles, Calif, USA, 1997.
[181] E. Ertin, U. Mitra, and S. Siwamogsatham, Maximum-likelihood based multipath channel
estimation for code-division multiple-access systems, IEEE Transactions on Communications,
vol. 49, no. 2, pp. 290302, 2001.
[182] V. Lotticia, A. DAndrea, and U. Mengali, Channel estimation for ultra-wideband communications, IEEE Journal on Selected Areas in Communications, vol. 20, no. 9, pp. 16381645, 2002.
[183] E. G. Strom and F. Malmsten, A maximum likelihood approach for estimating DS-CDMA multipath fading channels, IEEE Journal on Selected Areas in Communications, vol. 18, no. 1, pp.
132140, 2000.
[184] A. A. DAmico, U. Mengali, and M. Morelli, Multipath channel estimation for the uplink
of a DS-CDMA system, in Proceedings of IEEE International Conference on Communications
(ICC 02), vol. 1, pp. 1620, New York, NY, USA, AprilMay 2002.
[185] B. Mielczarek, M.-O. Wessman, and A. Svensson, Performance of coherent UWB Rake receivers
using dierent channel estimators, in Proceedings of International Workshop on Ultra Wideband
Systems (IWUWBS 03), Oulu, Finland, June 2003.
[186] M. Z. Win and R. A. Scholtz, On the energy capture of ultrawide bandwidth signals in dense
multipath environments, IEEE Communications Letters, vol. 2, no. 9, pp. 245247, 1998.
[187] L. L. Chong, The eect of channel estimation errors on wideband CDMA systems, Ph.D. thesis,
University of California, San Diego, La Jolla, Calif, USA, 2001.
[188] R. Tesi, M. Hamalainen, and J. Iinatti, Impact of the number of fingers of a selective Rake
receiver for UWB systems in modified Saleh-Valenzuela channel, in Proceedings of 4th Finnish
Wireless Communication Workshop (FWCW 03), Oulu, Finland, October 2003.

142

UWB propagation channels

[189] H. Niu, J. A. Ritcey, and H. Liu, Performance of UWB RAKE receivers with imperfect tap
weights, in Proceedings of IEEE Int. Conf. Acoustics, Speech, Signal Processing (ICASSP 03), vol. 4,
pp. 125128, Hong Kong, China, April 2003.
[190] K. C. Hwang and K. B. Lee, Performance analysis of low processing gain DS/CDMA systems
with random spreading sequences, IEEE Communications Letters, vol. 2, no. 12, pp. 315317,
1998.
[191] J. G. Proakis, Digital Communications, McGraw-Hill, New York, NY, USA, 1986.
[192] T. S. Rappaport, Wireless Communications: Principles and Practice, Prentice-Hall, Upper Saddle
River, NJ, USA, 1996.

Andreas F. Molisch: Mitsubishi Electric Research Labs, Cambridge, MA, USA;

Department of Electroscience, Lund University, Lund, Sweden
Email: [email protected]
Jurgen Kunisch: IMST GmbH, Carl-Friedrich-Gauss-Strasse 2, 47475 Kamp-Lintfort, Germany
Email: [email protected]
Robert Qiu: Center for Manufacturing Research, Tennessee Technological University,
P.O. Box 5077, Cookeville, TN 38505, USA
Email: [email protected]
Dajana Cassioli: RadioLabs, Dipartimento di Ingegneria Elettronica, Universit`a degli Studi di Roma
Tor Vergata, Via del Politecnico 1, 00133 Roma, Italy
Email: [email protected]
Michael Buehrer: Department of Electrical and Computer Engineering, Virginia Tech, Blacksburg,
VA 24061-0111, USA
Email: [email protected]
Marcus Pendergrass: Department of Mathematics and Computer Science, Longwood University,
201 High Street, Farmville, VA 23909, USA
Email: [email protected]
Istvan Z. Kovacs: Department of Communication Technology, Aalborg University, Niels Jernes Vej 12,
9220 Aalborg , Denmark
Gert F. Pedersen: Department of Communication Technology, Aalborg University, Niels Jernes Vej 12,
9220 Aalborg , Denmark
Email: [email protected]
Patrick C. F. Eggers: Department of Communication Technology, Aalborg University,
Jernes Vej 12, 9220 Aalborg , Denmark
Email: [email protected]
Sumit Roy: Department of Electrical Engineering, University of Washington, Seattle,
WA 98195-2500, USA
Email: [email protected]
Iyappan Ramachandran: Department of Electrical Engineering, University of Washington,
Seattle, WA 98195-2500, USA
Email: [email protected]

Signal processing
Ian Oppermann, Matti Hamalainen, Jari Iinatti,
Alberto Rabbachin, Ben Allen, Seyed A. Ghorashi,
Mohammad Ghavami, Olaf Albert,
and Christoph F. Mecklenbrauker

3.1. Introduction
Many dierent pulse generation techniques may be used to satisfy the requirements of a UWB signal. As discussed in the previous chapters, the FCC requires
that the fractional bandwidth is greater than 20%, or that the bandwidth of the
transmitted signal is more than 500 MHz, whichever is less. The FCC also stipulates peak power requirements [1]. Many possible solutions may be developed
within these restrictions to occupy the available bandwidth.
UWB systems have historically been based on impulse radio concepts. Impulse radio refers to the generation of a series of very short duration pulses, of the
order of hundreds of pico seconds. Each pulse has a very wide spectrum that must
adhere to the spectral mask requirements. Any given pulse will have very low energy because of the very low power levels permitted for typical UWB transmission.
Therefore, many pulses will typically be combined to carry the information for one
bit. Continuous pulse transmission introduces a complication in that, without further signal processing at the transmitter, strong spectral lines will be introduced
into the spectrum of the transmitted signal. Several techniques are available for
minimising these spectral lines, the most common of which are described later in
this chapter.
Impulse radio has the significant advantage in that it is essentially a baseband
technique. The most common impulse-radio-based UWB concepts are based on
pulse position modulation with time hopping (TH-PPM). Time-hopping, directsequence techniques, and multicarrier schemes are also described in this chapter.
However, the focus will be on impulse radio modulation schemes.
This chapter will address signal processing issues in UWB. The chapter will
address both impulse radio techniques and multiband techniques.

144

Signal processing
The 1st derivative of Gaussian pulse
1

Amplitude

0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
1 0.8 0.6 0.4 0.2

0 0.2 0.4 0.6 0.8

Time (ns)

= 0.25 ns
= 0.5 ns
= 0.75 ns

= 1 ns
= 2 ns

Figure 3.1. Gaussian monocycle in time domain.

3.2. Impulse radio schemes

3.2.1. Impulse radio UWB
Time-modulated ultra-wideband (TM-UWB) communication is based on discontinuous transmission of very short pulses or pulsed waveforms (monocycles), examples of which are presented in Figure 3.1. These pulses have the corresponding
ultra-wide frequency-domain characteristics presented in Figure 3.2. This type of
transmission does not require the use of additional carrier modulation as the pulse
will propagate well in the radio channel. The technique is therefore a baseband signal approach. This radio concept is referred to as impulse radio (IR).
One transmitted symbol is spread over N monocycles to achieve a processing
gain that may be used to combat noise and interference. This is similar to the
approach used for spread spectrum systems. The processing gain derived from
this procedure can be defined as
PG1 = 10 log10 (N).

(3.1)

The monocycle waveform can be any function that satisfies the spectral mask
regulatory requirements. Common pulse shapes include Gaussian, Laplacian, or
Rayleigh or Hermitean pulses. Data modulation is typically based on pulse position modulation (PPM). In [2], the authors also present pulse amplitude modulation (PAM) in UWB transmission. The UWB receiver is typically a homodyne cross-correlator that is based on an architecture that utilises a direct RFto-baseband conversion. Intermediate frequency conversion is not needed, which
makes the implementation simpler than that in conventional (super-)heterodyne
systems.

Ian Oppermann et al.

145
Gaussian monocycle

0
10

Power (dB)

20
30
40
50
60
70

0.5

1.5

2 2.5 3 3.5
Frequency (GHz)

4.5

= 0.5 ns
= 0.75 ns
= 2 ns
Figure 3.2. Gaussian monocycle in frequency domain.

Unlike spread spectrum systems, the UWB pulse does not necessarily occupy
the entire chip period. This means that the duty cycle can be extremely low. In a
low delay spread channel, the receiver is only required to listen to the channel for
a small fraction of the period between pulses. The impact of any continuous source
of interference is therefore reduced so that it is only relevant when the receiver is
attempting to detect a pulse. This leads to processing gain in a similar manner to
a spread spectrum systems ability to reduce the impact of interference. Processing
gain due to the low duty cycle is given by
'

PG2 = 10 log10

Tp
,
p

(3.2)

where
T p = chip period,
p = impulse width.

(3.3)

The total processing gain PG is the sum of the two processing gains [3]
PG = PG1 + PG2 .

(3.4)

In a discontinuous time-hopped system, consecutive pulses are transmitted

according to a pseudorandom (PR) time-hopping code. As the transmitted signal
is not continuous, UWB communication is resistant to severe multipath propagation. With a short pulse width and a relative long pulse repetition time (compared
to pulse width), the multipath components associated with the transmitted pulse

146

Signal processing
Impulse radio transmitter

Modulation

Impulse radio receiver

Pulse
generator

Crosscorrelator

Variable
delay

Variable
delay

Code
generator

Code
generator

Time
base

Time
base

LPF

BPF

DSP

Modulation
output

Figure 3.3. Block diagram of the TH-PPM UWB impulse radio concept by time domain corporation.

are attenuated before the next pulse is sent, reducing the inter-pulse interference.
If the time between pulses is greater than the channel delay spread, there is no ISI
between pulses and so no ISI between bits.
In the time domain, the transmitted Gaussian monocycle can be (ideally) approximated by the first derivative of the Gaussian function. Figures 3.1 and 3.2
shows the time and frequency domains of sample monocycles of dierent duration.
The Gaussian monocycle in the time domain v(t) is given by [3]
6

v(t) = 6A

e t 6(t/ p )2
e
,
3 p

(3.5)

where A is the pulse amplitude, p is the pulse width, and t is the time.
The corresponding function in the frequency domain is the Fourier transform
of v(t), given by
)

F v(t) = V ( f ) = j

A f p2
3

e (/6) f 2 p2
,
e
2

(3.6)

where f is the frequency.

The nominal centre frequency and the bandwidth of the signal depends on
the monocycles duration in time. The 3 dB bandwidth is approximately 116%
of the nominal centre frequency f0 = 1/ p [3]. Considering an example monocycle duration of 0.75 ns, as shown in Figure 3.1, the nominal centre frequency is
1.33 GHz and the half-power bandwidth is 1.55 GHz. The spectrum of Gaussian
monocycle is asymmetrical, as can be seen in Figure 3.2.
Figure 3.3 presents a block diagram of a time-hopping UWB impulse radio
concept that utilises pulse position modulation (concept by Time Domain Corporation, USA).

Ian Oppermann et al.

147

3.2.2. Fast stepped frequency chirps

Both the bandwidth and spectral mask limitations must be satisfied by the UWB
signal. There are many means of generating a very wideband signal. One technique is based on fast frequency chirps, which are commonly used in impulse
radar applications. It is possible to generate a wideband transmission by sweeping the transmitters oscillator in the frequency domain. A bandwidth of several
hundred MHz can be achieved with a 10 ns sweep time [4]. Wider bandwidths can
be achieved using this technique, for example, ground penetrating radar (GPR)
systems with 501200 MHz bandwidth have been documented in [5].
GPRs based on UWB technology are suitable for object detection in, for example, landmine sweeping and avalanche rescue operations because of the good
signal penetration ability and fine space resolution.
3.3. Multicarrier schemes
3.3.1. Multicarrier spread spectrum schemes
Another approach is to extend the techniques utilised for direct-sequence spread
spectrum or code-division multiple-access (CDMA) schemes that are used for
third generation mobile systems. Wideband CDMA systems with optional multicarrier (MC) techniques can be used to fill the available spectral mask.
There are three main techniques to generate an SS-MC transmission; multicarrier CDMA, multicarrier DS-CDMA, and multitone (MT) CDMA [6]. Each
of these techniques relies on, and benefits from, the properties of conventional
spread spectrum signals. However, multicarrier systems are reasonably complex
to implement. In particular, multicarrier systems require several mixers or digital
fast Fourier transform (FFT) techniques to place the dierent signal components
in the required bands.
Figure 3.4 shows a block diagram for an MC-CDMA system. The original data
stream is spread over the dierent subcarriers fi with each chip of pseudorandom
(PR) code Ci . The spectrum spreading is done in the frequency domain [6]. The
signal is de-spread at the receiver using corresponding chips gi of the spreading
code C. In UWB applications, the individual modulated carrier needs to fulfil the
500 MHz bandwidth requirements.
Figure 3.5 shows a block diagram for a multicarrier DS-CDMA system. This
method involves spreading the original data in the time domain after serial-toparallel conversion of the data stream [6]. The system needs to obtain the 500 MHz
minimum bandwidth requirement to be treated as UWB.
Figure 3.6 shows the block diagram of a multitone-CDMA system. The bandwidth of an MT-CDMA system is smaller than in the previous multicarrier systems
because of the small subcarrier spacing. This kind of multicarrier approach is the
closest to the original UWB idea. However, it causes the highest self-interference
due to the overlapping spectra.

148

Signal processing
cos(2 f1 t) g1

C1

cos(2 f1 t)

C2
Cn

..
.

LPF

cos(2 f2 t) g2

cos(2 f2 t)

LPF
.
.
.
LPF

cos(2 fn t) gn

cos(2 fn t)

f1 f2

fn

Figure 3.4. Block diagram and spectrum for multicarrier CDMA system.

C(t)

cos(2 f1 t) C(t)

cos(2 f1 t)

C(t) cos(2 f t)
Serial-to2
parallel
.
converter C(t) .. cos(2 fn t)

cos(2 f2 t) C(t)

.
cos(2 fn t) C(t)..

LPF
LPF
LPF

Parallel-toserial
converter

f1 f2

fn

Figure 3.5. Block diagram and spectrum for multicarrier DS-CDMA system.

C1
C2
Serial-toparallel C .
n .
.
converter

cos(2 f1 t)

cos(2 f1 t)

cos(2 f2 t)

cos(2 f2 t)
cos(2 fn t)

cos(2 fn t).
.
.

f1 f2 fn

Rake
combiner 1
Rake
combiner 2
Rake
combiner n

Parallel-toserial
converter

Figure 3.6. Block diagram and spectrum for multitone CDMA system.

Multicarrier technology is used in high data rate applications, for example, in

WLAN systems such as Hiperlan2, digital audio or video broadcasting (DAB and
DVB, resp.), and in asymmetric digital subscriber line (ADSL).
The advantage of multicarrier technologies over single-carrier systems is that
the data rate in each subcarrier is lower than for single-carrier systems. This eases
synchronisation of the spreading sequence at the receiver, and helps to avoid ISI.
The disadvantage is the increased complexity of the receiver requiring either multiple mixing stages or FFT processing.

Ian Oppermann et al.

149

Power spectrum magnitude (dB)

Spectral planGroup A
1
0
1
2
3
4
5
6
7
8
9
10

10

Frequency

11
109

Figure 3.7. Multiband frequency planGroup A.

MC-CDMA schemes spread the signal in the frequency domain. The signal
consists of overlapping, relatively narrow carriers that fill the available UWB signal spectrum. MC-DS-CDMA and MT-CDMA schemes apply spreading in the
time domain. MT-CDMA has a similar bandwidth to a basic DS-CDMA scheme
[6] with relatively small separation of the tones used ( f1 to fn in Figure 3.6). Consequently, the spreading factor employed for MT-CDMA schemes is much higher
than that for the MC schemes. The increased spreading factor increases the processing speed required at the receiver.
The conventional DS-SS technique without multicarrier properties can also
be characterised as a UWB technique if the chip rate is high enough. This calls for
extremely fast digital signal processing which may be impractical. Issues such as
synchronisation will also be a significant challenge.
3.3.2. Multiband UWB
One early proposal within the IEEE 802.15.3 working group for UWB signals [7]
utilised overlapping groups of UWB signals each with a bandwidth of 528 MHz.
This so-called multiband UWB ensures adherence to the FCC minimum bandwidth requirements and allows for ecient utilisation of the available spectrum.
Figure 3.7 shows the spectrum plan for the first group of UWB signals. The
subbands are spaced 470 MHz apart, and multiple 528 MHz signals may be
utilised. This allows for flexible coexistence with existing communications systems (such as WLAN systems). Figure 3.8 shows the second group of UWB signals
which overlap the first group of UWB signals by 235 MHz. This potentially enhances the systems flexibility with respect to coexistence, interference mitigation,
and multiple access.
With this proposal, each group of UWB signals is divided into lower sets (subbands 18) and upper sets (subbands 915). Only 7 subbands are used in the lower

150

Signal processing

Power spectrum magnitude (dB)

Spectral planGroup A + Group B

1
0
1
2
3
4
5
6
7
8
9
10

Frequency

10

11
109

Figure 3.8. Multiband frequency planGroups A and B.

set, which means one subband can be avoided for coexistence. The upper set is
used in parallel to the lower set to increase the bit rate.
3.4. Pulse shapes
This section describes the pulse shapes which may be used with UWB systems.
Whilst primarily focused on impulse radio systems, the pulse shapes can also be
applied to multiband systems.
3.4.1. Rectangular monocycle
The simplest example of an impulse is a rectangular monocycle which can be represented as follows:


PRe (t) =


&
1 %
U(t) U t T p ,
Tp

(3.7)

where T p denotes the pulse width and U() represents the unit step function. Thus,
by adjusting T p , the bandwidth of the signal can be controlled. Taking the Fourier
transform of PRe yields the familiar sin(x)/x function with nulls occurring at 1/T p
rad/s. The sin(x)/x function contains a DC term making the wireless transmission
of the rectangular monocycle impractical. The time and frequency domains of an
example rectangular pulse are shown in Figures 3.9 and 3.10 where the monocycle
has been normalised for unit energy.

Ian Oppermann et al.

151
104

Amplitude

5
4
3
2
1
0

10

15

20
1010

Time (s)

Figure 3.9. Rectangular monocycle in time domain.

105

3.5

Magnitude

3
2.5
2
1.5
1
0.5
0
0

0.5

1.5

2.5

3.5

Frequency (Hz)

4.5

109

Figure 3.10. Rectangular monocycle in frequency domain.

3.4.2. Rayleigh monocycle

The Rayleigh monocycle [8], which is derived from the first derivative of the
Gaussian pulse (see below), has the time-domain representation, PR (t), given by
0

PR (t) = AR

t (1/2)((t)/)2
e
,
2

(3.8)

where is the pulse width and is the centre of the pulse. Thus by adjusting ,
the bandwidth can be controlled. The term AR is a scaling factor which has been
introduced such that the total energy in the pulse is normalised to unity.

152

Signal processing
1.5

105

Amplitude

1
0.5
0
0.5
1
1.5
2

Time (s)

1010

Figure 3.11. Rayleigh monocycle in time domain.

1.2

105

Magnitude

1
0.8
0.6
0.4
0.2
0

0.2 0.4 0.6 0.8

1.2 1.4 1.6 1.8 2

1010

Frequency (Hz)

Figure 3.12. Rayleigh monocycle in frequency domain.

The frequency spectrum of the Rayleigh monocycle, PR ( f ), is

PR (t) = AR 2(2 f )e(1/2)(2 f ) e j(2 f +0.5) .

(3.9)

The eective time duration is defined as the time duration of the waveform that
contains 99.99% of the total monocycle energy. Here, 99.99% of the total Rayleigh
monocycle energy is in T p = 7 and centred at = 3.5.
Rayleigh monocycles do not have a DC component enabling them to be used
in impulse radio systems. The time and frequency domains of an example Rayleigh
monocycle are shown in Figures 3.11 and 3.12 where the monocycle has been normalised for unit energy.

Ian Oppermann et al.

153
108

Amplitude

3
2
1
0
1
2
3
4

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

108
Time (s)

Figure 3.13. Laplacian monocycle in time domain.

1
0.9
0.8
Amplitude

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

0.5

1.5

2.5

3.5

Frequency (Hz)

4.5

109

Figure 3.14. Laplacian monocycle in frequency domain.

3.4.3. Laplacian monocycle

The time-domain representation of the Laplacian monocycle [8], PL (t), is given by
!

 2 "

1
t
1
PL (t) =
2

et/ .

(3.10)

The time and frequency domains of an example Laplacian monocycle are

shown in Figures 3.13 and 3.14 where the monocycle has been normalised for unit
energy. It can be seen from the figure that the frequency domain contains a DC
term, therefore making the Laplacian monocycle impractical for use in impulse
radio applications.

154

Signal processing
4

109

Amplitude

2
1
0
1
2
3
4
1 0.8 0.60.40.2 0

0.2 0.4 0.6 0.8 1

108

Time (s)

Figure 3.15. Cubic monocycle in time domain.

20
18
16
Amplitude

14
12
10
8
6
4
2
0

Frequency (Hz)

10

108

Figure 3.16. Cubic monocycle in frequency domain.

3.4.4. Cubic monocycle

The time-domain representation of the cubic monocycle [8], Pc (t), is
Pc (t; ) =

t 3 t2 /(2)2
e
.
4

(3.11)

The time and frequency domains representations of an example cubic monocycle are shown in Figures 3.15 and 3.16, where the energy is normalised to unity.
The cubic monocycle does not contain a cubic term making it possible to be used
in impulse radio applications.

Ian Oppermann et al.

155
2

108

1.5

Amplitude

1
0.5
0
0.5
1
1.5
2
1 0.8 0.60.4 0.2 0

0.2 0.4 0.6 0.8 1

108
Time (s)

Gaussian pulse
Gaussian monocycle
Gaussian doublet
Figure 3.17. Gaussian waveforms in time domain.

3.4.5. Gaussian pulses

This class of pulses derives its name from the Gauss function that frequently occurs
in statistical theory and filter design. The familiar Gaussian pulse is a member of
this class of monocycles, as are its higher-order derivatives. The derivatives can
be generated by filtering the Gaussian pulse since filtering can be modelled by
the derivative function. Higher-order derivatives can be generated by successive
filtering.
3.4.5.1. The Gaussian pulse
A Gaussian pulse [9] is given by
PRe (t) =

1 (1/2)((t)/)2
e
,
2

(3.12)

where is the centre of the pulse and relates to the width of the pulse. Note that,
like the rectangular pulse, the Gaussian pulse also contains a DC term making it
impractical for use in wireless systems. However, higher derivatives do not contain
such a term and are therefore applicable. The time and frequency domains of an
example Gaussian pulse are shown in Figures 3.17 and 3.18.
3.4.5.2. Gaussian monocycle
The second derivative of the Gaussian pulse is called a Gaussian monocycle and
is often used in impulse radio systems. The time domain representation of the

156

Signal processing

Amplitude

0.8
0.6
0.4
0.2
0
0

4
5
6
7
Frequency (Hz)

10
108

Gaussian pulse
Gaussian monocycle
Gaussian doublet
Figure 3.18. Gaussian waveforms in frequency domain.

Gaussian monocycle, PG (t), is

!

t
1
PG (t) =
1

2 "

e(1/2)((t)/) .

(3.13)

The eective time duration, T p , for the Gaussian monocycle is T p = 7 with

the centre of = 3.5. The frequency spectrum of a Gaussian monocycle, PG ( f ; ),
is given by
2

PG ( f ) = (2 f )2 e(1/2)(2 f ) e j2 f ,

(3.14)

which does not contain a DC term.

The Gaussian monocycle has a single zero crossing in the time domain, whilst
higher-order derivatives have additional zero crossings. Therefore, higher-order
Gaussian pulses yield waveforms with lower relative bandwidths and higher centre
frequencies. The time and frequency domains of an example Gaussian monocycle
are shown in Figures 3.17 and 3.18.
3.4.5.3. Gaussian doublet
The Gaussian doublet [8] consists of two Gaussian pulses (the reversed amplitude
of each other) with a time gap of Tw between the maxima of each. The time domain representation of the Gaussian doublet is given by
PGD (t) =

&
1 % (1/2)((t)/)2
2
e
e(1/2)((tTw )/) .
2

(3.15)

Ian Oppermann et al.

157

Table 3.1. 10 dB bandwidth and the location of the centre frequency for Gaussian signals.
fc

Gaussian pulse
Gaussian doublet
2nd derivative of Gaussian pulse

1/T p
1/T p
1.73/T p

2/T p
2/T p
2.1/T p

3rd derivative of Gaussian pulse

2/T p

2/T p

Pulse waveform

The frequency spectrum of the Gaussian doublet, PGD ( f ), is

PGD ( f ) = 2AGD 2 sin( f Tw )e(1/2)(2 f ) e j[2 f (+0.5Tw )0.5Tw ] .

2

(3.16)

The factor AG is introduced so that the total energy of the pulse is normalised
to unity. The parameters Tw ,, and determine the pulse width. For Tw = 7, the
eective time duration would be T p = 14.
Owing to the similarity of the two basic Gaussian waveforms, the Gaussian
pulse and Gaussian doublet have similar spectral envelops. However, the spectral envelope of the Gaussian doublet has regular nulls. The distance between
nulls in the frequency domain is inversely proportional to the pulse separation
of the doublet (Tw ). Note that the bandwidth of the Gaussian doublet is smaller
than that of the Gaussian pulse and Gaussian monocycle, however its out-ofband energy is larger. The spectrum for all of the waveform is asymmetric about
the centre frequency. The 10 dB bandwidth and the location of the centre frequency for Gaussian signals are given in Table 3.1. Here, the 10 dB
bandwidth is

defined as B10 dB = fH fL , where the magnitude spectrum is 1/ 10 of its peak

value at frequencies fH and fL . The normal centre frequency is defined here as
fc = ( fH + fL )/2. The time and frequency domains of an example Gaussian doublet are shown in Figures 3.17 and 3.18.
3.5. Data modulation
A number of modulation schemes may be used with UWB systems. The potential
modulation schemes include both orthogonal and antipodal schemes.
3.5.1. Pulse amplitude modulation
The classic binary pulse amplitude modulation (PAM) can be presented using two
antipodal Gaussian pulses as shown in Figure 3.19. The transmitted binary baseband pulse amplitude modulated information signal s(t) can be presented as
x(t) = d j wtr (t),

(3.17)

158

Signal processing
0.8
0.6

Amplitude

0.4
0.2
0
0.2
0.4
0.6
0.8

0.2

0.4 0.6

0.8 1 1.2 1.4

Time (ns)

1.6 1.8

Bit 1, Ptype:1, T p = 1 ns
Bit 0, Ptype:1, T p = 1 ns
Figure 3.19. BPAM pulse shapes for 1 and 0 bits.

where wtr (t) represents the UWB pulse waveform, j represents the bit transmitted
(0 or 1), and

1,

j = 0,

1,

j = 1.

dj =

(3.18)

In Figure 3.19, the first derivative of the Gaussian pulse is shown, defined as
t
2
2
wG1 (t) =
e(t /2 ) ,
2 3

(3.19)

where is directly related with the pulse length T p by = T p /2.

3.5.2. On-off keying
The second modulation scheme is binary on-o keying (OOK). Using the following definitions

0,

dj =
1,

j = 0,
j = 1,

(3.20)

the waveform used for this modulation can be defined as in (3.19). The major
dierence between OOK and PAM is that nothing is transmitted in OOK when a
0 bit is chosen.

Ian Oppermann et al.

159

0.8
0.6

Amplitude

0.4
0.2
0
0.2
0.4
0.6
0.8

0.2

0.4 0.6 0.8 1 1.2

Time (ns)

1.4 1.6 1.8

Bit 1, Ptype:1, T p = 1 ns
Bit 0
Figure 3.20. OOK pulses used for 1 and 0 bits.

3.5.3. Pulse position modulation

With pulse position modulation (PPM), the chosen bit to be transmitted influences the position of the UWB pulse. That means that while bit 0 is represented
by a pulse originating at the time instant 0, bit 1 is shifted in time by the amount
of from 0. The signal can be represented as follows:


x(t) = wtr t d j ,

(3.21)

where d j assumes the following values, depending on the bit chosen to be transmitted,

0,

dj =
1,

j = 0,
j = 1,

(3.22)

and the other variables have been defined previously.

The value of may be chosen according to the autocorrelation characteristics
of the pulse. The autocorrelation function can be defined as [10]
(t) =

 +

wtr ()wtr (t + ).

(3.23)

For instance, if we want to implement a standard PPM with orthogonal signals, the optimum value for which we call opt will be the one which satisfies the

160

Signal processing
0.8

0.6

Amplitude

0.4
0.2
0
0.2
0.4
0.6
0.8

0.5

1
1.5
Time (ns)

2.5

Bit 0, Ptype:1, T p = 1 ns
Bit 1, Ptype:1, T p = 1 ns
Figure 3.21. PPM pulse shapes for 1 and 0 bits.

following equation:


opt =

 +

wtr ()wtr opt + = 0.

(3.24)

Figure 3.21 shows a particular case of PPM transmission where data bit 1
is sent delayed by a fractional time interval < 1, and data bit 0 is sent at the
nominal time.
3.5.4. Pulse shape modulation
Pulse position modulation (PPM) and pulse amplitude modulation (PAM) have
been considered for use in UWB impulse radio systems. Pulse shape modulation
(PSM) is an alternative where, instead of dierent pulse positions or pulse amplitudes, dierent pulse shapes are used to represent information bits. This requires a
suitable set of pulses to be used and, like traditional modulation schemes, higherorder modulation is possible by employing a larger pulse set. Modified Hermite
polynomial functions (MHPF) [11], wavelets [2], and prolate spheroidal wave
functions (PSWF) [12] have been proposed in the literature as pulse sets for PSM
systems and all exhibit orthogonal properties. The orthogonality of these pulses
is one of the most desirable properties since it enables optimum detection at the
receiver. Although other orthogonal pulse sets such as coded monocycles can also
be considered, their time and frequency specifications vary for dierent orders
of pulse. MHPF, wavelets, and PSWF wave functions, however, exhibit an almost
constant pulse width and frequency bandwidth with increased pulse order.
The application of orthogonal signal sets also enables alternative multipleaccess techniques to be considered. This can be attained by assigning groups of
orthogonal pulses to each user, where each user then uses the assigned set for pulse

Ian Oppermann et al.

161

shape modulation. The transmissions will then be mutually orthogonal and not
interfering with each other (assuming orthogonality is retained throughout the
link to the detector) and can therefore take place simultaneously and in the same
frequency band.
3.5.4.1. Modified Hermite polynomial functions
MHPF [11] are derived from Hermite polynomials. Hermite polynomials themselves do not possess orthogonality, however this is achieved by modifying them
which leads to the term modified Hermite polynomial functions and are defined
as follows:
hn (t) = et /4 hen (t) = (1)n et
2

2 /4

dn  t2 /2 
e
,
dt n

(3.25)

where n = 0, 1, 2, . . . , and < t < . It can be shown that MHPF satisfy the following dierential equations, where the dots refer to the dierentiation operation
with respect to time:


1 1
h n + n + t 2 hn = 0,
2 4
t
h n + hn = nhn1 ,
2
t
hn+1 = hn h n .
2

(3.26)

The equivalent Fourier transform of the above equations can be written as

follows:


1
H n + 16 2 n + 4 2 f 2 Hn = 0,
2
j8 2 f Hn + j H n = 4nHn1 ,
Hn+1 = j

(3.27)

1
Hn j2 f Hn .
4

The general formula for defining MHPF is given by

hn (t) = kn et

2 /4 2

n!

[n/2]

i=0

where [n/2] denotes the integer part of n/2.

1
2

i

(t/)n2i
,
(n 2i)!i!

(3.28)

162

Signal processing
10

104

8
6
Amplitude

4
2
0
2
4
6
8
5 4 3 2 1

0
1
Time (s)

n=0
n=1

1010

n=2
n=3

Figure 3.22. Modified Hermite wave functions in time domain.

From the above equation, order 0, 1, 2, and 3 MHPF are defined by the following equations:
h0 (t) = k0 et /4 ,
t 2 2
h1 (t) = k0 et /4 ,

! 2
"
t
2
2
h2 (t) = k0
1 et /4 ,

! 3

h3 (t) = k0

(3.29)

"

t t2 /4 2
e
.

The constants kn , where n = 0, 1, determine the energy, En , of the pulse given

by


kn =

En
.
n! 2

(3.30)

Figures 3.22 and 3.23 show the MHPF in the time and frequency domains
for n = 0, 1, 2, 3. The functions are normalised for unit energy. The figure shows
the pulse durations and bandwidths to be the same for all values of n, and they
have a nonzero DC component for n > 0. It is also observed that the number of
zero crossings in the time domain is equal to the pulse order, n. The increasing
centre frequency with pulse order can also be observed and this variable spectral
characteristic makes it dicult to apply to UWB impulse radio systems where the
transmissions should be within a certain spectral mask.

Ian Oppermann et al.

163

1.6

105

1.4

Amplitude

1.2
0
0.8
0.6
0.4
0.2
0

4
6
8
Frequency (Hz)

n=0
n=1

10

12

109

n=2
n=3

Figure 3.23. Modified Hermite wave functions in frequency domain.

A final characteristic of MHPF is the autocorrelation function. The width of

the autocorrelation function helps to ascertain the detection performance when
subject to jitter. The autocorrelation function of MHPF becomes narrower with
increasing pulse order. Thus, higher-order pulses will exhibit a greater sensitivity
to jitter.
3.5.4.2. Wavelets
Reference [2] proposes using wavelets as pulse shapes since they exhibit orthogonality and have the same time duration as the order is increased. Using an orthogonal wavelet, wavelet packets can be generated as follows. Consider 2 filters
of length 2N where h(n) and g(n) are reversed versions of the low and highpass
decomposition filter respectively
(referred to as quadrature mirror filters in filter
bank theory) normalised by 2 and N is the length of the wavelet. The following
sequence functions can then be used to produce a wavelet packet:

W2n (x) = 2

2N
1

h(k)Wn (2x k),

(3.31)

g(k)Wn (2x k),

(3.32)

k=0

W2n+1 (x) = 2

2N
1

k=0

where W0 (x) = (x) is the scaling function and W1 (x) = (x) is the wavelet
function. Compared to other types of wavelets, it is shown in reference [2] that
the Haar wavelet exhibits a high degree of orthogonality as the order is increased

2 3
Time

0.2
0
0.2

2 3
Time

0.2
0
0.2

2 3
Time

0.2
0
0.2

2 3
Time

Amplitude

0.2

Amplitude

Amplitude

Amplitude
Amplitude
Amplitude

0.2

Amplitude

Signal processing
Amplitude

164
0.2
0
0.2

2 3
Time

2 3
Time

2 3
Time

2 3
Time

0.2
0
0.2

0.2
0
0.2

0.2
0
0.2

Figure 3.24. The Haar wavelet [2] in time domain.

which can be seen in Figures 3.24 and 3.25. The spectral characteristics are also
shown in the figure along with the spectral characteristics of a Gaussian monocycle for comparison. The spectral characteristics show a narrow bandwidth for all
orders of packets considered, along with an increased centre frequency as the order is increased. The later characteristic makes it dicult to apply to UWB impulse
radio systems. Compared to other wavelets, it also has favourable autocorrelation
properties and will therefore be more robust to jitter. However, the construction
of this set may be a problem in practice as the Haar wavelets are not regular. Other
wavelets do, however, produce a more desirable frequency spectrum due to their
smoother time-domain characteristics. Examples of other candidate wavelets are
Biorthogonal 6.8 (Bior6.8), Coiflet order 2, Daubechies order 5 (Db5), and Symlet
order 5 wavelets.
3.5.4.3. Prolate spheroidal wave functions
In [12] Prolate spheroidal wave functions (PSWF) are used as the set of pulse
shapes in a UWB impulse radio system. PSWFs, (t), are defined as the solution

Ian Oppermann et al.

165

0.12
0.1

Amplitude

0.08
0.06
0.04
0.02
0

10

12

Frequency (GHz)
Gaussian
1st
2nd

3rd
4th
5th

6th
7th
8th

Figure 3.25. The Haar wavelet [2] in frequency domain.

of the following equation:

 T/2
T/2

(x)

sin (t x)
dx = (t).
(t x)

(3.33)

PSWFs can also be defined as the solution of the following dierential equation:

 d 
d
1 t2
+ c2 t 2 = 0,
dt
dt

(3.34)

where represents the energy of the part of PSWF that lies in [1, 1], denotes
the eigenvalue of (t), , and T are the bandwidth and pulse width of the PSWF,
respectively. The constant c = T/2 represents the number of degrees of freedom.
The functions produced as solutions to the above are almost time and band limited. They also have a double orthogonality property as shown below:


 T/2

T/2

m (t)n (t)dt = mn ,
(3.35)

m (t)n (t)dt = m mn ,

where n (t) represents a PSWF of order n. This is very useful since guarantees
unique demodulation at the receiver and demonstrates one of the strengths of
employing PSWF in impulse UWB impulse radio systems.

166

Signal processing

The computation of PSWF from the above results in a truncated infinite power
series, which can be obtained as follows [12].
The expression below is the generalised PSWF equation:


1/2

2n (c)/T S10n (c, 2t/T)

n (t) = n (, T, t) = 2  
2 31/2 ,
1
1
S
(c,
x)
dx
0n
1

(3.36)

where the denominator acts as a normalisation factor given by

#

1
1

$1/2

2

S10n (c, x) dx

2
.
2n + 1

(3.37)

For large c, almost all the energy of (t) lies in the interval [T/2, T/2]. The prolate
angular function of the first kind is given by

S10n (c, t)

dk (c)Pk (c, t),

for n even,

k=0,2,...

dk (c)Pk (c, t),

(3.38)
for n odd,

k=1,3,...

where Pk (c, t) is a Legendre polynomial. The coecients dk (c) satisfy the recurrence relation


n
k dk+2
(c) + k n (c) dkn (c) + k dkn2 (c) = 0

(3.39)

with
k =


k =

(k + 1)(k + 2)c2
,
(2k + 3)(2k + 5)


2k2 + 2k 1 c2
+ k(k + 1),
(2k 1)(2k + 3)

k =

(3.40)

k(k 1)c2
.
(2k 3)(2k 1)

In order to compute dkn (c), the infinite series is truncated by setting dkn (c) = 0 for
all k > 2N + 1, and the following equation is solved:
Odn = n dn .

(3.41)

Ian Oppermann et al.

167

Amplitude

0.5

0.5

1
1

0.5

Order 2
Order 3

0
Time (ns)

0.5

Order 4
Order 5

Figure 3.26. Prolate spherical wave functions [12] in time domain.

In (3.41), dn are the eigenvectors and n are the eigenvalues of O. Note that two
operators need to be defined to handle the odd and even coecients. The values
of dn defined by (3.41) should be suitably normalised which can be achieved by
using (3.37).
Figures 3.26 and 3.27 show PSWFs for n = 2, 3, 4, 5 in both time and frequency domains. It can be seen that all pulses have a zero DC component and that
the pulse duration and bandwidth do not change with the value of n. Thus PSWFs
are both time and band limited. The centre frequency is also shown to be much
less sensitive to pulse order therefore simplifying the control of these functions
compared to MHPF and wavelets. Additionally, as the order of pulses is increased,
the width of the main lobe of the autocorrelation function stays constant, thus,
unlike MHPF, these functions will be less sensitive to jitter.
3.6. Spectrum randomisation and multiple access
Continuous pulse transmission will lead to strong lines in the spectrum of the
transmitted signal. The regularity of these energy spikes may interfere with other
communication systems over short distances [13]. In order to minimise potential interference from UWB transmissions, a randomising technique is applied to
the transmitted signal. This makes the spectrum of the UWB transmission more
noise-like. The two main approaches to randomise the pulse train in impulse radio systems are time-hopping (TH) and direct-sequence (DS) techniques. A TH
technique randomises the position of the transmitted UWB pulses in time, while
the DS approach uses the technique found in spread spectrum systems to produce
a sequence of pulses with pseudorandom inversions. Neither the TH nor THE DS
techniques leads to further spectrum spreading.

168

Signal processing
30
25

Amplitude (dB)

20
15
10
5
0
5
10

4
6
8
Normalised frequency (GHz)

Order 2
Order 3

10

Order 4
Order 5

Normalised
power (dB)

Figure 3.27. Prolate spherical wave functions [12] in frequency domain.

10
0
10
20
30
40
50

0.5

1.5

2.5

3.5

Frequency (GHz)
Figure 3.28. Spectrum of UWB transmission without randomising techniques.

The spectrum of a modelled UWB transmission with and without a randomising overlay sequence is depicted in Figures 3.28 and 3.29. Figure 3.28 shows
the spectrum of a simple pulse train. The spectrum contains strong spectral lines
at multiples of the pulse repetition frequency. The envelope of the spectrum is that
of a single UWB pulse. Randomising the position in time of the generated pulses
using data modulation and other randomising techniques will aect the spectrum
in such a way that the energy spikes are spread leading to a smoother spectrum.
Figure 3.29 shows the spectrum of a pulse train which includes time-hoppingbased randomisation. A number of randomising techniques may be found in the
literature [14].

Normalised
power (dB)

Ian Oppermann et al.

10
0
10
20
30
40
50

169

0.5

1.5

2.5

3.5

Frequency (GHz)
Figure 3.29. Spectrum of UWB transmission with randomising techniques.

1 1

0 0 1

1 0

10 1

1 00

t
Time-hopping frame

t
Actual transmission time
due to PPM
Nominal transmission time
Figure 3.30. Time-hopping pulse position modulation technique.

3.6.1. TH-UWB
The data modulation is typically based on PPM using TH-UWB as the basis for a
communication system. This approach allows matched filter techniques to be used
in the receiver. The optimum time shift depends on the cross-correlation properties of the pulses used. The concept of TH-PPM is presented in Figure 3.30 where
only one monocycle per data symbol is used (no processing gain is achieved).
Figure 3.30 shows the construction of a single bit for TH-UWB systems.
When a PR code is used to determine the transmission time within a large
time frame, the spectra of the transmitted pulses become much more noise-like.
The time hopping randomises the signal in both time and frequency domains [14].
Pseudorandom time hopping also minimises collisions between users in multipleaccess systems, where each user has a distinct pulse shift pattern [15].
However, a consequence of the PR time modulation is that the receiver needs
accurate knowledge of the PR code phase for each user. One can imagine impulse
radio systems as time-hopping spread spectrum systems. UWB waveforms are
generated without any additional spreading. This simplifies the transceivers relative to conventional SS transceivers [16]. The data rate of the transmission can be
selected by modifying the number of pulses used to carry a single data bit. This in
turn has an eect on the process gain.

170

Signal processing

The pulse repetition time (or frame time) typically ranges from a hundred to
a thousand times the pulse (monocycle) width. The symbol rate can be defined
in terms of the number of monocycles used to modulate one data symbol in fixed
frame time as [15]
Rs =

1
1
=
[Hz],
Ts
Ns T f

(3.42)

where Rs = symbol rate, Ts = symbol time, T f = time-hopping frame, and Ns =

number of monocycles/data bit.
If the data rate in (3.42) is reduced whilst the time-hopping frame remains
constant, the number of monocycles per data bit is increased. This leads to an
increased processing gain.
3.6.1.1. Data modulation with time hopping
In TH mode, the pulse transmission instant is defined by the pseudorandom code.
One data bit is spread over the multiple pulses to achieve a processing gain due to
the pulse repetition. The processing gain is also increased by the low-transmission
duty cycle.
The TH spreading approach has been studied for PAM, PPM, and PSM. However, OOK cannot take the advantage of TH spreading because of the blank transmission in case of bit 0, and because it would create further problems for synchronisation. The information signal s(t) for the mth user can be analytically described for PAM modulation as

s(m) (t) =

w t kTd jT f cw

k= j =1

(m)
j

Tc dk(m) ,

(3.43)

for PPM modulation as follows:

s(m) (t) =

w t kTd jT f cw

k= j =1

(m)
j

Tc dk(m) ,

(3.44)

and for PSM modulation as follows:

s(m) (t) =

k= j =1

wd(m) t kTd jT f cw
k

(m)
j

Tc .

(3.45)

Figures 3.31, 3.32, and 3.33 show a single data bit for PAM, PPM, and PSM,

Amplitude

Ian Oppermann et al.

171

5
4
3
2
1
0
1
2
2
1
0
1
2
3
4
5

10

15

20 25 30
Time (ns)

35

40

45

50

Databit 1, TH-BPAM, Ptype:2, T p = 1 ns

Databit 0, TH-BPAM, Ptype:2, T p = 1 ns
Figure 3.31. Transmitted data bit for BPAM modulation with TH spreading.

5
4

Amplitude

3
2
1
0
1
2

10

15

20 25 30
Time (ns)

35

40

45

50

TH-PPM, Ptype:2, T p = 1 ns
Figure 3.32. Transmitted data bit for PPM modulation with TH spreading.

respectively. Note that in the case of PAM and PPM (Figures 3.31 and 3.32) the
same pulse shape has been used to represent either data bit 0 or 1, while in the
PSM case (Figure 3.33), two dierent pulse shapes have been considered. The TH
time shift and the PPM time shift combine to determine the transmission instant
of the pulse in the case of PPM. The pulse repetition frame is assumed to be 10 ns
long in all of these figures.

172

Signal processing
4
3
2
Amplitude

1
0
1
2
3
4
5

10

15

20 25 30
Time (ns)

35

40

45

50

Databit 0, TH-PSM, Ptype:1, T p = 1 ns

Databit 1, TH-PSM, Ptype:2, T p = 1 ns
Figure 3.33. Transmitted data bit for PSM modulation with TH spreading.

DS-UWB

1 bit (Td = N5 Tc )
Data bit

Spreading code in DS-SS system

Figure 3.34. Direct-sequence signal transmission.

3.6.2. Direct-sequence UWB

Much like the approach taken with spread spectrum techniques [13], DS-UWB
employs sequences of UWB pulses (analogous to chips). A user in the system is
assigned a pseudorandom sequence which controls pseudorandom inversions of
the UWB pulse train. A data bit is then used to modulate this sequence of UWB
pulses.
In the case of UWB systems, the pulse waveform takes the role of the chip in
a spread spectrum system. The DS spreading approach is suitable for PAM, OOK,
and PSM modulation schemes. PPM modulation is intrinsically a time-hopping
technique since the bit value is given by the position of the pulse in a transmission
slot. Figure 3.34 shows the bit structure for a DS-UWB signal. The rectangular
waveform is used to indicate the individual sequence elements.
The PAM and OOK data signal s(t) for user m is given by
s(m) (t) =

k= j =1

 (m)

w t kTd jTc c p

dk(m) ,

(3.46)

Ian Oppermann et al.

173

1.5
1

Amplitude

0.5
0
0.5
1
1.5

5
6
Time (ns)

10

Data bit 1, DS-BPAM, Ptype:2, T p = 1 ns

Data bit 0, DS-BPAM, Ptype:2, T p = 1 ns
Random code
Figure 3.35. Time window of a transmitted data bit for BPAM modulation with DS spreading.

where dk is the kth data bit, (c p ) j is the jth chip of the PR code, (cw ) j is the jth
code phase defined by the PR code, w(t) is the pulse waveform, T f is the pulse
repetition interval, Tc is the chip length, and N is the number of pulses per data
bit.
The pseudorandom sequence has values {1, +1} and the bit length is Td =
NTc .
3.6.2.1. Data modulation with DS-UWB
Figure 3.35 shows a single data bit with binary PAM when a data bit 1 or 0
is transmitted. The square wave represents the pseudorandom sequence, which
aects the polarity of individual pulses which make up the DS waveform. For clarity, only a small part of the data bit waveform is shown. Pulse type (Ptype in the
legend) describes the pulse waveform used. Ptype2 is the second derivative of the
Gaussian pulse.
Figure 3.36 shows a single data bit for OOK modulation when data bit 1
or 0 is transmitted. As mentioned above, in the case of data bit 0, the OOK
spreading signal is characterised by no transmission.
For PSM modulation, the information signal s(t) for the mth user can be presented as
s(m) (t) =

k= j =1

 (m) 

wdk (m) t kTd jTc c p

(3.47)

174

Signal processing
1.5

Amplitude

1
0.5
0
0.5
1
1.5

4
5
6
Time (ns)

10

Data bit 1, DS-OOK, Ptype:2, T p = 1 ns

Data bit 0, DS-OOK
Random code
Figure 3.36. Time window of a transmitted data bit for OOK modulation with DS spreading.

1.5

Amplitude

1
0.5
0
0.5
1
1.5

4
5
6
Time (ns)

10

Data bit 0, DS-PSM, Ptype:1, T p = 1 ns

Data bit 1, DS-PSM, Ptype:2, T p = 1 ns
Random code for data bit 0
Figure 3.37. Time window of a transmitted data bit for PSM modulation with DS spreading.

where the chosen bit dk for the mth user, defined according to (3.18), determines
the choice of the UWB pulse waveform to be transmitted.
Figure 3.37 shows a single data bit using PSM modulation when data bit 1
or 0 is transmitted. Ptype1 and Ptype2 in the legend represent the first and the
second derivatives of the Gaussian pulse.

Ian Oppermann et al.

175

3.6.3. Coexistence and pulse shaping

The problem of choosing a pulse type that will not cause undue interference to
other spectrum users is important. The impact that a particular pulse has on other
spectrum users will primarily be determined by the amount of energy the pulse has
in the pass-band of the victim receiver. Thus by considering the frequency domain
characteristics of the candidate pulse, a suitable choice can be made. However,
this approach may not be appropriate when the UWB transmitter is operating
in close proximity of the victim receiver since energy may be coupled into the
receiver by means other than through the receiver front end. Thus, the full spectral
characteristics of the pulse should be compared to that of the frequency plan of the
victim receiver. Some overlapping narrow bands causing serious interferences may
be deliberately avoided, such as 5 GHz in IEEE 802.11a.
In [17], the coexistence of two candidate UWB systems, that is, time-hopped
spread spectrum with pulse amplitude modulation (TH-SS PAM) and directsequence spread spectrum with pulse amplitude modulation (DS-SS PAM) with
GSM900, UMTS/WCDMA, and GPS, was investigated through simulation. The
study shows that the proper selection of a UWB pulse type and related pulse width
significantly impacts the degree of interference caused to the other spectrum users.
It was shown that for short pulses (which is the case for high data rate transmissions), higher-order Gaussian waveforms cause the least interference, whilst for
longer pulses (lower data rates), the basic Gaussian monocycle causes the least interference.
The interference environment is dynamic due to user mobility and usage.
Thus, it would be advantageous for a UWB system to dynamically adjust its pulse
shape according to the prevailing interference environment. Such flexibility would
also enable the system to adjust to regional variations in regulation that may be
demanded. In order to operate successfully, such a system would be required to
sense the interference environment and adjust the pulse shape accordingly, and
be programmable with regional regulatory variations as required. An example of
such an approach is outlined in [18].
3.7. Synchronisation
3.7.1. Basics
A survey of publications relating to acquisition in direct-sequence spread spectrum systems is given in [19]. Assuming an AWGN channel, the input signal for
the receiver in one user case is given by
r(t) =

Al s(1) t l + n(t),

(3.48)

l=1

where Al is the amplitude of radio channel path l, l is the delay of radio channel path l, L is the number of resolvable multipath components, and n(t) is the
additive white Gaussian noise component.

176

Signal processing

The path delay term l can be represented as a portion of the chip period, Tc ,
given as l = Tc . The signal waveform s(t) is dependent on the system concept,
that is, TH or DS. In both cases, the signal waveform includes the eect of the
spreading code.
The synchronisation stage must provide an estimate e for the timing oset ,
that is, e . The maximum likelihood (ML) algorithm generates several values e to
evaluate when the conditional probability density function of the received signal
p(r | ) achieves the maximum value [20]. So,


e = arg max p r | e .

(3.49)

As we assume white Gaussian noise channel, the conditional probability density

function is readily evaluated. For the optimal estimation e , we have to maximise
 

e =
=

 T0
0

 T0
0

r(t)s t, e dt
Al

w t kTb jT f

k=0 j =1

N



w t kTb jT f e dt + nc (t),

k=0 j =1

(3.50)
where nc (t) is the additive noise component, and the information signal is in the
form of TH-UWB described earlier. In practice, the algorithm is too complex to
implement, and some approximations must be used. In the so-called serial search
strategy, the phase of the local code e is changed step by step in equal increments.
In this way, the correct position within the uncertainty region can be found [20].
Serial search acquisition can be implemented using active correlation measurement technique or passive correlation measurement based on matched filtering
(MF).
3.7.1.1. Synchronisation schemes
UWB systems use very large bandwidths, and therefore a dense channel multipath
profile with many components can be distinguished from the received signal. The
multipath channel then introduces more than one correct synchronisation cell.
From the perspective of code synchronisation, this phenomenon causes problems:
the energy of the signal is spread over many multipath components, and the energy
of each path is very low. Therefore, the paths are dicult to acquire. Depending
on the receiver structure (type and length of rake), a number of paths should be
acquired.
In DS systems, the uncertainty region corresponds to a multiple of the code
length. In TH systems, the uncertainty region is divided into a number of cells
(C). The number of cells depends on the number of possible pulse positions combinations in a bit interval. Recently, an algorithm called chip level post detection

Ian Oppermann et al.

n(mTc )

nTc
MF

Sync

r(t)

177

CLPDI(m)

Th
Figure 3.38. MF with CLPDI.

integration (CLPDI) was proposed for code acquisition in direct-sequence CDMA

systems [21]. This algorithm is suitable for synchronisation in multipath environments and will be used as a general technique for synchronisation. Other simplified schemes may be adopted from CLPDI.
Once the CLPDI algorithm finds one of the possible synchronisation cells, an
additional sweep has to be performed to acquire the necessary number of paths.
The aim of this initial code acquisition in multipath channel is to find a starting
point to reduce the multipath search time. In this chapter, the method is applied
for time-hopping UWB system due to nature of UWB signals.
Figure 3.38 presents the synchronisation algorithm for a multipath environment proposed for DS-CDMA [21]. In DS systems, the impulse response of the
matched filter is the time-reversed replica of the spreading code. This means that
the impulse response has coecients given by the spreading code, and the delays
between the consecutive coecients are Tc . The MF output signal is proportional
to the autocorrelation function (ACF) of the spreading code. The sampling at the
output of the MF is made at least at the chip rate. The MF is followed by a threshold comparison, and if the threshold Th is crossed, the acquisition process ends.
There is detection if the threshold is crossed by the ACF at the zero delay. This occurs with probability of detection, Pd . If the threshold is crossed with some other
delay, a false alarm occurs. This happens with the probability of false alarm, Pfa .
False alarms may be catastrophic, and cause total miss of the correct code phase.
Because the MF gives its peak when the code is inside the filter, a multipath channel
leads to several peaks.
Chip level post detection integration performs post detection integration at
the chip level, that is, a number of consecutive samples at the output of the MF
are combined. CLPDI is performed over m samples and CLPDI output is sampled
at multiples of mT c as is presented in Figure 3.38. Because the sampling is done as
multiples of mT c, consecutive samples at the output of CLPDI are uncorrelated.
In addition, the uncertainty region, that is, the number of cells to be tested in the
acquisition, is decreased relative to the pure MF acquisition. The number of cells
in the uncertainty region is now reduced to Cm = C/m. In addition, the number of
potentially correct cells to be tested is reduced, as can be seen comparing Figures
3.39 and 3.40.
In time-hopping systems, the MF collects the samples together according to
delays between consecutive chips (pulses as presented in this chapter), that is, the
spreading code sets the delays between consecutive 1s in the impulse response.

178

Signal processing
600
500
400
300
200
100
0

50 100 150 200 250 300 350 400 450

Figure 3.39. MF output without noise as function of time in channel model 1, the number of pulses
in bit is N = 10 and the number of pulse position in frame is M = 16.

1400
1200
1000
800
600
400
200
0

50 100 150 200 250 300 350 400 450

Figure 3.40. CLPDI output without noise as function of time in channel model 1, N = 10, M = 16,
and m = 4.

Therefore, the MF waits until the chips (pulses) arrive in predetermined timeslots
inside frames.
3.7.2. Performance measures
The specifications of satisfactory performance measures for synchronisation will
depend on the particular application. The main issue is the time that elapses between the time the synchronisation starts and the time of acquisition. There are
two basic scenarioswhen there is an absolute time limit and when there is no
absolute time limit. If there is no time limit, the most interesting parameter is the
mean acquisition time, TMA = E{Tacq } and possibly the variance of the acquisition time T2 acq . This is the case when a data or pilot signal is always present, that
is, when the link operates continuously. The mean acquisition time is defined as
the expected value of the time which elapses between the initiation of acquisition
and the completion of acquisition. If there is a time limit, the parameters are the
acquisition time Tacq and the time limit Ts . If there is a time limit, a better performance measure is the probability of prompt acquisition Pr{Tacq Ts}, which is

Ian Oppermann et al.

179

called overall probability of detection Pdov . The complement of the probability of

ov = 1 P ov .
prompt acquisition is the overall probability of missing the code Pm
d
Missing the code occurs in systems where data transmission starts after a certain
time interval Ts from the initial system startup, that is, burst communication.
The acquisition time and, therefore, also TMA and Pdov are functions of several parameters. The most important of these are the probability of detection Pd
and the probability of false alarm Pfa . The probability of detection is the probability that the decision is the correct cell when the correct cell of the uncertainty
region is being tested. The probability of false alarm is the probability that the decision is the correct cell when a false cell is being tested. If the decision is based
on threshold comparison, threshold Th plays an important role concerning Tacq
because both Pd and Pfa are functions of this threshold. The other parameters are
the time spent evaluating one decision variable Ti , the penalty time Tfa , and the
number of cells in the uncertainty region C.
In this context, mean acquisition time is used as a performance measure. It
can be calculated for the DS CLPDI case as [21]


TMA =

L/m
PM
LTc + (N L) Tc + T fa P fa
L/m
1 PM

 +L/m1

NTc + (N L) Tc + Tfa Pfa

+
+ 1 i
N L/m
PM /m
i=0




i=0

i
iPM

+ mTc

(3.51)

(N L)(N L + m) Tc + Tfa Pfa

+ NTc ,
2N

where the uncertainty region N is the length of the code, no a priori information of
the correct code phase exists at the beginning of acquisition, Pfa is the probability
of false alarm at the output of the CLPDI, PM = 1 Pd where Pd is the probability
of detection in the correct code phase, Tfa = K p Tc is the penalty time caused by a
false alarm, and Tc is chip interval.
In a multipath propagation environment without the CLPDI block, the TMA
is evaluated with a similar equation [22].
3.7.2.1. Performance of CLPDI
Figure 3.41 presents the TH UWB simulation model with matched filter synchronisation. The figure also describes the outputs from dierent parts of the system.
Chip correlation (point A in Figure 3.41) is performed by correlating the received signal with the template waveform which has the same shape as the received
signal (after the antenna).
The pulse waveform in the transmitter in the simulations is an ideal Gaussian
pulse. The reference pulse in the receiver is therefore the second derivative of the

180

Signal processing
Code CRR

Sqn com

Tr
bSTX

Sig rx

Tc

Chip
correlator

Code
MF

C
Filtro

awn

Code
synchronisation

Ctemplate0

Th

Uindex slot(1)

Probability
function

Length of data bit in Tc

Pd t Pd Pfa t Pfa

41

Figure 3.41. Simulation model for TH-UWB synchroniser in an AWGN channel.

Code CRR

Th

(1q)

21

NTc slots
Synchronisation point
Figure 3.42. Output of the code matched filter when six bits are transmitted.

Gaussian pulse because of the eect of the transmitting and receiving antenna [22].
Only one sample per Tc is used to reduce the complexity in the simulations. This
is appropriate when the pulse width (T p ) is smaller than the chip length (Tc ), and
when chip synchronism is assumed. The output of the chip correlator is a sequence
that contains the information of the pulse locations within the received sequence
in every frame. The chip correlator output is fed into a code-matched filter (CMF).
The filter is matched to the whole spreading code (point B in Figure 3.41). In the
TH system this means that the delay between consecutive 1s in the impulse response is not fixed, but depends on the code phase as seen in Figure 3.42. After the
filter, a threshold comparison is used (point C in Figure 3.41). The threshold setting is dependent on the desired false alarm rate. Figure 3.42 presents an example
of the output of code-matched filter, as well as describing the correct synchronisation points.
In the CLPDI simulations, the CLPDI block is inserted after the code matched
filter in the block diagram of the TH-UWB synchroniser (see Figure 3.41).

Ian Oppermann et al.

181

3.7.2.2. Performance in Saleh-Valenzuela channels

There are a large number of resolvable paths in UWB systems, for example, more
than 100, because of the fine path resolution of the UWB channel. The following
simulation results only consider the most powerful paths for the code acquisition.
The number of cells or chip positions on which the receiver can be considered
synchronised is defined by the 20 strongest paths in the MF case. When m-CLPDI
is used, we need to make sure that the 20 synchronisation positions defined in
the MF case are included in the synchronisation cells for the m-CLPDI algorithm.
Since it is possible that the 20 synchronisation cells are nonconsecutive, the number of synchronisation cells in the m-CLPDI case can be bigger than 20/m. Once
one of the synchronisation cells is found, a new acquisition process starts. Verification extra time is added for each false alarm event occurred during the acquisition
process.
The simulations consider four modified Saleh-Valenzuela channel models
used by the IEEE 802.15.3 working group to measure performance. The pulse
width T p is fixed to 1 ns. The code length is 160, the number of frames per bit
is N = 10, the number of possible pulse positions in each frame is M = 16, the
penalty time in the case of false alarm is Tfa = 100Tb , and the required Pfa = 0.01.
The synchronisation process is performed over a data packet of 30 bits and the
channel is static over each packet. A new data block is transmitted every time the
synchronisation is reached or the data packet has been explored without synchronisation.
Figures 3.43, 3.44, 3.45, and 3.46 present the mean acquisition time of the
code MF for m = 4, 10, and 20 and the CLPDI block versus SNR for each of the
Saleh-Valenzuela channel models. SNR is defined as the Eb /N0 at the receiver. Eb is
the collective bit energy spread over all the paths in the channel. Using CLPDI, the
reduced code acquisition time is evident for all the channel models. The mean acquisition time increases in absolute time from both MF and CLPDI passing from
the IEEE 802.15.3a channel model 1 (CM1) to the CM4. The reason is the reduced
energy borne by the 20 strongest paths whilst moving from CM1 towards CM4.
Otherwise, passing from the CM1 to the CM4 relatively increases CLPDIs performance improvement relative to the MF case. As the channel moves from the CM1
to the CM4, the channel gets closer to the situation of having paths with equal average power that gives the best result, as has been shown for data detection using
diversity techniques [13]. In CM1, the extra gain achieved using m = 20 is not
remarkable. This is because of the reduced CM1 delay spread relative to the other
channel models.
3.8. Impulse radio demonstrator for 4-PPM
This section discusses an ultra-wideband radio testbed based on pulse position
modulation (PPM) for investigating the properties of short-range data communication. PPM was selected because of its low duty cycle which translates to high
power eciency [23]. The testbed is designed to realise data transmission at
6 Msymb/s over a distance of a few meters in indoor oce environments. The

182

Signal processing
Mean acquisition time, Pfa = 0.01, CM = 1
500
450

TMA /Tb

400
350
300
250
200
150
100
50
0

MF
CLPDI4

10 11 12 13
Eb /N0 (dB)

14 15 16

CLPDI10
CLPDI20

Figure 3.43. TMA /Tb as a function of Eb /N0 in channel model 1.

Mean acquisition time, Pfa = 0.01, CM= 2
500
450

TMA /Tb

400
350
300
250
200
150
100
50
0

MF
CLPDI4

10 11 12 13
Eb /N0 (dB)

14 15 16

CLPDI10
CLPDI20

Figure 3.44. TMA /Tb as a function of Eb /N0 in channel model 2.

focus of this eort is on the implementation of commercial-grade microwave circuitry and algorithms for ultra-wideband data transmission, especially concerning
mobile battery-driven devices. The testbed hardware is described in detail as well
as the two-stage approach used for receiver synchronisation.
The pulse repetition frequency is set to 60 MHz and each PPM symbol is repeated ten times. This gives a symbol rate of 6 Msymb/s. Each PPM symbol carries
two bits by means of four possible delays at {0, 2 ns, 4 ns, 6 ns}. This design results
in an uncoded bit rate of 12 Mb/s.

Ian Oppermann et al.

183
Mean acquisition time, Pfa = 0.01, CM= 3

500
450
400

TMA /Tb

350
300
250
200
150
100
50
0

MF
CLPDI4

10 11 12 13
Eb /N0 (dB)

14 15 16

CLPDI10
CLPDI20

Figure 3.45. TMA /Tb as a function of Eb /N0 in channel model 3.

Mean time code acquisition, Pfa = 0.01, CM = 4

500
450
400

TMA /Tb

350
300
250
200
150
100
50
0

MF
CLPDI4

10 11 12 13
Eb /N0 (dB)

14 15 16

CLPDI10
CLPDI20

Figure 3.46. TMA /Tb as a function of Eb /N0 in channel model 4.

In an initialisation step, the receiver estimates the pulse repetition frequency

of the transmitter and sets the local voltage-controlled crystal oscillator (VCXO) to
this estimate. Subsequently, an early-late tracker controls an eight-bit programmable delay shifter with a granularity of 65.1 ps and the local VCXO. The
whole synchronisation algorithm is implemented on a low-cost microcontroller
(C) running at a clock of 15 MHz.

184

Signal processing

Various measurement results obtained from the testbed are discussed and
compared to simulations in Matlab. We draw some early conclusions from this
work and give some recommendations for future investigations.
3.8.1. Motivation
The pioneering UWB industry has demonstrated that the basic physical layer technology works, but many system-level and operational issues remain that need to
be investigated before UWB technology can be integrated into commercial devices
[24].
A trial system was designed in hardware which employs a variant of pulse position modulation (PPM) for data transmission. The main purpose of the testbed
is a feasibility test for short-range data transmission using commercial low-power
components, to investigate the behaviour of the microwave circuitry, and to develop new algorithms for the ultra-wideband approach to data transmission.
Currently, there exists just a very small number of methods for generating signals having ultra-wide bandwidth: for example, chirp modulation, direct spreading by high-rate pseudorandom noise sequences [12, 25], fast frequency hopping
[26, 27], and transmission via short pulses [2831]. The latter option was chosen
because of its simplicity in circuitry and its inherently low power consumption. It
was recently found that such an orthogonal 4-ary PPM outperforms binary oset
PPM at all jitter levels in thermal and impulsive noise [32]. For the testbed, a transmission format was chosen which includes a 5th timeslot [31]. The extra timeslot
is inserted for aiding in synchronisation and tracking, compare Section 3.8.5.
The testbed is implemented using o-the-shelf electronic components only.
The testbed consists of a transmitter part and a receiver part. The data transmission is unidirectional. The testbed can be upgraded for duplex transmission and
extended to multiple access in a small network in the future, compare [29].
3.8.2. Transmitter
The testbed uses a PPM format with 4 timeslots of equal duration (2 ns each) and
an additional guard interval of longer duration (approx. 8.67 ns) resulting in a
pulse repetition frequency (PRF) of (4 2 ns+8.67 ns)1 = 60 MHz, cf. Figure 3.48.
Adopting the naming convention of [23], this modulation format can be called a
combination of x-frame PPM and x-pulse PPM. The transmitted PPM signal
s(t) can be modelled as
s(t) =

p t Ak/10 T kTPRF (t) ,

(3.52)

k=

where p(t) denotes the designed pulse shape, T = 2 ns is the timeslot duration,
and fPRF = 1/TPRF = 60 MHz is the PRF. The PPM symbols are denoted as Ak
{0, 1, 2, 3} and each symbol is repeated ten times, see (3.52). The slowly varying
delay (t) describes fluctuations in the PRF and clock jitter.

Ian Oppermann et al.

185

The transmitter design is shown in Figure 3.47a. It consists of a clock oscillator which defines the PRF of 60 MHz and generates a stable timebase for
the pulse positions within each symbol frame. A divider (denoted by (: 10) in
Figure 3.47) defines the number of repeated pulses sent per PPM symbol. A data
generator produces a pseudorandom noise (PRN) bit stream (binary symmetric
source) for measurement purposes. The data bits are mutually uncorrelated and
equi-probable which ensures that the pulse positions are uniformly distributed
within the 4 timeslots used for data transmission. The delay lines define the pulse
positions within a symbol. A symbol switch selects the pulse position to be transmitted during each frame. Finally, a pulse shaping device creates the waveform.
The pulse shaping is implemented by two high-frequency transistors (BFP 540F)
which shape the rising edge of the delayed 60 MHz signal to the required narrow
UWB pulse shape. A discone antenna designed for a center frequency of 5.8 GHz
emits the pulses.
In the following, we discuss the transmitter components and measurements
in detail.
Clock generator. The clock generator is a crystal oscillator that needs to be well
decoupled from the noise emitted by other parts of the transmitter. The overall
performance (bit error rate) of PPM is sensitive to the clock jitter variance. To
illustrate this, we show the eect of a small ripple (in the range of millivolts) in
the power supply of the crystal oscillator which can cause significant clock jitter.
Figure 3.49 shows oscillograms of the observed signal at the receiver side with and
without the supply voltage ripple.
Frequency divider. The divider is set to 10 and every PPM symbol is repeated ten
times. This enables the receiver to average received symbols over ten realisations.
This can be interpreted as a spreading gain of 10. In a future testbed enhancement,
this simple repetition code can be replaced by a time-hopping spreading sequence
to identify the user of interest in multiple-access scenarios.
Pseudorandom noise data generator. For the experiments, a two-bit wide pseudorandom noise (PRN) data generator is implemented. The data are generated for
measuring the link performance by counting bit errors for bit error rate (BER)
measurements. In a future implementation step, this data generator will be replaced by an external source providing, for example, encoded multimedia data
streams.
Delay lines. The demonstrator uses an integrated circuit 3D3215 (Data Delay Devices Inc.). This component implements five equally spaced taps of T = 2 ns delay
each. In the testbed, it is driven at almost its maximum usable frequency and some
delay tolerances have been observed. Delay variations are also due to dierences
in the capacitive loads due to the circuit layout. These load dierences have been
eliminated by introducing small parallel capacitors.
Symbol switch. The symbol switch consists of a symbol latch and a 4 to 1 multiplexer. The two data bits are latched first and then control the address of the

186

Signal processing
60 MHz
clock Digital delay IC. 3D3215, a T = 2 ns
T

T
Symbol switch

: 10

PRN
Data generator

RF
Discone
2 x antenna
BFP540F
CPLD XC2C32
(a)

(b)
Figure 3.47. (a) Block diagram of the transmitter; (b) Realisation.

multiplexer. The outputs of the 3D3215 delay taps are connected to the data inputs
of the multiplexer. It proves to be important to switch among the inputs when all
input signals have high state. If this condition is violated, an additional rising edge
is generated resulting in spurious pulses.
Figure 3.48 shows the measured 4-ary symbol pulses after the first pulse
shaper transistor stage. In this figure, we also observe a small spurious pulse. This
is caused by switching the multiplexer from an input with a low signal to one
with a high signal. In the testbed, these spurious pulses occur only once every
10TPRF 166.7 ns when the transmitter modulates a new pair of data bits which
is distinct from the previous bits.
The digital part of the PPM transmitter is integrated into a complex programmable logic device (CPLD, CoolRunner II, XILINX Inc.) which generates
very fast rising edges. The subsequent pulse shaper steepens the rising edges using
a high-pass filter.
Pulse shaper. For the pulse shaper, two stages of high-frequency transistors (BFP
540F) with grounded emitters are used. Every stage has a highpass RC filter at
its input and a shorted microstrip line at the output for shortening the amplified
pulse. The shorted microstrip line in the first stage is much longer than in the

Ian Oppermann et al.

Tek Stopped

187
1575170 Acqs

30 Nov 0415 : 23 : 01

TPRF 16.667 ns
T = 2 ns

Ch2 500 mV

M 2.0 ns 25.0 GS/s ET 40.0 ps/pt

A Ch1 / 1.76 V
(a)

Market 1 [T1]
Ref Lv1 34.38 dBm
0 dBm 6.97995992 GHz

RBW
VBW
SWT

3 MHz
3 MHz
58 ms

RF Att

10 dB

Unit

dBm

10
20
30

1 AP

40
50
60
70
80
90
100

Start 0 Hz
Date : 09.Dec.2004

1 GHz
14 : 35 : 19

Stop 10 Hz

(b)
Figure 3.48. (a) Modulated symbol pulses after the first pulse shaper stage. (b) Spectral power density
of the PPM signal.

second otherwise the amplitude of the pulse would decrease significantly. The collector resistance is chosen to produce a low output resistance and to allow a high
PRF. The UWB transmitter generates pulse peak voltages of nearly 3 V at 50
with a pulse width well below 100 ps (measured at half amplitude).
Antenna. Traditional broadband antennas are designed with multiple resonant
structures which cause excessive ringing and group delay distortion (e.g., logperiodic Yagi-Uda antenna). Several antenna types were designed and evaluated

188

Signal processing
Stopped

Ch1

39271 Acqs

200 mV Bw

19 Oct 04 08 : 42 : 06

M 10.0 s 125 MS/s 8.0 ns/pt

A Ch1 / 432 mV
(a)

Stopped

Ch1

207978 Acqs

200 mV Bw

19 Oct 04 09 : 06 : 26
t1 : 37.46 s
t2 : 13.62 ns
t : 37.47 s
1/t : 26.69 kHz

M 10.0 s 125 MS/s 8.0 ns/pt

A Ch1 / 376 mV
(b)

Figure 3.49. Influence of transmitter clock jitter on the received signal: (a) with; (b) without jitter.
This is an oscillogram with vertical histogram function.

by measurements to find a practical UWB antenna solution. Such an antenna solution should result in a narrow pulse response without ringing or introducing additional echoes. Such echoes occur due to imperfect matching between the transmitter output and the antenna input. Antenna mismatch causes part of the pulse
power to be reflected repeatedly between antenna connector and the second transmitters collector. A 6 dB attenuator is included between the second transistor and
the antenna connector for reducing such echoes.
UWB impulse radio transmitters need to be placed near (or even inside) the
antenna to cope with mismatch problems and the resulting echoes.

Ian Oppermann et al.

189
2nd LNA

Sampler f g = 10 MHz

r0 (t) V AGC

r1 (t)

V AGC

V AGC

V AGC

d Amplifer +
demapper

Wilk.
divider

Discone
antenna

LNA

Rectifier

r2 (t)

r3 (t)

r4 (t) V AGC

Data
2

e
1

Phase shifted
60 MHz

Digital delay 2ns +/ 100 ps

Figure 3.50. Block diagram of the data path of the receiver.

Power supply. Not shown in the block diagram is the power supply. Its quality
defines the clock precision and the accuracy of the pulse position amplitude. Noise
(Ripple) on the power supply causes clock jitter. Only a low impedance of the
power supply guarantees a constant supply voltage for the pulse shaper. This is a
stringent requirement for generating equally shaped narrow pulses with low jitter.
If switched power supply units are used, it is best to synchronise them. Otherwise mixing resulting from non-linearities causes low frequency noise on the
power supply lines which are hard to filter out.
3.8.3. Receiver
The receiver architecture is shown in Figures 3.50 and 3.51. The antenna is the
same type of discone antenna as that used by the transmitter. The average received
power level is expected to be the transmitted power decreased by approximately
60 dB [33]. Low-noise amplifiers (LNA) are used in the front end for increasing
the signal amplitude and to ensure sucient signal-to-noise ratio for the down
conversion. The LNAs consist of InGaP/GaAs MMIC (NBB-300, RF Micro Devices Inc.). A Wilkinson power divider splits the received signal into five branches
of equal strength. Five sampling mixers sample the input signal at dierent delays. Each of the five branches has one sampling mixer. The fifth branch of the
receiver aids the synchronisation, see Section 3.8.5. The sampling mixers contain
dual Schottky diodes and pulse shapers for generating the narrow LO (local oscillator) pulses. Low-pass filters are used for averaging the pulse energy coherently and

190

Signal processing

filtering out additive interference. Five equal voltage-controlled amplifiers amplify

the signal after down conversion. Five full-wave rectifiers are used for noncoherent
symbol reception. Included in the amplifier and decoder block is a circuit made of
several operational amplifiers realising the mapping of PPM symbols to bit pairs.
Another set of five low-pass filters is used for the synchronisation part of the
receiver (see Figure 3.51). It averages the energy of a high number of symbols and
suppresses interference. A five-channel analogue-to-digital converter (ADC) samples all inputs simultaneously. A C is used for executing the synchronisation algorithm and fulfills several other tasks in the testbed. A clock generator delivers
the timebase of the receiver: it is frequency controlled by the C.
Antenna and first LNA. The LNA performs a filtering together with the antenna as
well as a signal amplification. The gain has been set to a rather high value because
of the high-noise factor of the sampling mixer.
The same echo problem as in the transmitter exists in the receiver between
the antenna and the LNA. Here also an attenuator has been included to reduce the
echo generated in the antenna line. This attenuator could be omitted by a shorter
distance between receiver front end and antenna. This would only be possible by a
higher degree of integration of the front end.
Power divider. The power divider is a Wilkinson type with three stages and five
paths. Because of the high number of paths, the input impedance is chosen to be
25 . That causes the need for an impedance match between the output of the
first LNA stage and the input of the Wilkinson power divider. The transfomer is a
tapered microstrip line several wavelengths long.
The mutual layout coupling is a big part of the disparity of the practical realisation of the five output paths of the power divider. Nevertheless this power
divider was chosen due to its additional isolation between the outputs and lower
insertion loss between the input and each of the outputs. The isolation between
the outputs is required because of the limited LO to RF isolation of the sampling
mixer circuitry. This part will be much easier to realise if it is integrated onto a
chip. Then the isolation of the dierent arms can be realised by buer amplifiers.
Second LNA. The second LNAs together with the Wilkinson power divider help to
isolate the five sampling mixers from each other. Every sampling mixer generates
at its input the so-called kick-back pulses especially if the IF output voltage is nonzero.
Sampling mixer. Each of the five parallel sampling mixers consists mainly of a
pair of fast Schottky diodes (BAT15, Infineon Technologies) and a pulse transformer/shortener made up of a microstrip slotline connection. The slotline ends
in a short on one side and by the diodes and charge capacitors at the other. This
type of sampling mixer is well described in the literature and our design is adapted
from [34, 35]. A photograph of a single sampling mixer is shown in Figure 3.53.
The bank of sampling mixers is not required to be very amplitude linear because the outputs are evaluated relative to each other only. Therefore, the circuitry
for discharging the charge capacitors and biasing can be omitted in principle.

Ian Oppermann et al.

191
C
ATM ega8L

ADC
MAX155

f g = 25 kHz
a
y0 (t)

x0 (t)

y1 (t)

x1 (t)

y2 (t)

x2 (t)

y3 (t)

x3 (t)

y4 (t)

x4 (t)

e
15 MHz
V AGC

DAC
38

F (k)
Phase shifted
60 MHz Fine delay

16 bit
fPRF

:4
8 bit

DAC
16

S (k)
Delay step
3D3428

60 MHz
VCXO

Figure 3.51. Synchronisation and control components of the receiver. The inputs a, b, . . . , e are connected to the corresponding points in Figure 3.50.

The received PPM signal r(t) at the front end of the receiver can be modelled
as a linearly filtered version of (3.52) in additive noise,
r(t) =

q t Ak/10 T kTPRF (t) + n(t),

(3.53)

k=

where q(t) denotes the overall pulse shape, T = 2 ns is the timeslot duration,
fPRF = 1/TPRF = 60 MHz is the PRF, and n(t) denotes the noise. The pulse shape
q(t) results from the linear distortions between the antenna connectors at the
transmitter and the receiver.
The sampling mixers are clocked synchronously at the receivers estimate of
the PRF which is denoted by fPRF = (TPRF )1 .

192

Signal processing
Three-stage five-arm Wilkinson divider

2nd LNAs

Taper
50 Ohm
25 Ohm
1st LNA
with
additional
shielding
Figure 3.52. Front-end PCB of the UWB receiver.

Charge capacitors

Dual diode
Load resistance

Slot with shortened end

Bottom view
LO port
RF port
Air coils (IF port)
Top view

Figure 3.53. Detailed photographic images of the sampling mixer PCB.


The th sampling mixer ( = 0, 1, . . . , 4) is triggered at the delay T +  + (t)

is adjusted by the C using the delay step and fine delay, and | | T
where (t)
describes small constant delay deviations from the PPM symbol delay slots at T,
that is, we designed an early-late sampler with 0 3 100 ps and 1 2
100 ps.
After the sampling mixer in the th signal path of the receiver, we obtain

r (t) =

n=


r(t) t nTPRF T  (t)
,

 = 0, 1, . . . , 4.

(3.54)

The oscillogram shown in Figure 3.54 is measured at the sampling mixer output.
The horizontal scaling is 200 ns/division and the symbol duration is 10TPRF
166.7 ns. In this figure, we see approximately 12 spikes per division. The spikes
mark the LO feedthrough.

Ian Oppermann et al.

193

Tek Stopped single seq 1 Acqs

29 Nov 04 14 : 38 : 40

Ch2

50.0 mV

M 200 ns 10.0 GS/s 100 ps/pt

A Ch2 / 58.0 mV

Figure 3.54. Oscillogram of the sampling mixer output, 10 pulses per symbol, approx. 12 pulses per
200 ns division.

3.8.4. Local oscillator pulse generation

The only dierence between the five receiver paths is the position of the local oscillator pulses which control the diodes in the sampler circuit.
The main receiver clock of 60 MHz is the source of the local oscillator pulses.
The same digital delay line as the one used in the transmitter is also applied in the
receiver. The distance is adjusted to equidistance plus or minus a small fraction.
So two early and two late LO signals are generated. This is used to detect the exact
position in the time frame of the incoming signal. Further details can be found in
Section 3.8.5.
A CPLD generates the rising edges used in the pulse sharpening circuit. In a
later stage, it is foreseen to implement a time hopping sequence generator in the
CPLD for mitigating multiple-access interference.
In contrast to the transmitter, only one transistor stage is necessary in the
pulse shaper of the receiver because the amplitude and impedance requirements
are less stringent than in the transmitter.
Signal low-pass filter. High-input impedance and low-noise operational amplifiers
augment the signal amplitude of the now down-sampled signal. In front of them
LC low-pass filters are included. For low ringing, Bessel characteristic has been
chosen.
The large coils of the filters can pick up noise easily so care must be taken to
ensure proper shielding, especially at this low-amplitude stage.
Low-frequency amplifiers. The variable gain amplifiers (VGA) following the first
low-frequency stage amplify the received pulses to constant amplitude. The lowfrequency amplifier stages consist of three parallel dual channel VGA (AD604,
Analog Devices Inc.). One problem arising from this discrete realisation is to

194

Signal processing
Tek Stopped 187 Acqs

Ch1 500 Bw
mV

29 Nov 04 14 : 47 : 44

M 2.0 ms 1.25 MS/s 800 ns/pt

A Ch1 / 700 mV
(a)

Tek Stopped 343 Acqs

Ch1 500 Bw
mV

29 Nov 04 14 : 48 : 24

M 2.0 ms 1.25 MS/s 800 ns/pt

A Ch1 / 700 mV
(b)

Tek Stopped single seq 1 Acqs

Ch1 200 Bw
mV

29 Nov 04 14 : 49 : 43

M 1.0 ms 50.0 kS/s 20.0 s/pt

A Ch1 / 360 mV
(c)

Figure 3.55. Oscillograms of the receiver baseband signals, not frequency locked. (a) Output of the
VGA, (b) output of the full-wave rectifier, (c) output of the synchronisation low-pass filter.

Ian Oppermann et al.

195

ensure equal gain of all five paths used for every gain setting. This is controlled
by the C by means of an 8 bit DAC channel. An oscillogram of one channel is
given in Figure 3.55a. This kind of oscillogramwith all four transmitted pulses
in one receiver pathcan only be observed when transmitter and receiver work
with a frequency oset.
Full-wave rectifier. Before this stage the signal has both polarities. So depending
on the environment and antenna position, the bigger part of the pulse oscillation
can be either positive or negative. A clamp amplifier (AD8036, Analog Devices
Inc.) operates as a precision full-wave rectifier. It is used to flip all pulses to positive
voltage values (Figure 3.55b).
De-mapper. The PPM de-mapper compares the levels on all four signal paths and
selects the pair of bits that best represents the PPM symbol. It consists of operational amplifiers.
Synchronisation low-pass filter. For synchronisation, all five signal paths are lowpass filtered by 3rd-order RC filters. The cut-o frequency is approximately
25 kHz. This value is a trade-o between averaging over a large number of PPM
symbols and a low group delay. Low group delay is important for the delay tracking range of the synchronisation algorithm. A lower cuto frequency results in a
filter which averages over more PPM symbols. The improved signal to noise ratio
eases the detection of delay osets by the C (Figure 3.55c). For further improvement of the signal to noise ratio, a peak-value rectifier is placed before the input
of the synchronisation low-pass filter.
Analogue-to-digital converter (ADC). The low-pass filtered signal gets injected into
the analogue-to-digital converter. This ADC must sample all five input signals simultaneously. Therefore the MAX155 was selected.1
The ADC is entirely controlled by the C and works with an input clock of
3.75 MHz. It outputs 8- bit samples to be proccessed by the synchronisation algorithm executed by the C.
Microcontroller. The 8-bit C is an ATMega8L (ATMEL Inc.). This C is powerful
enough for delay tracking with update cycles below 60 s. In the testbed, it operates
on a 15 MHz clock derived from the VCXO.
The main task of the algorithm implemented in the C is the delay tracking.
The synchronisation algorithm itself is described and discussed in Section 3.8.5.
Other tasks of the C include hardware initialisation during power-up, oset voltage compensation of the ADCs, initial PRF acquisition (for details see page 61),
automatic gain control by setting the gain in the VGA (see page 56), and debugging support by tracing of key quantities via a serial output port and otherwise
unused DACs.
Digital-to-analogue converter (DAC). An eight-channel DAC is integrated into the
receiver circuitry for three dierent tasks. The first task is to set the gain of the
variable gain amplifier. In this way, the automatic gain control (AGC) loop is
1Can be replaced by the newer and faster MAX1304, MAXIM Inc.

196

Signal processing
Tek Stopped

Ch1 100
mV

10 Acqs

Bw

M 100 ms 500 S/s 2.0 ms/pt

A Ch1 / 708 V

Delay control (8-bit unsigned integer)

(a)

250
200
150
100
50
0
2000

2500

3000

3500

4000

4500

5000

t (samples)
(b)
Figure 3.56. (a) Oscillogram of the DAC output showing the delay shift control trajectory versus time.
(b) Simulation trace of 8-bit quantised delay shifter control values.

closed. The second task is to control the fine delay circuitry shown in Figure 3.51.
In this way, the receivers clock timing can be delayed in steps of TPRF /256. The
third DAC channel is used as a debug output. With that output the 8-bit address
of the phase delay table can be displayed on an oscilloscope easily (Figure 3.56).
The receiver phase is locked if a sawtooth is displayed. The period of the sawtooth gives the period of the oset frequency between transmitter and receiver.
Delay shifter. The receiver synchronisation subsystem uses a digitally controlled
delay shifter, comprising the delay step and fine delay blocks in Figure 3.50. The
delay shifter is controlled by an 8-bit input d {0, 1, . . . , 255} and is designed to

Ian Oppermann et al.

197

realise the delay d with a granularity of = 65.1 ps. This results in delays up to
(255/256)TPRF 16.60 ns.
The controlled delay shifter is implemented in two stages, delay step and fine
delay, respectively. The delay step is based on an 8-bit controllable delay circuit
(3D3428-0.25 from Data Delay Devices Inc.) having delay steps of 0.25 ns. This
granularity is too coarse for timing adjustment of the receiver. Therefore, a second
stage, fine delay, is serially concatenated with a range of delay shifts of approximately 1 ns. The fine delay is controlled by one output of the 8-bit DAC. This
combination realises a very fine delay adjustment in 256 quantised steps with a
stepsize of approximately 65.1 ps.
3.8.5. Receiver synchronisation
The guard interval in the symbol frame (see Figure 3.48) is introduced to aid synchronisation. In this design, it induces approximately 108% excess bandwidth into
the system. As a result, a discrete frequency component at the PRF is introduced
in the spectrum [23, 31]. This component can be extracted to derive the symbol
frame timing.
In (3.53), the PPM symbols Ak {0, 1, 2, 3} are assumed mutually independent, identically, and uniformly distributed, and (t) denotes a slowly varying delay for which the first-order probability density is uniform in [0, TPRF ]. The synchronisation algorithm first estimates fPRF (0) and subsequently tracks (t) and
fPRF (t) for t > 0 by means of an early-late discriminator.
As shown in Figure 3.50, the receiver front-end splits r(t) using the Wilkinson
power divider into five parallel signal paths. The signal on each path is amplified by
a factor of g, sampled at the receivers estimate (TPRF )1 of the PRF (see page 55),
low-pass-filtered with fg1 = 10 MHz and rectified. This low-pass filter serves as an
integrator for summing the energy of 10 consecutive pulses. The resulting y (t) is
passed on to the PPM symbol decoder by selecting the signal path having maximum magnitude. The synchronisation part of the receiver shown in Figure 3.51
passes y (t) through a second low-pass filter with fg2 = 25 kHz. The resulting x (t)
serves as input to the ADC:
 t


y (t) = 
g
x (t) =

 t0
0




hLP1 (t t
)r (t
) dt

,

(3.55)

hLP2 (t t
)y (t
) dt
.

The third-order low-pass filter characterised by hLP2 (t) is designed such that it effectively averages over a large number of pulses (approx. 3000 pulses are averaged).
Finally, the ADC quantises x (kTS ) to 8 bits and delivers the values x [k] to the C.
Pulse repetition frequency acquisition. Let x [k] be the 8-bit quantised ADC input
value of the th ADC at sample instant k where  {0, 1, . . . , 4}, see Figure 3.51.
Note that the ADC input signals are low-pass-filtered with 25 kHz cuto and the
sampling interval is approximatly TS = 58 s. Further, let x() [k] be the sorted ADC

198

Signal processing
100
90
80

J( f )

70
60
50
40
30
20
10
0

600

400 200

200

400

600

f f0
Figure 3.57. Simulated goal function J( f ) realisation versus frequency deviation.

input values at sample instant k such that

0 x(0) [k] x(1) [k] x(4) [k] 255.

(3.56)

The following goal function J( f ) is maximised for coarse frequency estimation for
the initial acquisition of the PRF:
'

n
1 x(1) [k] + x(2) [k] + x(3) [k] + x(4) [k]
J( f ) =
x(0) [k] .
n k=1
4

(3.57)

This goal function estimates the depth of the PPM symbol guard interval as the
dierence between the arithmetic mean of the four largest ADC input values and
the smallest one.
This goal function can be reformulated as

1 
x0 [k] + x1 [k] + x2 [k] + x3 [k] + x4 [k] 5x(0) [k] ,
4n k=1
n

J( f ) =

(3.58)

which shows that we do not need to sort the input ADC values. It suces to find

the index [k]
{0, 1, . . . , 4} of the minimum magnitude ADC value
x(0) [k] = min x [k] = x[k]
 [k]
0 4

(3.59)

for each sample k.

In the testbed implementation, we chose to average over n = 64 ADC samples. The C sweeps through the range of frequencies { fmin , fmin + 1, . . . , fmax }
represented by 16-bit integers. At each frequency f , the goal function (3.58) is
evaluated. In simulations, the properties of this goal function were evaluated numerically. A typical realisation of J( f ) obtained from simulations is shown in

Ian Oppermann et al.

199

Figure 3.57 versus frequency deviation from the true PRF. The goal function is not
smooth, but shows a large number of local extrema. We observe that the global
maximum value of J( f ) in this realisation is located very near the true PRF.
It turns out that J( f ) becomes small and insensitive to in (3.52) for frequencies f far from the true PRF. For frequencies f near to the true PRF, the local maxima of J( f ) become larger and larger. In the immediate neighbourhood around
the true PRF, the goal function value becomes sensitive to the delay in (3.53).
In the testbed, this scheme achieves significantly less than 100 Hz deviation on
average from the transmitters PRF of 60 MHz. This corresponds to an accuracy
better than 1.7 106 .
Finally, the C sets the VCXO to the frequency f achieving the maximum
J( f).
Delay lag estimator/corrector. The delay lag between transmitter and receiver is
defined as (t) := (t) (t) where (t) is defined in (3.52) and (t) is introduced
in (3.53).
The delay lag estimator receives the ADC samples x [k] for  = 0, . . . , 4 at

of the signal path with minimum observed
t = kTS . First, it finds the index [k]
magnitude,

[k]
= arg min x [k].

(3.60)

 {0,1,...,4}


If [k]
= 4, we conclude that the receiver has lost the symbol frame timing because
the signal path for  = 4 should always take samples from the guard intervals. In
this case, the delay lag estimator uses a parabola interpolation for the ADC values,
!

pk (m) = ak + bk m + ck m2 = x[k]+m
[k],


for m {1, 0, +1},

(3.61)

for estimating



 kTS
[k]
=

mod TPRF .

(3.62)

The algorithm first determines the parabola coecients and calculates


 +
[k]
:= 4 [k]

ak bk
.
ck

(3.63)


= 4, we conclude that the receiver is in a synchronised
On the other hand, if [k]
state because the guard interval is correctly located with respect to the receiver
timing. Now, we only need to track small delay errors. In this case, we use a variant
of an early-late discriminator for estimating (kTS ). Remember that the signal
paths for  = 0 and 3 are sampled early whereas those for  = 1 and 2 are sampled

200

Signal processing

late, see page 55:


[k]
:= Q

x1 [k] x0 [k]
x2 [k] x3 [k]
+Q
,
x1 [k] + x0 [k] + 1
x2 [k] + x3 [k] + 1

(3.64)

where Q() is a suitable quantiser function.


Finally, the obtained estimate [k]
is programmed via the control interface
of the two-stage delay shifter described on page 60. A lookup table is used for

onto a pair of control bytes for the delay step and fine
mapping each desired [k]

delay, such that [k] S [k] + F [k] as shown in Figure 3.51. A typical trajectory

from a simulation in noise is shown in Figure 3.56b and a corresponding
of [k]
trace from the testbed is shown in Figure 3.56a. The testbed trace is obtained with

which are regularly output on an
an oscilloscope by monitoring the values of [k]
additional DAC.
3.8.6. Complexity of UWB impulse radio systems
From todays viewpoint, the promise of low-complexity implementations of IRUWB systems can only be met by noncoherent receivers. These suboptimal implementations will have a far lower bit rate than the envisioned 480 Mb/s because the
ISI can not be equalised in time domain with current state of the art systems.
The low-complexity promise was stated by assuming that the complexity bottleneck is the frequency synthesiser with phase locked loops and narrowband filtering. For coherent IR-UWB systems, the counterparts are fine-granularity delay
shifters and low-jitter oscillators.
However, the use of IQ-demodulators or IF-digitalisation is not required for
IR-UWB systems. Here, sampling mixers are employed which do not consume
much DC power.
Concerning the synchronisation tasks in the testbed, we observe that the acquisition of the pulse repetition frequency is fairly easy and can be accomplished
at low computational cost. The early-late discriminator for fine delay tracking is
rather susceptible to disparities of the gains on dierent symbol paths in the receiver. Therefore, the C needs to calibrate the gains on all symbol paths during
power-up.
3.9. Conclusions
UWB systems may be primarily divided into impulse radio (IR) systems and multiband systems. Multiband systems oer the advantage of potentially ecient utilisation of spectrum.
However, IR systems however have the significant advantage of simplicity, and
so are potentially lower cost. In addition, IR is essentially a baseband technique.
The IR UWB concepts investigated support many modulation schemes including
orthogonal and antipodal schemes. However, the basic modulation must also include some form of spectrum randomisation techniques to limit the interference

Ian Oppermann et al.

201

caused by the transmitted pulse train. Both TH and DS randomisation techniques

were examined. Which modulation scheme to use depends on the expected operating conditions and the desired system complexity.
This chapter has introduced the application of impulses to UWB wireless
transmissions. A number of candidate pulse waveforms have been characterised in
the time and frequency domains. The application of orthogonal pulse waveforms
has then been introduced where these waveforms enable advanced modulation
and multiple-access schemes to be implemented. However, the success of these
schemes is determined by the extent of pulse distortion caused by the transmitter
and receiver circuitry and the propagation channel. Thus, distortion mitigation
techniques are required. Finally, the issue of coexistence of impulse radio with
other spectrum users has been discussed.
Acknowledgments
This work was funded by Siemens AG Austria PSE PRO RCD, all member companies of ftw., and Kplus within ftw. Project I0. The authors would like to thank Holger Arthaber at EMST, Vienna University of Technology, for measurement support
and reviewing the text.
Bibliography
[1] FCC press release, February 2002, http://www.fcc.gov/Bureaus/Engineering Technology/News
Releases/2002/nret0203.html.
[2] J. T. Conroy, J. L. LoCicero, and D. R. Ucci, Communication techniques using monopulse waveforms, in Proceedings of IEEE Military Communications Conference (MILCOM 99), vol. 2, pp.
11811185, Atlantic City, NJ, USA, OctoberNovember 1999.
[3] Time Domain Corporation, Comments of Time Domain Corporation, Docket 98-154. In the
Matter of Revision of Part 15 of the FCCs rules regarding ultra wideband transmission systems,
1998.
[4] G. F. Stickley, D. A. Noon, M. Chernlakov, and I. D. Longsta, Preliminary field results of an
ultra-wideband (10-620 MHz) stepped-frequency ground penetrating radar, in Proceedings of
IEEE International Geoscience and Remote Sensing (IGARSS 97), vol. 3, pp. 12821284, Singapore,
Republic of Singapore, August 1997.
[5] L. Carin and L. B. Felsen, Eds., Ultra-Wideband, Short-Pulse Electromagnetics 2, Conference Proceedings, Kluwer Academic/Plenum Press, New York, NY, USA, 1995, 605 pages.
[6] R. Prasad and S. Hara, An overview of multi-carrier CDMA, in Proceedings of IEEE 4th International Symposium on Spread Spectrum Techniques and Applications (ISSSTA 96), vol. 1, pp.
107114, Mainz, Germany, September 1996.
[7] IEEE, Draft Standard for Telecommunications and Information Exchange Between Systems
LAN/MAN Specific RequirementsPart 15.3: Wireless Medium Access Control (MAC) and
Physical Layer (PHY) Specifications for High Rate Wireless Personal Area Networks (WPAN),
Draft P802.15.3/D17, February 2003.
[8] C. J. Le Martret and G. B. Giannakis, All-digital PAM impulse radio for multiple-access through
frequency-selective multipath, in Proceedings of IEEE Global Telecommunications Conference
(GLOBECOM 00), vol. 1, pp. 7781, San Francisco, Calif, USA, NovemberDecember 2000.
[9] Y.-P. Nakache and A. F. Molisch, Spectral shape of UWB signals-influence of modulation format,
multiple access scheme and pulse shape, TR-2003-40, May 2003.
[10] J. G. Proakis and M. Salehi, Communication System Engineering, Prentice-Hall, Englewood Clis,
NJ, USA, 1994.

202

Signal processing

[11] B. Allen, M. Ghavami, A. Armogida, and H. Aghvami, UWBA ubiquitous wire replacement
technology?
[12] W. Zhuang, X. Shen, and Q. Bi, Ultra-wideband wireless communications, Wireless Communications and Mobile Computing, vol. 3, no. 6, pp. 663685, 2003.
[13] J. G. Proakis, Digital Communications, McGraw-Hill, New York, NY, USA, 1995.
[14] P. Withington, R. Reinhardt, and R. Stanley, Preliminary results of an ultra-wideband (impulse)
scanning receiver, in Proceedings of IEEE Military Communications Conference (MILCOM 99),
vol. 2, pp. 11861190, Atlantic City, NJ, USA, OctoberNovember 1999.
[15] M. Z. Win and R. A. Scholtz, Comparisons of analog and digital impulse radio for wireless
multiple-access communications, in Proceedings of IEEE International Conference on Communications (ICC 97), vol. 1, pp. 9195, Montreal, Quebec, Canada, June 1997.
[16] R. J. Fontana, J. F. Larrick, and J. E. Cade, An ultra wideband communications link for unmanned
vehicle applications, in Proceedings of Association for Unmanned Vehicle Systems International
1997 Conference (AUVSI 97), Multispectral Solutions, Baltimore, Md, USA, June 1997.
[17] M. Hamalainen, V. Hovinen, R. Tesi, J. H. J. Iinatti, and M. Latva-aho, On the UWB system
coexistence with GSM900, UMTS/WCDMA, and GPS, IEEE Journal on Selected Areas in Communications, vol. 20, no. 9, pp. 17121721, 2002.
[18] R. Zetik, J. Sachs, and R. Thoma, From UWB radar to UWB radio, in Proceedings of International Workshop on Ultra-Wideband Systems (IWUWBS 03), Oulu, Finland, June 2003, Center of
Wireless Communications (CWC), IEEE.
[19] M. Katz, Code acquisition in advanced CDMA networks, Ph.D. thesis, Acta Universitatis Ouluensis
C175, University of Oulu, Oulu, Finland, 2002.
[20] J. Iinatti, Performance of DS code acquisition in static and fading multipath channels, IEE Proceedings on Communications, vol. 147, no. 6, pp. 355360, 2000.
[21] J. Iinatti and M. Latva-aho, A modified CLPDI for code acquisition in multipath channel, in
Proceedings of IEEE Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC
01), pp. 610, San Diego, Calif, USA, SeptemberOctober 2001.
[22] F. Ramirez-Mireles and R. A. Scholtz, System performance analysis of impulse radio modulation, in Proceedings of Radio and Wireless Conference (RAWCON 98), pp. 6770, Colorado
Springs, Colo, USA, August 1998.
[23] R. Otte, L. P. de Jong, and A. H. M. van Roermund, Low-Power Wireless Infrared Communications,
Kluwer Academic, New York, NY, USA, 1999.
[24] W. Hirt and D. L. Moeller, UWB radio technologythe global view of a wireless system integrator, in Proceedings of International Symposium on Advanced Radio Technologies (ISART 02),
Boulder, Colo, USA, March 2002.
[25] L. Qinghua and L. A. Rusch, Multiuser detection for DS-CDMA UWB in the home environment, IEEE Journal on Selected Areas in Communications, vol. 20, no. 9, pp. 17011711, 2002.
[26] J. Foerster, V. Somayazulu, S. Roy, et al., Intel CFP presentation for a UWB PHY, Intels multiband UWB PHY proposal for IEEE 802.15.3a, Technical Report IEEE 802.15-03/109r1, Intel Corporation, Santa Clara, Calif, USA, 2003.
[27] C. Sandner, A. Wiesbauer, C. Grewing, et al., A 3 GHz to 7 GHz fast-hopping frequency synthesizer for UWB, in Proceedings of International Workshop on Ultra-Wideband Systems (IWUWBS
04), pp. 405409, Kyoto, Japan, May 2004.
[28] V. Vilnrotter, M. K. Simon, and T.-Y. Yan, The power spectrum of pulse position modulation
with dead time and pulse jitter, Technical Report 42133, Telecommunications and Mission Operations, Jet Propulsion Laboratory, Pasadena, Calif, USA, 1998.
[29] M. Z. Win and R. A. Scholtz, Ultra-wide bandwidth time-hopping spread-spectrum impulse radio for wireless multiple-access communications, IEEE Transactions on Communications, vol. 48,
no. 4, pp. 679689, 2000.
[30] A. F. Molisch, Y.-P. Nakache, P. Orlik, et al., Time hopping impulse radio, Technical Report IEEE
802.15-03/109r1, Mitsubishi Electric, Princeton University, Georgia Institute of Technology, and
New Jersey Institute of Technology, Princeton, NJ, USA, 2003.

Ian Oppermann et al.

203

[31] O. Albert and C. F. Mecklenbrauker, Low-power ultra-wideband radio testbed for shortrange data transmission, in Proceedings of International Workshop on Ultra-Wideband Systems
(IWUWBS 03), Oulu, Finland, June 2003, Center of Wireless Communications (CWC), IEEE.
[32] W. M. Lovelace and J. K. Townsend, The eects of timing jitter and tracking on the performance
of impulse radio, IEEE Journal on Selected Areas in Communications, vol. 20, no. 9, pp. 1646
1651, 2002.
[33] B. Kull and S. Zeisberg, UWB receiver performance comparison, in Proceedings of International
Workshop on Ultra-Wideband Systems (IWUWBS 04), pp. 2125, Kyoto, Japan, May 2004.
[34] B. Schiek, Mesysteme der Hochfrequenztechnik, Huthig, Heidelberg, Germany, 1984.
[35] D. Kress and R. Irmer, Angewandte Systemtheorie, VEB Verlag Technik, Berlin, Germany, 1989.

Ian Oppermann: Department of Electrical and Information Engineering, University of Oulu,

90014, Finland
Email: [email protected]
Matti Hamalainen: The Centre for Wireless Communications (CWC), University of Oulu,
Oulu, Finland
Email: [email protected]
Jari Iinatti: The Centre for Wireless Communications (CWC), University of Oulu, Oulu, Finland
Email: [email protected]
Alberto Rabbachin: The Centre for Wireless Communications (CWC), University of Oulu,
Oulu, Finland
Email: [email protected]
Ben Allen: Centre for Telecommunications, Kings College London, UK
Email: [email protected]
Seyed A. Ghorashi: Centre for Telecommunications, Kings College London, UK
Email: [email protected]
Mohammad Ghavami: Centre for Telecommunications, Kings College London, UK
Email: [email protected]
Olaf Albert: Siemens AG Austria, PSE PRO Radio Communication Devices, Erdberger Lande 2628,
1030 Vienna, Austria
Email: [email protected]
Christoph F. Mecklenbrauker: Das Forschungszentrum Telekommunikation Wien (ftw.),
Donau City Strasse 1/3, 1220 Wien, Austria
Email: [email protected]

Higher-layer issues: ad hoc and

sensor networks
Maria-Gabriella Di Benedetto, Luca De Nardis,
and Salvatore Falco

4.1. Introduction
Ultra-wide-band (UWB) radio has the potential of allowing simultaneous communication of a large number of users at high bit rates [13]. In addition, the high
temporal resolution inherent to UWB provides robustness against multipath fading and is particularly attractive for indoor local area network (LAN) applications.
UWB is also capable of recovering distance information with great precision. As
we will show later in this chapter, distance and position data can lead to better
organization of wireless networks, for instance, through better resource management and routing [4]. UWB signals spread, however, over very large bandwidths
and overlap with narrowband services. As a consequence, regulatory bodies impose severe limitations on UWB power density in order to avoid interference provoked by UWB onto coexisting narrowband systems [5]. It is therefore necessary to
take into account power considerations when designing UWB systems. Throughout this chapter we will show how the distance information made available by the
UWB technology can be exploited to achieve low power levels and increase network lifetime in the long term, while providing an adequate network performance
(in terms of data throughput) in the short term.
In the last few years, the increasing interest in applications based on the deployment of ad hoc networks triggered significant research eorts regarding the
introduction of the energy-awareness concept in the design of medium access control (MAC) and routing protocols. Ad hoc networks are in fact considered as a viable solution for scenarios in which fixed infrastructure, and consequently unlimited power sources, are not available. In such scenarios, an ecient management
of the limited power supply available in each terminal is a key element for achieving acceptable network lifetimes. This is particularly true for sensor networks, for
which long battery duration is one of the basic requirements, given the typical size
of such networks (up to thousands of terminals), as will be analyzed in Section 4.2.
Location information is another valuable way of achieving energy-awareness
in ad hoc networks. In Section 4.3 we first review location-aware routing protocols

206

Higher-layer issues: ad hoc and sensor networks

with focus on power eciency. We then address the problem of information exchange through the network by means of specifically designed protocols.
Next, we introduce in Section 4.4 a MAC protocol which foresees a dedicated
procedure for the acquisition of distance information and which is tailored on
UWB features.
The last section of the chapter, Section 4.5, analyzes the eect of mobility on
the behavior of the proposed MAC and routing strategies.
4.2. Power-efficient UWB networks
4.2.1. Introduction
This section briefly illustrates the current research trends regarding power-ecient
MAC, and power-ecient routing strategies.
4.2.2. Power-efficient MAC
As far as power-aware MAC design is concerned, most of the research activity took
the moves from existing solutions for medium access in wireless networks, and in
particular carrier sensing multiple access (CSMA), and out-of-band signaling.
CSMA is based on a channel sensing period performed by each terminal before starting transmission. The performance obtained by CSMA is however heavily
aected by two phenomena, the well-known hidden terminal and exposed terminal problems. In order to solve the hidden and exposed terminal problems,
solutions that substitute CSMA have been proposed. The multiple access with collision avoidance (MACA) protocol [6], for example, replaces the carrier sensing
procedure with a three-way handshake between transmitter and receiver. Following this approach, further modifications of the MACA protocol have been developed, such as MACAW [7] and MACA-by invitation (MACA-BI) [8].
Practical implementations of MAC protocols combine handshake and carrier sensing, as proposed in the floor acquisition multiple access (FAMA) protocol [9]. These protocols are commonly referred to as CSMA with collision avoidance (CSMA-CA). An example of CSMA-CA is the distributed foundation wireless
MAC (DFWMAC), which has been adopted for the MAC layer of the 802.11 IEEE
standard [10].
An alternative solution to CSMA-CA is oered by the out-of-band signaling
protocol [11]. This solution splits the available bandwidth into two channels: a
data channel used for data packet exchange, and a narrowband signaling channel
on which sinusoidal signals (referred to as busy tones) are asserted by terminals
that are transmitting and/or receiving, in order to avoid interference produced by
hidden terminals. In order to reduce the number of exposed terminals, the use
of two dierent busy tones for transmitting and receiving terminals was proposed
in [12].
CSMA-CA is by far the most common approach adopted in narrowband wireless LANs. This family of protocols suers, however, from several drawbacks in

Maria-Gabriella Di Benedetto et al.

207

terms of power consumption: packet collisions, which cause a waste of power

due to re-transmission, terminals blocked in idle state by active transmissions,
and overhearing, that is, power consumption in the reception of packets by nonintended destinations. Several protocols aiming at solving these drawbacks were
proposed. Among all, it is worth mentioning the power-aware multiple access
protocol with signaling for ad hoc networks (PAMAS) [13]. This protocol combines the CSMA approach with the out-of-band principle, in order to minimize
the time a terminal spends in the idle state without neither transmitting nor receiving. The protocol foresees, in fact, a control channel on which the handshake
between transmitter and receiver takes place, and a data channel on which data
packets are exchanged. The control channel allows terminals to determine when
they may safely switch to sleep mode, saving thus power, without aecting data
throughput or end-to-end delay. The main idea behind PAMAS is that if a node
detects the channel as busy, it goes in sleep mode rather than waste power in idle
mode without being able to exchange data packets. The protocol defines dedicated handshakes, which allow terminals to determine for how long they can keep
the radio interface switched o. Simulations show that this approach leads to a
power save of up to 70% compared to the standard MACA protocol. It should
be noted, however, that part of the dramatic advantage shown by PAMAS over
MACA is intrinsically due to the assumption of adopting a CSMA-CA approach
at the MAC level. PAMAS exploits, in fact, the presence of pause periods in the
terminal lifetime, due to busy channel or lost contention, to save energy without
aecting delay and throughput. If a dierent solution, for example, Aloha without carrier sensing, is adopted, the PAMAS-like approach is no longer a suitable
option.
4.2.3. Power-efficient routing
Power-ecient routing is another topic which is of great interest in relation to distributed ad hoc and sensor networks. Most of the power-aware protocols proposed
in the literature are based either on the definition of power-eective routing metrics or on the exploitation of additional information, such as location of terminals
in the network. In both cases, the final aim is to find source-destination paths that
minimize power consumption.
Regarding the definition of power-aware routing metrics, [14] provides a thorough analysis of the eect of power-aware metrics on network lifetime and fairness in energy consumption between dierent terminals in the network. The metrics taken into account in the analysis include transmission power, cumulative
transmitted and received power, residual power in each node, and their combinations. Interestingly, simulation results presented in [14] show that a straightforward minimization of transmitted power does not necessarily lead to a longer
lifetime for each terminal in the network. The adoption of a power-aware routing
metric may in fact give rise to paths composed of a high number of hops, involving
thus a higher number of terminals in each communication. This strategy provides
however fair power consumption between dierent terminals, increasing thus the

208

Higher-layer issues: ad hoc and sensor networks

network operation time before the first terminal runs out of power with the consequence of potentially causing a network partition.
For the specific case of UWB, a method for setting up connections by optimizing a power-dependent cost function is described in [15], and further refined
in [16, 17]. In [15], a communication cost is attached to each path, and the cost of
a path is the sum of the costs related to the links it comprises. The cost function is
expressed as the sum of two components as follows:

C(x, y) = C0 d + C1 R d .

(4.1)

The first component takes into account the synchronization cost for settingup a new link. If two nodes already share an active link, = 0 and there is no
synchronization cost. If two nodes do not share an active link, = 1 and a synchronization cost is added. The second component takes into account the cost
for transmitting data, and depends upon the requested data rate R. Both terms
are related to power consumption, and therefore depend upon the distance d between two nodes. Note that the evaluation of such a distance relies on the precise
ranging capabilities oered by the UWB technique. The parameter is related
to channel propagation characteristics and has commonly a value between 2 and
4. Constants C0 and C1 are used to weight the synchronization and transmission
components.
In [16] the proposed strategy is compared against traditional routing in a
scenario characterized by fixed terminals and full network connectivity. Results
show that the power-saving strategy, as expected, leads to multihop communication paths between terminals within reach of each other (physical visibility) and
by this way increases network performance by reducing average emitted power
and thus interference levels [16]. In [17] an improved version of the cost function
was proposed, in order to introduce additional parameters in the route selection
metrics. The general form of this cost function is given by

C(x, y) = C(power) + C(sync) + C(interference) + C(quality) + C(delay),

(4.2)

where the first two terms, related to power and synchronization, respectively, resemble the two terms defined in (4.1).
The UWB ranging capability is the basis for the definition of the above power
aware cost function; ranging information, however, can be even more eciently
exploited to build a network map through a distributed protocol, providing thus
the input for a location-based routing algorithm. This topic will be addressed
in Section 4.3, which deals with both distributed positioning and location-based
routing protocols.

Maria-Gabriella Di Benedetto et al.

209

4.3. Location-aware UWB networks

4.3.1. Introduction
As anticipated in Section 4.2, location information is valuable for achieving powereciency in ad hoc networks. In this section we investigate how location information can be exploited in UWB networks, thanks to the accurate ranging capability
oered by this technique. First, we review location-aware routing protocols available in literature, with focus on power-ecient protocols. Next, we address the
topic of how location information can be gathered and exchanged in a UWB network, by providing an example of a GPS-free positioning protocol suitable for a
UWB network, and finally we analyze the eect of UWB-based positioning on the
location-aware routing protocols originally designed for GPS-enabled terminals.
4.3.2. Location-aware routing protocols
As a general concept, all location-aware routing protocols pursue an improvement
of network performance by taking into account location information in the route
selection process. Depending on how this information is used, dierent aspects
of the routing performance can be optimized. Section 4.3.2.2, for example, shows
how location information can be used for increasing route stability. The main focus being here on power eciency, however, the remaining part of this section
addresses routing protocols that exploit location information for reducing routing
overhead, and thus power consumption.
4.3.2.1. Greedy perimeter stateless routing
The greedy perimeter stateless routing (GPSR) [18] protocol uses location information to reduce protocol overhead and obtain a good scalability when both terminal mobility and network size increase. GPSR adopts positional information as
the key metric in packet forwarding, and uses the following simple greedy forwarding strategy.
(1) Each packet is marked by the source terminal with the latest information
about the location of the destination.
(2) Each intermediate node forwards the packet to the neighboring visible node, that is, closest to the location of the destination stored in the
packet itself.
Note that greedy forwarding does not guarantee that a path between source
and destination is always detected, even if it exists. An example of greedy forwarding failure is presented in Figure 4.1. Node x, after receiving a packet from source
S intended for destination D, cannot find a neighbor which satisfies the greedy
forwarding rule, since both neighbors y and z are further than x from the destination D.
When greedy forwarding fails, the GPSR protocol switches from greedy forwarding to perimeter forwarding, in which a terminal is allowed to forward the

210

Higher-layer issues: ad hoc and sensor networks

D

x
y

z
S

Figure 4.1. Example of greedy forwarding failure. Arcs in the figure represent physical visibility between nodes. The greedy forwarding rule fails at node x since both neighbors y and z are further than
x from the destination D.

D
w
xD
y

Figure 4.2. Combination of perimeter forwarding and greedy forwarding. Arcs in the figure represent physical visibility between nodes. The selected path is represented by a thick line (dotted links:
perimeter forwarding, filled links: greedy forwarding).

packet to a neighbor which is further than itself from destination, in order to solve
the stall caused by greedy forwarding. The perimeter forwarding strategy uses the
so-called right-hand rule to forward packets around the area in which the greedy
approach fails, and can be described as follows (see Figure 4.2).
Suppose that x is the first terminal in which greedy forwarding fails.
(1) Terminal x records its position in the apposite L f field in the packet.
(2) Terminal x uses the information on its location and the location of destination D recorded in the packet to determine the xD line.
(3) Terminal x, based on the position of its neighbors, determines which is
the first edge in counterclockwise direction from the xD line (edge xy in
Figure 4.2).
(4) Terminal x forwards the packet to terminal y.
(5) Terminal y checks if it is closer to destination D than the position recorded in the L f field. If this is the case, y switches back to greedy forwarding,
otherwise repeats the procedure from step 2.

Maria-Gabriella Di Benedetto et al.

211

In the example in Figure 4.2, x enters the perimeter forwarding, and w switches back to the greedy forwarding. The grey area defines the perimeter over which
the algorithm switches from greedy forwarding to perimeter forwarding.
4.3.2.2. Location-aware long-lived route selection in
wireless ad hoc network
The location-aware long-lived routing (LLR) protocol [19] belongs to the family
of source-initiated, on-demand routing protocols. The protocol exploits location
information aiming at the minimization of route failures and, as a consequence,
of route reconstruction procedures. As in traditional source routing protocols, a
source terminal S in need of a route to a destination broadcasts route discovery
packets to its neighbors, and the latter will forward packets until the destination is
reached.
In LLR, the source S includes two additional information fields in each generated packet: its own position (XS , YS ) and radio transmission range Rs . Each
terminal A in physical connectivity with S that receives the packet uses such information and its own position (XA , YA ) and transmission range RA to evaluate two
figures.
(1) Forward movement limit (FML), that measures the maximum relative
movement between A and S which can be tolerated before the distance between
A and S is larger than the transmission range of A, defined as
FML = RA distance(A, S) = RA



XS XA

2

2

+ YS YA .

(4.3)

(2) Backward movement limit (BML), that measures the maximum relative
movement between A and S which can be tolerated before the distance between A
and S is larger than the transmission range of S, defined as
BML = RS distance (A, S) = RS



XS XA

2

2

+ YS YA .

(4.4)

FML and BML are then used to evaluate the normalized movement limit (NML),
which provides a single figure averaging FML and BML (note that FML = BML
when RS = RA ):
NML =

FML BML
.
FML + BML

(4.5)

Terminal A updates the packet by including the NML and substituting (XA ,
YA ) and RA to (XS , YS ) and RS , respectively, and forwards it to its neighbors.
The NML is an indicator of link stability: the higher the NML value, the lower
the probability of link failure due to terminal mobility. When the destination D

212

Higher-layer issues: ad hoc and sensor networks

receives several route discovery packets originated from the source S, it selects the
route characterized by the highest NML, in order to minimize the probability of a
route failure. In [19] two alternative ways are foreseen to evaluate the end-to-end
reliability of each route, depending on the way a route discovery packet is updated.
Additive NML. Each terminal adds the NML value relative to the last hop to the
value carried by the packet, and the best path is characterized by the maximum
overall NML.
Min-max NML. Each terminal compares the NML just evaluated with the value
recorded in the packet: if the current value is lower than the old one, the packet is
updated by overwriting the NML field, otherwise the field is left unchanged. In this
case the best route is determined by the worst link in each route (which determines
the NML value in the route discovery packet).
4.3.2.3. Distance routing effect algorithm for mobility (DREAM)
The DREAM algorithm [20] proposes the idea of using location information to
reduce the amount of routing overhead. The protocol combines both proactive
and reactive approaches, by relying on both periodic updates by each terminal
for the dissemination of location information and a flooding-like procedure for
sending a packet to the targeted destination.
When a source terminal S starts the procedure at t = t1 it is supposed to have
the following information:
(i) its own position (XS , YS ),
(ii) the positions of its one-hop neighbors,
(iii) the position of the destination D, (XD , YD ), at a given time t0 < t1 ,
(iv) the maximum speed v of the destination, or at least a probability density
function of the speed, p(v).
The source, based on this information, defines a geographical region in which the
routing packets should be forwarded, and determines the subset of neighbors that
are positioned inside this region, as shown in Figure 4.3. The forwarding region is
determined by the angle .
Depending on the information on destination speed, two cases are possible.
(1) v is known: in this case the distance x traveled by the destination in the
period t1 t0 is given by x = v (t1 t0 ), and, given a distance r between
S and D, one has
x
= arcsin .
r

(4.6)

(2) p(v) is known: in this case in order to determine a forwarding region

including D with probability p, the corresponding value of can be derived from [20]:
 +
r sin /(t1 t0 )

p(v)dv p.

(4.7)

Maria-Gabriella Di Benedetto et al.

213

v (t1 t0 )
D

Forwarding region

S
Figure 4.3. Definition of the forwarding region in the DREAM protocol.

Once has been determined, the terminal sends the packet indicating the neighbors lying within the forwarding region. Each intermediate terminal, when receiving a packet, evaluates if the packet should be forwarded by defining its own
forwarding region, and by checking if any neighbor lies within it. If no neighbor
lies within the forwarding region, the packet is not forwarded, reducing thus the
overhead by avoiding packet retransmissions in wrong directions.
As stated before, the procedure requires positional information regarding not
only the source, but also its neighbors and destination. This information is exchanged through a distributed dissemination algorithm, which constitutes the proactive routing protocol part.
Each terminal A periodically broadcasts update packets containing its own
position; two types of update packets are defined in DREAM:
(i) short-lived packets, for updating the location tables in As neighbors,
(ii) long-lived packets, for updating the location tables in terminals that are
not in direct connectivity with A.
Note that the lifetime of each packet is defined in terms of physical distance reached
from the terminal originating the packet: each terminal B receiving an update
packet generated by terminal A checks both As position (recorded in the packet)
and its own position: if the distance between A and B is higher than the packet
lifetime, the packet is discarded, otherwise, it is forwarded. Short-lived and longlived packets are emitted by each terminal with dierent frequencies, taking into
account the so-called distance eect, which is represented in Figure 4.4 and can be
described as follows.
Given a terminal A moving at speed v, and two terminals B and C that at
t = t0 are positioned at distances DAB and DAC DAB from A, respectively, terminal B will experience a faster variation of the forwarding region to destination A
than terminal C, due to As movement in the time interval t1 t0 (i.e., the angle
B will increase faster than the angle C to keep A in the forwarding region). As a

214

Higher-layer issues: ad hoc and sensor networks

DAC

v (t1 t0 )
A
DAB

C
C

Figure 4.4. Example of distance eect. Given two nodes B and C at distances DAB and DAC , with
DAC > DAB , the angles B and C as derived either from (4.6) or (4.7) are such that B > C , and B
needs thus more frequent updates than C on the position of A, in order to keep A within the forwarding
region.

consequence, B will need much more frequent updates than C regarding As position: such updates will be provided by means of short-lived packets that will reach
B without increasing the routing overhead all over the network. C will receive updates at lower frequency by means of long-lived packets.
Each terminal is thus expected to adapt the emission frequencies of both longlived and short-lived packets to its own speed: fixed terminals will use the lowest
frequencies, while highly mobile terminals will send frequent updates.
4.3.2.4. Location-aided routing (LAR)
The LAR protocol [21, 22] is a typical on-demand routing protocol. In order to
find a route between source and destination terminal, it relies on a flooding-based
route discovery procedure originated by the source by means of a broadcast route
request (RRQ) packet. The packet is forwarded by other nodes all over the network
until either it reaches the destination (in which case the connection enters in the
found status) or a timeout expires leading to a route discovery failure. When the
destination receives an RRQ packet from the source, it replies with a route reply
(RRP) packet which proceeds backward on the selected path, until it reaches the
source, that can then start the transmission of data (DATA) packets. A node which
detects a connection failure while sending or receiving RRP or DATA packets starts
an alarm procedure based on the transmission of broadcast route reconstruction
(RRC) packets. When an RRC packet reaches the source of the connection, the
transmission of DATA packets is stopped, and the source decides either to drop
the connection or to start searching for a new path to destination.
The major drawback of a flooding-based on-demand protocol is the huge
amount of routing overhead generated during path search procedures. The
location-aided routing exploits location information in order to reduce the
amount of routing overhead, although in a dierent way from the DREAM algorithm described in Section 4.3.2.3. The LAR protocol uses in fact the location
information during connection set-up in order to reduce the number of control

Maria-Gabriella Di Benedetto et al.

215

v (t1 t0 )
D
(xD (t0 ), yD (t0 ))

S
Expected zone

Figure 4.5. Definition of expected zone in the LAR protocol. If source S knows the position of destination D in t = t0 , (xD (t0 ), yD (t0 )), and the maximum speed v of D, it can define the region in which D
can lie within in t = t1 (expected zone) as a circle of radius dMAX = v (t1 t0 ) centered in the position
of D in t = t0 .

packets, while no position information is used during DATA packets transmission,

since they are sent along the path found during the set-up phase. The basic location information required by LAR consists in
(i) source position,
(ii) destination position,
(iii) maximum terminal speed.
Such information is exploited during the route discovery procedure as follows.
Suppose that a terminal S starts a route discovery procedure to destination D at
time t = t1 , and that the last information update regarding Ds location was received by S at t = t0 . Based on the estimation of the maximum speed v of terminal D, S can evaluate the maximum distance traveled by D since the last location
update. Such a distance is given by dMAX = v (t1 t0 ). As a consequence, the
current position occupied by D lies in a circular region of radius dMAX centered on
(xD (t0 ), yD (t0 )), referred to as the expected zone, represented in Figure 4.5.
The expected zone indicates which zone of the network should be reached
by RRQ packets. The key idea in LAR is to exploit this information to reduce the
amount of RRQ packets flooding through the network, by allowing forwarding of
packets generated by the source only in the direction of the expected zone containing the destination. The region of the network in which forwarding is allowed is
referred to as request zone. An intermediate terminal is allowed to forward an RRQ
packet only if it lies within the request zone defined by the source of the connection
request.
Note that the request zone may be defined in several ways, the only constraint
being that such zone must include both the position of the source S and the expected zone. A smaller request zone leads to a stronger reduction of the routing
overhead, and can thus achieve a higher power eciency; on the other hand, it
leads to a lower number of neighbors involved in the route discovery procedure,
and may cause a lower percentage of successful connections. Figures 4.6 and 4.7
show two dierent definitions of the request zone, and the eect on the number of
neighbors involved in the route discovery procedure. In Figure 4.6 a conic request

216

Higher-layer issues: ad hoc and sensor networks

v (t1 t0 )

Expected zone

Request zone
Figure 4.6. Definition of conic request zone in the LAR protocol. The nodes that fall within the request
zone defined by source S and destination D are represented in grey, while the nodes that are not involved
are represented in black.

v (t1 t0 )

Expected zone
Request zone
Figure 4.7. Definition of spherical request zone in the LAR protocol. The nodes that fall within the
request zone defined by source S and destination D are represented in grey, while the nodes that are
not involved are represented in black.

zone is considered, similar to the forwarding region defined in the DREAM protocol, while a spherical request zone is presented in Figure 4.7.
In both figures nodes that are involved in the route discovery procedure are
represented in grey, while the nodes which are not involved are black. It is evident
that the conic request zone reduces the overhead much more eectively than the
spherical one, since a lower number of nodes forwards the RRQ packets. On the
other hand, for particular topologies the conic request zone may lead to connection set-up failures which could be avoided with a larger request zone. An example
of route discovery failure is shown in Figure 4.8, which shows a situation in which
no intermediate nodes in physical connectivity with the source fall within the request zone.
The choice of size and shape of the request zone is thus the result of a trade-o
between eectiveness in overhead reduction and probability of successful connection set-up.

Maria-Gabriella Di Benedetto et al.

217

S
D

v (t1 t0 )

Expected zone

Request zone
Figure 4.8. Route discovery failure due to a conic request zone in the LAR protocol. The nodes that
fall within the request zone defined by source S and destination D are represented in grey, while the
nodes that are not involved are represented in black. Note that a route discovery procedure generated
in S and directed to destination D fails, since none of the grey nodes is in physical connectivity with the
source S. A larger request Zone, involving black nodes, would instead lead to a success.

A more advanced trade-o can be obtained by allowing the source node to

perform several attempts of route discovery to a given destination and adopting
in each attempt an increasingly larger request zone: the result would be a higher
percentage of successful connection set-ups, at the price of a higher overhead, obtained as the sum of the routing packets transmitted in each attempt.
This approach neglects, however, the eect on the latency in connection setup: since each attempt would require a significant amount of time, the interval
between connection request and set-up may turn out to be unacceptable in case of
delay-sensitive applications.
As in the case of the DREAM protocol, information about the position of the
destination is required at the source in order to reduce routing overhead, and thus
location information must be disseminated through the network. Oppositely to
DREAM, however, in LAR such dissemination is performed by piggybacking location information in all routing packets, without any additional control packet. At
the beginning of network operations, terminals will be thus forced to find routes in
the absence of location information. In such a situation basic flooding is adopted.
4.3.3. From GPS to UWB positioning
The protocols described in Section 4.3.2 share the common assumption that each
terminal retrieves its own location information from GPS. As a consequence, protocols focus on how to disseminate location information, assuming the problem
of retrieving it as solved.
The adoption of the UWB technology in place of GPS as the basis for retrieving position information opens new doors to location-based applications (such as
indoor deployment), but also poses new challenges.

218

Higher-layer issues: ad hoc and sensor networks

Since UWB can only provide ranging information, in fact, distributed processing of ranging measurements is required in order to build a network map. The
problem of how to build such a map is not trivial, in particular in the case of a pure
ad hoc network, where no fixed reference points are available. The self-positioning
algorithm (SPA) [23], proposes a solution to this problem. The SPA is composed
of two steps. First, each node in the network attempts to build a coordinate system
centered on itself, and to determine the position of its neighbors in this system.
Second, the node-centered coordinate systems converge to a global network-wide
coordinate system.
In the first phase, each node Ni tries to build a coordinate system by
(1) detecting the set of its one-hop neighbors KNi , by using beacons,
(2) evaluating the set of distances from its neighbors DNi ,
(3) broadcasting DNi and KNi to its one-hop neighbors.
Since all nodes perform the above procedure, node Ni knows the distance from
all its one-hop neighbors, the IDs of its two-hop neighbors, and a subset of the
distances between the one-hop neighbors and the two-hop neighbors. This information is used by Ni to build its own coordinate system.
In the second phase, all coordinate systems evolve into a network-wide coordinate system by choosing the same orientation of the x and y axes. Iteratively,
pairs of nodes rotate and align their coordinate systems. Using this distributed approach, a network-wide coordinate system is eventually achieved. Note that nodes
that are not able to build their own node-centered coordinate system can obtain
their position in the network-wide coordinate system if they are in range with
three nodes that already received the coordinate system.
UWB-based positioning poses thus additional requirements, with respect to
GPS, to location-aided routing protocols.
(i) Determination of a network-wide coordinate system may require a long
time. Furthermore, occasional large errors in position information due
to lack of connectivity or unfavorable topology may occur. The routing
algorithm must be then capable to find routes and establish connection
even for incomplete or lacking positioning information.
(ii) Estimation of absolute terminal speeds is in general not available.
The GPSR protocol requires a location service in order to provide the position
of the destination to the source, in order to mark the packet with such information.
When this location service is not available from start, as in UWB-based positioning, each data packet must be forwarded by means of flooding, with a high routing
overhead.
LLR can be adapted to a UWB-based positioning system with minor changes,
but the protocol is inherently tailored for scenarios characterized by terminal mobility (in which route stability is the main concern), and does not oer major
advantages in terms of power eciency in networks composed of still or slowly
mobile terminals.
In the DREAM protocol, both selection of routes and dissemination of information rely on positioning to work properly: in particular, the proactive dissemination algorithm exploits the capability of a terminal to determine its physical

Maria-Gabriella Di Benedetto et al.

219

distance from terminals which are not reachable, to evaluate the validity of an update packet. If location information is not available or subject to errors, both the
proactive and reactive parts of the protocol are aected. Furthermore, lack of speed
information poses an additional issue to the correct behavior of the protocol.
Oppositely, LAR protocol inherently oers a backup solution when no positioning information is available. In fact, in the case of absence of information
regarding the position of the destination, the protocol switches to a basic flooding
scheme, resembling a dynamic source routing [24] protocol. In this case the lack
of positioning information results in a reduced eciency of the protocol.
The above considerations led to the choice of adopting LAR as the routing
protocol for the integrated power-ecient and location-aware solution presented
in the next section.
4.4. Power-efficient and location-aware medium access control design
4.4.1. Introduction
The analysis carried out in Sections 4.2 and 4.3 showed that both distance and position information can be exploited in order to optimize network performance by
reducing power consumption and enabling location-aware path selection strategies. In the case of UWB networks, in particular, it was highlighted in Section 4.3
that position information must be derived by means of a dedicated protocol from
distance information. The design of a medium access control protocol capable of
providing accurate distance information is thus a fundamental step for the development of advanced location-aware strategies in UWB networks.
In this section we will introduce a MAC protocol which foresees a dedicated
procedure for the acquisition of distance information, and is furthermore tailored
for the characteristics of UWB in a low data rate application scenario: the so-called
uncoordinated, wireless, baseborn medium access for UWB communication networks (UWB)2 .
4.4.2. The (UWB)2 MAC protocol
As already mentioned in Section 4.2, CSMA is almost universally adopted in MAC
for narrowband wireless networks, since it overcomes the drawback of frequent
packet collisions experienced in the case of plain Aloha [25]. The CSMA approach,
however, shows significant advantages over the Aloha approach under the two following key hypotheses:
(1) simultaneous transmissions of multiple packets result in destructive collisions because of multiuser interference,
(2) high trac is oered to the network.
Although the above hypotheses are reasonable in a traditional narrowband wireless LAN, we will discuss in the following why this may not longer be the case for a
UWB network, in particular when low data rate applications, such as those typical
of sensor networks, are considered.

220

Higher-layer issues: ad hoc and sensor networks

Probability of correct packet transmission

1
R = 1 Mbps

0.9
0.8

R = 2 Mbps

0.7
R = 4 Mbps

0.6
0.5

R = 6 Mbps

0.4
R = 8 Mbps

0.3
0.2
0.1
0

R = 10 Mbps
0

50

100

150

Number of packets on the channel

Figure 4.9. Probability of successful packet transmission as a function of the number of packets on
the channel. Dierent solid lines correspond to dierent data rates from 110 Mbps. The dashed line
shows simulation results in the specific case of R = 10 Mbps. Drawn from [26].

We will refer throughout this analysis to the specific case of time-hopping impulse radio UWB (TH-IR-UWB), for which the (UWB)2 was originally designed.
The applicability of the ideas exposed here to other UWB variants, such as direct
sequence UWB or multiband OFDM UWB remains to be analyzed.
As explained in previous chapters, UWB in its impulse radio version is based
on the emission of very short pulses which are modulated in position (pulse position modulation (PPM)) or amplitude (pulse amplitude modulation (PAM)) by
the information bits. The duty cycle of emitted signals depends thus on the ratio
between the pulse repetition period (PRP), that is, the average interval between
two consecutive pulses, and the duration of a pulse. In the case of low data rate
transmissions, where bit rates in the order of 100 kbps, or below, can be considered as an acceptable target, this corresponds to signal duty cycles as low as 106 :
the adoption of short pulses oers thus an intrinsic protection from MUI.
Simulation results presented in [26] suggest in fact that in the case of LDR
scenarios, the huge bandwidth adopted for transmissions translates in very short,
rare pulses, and thus in a low probability of collisions between pulses emitted by
dierent terminals. Under this condition the probability of packet error is negligible even in presence of 100 simultaneous packet transmissions (see Figure 4.9,
drawn from [26]).
In the case of PPM, the UWB signal is quasiperiodic due to the rather small
PPM shift value which is usually adopted. In order to mitigate energy peaks at
multiples of the average pulse repetition frequency, time intervals between UWB
pulses must be randomized. This is achieved by adopting TH codes, which introduce additional pseudorandom delays in pulse transmissions. The adoption of the
TH principle also oers an additional degree of freedom for multiple access, which

Maria-Gabriella Di Benedetto et al.

221

can be obtained by assigning dierent TH codes to dierent users, thus adopting

a TH code division multiple access (TH-CDMA) scheme.
The above considerations led to the definition of the (UWB)2 protocol.
(UWB)2 is a multichannel MAC protocol. Multichannel CDMA MAC algorithms, commonly referred to as multicode, have been intensively investigated for
direct sequence (DS) CDMA networks [2729]. Note, however, that although in
the last years most of the research eorts were focused on DS CDMA, frequency
hopping (FH) CDMA and TH-CDMA also provide viable solutions.
The (UWB)2 protocol applies indeed the multicode concept to the specific
case of a TH-IR UWB system. (UWB)2 adopts a hybrid scheme based on the combination of a common control channel, provided by a common TH code, with
dedicated data channels associated to transmitter TH codes. The adoption of a
hybrid scheme can be motivated as follows:
(1) it simplifies the receiver structure, since data transmissions (and corresponding TH codes) are first communicated on the control channel;
(2) it provides a common channel for broadcasting. This is a key property
for the operation of protocols in higher layers. Broadcast messages are,
for example, required for routing and distributed positioning protocols.
As regards code assignment, a unique association between MAC ID and Transmitter Code can be obtained by adopting the algorithm described in [30] which
avoids implementing a distributed code assignment protocol.
The multiple access capability warranted by the TH codes is used by the
(UWB)2 protocol in data transmission. The protocol relies for the access to the
common channel on the high MUI robustness provided by the processing gain of
UWB.
(UWB)2 is designed for distributed networks dedicated to low data rate applications. As a consequence, it does not assume that synchronization between
transmitter and receiver is available at the beginning of packet transmission, because of clock drifts in each terminal during inactivity periods. As a consequence,
a synchronization trailer long enough to guarantee the requested synchronization
probability is added to each packet. The length of the trailer depends on current
network conditions, and is provided to the MAC by the synchronization logic.
In the view of allowing the introduction of the location-related functions presented in the previous sections, in particular distributed positioning and locationaware routing, (UWB)2 also exploits the ranging capability oered by UWB. Distance information between transmitter and receiver is in fact collected during control packets exchange.
The procedures adopted in (UWB)2 for transmitting and receiving packets,
described in [26], have in fact two main objectives:
(i) to exchange information such as the adopted synchronization trailer,
that is, hopping sequence and length,
(ii) to perform ranging. Since no common time reference is available, a twoway handshake is required to collect distance information by estimating
the round-trip time of signals in the air.
An example of such procedure is shown in Figure 4.10, where a transmitter Tx and

222

Higher-layer issues: ad hoc and sensor networks

Rx

Tx
t0

dTxRx = C = C

LE
t0 +

t2 t0
2

t 1 = t0 + +

LC

t2 = t0 + 2 +
dRxTx = C = C

t0 + 2 + 2

DATA
Time

Time

t3 t1
2

t3 = t0 + 3 + 2

Figure 4.10. Example of two-way ranging procedure.

receiver Rx set up 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 the receiver Rx, and finally by the
DATA packet on the transmitter code of transmitter Tx. Thanks to the adoption
of a delay known to both Tx and Rx, this handshake allows the determination
of the propagation delay and thus of the distance TxRx to both the devices
involved in the communication.
The ranging procedure made available by (UWB)2 provides the information
that is required by the positioning and routing algorithms, and thus enables the
adoption of a location-aware, power-ecient path selection strategy, based on the
combination of three key components: the (UWB)2 MAC protocol [26], the selfpositioning algorithm [23], and the location-aided routing protocol [21].
Performance analysis of such strategy in specific test cases, characterized by
dierent degrees of mobility and dierent mobility models, will be the subject of
the next section.
4.5. Performance analysis in specific test cases
4.5.1. Introduction
In this section we will analyze the performance of the solution presented in the
three previous sections of this chapter, composed of the (UWB)2 MAC protocol,
of a distributed positioning protocol, and of the LAR routing protocol, combined
with a power-aware routing metric. The analysis will focus on two main aspects:
(i) eectiveness of the proposed solution in terms of network lifetime and
throughput,
(ii) eect of mobility on system performance.
To this aim, we first address the problem of mobility modeling, by comparing
several mobility models and determining their eect on network topology. Next,
we define three test cases, each of them with dierent node mobility characteristics. The behavior of the proposed MAC and routing strategy within each test case
is finally studied.

Maria-Gabriella Di Benedetto et al.

223

4.5.2. Mobility models

Introducing mobility within the simulated scenarios allows to test the proposed
protocols under more realistic conditions. This is particularly true for routing protocols, where performance greatly varies with the topology of the network. Since
dierent mobility models lead to dierent topologies, however, it is important to
select the mobility model that best fits the reference scenario, that is, the typical
mobile user that each node in the network represents.
Dierent mobility models are presented in literature. An interesting subclass
is composed by group mobility models. This class of models was introduced to
emulate particular node behaviors, such as crowds moving towards a common
destination or rescue (and similar) squads, where each node is bound to show
some degree of mobility similarity with other nodes belonging to the same group.
4.5.2.1. Mobility models in literature
Several mobility models for ad hoc network are proposed in the literature: (1)
the Random Mobility or Brownian model [31], with no relation between speed
and direction of the node in two subsequent timeslots; (2) the Random Direction
model [32], with nodes keeping the same velocity during the whole simulation;
(3) the Ko mobility model used for evaluating the LAR protocol [21], according
to which the path walked by the node is formed by sections with exponentially
distributed length and random direction; (4) the Markovian model [33], which
has many moving states and a transit matrix used to determine whether the node
should keep or change the current motion direction; (5) the Random Waypoints
[24], whose movement paths are composed of segments with random speed, direction, and duration, separated by stand-still periods, simulating pausing intervals; (6) the Inertia [34], in which the node moves along a random direction, with
random speed, for a random time, and then decides whether to keep the same
movement characteristic for the next movement segment (the nodes inertia), or
to select a new direction, speed, and duration.
All the previous models are, clearly, inadequate to simulate a group movement, since each node has its own movement pattern, which has no relationship
with those exhibited by other nodes; in fact, these models were mainly developed
with the aim of testing trac loads oered to the system, rather than reproducing
realistic behaviors.
The need for group mobility models, showing correlated movements among
the nodes and oering flexibility for the implementation of particular behaviors,
has lead to the design of other models; among these, the exponential correlated
random (ECR) model [35]. Other approaches are proposed in literature, in order to bypass the limitation of ECR; those reviewed here are the Reference Point
Group Mobility model [36], and the Reference Velocity Group Mobility model
[37]; besides these models, a new proposal, the Kerberos mobility model, has been
designed and is presented in this section.
An important note on the choice of the mobility model is that, besides models
designed with other aims, there exist no good or bad mobility models: each

224

Higher-layer issues: ad hoc and sensor networks

model is dierent. Each mobility model may or may not show a global, mobilityrelated property, that could in turn be exploited by the system. The capability of
taking into account the characteristics of the mobility model during the design
phase marks, therefore, the dierence among a general/unoptimized system, and
a well-designed one.
4.5.2.2. Nongroup mobility models
In order to make a comparison between group and nongroup mobility models,
simulations were performed, which adopt two nongroup mobility models: the
Random Waypoints [24], and the Inertia [34] models.
In the Random Waypoints model, each node selects a random direction,
speed, and movement duration; at the end of the movement, the node stands still
for a random time; after the pause time, the node selects a new random direction, speed, and movement duration, and starts walking along the new path. The
Random Waypoints model shares a particular mobility pattern with other models
adopting the random direction approach: the density wave eect, with a greater
concentration of the nodes in the central area.
According to the Inertia model, after the end of its movement segment, a node
must choose the direction for the next segment. The two alternative choices for the
next segment are: (1) the node keeps the current direction, with probability , or
(2) the node selects randomly a new direction, with probability 1 ; the weight
of the two alternatives can be dierent, with a higher probability of keeping the
current direction ( > 1/2). The name Inertia refers indeed to the node property
of seldomly changing its direction.
4.5.2.3. Group mobility models
Group mobility models, as previously stated, require that nodes movements are,
somehow, related. In particular, two kinds of relations can be imposed on the
nodes: (1) the nodes must show some relationship in their directions and/or velocities, or (2) the nodes must be somehow close to each other. Models of the first
kind are better suited to simulate the movement of a crowd heading towards a
given direction; such models show the network partition property, with nodes of a
given group emerging by the compound of all network nodes and forming, in vast
areas, distinguishable subnetworks of nodes with similar velocities and directions.
When, on the contrary, the relationship among group nodes is a distance one,
the resulting geographical distribution is characterized by the concentration of the
nodes in a limited area; in this case, the nodes usually do not show any relationship
in velocity or direction.
Reference point group mobility. The Reference Point Group Mobility (RPGM)
model was developed by Hong et al. in [36]. To represent the group mobility behavior of the mobile nodes, the model defines, for each group, a logical reference
point, whose movement is followed by all nodes in the group. The path the reference point moves along defines the entire group mobility behavior, including

Maria-Gabriella Di Benedetto et al.

225

location, speed, direction, acceleration, and so forth. Thus the group trajectory is
determined by providing a path for the reference point, with any node in the group
randomly placed in the neighborhood of its reference point.
Once next position (x, y) of the reference point is determined (the path of the
reference point can be selected with any nongroup-oriented mobility model), each


other node position is calculated adding a random motion vector RM to (x, y).
The length and direction of this module can be determined through any policy;
the authors of the model suggest a uniform distribution within a given range from
the reference point for the length, and a uniform distribution between 0 and 360
for the direction.
Main RPGM characteristic is that the groups are characterized by physical
proximity, that is, all the nodes are close to the reference point. Obviously, the
way this physical proximity translates itself into actual distribution of the nodes
within deployment area depends on the values actually used for the distribution


of the length of the RM random motion vector.
One drawback of the direct implementation of such a mobility model is that
it does not implement a check on the maximum/minimum node velocity, because
of the position-based approach. Since the resulting position of a node is calculated
as the sum of two vectors, the reference point position (x, y) and the random mo

tion vector RM, it is possible, for given values of these two parameters, that the
node should have a speed higher than its maximum; a similar case holds for the
minimum velocity.
About the resultant distribution of the nodes, the adoption of this approach
typically distributes the nodes inside a circular area centered around the reference
point: the wider the distance from the reference point, the more vast the area in
which the nodes are distributed, the looser the bonds between the nodes in the
group.
Reference velocity group mobility. The Reference Velocity Group Mobility (RVGM)
model has been presented in [37] as an evolution of RPGM. The fundamental concept proposed is that movement similarity is a more fundamental characteristic for
group mobility than physical proximity; thus nodes being part of the same group
should show close velocities and directions, rather than physical proximity.
Each group, therefore, is characterized by a group velocity. The nodes that are
members of the group have velocities close to the characteristic group velocity,
with slight deviations. Hence, the characteristic group velocity is also the mean
group velocity. The distributions of both the group velocity and the velocity deviation can be of any type, in order to model the various possible mobility patterns.
Main characteristic for RVGM is that, in a scenario with infinite dimensions
area, the network will eventually incur a complete partition, with each subnetwork
corresponding to a group; in a finite-dimensions area this behavior is not fully
developed, because of the presence of area boundaries, but is latently present in
the movement patterns.

226

Higher-layer issues: ad hoc and sensor networks

The main reason why RVGM was evolved from RPGM is that the former
model is better suited when dealing with mobility or partition prediction purposes;
representing the nodes by their physical coordinates, RPGM does not allow to easily detect group movement patterns and the trend in network topology changes.
About the resultant distribution of the nodes, the adoption of this approach
typically distributes the nodes to cover the whole area with nodes with similar
velocities and directions.
Kerberos mobility model. The basic concept in Kerberos mobility model is that
each node in a group is allowed to move around freely, provided it keeps contact
with the other nodes in its group.
This basic concept can be translated into the adoption of a condition of physical proximity between the given node and the other nodes of its group: when this
condition is met, the node moves according to a nongroup mobility model (e.g.,
Inertia), while, when the node is far from its fellows, it is obliged to move closer to
them.
The dierence between Kerberos and RPGM models is that, in the latter, each
node follows the reference point and, for this reason, is close to its fellows; this is
how physical proximity arises in RPGM. In the Kerberos model, on the contrary, it
is explicitly required for each node to monitor its distance from its fellows, and to
stay close to them.
Kerberos requires that, after each mobility update, each node checks how many
nodes of its own group it is connected to: this set of nodes forms the fellowship of
the node. Two nodes in the same fellowship can be connected through a direct link
when the destination node is within range, or through relaying of intermediate
nodes, all belonging to the group, which are at the same time, members of the
fellowship. If the number of nodes in the fellowship is greater than half the number
of nodes forming the group, the node is allowed to move freely; otherwise, the
node is separated by the main chunk of group (if any exists), and is compelled to
move towards the closest group node not belonging to its fellowship.
Notes on group mobility models. The dierent approaches adopted by RPGM,
RVGM, and Kerberos, are characterized by dierent and characterizing patterns
in the disposition of the nodes belonging to a single group. In the case of RPGM,
since the nodes are bound to stay within a given range from the same point, each
group forms a globular pattern, centered around the reference point; this usually
means a high level of connectivity between the nodes in the group. On the other
hand, in the case of Kerberos, there are higher probabilities for patterns in which
the nodes form a chain or a line; this happens because, once connectivity requirements are verified, each node is allowed to move freely, and therefore even further
from the group itself.
However, Kerberos mobility model is closer to RPGM than to RVGM; it is
possible, changing the connectivity requirement into a maximum distance from
a chosen node (the group leader), to obtain mobility patterns very similar to
RPGM.

Maria-Gabriella Di Benedetto et al.

227

4.5.2.4. Mobility model metrics and test scenario

In order to analyze the impact of dierent mobility models on the performance
of a routing protocol, it is appropriate to simulate the mobile network and to test
the protocol. It is also possible, however, to analyze the topological characteristics
of networks adopting the dierent mobile models. These topological properties
have, obviously, a deep impact on routing performance: for example, the average
number of nodes within range will aect the route availability between any two
given nodes, so that those mobility models that scatter the nodes over the whole
simulation area are expected to have a lower route availability than those mobility
models which concentrate the nodes in smaller areas.
In [38], the authors provide a number of mobility model metrics, to be applied to the evaluation of mobility models impact on routing performance.
(i) Number of link changes: the number of transitions from the state connected to the state disconnected and vice versa, for the link between
any pair of nodes in the network. Its average over the number of links in
the network is the average number of link changes, and is calculated separately for links between nodes of the same group and for links between
nodes of dierent groups.
(ii) Link duration: the average duration of the link between two nodes, that
is, the average time that two nodes remain continuously within range.
Its average over the number of links in the network is the average link
duration, and is calculated separately for links between nodes of the same
group and for links between nodes of dierent groups.
(iii) Path availability: the fraction of time during which a path of links within
transmission range is present between two nodes in the network. Its average over the number of links in the network is the average path availability, and is calculated separately for paths between nodes of the same
group and for paths between nodes of dierent groups.
We propose an additional metric, in order to evaluate the probability that any two
given nodes are within range.
(iv) Link availability: the percentage of time during which link between two
given nodes is active, that is, the two links are within transmission range.
Its average over the number of links in the network is the average link
availability, and is calculated separately for links between nodes of the
same group and for links between nodes of dierent groups.
The simulation scenario consisted in a network of 16 nodes, divided into 4
groups of 4 nodes each. The deployment area was a square with 100 m sides, with
transmission range equal to 20 m. The simulations have been performed with different values for node maximum velocity, from 1 m/s to 10 m/s, and lasted 2 hours.


As regards the mobility model parameters, RPGM RM maximum length has been
set to 30 m; Kerberos simulations were performed adopting Inertia as mobility
model when a node is free to move.

Higher-layer issues: ad hoc and sensor networks

Average number of link changes

228
550
500
450
400
350
300
250
200
150
100
50
0

Waypoints Inertia

RVGM

RPGM

Kerberos

Intragroup
Intergroup
Figure 4.11. Average number of link changes for dierent mobility models.

4.5.2.5. Mobility models simulation results

Number of link changes. Figure 4.11 shows the average number of link changes for
all the mobility models investigated, and for both intergroup and intragroup cases;
the values are expressed in average number of changes per minute.
Kerberos shows a high number of changes in links between nodes belonging
to the same group. This is the result of two behaviors:
(i) when a node is connected to the group, it is allowed to move freely, without any constraint based on the positions of the other nodes of its group.
This means that nodes in the same, connected, group may be headed
along dierent directions; the result is a high number of link breaks or
set-ups;
(ii) when a node detects that its group is disconnected, it is compelled to
move towards the other nodes in its group: the result is the set-up of one
or more links with group nodes.
The second most notable result in Figure 4.11 is that the dierence between
intra and intergroup links is higher in group mobility models than in nongroup
ones: a foreseeable eect of the group bound observed by the former models, since
the latter do not make distinction between group and nongroup mobility models.
In the evaluation of a routing protocol, this mobility model metrics can give
an estimation of the control trac that the protocol produces each time a link is
established or is broken: this includes control trac to update network topology
knowledge, alternative route checking, and so forth.
Link duration. The average link duration of each mobility model is shown in Figure 4.12, for intra and intergroup links; the values are expressed in seconds.
This mobility model metrics tells apart the group by the nongroup mobility
models: the former show a high dierence in average link duration between intra

Maria-Gabriella Di Benedetto et al.

229

80

Average link duration

70
60
50
40
30
20
10
0

Waypoints Inertia RVGM RPGM Kerberos

Intragroup
Intergroup
Figure 4.12. Average link duration, in seconds, for dierent mobility models.

and intergroup links. In the case of RVGM, the higher duration for intragroup
links is derived from the circumstance that, if two nodes form an intragroup link,
they have a large probability to have similar velocities and directions. In the case of
Kerberos and, above all, RPGM, the physical proximity property is what make the
link more stable: the nodes forming the intragroup link are very likely to remain
close in the future. Kerberos shows the lowest average link duration among group
mobility models, because group nodes neither share similar velocities and directions (as in RVGM), nor are compelled to stay within a maximum distance from
a common point; however, the constraint on group connectivity allows a node to
break its intragroup link only if it has, at least, another intragroup link, otherwise
it will depart from its group. Between the nongroup mobility models, Inertia has
a slightly higher average link duration than Random Waypoints. This is due to the
fact that the two nodes composing the link will be more likely to keep on moving
in the same direction if they are modeled with Inertia, rather than with Random
Waypoints.
This metric shows dierent information, with respect to the number of link
changes, because it measures the length of a connection after it has been established, while the other metric measures the frequency of link establishment. Taking RPGM and Random Waypoints as examples, these models have average number of link changes values that are very close, but their values for average link duration are far apart: this happens because both establish links with the same frequency, but those exhibited by RPGM are built by nodes that are compelled to stay
close for a long time, while in Random Waypoints the nodes usually break the link
shortly after they build it.
Regarding routing, this metric gives an estimation of the average reliability of
a link. This parameter impacts the procedures that must select the most reliable

230

Higher-layer issues: ad hoc and sensor networks

70

Average link availability

60
50
40
30
20
10
0

Waypoints Inertia RVGM RPGM Kerberos

Intragroup
Intergroup
Figure 4.13. Average link availability, in percentage of time, for dierent mobility models.

route among several possible: it is clear that a route mainly composed of intragroup links will be much more reliable than another mainly composed of intergroup linksunder the hypothesis of the adoption of a group mobility model.
Link availability. Figure 4.13 shows the results for the average link availability metric, expressed in percentage of time in which a given link is established.
It is clear how group and nongroup mobility models greatly dierentiate themselves, as regards the availability of a link. Nongroup mobility models do not distinguish between intra and intergroup links; also, the percentage of availability of
a link is quite low, since the link is formed by nodes that move in a completely
independent fashion. Group mobility models, on the contrary, highly privilege intragroup links over the others: the similarity in the movement of the RVGM, the
physical proximity requirement of the RPGM, and the group connectivity constraint of Kerberos, all result in a more probable presence of a link between nodes
of the same group; on the other side, nodes from dierent groups are compelled
to have dierent movement behaviors, ending in less frequent links.
Through the average link availability mobility model metrics, it is possible to
determine the expected number of links a node has. The connectivity of a node
is an important parameter for a routing protocol: it can be used to calculate the
expected load on each link, to estimate the capability of performing relaying for
other nodes trac.
Path availability. Simulation results for the average path availability are shown in
Figure 4.14. For each possible pair of nodes in the network, simulations have been
performed to determine the probability of existence of a path, composed of one or
more links. The intragroup paths are those in which the starting and the ending
nodes of the path are members of the same group.

Maria-Gabriella Di Benedetto et al.

231

100
Average path availability

90
80
70
60
50
40
30
20
10
0

Waypoints Inertia RVGM RPGM Kerberos

Intragroup
Intergroup
Figure 4.14. Average path availability, in percentage of paths, for dierent mobility models.

Once again, it is possible to tell apart the group from the nongroup mobility
models, because of the dierence between inter and intragroup values, and of the
value of the latter case. As expected, nongroup mobility models show no dierence
between intra and intergroup paths, since all the nodes are moved independently
from their group fellows. The dierence between Random Waypoints and Inertia
is given by the fact that Random Waypoints scatters the nodes all over the simulation area: the probability of network partition (which implies a dramatic decrease
in average path availability values) is therefore high, but is lower than in Inertia,
where some nodes tend to be separated by the remaining network for long times.
As regards the group mobility models, note that
(i) Kerberos shows the highest average path availability for intragroup paths.
With the adoption of this mobility model, the network shows highly
connected groups that span over wide areas. intragroup links are less
frequent, but they allow a large number of paths to exist;
(ii) RPGM shows a lower average path availability for intragroup paths, but
higher than Kerberos for intergroup ones. Physical proximity requirement allows for a few group partitions, but the groups cover wider areas than Kerberos, thus increasing the probability of an intergroup link,
which sustains a number of intergroup paths;
(iii) RVGM results are due to the aforementioned property of similarity in
movements, rather than in physical proximity, for the nodes of a group.
The resulting values are similar to an average between the nongroup and
the remaining group mobility models.
As one could expect, the path availability mobility model metrics is the one
with the most direct impact on routing performance. It, in fact, directly translates
in the probability of existence of a path between any two given nodes. Low path

232

Higher-layer issues: ad hoc and sensor networks

probabilities could require routing strategies that invest few resources in path discovery, since there is a higher possibility that this procedure ends in a failure.
Simulation results show that dierent mobility models have quite dierent
properties, as regards the path availability: comparing, for example, the Random
Waypoints and the RVGM, the latter shows a higher availability for intragroup
path than the former, while the opposite holds for intergroup paths. The actual
answer on the global path availability depends on the ratio between intra and
intergroup links in the actual network, that is, on the ratio between group and
network cardinality. It is important to underline, however, that in those scenarios
in which group mobility models are more suited, there is, typically, also a traffic relationship between group nodes; that is to say, a node will be more likely to
communicate with nodes of the same group, rather than nodes of other groups:
the eect of this property is that the actual percentage of routes found depends also
on the ratio between intra and intergroup path, weighted with the ratio between
intra and intergroup communications.
4.5.3. Test cases
We will investigate three dierent test cases which are characterized by the following common scenario: a network of N terminals deployed in initial random
positions. Each terminal periodically generates a connection request to a random
destination, following a Poisson distribution with average time between two subsequent requests. Each connection request is characterized by a constant bit rate Rc
and a total number of bits to be transferred which is randomly selected in the interval [1, MaxDimBit]. Each terminal is furthermore characterized by a total amount
of energy ETOT , which is reduced after each packet transmission or reception based
on the following energy model [39]:


ETX = Estart + L ETx-bit-rate Rb + ETx-bit-prop Rb d ,





ERX = E start + L ERx-bit-fixed + ERx-bit-rate Rb ,

(4.8)

where L is the length of the packet.

Note that, although this model is not specific for UWB, it addresses a class of
devices which is close, in terms of achieved bit rate and complexity, to the UWB
devices foreseen for low bit rate networks.
All test cases share furthermore the settings adopted for positioning and routing protocols, in particular data and control packet sizes. Finally, the initial amount
of energy ETOT for each node is the same in all the three test cases, although its actual value depends on the transmission range.
The three test cases dier however in the mobility pattern of the terminals.
(i) Test case 1: network of still terminals in random positions.
(ii) Test case 2: network of mobile terminals in initial random positions,
following the inertia mobility model.

Maria-Gabriella Di Benedetto et al.

233

(iii) Test case 3: network of mobile terminals in initial random positions,

following the Kerberos mobility model.
The initial selection of terminal positions can heavily influence the network
performance, in particular in the scenario considered in test case 1: as a consequence, in order to determine the average performance of the strategies, Nruns
simulation runs were executed, with terminals deployed in random positions in
an area of Aside xAside m2 . Given a set of terminals in random coordinates, however,
network topology is also determined by the transmission range, which influences
the network connectivity. For this reason, tests were executed for three dierent
transmission ranges, RTX-low , RTX-med , and RTX-high , leading to low, medium, and
high network connectivity, respectively. The common settings to all tests cases are
presented in Table 4.1, while Table 4.2 presents the initial energy per node for each
transmission range.
In each test case four dierent strategies were compared, which combine in
dierent ways the two key components of the proposed solution: location-aided
routing algorithm and power-aware routing metric. The distinction between the
routing protocol and the routing metric allows in fact to evaluate the eect of each
of the two components on system performance.
The four strategies are as follows:
(i) dynamic source routing with hop minimization (DSR + hop),
(ii) dynamic source routing with cost function minimization (DSR + cost),
(iii) location-aided routing with hop minimization (LAR + hop),
(iv) location-aided routing with cost minimization (LAR + cost).
As previously stated, the performance of each strategy was analyzed under
conditions of limited available energy taking into account two aspects: long-term
system performance, that is, how long the network is alive and thus capable of
transferring data, and short-term system performance, that is, how network behaves during its life. The following routing strategy metrics were selected to measure long-term and short-term performance.
(i) Number of found connections.
(ii) Number of DATA packets delivered to destination.
(iii) Percentage of found connections.
(iv) End-to-end throughput for DATA packets.
A found connection is defined as a connection in which the destination terminal,
following the reception of a routerequest (RRQ) packet, is able to send back to
the source terminal a routereply (RRP) packet. The total number of found connections is a good indicator of the long-term behavior of the network since, given
the constant rate of connection requests generation, a longer network lifetime
will lead to a higher number of requested Connections and, eventually, of found
connections. Since each connection, however, is set up in order to transfer DATA
packets, we also consider the total number of DATA packets transferred during
network lifetime, in order to determine which is the impact of each strategy at
packet level.
The percentage of found connections, oppositely, allows to measure the network behavior in the short-term, by measuring how good each connection is served

234

Higher-layer issues: ad hoc and sensor networks

Table 4.1. Simulation settings common to all test cases.
Parameter

Value

Rb
Rc

1 Mbps
50 kbps
50 s

MaxDimBit
N

1 Mbit
20
4

Estart
ETx-bit-rate
ETx-bit-prop

2.76 105 J
3.25 107 J
1.25 1011 J

ERx-bit-fixed
ERx-bit-rate

1.13 107 J
2.79 107 J

Adim
RTX-low
RTX-med

80 m
20 m
40 m

RTX-high
RRQ size
RRP size

60 m
760 bit
760 bit

DATA size
ACKDATA size
RRC size

5000 bit
250 bit
250 bit

by the network. The end-to-end throughput for DATA packets provides the same
kind of short-term information at the packet level.
Note that the evaluation of the short-term performance behavior is fundamental in the comparison of the dierent strategies since, from a theoretical viewpoint, a strategy could lead to an overall high number of found connections, thanks
to a long lifetime which allows for a high number of requested connections, but at
the price of a low percentage of found connections.
4.5.3.1. Test case 1
This test case analyzed a scenario with terminals in fixed locations, which are unknown at start of operations.
Since terminals could not move, network topology only changed when a terminal ran out of energy: as a consequence, in this scenario, performance was
significantly aected by the transmission range. Furthermore, as we will see, dierent strategies were aected in a dierent way by variations in network connectivity.
Figure 4.15 presents the number of found connections under conditions of
low network connectivity. The LAR + hops strategy achieves the highest number
of found connections, but the performance of the four strategies is quite similar,
since the main factor in determining network behavior is the degree of network
connectivity. This conclusion is also supported by the other indicators, that is, the

Maria-Gabriella Di Benedetto et al.

235

Table 4.2. Initial energy amount per node for each of the three transmission ranges considered in
simulations.
Energy ETOT

Range
RTX-low
RTX-med
RTX-high

25 J
250 J
800 J

500
450
Found connections

400
350
300
250
200
150
100
50
0
DSR + hops DSR + cost LAR + hops LAR + cost
Strategies
Figure 4.15. Found connections for test case 1 with RTX-low .

number of DATA packets, the percentage of found connections, and the DATA
throughput, presented in Figures 4.16, 4.17, and 4.18.
Interestingly, the adoption of the power-aware cost function in place of number of hops leads to slightly worse performance in terms of number of found connections and of delivered DATA packets, indicating a shorter network lifetime.
This can be explained by observing that when the transmission range is low, the
energy costs in transmission and reception which are not related to propagation
are more relevant in the overall energy consumption, thus eliminating the advantage of increasing the number of hops per connection. As an example, let us consider the case of the transmission of a DATA packet over a distance d = 10 m. The
overall energy cost of the transmission (ETX + ERX ), based on the model in (4.8) is
given by


E = 2 Estart + L ERx-bit-fixed + ERx-bit-rate Rb + ETx-bit-rate Rb



+ L ETx-bit-prop Rb d





= 5.52 105 + 5000 7.17 107 + 5000 1.25 107
= 4.27 103 J.



(4.9)

236

Higher-layer issues: ad hoc and sensor networks

103
45

Received DATA packets

40
35
30
25
20
15
10
5
0

DSR + hops DSR + cost LAR + hops LAR + cost

Strategies

Figure 4.16. Received DATA packets for test case 1 with RTX-low .

Percentage of found connections

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

DSR + hops DSR + cost LAR + hops LAR + cost

Strategies

Figure 4.17. Percentage of found connections for test case 1 with RTX-low .

If two hops over d = 5 m are used in place of the single hop, we get the overall
cost:


E = 4 Estart + 2 L ERx-bit-fixed + ERx-bit-rate Rb + ETx-bit-rate Rb



+ 2 L ETx-bit-prop Rb d







= 1.104 104 + 2 5000 7.17 107 + 2 5000 7.8125 109
= 7.35 103 J,

which is higher than the cost of the single hop at longer distance.

(4.10)

Maria-Gabriella Di Benedetto et al.

237

End-to-end DATA throughput

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
DSR + hops DSR + cost LAR + hops LAR + cost
Strategies
Figure 4.18. End-to-end DATA throughput for test case 1 with RTX-low .

As far as routing algorithms are concerned, results show that, independently

of the selected routing metric, the performance increase guaranteed by the adoption of LAR is far from dramatic. This can be explained by observing that the main
advantage of LAR is the reduction of overhead by avoiding broadcast packets being
forwarded in wrong directions. Under the condition of low connectivity, the number of broadcast packets is inherently limited by the low number of links available
in the network, thus achieving the same eect pursued by LAR, at the price of a
low percentage of found connections.
If we consider a higher transmission range, on the other hand, the above considerations are no longer valid, and we would thus expect an advantage in terms of
network lifetime by adopting the power-aware cost function and the LAR protocol.
The results for transmission range set to RTX-med are presented in Figures 4.19
and 4.20 for found connections and delivered number of DATA packets, respectively. Results show that in both cases the LAR + cost strategy leads to the best
performance, increasing by a factor of 2 both the number of found connections
and the number of DATA packets. Furthermore, the adoption of the power-aware
cost function leads to better performance independently of the selected routing
algorithm.
Noticeably, also the short-term performance indicators, presented in Figures
4.21 and 4.22, are improved by the adoption of LAR routing and power-aware cost
function.
The advantage of the LAR + cost solution is confirmed when the RTX-high
transmission range is considered: Figure 4.23 shows that the number of found connections is increased by a factor of 2 by adopting the LAR + cost strategy.
The increase in found connections obtained by adopting the cost function is
around 10%, as it was in the case of transmission range set to RTX-med . This is coherent with the fact that, given the number of terminals and the size adopted in the

238

Higher-layer issues: ad hoc and sensor networks

102
20

18
Found connections

16
14
12
10
8
6
4
2
0

DSR + hops DSR + cost LAR + hops LAR + cost

Strategies

Received data packets

Figure 4.19. Found connections for test case 1 with RTX-med .

104
18
17
16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
0
DSR + hops DSR + cost LAR + hops LAR + cost

Strategies
Figure 4.20. Received DATA packets for test case 1 with RTX-med .

test cases, the average distance between two terminals is 40 m. As a consequence,

the maximum advantage from the cost function is obtained when the transmission range reaches 40 m, that is, for RTX-med . Further increases of the transmission
range do not lead to any additional gain, since the energy saving obtained with the
cost function is not related to the maximum transmission distance reachable by a
terminal, but to the actual distance between terminals.

Maria-Gabriella Di Benedetto et al.

239

Percentage of found connections

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
DSR + hops DSR + cost LAR + hops LAR + cost
Strategies
Figure 4.21. Percentage of found connections for test case 1 with RTX-med .

1
End-to-end DATA throughput

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

DSR + hops DSR + cost LAR + hops LAR + cost

Strategies

Figure 4.22. End-to-end DATA throughput for test case 1 with RTX-med .
Table 4.3. Inertia mobility settings for test case 2.
Parameter

Value

VMAX

6 m/s
0.5

4.5.3.2. Test case 2

This test case analyzed a scenario in which terminals were randomly deployed, and
then moved following the Inertia mobility model, with mobility settings presented
in Table 4.3.

Higher-layer issues: ad hoc and sensor networks

Found connections

240

102
16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
0
DSR + hops DSR + cost LAR + hops LAR + cost

Strategies
Figure 4.23. Found connections for test case 1 with RTX-high .

The movement capability of terminals had a strong impact on network behavior, in particular in scenarios where network connectivity was more aected by
mobility. This is the case when low transmission range RTX-low is considered: the
introduction of mobility led indeed to results which are completely dierent from
those obtained in test case 1 with still terminals.
Figure 4.24 shows the number of found connections in this scenario. It is evident that in this case, the introduction of the LAR algorithm increases network
performance also with transmission range set to RTX-low : in particular, the comparison between DSR + hops and LAR + hops shows an increase of about 20% in
the number of found connections.
Note that in conditions of low connectivity and mobility, the adoption of cost
instead of hops does not bring any advantage. As already observed for test case 1, in
fact, the low transmission range leads in most cases to a higher energy consumption when moving from a single hop to two or more hops. In presence of mobility,
furthermore, the cost-based strategies suer from an additional increase of energy
spent in signaling connection failures due to mobility by means of broadcast RRC
packets. In fact, since the cost function leads to a higher average number of hops,
as shown in Figure 4.25, the paths selected with this metric are more subject to
failures caused by terminal mobility.
If we consider a transmission range set to RTX-med , we observe, on the contrary,
that the adoption of the cost function significantly increases the number of found
connections. In this condition the best solution is thus the LAR + cost strategy, as
shown in Figure 4.26.
This conclusion is also confirmed by the other long-term parameter, that is,
the number of received DATA packets, shown in Figure 4.27.
The short-term parameters confirm that LAR + cost is overall the best choice:
the throughput, presented in Figure 4.28, is almost identical for both LAR-based

Found connections

Maria-Gabriella Di Benedetto et al.

241

102
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
0
DSR + hops DSR + cost LAR + hops LAR + cost

Strategies
Figure 4.24. Found connections for test case 2 with RTX-low .

Average number of hops

3.5
3
2.5
2
1.5
1
0.5
0
DSR + hops DSR + cost LAR + hops LAR + cost
Strategies
Figure 4.25. Average number of hops for test case 2 with RTX-low .

strategies, and the same is true for the percentage of found connections, in Figure
4.29.
The advantage obtained by the LAR-based strategies on the short-term parameters is also influenced by the longer network lifetime. As an example, Figure
4.30 shows the evolution of the percentage of found connections over time for
a single simulation run. The plot shows that the dierence in the average value
(Figure 4.29) is due to the capability of LAR-based strategies to keep the network
in steady state for a longer period of time.

242

Higher-layer issues: ad hoc and sensor networks

102
18

16
Found connections

14
12
10
8
6
4
2
0

DSR + hops DSR + cost LAR + hops LAR + cost

Strategies

Figure 4.26. Found connections for test case 2 with RTX-med .

104

11

Received DATA packets

10
9
8
7
6
5
4
3
2
1
0

DSR + hops DSR + cost LAR + hops LAR + cost

Strategies

Figure 4.27. Received DATA packets for test case 2 with RTX-med .

The case of high network connectivity confirms LAR-based strategies as the

most ecient ones (see Figure 4.31), but in this case the performance gain achieved
with the cost function is not significant.
This can be explained by considering that, as already observed for test case
1, an increase of transmission range over the average distance between two terminals does not bring any additional advantage for cost-based strategies. At the
same time, the negative eect of connection failures due to mobility increases with
transmission range, since RRC packets are transmitted at maximum power; this
means that each RRC packet will consume higher energy, and that each packet will

Maria-Gabriella Di Benedetto et al.

243

1
End-to-end DATA throughput

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
DSR + hops DSR + cost LAR + hops LAR + cost
Strategies
Figure 4.28. End-to-end DATA throughput for test case 2 with RTX-med .

Percentage of found connections

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
DSR + hops DSR + cost LAR + hops LAR + cost
Strategies
Figure 4.29. Percentage of found connections for test case 2 with RTX-med .

be forwarded for a higher number of times, thanks to the higher network connectivity. As a consequence, the eciency of cost-based strategies is reduced.
4.5.3.3. Test case 3
This test case analyzed a scenario in which terminals were randomly deployed,
and then moved following the Kerberos mobility model, with mobility settings
presented in Table 4.4.
The analysis of the performance of the four strategies in the case of the Kerberos mobility model is somewhat more dicult than in the other cases; the peculiar spatial distribution of terminals generated by this group mobility model must

244

Higher-layer issues: ad hoc and sensor networks

Percentage of found connections

1
0.9
0.8
0.7
0.6
0.5
0.4

1000 2000 3000 4000 5000 6000 7000 8000 9000

Time (s)
DSR + hops
DSR + cost

LAR + hops
LAR + cost

Figure 4.30. Percentage of found connections for test case 2 with RTX-med (single run).

Found connections

102

12
11
10
9
8
7
6
5
4
3
2
1
0

DSR + hops DSR + cost LAR + hops LAR + cost

Strategies

Figure 4.31. Found connections for test case 2 with RTX-high .

be taken into account for a correct evaluation of the simulation results, and the
distinction between intra and intergroup network topology properties must be
considered.
The global number of found connections in the case of low transmission range
is presented in Figure 4.32. The figure shows that, as much as in the case of test
case 2, the presence of mobility allows for a higher network connectivity and thus
leads to better results than in the case of still terminals. For all the four strategies,
however, performance is significantly worse than in the case of the Inertia mobility model, as confirmed by the number of received DATA packets, presented in
Figure 4.33.

Maria-Gabriella Di Benedetto et al.

245

Table 4.4. Kerberos mobility settings for test case 3.

Parameter

Value

VMAX
Kerberos range
Nneigh

6 m/s
RTX VMAX /3 m
4

102
10

9
Found connections

8
7
6
5
4
3
2
1
0
DSR + hops DSR + cost LAR + hops LAR + cost
Strategies
Figure 4.32. Found connections for test case 3 with RTX-low .

Received DATA packets

103
25

20
15
10
5
0

DSR + hops DSR + cost LAR + hops LAR + cost

Strategies

Figure 4.33. Received DATA packets for test case 3 with RTX-low .

This is due to the fact that the Kerberos model does not lead to a uniform
distribution of nodes, but creates groups of nodes. Figure 4.34 shows in fact the

246

Higher-layer issues: ad hoc and sensor networks

45

Average distance (m)

40
35
30
25
20
15

5
Time (s)

Inertia (global)
Inertia (intragroup)

10
103

Kerberos (global)
Kerberos (intragroup)

Figure 4.34. Average distances between two nodes in the network (global) and between two nodes
in the same group (intragroup) for Inertia and Kerberos, the latter with Kerberos range = RTX-low
VMAX /3 m.

average distance between two nodes in the network and the average distance between two nodes in the same group for both Inertia and Kerberos. As one could
expect, the two distances are similar in the case of Inertia, since no special group
behavior is defined in the model. Oppositely, in the case of Kerberos a large dierence between the two values can be observed.
The lower performance observed for all strategies in test case 3 is due to the
higher connectivity which characterizes the Kerberos model. Results in Section
4.5.2.5 show in fact that this mobility model guarantees a higher average connectivity than the Inertia model; this translates in a higher number of forwarded RRQ
packets, and thus a shorter network lifetime. The higher connectivity, on the other
hand, leads to a better performance the short term, as presented in Figure 4.35,
showing the percentage of found connections for the first 5000 seconds of a simulation run for the same strategy (DSR + hops) with the two mobility models.
The higher network connectivity also determines in this test case a larger performance gap between DSR and LAR-based strategies. The average increase in
number of found connections obtained by switching to LAR is in fact in the order
of 100% for test case 3, whereas it was only around 20% in test case 2 with the
same settings (see Figure 4.24).
Note that the cost-based strategies are penalized by both the short transmission range and the negative eect of mobility on route duration, and achieve thus
lower performance than the hop-based strategies, similarly to what was observed
in test cases 1 and 2.
The results with medium transmission range RTX-med , presented in Figures
4.36 and 4.37, confirm that the adoption of the LAR strategies increases the network lifetime. In this case too, the adoption of the cost-based metric has a negative

Maria-Gabriella Di Benedetto et al.

247

Percentage of found connections

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2

10

15

20

25

30

Time (s)

35

40

45

50
102

DSR + hops (Inertia)

DSR + hops (Kerberos)
Figure 4.35. Percentage of found connections in a single simulation run for the DSR + hops strategy
with Inertia and Kerberos mobility models.

102
11

Found connections

10
9
8
7
6
5
4
3
2
1
0
DSR + hops DSR + cost LAR + hops LAR + cost
Strategies
Figure 4.36. Found connections for test case 3 with RTX-med .

eect on network performance, so that the LAR + hops strategy is still the overall best solution in this scenario. Note that this conclusion diers from what was
obtained for the same transmission range in test cases 1 and 2. In particular, in
test case 2 the negative eect of mobility was compensated by the energy saving
achieved with a higher number of hops, so that for the transmission range RTX-med ,
the LAR + cost led to the best performance (Figure 4.26). This is no longer true in
test case 3, due the peculiar mobility pattern characterizing the Kerberos model. In

248

Higher-layer issues: ad hoc and sensor networks

103
50

Received DATA packets

45
40
35
30
25
20
15
10
5
0

DSR + hops DSR + cost LAR + hops LAR + cost

Strategies

Figure 4.37. Received DATA packets for test case 3 with RTX-med .

102
11

Found connections

10
9
8
7
6
5
4
3
2
1
0
DSR + hops DSR + cost LAR + hops LAR + cost
Strategies
Figure 4.38. Found connections for test case 3 with RTX-high .

Section 4.5.2.5 it was show in fact that the Kerberos mobility model is characterized by a shorter intergroup link duration if compared to Inertia: as a consequence,
the adoption of intergroup links in a connection has a stronger negative eect on
route stability. The cost-based strategies are more sensible to this eect, since the
selected routes are characterized by a high number of hops, and there is thus a
significant probability for a route to include one or more intergroup links. The
lower route stability is clearly indicated by the fact that, despite the highest number of found connections, the LAR + cost strategy is capable of delivering a lower
number of DATA packets than the LAR + Hops strategy: this is due to the fact that

Received DATA packets

Maria-Gabriella Di Benedetto et al.

249

103
70
65
60
55
50
45
40
35
30
25
20
15
10
5
0
DSR + hops DSR + cost LAR + hops LAR + cost

Strategies
Figure 4.39. Received DATA packets for test case 3 with RTX-high .

routes selected with the LAR + cost strategy have a shorter duration and thus the
corresponding connections are able to deliver a lower number of packets before
being broken for a lack of connectivity.
Results with high transmission range RTX-high (Figures 4.38 and 4.39) further
confirm that cost-based strategies are heavily aected by the mobility pattern of
the Kerberos model. Although in fact the higher network connectivity allows a
higher average number of packets to be delivered in each connection for all the
four strategies, yet a significant gap can be observed between hop-based and costbased strategies.
Quite interestingly, as the transmission range (and consequently the Kerberos
Range) increases, the results in test case 3 approach those obtained in test case 2;
this is coherent with the fact that, as the Kerberos range increases, there is a lower
and lower probability for a node to remain isolated from its own group and thus to
be forced to modify its mobility pattern. As a consequence, the Kerberos mobility
model will fall back more and more to the standard Inertia mobility model.
4.5.4. Conclusions
The analysis carried out throughout this section had the objective of testing the
eectiveness of the location-based, power-aware approach for MAC and routing
in low bit rate UWB networks introduced in the previous sections of this chapter.
The results of such analysis highlight that the exploitation of ranging and locationing information provided by the UWB physical layer may eectively extend
network lifetime without significant eects on short-term network performance.
The analysis also pointed out the direct relation between the scenario in which
the network is deployed and the performance of the selected routing strategy. In
particular, network connectivity, transmission range, and mobility pattern of the

250

Higher-layer issues: ad hoc and sensor networks

nodes are all factors which can influence the performance of a routing strategy
and, most important, may have a dierent eect on dierent strategies.
Results showed in fact that, depending on the scenario, the best network performance can be obtained by means of either a full exploitation of the positional
information provided by UWB in both routing algorithm and metric (as in test
cases 1 and 2), or a partial one, limited to the adoption of the location information in the routing algorithm (as in test case 3).
The flexibility guaranteed by the cost function defined in Section 4.1, however, enables a smooth transition from a fully power-aware metric to a traditional
hop-based metric, and thus allows for a fine adaptation of the MAC and routing
strategy to any network scenario, ranging between the two extreme cases of sparse
networks of fixed nodes and dense networks of highly mobile nodes.
Bibliography
[1] M.-G. Di Benedetto and G. Giancola, Understanding Ultra Wide Band Radio Fundamentals, Prentice Hall, Upper Saddle River, NJ, USA, 2004.
[2] M.-G. Di Benedetto and B. Vojcic, Ultra wide band wireless communications : a tutorial, Journal
of Communications and Networks, vol. 5, no. 4, pp. 290302, 2003.
[3] M. Z. Win and R. A. Scholtz, Ultra-wide bandwidth time-hopping spread-spectrum impulse radio for wireless multiple-access communications, IEEE Transactions on Communications, vol. 48,
no. 4, pp. 679689, 2000.
[4] L. De Nardis and M.-G. Di Benedetto, Medium access control design for UWB communication
systems: review and trends, Journal of Communications and Networks, vol. 5, no. 4, pp. 386393,
2003.
[5] Federal Communications Commission, Revision of Part 15 of the Commissions rules regarding
ultra-wideband transmission systems, Report 98153, Federal Communications Commission,
Washington D.C., USA, 2002.
[6] P. Karn, MACAA new channel access protocol for packet radio, in Proceedings of ARRL/CRRL
Amateur Radio Ninth Computer Networking Conference, pp. 134140, Ontario, Canada, September 1990.
[7] V. Bharghavan, A. Demers, S. Shenker, and L. Zhang, MACAW: A media access protocol for wireless LANs, in Proceedings of Conference on Applications, Technologies, Architectures and Protocols
for Computer Communication (SIGCOMM 94), pp. 212225, London, UK, AugustSeptember
1994.
[8] F. Talucci and M. Gerla, MACA-BI (MACA by invitation). A wireless MAC protocol for high
speed ad hoc networking, in Proceedings of 6th IEEE International Conference on Universal Personal Communications Record, vol. 2, pp. 913917, San Diego, Calif, USA, 1997.
[9] C. L. Fullmer and J. J. Garcia-Luna-Aceves, Floor acquisition multiple access (FAMA) for packet
radio networks, in Proceedings of Conference on Applications, Technologies, Architectures and
Protocols for Computer Communication (SIGCOMM 95), pp. 262273, Cambridge, Mass, USA,
AugustSeptember 1995.
[10] B. P. Crow, I. Widjaja, J. G. Kim, and P. T. Sakai, IEEE 802.11 wireless local area networks, IEEE
Communications Magazine, vol. 35, no. 9, pp. 116126, 1997.
[11] F. A. Tobagi and L. Kleinrock, Packet switching in radio channels: part IIThe hidden terminal problem in carrier sense multiple access and the busy tone solution, IEEE Transactions on
Communications, vol. 23, no. 12, pp. 14171433, 1975.
[12] J. Deng and Z. J. Haas, Dual busy tones multiple access (DBTMA): a new medium access control
for packet radio networks, in Proceedings of IEEE International Conference on Universal Personal
Communications (ICUPC 98), vol. 2, pp. 973977, Florence, Italy, October 1998.

Maria-Gabriella Di Benedetto et al.

251

[13] S. Singh and C. S. Raghavendra, PAMAS: Power aware multiaccess protocol with signalling for
ad hoc networks, ACM Computer Communication Review, vol. 28, no. 3, pp. 526, 1998.
[14] C.-K. Toh, Maximum battery life routing to support ubiquitous mobile computing in wireless
ad hoc networks, IEEE Communications Magazine, vol. 39, no. 6, pp. 138147, 2001.
[15] M.-G. Di Benedetto and P. Baldi, A model for self-organizing large-scale wireless networks,
in Proceedings of International Workshop on 3G Infrastructure and Services, pp. 210213, Athens,
Greece, July 2001.
[16] L. De Nardis, P. Baldi, and M.-G. Di Benedetto, UWB ad-hoc networks, in Proceedings of IEEE
Conference on Ultra Wideband Systems and Technologies (UWBST 02), pp. 219223, Baltimore,
Md, USA, May 2002.
[17] P. Baldi, L. De Nardis, and M.-G. Di Benedetto, Modeling and optimization of UWB communication networks through a flexible cost function, IEEE Journal on Selected Areas in Communications, vol. 20, no. 9, pp. 17331744, 2002.
[18] B. Karp and H. T. Kung, Gpsr: greedy perimeter stateless routing for wireless networks, in Proceedings of 6th ACM/IEEE International Conference on Mobile Computing and Networking (ACM
MOBICOM 00), pp. 243254, Boston, Mass, USA, August 2000.
[19] D. Kim, Y. Choi, and C. K. Toh, Location-aware long lived-route selection in wireless ad hoc
network, in IEEE 52nd Vehicular Technology Conference (VTC 00), vol. 4, pp. 19141919, Boston,
Mass, USA, September 2000.
[20] S. Basagni, I. Chlamtac, V. R. Syrotiuk, and B. A. Woodward, A distance routing eect algorithm for mobility (DREAM), in Proceedings of ACM/IEEE International Conference on Mobile
Computing and Networking (ACM MOBICOM 98), pp. 7684, Dallas, Tex, USA, 1998.
[21] Y.-B. Ko and N. H. Vaidya, Location-aided routing (LAR) in mobile ad hoc networks, in Proceedings of 4th Annual ACM/IEEE International Conference on Mobile Computing and Networking
(MOBICOM 98), pp. 6675, Dallas, Tex, USA, October 1998.
[22] Y.-B. Ko and N. H. Vaidya, Optimizations for location-aided routing (LAR) in mobile ad hoc
networks, Tech. Rep. 98-023, CS Department, Texas A & M University, Dallas, Tex, USA, 1998.
[23] S. Capkun, M. Hamdi, and J. P. Hubaux, GPS-free positioning in mobile Ad-Hoc networks, in
Proceedings of 34th Annual Hawaii International Conference On System Sciences, pp. 34813490,
Maui, Hawaii, USA, January 2001.
[24] D. B. Johnson and D. A. Maltz, Dynamic source routing in ad hoc wireless networks, in Mobile
Computing, vol. 353, pp. 153181, Kluwer Academic Publishers, Dordrecht, The Netherlands,
1996.
[25] D. P. Bertsekas and R. Gallager, Data Networks, Prentice Hall, Englewood Clis, NJ, USA, 2nd edition, 1992.
[26] M.-G. Di Benedetto, L. De Nardis, M. Junk, and G. Giancola, (UWB)2 : uncoordinated, wireless, baseborn medium access for UWB communication networks, Springer Mobile Networks and
Applications, vol. 10, no. 5, pp. 663674, 2005.
[27] D. Raychaudhuri, Performance analysis of random access packet-switched code division multiple access systems, IEEE Transactions on Communications, vol. 29, no. 6, pp. 895901, 1981.
[28] S. Dastangoo, B. R. Vojcic, and J. N. Daigle, Performance analysis of multi-code spread slotted ALOHA (MCSSA) system, in Proceedings of IEEE Global Telecommunications Conference
(GLOBECOM 98), vol. 3, pp. 18391847, Sydney, NSW, Australia, November 1998.
[29] E. S. Sousa and J. A. Silvester, Spreading code protocols for distributed spread-spectrum packet
radio networks, IEEE Transactions on Communications, vol. 36, no. 3, pp. 272281, 1988.
[30] M. S. Iacobucci and M.-G. Di Benedetto, Computer method for pseudo-random codes generation, National Italian patent, RM2001A000592.
[31] Y.-C. Hu and D. B. Johnson, Caching strategies in on-demand routing protocols for wireless ad
hoc networks, in Proceedings of 6th Annual International Conference on Mobile Computing and
Networking (MOBICOM 00), pp. 231242, Boston, Mass, USA, August 2000.
[32] E. M. Royer, P. M. Melliar-Smith, and L. E. Moser, An analysis of the optimum node density
for ad hoc mobile networks, in Proceedings of IEEE International Conference on Communications
(ICC 01), vol. 3, pp. 857861, Helsinki, Finland, June 2001.

252

Higher-layer issues: ad hoc and sensor networks

[33] B. Liang and Z. J. Haas, Predictive distance-based mobility management for multidimensional
PCS networks, IEEE/ACM Transactions on Networking, vol. 11, no. 5, pp. 718732, 2003.
[34] S. Basagni, I. Chlamtac, and V. R. Syrotiuk, Dynamic source routing for ad hoc networks using
the global positioning system, in Proceedings of IEEE Wireless Communications and Networking
Conference (WCNC 99), vol. 1, pp. 301305, New Orleans, La, USA, September 1999.
[35] M. Bergamo, R. Hain, K. Kasera, D. Li, R. Ramanathan, and M. Steenstrup, System design specification for mobile multimedia wireless network (MMWN) (draft), Tech. Rep., DARPA project
DAAB07-95C -D156, October 1996.
[36] X. Hong, M. Gerla, G. Pei, and C.-C. Chiang, A group mobility model for ad hoc wireless networks, in Proceedings of 2nd ACM International Workshop on Modeling and Simulation of Wireless
and Mobile Systems (MSWiM 99), pp. 5360, Seattle, Wash, USA, August 1999.
[37] K. H. Wang and B. Li, Group mobility and partition prediction in wireless ad-hoc networks, in
Proceedings of IEEE International Conference on Communications (ICC 02), vol. 2, pp. 10171021,
New York, NY, USA, AprilMay 2002.
[38] F. Bai, N. Sadagopan, and A. Helmy, Important: a framework to systematically analyze the impact of mobility on performance of routing protocols for adhoc networks, in Proceedings of 22nd
Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM 03),
vol. 2, pp. 825835, San Francisco, Calif, USA, MarchApril 2003.
[39] R. Min and A. Chandrakasan, A framework for energy-scalable communication in high-density
wireless networks, in Proceedings of International Symposium on Low Power Electronics and Design
(ISLPED 02), pp. 3641, Monterey, Calif, USA, August 2002.

Maria-Gabriella Di Benedetto: INFOCOM Department, School of Engineering,

University of Rome La Sapienza, 00184 Rome, Italy
Email: [email protected]
Luca De Nardis: INFOCOM Department, School of Engineering,
University of Rome La Sapienza, 00184 Rome, Italy
Email: [email protected]
Salvatore Falco: INFOCOM Department, School of Engineering,
University of Rome La Sapienza, 00184 Rome, Italy
Email: [email protected]

Spatial aspects of UWB

Thomas Kaiser, Jurgen Kunisch, Werner Sorgel,
Christian Waldschmidt, Werner Wiesbeck, F. Argenti,
T. Bianchi, L. Mucchi, L. S. Ronga, Alain Sibille,
Sigmar Ries, Christiane Senger, W. Pam Siriwongpairat,
Masoud Olfat, Weifeng Su, K. J. Ray Liu, and Feng Zheng

5.1. Introduction
The aim of this part is to discuss the spatial aspects of the UWB radio channel from
various perspectives. First, in Section 5.2, a model for the UWB radio channel by J.
Kunisch, a spatial model is presented based on measurement data that were obtained in oce environments with line-of-sight, non-line-of-sight, and intermediate conditions. Basically, the goal was to arrive at expressions for the space-variant
impulse response or, equivalently, for its space-variant transfer function such that
impulse responses or transfer functions belonging to adjacent locations could be
correlated properly. The distinguishing feature of the ultra-wideband indoor radio channel is that certain individual paths are recognizable and resolvable in the
measurements. This is adequately reflected by the proposed spatial model. Moreover, the model covers also movements of the receiving or transmitting antennas
on a small scale (several wavelengths). At the end, the relevant model equations
are presented in a compact, algorithmic-like form.
Section 5.3, UWB antenna arrays by W. Sorgel et al., introduces state-of-theart UWB antenna array concepts and their applications. Then, the transient radiation behaviors of UWB antenna arrays in the frequency and in the time domain
are discussed by deriving the antenna arrays, transient responses and their quality
measurements such as dispersion and ringing. The theoretical modeling is substantiated by experimental results, and then this section finishes with simulation
results for a linear exponentially tapered slot antenna array (Vivaldi antennas) for
time-domain beam steering.
In Section 5.4, UWB polarization diversity by F. Argenti et al., two transmitting
antennas with orthogonal polarization are proposed in order to uniformly cover
an indoor area. A single-antenna receiver results into a 2 1 MISO (multiple-input
single-output) system so that space-time coding becomes applicable. The goal is
to employ the polarization diversity in order to obtain a quality constant symbol
detection while the receiver moves around within the covered area.

254

Spatial aspects of UWB

Section 5.5, spatial diversity by A. Sibille, addresses several basic issues pertaining to spatial diversity in UWB systems. It mainly concentrates on impulsebased radio, but it also evokes eects concerning frequency-domain modulation
schemes such as OFDM. The main subjects are the impacts of electromagnetic
coupling between sensors and of channel properties (like fading and angular variance) on spatial diversity.
Beside the two main pillars of MIMO signal processing, namely, spatial multiplexing and space-time coding, the third and most classical one is beamforming,
which is revisited in Section 5.6, UWB beamforming and DOA estimation by S. Ries
et al. Beamforming for UWB signals has some special properties that are dierent
from the narrowband case. For instance, because of the absence of grating lobes
in the beampattern, the spacing of the array elements is not limited by half of the
wavelength, so that high resolution can be achieved with only a few array elements.
Interestingly, the short duration of UWB pulses is also beneficial for digital beamforming. This means that despite enormous sampling rate requirements, digital
beamforming of UWB signals is not too far away to become reality. At the end
of this section, the principal feasibility of direction-of-arrival (DOA) estimation is
shown by an illustrative example.
In Section 5.7, performance analysis of multiantenna UWB wireless communications by W. P. Siriwongpairat et al., the performance of UWB-MIMO systems
using dierent models for the wireless channels and dierent modulation schemes
is presented. In particular, the performance merits of UWB-ST-coded systems
employing various modulation and multiple-access techniques, including timehopping (TH) M-ary pulse-position modulation (MPPM), TH binary phase-shift
keying (BPSK), and direct-sequence (DS) BPSK are pointed out. At the end, the
application of multiple transmit and/or receive antennas in an UWB-OFDM system is discussed.
Section 5.8, channel capacity of MIMO UWB indoor wireless channels by F.
Zheng and T. Kaiser, concludes this part. Therein the channel capacity will be evaluated in terms of ergodic capacity and outage probability for UWB indoor wireless systems with multiple transmit and receive antennas (multiple-input multipleoutput, MIMO). For some special cases, analytic closed-form expressions for the
capacity of UWB wireless communication systems are given, while for other cases
the capacity is obtained by Monte Carlo simulation approach. Several conclusions
can be drawn from these results. For instance, the UWB-MIMO communication
rate supportable by the channel increases linearly with the number of transmit or
receive antennas for a given outage probability, which reminds on the significant
data rate increase of MIMO narrowband fading channels.
5.2. A model for the ultra-wideband space-variant
indoor multipath radio channel
5.2.1. Introduction
The foundation of all wireless systems is the radio channel. Consequently, a component which is of paramount importance for the design of any wireless system

Thomas Kaiser et al.

255

is a channel model that reproduces the radio propagation aspects relevant to this
particular system. In this section, we will construct, in a step-by-step fashion, such
a model of the ultra-wideband radio channel. The model is based on measurement data (cf. [1]) that were obtained in oce environments with line-of-sight,
non-line-of-sight, and intermediate conditions. We will also give site-specific parameter profiles that have been derived from these measured data.
In particular, the model will apply to the indoor multipath channel in the FCC
frequency band [2] and it will cover, to some amount, space-variant properties of
the channel. The typical application focuses on inter- or intraoce communication with distances between transmitter (Tx) and receiver (Rx) below approximately 10 m, for example, WPAN1 systems.
Basically, the aim is to arrive at expressions for the space-variant impulse response or, equivalently, its space-variant transfer function such that impulse responses or transfer functions belonging to adjacent locations will be correlated
properly. From such a basic description, more specialized models may be derived
easily, for example, a discrete channel model that incorporates eects of modulation and demodulation.
A distinguishing feature of the ultra-wideband indoor radio channel is that
certain individual paths are recognizable and resolvable2 in the measurements. In
contrast to dense multipath contributions, the corresponding echoes exist individually over distances larger than a wavelength. Consequently, an important requirement for our model is that it will incorporate this property: the delay and
amplitude changes of individual paths corresponding to changing positions of the
transmitter or receiver will be accounted for to produce a realistic Doppler behavior (e.g., delay-Doppler spectrum).
Another important aspect of the model which again is related to space variance is that it is focused on small-scale rather than large-scale behavior. Our model
will be suited to describe movements of the receiving or transmitting terminal antenna within certain areas measuring several wavelengths in diameter, but it will
not be suited to derive, for example, power laws for movements on a larger scale,
say, across a large building. This limitation is a consequence of the limitations of
the underlying measured data.
The model is of a hybrid nature. By that we mean that it combines a statistical approach (to model dense multipath) and a simple quasideterministic method
(to model strong individual echoes). Therefore, the model is basically a statistical
modelin the sense that for a model realization certain values are drawn from
prescribed distributions with prescribed parameters, but it also has a minor deterministic componentin the sense that for a model realization certain values are
derived from a fixed algorithm that does not depend on statistical distributions but
1WPAN stands for wireless personal area networks.
2The notion of resolvability is somewhat fuzzy here because the capability to resolve a path is
mostly an attribute of the superresolution algorithm used to estimate echo positions, rather than an
attribute of the path itself. A path may be considered resolvable, for example, if it can be resolved by an
optimal estimator. This in turn poses the question of what is meant by an optimal estimator (e.g., in
terms of resolution and accuracy).

256

Spatial aspects of UWB

on a description of the environment. In our case, however, we do not use a description of the real environment (like it would be the case with ray tracing/launching
models), rather a trivial surrogate environment will be usedhence the designation quasideterministic.
We will start out by reviewing some important properties of the ultra-wideband radio channel in the next section. The following Section 5.2.3 will elaborate
on how these properties influence the design of the model. Finally, the model equations will be presented compactly in an algorithm-like form.
5.2.2. The ultra-wideband radio channel
What, then, are those properties specific to ultra-wideband channels that necessitate the design of dedicated ultra-wideband channel models beyond reparameterizing standard models that are applicable to plain wideband conditions?
The indoor ultra-wideband radio channel diers from the plain wideband
channel in at least two respects.
Spatial resolution. For indoor environments, the ratio of the spatial resolution
(proportional to the inverse bandwidth) and the characteristic geometric dimensions of a room are such that echoes caused by certain paths may be appreciably
distinguishable from each other.3 This applies in particular to early echoes corresponding to paths with small excess delays. For example, in LOS conditions, the
direct path may be shorter by more than the spatial resolution than the second
shortest path, causing a gap in between the corresponding two echoes. With increasing delay, the arrival probability for paths increases because the corresponding path distances get longer, which in turn increases the probability of interactions with the environment. Consequently, the onset of the inevitable dense multipath regime is shifted towards larger delays. For medium to large excess delays,
the power delay profile (PDP) fits quite well into the exponential decay scheme
that has been observed for wideband channels (cf., e.g., [3]). For small excess delays, however, ultra-wideband channel PDPs tend to have a more spiky appearance
caused by individual echoes. Yet the first path corresponding to LOS conditions
may be followed by some trailing multipath contributions caused, for example, by
the mechanical antenna support.
Depending on the environment and Tx-Rx distance, the direct path, floor,
ceiling, and wall reflections may be recognizable. An example for an LOS ultrawideband channel PDP illustrating these points is given in Figure 5.1. Furthermore, Figure 5.2 shows 150 adjacent power delay profiles for the same channel obtained by displacing one antenna over a linear distance of 150 cm. The individual
3The term path is used here to denote a trajectory along which energy travels, that is, paths

are related to the physical process of wave propagation. In contrast, the term echo is used here to
denote certain features observable in an impulse response. An echo may be caused by superposition of
several contributions arriving via dierent pathsunresolved superposition shows as fading. An echo
can also result as an artifact of signal processing (countermeasures: e.g., windowing, superresolution
algorithms, etc.). In this sense, echoes are related to the observation of the wave propagation process.

Mag. (rms(S21 )) (dB (10 GHz))

Thomas Kaiser et al.

257
70
80
90

100
110

10

20

30

40 50 60
Delay (ns)

70

80

90

70

80

90

Mag. (rms(S21 )) (dB (10 GHz))

(a)
70
80
90
100
110

10

20

30

40 50 60
Delay (ns)
(b)

Figure 5.1. Power delay profile derived from 30 30 impulse responses on a 30 cm 30 cm grid for
LOS conditions: (a) measurement; (b) model.

echoes and the dependence of the corresponding delays from the spatial position
(including a slight curvature of the LOS trace due to an appreciable change of the
Tx-Rx distance over the 150 cm of antenna displacement) are clearly visible. The
trace slanted to the right (beginning approximately at 16 nanoseconds for displacement = 0 m) corresponds to a strong wall reflection. Figures 5.3, 5.4, 5.5, and 5.6
show the response of the channel to an impulse with a waveform that corresponds
to a Kaiser-Bessel frequency-domain window. All figures show a comparison of
measurement and model results, which, due to the hybrid model nature, are expected to be similar, but not identical.
Frequency behavior. For ideal constant gain antennas, the free-space channel power gain as a function of the frequency f follows an f 2 law which may be neglected for small fractional bandwidths. For an ultra-wideband channel extending
from 3 to 11 GHz, however, the f 2 law amounts to a dierence of approximately

1.4

65

1.2

70

75

0.8

80

0.6

85

0.4

90

0.2

95

10

15
20
Delay (ns)

Mag. (S21 ) dB (10 GHz)

Spatial aspects of UWB

Antenna displacement (m)

258

25

1.4

65

1.2

70

75

0.8

80

0.6

85

0.4

90

0.2

Mag. (S21 ) dB (10 GHz)

Antenna displacement (m)

(a)

95

10

15
20
Delay (ns)

25

(b)
Figure 5.2. Power delay profiles for 150 cm displacement of one antenna almost perpendicular to the
line of sight: (a) measurement; (b) model.

11 dB between the lower and upper bandlimitsthe corresponding figure for

W(ideband)CDMA4 is in the order of 0.02 dB. Strongly depending on the considered environment and the employed antennas, dierent exponents or other
frequency dependencies may occur (cf. [1, 4]). A gain decay with increasing frequency may be expected to cause a reduction of the eective bandwidth as the
larger part of the received signal energy concentrates towards lower frequencies. To
take full advantage of the radio channel, this property has to be considered in both
antenna and signal design, and consequently by the channel model (cf., e.g., [5]).
5.2.3. Model description
This section gives an overview of the model elements.
4UTRA: 5 MHz bandwidth centered around approximately 2 GHz.

Thomas Kaiser et al.

259

Real (S21 ) (20 GHz)

104

5
4
3
2
1
0
1
2
3
4
5

20

40

60

80 100 120
Delay (ns)

140

160

140

160

(a)

Real (S21 ) (20 GHz)

104

5
4
3
2
1
0
1
2
3
4
5

20

40

60

80 100 120
Delay (ns)
(b)

Figure 5.3. Example for a single impulse response (LOS): (a) measurement; (b) model.

5.2.3.1. Modeled quantity

The primary quantity to be modeled is the space-variant transfer function H( f , rt ,
rr ), where f , rt , and rr denote frequency, transmitter, and receiver positions, respectively. From H other relevant quantities may be derived easily, including the
space-variant impulse response h(, rt , rr ) and the baseband impulse response
hB (, rt , rr ), and so forth.
By assuming trajectories rt (t) or rr (t), space variance may be converted to
time variance caused by motion of terminals.
5.2.3.2. Modeling approach
In the last section, it was pointed out that the power delay profile roughly follows
an exponential decay rule. Furthermore, in a dense multipath region, the samples

260

Spatial aspects of UWB

Real (S21 ) (20 GHz)

104

5
4
3
2
1
0
1
2
3
4
5
10

12

14

16
18
Delay (ns)

20

22

24

20

22

24

(a)

Real (S21 ) (20 GHz)

104

5
4
3
2
1
0
1
2
3
4
5
10

12

14

16
18
Delay (ns)
(b)

Figure 5.4. Example for a single impulse response (LOS, detail): (a) measurement; (b) model.

of the baseband impulse response reasonably follow a Rayleigh amplitude statistic [1]. Therefore, the approach from [3] is adopted as a starting point. To account
for the ultra-wideband specific channel behavior, the model proposed here is based
on the following rules.
Rule 1. For given rt and rr , a number of individual echoes (arrival times and
complex amplitudes) are generated according to a certain algorithm described below.
Rule 2. With each individual echo, a diuse multipath cluster is associated with
Rayleigh amplitude echoes generated in every bin, an exponential echo
power decay law, and uniformly distributed phase.
Rule 3. For any rt and rr , the arrival times of an individual echo and its associated multipath cluster are equal.
Rule 4. The cluster height (the maximum ray power expectation value within
the given cluster) is set to be a certain amount (which is a model parameter) below the power of the associated individual echo.

Thomas Kaiser et al.

261

Real (S21 ) (20 GHz)

104

1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1

20

40

60

80 100 120
Delay (ns)

140

160

140

160

(a)

Real (S21 ) (20 GHz)

104

1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1

20

40

60

80 100 120
Delay (ns)

(b)
Figure 5.5. Example for a single impulse response (NLOS): (a) measurement; (b) model.

Rule 5. The cluster decay parameter is unique for all clusters and is chosen to
have a similar value as the decay parameter of the overall resulting single
cluster.
Rule 6. All of these clusters are generated according to the same parameters.
Rule 7. For each individual echo, only a single multipath cluster realization is
generated, which is to be used for all transmitter and receiver positions
rt and rr .
Rule 8. The individual echoes and their associated clusters are coherently superimposed.
Rule 9. A frequency-dependent power decay corresponding to the bandwidth is
enforced.
From the above rules, it is obvious that the model uses some important elements of the well-known Saleh-Valenzuela (SV) model [3], among these the multiple clusters approach. Unlike SV, however, cluster arrival times are not drawn

262

Spatial aspects of UWB

Real (S21 ) (20 GHz)

104

1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
10

12

14
16
Delay (ns)

18

20

18

20

(a)

Real (S21 ) (20 GHz)

104

1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
10

12

14
16
Delay (ns)
(b)

Figure 5.6. Example for a single impulse response (NLOS, detail): (a) measurement; (b) model.

from a Poisson distribution, rather (cf. Rules 13) they are to be determined by a
dedicated algorithm that also produces the arrival times of individual echoesthis
algorithm is essential to ensure proper space-variant properties of the model. Rule
4 is required to make the individual echoes stand out over their associated multipath clusters; an individual echo plus its associated cluster may be considered the
impulse response of that path that corresponds to the individual echo. Rules 5 and
6 then follow from the simple (but also weak) assumption that impulse responses
along dierent paths are similar regarding power decay as a function of delay. Rule
7 again is related to the interpretation of each cluster as per-path impulse response:
changing transmitter or receiver position on a small scale will mainly aect each
clusters delay and phases, with only minor disturbance of its shape.
Using a tailored algorithm to generate the arrival times and amplitudes of
individual echoes and their clusters (instead of SVs homogeneous Poisson process with exponential power decay of cluster amplitudes) oers several advantages;

Thomas Kaiser et al.

263

in particular, the positions of the individual echoes and the associated multipath
clusters may be adjusted to render a realistic Doppler spectrum (in wave number domain, corresponding to angle-of-arrival domain) which is consistent with
measurement data.
5.2.3.3. Generation of individual echoes
Viewed from the receiver position, and assuming that each individual echo corresponds to a wavefront arriving from a certain direction, each of the individual
echoes may be thought of as being caused by a wave emitted from a virtual source
situated at a certain distance in that direction from the receiver, assuming freespace propagation. In other words, the transmitter may be replaced by an equivalent ensemble of virtual sources.
In contrast, in the transmitter perspective, only signals departing into certain
directions will eventually arrive at the receiver. In this view, each of these directions
may be thought of as corresponding to a virtual sink such that all signals absorbed
in virtual sinks are summed up to form the total signal arriving at the receiverthe
receiver may be replaced by an equivalent ensemble of virtual sinks.
Both pictures, and especially directions of departure and arrival, are linked by
the fact that both virtual sources and sinks are determined by interactions with the
enviroment along paths that connect transmitter and receiver.
The proposed model employs this approach and generates a set of virtual
sources from which the individual echoes are derived self-consistently for dierent
terminal positions. It is reasonable to assume that many of the strong low excess
delay individual echoes correspond to important low-order interactions, for example, direct transmission and single floor, ceiling, and wall reflections. Therefore, as
a first-order approximation, positions of potential virtual sources are generated as
image positions of the transmitter position rt up to a certain order in a generic
room with respect to given mirror positions.5 Each wall is assumed to act as a mirror, and no other mirrors are considered. Only a subset of all such virtual sources
is actually used.
Essentially, this approach represents a trivial ray-tracing type of method applied to an extremely simplistic environment model. Although this is far from a
rigorous deterministic treatment, this approach is justified by its ability to introduce reasonable spatial correlation of the modeled impulse responses.
Adaptation of the model to a specific environment requires a proper selection scheme. While general selection schemes are still subject to investigations, the
subsets for the measurement-derived scenarios under consideration here are given
in Table 5.1. Obviously, more general schemes to generate virtual source positions
beyond using images can be devised.
5This is not to say that all important interactions are specular reflections. The image approach
is used because it captures basic characteristic dimensions of generic rooms, and for its algorithmic
simplicity.

Spatial aspects of UWB

264

Parameter
N
R
B
F
d0
0
GMP
GMP,LOS
G

N2
xt
yt
zt
xr
yr
zr
X0 , X1
Y0 , Y1
Z0 , Z1
Q
K = (k)

Unit
ns1
GHz
GHz
m
dB
dB
dB
ns
dB
m
m
m
m
m
m
m
m
m

Oce LOS
2048
12.8
10
6
0.8
3.0
20.0
13.0
5.0
10.0
1.1
118
1.78
4
1.5
3.0, . . . ,4.5
1.5, . . . , 1.8
1.5
0, 5.06
0, 5.00
0, 2.75
3
171, 179, 164,
115, 173, 177

Oce NLOS
2048
12.8
10
6
1.0
2.0
16.0
N/A
20.0
9.5
1.01
123
0.91
4.39
1.5
0.15, . . . , 1.65
1.5, . . . , 1.8
1.5
0, 5.06
0, 5.12
0, 2.75
3

Oce-oce
2048
12.8
10
6
3.5
3.0
19.0
0.0
21.0
12.5
1.3
121
1.32
1.8
1.5
3.9, . . . , 4.2
2.5, . . . , 4.0
1.5
5.06, 5.06
0, 5.12
0, 2.75
3
172, 221, 123,
179, 228, 130,
173, 32, 312,
110, 174, 170

Table 5.1. Model parameters.

172, 123,173,
130, 124, 215

Number of samples of hB , H
Sampling rate in delay domain
Bandwidth of hB
Center frequency
Reference distance for individual echo power law
Path loss exponent for individual echo power law
Cluster gain with respect to associated individual echo
Additional cluster gain for LOS individual echo
Basic gain
Multipath cluster exponential decay parameter
Frequency-domain decay exponent
Normalized measurement I/Q noise variance
Nominal transmitter coordinates

Range of receiver coordinates

Covered by measurement

Mirror positions

Maximum image order per space dimension

Set of selected virtual source indices

Thomas Kaiser et al.

265

It should be noted that despite the apparent asymmetry with respect to transmitter and receiver positions (virtual source coordinates are determined from
transmitter coordinates), a virtual source determined as transmitter image just
represents an environment interaction: a reflection, which in itself is a symmetric
process with respect to receiver and transmitter positions. Virtual sources just aid
in geometric calculations and could be replaced by virtual sinks (obtained from
images of the receiver position rr ). In particular, this implies that transmission
from various transmitter positions (in a vicinity of the original transmitter position) to a given fixed receiver position can be modeled by using the same set K
(see below for a definition of K) of virtual source indices, applied to a set of virtual
sinks determined from the receiver position. If both variations of transmitter and
receiver positions are to be modeled, either virtual sources or sinks can be used
equally well.
5.2.4. Model summary
This section details all of the model equations. A complete list of all model parameters used below is given in Table 5.1, including measurement-based values
for these scenarios.
Oce LOS. Transmission within an oce of approximately 5 m 5 m 2.75 m,
line-of-sight condition between Tx and Rx.
Oce NLOS. Transmission within an oce of approximately 5 m 5 m 2.75 m,
line of sight between Tx and Rx obstructed by metal cabinet of approximately
0.6 m 1.8 m 2 m.
Oce-oce. Transmission through light plasterboard wall between two adjacent
oces of approximately 5 m 5 m 2.75 m.
The following steps are used to generate samples hd (n, rt , rr )  hB (n/R, rt , rr ),
for n = 0, . . . ,N 1, of the baseband impulse response hB .
5.2.4.1. Definition of virtual sources
Let rt  (xt , yt , zt ). A grid of virtual source coordinates is defined by


xq (m, n, p)  xt , X0 , X1 , m ,


yq (m, n, p)  yt , Y0 , Y1 , n ,

(5.1)

zq (m, n, p)  zt , Z0 , Z1 , p ,
for m, n, p = Q . . . Q, where


(x, a, b, q)  x + q(b a) + 1 (1)q

 a + b 2x

(5.2)

This defines a total of (2Q + 1)3 virtual sources, which may be numbered from

266

Spatial aspects of UWB

1 to (2Q + 1)3 by a scalar index k according to the correspondence

(m, n, p) k = 1 + m + Q + (n + Q)(2Q + 1) + (p + Q)(2Q + 1)2 .

(5.3)

The point
(m, n, p) = (0, 0, 0) kLOS 

(2Q + 1)3 + 1
2

(5.4)

corresponds to the line-of-sight path. In the following, the single index k is used
instead of m, n, p.
5.2.4.2. Generation of individual echo parameters
From the (2Q + 1)3 virtual sources generated above, only a subset K is selected to
generate individual echoes. For


rr  x r , y r , zr ,

(5.5)

the arrival time (delay) of individual echo k K is given by



k, rt , rr 

d k, rt , rr
,
c0

(5.6)

where


d k, rt , rr =

6


xq (k) xr

2

+ yq (k) yr

2

+ zq (k) zr

2

(5.7)

is the distance between virtual source k and position rr , and c0 is the free-space
propagation velocity.
If k does not correspond to the direct (LOS) path, that is, k = kLOS , the power
gain G(k, rt , rr ) is given by6


G k, rt , rr

c0
=
4d0 F

2  

d k, rt , rr
d0

 

(5.8)

where is a model parameter. Otherwise,



G k, rt , rr =

c0

4d k, rt , rr F


2

(5.9)

The amplitude gain is then given by



A k, rt , rr  G k, rt , rr

1/2

(5.10)

6Conceptually, the power law is attributed partly to the strength of the virtual source rather than

solely to attenuation due to (free-space) propagation between the virtual source and the observation
point. Assuming a certain power law is equivalent to adopting a certain dependence between virtual
source strength and distance to the observation point.

Thomas Kaiser et al.

267

5.2.4.3. Generation of associated multipath clusters

For individual echo k, multipath power values


b(k, n)2 ,

for n = 1, . . . , N 1,

(5.11)

are drawn from an exponential distribution with parameter



2
2 3
n
E b(k, n) = 10GMP /10 dB exp
,
R

(5.12)

where the delay index n corresponds to delay = n/R.

If k = kLOS , then GMP + GMP,LOS is used instead of GMP in (5.12). From
|b(k, n)|2 , the complex multipath amplitudes (normalized to the amplitude of the
associated individual echo) are
b(k, 0)  1

(associated individual echo),


2


b(k, n)  b(k, n) exp jk,n ,

n = 1, . . . , N 1,

(5.13)

with uniform i.i.d. phases k,n . By applying a DFT to b(k, n), samples B(k, l) of the
normalized and delay-less transfer function of the kth cluster are obtained:
B(k, l) =

N
1

b(k, n) exp

n=0

j2nl
,
N

l = 0, . . . , N 1,

(5.14)

where the frequency index l corresponds to the frequency

f (l)  F +

R
N

mod l +

N
N
.
,N
2
2

(5.15)

5.2.4.4. Superposition of individual echoes and

associated multipath clusters
Samples of the overall transfer function are obtained by denormalizing, shifting,
and summing over all clusters:


Hd l, rt , rr  H f (l), rt , rr = 10G/20 dB



A k, rt , rr B(k, l) exp j2 f (l) k, rt , rr .

(5.16)

kK

Note that in the sum on the right-hand side of (5.16) the term B(k, l) does
not depend on rt and rr , in accordance with Rule 7 formulated above.

268

Spatial aspects of UWB

5.2.4.5. Filtering and frequency-dependent power decay

A frequency-dependent gain decay and bandwidth limitation is applied by the substitution



 B
f (l)
for  f (l) F  ,
F
2
Hd (l)  0 otherwise.

Hd (l)  Hd (l)

(5.17)

Optionally, at this stage another filter may be applied which corresponds to a different frequency-domain window or a prescribed transmitted impulse form.
5.2.4.6. Noise addition
An inverse DFT of Hd (l, rt , rr ) yields samples


hd n, rt , rr =



N 1

j2nl
1 
Hd l, rt , rr exp
,
N l=0
N

n = 0, . . . , N 1,

(5.18)

of the normalized bandlimited baseband impulse response which may be subjected

to observation noise


hd n, rt , rr  hd n, rt , rr + sr n, rt , rr + jsi n, rt , rr ,

(5.19)

where sr (n, rt , rr ) and si (n, rt , rr ) are i.i.d. zero-mean Gaussian random variables
with variance N2 .7
5.2.4.7. Normalization
Finally, samples of the baseband impulse response hB are obtained by


hB



n
, rt , rr = Rhd n, rt , rr ,
R

n = 0, . . . , N 1.

(5.20)

Equation (5.20) assumes hB is nonzero only in an interval which is shorter than

N/R (no aliasing in delay domain), which implies a constraint for N and R. Note
that hB has dimension s1 .
5.2.5. Summary
We have constructed a model of certain ultra-wideband indoor radio channels.
The model is based upon Saleh-Valenzuela multipath clusters. Unlike SalehValenzuela, cluster arrival times are modeled in a quasideterministic fashion;
7Note that the values specified in Table 5.1 have been determined such as to reproduce the mea-

surement noise of the underlying measurement data.

Thomas Kaiser et al.

269

furthermore clusters are augmented by adding strong individually recognizable

paths. An essential new element is the use of a simple geometric model of a generic
environment to provide consistent space-variant behavior of such strong individual paths and their associated multipath clusters for small-scale movements of the
terminals. The model explicitly includes a global frequency decay of the resulting
transfer functions.
5.3. UWB antenna arrays
5.3.1. Introduction
Since the United States FCC has opened the spectrum from 3.1 GHz to 10.6
GHz, that is, a bandwidth of 7.5 GHz, for unlicensed use [6] with up to 41.25
dBm/MHz eective isotropic radiated power (EIRP), numerous applications in
communications and sensor areas are showing up. All these applications have in
common that they spread the necessary energy over a wide frequency range in this
unlicensed band in order to radiate below the limit. The resulting ultra-wideband
(UWB) systems are mainly limited by the strict power regulations, which protect
the wireless systems assigned to frequencies within the UWB band. The usage of
highly ecient UWB antennas and antenna systems with multiple antennas is an
important option in order to cope with the challenging power restriction and enables features like exploitation of spatial diversity, beamforming, and direction estimation. The basis for this is the understanding of the characteristics and specialties of UWB antenna arrays. The theory of wideband antenna arrays is the subject
of many textbooks, which trade this problem usually from a frequency-domain
point of view.
However, frequency-domain theory alone is not sucient for assessing the influence of UWB antenna arrays on the system performance. Therefore, the treatment of the transient radiation behavior of UWB antenna arrays in time domain
is given by deriving the antenna arrays transient response and analyzing its quality
measures like dispersion and ringing. The influence of the single elements transient response, the geometry of the array, the coupling between neighboring elements, and the feeding network are discussed in the context of a typical array
geometry: the linear array. The theoretical modeling is illustrated by experimental
results. The section finishes with simulation results for the application of a linear
array of Vivaldi antennas, exponentially tapered slot antennas (TSA), for timedomain beam steering.
5.3.2. Modeling UWB antenna arrays
5.3.2.1. Transfer function and impulse response
In general, the electrical properties of antennas are characterized by input impedance, eciency, gain, eective area, radiation pattern, and polarization properties [7]. For narrowband applications, it is possible to analyze these at the center

270

Spatial aspects of UWB

frequency of the system. For larger bandwidths, all of them become more or less
frequency dependent, but the straightforward evaluation of the named parameters
as functions of frequency is not sucient for the characterization of the transient
radiation behavior. One proper approach to take transient phenomena into account is to model the antenna as a linear time-invariant transmission system that
translates the exciting voltage utx at the terminal into the radiated electrical far
field erad . This system can be fully characterized by its transient response h(, ).
Assuming free-space propagation, this can be written according to (5.22) as shown
in [8]. The dimension of the antennas normalized transient response h is m/s. Together with the time-integrating convolution h, this relates to the meaning of a
normalized8 eective height:
urx
einc
= h(, ) 
,
ZL0
ZF0


erad (, ) 1
r
utx

= t
a(, ) 
r
c0
ZF0
ZL0


1
r

utx
= t
.
h(, ) 
r
c0
t
ZL0

(5.21)

(5.22)

The eective antenna gain Ge can be readily calculated from the antennas
response in frequency domain H:
Ge ( f , , ) =
etot =


2 
H( f , , )2 ,
c0

ei

1
i

ri

ri

hi (, )
c0
t

(5.23)
ZF0
utx,i .
ZL0

The voltages utx,i at the port of each radiator are determined by the properties of
the signal source, the feed network (signal distribution and matching), and the
mutual coupling between the single radiators. The presence of electromagnetic
coupling of the single radiators can be modeled using the self and mutual coupling impedance matrix Zc [9]. It can be transformed into a scattering parameter
matrix Sc , which describes the N element array as a linear N-port, see (5.24). If a
single source signal is assumed, which is distributed by a linear feed network on the
N single radiators, the array together with the feed network behaves as single antenna with a transfer function for the radiated electric field. This transfer function
can be obtained by analyzing the linear network consisting of feed network, mutual coupling network, and coupling to the radiated far-field. The latter is modeled
as a directional and polarized far-field port of the array [10]. This model includes
8Normalized means that the regarded voltages are normalized to the square root of the antennas

characteristic impedance  ZL0 and the electric field is normalized to the square root of the free-space
characteristic impedance ZF0 .

Thomas Kaiser et al.

271

all relevant eects due to mismatch and mutual coupling [11]. The impedance
matrix can be obtained from measurements or numerical simulations. Analytic
expressions are available for some special array geometries of narrowband radiators like linear arrays of dipole antennas [9]. Applying multiple sources, this type
of model has been used with great success in order to describe the properties of
multiple-input multiple-output (MIMO) antenna arrays [11]:


Sc = Z1/2
L0 Zc + ZL0

1 

1/2
Zc ZL0 ZL0
.

(5.24)

In (5.24) ZL0 denotes the diagonal matrix of the port impedances to which
the scattering parameters are normalized. Assuming a single source signal utx and
identical single elements that are identically oriented, the transfer function har of
the antenna array can be written as a convolution of the single elements transient
responses h and the time-domain array factor far :
har = h far .

(5.25)

The time-domain array factor (5.25) far is determined by the array topology, the
feeding network, and the mutual coupling of the elements. For the synthesis of
ultra-wideband arrays, the combination of the time-domain array factor and the
transient response of the single element is of high importance. Therefore, (5.25) is
analyzed for dierent parameters. It is mandatory to introduce realistic transient
responses for the single elements in order to get realistic results for the dispersive
behavior of the whole array. In order to separate the influences of the dierent
phenomena, possible simplifications and the related assumptions are shown. The
distortions that the mutual coupling introduces due to reflections and reradiation
can be neglected if the distance between the elements is large enough. The amount
of coupling also depends on the type of the single radiators: omnidirectional radiators are more likely to couple than more directive ones.
It is assumed that the mutual coupling and the antenna mismatch can be neglected in the bandwidth of interest. Then the array factor can be written as the
scalar sum of the path transfer functions Wi (t) of the feeding network for each
element i:
far =

Wi (t) t r,i .

(5.26)

i=1

The time delay r,i refers to the relative delay for the signal propagating from
the single element i to an observation point in the arrays far-field zone in direction
r. The delay r,i is the dierence between the time of arrival (TOA) to the same observation point for radiation from element i and radiation from a reference point
in the array structure, which is usually the center of the array. The weighting functions Wi (t) are the transfer functions of the active or passive feeding network of
the array. In the case of a passive power divider network, the weighting functions

272

Spatial aspects of UWB

Wi (t) are bound to

N



Wi (t)2 1.

(5.27)

i=1

In the case of an ideal dispersion free feeding network, they reduce to Wi (t) =
wi (t feed,i ). This leads to a simplified formulation for the time-domain array
factor:
far =

wi t r,i feed,i .

(5.28)

i=1

As well as a tapered amplitude distribution is used in frequency domain in

order to lower the sidelobes, the wi can be used in time domain in order to lower
the ringing for the o-boresight directions. In the following calculations for an
ideal feeding network, a uniform amplitude distribution with
1
wi =
N

(5.29)

is assumed.
5.3.2.2. Evaluation
The transient response of the antenna array can be evaluated with respect to the
peak value p and the full width at half maximum (FWHM) w0.5 of its analytic9
envelope |h+ar | [12]. The peak value indicates what maximum output voltage of
the array might be expected for a given incident field. The FWHM is a measure
in which time interval the main energy of the radiated signal is concentrated for a
given direction:


p(, ) = max h+ar (t, , ),

t

FWHM(, ) =

max

|h+ar (t,,)|= p(,)/2

min

|h+ar (t,,)|= p(,)/2

t.

(5.30)

For a more detailed analysis of the antennas transient response, see [13, 12].
The FWHM can also be used to define a time-domain far-field criterion [14]. In
the far-field region, it is assumed that for the boresight direction the propagation
delays are equal for all elements. The maximum error at a given distance r is the
dierence in propagation delay between the center of the array and the outmost
element at D/2 = (1/2)(N 1)d is
r
t =
c0

'

1+

D2
1 .
4r 2

9()+ denotes the complex analytical signal in time domain.

(5.31)

Thomas Kaiser et al.

273

The requirement that this error will be less than a fraction of the duration of the
impulse response FWHM results in
rFF >

D2
.
8c0 FWHM

(5.32)

There is up to now no standardized value for . Practical values range from =

1/8, . . . , 1/3 depending on the required accuracy. For an array with N = 4 elements, spacing d = 40 mm, and single elements for the FCC frequency range
with a typical FWHM = 115 picoseconds, this range of yields an rFF = 4.6, . . . ,
1.8 m. In frequency domain, the well-known far-field criterion for harmonic excitation with f = 10.6 GHz yields for the same array rFF > 2D2 / = 1.8 m. For
smaller r, the transient response becomes more and more dependent on the distance har = har (r, , ).
5.3.3. Uniform spaced linear array
In Figure 5.7, the orientation of a single Vivaldi element is shown together with the
employed spherical coordinate system. The xy-plane coincides with the H-plane
and the xz-plane coincides with the E-plane10 of the antenna. The single-element
Vivaldi antenna consists of an exponentially tapered slot, etched onto the metallic
layer on a Duroid 5880 substrate. The narrow side of the slot is used for feeding
the antenna, the opening of the taper points to the direction of the main radiation.
The Vivaldi is designed with a Marchand balun feeding network, which has been
developed for the frequency range of 3.110.6 GHz with a return loss better than
10 dB. The size of this antenna is 78 75 1.5 mm3 . The radiation is supported by
the travelling wave structure. The propagation velocity on the structure varies only
slightly with frequency. The CW gain is calculated from the measured directional
transient response h(, ) according to (5.23). The maximum gain of the single
element is 7.9 dB at 5.0 GHz [15].
In Figure 5.8, the geometry of the linear array is shown. The single elements
are arranged along the y-axis and are separated with equal spacing d. An observation point in the far-field region is visible with the same direction for each
element. The propagation paths are parallel rays emerging from the single elements as shown in Figure 5.8. In (5.33) the delay for element i is given relative to
the center of the array dependent on the angle :


r,i =

ri r1
1 N 1
=
i d sin .
c0
c0
2

(5.33)

Similar expressions like (5.33) are readily obtained for other geometries like a circular array or a rectangular array following similar geometrical considerations.
10The H-plane contains the magnetic field vector H and the direction of maximum radiation.

Accordingly, the E-plane contains the electric field vector E and the direction of maximum radiation.

274

Spatial aspects of UWB

z

y
x

Figure 5.7. Coordinate system with a single Vivaldi element.

d
1

N
y

rN
ri
r3
r2
r1

i = (ri r0 /c0 )

x
Figure 5.8. Geometry of a uniform linear array.

5.3.3.1. Results for the uniform linear array

For the uniform linear array (ULA), the main parameters are the number of elements N, the spacing of the elements d, the distribution of the excitation, especially
a time delay between neighboring elements .
In Figure 5.9, the measured magnitude of the transient response analytic envelope is shown for the single Vivaldi element in H-plane. Its high peak value
(pmax = 0.35 m/nanosecond) and the short FWHM = 115 picoseconds of the
transient responses envelope stand for very low dispersion and ringing. For oboresight directions | | > 30 , the antenna exhibits no unique center of radiation.
Also the amplitude decreases in these directions. For = 180 , an appreciable
backward radiation occurs with p( = 180 ) = 0.08 m/nanosecond. The transient
response for this direction is more dispersive and shows a width of FWHM = 807
picoseconds. The measured transient response of one single Vivaldi element is
taken for the simulations of the arrays transient response according to (5.25). The
measurement bandwidth was 20 GHz. For details on the measurement procedure,
see [13, 12]. Assuming a uniform amplitude distribution and negligible coupling,

Thomas Kaiser et al.

275
0

10
15
20

2
Time (ns)

180
90

1
0

180

90

|h|/(1 m/ns) (dB)

0
10
20
30
3

25
30

(deg)

(a)
0

10
15
20

2
Time (ns)

180
90

1
0

180

90

|h|/(1 m/ns) (dB)

0
10
20
30
3

25
30

(deg)

(b)
Figure 5.9. (a) Measured transient response of single Vivaldi element. (b) Simulated transient response of linear array with N = 4 Vivaldi elements with spacing d = 4 cm. The simulation is based on
the measurement of the single element.

the maximum peak value of the arrays transient response



pmax,ar = max par (, ) = N pmax,i

,

(5.34)

increases N. In the same manner as the main peak increases,

the peak am
plitude for the o-boresight directions is decreased 1/ N because for these
directions the impulse responses of the
single elements are resolved in time and
the single element i is excited with 1/ N according to (5.29) as can be seen in
Figure 5.11. Depending on the duration and the spectral properties of the excitation, this ringing can cause significant sidelobes.
Neglecting the mutual coupling, the array gain increases linearly with the
number of elements Garray = NGi . The spacing d of the array elements has a significant impact on the transient response: the maximum dimension of the array determines the maximum ringing for the o-boresight directions in terms of
FWHM as can be seen from Figure 5.12. Larger element spacing narrows the beam

276

Spatial aspects of UWB

0

10
15
20

2
Time (ns)

180
90

1
0

180

90

|h|/(1 m/ns) (dB)

0
10
20
30
3

25
30

(deg)

(a)
0

10
15
20

2
Time (ns)

180
90

1
0

180

90

0
(deg)

|h|/(1 m/ns) (dB)

0
10
20
30
3

25
30

(b)
Figure 5.10. (a) Simulated transient response of a linear array with N = 4 Vivaldi elements with
spacing d = 8 cm. (b) N = 7 Vivaldi elements with spacing d = 4 cm. The simulations are based on the
measurement of the single element.

with constructive interference of the single-element signals whereas the peak value
p() for the o-boresight directions depends on the number of elements. The antenna gain in the frequency domain is strongly aected by the element spacing
(see Figure 5.13). For element spacings with d > , frequency-selective grating
lobes emerge. Therefore, an element spacing d < 28 mm is required for a gratinglobe-free operation below 10.6 GHz. As seen before for short pulses, these grating
lobes transform into a ringing behavior.
5.3.4. Beam steering
The application of phased arrays is an established technology for steering the antenna beam electronically. For ultra-wideband applications, the steering of the antenna main beam direction is an interesting option for location aware devices in
order to scan sequentially the surrounding space and to cancel interference from
other devices. This can be achieved by introducing a mutual delay between

Thomas Kaiser et al.

277
1.2

|hn , max | (m/ns)

1
0.8
0.6
0.4
0.2
0

150 100

50

50

100

150

(deg)
N =1
N = 2, d = 4 cm
N = 4, d = 4 cm

N = 7, d = 4 cm
N = 12, d = 4 cm

2
1.8

FWHM (ns)

1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0

150 100

50

50

100

150

(deg)
N =1
N = 2, d = 4 cm
N = 4, d = 4 cm

N = 7, d = 4 cm
N = 12, d = 4 cm

Figure 5.11. Pattern for the peak value p() and FWHM w0.5 () of the array impulse response for
varying number of elements N; d = 4.

neighboring elements. The eect can be visualized best by looking at the timedomain array factor far according to (5.28). On Figure 5.14(a) it is shown for
= 0 picosecond. Each line in the diagram represents the angle-dependent delay (t r,i feed,i ). The directions where the lines cross yield the maximum
radiation. Now let feed,i = i, then the lines are shifted in time with a fixed mutual delay . Due to this shift, the crossing points of the single-element radiation
curves shift too. For small shifts, the crossing in backward direction will be attenuated by the convolution with the single-element response. For increasing , the

278

Spatial aspects of UWB

0.8

|hn , max | (m/ns)

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

150 100

50

50

100

150

(deg)
N =1
N = 4, d = 2 cm
N = 4, d = 4 cm

N = 4, d = 8 cm
N = 4, d = 16 cm

2
1.8

FWHM (ns)

1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0

150 100

50

50

100

150

(deg)
N =1
N = 4, d = 2 cm
N = 4, d = 4 cm

N = 4, d = 8 cm
N = 4, d = 16 cm

Figure 5.12. Pattern for the peak value p() and FWHM w0.5 () of the array impulse response for
varying d; N = 4.

two crossing points move towards each other until they coincide for = d/c0
at = 90 . This is the special case of an end-fire array (see Figure 5.15(a)). If
the mutual delay is larger, no crossing occurs anymore (see Figure 5.15(b)). The
directions for maximum radiation are calculated by


max = arcsin

c0
.
d

(5.35)

Thomas Kaiser et al.

279
15

12
10
f (GHz) 8

180

Gain (dBi)

10

15
5
5

90

6
4
180

90

0
5

(deg)

(a)
15

12
10
f (GHz) 8

180

Gain (dBi)

10

15
5
5

90

6
4
180

90

0
(deg)

(b)
Figure 5.13. (a) Simulated gain of a linear array with N = 4 Vivaldi elements with spacing d = 4 cm.
(b) N = 4 Vivaldi elements with spacing d = 8 cm. The simulations are based on the measurement of
the single element.

The sensitivity to changes of is obtained as

max
1
180 c0

=
.

d 1 c /d2
0

(5.36)

For example, a linear array with spacings d = 4 cm exhibits in the main beam
direction a sensitivity to a variation of of 0.43 /picosecond. This determines the
requirements on the timing accuracy of an active, true-time delay beam-steered
array. As expected from the time-domain array factor, the dispersive behavior of
the scanning array cannot be better than that of the single element for the scanning
direction max . It is worsening for the directions aside max . This is due to the
eect that the resolved transient responses of the array factor begin to lie within
the mainlobe of the single element. Therefore, they may encounter an appreciable
amplification.

280

Spatial aspects of UWB

0.5

Time (ns)

0
0.5
1
1.5
180

90

0
(deg)

90

180

90

180

(a)
0.5

Time (ns)

0
0.5
1
1.5
180

90

0
(deg)
(b)

Figure 5.14. (a) Time-domain array factor far of a linear array with N = 7 elements, spacing d = 4 cm
(solid and dashed lines). The solid lines hold for an array with N = 4 elements and spacing d = 8 cm;
= 0 picosecond. (b) N = 7 elements with spacing d = 4 cm and a delay between neighboring
elements of = 80 picoseconds.

5.3.5. Summary
The application of UWB antenna arrays has been analyzed on the basis of the arrays transient response, which provides all relevant information about their transient radiation and reception behavior. With respect to a given array topology,
simplified expressions for the time-domain array factor have been introduced. Together with measurements of the transient response of a single element, the parameters of linear arrays are analyzed. It can be concluded that the appearance
of grating lobes for sparse arrays in narrowband applications translates into additional dispersiveness for o-boresight directions. In time domain, this means
additional ringing and therefore a higher FWHM than within the main beam. The
maximum of the FWHM depends on the size of the array. The dispersive properties of a UWB array has to be taken into account for beam-steering purposes. Also

Thomas Kaiser et al.

281
0.5

Time (ns)

0
0.5
1
1.5
180

90

0
(deg)

90

180

90

180

(a)
0.5

Time (ns)

0
0.5
1
1.5
180

90

0
(deg)
(b)

Figure 5.15. (a) Time-domain array factor far of a linear array with N = 7 elements, spacing d = 4 cm
and a delay between neighboring elements of = 133 picoseconds. (b) N = 7 elements with spacing
d = 4 cm and a delay between neighboring elements of = 200 picoseconds.

the timing accuracy of the delay network single elements has to be considered. The
presented analysis is independent of the signal that might be applied to the array.
The spatio-dispersive properties of the array and their impact on a given signal can
be tested in a simulative design study with the presented tools.
5.4. UWB polarization diversity
5.4.1. Introduction
In the early days of radio communication systems, baseband transmission has been
the first form of delivering information between remote locations by means of
electromagnetic waves. Several problems related to these first systems were overcome by the invention of heterodyning and superheterodyning methods, which
introduced the modulation concept [16]. Nowadays, the advances in the field of

282

Spatial aspects of UWB

0
10
20
30

2
180

Time (ns)

90

0
0

90
180

(deg)

(a)
1
0.9
|hn , max | (m/ns)

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

150 100

50

50

100

150

(deg)
N
N
N
N
N

=1
= 7, d
= 7, d
= 7, d
= 7, d

= 4 cm,
= 4 cm,
= 4 cm,
= 4 cm,

= 0 ns
= 0.06 ns
= 0.08 ns
= 0.2 ns

(b)
Figure 5.16. (a) Simulated transient response of a linear array with N = 7 Vivaldi elements with
spacing d = 4 cm and delay between neighboring elements of = 80 picoseconds. (b) Pattern for the
peak value p() of the array impulse response for varying .

transmitter and receiver technology give to impulse radio a new important role in
the scenario of communication systems [17].
Ultra-wideband (UWB) signals are characterized by a band of several hundreds of MHz up to several GHz. The Federal Communications Commission
(FCC) issued in February 2002 a Report and Order (R&O) [6] in which defines
as UWB any signal that occupies, on an unlicensed basis, more than 500 MHz in

Thomas Kaiser et al.

283

UWB EIRP emission level in dBm

40
45
50
55

1.99

3.1

60
65
70
75
80

GPS
band
0.96

1.61

100

101
Frequency in GHz

Figure 5.17. Mask of the allowed PDS.

the band between 3.1 and 10.6 GHz and whose power spectrum density (PSD)
meets a given mask designed to introduce no interference to already deployed services, such as global positioning system (GPS) and personal communication services (PCS) digital cellular systems. The mask is shown in Figure 5.17. Early UWB
systems used the available band to transmit baseband impulses characterized by
a very narrow time duration [18]. In such systems, particular research eort has
been devoted to a proper design of the impulse shaping filter such that to concentrate as much energy as possible in the allowable bandwidth with a tolerable
PSD. An advantage of this technique is that fine resolution of multipath is feasible,
resulting in a reduced fading. This fact, along with the limited allowed transmission power due to the imposed emission constraint, makes the UWB technology
a valid candidate for indoor short-range transmission, typical of wireless personal
area network (WPAN). Preliminary and simple computations of capacity demonstrate that data rates greater than 100 Mbps at 10-meter distancethus, much
higher than already standardized short-range techniques, such as Bluetooth and
IEEE 802.11bwill be accomplished.
Recently, a dierent concept of UWB signal has emerged [1921]. Instead of
transmitting single impulses occupying a large bandwidth, the available spectrum
is divided into several bands that can be accessed separately by a single or multiple
users. Due to FCC definition, each band must be at least 500 MHz. Usually, a number of bands between 3 and 10 are taken into consideration. This method, referred
to as multiband UWB, has several advantages with respect to impulse radio UWB. A
flat PSD can be accomplished in each band, so that higher power can be conveyed
into the channel with respect to impulse-based UWB. Choosing proper positions
of the bands allows regional regulation constraint on spectrum occupancy to be
met. Also from an implementation point of view, the multiband approach is attracting since less stringent requirements can be imposed on the ADC sampling
rate and dynamic range, so that a software-defined receiver becomes feasible. Different combinations of CDMA and OFDM have been proposed for fully exploiting

284

Spatial aspects of UWB

the 500 MHz band and several band-hopping methods have been foreseen to allow
several users to access the channel as well several piconets to coexist.
Another important aspect of UWB systems is antenna technology and design
related to minimum space occupancy and portability as well as space diversity
exploitation [22, 23]. Critical points in antenna design are the required huge bandwidth and the tradeo between antenna gain, antenna dimensions, and flatness of
the antenna-to-antenna response [24]. More attention must be paid to the position of the transmitting and receiving antennas and their relative orientation since
this influences the antenna-to-antenna frequency response. For transmission between several devices and a base station, more than one antenna can be foreseen at
the base station, but only one, for dimension containment, at the portable devices.
Exploiting such diversity is an important issue of UWB systems.
In this study, a UWB transmission system able to provide a fast, local-range,
wireless connection on an unlicensed basis by using the same frequency spectrum
assigned to other narrowband communication standards is considered. Several
channel models have been presented in the literature [2527], but only a few consider the distortion eect caused by a real UWB antenna.
Based on the observation that the antenna-to-antenna frequency response
strongly depends on the mutual orientation, a technique using two transmitting
antennas with orthogonal orientation to create diversity in the received UWB signal is proposed. The receiver is equipped with a single antenna whose orientation
and position are arbitrary within the coverage area, thus resulting in a special case
of a 2 : 1 MIMO system. Space-time11 encoding is used at the transmitter to exploit
polarization diversity without increasing the signal bandwidth [28]. The performance of dierent multiple-access schemes for the downlink is investigated. It will
be demonstrated that the proposed system allows reliable detection independently
of the receiving antenna orientation and a uniform coverage of the indoor area.
This section is organized as follows. Section 5.4.2 introduces the system of
antennas used in the transmit/receive devices and reviews Alamoutis method to
exploit polarization diversity. Section 5.4.3 presents the model of the proposed
multiband UWB system, the methods for multiple access to the channel, and the
model of the transmitted signal. Section 5.4.4 describes the UWB channel model
that will be used in the simulations. Section 5.4.5 presents a derivation of the
conditional probability of error of the proposed system given the knowledge of
the channel status, whereas in Section 5.4.6 a semianalytical approach is used to
derive the average probability of error. Implementation issues are discussed in
Section 5.4.7. Concluding remarks are given in Section 5.4.8.
5.4.2. Antenna issues and polarization diversity
The study and design of UWB antennas has a key role in assessing the overall
performance of a communication system. In this section, the specific antenna
11The term polarization time would be more appropriate in this context. Here, we use the term

space time since it is customary in multiple antenna systems.

Thomas Kaiser et al.

285

0.03
0.02
0.01
0
0.01
0.02
0.03

0.06 0.04 0.02

0.02

0.04

0.06

Figure 5.18. Shape of the slot antenna designed for UWB communications (the dimensions are in
meters).

considered for our analysis is described. The antenna has been designed for impulse radio UWB systems, but can be used also for our multiband UWB system.
The complete antenna design and analysis is described in [24]. It is a slot
antenna, having a length of approximately six inches. The shape of the antenna
is shown in Figure 5.18. Although a particular type of antenna is considered, the
proposed method can be applied to any antenna shape. In [24], a tool for achieving the three-dimensional radiation pattern of the antenna is presented. Once the
relative positions and orientations of the receiving and transmitting antennas are
given, the antenna-to-antenna transfer function, computed on a preassigned frequency grid, can be evaluated. As it will be shown, these transfer functions are
strongly dependent on the relative orientations of the antennas.
To avoid performance loss due to the orientation-variant radiation pattern,
two antennas with orthogonal orientation at the transmission side are used. It will
be demonstrated that such a system is able to deliver a suciently high level of
signal independently of the orientation of the receiving antenna.
In Figure 5.19, the specific indoor environment used to simulate the proposed system is shown. It sketches a quite large environment (16 16 7 meters),
with a staircase leading to an intermediate floor. The two base-station antennas,
with orthogonal orientation, are placed on the ceiling of the room. Seven dierent possible locations and orientations of the portable UWB device are shown.
In Figure 5.20, the amplitude antenna-to-antenna transfer functions evaluated for
each position and for each transmitting antenna (denoted as H and V polarizations) are given. As can be seen, a common trend for all polarizations and all locations is that the amplitude response decreases with frequency, so that the highest
frequency bands are expected to yield the main contribution to the total bit error
rate.
Diversity can be created in the time, frequency, or space domain to combat
negative eects of the channel due to fading [29]. Also cross-polarized antennas
can be used to increase the diversity order at the receiver [3033]. In a space diversity scheme, that is, based on an antenna separation equal to ten times the wavelength, a uniform coverage is not guaranteed with real antennas due to the frequency distortion induced by orientation. With polarization diversity, this problem can be avoided.

286

Spatial aspects of UWB

0
1
2
3
4
5
6
7

2
0
4
6
4 2
8 8 6

2 4

Figure 5.19. Scheme of the indoor environment. The position of the base-station antennas and seven
dierent possible locations and orientations of the UWB device are shown.

But, diversity may decrease the resources available for user data transmission. Recently, space-time coding methods have been proposed to exploit diversity
without wasting system resources. In this study, Alamoutis scheme [34], a wellknown two-branch space-time block-coding (STBC) method, is considered. It allows two coded symbols to be transmitted in two consecutive time intervals over
two channels that, in this case, are related to the orthogonally oriented antennas.
Alamoutis method is now briefly reviewed. Let s0 and s1 be the complex symbols
to be transmitted in the consecutive time intervals (0, T) and (T, 2T). The symbols
transmitted over the antennas in the time intervals {(0, T), (T, 2T), . . .} are
)

Antenna V : cV 1 , cV 2 , . . . = s0 , s1 , . . . ,
*

(5.37)

Antenna H : cH1 , cH2 , . . . = s1 , s0 , . . . .

Let hH and hV be the complex gains introduced by the channel in the links from
the H and V antennas to the receiver. In this scheme, it is assumed that the channel coecients do not vary within the period (0, 2T). Hence, the received signal
sequence {r0 , r1 , . . .} is
)

r0 , r1 , . . . = s0 hV + s1 hH + n0 , s1 hV + s0 hH + n1 , . . . ,

(5.38)

where n0 and n1 are the AWGN contributions. The space-time decoder, which has
knowledge of the channel status, builds two decision variables by combining the
received symbols as follows:
)
* )
*

s0 , s1 , . . . = r0 hV + r1 hH , r0 hH r1 hV , . . .
2  2  2 

= s0 hV  + hH  + n0 h
V + n1 h H ,
 2  2 
3
s1 hV  + hH  + n0 h n hV , . . . .
H

As can be seen, a factor two of diversity gain is achieved.

(5.39)

Frequency response

Position 1

3
H pol.
V pol.

3.5

103

2
1

3.5

4.5

103

3.5

4.5

Frequency response

3.5
3
2.5
2
1.5
1
0.5
0

H pol.
V pol.

3.5

4.5

Frequency (Hz)
103

5.5

109

Position 6
H pol.
V pol.

2.5
2
1.5
1
0.5
0

6
109

103

Position 4

5.5

Frequency (Hz)

103

0.5

4 4.5 5 5.5 6
9
Frequency (Hz) 10

Position 5

3.5

1.5

6
109

H pol.
V pol.

2.5

5.5

Frequency (Hz)

3.5
3
2.5
2
1.5
1
0.5
0

1
0.5

0.5

H pol.
V pol.

1.5

Position 3

1.5

Position 2

H pol.
V pol.

103

2.5

4 4.5 5 5.5 6
9
Frequency (Hz) 10

2.5

Frequency response

Frequency response

103

Frequency response

4
3.5
3
2.5
2
1.5
1
0.5
0

287

Frequency response

Frequency response

Thomas Kaiser et al.

3.5

4.5

Frequency (Hz)

5.5

109

Position 7
H pol.
V pol.

3.5

4.5

Frequency (Hz)

5.5

109

Figure 5.20. Amplitude transfer functions in the dierent positions of the UWB device within the
room.

288

Spatial aspects of UWB

Table 5.2. Boundary and center frequencies (GHz) of the proposed UWB system.
Band

Low

Center

High

1
2

3.1
3.725

3.35
3.975

3.6
4.225

3
4

4.35
5.225

4.6
5.475

4.85
5.725

In the following section, the proposed multiband UWB system with polarization diversity is described.
5.4.3. System model: multiband UWB
The proposed system uses a multiband approach to generate the UWB signal. It
will be assumed that M bands can be accessed by the users or the devices connected
to the system. According to the FCC recommendations, each band must be at least
W Hz wide, where W > 500 MHz. The boundary and center frequencies of the
proposed UWB system, for which we have chosen M = 4, are shown in Table 5.2.
The multiband approach allows the channel to be shared by the dierent users
by means of frequency-division multiplex (FDM). Each band is divided into temporal slots and accessed by means of time-division multiplex (TDM). Note that a
user device can access contemporarily more than one band only if it disposes of a
multicarrier demodulator.
Due to the large frequency width, multipath propagation aects the reception of signals in each band. Orthogonal frequency-division multiplex (OFDM)
has been proposed to combat channel degradation due to a frequency-selective
channel [35, 36]. The transmitted symbol in each time/frequency tile, created by
the mixed FDM/TDM multiple access, is obtained by applying OFDM to a set
of user information bits. Since OFDM creates a further splitting of the available
spectrum, several choices for assigning frequency bins (belonging, eventually, to
dierent bands) to active users can be considered. The dierent strategies will be
discussed in the following subsection.
5.4.3.1. Multiple access in multiband UWB
Assume that each band is divided into N subcarriers. In the proposed system, N =
128 has been assumed. Hence, MN subcarriers are available to be assigned to the
total number, say K, of active users. For the sake of simplicity, a progressive index
spanning all the subcarriers, even if they belong to dierent bands, is used. Hence,
the nth subcarrier, n = 0, 1, . . . , MN 1, has the frequency
fn = fq,low + f p,

(5.40)

where n = qN + p, fq,low is the lower bound frequency assigned to the qth band,
and f = W/N is the subcarrier spacing.

Thomas Kaiser et al.

289

It is assumed that each subcarrier is exclusively assigned to each user, that is,
it cannot be shared between two, or more, users. Let
)

I(k, l) = ik,l,m , m = 0, 1, . . . , MI 1 ,

k = 0, 1, . . . , K 1, l = 0, 1, 2, . . . ,
(5.41)

be the subset of carriers assigned to the kth user in the lth symbol time. The number of carriers MI assigned to each user in each symbol time depends on the type
of multiple access that is used. Some simple methods of accessing the channel are
now illustrated.
In a purely FDM access, all users transmit contemporarily using dierent subcarriers. Assuming that evenly spaced subcarriers are assigned to dierent users,
the set of assigned subcarriers can be expressed as
4

IFDM (k, l) = k + mK, m = 0, 1, . . . ,

MN
1 .
K

(5.42)

In a purely TDM access, time slots are assigned cyclically to dierent users,
but, in a slot, all subcarriers are assigned to a user, that is,

m,

m = 0, 1, . . . , MN 1, l = k, k + K, k + 2K, . . . ,

l = k, k + K, k + 2K, . . . .

ITDM (k, l) =

(5.43)

In a mixed FDM/TDM access, a set of adjacent subcarriers is assigned to a user

at each slot time, but the set of assigned subcarriers changes at each time interval.
We have
4

IFDM TDM (k, l) = m + rem(k + l, K)

MN
MN
, m = 0, 1, . . . ,
1 ,
K
K
(5.44)

where rem(p, q) is the remainder of the division between the integers p and q.
With a suitable choice of system parameters, the subcarriers assigned to a user do
not span more than one band at each time interval, and, as a particular case, all
subcarriers within a band may be assigned to the same user.
A graphical representation of the three access methods used in this study is
shown in Figure 5.21.
5.4.3.2. Signal model
(l)
be the data
In this section, the model of the transmitted signal is presented. Let dk,n
symbol transmitted by the kth user in the lth symbol time over the nth subcarrier.
(l)
In the following, it will be assumed that dk,n
belongs to a given constellation of
(l)
(l)
symbols. After Alamoutis STBC, two sequences, {cH,k,n
} and {cV ,k,n }, are created

Spatial aspects of UWB

Frequency

290

Time
User 1
User 2

User 3
User 4

Frequency

(a)

Time
User 1
User 2

User 3
User 4

Frequency

(b)

Time
User 1
User 2

User 3
User 4
(c)

Figure 5.21. Assignments of the time-frequency tilings with the dierent access methods: (a) TDM,
(b) FDM, and (c) mixed TDM/FDM.

Thomas Kaiser et al.

291

and transmitted over the H and V polarizations of the antenna. By using the definitions given in the previous section, we have
(l)
j2 fn t
s(l)
X,k,n (t) = cX,k,n e

(5.45)

for t [lT, (l + 1)T], where T is the symbol duration time and X denotes either
the H or the V antenna. Hence, the signal transmitted over the qth band by the X
antenna is
S(l)
X,q (t) =

(q+1)N 1

K
1

(l)
cX,k,n
e j2 fn t .

(5.46)

k=0 nI(k,l), n=qN

Consider now the eect of the channel on the transmitted signal. It can be modeled
as the convolution of the antenna-to-antenna transfer function with a random
process taking into account dierent channel characteristics (multipath, fading,
time variability, etc.). The slicing of the spectrum into several narrow subbands
makes the channel in each subcarrier flat. Assuming a slow motion of the mobile
user, the eect of the channel on the nth subcarrier within the lth time interval is
modeled as a complex constant that will be denoted as
(l)

(l) jX,n
h(l)
.
X,n = X,n e

(5.47)

A description of the channel model and of the statistics assumed to derive the
variable h(l)
X,n is given in the next section.
The signal received by the user, equipped with a single antenna, in the nth
band and in the lth symbol time interval is given by
R(l) (t) =

M
1

q=0

(l)
S(l)
H,q (t) + SV ,q (t)

K
1

(l)

jH,n (l)
(l)
cH,k,n e j2 fn t
H,n e

(5.48)

k=0 nI(k,l)

K
1

(l)

jV ,n (l)
(l)
cV ,k,n e j2 fn t + w(t),
V ,n e

k=0 nI(k,l)

where w(t) is AWGN.

To extract the information delivered by the nth subcarrier, the receiver performs the correlation with e j2 fn t , that is,
(l)
=
rk,n

 (l+1)T

j2 fn t
R(l)
dt
m (t)e
lT
(l)
(l)
jH,n (l)
jV ,n (l)
(l)
cH,k,n + (l)
cV ,k,n
H,n e
V ,n e

(5.49)
(l)
+ wk,n
.

292

Spatial aspects of UWB

By applying Alamoutis decoder to two consecutive received symbols, we have

9
d
9
d

(l)
k,n

(l+1)
k,n



(l)
(l)
(l) (l)
(l+1) (l) jH,n
= rk,n V ,n e jV ,n + rk,n
H,n e
,


(l)
(l)
(l) (l)
(l+1)
jV ,n
= rk,n H,n e jH,n rk,n
(l)
V ,n e

(5.50)

and by using (5.49) in (5.50) we have

(l)
k,n

(l+1)
k,n

9
d
9
d




(l)
V ,n
(l)
V ,n

2
2

2 

2 

+ (l)
H,n
+ (l)
H,n

(l)

(l)
(l) (l) jV ,n
(l+1)
dk,n
+ wk,n
V ,n e
+ wk,n
(l)

(l+1)
(l) (l) jH,n
(l+1)
dk,n
+ wk,n
H,n e
wk,n

(l)

jH,n
(l)
,
H,n e

(l)

jV ,n
(l)
.
V ,n e

(5.51)
In the next section, the model used to derive the coecients h(l)
X,n is presented,
whereas in the following section the probability of error of the system is derived.
5.4.4. The UWB channel
Models for describing the eect of the channel are usually extracted from sets of
measurements conducted in dierent indoor/outdoor environments [2527, 37].
The set of measurement should be suciently large so that a general model, able
to capture the main characteristics of the channel, can be derived.
In a wireless connection, the received signal is characterized by the superposition of replicas of the transmitted signal due to the reflections and diraction of
the EM wave around the objects (walls, furniture, persons, etc.) that are present in
the environment. Hence, the received signal may be either attenuated or amplified
according to destructive or constructive arrival of dierent echoes. The dierent
replicas can be discriminated with a time resolution approximately equal to the
inverse of the bandwidth of the system. In narrowband systems, a high number of
replicas fall into the bin representing the time resolution of the system. Hence, for
the central limit theorem, the received signal is assumed multiplied by a complex
gain distributed as a Gaussian random variable, that is, its amplitude is Rayleigh
distributed. Given the transmission environment, the model could be derived not
only on a statistical basis but also from geometrical considerations. For example,
if it is known that the transmitter and the receiver are in line of sight (LOS), then
the direct path is assumed to prevail over the others paths and the global received
signal becomes Ricean distributed.
The model used for narrowband signals is not valid for UWB signals. The time
resolution of the system is so high that the number of paths impinging the receiver
antenna in a time-resolution bin is not so large to justify the use of the central limit
theorem. The IEEE 802.15.3a task group [38], aiming at defining a standard for a
wireless interface based on UWB, has evaluated how a certain number of indoor
wireless channel models fitted the set of UWB channel measurements. Among
available models, that of Saleh and Valenzuela [37] has been found particularly

Thomas Kaiser et al.

293

suitable to describe the behavior of the UWB channel. In such a model, the rays
of the transmitted signal are characterized by arriving at the receiver grouped in
clusters. A power delay profile of clusters and of rays within a cluster must be
defined as well as statistics for interarrival times between clusters and interarrival
time of rays within a cluster.
Based on this model and on a set of propagation measurements, Foerster
et al. [26] introduced a modified model for UWB signals. The impulse response of
the channel is assumed equal to

h(t) = Y

J
P

j,p t T j j,p ,

(5.52)

j =0 p=0

where T j is the arrival time of the first ray of the jth cluster, j,p is the arrival time
of the pth path within the jth cluster (measured starting from the arrival of the
first ray), j,p is the complex gain of each path, Y models the shadowing of the
link. The interarrival times between clusters (T j T j 1 ) and the interarrival times
between rays within a cluster ( j,p j,p1 ) are exponentially distributed with arrival rates and , respectively. The power delay profiles of clusters and of rays
within a cluster are exponentially distributed with decay factors and , respectively. The path complex gain can be written as j,p = j j,p , where j and j,p are
distributed as log-normal random variables, that is, 20 log10 j and 20 log10 j,p are
normal random variables, that is, N (0, 12 ) and N (0, 22 ), respectively. Hence, also
j,p is log-normally distributed, that is, 20 log10 j,p is N ( j,p , 12 + 22 ), where j,p
is computed by normalizing the path energy with respect to the first arrived path
energy. This normalization takes into account the power delay profiles with decay factors and and the values of the realization of the interarrival times (see
[26] for more details). Finally, Y is a log-normal distributed shadowing, that is,
20 log10 Y is N (0, Y2 ).
In the present study, a mixed deterministic/statistic modeling of the channel
is used.
It is assumed that an LOS link between the transmitter and the receiver exists. This link is modeled as a deterministic transfer function depending on the
frequency as well as on the relative orientation of the transmitting and receiving antennas. The seven positions and orientations sketched in Figure 5.19, whose
transfer functions are plotted in Figure 5.20, have been taken into consideration in
the simulations.
This deterministic model, however, does not include the several echoes received after reflections and diraction around objects in the room, that is, the
non-LOS (NLOS) part of the received signal. Dealing with such echoes in a deterministic way aects the generality of the environment and its modeling. Hence,
a statistic modeling for this NLOS component has been considered. For this purpose, the model described in [26] and previously reviewed has been used. The
values of the parameters of the models that have been used in the simulations are
taken from [26] and are reported in Table 5.3 It should be noted that, due to the

294

Spatial aspects of UWB

Table 5.3. Parameters of the channel model.

0.0667

(1/ns)

2.1
14
7.9

(1/ns)

1
2
Y

3.4
3.4
3

(dB)
(dB)
(dB)

nature of the UWB transmission, the parameters of this model do not refer to a
baseband representation.
The LOS and the NLOS components are linearly combined according to a
parameter that allows us to vary the percentage of energy deriving from the LOS
component. Let aX,n be the transfer function of the LOS link from the X (either
H or V ) to the receiving antenna, measured at the nth subcarrier frequency. This
function depends on the positions of the receiver and on the relative orientation
of the transmitting/receiving antennas but does not depend on time. Samples of
(l)
be a realization of the
this function are plotted in Figure 5.20. Moreover, let gX,n
coecients obtained from the NLOS model at the nth subcarrier frequency, valid
for the X antenna link and at the lth symbol time. The coecients h(l)
X,n used in
Section 5.4.3.2 are given by


(l)
h(l)
X,n = aX,n + 1 gX,n bn .

(5.53)

The function bn is the antenna-to-antenna transfer function when the transmitting and receiving antennas are favorably coupled and at a distance of three meters
(the minimum distance in the walk depicted in Figure 5.19). This function has
been introduced to simulate the reception of the NLOS component by our specific
antenna system. In fact, it yields a frequency shaping to the NLOS component dependent on the antenna characteristics and similar to that encountered by the LOS
component.
A feature that must also be considered for a realistic simulation of our system is the correlation between the signals transmitted from the two orthogonally
disposed antennas. Since the correlation coecient between the two channels depends on several factors, quite dicult to be modeled, in this study the two ex(l)
and gV(l),n , are either
treme cases, in which the channels, that is, the variables gH,n
completely correlated or completely uncorrelated, are considered. The experimental results will demonstrate that the proposed system yields a gain with respect to
a single antenna system in both cases.
5.4.5. Probability of error
The probability of error of the UWB communication system depends on the
signalling used and on the status of the channel. Since the multiband UWB

Thomas Kaiser et al.

295

communication link provides M macrobands, each one divided into N OFDM

subcarriers, the probability of error for a single subcarrier is first analyzed. By assuming that a single antenna is used, either H or V , and that the modulation is
binary antipodal, the expression for the error rate as a function of the received
SNR b,n at the nth subcarrier is given by


(l)
P2 b,n

'6

(
(l)
2b,n

=Q

(5.54)

where



2 E
 b

(l)
(l)
b,n
= hX,n 

(5.55)

N0

and h(l)
X,n is the channel coecient seen from the X antenna to the receiver at the
generic nth subcarrier frequency, given in (5.53). This channel coecient takes
into account both the antenna orientation eect as well as the multipath nature of
the UWB channel.
In the case of space-time transmit diversity with order two at the receiver,
the probability of error measured at a single OFDM subcarrier is again given by
(l)
(l)
) in (5.54), where now, however, b,n
is given by
P2 (b,n


2

2  E

 (l) 
 (l) 
(l)
b,n
= hH,n  + hV ,n 

N0

(5.56)

This result is true if the channel coecients from the two antennas are independent of each other and if they do not vary during two consecutive bit intervals.
The probability of error seen by the kth user of interest is the average over all
the subcarriers. Let
%

(l)
(l)
(l)
k = b,n1 , . . . , b,nMI

&T

n1 , . . . , nMI I(k, l),

(5.57)

be the vector of SNRs over the MI subcarriers assigned to the kth user at the lth
time interval. MI and I(k, l) are defined in Section 5.4.3.1.
The desired probability of error is then given by


P2 (l)
=
k

1
MI

(l)
P2 b,n
.

(5.58)

nI(k,l)

In order to obtain the unconditional error probability, P2 ((l)

k ) has to be averaged over the joint probability density function of (l)
,
that
is,
the following
k
integral has to be calculated:12


P2 =

 

(l)
(l)
P2 (l)
k p k dk .

(5.59)

12The dependence on the time index l is here dropped, because the probability of error is assumed

not to be dependent on the instant of observation.

296

Spatial aspects of UWB

Unfortunately, a closed form of the integral in (5.59) is not easy to achieve due to
(l)
the complex formulation of the random variables b,n
that depend instantaneously
(l)
on the channel complex value hX,n seen between the transmitter and the receiver.
Due to the problems related to solving (5.59), a semianalytical approach,
based on Monte Carlo simulations, has been used to evaluate the total average
probability of error. A set of channel coecients {h(l)
X,n } for each antenna was generated, so that (5.58) could be computed. Then, the total average probability of
error P2 is approximated by
P2 =

N
1 r  (l) 
P2 k,z ,
Nr z=1

(5.60)

(l)
where (l)
k,z is the zth realization of the vector k and Nr is the number of Monte
Carlo trials.
The numerical results are reported in the next Section 5.4.6.

5.4.6. Results
The performance of the proposed system has been evaluated by means of a semianalytical approach. Nr = 100 independent realizations of the UWB channel model
described in Section 5.4.4 have been generated, considering all the seven positions
and orientations reported in Section 5.4.2. Hence, the conditional probability of
error relative to each channel realization relying on the SNR expressed in (5.55)
for the single-antenna case and on (5.56) for the polarization diversity case, respectively, was evaluated. Finally, the overall performance of the dierent systems
by averaging the probability of error over the Nr channel realizations was obtained.
Four systems based on dierent antenna configurations and channel hypotheses are considered. The first two systems, denoted as H pol . and V pol ., refer to the
cases in which only one of either the H or V antenna is employed at the transmitter. The name HV corr . denotes the system using two antennas with orthogonal
orientations and Alamoutis coding, assuming that the NLOS components of the
(l)
and gV(l),n , are completely correlated. Conversely, with
two channels, that is, gH,n
(l)
and gV(l),n are completely
HV ind ., an analogous two-antenna system, where gH,n
uncorrelated, is denoted.
The results of simulations considering the TDM access scheme, described in
Section 5.4.3.1, are shown in Figure 5.22. In this case, we suppose that LOS path
carries the 80% of total transmitted power, that is, = 0.8. The error probability
relative to each receiver position and orientation as well as the error probability
obtained by averaging over all the seven positions are shown. As can be seen, the
proposed system gains almost 5 dB in the average.
As can be seen, the performance of the single-antenna systems varies widely
for the dierent receiver positions and antenna orientations as well as considering either the H or V antenna at the transmitter. The proposed polarizationdiversity schemes allow us to achieve a performance very close to that of the best

Thomas Kaiser et al.

Position 1

100
101
102
103

H pol.
V pol.
HV corr.
HV ind.
10

Position 2

100

BER

BER

297

101
102

20

103

30

Eb /N0 (dB)
Position 3

101
H pol.
V pol.
HV corr.
HV ind.

102
103

10

101
H pol.
V pol.
HV corr.
HV ind.

102

20

103

30

Position 5

BER

BER

103

H pol.
V pol.
HV corr.
HV ind.
0

10

H pol.
V pol.
HV corr.
HV ind.

103

30

Eb /N0 (dB)

BER

BER

103

H pol.
V pol.
HV corr.
HV ind.
0

10

Eb /N0 (dB)

30

101
H pol.
V pol.
HV corr.
HV ind.

102

20

20

Average

100

101
102

10

Eb /N0 (dB)

Position 7

100

30

101
102

20

20

Position 6

100

101
102

10

Eb /N0 (dB)

Eb /N0 (dB)
100

30

Position 4

100

BER

BER

100

H pol.
V pol.
HV corr.
HV ind.
10
20
Eb /N0 (dB)

30

103

10

20

30

Eb /N0 (dB)

Figure 5.22. Performance of single-antenna (H pol. and V pol.) and polarization-diversity (HV corr.
and HV ind.) systems. The dierent positions/orientations described in Section 5.4.2 are considered.
The right bottommost figure plots the average of the other curves. TDM access scheme with LOS carrying 80% of the total transmitted power has been used.

298

Spatial aspects of UWB

transmit antenna, that is, the antenna showing the most favorable amplitude response with respect to the current receiver position and orientation. As a result,
the average error probability of the polarization-diversity schemes outperforms
that of both the H pol. and the V pol. schemes. Moreover, the behavior of the
polarization-diversity scheme appears to be independent of the correlation between the NLOS components of the two channels. The diversity gain showed in
the average performance is to be ascribed mainly to the dierent frequency responses due to the orthogonal orientations of the two transmit antennas. It is
worth noting that this fact allows us to avoid any space separation between the
two antennas.
The independence of the polarization diversity gain from channel correlation as well as the overall system performance appear to be strictly related to the
percentage of power carried by the LOS path. In Figure 5.24, the average error
probability curves of the four proposed systems using the TDM access scheme
are compared considering two dierent LOS conditions. In particular, the cases in
which LOS path carries either 20% or 80% of total transmitted power are taken
into consideration.
The benefits of polarization diversity with respect to the single-antenna system are evident for both LOS conditions. Moreover, it is remarkable that even
for the 20%-LOS case the performance gain obtained with polarization diversity
seems to be independent of the correlation of channel multipath components. On
the other hand, by comparing Figures 5.23(a) and 5.23(b), it is apparent that the
overall system performance depends on the LOS power, that is, there is a slight
performance improvement when LOS power increases.
The last issue that has been considered is the eect of the access scheme on the
overall performance of the system. In Figure 5.24, the error probability obtained
with the TDM, FDM, and FDM-TDM access schemes are compared, considering
the HV ind. system and the 80%-LOS case. As can be seen, the performance of the
proposed system is independent of the access scheme. This result can be ascribed
to the fact that we have tested an uncoded system. The eect of the dierent access
methods is to distribute the errors in dierent ways over a group of symbols and
this has a small impact on the average bit error rate of the system, unless some
channel coding is used to protect information from channel errors.
5.4.7. Implementation issues
This section mainly focuses on the dependence of the frequency response of a
practical UWB antenna from the mutual orientation of the transmitter and the
receiver. However, the real communication environment and the terminal design
present several open problems and technological challenges.
5.4.7.1. Signal processing
The first and most important issue is related to the implementation of ultra-wideband signal processing stages in the terminals. The impulse radio [18] approach

Thomas Kaiser et al.

299
100

BER

101

102

103

10

15
20
Eb /N0 (dB)

H pol.
V pol.

25

30

25

30

HV corr.
HV ind.
(a)

100

BER

101

102

103

10

15

20

Eb /N0 (dB)
H pol.
V pol.

HV corr.
HV ind.
(b)

Figure 5.23. Performance comparison between dierent LOS conditions. LOS path carries (a) 80%
and (b) 20% of the total power.

provides a simple way to implement UWB transmission but it provides a limited

control over the spectral content of the transmitted signal, giving rise to coexistence problems with other services. The UWB multiband paradigm oers a good
spectrum utilization and sophisticated reception schemes, but requires very complex RF analog stages.
To obtain a good control over the UWB radio signal, digital signal processing
techniques are envisaged. The analog-to-digital conversion can be performed in

300

Spatial aspects of UWB

100

BER

101

102

103

10

15

20

25

30

Eb /N0 (dB)
TDM-LOS 80%
FDM-LOS 80%
FDM-TDM-LOS 80%
Figure 5.24. Performance comparison among dierent multiple access methods. The HV ind. system
with the LOS carrying the 80% of the total power is considered.

UWB
antenna
Tx baseband
signal processing

DAC

SSPA

Rx baseband
signal processing

ADC

LNB

Digital domain
> 12 Gsps

Analog domain
(3-6 GHz)

Figure 5.25. Single-band UWB conversion stages.

two ways for the multiband approach:

(i) single-branch conversion stages;
(ii) multibranch conversion stages.
The first case, represented in the Figure 5.25, employs a single ultra-wideband
direct conversion from the digital domain to the analog one and vice versa. The
approach is subject to the availability of fast conversion devices and presents the
advantage of a perfectly coherent signal synthesis in the UWB frequency bands
(36 GHz).
In the second case, shown in Figure 5.26, the RF section is subdivided into 4
branches of 500 MHz each. The requirements for ADC and DAC devices are less
stringent in this case but the up-down conversion sections of the branches need to

Thomas Kaiser et al.

Tx baseband
signal processing

Rx baseband
signal processing

301

DAC

f1

DAC

f2

DAC

f3

DAC

f4

ADC

f1

ADC

f2

ADC

f3

ADC

f4

UWB
antenna

SSPA

LNB

Analog domain
4 branches
500 MHz each

Digital domain
4 1 Gsps

Figure 5.26. Multiband UWB conversion stages.

be properly synchronized in order to obtain a coherent synthesis of the transmitted

UWB signal.
5.4.7.2. Antennas location
Another important implementation issue is related to the positioning of antennas
on the communication environment. As stated in the previous sections of this
chapter, a polarization diversity is introduced to mitigate the frequency-variant
radiation pattern. This approach eliminates to need of a minimum separation of
the antennas involved in the space-time coding scheme, thus allowing a colocation
of orthogonally polarized sensors.
5.4.7.3. Power management
The main target of the orientation diversity scheme is to provide a quasiconstant
quality of UWB radio coverage in an indoor environment. The transmitted power
is balanced between the two orthogonally oriented transmitting antennas in order
to provide a constant RF transmitted power.
As in many other wireless systems, a closed-loop power control must be implemented, in order to compensate the fading eect induced by mobility of the
terminals and the scatterers and to limit the interference to other services in the
same bandwidth. It is worth noting that such a power control is not able to mitigate the frequency distortive eect described in the previous sections and does not
interfere with the proposed diversity schemes since it does not modify the power

302

Spatial aspects of UWB

balance between the two antennas. For stability reasons, it should be realized with
a slow time-constant power measurement scheme, averaging the fast variation of
the combined received power.
5.4.8. Summary
The performance of a multiband UWB system with polarization diversity is investigated. The scope of the present work is to avoid the degradation of the frequency response between the transmit and receive antennas due to their mutual
orientation. A mixed deterministic and statistic UWB channel model is proposed,
which includes both LOS and NLOS components. Polarization transmit diversity is used to guarantee high performance at the receiver independently of the
antenna-to-antenna orientation. Three dierent multiple-access schemes are considered, FDM, TDM, and hybrid FDM-TDM. An improvement with respect to
a single-antenna system is obtained for all access schemes without the need of a
minimum separation of the transmit antennas. Some implementation aspects are
analyzed showing the advantages of the multiband approach which reduces signal processing complexity and makes the software radio concept feasible also for
UWB applications. Finally, it is worth noting that the methodology used in this
work can be applied to dierent environments and antenna shapes.
5.5. Spatial diversity
This work investigates spatial diversity in the context of ultra-wideband (UWB)
communication systems. For this purpose, several spatial correlation functions are
defined, which may help predict the performance of a diversity-based multisensor
receiver in a given space-variant channel, and for a given pulsed or multiband
modulation scheme. According to the results, full correlation of output signals,
or of the contrary very fast decorrelation, can be obtained, depending on the nature of the receiver operating principle. Maps of spatially variant signals without
or with diversity combining schemes are shown for simulated discrete multipath
channels, and also for measured channels in a few environments. Finally electromagnetic issues relative to the coupling between close UWB sensors are considered.
5.5.1. Introduction
5.5.1.1. Context of the work
Spatial diversity is a well-known method to combat fading, which has been pioneered as early as the late twenties [39, 40]. The word diversity itself first suggests that radio systems employing such a technique intend to take advantage of
the signal variability from one sensor to another, in order to improve the link robustness. A second simple idea is that by using several antennas at reception, we

Thomas Kaiser et al.

303

Output
signal

PAM

+1
+1
1
+1
+1
1
1
1
1
+1

+1

+1

Noise

PPM

DS-CDMA

Noise

IFFT

S/P

(a)

..
.
Output
signal

(b)
Figure 5.27. Matched-filter correlation and detection for (a) pulse-based and (b) MBOFDM physicallayer schemes.

may try to capture more energy from the ambient medium, thereby improving
the signal-to-noise (SNR) or signal-to-interference (SIR) ratio. Both contribute to
performance improvement, let it be through the range, the throughput, or more
generally speaking through network capacity. However, since the early days of
diversity, it has been appreciated how far smart antenna approaches could go, well
beyond such simple ideas as expressed above, and could concern all sorts of radio
signal processing issues involving multiple antennas. In the present work, we will
address the as yet little investigated issue of diversity in ultra-wideband systems,
and will limit ourselves to the two time-honoured eects mentioned above. The
paper attempts to define spatial diversity descriptors which are suited to the way
the received signals are processed in a specific receiver. For this purpose, some of
the main candidate modulation schemes for standardization are considered, operating either in the time domain (pulsed schemes) or in the frequency domain
(OFDM schemes).
We consider the linear system model depicted in Figure 5.27, where a singleantenna transmitter communicates with a multiple-antenna receiver through the
radio environment. After going through antenna and front-end filters, the signal
is correlated with a reference template or FFT processed at baseband, resulting
in an analog output signal. Spatial diversity descriptors are built from this signal,
which de facto is template dependent. In a digital architecture, this signal is the
decision variable, but digital quantization issues are here irrelevant. Such a model

304

Spatial aspects of UWB

is quite general, and is suited to various UWB physical-layer schemes as will be

seen below.
5.5.1.2. System model
The output signals can very generally be written as the result of a correlation operation between the received signal and a reference waveform (template):


S() =

F r(t) ref(t )dt,

(5.61)

where the dependence expresses the clock synchronisation dependence, and F()
is a linear operator applied to the received signal. According to the choice of the
template waveform, a wide selection of basic UWB physical-layer schemes can be
covered:
(i) pulse-amplitude modulation (PAM), where F() = Id (), and where the
template is a single pulse, for example, reproducing the temporal shape
of the received signal for an ideal dispersionless channel (matched filtering);
(ii) binary pulse-position modulation (BPPM), where F() = Id (), and
where the template is composed of two pulses of opposite polarities, separated by a (small) time delay;
(iii) direct-sequence code-division multiplexing access (DSCDMA), where
F() = Id (), and where the template is composed of a repetition of a
single pulse, multiplied by a PN sequence of 1;
(iv) switched multiband (SMB), where F() = Id (), and where there are
now as many template waveforms as the number of bands, each template
being chosen in order to comply with the spectral requirements of its
own band;
(v) multiband-OFDM (MBOFDM), where F() = FFT(), which is equivalent to a many-tone filter bank per band, with a template vector of samplers associated to each band at the frequency-domain FFT output.
The signals can be either complex (baseband) or real (true signals) without restriction to generality. In what follows, complex signals will be used for multiband
schemes, and real signals for pulse-based schemes.
5.5.1.3. Definition of spatial diversity descriptors
Since spatial diversity may be a useful approach to improve radio systems performance, it appears necessary to define descriptors which might be helpful to
predict the potential improvement expected from spatial diversity. In the following, we will make a substantial use of normalized intersensor correlations, which
relate the output signal on one antenna to its value on another antenna as follows:


E S1 S
=  n2  ,
E S1 

(5.62)

Thomas Kaiser et al.

305

where E() stands for the expectation operator, and S1 and Sn are output signals
upstream of the decision stage, for sensors 1 and n, respectively.
From a practical point of view, 1n gives us an indication about what we can
expect for the signal output issued from another sensor, in comparison to what it
is on a given sensor, in a statistical sense for a given statistical set (environment,
scenario). From the definition quoted above, we define several ways to compute
intersensor correlations, depending on the way the receiver is operated.
(i) In the case of SR (synchronized reference), it is assumed that optimum
synchronisation is achieved on the reference sensor (sensor 1), resulting in the
maximal value of the correlator output for this sensor. For sensor n, it is assumed
that the clocking times are the same as sensor 1:
 

SR =

 
 2 
E S1 

E S1 max 1 Sn max 1



(5.63)

with Sk () = rk (t)ref(t )dt and max 1 = arg(max S1 ()).

(ii) The case of SS (synchronized sensors) is similar to SR , except that the
clocking times on sensor n are adjusted in order to maximize the correlator outputs also on this sensor:
 

SS =


 2 
E S1 

E S1 max 1 Sn max n



(5.64)

with max 1 = arg(max S1 ()), max n = arg(max Sn ()).

(iii) RD (RAKE decorrelation) assumes perfect (total) RAKE combining on
the reference sensor 1, achieved by filtering with the conjugate transpose of the
channel transfer function. Signals from other sensors are subjected to the same
filter, which is expected to decorrelate signals very eectively [41]. Using (5.61),
we have FRD,1 (rk (t)) = h (r1 , ) rk (t), where stands for the convolution, and
 

RD =

 
 2 
E S1 

E S1 max 1 Sn max 1



(5.65)

with Sk () = FRD,1 (rk (t))ref(t )dt, max k = arg(max Sk ()).

(iv) RC (RAKE combining) assumes perfect (total) RAKE combining on each
individual sensor. We now have
 

RC =



 2 
E S1 

E S1 max 1 Sn max n



with Sk () = FRD,k (rk (t))ref(t )dt, max k = arg(max Sk ()).

(5.66)

306

Spatial aspects of UWB

(v) In the case of MBOFDM, the FFT stage output is a vector of complex
signals. We thus naturally define the single-band correlation as

SBO



=  2  ,
E S1 

E S1 Sn

(5.67)

which is subsequently averaged over all subbands to give MBO .

(vi) Switched multibands do not dier from pulsed-based modulations, except through the pulse duration which is longer. Since the mean frequency is
switched at regular times in order to cover the whole bandwidth, however, a specific intersensor correlation SMB is defined by averaging the correlation over all
subbands.
(vii) In addition, a narrowband correlation NB = E(H1 (F)Hn (F))/E(|H1 (F)|2 )
has been used, for the sake of comparison. F is the mean received signal frequency,
and Hk is the frequency-domain transfer function between transmitter and receiver sensor k, including the channel.
5.5.2. Signal waveforms and channel model
5.5.2.1. Signal waveforms
In the present work, we assume ideal antennas at both transmitter and receiver,
which means here that these antennas are frequency independent when operated
in transmission with no gain (isotropic antennas). It is well known however that
any antenna obeys the relation Aem = 2 G/4 between the gain G and the eective
area in reception, for an impedance-matched load and a wavelength . This implies
such an ideal antenna to act as an integrator in the time domain, and also that
the transmitted and received signals dier [42]. As a representative example, a
pulsed waveform whose 3 dB bandwidth is close to 3 GHz is shown in Figure 5.28,
and has been chosen to respect the FCC mask. In a noise-limited receiver with
a perfect channel, the optimal template for the correlator is the received signal
itself, therefore we choose this same signal for the template in the computation of
correlator outputs for intersensor correlations.
5.5.2.2. Channel model
We assume a space-variant discrete channel model. For a given position of the
receiving and transmitting antennas, the channel is described as a discrete sum
of plane wave multipaths, each characterized by its delay, its amplitude, and its
direction of arrival (DOA). The channel impulse response (CIR) can be written as
a function of delay and receiver antenna position r as follows:
h(r, ) =

Ai i (r)

with i (r) = i0

rui
,
c

(5.68)

Thomas Kaiser et al.

307
0
Amplitude (dB)

FCC
10
20
30
40
50

10

Frequency (GHz)
Received
Transmitted
(a)
10

Signal

5
0
5
10

0.2

0.4
Time (ns)

0.6

0.8

Received
Transmitted
(b)
Figure 5.28. (a) Spectra of the transmitted and received signals for ideal antennas, in comparison with
the FCC mask; (b) temporal waveforms.

where i0 is a fixed path delay, ui is a unitary vector along the DOA at the receiver,
and c is the velocity of light.
As compared to usual wideband channel models, the present model has the
following dierences.
(i) When we deal with real and not complex signals, the CIR has to be real as
well and not complex. This means in particular that the path amplitudes
are real, positive or negative. According to physical intuition, the signs
and amplitudes of the received paths are related to the elementary events
experienced by the pulsed waves, that is, specular metallic reflections,
diraction, transmission, and so forth.
(ii) We expect a much greater number of paths, due to the extremely large
bandwidth. The multipath density has been experimentally ascertained
[4350].

308

Spatial aspects of UWB

The statistics and related parameters of path amplitudes, delays, and DOAs are obviously a crucial issue for the pertinence of the channel model. For the purpose of
the present work, we make simplifying assumptions, which should not dramatically aect our conclusions.
(i) We specify fixed-delay bin durations, in which the number of occurrences of path times of arrival (TOA) is distributed according to a Poisson law. The TOAs in each bin are randomly distributed.
(ii) Path amplitudes are governed by a Ricean distribution, whose K factor
is randomly generated within certain limits (uniform law); the signs of
path amplitudes are also randomly chosen (uniform law). In addition,
the decay of amplitudes with delay (i.e., path length) is governed by an
attenuation exponent.
(iii) Path DOAs are governed by a Gaussian distribution in azimuth and elevation.
In the practical simulation examples below, elevations will be systematically nil.
The major changes in the parameters will fall on the azimuth spread, and on the
multipath density. Most simulations will make use of omnidirectional scenario,
that is, for which the azimuth distribution function will be flat over 360 . In some
cases, we will make use of clustered channels, that is, for which several multipath
distributions with diering statistical parameters are superimposed. Here we have
selected 3 representative channels.
Channel 1. It has a dominant multipath cluster concentrated in a small delay
range and in a small angle range, followed by a trailing multipath cluster of lesser
amplitude, and scattered in angle. This channel may describe an indoor line-ofsight (LOS) situation.
Channel 2. It has an exponential-like power delay profile over a fairly large range
of delays, with a moderately dense multipath structure. This channel may describe
a relatively gentle indoor LOS or NLOS (non-LOS) case.
Channel 3. It is similar to channel 2, except that the multipath density is much
higher (bad NLOS channel).
Typical realisations of these channels are shown in Figure 5.29.
5.5.3. Spatial correlations in the extreme UWB regime
Assuming a discrete channel model, it is possible to evaluate the various spatial
diversity descriptors considered in Section 5.5.1.2. According to the discussion of
Section 5.5.2.1, we can take the received pulse pr (t) of an ideal channel itself as the
reference template. Then in the absence of receiver filtering and for a single pulse
transmitted at time 0 and received at sensor position rk , the received signal writes


rec rk , t =

 

Ai pr t i rk .

(5.69)

Thomas Kaiser et al.

309

Amplitude

1
0.5
10

15

0.5
1

20

40
60
Delay (ns)

80

100

(a)

Amplitude

1
0.5
12

17

0.5
1

20

40
60
Delay (ns)

80

100

(b)

Amplitude

1
0.5
10

15

0.5
1

20

40
60
Delay (ns)

80

100

(c)
Figure 5.29. Impulse response of (a) channel 1, (b) channel 2, and (c) channel 3.

Therefore,
Sk () =

 

Ai pr t i rk

pr (t )dt =

 

Ai :
pr i rk ,

(5.70)

with


pr () =

pr (t)pr (t )dt

(pulse autocorrelation).

(5.71)

310

Spatial aspects of UWB

The intersensor correlation makes an extensive use of the quantity

  

 

E S1 a Sn b

'
=E

 

Ai A j pr a i r1

(
 

pr b j rn

(5.72)

i, j

where the values of a , b dier for the various intersensor correlations. In the next
five subsections, we will restrict computations to the extreme UWB regime, for
which all interpath delay lags are well larger than the total pulse autocorrelation
duration.
5.5.3.1. Synchronized reference
In the case of SR , we have a = b = iM (r1 ), where iM = arg(max |Ai |). Then
 

E S1 max 1 Sn max 1



'
=E

 

 

Ai A j pr iM r1 i r1

 

(
 

pr iM r1 j rn

i, j


  
2
 
= E AiM  pr (0) pr iM r1 iM rn

'
'
((
2

rn r1 uiM
= E AiM  pr (0) pr


 ((
'
2

d
1n cos 1n,iM

,
= E AiM  pr (0) pr
'

(5.73)
where 1n,iM is the angle between (rn r1 ) and uiM and d1n = |rn r1 |. Finally,


SR =

2



 
E AiM  pr (0) pr d1n cos 1n,iM /c

 
 
 p (0)2 E Ai 2
M
r

(5.74)

The statistical set will aect the value of SR through the distribution of DOA angles
and the distribution of wave amplitudes. Assuming that the angular and amplitude
stochastic variables are independently distributed yields


 

E pr d1n cos 1n,iM /c

.
SR =
pr (0)

(5.75)

When the statistical set describes a single location where the receiving antenna positions occupy a small-scale space area or volume, we can assume that the angles
of arrival do not change over this volume, in which case the expectation operator
E() above disappears. According to the cosine between this angle and the vector joining two spatial diversity antenna positions, we find for the synchronized

311
Normalized correlation

Thomas Kaiser et al.

0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8

SR -single loc

10
15
Distance (cm)

20

25

20

25

Normalized correlation

(a)

0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8

10
15
Distance (cm)

SR
MBO

SS
RC
(b)

Figure 5.30. Intersensor correlations in the extreme UWB regime: (a) single-location averaging; (b)
multiple-location averaging.

reference intersensor correlations a series of pulse autocorrelation functions that

are identical except for a dilation factor on the distance axis. Such functions are
shown on Figure 5.30 for the pulse waveform of Figure 5.28.
When the statistical set describes several locations with a similar path loss (in
order to guarantee the angular and amplitude stochastic variables independence),
we have to take into account the angular distribution function. Assuming the statistical set is very large, we can approximate the discrete sums by integrals, yielding
the expression


SR =

P () pr d1n cos()/c d

,
pr (0)

where P is the distribution function of the angles of arrival.

(5.76)

312

Spatial aspects of UWB

In the case of a uniform distribution (the locations cover equally all

possible
orientations with respect to the Tx-Rx LOS direction), we have SR =

pr (d1n cos()/c)d/2 pr (0). Again this expression is shown on Figure 5.30.
5.5.3.2. Synchronized sensors
In the case of SS , we have a = iM (r1 ), b = iM (rn ), where iM = arg(max |Ai |).
Then
 

E S1 max 1 Sn max n



'

=E

(
  
    
 


Ai A j pr iM r1 i r1 pr iM rn j rn

i, j


2 
2 
= E AiM   pr (0)

(5.77)
resulting in SS = 1. Therefore, whatever the statistical set, the receiver correlator
outputs are fully correlated. This can be very simply explained by saying that if the
pulses are synchronised on each receiver branch, the output signals are equal.
5.5.3.3. RAKE decorrelation
In the case of RD , the computation of intersensor correlations first requires expressing the combined eects of channel and RAKE filtering. Let us specialize to
the case where the total RAKE filter is derived from the channel for the reference
sensor position r1 , but is nevertheless applied to a dierent antenna position r:


h r1 , h(r, ) =

i, j

 

Ai A j i r1




j (r)

 

(5.78)

Ai A j + i r1 j (r) .

i, j

Therefore, the received signal writes, for any receiving antenna position rk ,


rec rk , t =

 

 

Ai A j pr t + i r1 j rk ,

i, j

Sk () =

i, j

 

 

Ai A j pr t + i r1 j rk


 

pr (t )dt

(5.79)

 

Ai A j pr + i r1 j rk .

i, j

Let us write the quantity

  

 

E S1 a Sn b
'

=E

 

 

Ai A j Ak Al pr a + i r1 j r1

 

(
 

pr b + k r1 l rn

i, j,k,l

(5.80)

Thomas Kaiser et al.

313

In the case of RD , a = arg(max(S1 ())) = 0, since the eect of the time reversed
RAKE filtering is to combine constructively all multipaths at this delay. In practice, there would be a true unique delay allowing for the respect of the causality
principle. Furthermore, in the case of RD , we also have b = 0. Thus


E S1 (0)Sn (0)
'

=E

  

 

  

(
 

Ai A j Ak Al pr i r1 j r1 pr k r1 l rn

i, j,k,l

'
=E

'

Ai A j Ak Al pr i0 j0

i, j,k,l

r1 ui u j
c

(

(5.81)


(
r1 uk rn ul
pr k0 l0
.
+

In the extreme UWB regime where multipaths are so much separated in delay that
the pulse autocorrelations attached to each multipaths never overlap whatever the
receiving sensor position, only terms for which i = j and k = l do not cancel.
Therefore,


'

E S1 (0)Sn (0) = pr (0)E

' 
 ((
 2  2
Ai 
Ak  p uk rn r1

'

'


 ((
 2  2
d1n cos 1n,k





= pr (0)E
Ai
Ak pr
,

(5.82)

where 1n,k is the angle between uk and rn r1 . The term i |Ai |2 is the total energy
in the impulse response, which can be considered constant whatever the sensor
position in a small-scale volume, and also constant for a statistical set of several
locations having the same path loss. Therefore, it can be factored from the expectation, and disappears after normalization:

RD =

 +  2 

 
  p d1n cos 1n,k /c
k Ak
r
 +  2

.
Ak  p (0)
E
k

(5.83)

Assuming a continuous distribution and the independence between amplitude

and angular stochastic variables yields
RD =





A()2 P () p d1n cos()/c d
r
.

2
p (0) A() P ()d

(5.84)

These expressions dier slightly from the equivalent ones for SR by the fact that

314

Spatial aspects of UWB

all multipaths come into play in RD . For this reason, it may be predicted that there
will be less dierence between a single location and multiple locations, stemming
from the fact that even for a single location there are already many DOAs involved.
Assuming a Clarkes scenario where |A()|2 P () does not depend on will
again lead to the expression


RD =

pr d1n cos()/c d

= SR .
2 pr (0)

(5.85)

5.5.3.4. RAKE combining

In the case of RC , we have adaptive RAKE filtering for each sensor location, resulting in moderate changes in the above expressions:


E S1 (0)Sn (0)
'

=E

  

 

Ai A j Ak Al pr i r1 j r1

  

 

pr k rn l rn

i, j,k,l

'

'

r1 ui u j
Ai A j Ak Al pr i0 j0
=E
c
i, j,k,l

'

pr

(

r u ul
k0 l0 n k
c

(5.86)

 ((

In the extreme UWB case,

'
(
 2  2

2






E S1 (0)Sn (0) = pr (0) E
Ai
Ak
,


(5.87)

and finally
RC = 1.

(5.88)

We again arrive at the simple and intuitive conclusion that if total RAKE combining is applied on each receiver branch, the correlator outputs are equal and
consequently totally correlated.
5.5.3.5. Multiband OFDM
In the case of SBO , we have, for a given sensor position r,
S=

Ai exp ji (r) ,

(5.89)

where /2 is the vector of frequencies. Thus for two positions r1 and rn of the

Thomas Kaiser et al.

315

sensor we have


'

E S1 Sn = E
'
=E
'
=E

 

Ai A j exp ji r1

(
 

exp j j rn

i, j

Ai A j

i, j

Ai Aj

i, j

  

 

exp jk i r1 j rn

(5.90)
(

exp jk i0 j0

r1 ui rn u j 
.
c

In the extreme UWB case again we have i = j otherwise the enormous frequency
diversity achieved by the ultra-wideband will average the exponential to zero. Thus


'

E S1 Sn = E

 (


 2
rn r1 ui
Ai 
exp j
.
k

(5.91)

If we assume independence between amplitude and angular variables, this writes



'

E S1 Sn = E

( '
'
'
 (((
 2

rn r1 ui
Ai  E
exp jk
,
i

that is,

'

SBO =

1
E
exp
NF
k

'

jk d1n cos 1n,k

c

(5.92)

 ((

where NF is the number of frequency tones. With a continuous approximation of

the angular scenario,
+ 

SBO =

P () exp jk d1n cos()/c d


.
NF P ()d

(5.93)

In the case of a uniform angular distribution, this yields

MBO =



1

,d ,
NF k NBC k 1n

(5.94)

with
1
NBC (, d) =
2

exp

jd cos()
d
d = J0
c
c

(5.95)

J0 being a Bessel function of the first kind.

Finally, we arrive at MBO by averaging the correlations over all subbands:
MBO =

NB

SBO,m
m=1

NB

(5.96)

316

Spatial aspects of UWB

where NB is the number of subbands. This expression is nothing else than the
narrowband intersensor correlation, averaged over all frequency tones. Because
of this averaging, the oscillating dependence of the Bessel function will tend to
washout, as the peak positions scale with the inverse of the frequency. This explains
why beyond about 10 cm distance in the graph of Figure 5.30 the correlation is
virtually flat and equal to zero. Basically the same kind of mechanism operates to
explain the fast decorrelation seen on SR or RD . Those quantities involve the pulse
autocorrelation, which is the Fourier transform of the power spectral density, and
therefore exhibits small and evanescent time-domain sidelobes in the UWB case
for the used waveform.
When the extreme UWB regime is not realized, it is no longer possible to
separate distinct multipaths by a duration greater than the pulse autocorrelation
and greater than the time of flight between the distant sensors. Therefore, there
is in the data a superposition between signals originating from several distinct
multipaths. Since analytical expressions would be tedious to obtain, the following
section presents results of simulations in this intermediate UWB regime, allowing
to assess the dierences with the extreme UWB regime.
5.5.4. Spatial correlation functions results
5.5.4.1. Pulse-based modulation schemes
According to Section 5.5.1.2, we present below simulation results for the 5 kinds
of intersensor correlations SR , SS , RD , RC , and MBO . In addition, we have plotted
two narrowband correlations at the center frequency: the first one is the Clarke
scenario NBC correlation [40], the second one (NB ) has been simulated in the same
way as the other correlations with the same set of channels. In all these simulations,
the sensors are perfectly uncoupled (virtual array), and the intersensor distance
has been varied from 0 to 25 cm by steps of 0.5 cm.
The curves of Figure 5.31 have been computed over a statistical set of 100 realisations of CIRs, corresponding to 100 dierent locations with randomly chosen
orientations. Thus they express the location variability of the channel. A few observations can already be drawn from these data.
(i) The correlations do not significantly depend on the channel, at least in
the conditions of the simulations (signal BW of Figure 5.28, 3 GHz).
(ii) There are only moderate dierences with ideal channel results of the
previous section (Figure 5.30). In particular SS and RC are very close to
1, while SR , RD , and MBO exhibit a fast decorrelation. SR exhibits the
smallest and least oscillating values versus intersensor distance.
(iii) Narrowband correlations exhibit sustained oscillations versus intersensor distance, in close agreement with an ideal Clarke scenario.
It may seem surprising that, given the high multipath density of channel 3,
the plots of Figure 5.31 closely resemble those of the extreme UWB regime
(Figure 5.30). The first reason for that is that all output signals have similar spectral
contents whatever the channel, determined by the pulse shape. The second reason

317

Normalized correlation

Normalized correlation

Thomas Kaiser et al.

0.5
0
0.5

10
15 20
Distance (cm)

SR
MBO
RD
NB

25

1
0.5
0
0.5

SS
RC
NBC

10
15 20
Distance (cm)

SR
MBO
RD
NB

SS
RC
NBC

(b)

Normalized correlation

(a)

25

1
0.5
0
0.5

10
15 20
Distance (cm)

SR
MBO
RD
NB

25

SS
RC
NBC

(c)
Figure 5.31. Intersensor correlations with multiple-location averaging for (a) channel 1, (b) channel
2, and (c) channel 3.

is better understood by first looking at simulation results for single locations, that
is, for which the statistical set only expresses the small-scale spatial variance for a
unique set of DOAs-DODs. It can be seen, for example, for channel 2 (Figure 5.32)
that there is a large variability from one location to another, and SR , RD , and MBO
do not behave quite identically. In particular, RD exhibits by far the fastest decorrelation with increasing distance, and its behaviour is significantly dierent from the
extreme UWB regime of Figure 5.30. The survival of the decorrelation for RD on
a single location is due to the great path diversity inherent to RAKE decorrelation,
and is an interesting feature since this case corresponds to the practical situation
of a receiver in a given position. This ecient decorrelation also applies to a lesser
extent to MBO , since OFDM signal outputs contain all multipaths and again exhibit path diversity. When multiple locations define the statistical set, the angular

Spatial aspects of UWB

1.5

Normalized correlation

Normalized correlation

318
1
0.5
0
0.5
1
1.5

10

15

20

25

1
0.8
0.6
0.4
0.2
0
0.2
0.4

10

15

20

25

Distance (cm)

Distance (cm)

(b)

Normalized correlation

(a)
1.5
1
0.5
0
0.5
1

10

15

20

25

Distance (cm)
(c)
Figure 5.32. Intersensor correlations with single-location averaging for channel 2; (a) SR , (b) RD ,
and (c) MBO .

variability is sucient to randomize the intersensor correlations, which explains

why all three channels have near similar behaviours (Figure 5.31).
It is possible to evaluate the BW sensitivity of these results by modifying the
pulse waveform, keeping it spectrally centred around 5 GHz. It can be seen on
Figure 5.34 that we recover a narrowband behaviour for SR by reducing the bandwidth below about 400 MHz, at least on the explored range of distances. The bandwidth is indeed directly related to the coherence distance of the pulse waveform.
The long coherence distance in narrowband is thus responsible for the oscillating character of NBC , while in the UWB case the very short coherence distance is
responsible for the near-zero correlation at relatively short distances. When the
bandwidth is reduced, SS and RC tend to become less than 1, which is the result of
the onset of small-scale fading.
5.5.5. Space-variant channels
5.5.5.1. Modeled channels
In the narrowband case, the spatial variance of the signals is the Fourier transform of the complex amplitude of incoming waves in wavevector space. In ultrawideband, this spatial variance results from such Fourier transforms, weighted by

Thomas Kaiser et al.

319

Normalized correlation

1.5
1
0.5
0
0.5
1
1.5

10

15

20

25

20

25

Distance (cm)
(a)

Normalized correlation

1.5
1
0.5
0
0.5
1
1.5

10

15

Distance (cm)
(b)
Figure 5.33. Intersensor correlations with single-location averaging for channel 3: (a) SR ; (b) RD .

the frequency-dependent complex gain of the waves. This introduces extra spatial features with respect to the narrowband case. Such features can be seen very
clearly by plotting maps of the receiver signal outputs versus the sensor position.
In Figure 5.35 can be seen such a map of the maximum correlation output in
the narrowband case (centre frequency) plotted in linear scale, and in the UWB
single pulse case for channel 3 (the spatial increment in x, y directions is 2 cm).
The former is characterized by Rayleigh fading and a visually quasirandom appearance, due to the large number of interacting waves which destroy coherence.
The latter exhibits visible structures and an amplitude distribution close to Rice.
Such structures can be expected in a dense channel on the basis that the received
pulses may overlap each other, and this overlap changes with the sensor position
through the wave directions. When the overlap favours a constructing interference
of the pulses, the signal amplitude at the correlator output is enhanced, which
corresponds to the dark squares. This is the UWB equivalent of narrowband fading.
In the case of channel 1 (LOS-like), the structures are linear and larger, which
directly stems from the smaller angular spread of the main paths (Figure 5.36).

320

Spatial aspects of UWB

Normalized correlation

0.5

0.5

10
15
Distance (cm)

5.6 GHz
3 GHz

20

25

1.4 GHz
0.4 GHz
(a)

Normalized correlation

0.5

0.5

10
15
Distance (cm)

20

25

(b)
Figure 5.34. Intersensor correlations with multiple-location averaging for channel 2 as a function of
the received signal bandwidth (inset): (a) SR ; (b) RD .

5.5.5.2. Measured channels

Several UWB channel measurements have been carried out in recent years in order to assess various characteristics of the channel such as multipath statistics
[43, 45, 46], DOA-DOD [47, 48], and spatial structure [49]. Here we show a few
results which have been obtained at ENSTA and by RadioLabs researchers [46, 50].
The former made use of specially designed shaped monocones [51], with a measuring bandwidth from 2 to 10 GHz (vector network analyser). The latter made
use of commercial biconical antennas, and measurements were carried out with
a bandwidth of 2.4 GHz around 4.78 GHz using a pulse generator and a digital
sampling oscilloscope. Depending on the scenario, it can be seen that a variety of
small-scale spatial structures can be seen, with little or very little fading in the LOS
one, and stronger fading in NLOS scenarios.

0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

321
6
5
4
3
2

Y (m)

Y (m)

Thomas Kaiser et al.

1
0

0.1

0.2 0.3
X(m)

0.4

0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

4.5
4
3.5
3
2.5
2
1.5
1
0.5
0

0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

2
1.5
1
0.5
0

0.1

0.2 0.3
X(m)

0.2 0.3
X(m)

0.4

(b)

0.4

100
Cumulated probability

Y (m)

(a)

0.1

101
102
103

50 40 30 20 10

Amplitude (dB)
Narrow band
UWB

(c)

(d)

Figure 5.35. Channel 3 small-scale structures of the received signal in the UWB (a)-(b) and in the
narrowband (c) case, and (d) amplitude distribution.

5.5.6. Diversity combining

5.5.6.1. Combining in the time domain
According to the above results, it appears that the channel spatial variance may
be quite rich especially in NLOS and dense multipath scenarios. This richness can
be exploited through adequate receiver and even transmitter or MIMO architectures [52, 53]. A simple approach is to combine received signals with maximum
ratio combining (MRC) weights, subsequent to proper synchronization in each
output branch. The results of Section 5.5.4 show that the correlation between synchronized signals even on close sensors approaches 1, meaning that we can expect
eciently gaining energy and SNR from this combining approach. In the simulations below, a circular array of 5 cm radius with 9 sensors (including a central one)
was scanned on the square area of 50 50 cm. Several evaluations of the average
combining capability were carried out for subarrays of two sensors (including the
central one), extracted from the 9-sensor array (5 cm intersensor distance). The

Spatial aspects of UWB

0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

0.1

0.2 0.3
X (m)

0.4

3.5
3
2.5
2
1.5
1
0.5
0

Y (m)

Y (m)

322
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

0.1

0.4

(b)

Y (m)

(a)

0.2 0.3
X (m)

2
4
6
8
10
12
14
16
18
20

0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

0.1

0.2 0.3
X (m)

0.4

2
4
6
8
10
12
14
16
18
20

(c)
Figure 5.36. UWB small-scale structure of the received signal for (a) channel 2 and (b) channel 1 (single sensor-UWB), and narrowband small-scale structure for (c) channel 1 (single sensor-narrowband),
in linear scale.

interest of these 8 subarrays is that they cover all possible orientations in the plane.
(i) MRC combining without resynchronization (same clocking for the two
sensors). The combining performance is averaged on the 8 subarrays for
each array position.
(ii) MRC combining after optimal synchronization of each sensor. The combining performance is averaged on the 8 subarrays for each array position.
(iii) MRC combining without resynchronization, but after selection of the
best subarray among the 8 ones per array position.
The first approach is very ineective, which stems from the near-zero correlation
for SR (average output signal on sensor 2 is close to zero if synchronization performed on sensor 1). The second approach is much more eective, with a net gain
typically 35 dB for 2 sensors, resulting from the near-1 correlation for SS . The
third approach, although suboptimal, is nearly as performant as the latter.
5.5.7. Electromagnetic coupling issues for close sensors
In all spatial diversity considerations above, the sensors were electromagnetically
uncoupled, either a simulation assumption or an experimental fact when measurements involved a virtual array. However, in practical applications, it is likely that

Thomas Kaiser et al.

323

sensors will be at short distances, because they will have to be integrated on a small
terminal or on a limited size access point. We can expect significant electromagnetic coupling in such a situation, and it is necessary to assess the importance of
such an eect, especially in the case of UWB which is not as well known as narrowband. According to the previous sections, it appears that in many practical cases
of interest spatial diversity will bring extra energy to the useful signal. Therefore,
we will focus on the influence of coupling on extra energy capture by diversity.
Replacing a single electrical generator (the receiving antenna) by, for example,
an array of two radiators intuitively seems to imply a doubling of the received
power (3 dB gain as seen in the last section), provided suitable impedance matching by a network was achieved. This assumption is only true under the condition of electromagnetically uncoupled sensors. Otherwise electromagnetic shortcircuiting reduces the useable power delivered to the loads. While two far away
sensors deliver at most 2 |E|2 /8Ra , where Ra = Re(Za ) is the antenna radiation
resistance, in the extreme case of very strong coupling, for instance, two short circuited sensors will deliver at most |E|2 /8Ra , that is, there will be no power gain
(Figure 5.40). Thus electromagnetic coupling appears to be the true limitation to
the spatial diversity advantage in UWB systems, rather than intersensor correlation which does not have the meaning it has in narrowband systems as discussed
above.
Obviously coupling will very much depend on the geometry and radiation
characteristics of the radiators. Nevertheless, it is interesting to consider the case
of canonical radiators, and of radiators arranged in an array where the coupling
may be expected to be large. A good example might be an array of dipoles, since
dipoles are probably the most emblematic and most employed radiators, and since
when arranged parallel to each other they are expected to be strongly coupled
as the result of their direction of radiation. Unfortunately, ordinary dipoles (or
monopoles) are resonant (narrowband) radiators, typically of length /2. Shaped
monocones on a ground plane have been designed, optimized, fabricated, and successfully tested, whose bandwidth starts at 3 GHz up to 10 GHz and beyond [51].
They are derived from conventional monopoles, but are much less resonant. Assemblies of two such monopoles in an array of varying intersensor distance (12 cm,
8 cm, and 4 cm) allow to investigate the importance of coupling, through electromagnetic simulations (method of moments, WIPL commercial software). A fairly
large ground plane of constant size 30 16 cm was used, in order to avoid any
detrimental influence of its finiteness on the antenna impedance and for a fair
comparison between arrays with varying intersensor distances. The smallest investigated distance should be appreciated in comparison with the monopole diameter of 2.3 cm (Figure 5.41). Each radiator is fed through an exponentially shaped
transition region, between a feeding coaxial line and the monopole itself. The
feeder, together with the monopole shape, is indeed a critical issue for the proper
impedance matching of the antenna to 50 [54]. This array electrically behaves as
a linear two-port device, characterized by a scattering S matrix, or equivalently by
an impedance or admittance matrix. In all cases, the presence of electromagnetic
coupling between sensors is characterized by o-diagonal entries. The matrices are

Spatial aspects of UWB

Y (m)

324
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

105

16
14
12
10
8
6
4
2
0

0.1

0.2

0.3

X(m)

Y (m)

(a)

0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

104

0.1

0.2

1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0

0.3

X(m)
(b)
10

0.25

Y (m)

0.2
0.15

0.1
0.05
0

0.1

0.2

X (m)
(c)
Figure 5.37. Small-scale fading structure in three ENSTA environments ((a) NLOS and (c) two LOS
ones, linear scale). Spatial steps are 2 and 3 cm, respectively. The scenario in (b) contains two large
reflectors forming a corridor, which results in a regular interference structure.

also symmetric by virtue of reciprocity, and in addition the diagonal elements are
here equal because of the presence of a symmetry plane between the radiators.
Several features can be seen in Figure 5.42.
(i) The return loss is basically below 15 dB in the 36 GHz band, which
is traditionally considered to be a good impedance matching for an antenna.

325

0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.1

0.2

0.3

0.4

0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.4

0.3

0.2

Y (m)

X(m)

(a)

(b)

Y (m)

0.07
0.06
0.05
0.04
0.03
0.02
0.01
0

Y (m)

X (m)

Thomas Kaiser et al.

0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45

0.1 0

0.035
0.03
0.25
0.02
0.15
0.01
0.005
0

0.06
0.05
0.04
0.03
0.02
0.01
0.4

0.3

0.2

0.1 0

X(m)
(c)
Figure 5.38. Small-scale fading structure in three Radiolabs laboratory environments (a) LOS (b)
NLOS, and (c) NLOS (linear scale, spatial step: 2 cm).

(ii) The transmission coecient from one radiator to the other by electromagnetic coupling is below 20 dB for the most distant array, and rises
to below 10 dB for the shortest distance at the smallest frequency.
(iii) The transmission coecient decreases with increasing frequency, for example, from 11 dB to 24 dB between 3 and 8 GHz for the smallest
array.
The first remark shows that even in the smallest array, electromagnetic coupling
remains moderate. This is of course a favourable feature for spatial diversity applications. Nevertheless, it is useful to analyse the eect of this residual coupling in
a quantitative manner. For this purpose, we have computed the power loss in two
ways.
(i) The array is operated in transmission, that is, one radiator is excited by a
generator while the other behaves as a parasitic element delivering some loss power
into its 50 matched load. An eciency can therefore be defined, equal to the
ratio of the radiated power to the total power supplied to the excited radiator.
It can be seen that even in the worst case of the smallest array, the eciency is
better than 0.4 dB at 3 GHz, rapidly growing to better than 0.1 dB at 6 GHz.
This means that a very small power is lost to the load of the parasitic radiator
(Figure 5.43).

Spatial aspects of UWB

0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

2
4
6
8
10
12
14
16
18

0.1

0.2

0.3

Y (m)

0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

0.4

2
4
6
8
10
12
14
16
18

5 cm

0.1

0.2

0.3

(a)

(b)

0.2

0.3

0.4

Two sensors (MRC)- UWB

2
4
6
8
10
12
14
16
18

0.1

X (m)

0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

X (m)

0.4

Y (m)

Y (m)

Y (m)

326

0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0

0.1

0.2

X(m)

X(m)

(c)

(d)

0.3

0.4

2
4
6
8
10
12
14
16
18
20

Figure 5.39. Small-scale fading structure of two sensor subarrays (channel 3); inset: array geometry
showing one subarray; (a) single sensor reference (single sensor-UWB); (b) combining without resynchronization (two sensors (MRC)-UWB); (c) combining after optimal synchronization (two sensors
(MRC)-UWB); (d) selection, no resynchronization (two sensors (MRC)-UWB) (all scales in dB, spatial
steps 2 cm).

E
Za

Za

2Za

2Za
Za

Za

Za
(a)

(b)

Figure 5.40. (a) Two isolated sensors versus (b) two extremly coupled sensors.

(ii) The array is operated in reception. An arbitrary scenario was here considered, in which waves of constant amplitude 1 V/m are incident on the array

Thomas Kaiser et al.

327

S parameter (dB)

Figure 5.41. Schematic view of the dual UWB arrays, with intersensor distances of 12 cm and 4 cm.
0
10
20
30

4
6
Frequency (GHz)

S11
S12

S parameter (dB)

(a)
0
10
20
30

4
6
Frequency (GHz)

S11
S12
(b)
Figure 5.42. S parameters of the arrays: (a) 12 cm distance; (b) 4 cm distance.

between 0 and 90 elevation and 0 to 360 azimuth. The total received power delivered to the loads can be seen on Figure 5.44. There is virtually no dierence between both arrays, meaning that the total received power is not aected by sensors
proximity. Actually when comparing the average received powers for intersensor
distances of 12 and 4 cm, the dierence is less than 2%. As a matter of comparison,
two /4 monopoles distant from /10 lose 45% of received power.
It is interesting to remark the decrease in frequency of the received power.
This feature is typical of UWB, and is due to the frequency dependence of the
antenna eective area for a constant gain antenna, scaling as 2 as recalled in

328

Spatial aspects of UWB

0
Eciency (dB)

0.1
0.2
0.3
0.4
0.5
0.6
0.7

3
4
5
6
Frequency (GHz)

12 cm
4 cm

Mean received power (R.U.)

Figure 5.43. Eciency of the dual sensor array operated in transmission, as defined in the text.

40
35
30
25
20
15
10
5
0

4
5
6
Frequency (GHz)

12 cm
4 cm
Figure 5.44. Mean received power for a hypothetical omnidirectional scenario of incident waves.

Section 5.5.2.1. Obviously low frequencies are more favourable to the link budget. This characteristic explains why the transmission coecient between the two
sensors decreases with increasing frequency, since in the case of power fed to one
radiator, the second behaves as a very close receiver.
In order to conclude this section, let us mention that the radiation patterns
have also been investigated in detail. It turns out that the shadowing eect of
one loaded UWB monopole on the other, although not negligible, is basically minor and does not change the general characteristics of the elementary radiators
radiation patterns (Figure 5.45).
5.5.8. Summary and conclusion
This work has attempted to address spatial diversity in UWB systems by taking into
account the way this diversity could be exploited in a practical receiver. This has

Thomas Kaiser et al.

329
Array

Single radiator

3 GHz

6 GHz

Elevation 30
Elevation 60

Elevation 0

Figure 5.45. Copolarized radiation pattern showing the moderate modification due to the presence
of the nearby radiator at 4 cm (dotted line: elavation 0 , full line: elevation 30 , dashed line: elavation
60 ).

led to the definition of various spatial diversity descriptors, which are intersensor correlations between the output signals of the various diversity branches of the
receiver. According to the way the synchronisation is carried out on these signals,
or on the modulation scheme, full correlation or full decorrelation is found. The
former is useful in order to gather as much energy as possible and improve SNR.
The latter is useful in, for example, MIMO approaches where such a decorrelation
can allow spatial multiplexing of data flows [41]. The distant-dependent intersensor correlations depend on the channel multipath density and angular scenario,
especially when they are computed on single locations (small-scale averaging).
The signal bandwidth is also an important parameter, narrowband correlations
being recovered below a frequency which depends on the multipath density, and
on the intersensor distance. Maps of the spatial variability of receiver output signals have been presented, both for simulated channels using a discrete mutipath
model, and for a few experimental measurements carried out by ENSTA and by
RadioLabs [46, 50]. Although fading is much less prominent than in narrowband
Rayleigh fading conditions, significant variations of the signal are seen, especially
in NLOS case where multipath dense and angularly spread scenarios lead to irregularly structured maps of output signal amplitude. Improvement towards less
spatial variance is seen when diversity combining is carried out, through adequate
combining schemes. The influence of electromagnetic coupling between sensors
has also been evaluated, through electromagnetic simulations of UWB shaped
monocones for which copolarized radiative coupling is a priori the greatest. Such
coupling phenomena may act as a short-circuiting mechanism between the sensors, annihilating the benefit of diversity combining. It appears that coupling has a

330

Spatial aspects of UWB

rather small detrimental eect on, for example, the transmission eciency of dual
sensors arrays made of such monocones, or on the total received power, or on the
radiation patterns of each monocone. Distances down to 4 cm appear acceptable,
for the FCC band 3.110.6 GHz.
5.6. UWB beamforming and DOA estimation
5.6.1. Introduction
This section is split into three parts. First, we briefly introduce the general properties of UWB beamforming, which dier from the narrow- and broadband cases.
The results are basically also valid for broadband pulses as long as the wavefront
travel time across the array is greater than the pulse duration. Second, the major
part of this section is devoted to digital UWB beamforming through investigating
the impact of sampling on the beamformers performance. At the end, we demonstrate by real data that DOA estimation could be principally feasible even in typical
dense multipath environments if clusters are in focus and not paths.
5.6.2. Ideal beamforming for UWB signals
In this section, we will focus on special properties of the beampattern for short
UWB pulses that are dierent from the well-known narrowband beampattern.
To start, we consider the case of a linear equispaced array, consisting of N
equidistant omnidirectional antennas. If c is the propagation speed, is the angle
of incidence of an impulse signal s(t) measured with respect to broadside direction
and d is the distance between two sensors, then the signal recorded at the nth
sensor is given by




d
yn (t) = s t + n sin() = s t + n ,
c

n = 0(1)N 1,

(5.97)

where = (d/c) sin(). Note that in this investigation we omit the influence of a
real UWB channel characterized by multipath propagation.
The ideal delay-and-sum or time-delay beamformer produces
N 1


1
hn yn t n ,
y(, , t) = +
hn n=0

(5.98)

where = (d/c) sin() are the steering delays, is the steering angle, and hn are
the weighting coecients. In order to achieve a time-independent beampattern,13
the total energy of the beamformer output is often used,
'

BP(, ) =



 y(, , t)2 dt

(1/2

(5.99)

13A very similar looking alternative beampattern is defined as [55] BP(, ) = max | y(, , t)|.
t

Thomas Kaiser et al.

331

which reduces to the conventional narrowband beampattern if s(t) is of narrowband type. The problem is that for short UWB pulses, an exact analytical evaluation of (5.99) is in general not possible. For some special UWB pulses, approximate formulas for the beampattern have been developed [56].
Some insight can be gained by computing the spectrum of the beamformer
output. It is given by
N 1
1
hn e j 2n f (d/c)(sin()sin()) ,
Y (, , f ) = S( f ) +
hn n=0

(5.100)

where the sum is recognized as the weighted discrete aperture function or the
Fourier transform of a window sequence hn . Such functions have been investigated extensively for fixed frequency f , and their properties are well known, see
[57].
Introducing the directivity function di defined by
N 1
1
hn ej 2nx
di(x) = +
hn n=0

(5.101)

for convenience, (5.100) can be expressed as



Y (, , f ) = S( f ) di f


d
sin() sin() .
c

(5.102)

Note that di(x) is periodic with period 1.

Now, by using Parsevals relation




 y(, , t)2 dt =



Y (, , f )2 df ,

(5.103)

it is obvious that the UWB beampattern can also be analysed in the frequency
domain. Under the assumption that s(t) is a real-valued bandlimited energy signal
with frequencies in the band from B1 to B2 , the beampattern is expressed as
BP(, ) =
2

=2



Y (, , f )2 df

 B2
B1

 


 

 2
S( f )2  di f d sin() sin()  df .


c

(5.104)

Splitting the frequency band into a number of M nonoverlapping adjacent narrow frequency bands Bi from Bi1 to Bi2 with bandwidth Bi , we obtain by the

332

Spatial aspects of UWB

generalised mean value theorem for integrals

BP(, )2 = 2

 

M  Bi

2 
 

 2
S( f )2  di f d sin() sin()  df


c
i=1

Bi1

M '  Bi

i=1

Bi1

( 





 2
S( f )2 df  di fi d sin() sin()  ,


c

(5.105)

where the frequency fi lies in the frequency band Bi . Hence, the UWB beampattern is approximately a weighted superposition of narrowband beampatterns for
dierent frequencies fi .
This property can be exploited to obtain a rough estimate for the mainlobe
width of the UWB beampattern.
The mainlobe width or beamwidth in the u = sin() variable of the classical narrowband beampattern with uniform array weighting for a wavelength is
usually given by [58]
u=

,
Nd

(5.106)

where N is the number of array elements and d is the distance between two elements. Since the mainlobes are averaged from 1 to 2 with corresponding mainlobe widths from
u1 =

1
2
 u2 =
,
Nd
Nd

(5.107)

it is reasonable to assume that the average resulting mainlobe has a width given by
u=

(1 + 2 )/2
,
Nd

(5.108)

corresponding to the virtual center frequency (B1 + B2 )/2 of the band from B1 to
B2 . This is also reported in [55]. It is not possible to give a more precise general
estimate, since the averaging is weighted with the specific pulse energy spectrum.
Hence, for concrete cases, simulations are the tool for more precise mainlobe estimates; sometimes, for special pulse shapes, it is possible to obtain analytical approximations, and this is certainly an interesting field for future investigations.
Regarding the grating lobes, which are well known in the narrowband case,
there is also an averaging eect for the UWB beampattern that mitigates them
[56, 55]. A simple explanation of this phenomenon is that the grating lobes occur at dierent angles for dierent signal frequencies in contrast to the mainlobe,
which is always located at the same angle. Hence, for a UWB signal, the grating
lobes are averaged out to some extent, whereas the mainlobes are added coherently.
However, it seems dicult to extract more precise information from (5.105).

Thomas Kaiser et al.

333

In order to still gain more insight, a time-domain discussion seems to be more

suitable. If s(t) is a real-valued energy signal, we can write




 y(, , t)2 dt
=

 '

(2

N 1


1
hn yn t n,n
hn n=0

dt

N 1
N 1


1 1 
+
s t + n n s t + l l dt
hn n=0 hl l=0


N
1 N
1



1
hn hl
s t + n n s t + l l dt,
2

hn n=0 l=0
(5.109)

= +

or by introducing the usual definition of the autocorrelation function

ACFs () =

s(t)s(t + )dt,

(5.110)

we obtain


N
1 N
1





 y(, , t)2 dt =  +1 
hn hl ACFs (l n)
2

hn

n=0 l=0

(5.111)

for the UWB beampattern. Since ACFs () has its maximum for = 0, the beampattern has as maximum value
BP(, )2 = ACFs (0).

(5.112)

Further, since the modulus |ACFs ()| of an autocorrelation function is majorized

by a decreasing function in , no true grating lobes can appear in the beampattern.
Of course, this property is of great importance for UWB array design, since it
allows one to choose the antenna distance without the /2 limitation, enabling the
achievement high resolution with only a few array elements.
More precisely, we can express the beampattern as
1
BP(, )2 =  + 2
hn

' N 1

n=0

h2n ACFs (0)

N
1 N
1



1
+  + 2
hn hl ACFs (l n) .
hn n=0 l=0
l=n

(5.113)

(5.114)

334

Spatial aspects of UWB

Now, for pulses with duration T, which means that ACFs () is negligible for | | >
T, if
T < ,

(5.115)

then all terms in the second sum (5.114) disappear, and a fixed sidelobe level
+
+

h2n
2 ACFs (0)
hn

(5.116)

remains. As an example, in the case of an angle of incidence of 90 , the fixed sidelobe level is reached at the steering angle of 90 if
T<

2d
.
c

(5.117)

If the pulse has a shorter duration, the fixed level is already reached at a smaller
stearing angle, which can be computed from (5.115).
An additional important property of UWB beamforming can be derived as
follows. By Cauchys inequality for sums [59], the ratio
+
+

h2n
2
hn

(5.118)

is always 1/N and takes its minimum value 1/N if and only if all the weighting
coecients hn are equal. This is an important result, because it means that
(i) for impulse beamforming, an unequal weighting of the array elements
is not only useless, but detrimental since it increasesthe sidelobe level;
(ii) the maximum ratio of mainlobe to sidelobe level is N.
Furthermore, this result can be extended to the more general case where hn is
substituted by a linear time-invariant system with impulse response hn (t), see [60].
In order to gain more insight for typical UWB signals, we will evaluate the
beampattern by numerical simulations. Most often found in the literature is the
twice dierentiated Gaussian impulse
'

t
g(t) = 1 16
T

2 (

e8(t/T) ,
2

(5.119)

where the nominal duration T is set to 2 1010 second, leading to a 3 dB

bandwidth from 5 GHz to 11.5 GHz. An alternate UWB waveform is a pulse form
adapted to the FCC (Federal Communications Commission) mask. The FCC pulse
is a combination of 6 rect functions in frequency domain multiplied by a Hamming window, which leads to the following pulse in the time domain:
gFCC (t) = 2

6

i=1

ai bi sinc bi t cos 2 fi t w(t),

(5.120)

Thomas Kaiser et al.

335
0

Level (dB)

2
3
4
5
6
7
8
90

45

45

90

Angle (deg)
Figure 5.46. Beampattern of g(t) (solid line) and gFCC (t) (dash-dotted line) impulses arriving at 90
on array 1 with 2 elements at distance 14/2 and of narrowband signal with wavelength (dashed line).

where


w(t) = 0.54 + 0.46 cos




2t
(Hamming window),
10 ns

a1 , . . . , a6 = [0.43, 0.02, 0.22, 0.27, 0.86, 0.27] 102 ,



(5.121)

b1 , . . . , b6 = [1.5666, 0.983, 0.369, 1.146, 7.144, 1.575](GHz),



f1 , . . . , f6 = [0, 1.27, 1.95, 2.71, 6.85, 11.21](GHz).

In the simulations, the first derivative of this transmit pulse will be used, since the
receive antenna is approximated in first order as a dierentiator. The wavelength of
the corresponding sinewave is chosen according to a nominal center frequency of
6.85 GHz. Two typical array configurations are considered throughout this section:
(i) array 1 has 2 elements at distance 7,
(ii) array 2 has 8 elements at distance ,
so that both arrays use the same aperture. Note that in the narrowband case, grating lobes will appear in the beampattern. In Figures 5.46, 5.47, 5.48, and 5.49,
BP(, ) is evaluated for signals arriving at 0 or 90 , and the beampatterns corresponding to g(t), gFCC (t) and a sinewave of frequency 6.85 GHz are plotted in
the same figure for comparision.
Figures 5.46, 5.47, 5.48, and 5.49 show the striking feature of impulse beamforming, namely, the mitigation of grating lobes [56, 55].
A further interesting property that can be observed in Figures 5.46, 5.47, 5.48,
and 5.49 is the dependence of the beampattern and especially its sidelobes on
the pulse shape, which can also be expected from (5.111). This is the object of
further research and may have influence on pulse shaping in future UWB systems.

336

Spatial aspects of UWB

0
1

Level (dB)

2
3
4
5
6
7
8
90

45

45

90

Angle (deg)
Figure 5.47. Beampattern of g(t) (solid line) and gFCC (t) (dash-dotted line) impulses arriving at 0 on
array 1 with 2 elements at distance 14/2 and of narrowband signal with wavelength (dashed line).

Level (dB)

10

15

20
90

45

45

90

Angle (deg)
Figure 5.48. Beampattern of g(t) (solid line) and gFCC (t) (dash-dotted line) impulses arriving at 90
on array 2 with 8 elements at distance and of narrowband signal with wavelength (dashed line).

5.6.3. Digital UWB beamforming

Even if the sampling of UWB signals with a bandwidth of several GHz is far away
from mass-market applications, it can be realized for a short duration even today
with sophisticated storage oscilloscopes. Hence, at least for research purposes, it is
possible to realize UWB beamforming based on sampled signalsso digital beamforming becomes of relevance. The intuitive first approach is the rather simplistic
digital time-delay beamforming. Of course, numerous other methods for UWB
beamforming are known and are worth being considered as well, see, for example, [6163, 55] and the literature cited therein. Since the computational requirements of such sophisticated approaches are often much higher, digital time-delay
beamforming (with interpolated signals) is an attractive and a feasible method for

Thomas Kaiser et al.

337
0

Level (dB)

10

15

20
90

45

45

90

Angle (deg)
Figure 5.49. Beampattern of g(t) (solid line) and gFCC (t) (dash-dotted line) impulses arriving at 0
on array 2 with 8 elements at distance and of narrowband signal with wavelength (dashed line).

systems with a large number of antennas and/or with very high sampling frequencies. However, there is one critical point: for steering a beam to arbitrary directions, which is desirable for many reasons (direction estimation, adaptive beamforming), signal values are needed between the sample points. To cope with this
problem, all digital time-delay beamformers apply some kind of interpolation.
The simplest method is to use nearest-neigbour interpolation (NN interpolation), thus replacing the signal value at the desired steering delay with the signal value at the nearest sampling point. For narrowband beamforming, nearestneighbor interpolation is known to produce large errors. This detrimental eect
can be reduced by heavily oversampling the input signals [64]. However, due to
the immense bandwidth of UWB signals, this seems to be unrealistic in the near
future. A better method is to use linear interpolation between the two nearest
sampling points, which requires two multiplications for each desired signal value.
This approach still produces large errors in the narrowband case without adequate
oversampling [65]. It should be mentioned that these two interpolation methods
are the optimum methods for the case that only one or two signal samples can be
used to approximate a time-shifted signal value [66]. In the next section, a precise
treatment of this topic will be presented, and it will be shown that UWB impulses
give better results than in the narrowband case.
5.6.3.1. The digital interpolation beamformer
A thorough investigation of digital time-delay beamforming using interpolation of
signal samples will now be given. Recall that ideal time-delay beamforming means
appropriately delaying and summing sensor signals. The ideal time-delay beamformer produces
y(, , t) =

N 1

1 
s t + n n .
N n=0

(5.122)

338

Spatial aspects of UWB

No weighting, that is, hn = 1, for all n, will be applied, since the derivation
in the last section has shown that it is detrimental for the beamforming of short
pulses.
Now, in a digital realization, the signals at each antenna are sampled at a rate
fs > 2 fc , where fc is the cuto frequency resulting in signal samples of the form


k
+ n .
fs

(5.123)

Based on these sample values, the ideal beamformer output (5.122) has to be approximated.
Note that in general the desired time instants for beamsteering at angle ,
t + n n ,

(5.124)

do not coincide with the discrete sampling instants

k
+ n .
fs

(5.125)

There is a time dierence n between the ideal steering delay n and its preceding
sample point kn ()/ fs .
We will therefore interpolate the original signals in each channel by the sample values. This has to be developed with some care. The easiest way is to start
with perfect reconstruction using the sinc-kernel, hence to use Shannons sampling theorem (see Appendix 5.6.A), and finally replacing the sinc-kernel with a
short time-limited interpolation kernel.
It follows from (5.155) that


s t + n =

k=



k
+ n sinc fs t k
fs

(5.126)

if the sampling frequency fs is higher than 2 fc . Now, since the time origin is arbitrary, we can replace t by t n , which results in


s t + n n =

k=

 


k
+ n sinc fs t n k .
fs

(5.127)

Applying this procedure for each channel, the ideal time-delay beamformer
can be implemented using signal samples
y(, , t) =



N 1
 


1
k
s
+ n sinc fs t n k .
N n=0 k= fs

(5.128)

Thomas Kaiser et al.

339

Replacing now sinc by a timelimited kernel ker with finite duration T the
complete expression of the sample approximation of the ideal time-delay beamformer, or, in other words, the digital time-delay interpolation beamformer (DTDIBF) is given by
yd (, , t) :=

n )+T  

N 1 k= fs (t
 


1
k
s
+ n ker fs t n k ,
N n=0 k= f (tn )T +1 fs
s

(5.129)

where the summation limits for the finite-time kernel are used.
The quality of the DTDIBF in relation to the ideal beamformer is given by the
following theorem (for the proof, see Appendix 5.6.B).
Theorem 5.6.1. Let ker be a time-limited reconstruction kernel, and let m be an integer 1. Then the conditions


di KER( f ) 

= i k

df i
f =k

(k Z; i = 0, . . . m 1)

(5.130)

on the kernels Fourier transform KER( f ) imply the following error bound for the
dierence between the ideal time-delay beamformer output and the DTDIBF output:


 y(, , t) yd (, , t) Mm (ker)
C

m!

fs
2 fc

m



s(t) ,
C

(5.131)

if s(t) is bandlimited with cuto frequency fc , or in a version more appropriate for the
UWB case
 m





 y(, , t) yd (, , t) Mm (ker) ( fs )m  d s(t)  ,

C
m 

m!

dt

for s(t) C (m) (R).

(5.132)

Mm (ker) is the mth absolute moment of a kernel ker, which is defined by

Mm (ker) = sup



(t k)m ker(t k).

t R k=

(5.133)

Note also the other notations given in Appendix 5.6.A.

As can be expected from the results of single channel interpolation [67], the
maximum deviation of the digital from the ideal beamformer output depends on
the cuto frequency versus sampling rate, the order m and Mm (ker)/m!, which depends on the specific kernel, and can therefore be predicted using the curves given
in Figure 5.50. For more details, see (5.139) below.

Spatial aspects of UWB

Normalized error bound magnitude (dB)

340
30
20
10
0
10
20
30
40
50
60

Ratio sampling rate/bandwidth

m=1
m=2
Figure 5.50. Normalized error bound in dB for kerm , m = 1 and 2.

A second remark is that the use of the error bound from (5.132) should be preferred in the UWB case, because the estimate in (5.131) is based on the Bernstein
inequality [68], which is saturated (worst case) for narrowband signals with cuto
frequency fc . To give a concrete example, for m = 2, a complex exponental signal with amplitude 1 and cuto frequency 0.5, the evaluation of (2 fc / fs )m sC
results in 2 ( fs )2 , while for the UWB pulse s(t) = sinc(t) with the same cuto frequency, ( fs )m dm s(t)/dt m C can be shown to have the value ( 2 /3)( fs )2 ,
hence the value is only 1/3 of the value for the (worst) narrowband case. A similar
result is valid if the C -norm is replaced by the L2 (R)-norm.
A further consequence is as follows. Since matched filtering reduces the effective bandwidth, it should in view of (5.131) be carried out before the digital
beamforming process.
We will mention here as concrete examples of kernels with finite duration
only the unique optimum solutions for m = 1 and m = 2, since the number of
samples for computing an interpolated value should be kept to a minimum for
UWB beamforming. For higher m, which might be interesting in the future, see
[66, 67].
For m = 1, the corresponding kernel is the rect-function

1,

|t | 0.5,

0,

|t | > 0.5,

ker1 (t) = rect(t) =

(5.134)

with Fourier transform

KER1 ( f ) = sinc( f ),

(5.135)

Thomas Kaiser et al.

341

and for m = 2, the corresponding kernel is the (t)-function

1 |t |,

ker2 (t) = (t) =

0,

|t | 1,
|t | > 1,

(5.136)

with Fourier transform

KER2 ( f ) = sinc2 ( f ).

(5.137)

Further, the moments of these kernels are given by



M1 ker1 = 0.5,

M2 ker2 = 0.25.

(5.138)

It is observed that for m = 1 and m = 2, these optimum methods are well known,
as ker1 reconstruction is sample and hold and ker2 reconstruction is linear interpolation [67].
The performance of these kernels for bandlimited signals can be compared by
plotting the normalized error bound, for sC = 1 in dB,

 m 
 M (ker)

fs
 m

m

20 log 
()
 m!

B

(5.139)

as a function of the ratio sampling rate versus bandwidth (Figure 5.50). The tradeo between the sampling rate and the number of samples used for signal interpolation is clearly observed. However, Figure 5.50 allows one to choose the corresponding sampling rate and kernel to fulfill a specified normalized interpolation
error.
As a final remark, it should be mentioned that the new modified linear interpolation with recursive 1-pole multiplier requires less preprocessing as described
in [69] and the revitalized linear interpolation gives better results than using ker2
as expected with only slightly more processing load.
5.6.3.2. The beampattern of the digital time-delay interpolation
beamformer in the narrowband case
In many situations, time-delay beamforming is applied to narrowband signals with
unknown frequencies in a certain frequency band of interest. Even for impulse
beamforming, this case is important, because it applies to the presence of narrowband interference. Therefore, digital beamforming will be investigated for this
specific case, and the serious phenomenon denoted as pseudograting lobes will
be analysed. We will use the complex signal representation to make the derivations
less intricate; as usual, the results can be applied for real signals too.
Following the derivation of the digital time-delay beamformer in the general
case, if the input signal at antenna n is a complex exponential arriving at an angle

342

Spatial aspects of UWB

with an unknown frequency f , where | f | < fc , with fc as the maximum frequency

of the beamformers target frequency band


d
s t + n sin() = ej2 f (t+n(d/c) sin()) ,
c

(5.140)

then the digital beamformer produces

yd (, , t) =

N 1


1 j2 f ((k/ fs )+n )  
e
ker fs t n k .
N n=0 k=

(5.141)

A better understanding of this relation can be obtained by rewriting it (the

detailed derivation can be found in Appendix 5.6.C) as
yd (, , t) =



N 1

f
1 j2 f (t+n n ) j2k fs (tn )
e
e
KER k +
.
N n=0
fs
k=

(5.142)

Hence, the squared magnitude of the digital beamformers output using the
di abbreviation can be written as


 yd (, , t)2


 


2

d
f
d


j2tk fs

=
e
KER k +
di f sin() f + k fs sin() 
.


fs
c
c
k=

(5.143)

This is the narrowband digital interpolation beampattern. The spectrum repetition due to the sampling operation is clearly observed, and it is also presented
in the directivity function. For a closer analysis, it is useful to express (5.143) in
the following way:


 yd (, , t)2





  


f
d
d

= KER
di
f
( f ) sin()
sin()

fs
c
c





 weight
ideal narrowband di function
2



 





f
d
d

+ ej2tk fs KER k +
di f sin() f + k fs sin() 
,

f
c
c
s
k=0









undesired di functions
weights

(5.144)

Thomas Kaiser et al.

343

or, by neglecting cross-terms,


2 
2

 


 




 




2 


f
d
d

 yd (, , t) KER

 
di f c sin() ( f ) c sin() 


f


s 






 


 weight  
ideal narrowband di function

2 
2

 


 













d

f  
d
 .
KER k +
 
+
di
f
f + k fs
sin()
sin()




f
c
c


s


k=0 






 



 
undesired di functions
weights

(5.145)
Note that if KER( f ) = rect( f ), which means that Shannons sampling series
(5.155) is used for the reconstruction, then all the terms except the one for k = 0
in the sum vanish, and the ideal narrowband beampattern is recovered.
If this is not the case, (5.145) reveals that the digital interpolation beampattern
consists of the ideal narrowband beampattern and a sum of undesired terms due to
imperfect interpolation. These terms take their maximum values if the argument
of the di functions is an integer (under the assumption of an equispaced array).
Hence, the equation
M=




d
f sin() f + k fs sin()
c

(5.146)

with integer M describes the angular distribution of the peaks in the digital beampattern, since for the solutions of this equation in M, k, , and , the 1-periodic
function di takes its maximum. Except for k = 0 and M = 0, this means that
there is a peak in the beampattern at an undesired angle, weighted relative to the
mainlobe by

'
(
' (2

f
f 


KER k +
 .
/KER

fs
fs 

(5.147)

This disturbing phenomenon has been called pseudograting lobes (PGLs), and
formula (5.145) shows that it is inherent in any reconstruction procedure by timelimited kernels. The angle related to the undesired peaks can be predicted exactly
by solving (5.146) for a given angle of incidence for all k = 0. Also, since KER( f )
decays with frequency, the main contributions will come from the first small values
of k, especially from k 1.
However, the term pseudograting lobes from [65], where it is discussed in
detail for linear lowpass interpolation, is misleading, since the solution of (5.146)
for M = 0 and k = 0 is in fact an undesired mainlobe. The position of these

344

Spatial aspects of UWB

0

Level (dB)

10

15

20
90

45

45

90

Angle (deg)
Figure 5.51. Ideal beampattern of g(t) arriving at 90 on array 2 (solid line) and its digital beampattern using NN interpolation with fs = 20 GHz (dash-dotted line) and fs = 40 GHz (dashed line).

undesired mainlobes is calculated from (5.146) solving for M = 0 as

'

k, f = arcsin sin()

1 + k fs / f

(
 .

(5.148)

This eect is important since it means that the use of a random array or an aperiodic array, which has no grating lobes [58], does not help eliminate the errors in
the digital beampattern due to the undesired mainlobes.
Hence, in the narrowband case, these undesired mainlobe errors can only be
reduced by improving upon the interpolation by choosing good reconstruction
kernels with high spectral damping around the integers, or choosing a higher sampling frequency, as can be predicted from (5.132).
Again, for a UWB signal, the situation improves: (5.148) shows that the positions of the undesired mainlobes are frequency dependent, and therefore an averaging eect over frequency occurs, which reduces the undesired mainlobes. However, it seems dicult to extract precise information from (5.148) so that it is better
to rely on the results concerning interpolation.
5.6.3.3. Simulation results
In Figures 5.51, 5.52, 5.53, 5.54, 5.55, 5.56, and 5.57, the digital beampatterns for
the two UWB pulses are compared to the ideal analog beampattern. The simulations are preliminarily restricted to the case of array 2. In Figures 5.51 and
5.52, nearest-neigbour interpolation is used in order to approximate the true-time
delays at sampling frequencies fs = 20 GHz and fs = 40 GHz. In Figures 5.53
and 5.54, linear interpolation is used at sampling frequencies fs = 20 GHz and
fs = 40 GHz. Figures 5.55 and 5.56 show linear interpolation and the revitalized
linear interpolation [69] is compared at sampling frequency fs = 30 GHz.

Thomas Kaiser et al.

345
0

Level (dB)

10

15

20
90

45

45

90

Angle (deg)
Figure 5.52. Ideal beampattern of gFCC (t) arriving at 90 on array 2 (solid line) and its digital beampattern using NN interpolation with fs = 20 GHz (dash-dotted line) and fs = 40 GHz (dashed line).

Level (dB)

10

15

20
90

45

45

90

Angle (deg)
Figure 5.53. Ideal beampattern of g(t) arriving at 90 on array 2 (solid line) and its digital beampattern using linear interpolation with fs = 20 GHz (dash-dotted line) and fs = 40 GHz (dashed line).

In Figure 5.57, the digital beampattern of a narrowband signal with frequency

f = 10 GHz arriving at 90 on an array designed for /2 and its digital beampattern using NN interpolation with fs = 20 GHz and linear interpolation with
fs = 20 GHz are shown. Here, two fully developed undesired mainlobes are visible
at 90 and 19.4 , corresponding to k = 1 in (5.148), clearly demonstrating this
phenomenon in the narrowband case.
The use of UWB signals reduces the errors significantly in comparision to the
narrowband case, and the digital beampattern of the UWB pulses are quite close
to the ideal pattern, at least in the case of fs = 40 GHz. This is due to the fact
that in the narrowband case, the interpolation errors from dierent antenna signals can add constructively, while in the UWB case, the averaging eect reduces
the interpolation errors. In fact, at fs = 40 GHz, the cheapest method, namely,

346

Spatial aspects of UWB

0

Level (dB)

10

15

20
90

45

45

90

Angle (deg)
Figure 5.54. Ideal beampattern of gFCC (t) arriving at 90 on array 2 (solid line) and its digital beampattern using linear interpolation with fs = 20 GHz (dash-dotted line) and fs = 40 GHz (dashed line).
0

Level (dB)

10

15

20
90

45

45

90

Angle (deg)
Figure 5.55. Ideal beampattern of g(t) arriving at 90 on array 2 (solid line) and its digital beampattern using linear interpolation with fs = 30 GHz (dash-dotted line) and revitalized linear interpolation
with fs = 30 GHz (dashed line).

nearest-neigbour interpolation, which needs no multiplications, leads to an acceptable beampattern. This means that fortunately digital delay-and-sum beamforming of UWB signals is not complicated from the viewpoint of pure digital
signal processing complexity.
5.6.4. Direction-of-arrival estimation in UWB environments
After discussing UWB beamforming in a rather ideal environment, that is, no
multipath propagation, we will now focus on estimating the relevant DOAs in
realistic UWB scenarios. Note that indoor environments for high-data-rate UWB
communication systems are characterized by significant multipath propagation.
Referred to UWB beamforming, this means that several wavefronts coming from

Thomas Kaiser et al.

347
0

Level (dB)

10

15

20
90

45

45

90

Angle (deg)
Figure 5.56. Ideal beampattern of gFCC (t) arriving at 90 on array 2 (solid line) and its digital beampattern using linear interpolation with fs = 30 GHz (dash-dotted line) and revitalized linear interpolation with fs = 30 GHz (dashed line).

Level (dB)

10

15

20
90

45

45

90

Angle (deg)
Figure 5.57. Ideal beampattern of narrowband signal with frequency f = 10 GHz arriving at 90 on
array designed for /2 (solid line) and its digital beampattern using NN interpolation with fs = 20 GHz
(dash-dotted line) and linear interpolation with fs = 20 GHz (dashed line).

dierent directions will simultaneously impinge on the antenna array. Hence, the
beamformer should either suppress these additional directions or better, if possible, should eciently combine them. For that reason, a crucial question before
entering the subject of multipath-capable UWB beamformers is whether the most
relevant DOAs could be prinicipally determined in typical UWB environments.
The aim of this section is to highlight this issue in an exemplary way in order to
introduce it into this quite open field of UWB DOA estimation.
Despite the huge UWB bandwidth, the dense multipath propagation makes it
impossible to resolve any path in an indoor environment. For that reason, estimating the DOA of each path is impossible too. Fortunately, the UWB propagation is

348

Spatial aspects of UWB

marked by distinct clustering meaning that a couple of paths coming from one major direction are grouped into a cluster so that a DOA could be assigned to cluster
(on behalf of a path). As a special case in LOS (line-of-sight) conditions, the major goal should be to estimate the DOA of the LOS path and in NLOS (without a
dominant LOS) scenarios the rough DOA of a cluster (see [70]) is sucient.
A simple but not optimal way to estimate the DOA is by utilizing a beamformer scanning in a circle and selecting the arrival angles associated with the
largest magnitudes as the relevant DOAs. Therefore, for this task the digital beamformer introduced in the previous section is applied. The weights of the dierent
branches are set to one and the delays n = n = (nd/c) sin() are varied continuously in order to steer the mainlobe in the desired direction . The beamformer
output is given by


'

BP t0 , t, :=

t0 +t
t0



 y(, , t)2 dt

(1/2

(5.149)

where t0 is the starting point, t is the length of the observation interval, and is
the variable representing the DOAs and picked by




 t0 , t = max BP t0 , t, .

(5.150)

First we consider a two-path channel



h(t, ) = 1 t, 1 + 2 t, 2

(5.151)

with the same path delay but dierent attenuations 1 = 2 = 1, and dierent impinging directions 1 = 45 and 2 = 22.5 . This transmit pulse is of Gaussian
shape (5.119) so as to gain a first impression of close to ideal DOA estimation and
the antenna spacing is d = = 4.4 cm. Figure 5.58 shows the resulting normalized
(scaled by its maximum) beampattern. Observe that not only the magnitude deviation (meaning more robustness against additional noise) but also the accuracy
of the DOA estimation is improved with increasing number of antennas. Next,
we consider a more realistic UWB impulse response, which is generated according
to the IEEE standard channel model for UWB NLOS indoor environments [71].
This NLOS impulse response consists of more than 100 paths and has a length of
80 nanoseconds with 10-nanosecond delay spread. The resulting beampattern is
shown in Figure 5.59 with t = 50 nanoseconds. Every circle at the ground surface marks time and angle of arrival for a ray with positive amplitude of the impulse
response whereas every cross represents a negative one. The six significant regions
with large magnitudes correspond to the six major clusters, so it can be seen that
the DOA of a cluster could be satisfactorily resolved in angle and also in time.
In the last example a real LOS scenario is under investigation. The channel
impulse responses were measured in labaratories of the IMST GmbH (Germany,
Kamp-Lintfort) with fs = 10 GHz and d = 5 cm. The true DOA of the LOS component is indicated by a black vertical line in Figure 5.60, and for N = 4 and

Thomas Kaiser et al.

349
0

BP(0, , )

10

15

90

45

45

90

N =2
N =4
N =8
Figure 5.58. Normalized beampattern for N = 2 (solid line), N = 4 (dashed line), and N = 8 (dashdotted line).

BPN (dB)

0
2
4
6
8

200
100
t (ns)

0 90

45

45

90

( )

Figure 5.59. Normalized beampattern for IEEE standard model CM3 impulse response, N = 4.

t = 10 nanoseconds the estimated DOA is again the maximum of the normalized beampattern. It can be seen that both are in good agreement.
5.6.5. Conclusions
The aim of this section was to introduce UWB beamforming with special focus on
digital UWB beamforming, which is today and will be in the future a critical task
because of the enormous UWB bandwidth and the associated extremly high sampling frequency. Digital beamforming was investigated for rather ideal channels,
that is, neither noise nor multipath propagation has been yet considered; however,

350

Spatial aspects of UWB

0
BP (dB)

10
20
30
40
50

150
100
50
t (ns)

0 90

45

45

90

( )

Figure 5.60. BP for N = 4, t = 10 nanoseconds, d = 5 cm in an LOS scenario.

it seems that increasing the bandwidth is beneficial for mitigating errors caused
by sampled data. Measurements have shown that despite the huge bandwidth,
each single path cannot be resolved in the channel impulse response meaning that
beamforming towards a single path is principally not successful. For that reason,
at the end of the section, the estimation of the direction of arrivals is exemplary
investigated by common models and real data. The obtained figures indicate that
DOA estimation could be achieved by considering clustersso a group of paths
on behalf of paths. Nevertheless, much more research is required to substantiate
these first qualitative observations by quantitative ones. Future research will therefore concern UWB DOA estimation and in particular UWB (digital) beamforming
in noisy multipath environments.
Appendices
5.6.A. Mathematical notations and preliminary results
In the following, a real- or complex-valued signal s(t) defined on the real line R is
of finite duration T, if it vanishes outside the interval [T, T]. The Fourier transform (spectrum) of a function s(t) L1 (R), the set of absolutely integrable functions, will be denoted by
S( f ) =

s(t)ej2 f t dt

( f R).

(5.152)

The same notation will be used for the Fourier transform if s(t) L2 (R), the set
of square-integrable functions on the real line. s(t) will be called bandlimited with
bandlimit (or cuto frequency) fc if S( f ) vanishes outside the interval [ fc , + fc ].
A function g(t) is an entire function of exponential type if it is analytic in
the complex plane together with


g(z) Ae |z|

for a constant A and any complex z.

(5.153)

Thomas Kaiser et al.

351

The Paley-Wiener theorem [68] states that g(t) is an entire function of exponential type B and belongs to L2 (R) if and only if g(t) is bandlimited with
bandlimit B/2 and G( f ) L2 (R), and the following applies:
g(z) =

 B/2
B/2

G( f )ej2 f z df .

(5.154)

C(R) denotes the set of all functions g(t), uniformly continuous and bounded
on the real line, endowed with the supremum norm g(t)C = suptR |g(t)| and
C r (R) is the set of functions g(t) such that di g(t)/dt i C(R), 0 i r, where
di g(t)/dt i is the ith derivative of g(t). Square brackets  x mean the greatest integer
less than or equal to x, and the Kronecker symbol i is defined as i = 1 for i = 0,
i = 0 for i = 0. The sinc function is given by sinc(t) = sin(t)/(t) with the usual
continuous completion.
We will use the space of entire signals of exponential type B bounded on the
real line and denote this space by BL(B). This signal space is very convenient, since
it contains all important signal types occuring in the applications, namely, the bandlimited absolutely integrable or square-integrable signals with highest frequency
fc = B/2. This includes bandwidth B, as well as the CW signals ej2 f t for all frequencies f fc ; for further properties, see [68, 72] (where this space is denoted

). We will use this signal space to formulate error bounds in the supremum
by BB
norm.
Using this notation, one version of the famous Shannon sampling theorem
states that if a signal s(t) is in BL(B), then it can be reconstructed without error by
the Shannon sampling series
s(t) =

k=



k
sinc fs t k ,
fs

(5.155)

if the sampling frequency fs is higher than B. The condition fs > B is necessary

since BL(B) contains signals that are not absolutely integrable [68].
5.6.B. Proof of Theorem 5.6.1
For all t R, we have


 y(, , t) yd (, , t)

 N 1


1

 N
 


1
k

 
s t n + n
s
+ n ker fs t n k 
= 

N  n=0
fs
n=0 k=




N 1


 


1 
k


s t n + n
s
+
n
ker
f
t
n k .

s

N n=0 
f
s
k=
(5.156)

352

Spatial aspects of UWB

Now, setting t
= t n and observing that sC = s( +n )C , the inner terms
can be estimated using [67, Theorem 2], and (5.132) follows. Equation (5.131) is
a version of (5.132) for bandlimited signals, because if the signal s(t) is in BL(B),
then the Bernstein inequality [68] states that
 m

 d s(t) 
m


 dt m  (B) sC ,
C

m 0.

(5.157)

Using this in (5.132) yields (5.131).

5.6.C. Proof of equation (5.143)
Applying Poissons summation formula [73], it holds that

ker(x k) =

k=

KER(k)ej2kx ,

(5.158)

k=

where the summation limits on the left-hand side are formally extended to infinite
values and uniform convergence of the series on the left-hand side is assumed to
be true. The next step

k=

ej2( f / fs )(xk) ker(x k) =

KER k +

k=

f j2kx
e
fs

(5.159)

follows by the modulation theorem for Fourier transforms. Setting x = fs t and

multiplying by ej2 f t gives the identity

ej2 f (k/ fs ) ker fs t k = ej2 f t

k=

KER k +

k=

f j2k fs t
e
.
fs

(5.160)

Setting t n for t in (5.160), we obtain

 

ej2 f (k/ fs ) ker fs t n k = ej2 f (tn )

k=

k=

KER k +

f j2k fs (tn )
e
.
fs
(5.161)

Using this equation in (5.141) gives (5.142) and combining the exponentials and
changing the summation order yields
yd (, , t) = ej2 f t

k=

ej2k fs t KER k +


1
f 1 N
ej2( f (n n )k fs n ) ,
fs N n=0
(5.162)

so that evaluation of the inner sum results in (5.143).

Thomas Kaiser et al.

353

5.7. Performance analysis of multiantenna

UWB wireless communications
5.7.1. Introduction
The idea of employing multiple-input multiple-output (MIMO) technology in
ultra-wideband (UWB) system has gained considerable interest recently. By the
use of space-time (ST) coding [7478], MIMO systems exploit both spatial and
temporal diversities, and hence promise to improve the system performance significantly. In this section, we present a performance evaluation for UWB-MIMO
multiuser systems in case of Nakagami frequency-selective fading environment.
First, we quantify the performance merits of single-carrier UWB ST-coded systems regardless of specific coding schemes. We consider various modulation and
multiple-access techniques, including time-hopping (TH) binary pulse-position
modulation (BPPM), TH binary phase-shift keying (BPSK), and direct-sequence
(DS) BPSK. Second, we show that the combination of ST coding and RAKE receiver is able to exploit spatial diversity as well as multipath diversity, richly inherent in UWB environments. Utilizing the real orthogonal design (ROD) [77] as
the engine code for UWB-ST codes, DS-UWB-MIMO transmission is shown to
be a promising scheme that provides superior performance for both single- and
multiple-user scenarios.
Another approach for combating the detrimental eects of multipath fading
in UWB systems is to use orthogonal frequency-division multiplexing (OFDM).
With OFDM modulation, the duration of each symbol is much longer than the
delay spread, so the channel response at each subcarrier could be considered almost flat. Recently, in [7983], space-frequency (SF) coded MIMO-OFDM has
been proposed to exploit spatial as well as frequency diversities in multipath environment. It has been shown in [8486] that UWB-OFDM is able to make benefit
of the multipath diversity. Here, we will discuss the application of SF coding in
UWB-OFDM systems to gain both spatial and frequency diversities.
5.7.2. UWB communications
According to the Federal Communications Commission (FCC) definition, UWB
technology is a transmission scheme that occupies a bandwidth of more than 20%
of its center frequency, or nominally more than 500 MHz. The ultra-wideband nature oers several advantages over narrowband technology including high data
rate, low power consumption, extensive multipath diversities, and support for
multiple access. These unique characteristics of UWB make it a viable candidate
for future short-range wireless communications. Recently, the two leading proposals for IEEE 802.15.3a wireless personal area networking (WPAN) standard [87]
have been based on single-carrier (SC) UWB and multicarrier (MC) multiband
(MB) OFDM. The signal model for both SC-UWB and MC-UWB systems will be
given in the subsequent subsections.

354

Spatial aspects of UWB

TH-BPPM
Tf

Tw

Tc

TH-BPSK
Tc

Tc

DS-BPSK
Figure 5.61. UWB signals with various modulation and multiple-access techniques.

5.7.2.1. Single-carrier UWB

SC-UWB, also referred to as carrier-free and impulse communications, is implemented by directly modulating a sequence of impulse-like waveforms that occupies several GHz bandwidth. In order to allow multiuser transmission without catastrophic collisions, the multiple-access approaches such as TH and DS spreading
have been exploited. Figure 5.61 illustrates an example of TH-UWB and DS-UWB
signals.
The TH-UWB utilizes low duty-cycle pulses, where the time spreading between the pulses is used to provide time multiplexing of users. The uth users
transmitted waveform at the kth frame can be described as [8890]


xu (k; t) = Eu au (k)w t kT f cu (k)Tc bu (k)Td ,

(5.163)

where w(t)

is the transmitted monocycle of duration Tw , and T f is the frame
interval with T f  Tw . The monocycle is normalized to have unit energy, and
each frame is transmitted with energy Eu . To accommodate the TH sequences in
multiple-access environments, T f is further divided into Nc segments of Tc seconds, where Nc Tc T f . The TH sequence of the uth user is denoted by {cu (k)},
0 cu (k) Nc 1. The modulation delay is represented by Td and the data sequence is either {au (k)} or {bu (k)}, depending on the modulation techniques. In
BPPM, the data are conveyed by the positions of the pulses, whence au (k) = 1
for all k and bu (k) {0, 1}. BPSK, on the other hand, alternates the polarities
of the pulses in response to the information. Accordingly, the data sequence is
au (k) {1, 1} whereas bu (k) = 0 for all k.
In contrast to TH approach, DS-UWB employs a train of high duty-cycle
pulses whose polarities follow pseudorandom code sequences. The transmitted
DS-BPSK signal model can be described as [91, 92]


xu (k; t) =

N 1

c

  

Eu
du (k)
cu nc w t kT f nc Tc ,
Nc
nc =0

(5.164)

where du (k) {1, 1} is the data, {cu (nc )}Nncc=01 {1, 1} represents the spreading

Thomas Kaiser et al.

355


sequence, and Tc Tw denotes the hop period. The factor 1/Nc is introduced
such that the sequence of Nc monocycles has unit energy, and hence the transmitted energy per frame is Eu .
5.7.2.2. Multicarrier UWB
As opposed to the conventional SC-UWB in which a single bandwidth is used
for all data transmission, MC-UWB system uses multiple-carrier frequencies to
transmit the information. Since utilizing the whole available UWB spectrum (from
3.1 GHz to 10.6 GHz) leads to several disadvantages, such as high transceiver complexity, requirement on high-speed analog-to-digital converters, and high power
consumption, MC MB-UWB approach [87] has been introduced.
In MB-UWB system, the available bandwidth is divided into several subbands
with smaller bandwidth. Each subband occupies a bandwidth of at least 500 MHz
in compliance with the FCC regulations. In order to eciently capture the multipath energy, which is richly inherent in UWB environment, OFDM technique
has been used to modulate the information in each subband. The major dierence
between MB-OFDM and traditional OFDM scheme is that the MB-OFDM transmission is not performed continually on every subband. Instead, the modulated
OFDM signal is transmitted over dierent subbands at staggered time in order
to enable full frequency diversity and accommodate multiuser communication.
Specifically, consider an MB-OFDM system with N subcarriers and the subband
bandwidth of Bsb . The uth users transmitted complex continuous MB-OFDM signal at the sth symbol period can be modeled as
xus (t) =

N
1

dus (k)pus (k; t sT),

(5.165)

k=0

in which T represents OFDM symbol duration, dus (k) denotes the complex symbol to be sent over the kth subcarrier, and { pus (k; t)}kN=01 is a set of N-orthogonal
waveforms. Typically, pus (k; t) is given by




pus (k; t) = w(t) exp j2 fus + k f t ,

(5.166)

where w(t) is a lowpass pulse of duration T, and f = 1/T = Bsb /N is the frequency spacing between the adjacent subcarreirs. The carrier frequency, fus , specifies subband, in which the signal is transmitted during the sth OFDM symbol
duration. These carrier frequency sequences are based on time-frequency codes,
which are uniquely assigned to dierent users so as to minimize the multipleaccess interference.
5.7.3. MIMO communications
In a point-to-point MIMO system, multiple antennas are deployed at both transmitter and receiver, as shown in Figure 5.62. At the transmitter, the data sequence

356

Spatial aspects of UWB

Mod.

RF

RF

ST or SF
encoder
Mod.

Demod.
.
..

.
..
RF

RF

ST or SF
decoder
Demod.

Figure 5.62. A point-to-point MIMO communication system.

is divided into blocks. Each block is encoded into a codeword matrix D. Each
column of D contains a sequence of symbols that will be sent from each transmit
antenna over a series of time slot or frequency tones (depending on the underlined
modulation scheme, e.g., single carrier or multicarrier). These symbol streams are
modulated with a pulse-shaping function, translated to the passband via parallel
radio frequency (RF) chains, and then simultaneously transmitted over all transmit antennas. After down conversion, matched-filtering and demodulation processes, the receiver jointly decodes the received signals across all receive antennas.
The design of matrix D is originally studied for flat-fading channels. In this
case, the row and column indices of D indicate the dimensions of space and time,
and hence D is termed space-time (ST) code. For a MIMO system with Nt transmit
and Nr receive antennas, D is an K Nt matrix whose (k, i)th element, denoted by
di (k), represents the symbol transmitted at transmit antenna i over time slot k.
Here, K represents the number of time slots for one codeword transmission. The
MIMO channel is described by an Nt Nr matrix A. The (i, j)th component of A,
denoted by i j , is the channel fading coecient from the ith transmit to the jth receive antenna. The received signal at each receive antenna is a noisy superposition
of the Nt transmitted signals degraded by the channel fading. Consider the case
when the channels are quasistatic, that is, they remain constant during the transmission of an entire codeword. The received signal can be described as an K Nr
matrix


Y=

E
DA + N,
Nt

(5.167)

where N is the matrix of additive complex Gaussian noises, each with zero mean
and variance N0 /2 per dimension. For normalization purposes, the fading coecient for each transmit-receive link is assumed to have unit variance, and the ST
code satisfies the energy constraint E[D2 ] = KNt . Here, E[X] and
 X denote
the expectation and Frobenius norm14 of X, respectively. The factor 1/Nt ensures
transmit energy identical to single-antenna transmission. Assuming that the channel state information (CSI) is perfectly known at the receiver, the receiver performs
 that minimizes
maximum likelihood (ML) decoding by choosing the codeword D
14The Frobenius norm of an M N matrix X = (x ) is defined as [93]
mn





X2 = tr XH X = tr XXH =

' M 1 N 1
(


xmn 2 .
m=0 n=0

Thomas Kaiser et al.

357

the square Euclidean distance between the hypothesized and actual received signal
matrices, that is,


2


E




D = arg min Y
DA
.

D
Nt

(5.168)

The upper bound of the average pairwise error probability (PEP) between D and
 is of the form [74, 75]
D


 Gc
P D  D

4Nt

1Gd

(5.169)

where = E/N0 is the average signal-to-noise ratio (SNR) at each receive antenna.
The quantities Gd and Gc depend on the distribution of channel fading coecients
and the structure of D. They characterize the performance of ST-coded MIMO
system as follows. The exponent Gd determines the slope of the error probability
curve plotted as a function of SNR (measured in dB). The factor Gc displaces the
performance curve rather than alternating its slope. The minimum values of Gd
and Gc over all pairs of distinct codewords are called diversity gain and coding
gain, respectively.
In wideband system, the channel fading is frequency selective. To exploit the
additional frequency diversity in wideband MIMO communications, SF code incorporating with OFDM modulation has been introduced. The strategy of SF coding is to distribute the symbols across space (transmit antenna) and frequency
(OFDM tones). The codeword D in this case represents SF codeword of size N Nt ,
where N is the number of OFDM tones. Note that the design of SF code is restricted to one OFDM symbol duration (T seconds), and the MIMO channel is
normally assumed constant over a period of T. The performance criteria for SF
coded MIMO-OFDM can be characterized in a similar fashion as that of ST system.
5.7.4. Single-carrier UWB-MIMO system
The application of ST coding in SC-UWB communication was first proposed in
[94], where a specific ST code is used for 2-transmit-antenna system. In this section, we will introduce a general SC-UWB-MIMO system model, in which all ST
coding schemes can be used directly. Under this model, we will provide the analysis
of the receiver processing and system performances.
5.7.4.1. Transceiver design
We consider an SC-UWB-MIMO multiuser system with Nu users, each equipped
with Nt transmit antennas, and a receiver with Nr receive antennas as shown in
Figure 5.63.
For each user, the sequence of binary symbols is divided into blocks of Nb
symbols. Each block is encoded into an ST codeword to be simultaneously sent

358

Spatial aspects of UWB

User 0

ST
encoder

UWB
mod.

.
.
.

Nt 1
.
.
.
User Nu 1

ST
encoder

UWB
mod.

.
.
.

Nr 1

.
.
.

Rake
receiver

Maximum
likelihood
decoder

Rake
receiver

Nt 1
Figure 5.63. A SC-UWB-MIMO multiuser communication system.

over all transmit antennas. The ST codeword matrix can be expressed as a K Nt

matrix

Du =

du0 (0)
du0 (1)
..
.

du1 (0)
du1 (1)
..
.

du0 (K 1) du1 (K 1)

..

duNt 1 (0)
duNt 1 (1)
..
.
duNt 1 (K 1)

(5.170)

where dui (k) {1, 1} represents the binary symbol transmitted by the uth user
at transmit antenna i over time slot k. Since K time slots are required for Nb symbols transmission, the code rate is R = Nb /K. The transmitter converts the ST
 u (t) = [xui (k; t)] by modcodeword matrix into a K Nt UWB ST signal matrix X
i
i
ulating the symbol du (k) into UWB signal xu (k; t). The UWB signal depends on
particular multiple-access and modulation techniques. With TH approach, xui (k; t)
is modeled as (see (5.163))


xui (k; t)

1 dui (k)
Eu
w t kT f cu (k)Tc
Td
Nt
2

(5.171)

for BPPM, and



xui (k; t)



Eu i
du (k)w t kT f cu (k)Tc
Nt

(5.172)

for BPSK. With DS-BPSK technique (see (5.164)), the UWB signal becomes


xui (k; t)

=


N 1

c

  

Eu i
du (k)
cu nc w t kT f nc Tc .
Nt Nc
nc =0

(5.173)

Note that the factor 1/Nt is introduced to ensure that the total transmitted energy
of the uth user is Eu during each frame interval, independent of the number of
transmit antennas.

Thomas Kaiser et al.

359

The MIMO channel coecients are assumed to be real, mutually independent, and quasistatic. Ignoring the Doppler eect, the channel impulse response
from transmit antenna i of user u to receive antenna j can be modeled as
ij

hu (t) =

L
u 1

ij

u (l) t u (l) ,

(5.174)

l=0
ij

where u (l) is the multipath gain coecients, Lu denotes the number of resolvable
paths, and u (l) represents the path delays relative to the delay of the desired users
first arrival path. Without loss of generality, we let the first user be the desired
user, and assume 0 (0) = 0. In order to simplify the analysis, we assume that the
minimum resolvable delay is equal to the pulse width. The amplitude of the lth
ij
path, |u (l)|, is modeled as a Nakagami-m random variable with average power
u (l), that is, [95]


p|iuj (l)| (x) =

2
m
(m) u (l)

m

x2m1 exp

m 2
x ,
u (l)

(5.175)

where () denotes the Gamma function, m is the fading parameter (m 1/2),

ij
and u (l) = E[|u (l)|2 ]. For each user, the time delay u (l) and the average power
u (l) are assumed identical for every transmit-receive link.
Due to the eect of propagation channel and the variation of antenna char
is
acteristics caused by large bandwidth, the shape of transmitted monocycle w(t)
modified to its second derivative at the receive antenna output [96]. Denote the
received monocycle waveform by w(t) and let xui (k; t) be of the form similar to the

with w(t).
transmitted waveform xui (k; t) by replacing w(t)
The signal received at each receive antenna consists of mulitpath signals from
all active users and thermal noise. With the first user being the desired user, the
received signal at receive antenna j can be expressed as
j

r j (t) = y0 (t) + nMU (t) + n j (t),

j

where nMU (t) =

+Nu 1
u=1

(5.176)

yu (t) is the multiuser interference, and

yu (t) =

N
1 L
t 1 K
u 1

ij

u (l)xui k; t u (l)

(5.177)

i=0 k=0 l=0

represents the signal from the uth user. The real additive white Gaussian noise
process, n j (t), is assumed to have zero mean and two-sided power spectral density
N0 /2. By choosing the signal parameters such that Nc Tc + maxu {u (Lu 1)} T f ,
we can guarantee that there is no intersymbol interference (ISI).
As depicted in Figure 5.63, an SC-UWB-MIMO receiver comprises Nr RAKE
receivers and an ML decoder where the decoding process is performed jointly

360

Spatial aspects of UWB

across all Nr receive antennas. The RAKE receiver employs L (L maxu {Lu }) correlators, each using the delayed versions of the received monocycle as the reference
waveform. Denote the reference signal by vk
(t). The output of the l
th correlator
at receive antenna j is given by
j

yk
(l
) =

vk
t 0 (l
) r j (t)dt  yd,k
(l
) + ntot,k
(l
),

(5.178)

where yd,k
(l
) denotes the correlator output corresponding to the desired transj
j
j
mitted information, and ntot,k
(l
)  nMU,k
(l
) + nk
(l
) is the combination of multiuser interference and sample noise. In what follows, we analyze the receiver assuming dierent modulation and multiple-access techniques employed. For
 notation convenience, we denote the autocorrelation function of w(t) as (s)= w(t
s)w(t)ds.
TH-BPPM. The design of TH-BPPM receiver depends on the choice of the modulation delay, Td . With Td = arg minTd (Td ), the correlation waveform adopted
at each RAKE receiver is modeled as


vk
(t) = w t k
T f c0 (k
)Tc w t k
T f c0 (k
)Tc Td .

(5.179)

Substituting (5.171) and (5.179) into (5.178), after some manipulations, the correlator output is given by


j
yk
(l
)

= 1 Td



Nt 1
E0
ij
j
di (k
)0 (l
) + ntot,k
(l
).
Nt i=0 0

(5.180)

Next we arrange the correlator outputs in the matrix form



Y j = 1 Td


 E0

Nt

D0 A0 + Ntot ,

(5.181)

where D0 is the desired users ST symbol defined in (5.170). Both matrices Y j and
j
j
j
Ntot are of size K L whose (k, l)th elements are yk (l) and ntot,k (l), respectively.
j
The multipath gain coecient matrix A0 of size Nt L is formatted as

j
A0

0j

0j

0 (0)
1j
0 (0)
..
.
(Nt 1) j

0 (1)
1j
0 (1)
..
.
(Nt 1) j

(0) 0

..

0j

0 (L 1)
1j
0 (L 1)
..
.
(Nt 1) j

(1) 0

(L 1)

(5.182)

Given the CSI on MIMO channels, the output of ML decoder is

 0 = arg min
D
D0


2

  E0
j
 j 
Y 1 Td
D
A
.
0 0


Nt


N
r 1 
j =0

(5.183)

Thomas Kaiser et al.

361

TH-BPSK. The reference waveform used at the TH-BPSK receiver is the delay
version of the received monocycle, that is,


vk
(t) = w t k
T f c0 (k
)Tc .

(5.184)

Substituting (5.172) and (5.184) into (5.178), we obtain


j
yk
(l
)

Nt 1
E0
ij
j
di (k
)0 (l
) + ntot,k
(l
).
Nt i=0 0

(5.185)

The correlator outputs can be written in the matrix form as


j

Y =
j

E0
j
j
D0 A0 + Ntot ,
Nt

(5.186)

in which Y j , A0 , and Ntot are in the same forms as the ones stated in (5.181). The
decision rule for the ML decoder can be written similar to (5.183) as
 0 = arg min
D
D0


N
r 1 
j =0

 j
Y


2


E0
j
D0 A0 
.

Nt

(5.187)

DS-BPSK. The DS-BPSK receiver adopts the monocycle sequence



v (t) =
k

Nc 1
  

1
c0 n
c w t k
T f n
c Tc
Nc n
c =0

(5.188)

as the reference waveform. From (5.173), (5.188), and (5.178), we have


j
yk
(l
)

Nt 1
L
0 1
E0
ij
j
d0i (k
)
0 (l) f (l, l
) + ntot,k
(l
),
Nt i=0
l=0

(5.189)

where
f (l, l
) 

Nc 1
c 1
  N
  


1
c 0 nc

c0 nc nc nc
Tc + 0 (l) 0 (l
) .
Nc nc
=0
nc =0

(5.190)

We rewrite the correlator outputs in the matrix form as



Yj =

E0
j
j
D0 A0 F + Ntot ,
Nt

(5.191)

362

Spatial aspects of UWB

in which F is an L0 L matrix whose (l, l

)th element is f (l, l
). The multipath gain
j
coecient matrix A0 of size Nt L0 is of a form similar to (5.182). Subsequently,
the decision rule can be stated as
 0 = arg min
D


N
r 1 

D0

 j
Y


j =0

2


E0
j 
D0 A0 F
.

Nt

(5.192)

5.7.4.2. Performance analysis

To evaluate the performance of SC-UWB MIMO systems, we first calculate the
j
statistics of noise and interference. According to (5.178), the noise sample nk
(l
)


is Gaussian distributed with zero mean and variance n2  (N0 /2) vk2
(l
)dt.
j
To investigate the distribution of multiuser interference, we express nMU,k
(l
) (see
(5.176) and (5.178)) as
j

nMU,k
(l
) =

N
1
u 1 N
t 1 L

u=1

i=0 l=0

ij

u (l)niu,k
(l, l
),

(5.193)

where niu,k
(l, l
)  vk
(l
)xui (t u (l))dt. Using the same approach as in [96],
one can show that niu,k
(l, l
) is approximately Gaussian random variable with zero
mean and variance
% j
2 & E 1
E nu,m (k) = u
Nt T f

 0

12

w(t s)v(t)dt ds 

Eu 2
.
Nt a

(5.194)

Assuming Nakagami-m fading coecients, independent of the transmitted signals, and using central limit theorem, we can show that for suciently large L, Nt
j
and Nu , nMU,k
(l
) is approximately Gaussian random variable with zero mean and
+
+
j
variance a2 Nu=u 1 1 Eu Ll=01 u (l). Hence, ntot,k
(l
) is zero-mean Gaussian random
variable with variance
n2tot = a2

N
u 1

Eu

u=1

L
1

u (l) + n2 .

(5.195)

l=0

Since the total noise and interference can be approximated with Gaussian distribution, PEP can be evaluated in a similar fashion as in the conventional narrowband
MIMO system. The values of n2 , a2 , and PEP depend on particular modulation
and multiple-access techniques, and will be given in the following subsections.
TH-BPPM. From (5.179), we have


vk
t 0 (l
)

2



dt = 2 1 Td .

(5.196)

Thomas Kaiser et al.

363

)]N , and a2 is obtained from

(5.194) as
The noise variance becomes n2 = [1(T
 d 2 0

2
2
2
2
a = 2(a d ), in which a  (1/T f ) (s)ds and d2  (1/T f ) (s)(s
Td )ds.
 0 are two distinct transmitted ST codewords, following
Suppose that D0 and D
similar calculation steps as in [97], the PEP conditioned on the channel coecient
j
matrix A0 is given by


0 |
P D0  D

j
A0

;
<
<
<
= Q =

2Nt

2
 j


 0 A0  ,
 D0 D

N
r 1 

(5.197)

j =0

where Q(x) is the Gaussian error function,15 and = [1 (Td )]2 E0 /(2n2tot ), which
can be expressed as
!

a2 d2

= 4
 2
1 Td

N
u 1
u=1

L1
Eu
2N0 /E0
 
u (l) + 
E0 l=0
1 Td

"1

(5.198)

Applying the inequality Q(x) (1/2) exp(x2 /2), for x > 0, the conditional PEP
in (5.197) can be upper-bounded by


 1
 0 | A0j exp
P D0  D

'

Nr 1 
 2


 0 A0j 
 D0 D
 .

4Nt j =0

(5.199)

 0 )T (D0 D
 0 ), where ()T denotes
For convenience, let us define Z = (D0 D
transpose operation. Notice that
1 

T j
 2 L
j

 A0j 
 D0 D
 =
a0 (l) Za0 (l),

(5.200)

l=0
j

where a0 (l) represents the lth column of A0 . Since Z is a real symmetric matrix,
there exists a set of nonnegative eigenvalues {i }Ni=t 01 and the corresponding normalized eigenvectors {vi }Ni=t 01 such that
Z = VVT ,

(5.201)

where V  [v0 v1 vNt 1 ] is an orthonormal matrix, and is a diagonal matrix whose diagonal elements are the eigenvalues of Z. Substituting (5.201) into
(5.200), we have
1 N
t 1


2
 2 L

 0 A0j 
 D0 D
 =
i i j (l) ,

l=0 i=0
15The Gaussian error function is defined as Q(x) = (1/ 2)  exp(s2 /2)ds.
x

(5.202)

364

Spatial aspects of UWB

in which i j (l)  (a0 (l))T vi . Since {0 (l)}Ni=t 01 are independent identically distributed (i.i.d.) and V is orthonormal, {i j (l)}Ni=t 01 are i.i.d. random variables
whose magnitudes are approximately Nakagami-m distributed with parameter
m
 = Nt m/(Nt m m + 1) and average power 0 (l) (see [95, page 25]). By the
use of characteristic functions, |i j (l)|2 is approximately Gamma distributed, that
is, [98]
j

ij

p|i j (l)|2 (x) =


1
m
(m)
 0 (l)

m

xm 1 exp


m
x .
0 (l)

(5.203)

Substituting (5.202) into (5.199) and averaging (5.199) with respect to the distribution of |i j (l)|2 , the resultant PEP upper bound is given by


0
P D0  D

 r
 mN

L@
1 N@
t 1 
l=0 i=0

0 (l)
1+
i
4Nt m


(5.204)

For high SNR environments, the bound can be further simplified to



 0 G0 (m)

P D0  D
C

4Nt

1mrN
 rL

(5.205)

L1
r 1
where G0 (m)
  (m)
 1 ( l=0 0 (l))1/L ( i=0 i )1/r , r is the rank, and {i }ir=01 represent nonzero eigenvalues of matrix Z. For a single-user system, since there is no effect of multiple-access interference, reduces to [1 (Td )]E0 /2N0 . Consequently,
 r L and G0 (m)
 over all pairs of distinct codewords dethe minimum values of mrN
fine diversity gain and coding gain of TH-BPPM MIMO system.

TH-BPSK. Since the reference signal for TH-BPSK system is the shifted monocycle whose energy is unity, n2 = N0 /2 and a2 = a2 . Following the similar calculations as in BPPM, the upper bound of the PEP can be expressed as


 0 G0 (m)
P D0  D


4Nt

1mrN
 rL

(5.206)

where G0 (m)
 is of the same form as the one defined in (5.205), and
!

 1
N
L1
u 1
E0
Eu
E0
2
=
2

(l)
+
u
a
2n2tot
E
N
0
0
u=1
l=0

"1

(5.207)

which becomes E0 /N0 for the single-user system.

DS-BPSK. From (5.188), we can show that n2 = N0 /2 and a2 = a2 . As in THUWB, the upper bound of the PEP conditioned on the channel matrix is given
by


 0 | A0
P D0  D

1
exp
2

'

Nr 1 
2



 0 A0j F
 D0 D
 ,
4Nt j =0

(5.208)

Thomas Kaiser et al.

365

where = E0 /(2n2tot ) is in the same form as (5.207). By the use of eigenvalue

decomposition as in (5.201), we obtain

2
  j T

 T


j

 0 A0j F
 D0 D
 = tr FT A0 ZA0 F = tr B j B j ,

(5.209)

where tr() stands for the trace operation, and B j  VT A0 F is an Nt L matrix.

Let Ix denote an x x identity matrix, represent the tensor product, and v(X)
stack the columns of X in a single column vector. Then, we have
N
r 1 

r 1
2 N
 T



 0 A0j F
 D0 D
 =
tr B j B j = bT b,

j =0

(5.210)

j =0

where
 INr L ,

b  vT B0 vT BNr 1

&T

(5.211)

Denote the correlation matrix of b by R = E[bbT ]. Since R is nonnegative definite,

it has a symmetric square root U such that R = UT U [93]. Let q = (UT )1 b.
Observe that E[qqT ] = INt Nr L , and hence the components of q are uncorrelated.
Now, the upper bound of the conditioned PEP in (5.208) can be reexpressed as


j

 0 | A0 exp
P D0  D

T
q UUT q .
4Nt

(5.212)

Assuming that R is full rank, U is also full rank [93]. Therefore, maximum diversity
gain can be achieved by maximizing the rank of . Note that
rank() = Nr Lrank() = Nr Lrank(Z),

(5.213)

where rank(X) stands for the rank of X. Hence, the rank criterion for DS-UWB ST
system is identical to that of TH-UWB ST system. In order to quantify the coding
gain, it might be necessary to evaluate the statistics of q which is dicult to obtain
for Nakagami fading distribution. The performance of DS-UWB ST system can be
further investigated by computer simulations.
5.7.4.3. SC-UWB using ROD ST code
In this section, we consider two-transmit-antenna system employing ROD ST coding scheme [77]. Generalization to UWB ST systems with higher number of transmit antennas is straightforward. The user subscript u is omitted for notation simplicity. Based on ROD structure, the ST code is modeled as

d0
D= 1
d

d1
.
d0

(5.214)

366

Spatial aspects of UWB

Since two time slots are used for the transmission of two symbols (d0 and d1 ), the
code is of full rate R = 1. The corresponding matrix Z is given by
Z=

1



di 9
d

i 2

I2 = 4

i=0

1



i

di 9
d I2 .

(5.215)

i=0

For ROD code with rate 1/K, where K 2 is an even integer, the codeword D of
size K 2 is modeled as

1
D = d
1

T T
1

1
1

1
1 1

(5.216)

K 2

In this case, the data symbol d is transmitted repeatedly over K frames from both
transmit antennas. The matrix Z becomes
d )2 KI2 = 4K(d 9
d )I2 .
Z = (d 9

(5.217)

Observe that both full- and reduced-rate codes result in two equal eigenvalues 0 =
1  and the matrix Z = I2 of full rank (r = 2). Substituting the eigenvalues
into (5.205), we arrive at



P D  D

L@
1 
l=0

 r
 2mN

(l)
m
 8

(5.218)

where m
 = 2m/(m + 1). With channel parameters and Nr being fixed, (5.218)
depends only on the value of . The higher the , the better the performance.
To compare the performance of various systems, we assume that the energy
per bit Eb is fixed. For simplicity, we also assume that all users have equal transmitted energy per frame (E). Expressing E in term of Eb , we have E = Eb for full-rate
and E = Eb /K for 1/K-rate codes. Assuming one erroneous symbol, the eigenvalues for 1/K rate are K times larger than those for full-rate code. We denote the

eigenvalue of full-rate code as .

In single-user case, is [1 (Td )](Eb /2N0 ) for BPPM and (Eb /N0 ) for
BPSK, regardless of the code rate. As a result, reducing the code rate does not
improve the performance of single-user systems. In addition, of TH/DS-BPSK
systems are 2[1 (Td )]1 times that of TH-BPPM system. Since [1 (Td )] < 2,
TH/DS-BPSK systems tend to outperform TH-BPPM system for every code rate.
On the other hand, we show in [99] that the multiuser system with reduced rate
provides higher value of , and hence likely to performs better than that with full
rate.
Figures 5.64 and 5.65 show the BER performance of TH and DS UWB systems in single-user and asynchronous multiuser environments. We employ UWB
signals with T f = 100 nanoseconds, Tw = 0.8 nanosecond, and random TH or

Thomas Kaiser et al.

367
100
101

BER

102
103
104
105

10 12 14

16

18 20

Eb /N0 (dB)
TH-BPPM, Nt = 1, Nr = 1
TH-BPPM, Nt = 2, Nr = 1
TH-BPPM, Nt = 2, Nr = 2
TH-BPSK, Nt = 1, Nr = 1
TH-BPSK, Nt = 2, Nr = 1

TH-BPSK, Nt = 2, Nr
DS-BPSK, Nt = 1, Nr
DS-BPSK, Nt = 2, Nr
DS-BPSK, Nt = 2, Nr

=2
=1
=1
=2

(a)
100
101

BER

102
103
104
105

10 12 14

16 18

20

Eb /N0 (dB)
TH-BPPM, Nt = 1, Nr = 1
TH-BPPM, Nt = 2, Nr = 1
TH-BPPM, Nt = 2, Nr = 2
TH-BPSK, Nt = 1, Nr = 1
TH-BPSK, Nt = 2, Nr = 1

TH-BPSK, Nt = 2, Nr = 2
DS-BPSK, Nt = 1, Nr = 1
DS-BPSK, Nt = 2, Nr = 1
DS-BPSK, Nt = 2, Nr = 2
(b)

Figure 5.64. TH and DS UWB systems: (a) single user; (b) multiuser (Nu = 5).

spreading sequence. The received monocycle is modeled as the second derivative of

+
the Gaussian pulse [96]. The channel parameters are m = 2 and lL=u 0 1 u (l) = 1,
with delay profile following [100]. The number of fingers for the RAKE receiver is
chosen to be L = 4.

368

Spatial aspects of UWB

100
101

BER

102
103
104
105

10 12 14

16 18

20

Eb /N0 (dB)
TH-BPPM, Nr = 1, R = 1
TH-BPPM, Nr = 1, R = 1/2
TH-BPPM, Nr = 2, R = 1
TH-BPPM, Nr = 2, R = 1/2
TH-BPSK, Nr = 1, R = 1
TH-BPSK, Nr = 1, R = 1/2

TH-BPSK, Nr
TH-BPSK, Nr
DS-BPSK, Nr
DS-BPSK, Nr
DS-BPSK, Nr
DS-BPSK, Nr

= 2, R = 1
= 2, R = 1/2
= 1, R = 1
= 1, R = 1/2
= 2, R = 1
= 2, R = 1/2

(a)

100
101

BER

102
103
104
105

10 12 14

16 18

20

Eb /N0 (dB)
TH-BPPM, Nr = 1, R = 1
TH-BPPM, Nr = 1, R = 1/2
TH-BPPM, Nr = 2, R = 1
TH-BPPM, Nr = 2, R = 1/2
TH-BPSK,Nr = 1, R = 1
TH-BPSK, Nr = 1, R = 1/2

TH-BPSK, Nr
TH-BPSK, Nr
DS-BPSK, Nr
DS-BPSK, Nr
DS-BPSK, Nr
DS-BPSK, Nr

= 2, R = 1
= 2, R = 1/2
= 1, R = 1
= 1, R = 1/2
= 2, R = 1
= 2, R = 1/2

(b)
Figure 5.65. TH and DS UWB-MIMO systems with ROD ST codes of dierent rates: (a) single user;
(b) multiuser (Nu = 5).

Thomas Kaiser et al.

369

We can see from Figure 5.64 that MIMO systems outperform SISO systems,
regardless of the modulation and multiple-access techniques. Consider the singleuser case illustrated in Figure 5.64(a). At any fixed SNR, the performances of THBPSK and DS-BPSK systems are close to each other, and both BPSK systems yield
superior performances to TH-BPPM. In Figure 5.64(b), we show the performances
when 5 asynchronous users are active. In low SNR regime, TH/DS-BPSK systems
outperform TH-BPPM scheme, and both BPSK systems yield close performances.
However, due to the multiple-access interference, the BER of TH multiuser systems slightly drop with increasing Eb /N0 , and a high error floor can be noticed at
high SNR. On the other hand, even in multiple-access scenarios, we can still see
considerable improvement of DS-BPSK ST system.
In Figure 5.65, we compare the performance UWB systems employing ROD
ST codes with full and half rates. Both figures illustrate that BPSK provides lower
BER than BPPM scheme, regardless of the code rate. From Figure 5.65(a), we can
see that the performances of full- and half-rate ROD codes are close to each other
for every modulation schemes. This supports the theoretical results that for single user system, decreasing the rate of ROD ST code does not improve the performance. Unlike the single-user case, the results in Figure 5.65(b) confirm our
expectation that when the code rate is lower, both TH-BPSK and TH-BPPM multiuser systems achieve better performances, especially in high SNR regime. However, for DS-BPSK multiple-access systems, the BER improvement obtained from
reducing the code rate is insignificant. This is because for Nu = 5, the eect of
multiuser interference to DS system is considerably small, and ROD ST code provides close to maximum achievable performance without decreasing the code rate.
5.7.5. Multicarrier UWB-MIMO system
In this section, multiantenna UWB-OFDM system model, including transceiver
structure and SF coding, will be described. Based on this model, we will quantify
the performance criteria in terms of diversity order and coding gain. Finally, some
simulation results will be shown for the SF-coded UWB-OFDM systems.
5.7.5.1. Transceiver design
We consider a scenario that has been proposed for IEEE 802.15.3a WPAN standard. The available UWB spectrum of 7.5 GHz is divided into several subbands,
each with bandwidth Bsb = 528 MHz (multiband (MB) OFDM). Each user uses
one subband per transmission. At each user, the signals from all transmit antennas share the same subband. Within this subband, the OFDM modulation with
N = 128 subcarriers is used at every transmit antenna. Denoting the OFDM
block period by T, the frequency separation between two adjacent subcarriers is
f = Bsb /N = 1/T. In what follows, we will investigate a peer-to-peer MB-UWB
MIMO-OFDM communication system with Nt transmit and Nr receive antennas,
as shown in Figure 5.66.

370

Spatial aspects of UWB

FFT

0
D/A
add
prefix ..
exp( j2 f s t)
.

SF
encoder

FFT
add
prefix

D/A

A/D

exp( j2 f s t)

Nt 1 Nr 1

exp( j2 f s t)

A/D

exp( j2 f s t)

..
.

FFT
remove
prefix
FFT
remove
prefix

Maximum
likelihood
decoder

Figure 5.66. A peer-to-peer MB-UWB MIMO-OFDM communication system.

First, the transmitter partitions the input bit sequence into blocks of Nb bits.
Each block is mapped onto an N Nt SF codeword matrix

D=

d0 (0)
d0 (1)
..
.

d1 (0)
d1 (1)
..
.

d0 (N 1)

d1 (N 1)

..

dNt 1 (0)
dNt 1 (1)
..
.
dNt 1 (N 1)

(5.219)

where di (k) represents the complex symbol to be transmitted over subcarrier k by

transmit antenna i. Next, the transmitter applies N-point IFFT over each column
of D. In order to avoid the ISI, a cyclic prefix is added to the output of the IFFT
processing before passing the discrete OFDM signal through a digital-to-analog
converter. The analog baseband OFDM signal is then modulated by the subband
carrier frequency. As depicted in Figure 5.66, f s denotes the subband carrier frequency of the sth OFDM symbol. Finally, the modulated MB-OFDM signals are
simultaneously sent from all Nt transmit antennas. Since Nb information bits are
transmitted in T seconds, the transmission rate is R = Nb /T = Nb Bsb /N.
The receiver performs down conversion, matched filtering, discards the cyclic
prefix, and applies the FFT. In this way, OFDM converts a frequency-selective fading channel into N parallel flat-fading channels whose fading coecients are equal
to the channel frequency response at the subcarriers center frequency. Recall the
channel impulse response in (5.174). Its corresponding frequency response is given
by

Hij( f ) =

L
1

i j (l) exp j 2 f (l) ,

j  1.

(5.220)

l=0

Consequently, the received signal at the kth subcarrier at receive antenna j is


j

y (k) =

Nt 1
E
di (k)H i j (k) + n j (k),
Nt i=0

(5.221)

Thomas Kaiser et al.

371

where n j (k) is complex additive white Gaussian noise with zero mean and power
spectral density N0 /2 per dimension, and
H i j (k) =

L
1

i j (l) exp j 2k f (l)

(5.222)

l=0

is the frequency response of the channel at subcarrier k. The SFcodeword matrix

is normalized to have energy E[D2 ] = NNt , and the factor E/Nt guarantees
that the average transmitted energy is E, independent of the number of transmit
antennas.
For subsequent performance evaluation, we provide the matrix representation
of (5.221) as follows. Based on the formulation in [82], we rewrite the received
signal at receive antenna j in the matrix form as

j

Y =

E
SD H j + N j ,
Nt

(5.223)

where SD is the data matrix of a form

%

SD = S0

S1

&

SNt 1 ,

(5.224)

in which


Si = diag di (0), di (1), . . . , di (N 1) ,

(5.225)

and diag() forms a diagonal matrix from its arguments. The channel vector H j of
size NNt 1 is given by
Hj =

%

H0 j

T

H1 j

T

H(Nt 1) j

T &T

(5.226)

(5.227)

where
%

Hi j = H i j (0)

H i j (1)

H i j (N 1)

&T

The received signal vector Y j and the noise vector N j are of size N 1 whose kth
elements are y j (k) and n j (k), respectively. From (5.223), the decision rule at the
ML decoder can be stated as
 = arg min
D
D


N
r 1 
j =0


Y


2

 
E

INr SD H j 
.

Nt

(5.228)

372

Spatial aspects of UWB

5.7.5.2. Performance analysis

Following the computation steps as in [97], the PEP conditioned on the channel
matrix is given by


 |H
P D  D


j

;
<
<
<
= Q =

2Nt

1
exp
2

'

N
r 1


2
S H j 

j =0

Nr 1



S H j 2 ,

4Nt j =0

(5.229)

where = E/N0 is the average SNR at each receive antenna, and S = SD

SD represents the dierence of codeword matrices. For convenience, let us denote an Nt L 1 channel vector a j = [(a0 j )T (a(Nt 1) j )T ]T , in which ai j =
[i j (0) i j (1) i j (L 1)]T . According to (5.222), we can express (5.226) as


H j = INt W a j ,

(5.230)

where W is an N L Fourier matrix, defined as

W=

(0)
..
.

(1)
..
.

(L1)
..
.

(N 1)(0)

(N 1)(1)

..

(N 1)(L1)

(5.231)

in which = exp( j 2 f ). As a consequence, the term S H j 2 in (5.229) can

be expressed as
 


 
S H j 2 = S IN W a j 2
t


 j
 H 
= aj
INt WH H
S S INt W a .

(5.232)

Assume that the MIMO channel is spatially uncorrelated, that is, the channel taps
i j (l) are independent for dierent i and j. Also, let each link have thesame
 power
i j (l)|2 ] = (l). Denote R1/2 = I diag(
|

delay
profile,
that
is,
E[
Nt
0
1

L1 ), and let q j = (R1/2 )1 a j . We can show that the elements of q are uncorrelated Nakagami-m distributed with normalized power = 1. Substituting
a j = R1/2 q j into (5.232), we have





  
S H j 2 = q j H IN UH H S IN U q j ,
S
t
t


(5.233)

Z

where
U  W diag



0 , 1 , . . . , L1 .

(5.234)

Thomas Kaiser et al.

373

Since Z is a symmetric matrix of size Nt L Nt L, it can be decomposed into Z =

VVH , where V  [v0 v1 vNt L1 ] is a unitary matrix, and = diag{0 , 1 , . . . ,
Nt L1 } is a diagonal matrix whose diagonal elements are the eigenvalues of Z.
After some manipulations, we arrive at
N
t L1


 j 2
S H j 2 =
n n  ,

(5.235)

n=0
j

where n  vnH q j . Since V is unitary and the components of q j are i.i.d., {n } are
independent random variables whose magnitudes are approximately Nakagamim
 distributed with parameter m
 = Nt Lm/(Nt Lm m + 1) and average power
 is approximately
= 1 (see [95, page 25]). Note that when Nt L or m is large, m
one. Substituting (5.235) into (5.229) and averaging (5.229) with respect to the
j
distribution of |n |2 (see (5.203)), the resultant upper bound of the PEP is given
by



P D  D

 r
 mN

N@
t L1 
n=0

1
1+
n
4Nt m


(5.236)

Assume the rank and nonzero eigenvalues of matrix Z to be r and {n }nr =10 , respectively. For high SNR environments, the bound can be simplified to


 Gc
P D  D

4Nt

1Gd

(5.237)

in which
Gc = (m)


' r 1
@

(1/r

(5.238)

n=0

defines the coding gain, and

Gd = mrN
 r

(5.239)

defines the diversity gain of UWB MIMO-OFDM system. From (5.233), we observe that


rank(Z) = rank S INt U

H

S INt U



 

= rank S INt U .

(5.240)

Since S is of size N NNt whereas U is of size N L, rank(S (INt U))

min{N, Nt L}. Therefore, we can quantify the maximum achievable diversity gain
as
)

Gd = min mN
 rN .
 t Nr L, mN

(5.241)

374

Spatial aspects of UWB

5.7.5.3. MC-UWB using SF codes

In this section, we compare the performance of UWB MB-OFDM systems employing various SF codes. Each SF code has a form
%

D = GT0

GT1

GTP1

&

0T(N PNt )Nt ,

(5.242)

where {1, 2, . . . , L} denotes the frequency diversity order, and 0mn stands for
all-zero matrix of size m n. The matrix G p is of size Nt Nt whose structure depends on particular SF code. Let us consider a system with two transmit antennas.
Based on the SF code in [82], G p is given by


G p = INt 11

x p,0
x
p,1

x p,1
,
xp,0

(5.243)

in which 11 is an all-one matrix of size 1, and x p,i is selected from BPSK or

QPSK constellations. We also exploit a full-rate SF code [83] whose G p is specified
as


x p,0
G p = Nt
01

01
,
x p,1

(5.244)

where 1 matrix x p,i is defined as follows. For simplicity, we omit the subscript p
and denote K = Nt . Let s = [s1 s2 sK ] be a vector of BPSK or QPSK symbols.
The 1 K matrix x  [x0T x1T ] is specified as


1
x = sV 1 , 2 , . . . , K ,
K

(5.245)

in which V is a Vandermonde matrix16 with k = exp(j(4k 3)/(2K)) for K =

2s (s 1) and k = exp(j(6k 1)/(3K)) for K = 3 2s (s 0) [83].
Based on the same channel model as in Section 5.7.4, the performances of
MB-OFDM system with spectral eciency of 1 bit/s/Hz are illustrated in Figures
5.67(a) and 5.67(b). In Figure 5.67(a), we show the performance of the SF codes
with a frequency diversity of order = 2. We observe that regardless of particular SF coding scheme, the spatial diversity gained from UWB-MIMO architecture
does improve the system performance significantly. In addition, the performance
can be further improved with the choice of SF code and permutation scheme.
In Figure 5.67(b), we compare the performance of MB-OFDM system with
dierent frequency diversity orders. Here, we utilize the full-rate SF code [83] with
= 2, 3, and 4. We can see that by increasing the frequency-diversity order, the
system performance can be improved. This observation is in accordance with our
theoretical result in (5.237). Therefore, with a properly designed SF code, we can
eectively exploit both spatial and frequency diversities in UWB environment.
16A Vandermonde matrix with variables , , . . . , is a K K matrix whose kth (k =
1 2
K
1, 2, . . . , K) row is defined as [1k1 2k1 Kk1 ].

Thomas Kaiser et al.

375
100
101

BER

102
Nt = 1
Nr = 1

103
Nt = 2
Nr = 1

Nt = 2
Nr = 2

104

10 12 14

16 18

20

Eb /N0 (dB)
SF code with random permutation
Full-rate SF code with random permutation
Full-rate SF code with optimum permutation
(a)
100
101

Nt = 2
Nr = 1

BER

102

Nt = 1
Nr = 1

103
Nt = 2
Nr = 2

104

=2
=3

10 12 14

16 18

20

Eb /N0 (dB)

=4
(b)
Figure 5.67. Performance of MB-OFDM: (a) with dierent SF codes; (b) with dierent diversity orders.

5.7.6. Summary
In this section, we provide performance analysis for multiantenna UWB communication systems with various transmission schemes. Based on Nakagami-m
frequency-selective fading channels, the performance metrics (diversity and coding gains) of UWB-MIMO systems are quantified regardless of specific coding

376

Spatial aspects of UWB

schemes. For SC-UWB, we show that the use of ST coding in combination with
RAKE receiver architecture is able to exploit spatial as well as multipath diversity,
inherent in UWB environments. Comparing various modulation techniques, DSUWB MIMO scheme is shown to achieve superior performance in both singleuser and multiuser scenarios. For MC-UWB, we show that SF codes in addition
to OFDM modulations are able to provide spatial as well as frequency diversities,
thereby significantly improving the system performance.
5.8. Channel capacity of MIMO UWB indoor wireless systems
5.8.1. Introduction
In this section, we will investigate the problem of channel capacity calculation for
UWB indoor wireless systems. In [101103], the channel capacity of M-ary PPM
UWB systems is discussed. However, the random fading property of UWB channel gains is not taken into account there. To the best of the authors knowledge,
there are no other reports appeared in the literature to address this problem up
to now. We aim to make further development towards this problem in the sense
of information theory. Our concern will be focused on the systems having multiple transmit and receive antennas (MIMO). The work by Foschini and Gans [104]
and Telatar [105, 106] has shown that for a conventional narrowband fading channel, a MIMO system can achieve capacity of approximately min(NT , NR ) times
that of a separate channel, where NT and NR are the numbers of transmit and receive antennas, respectively. The measured data rates are found to be about 30%
smaller than that being anticipated from an idealized model [107109], showing
that the capacity gains over single-antenna systems are still promising. Therefore,
it is of importance to investigate how the capacity of MIMO UWB channels behaves.
As channel capacity depends on the statistical properties of the channel, we
need first to investigate the statistics for the multipath fading of UWB indoor wireless system, which is our starting point.
In Sections 5.8.3, 5.8.4, 5.8.5, and 5.8.6, we discuss the frequency-flat case,
while in Sections 5.8.7, 5.8.8, and 5.8.9, we consider the frequency-selective case.
In Section 5.8.10, we investigate the eect of line-of-sight transmission on UWB
channel capacity. Note that in the frequency-flat case, a UWB communication
channel diers from a classic narrowband frequency-flat communication channel in that the statistics of the amplitude fading for both cases is dierent. Note
also that the frequency-flat case is rather unrealistic but serves as a good starting
point.
This section gives a summary of our researching results on this topic. The
details will be provided elsewhere.
Notation 1. Throughout this section, I is an identity matrix whose dimension is
implied by context, Pr denotes the probability of an event, PA (x) and pA (x) represent the cumulative distribution function (cdf) and probability density function
(pdf), respectively, of random variable A, EA ( f (A)) stands for the expectation of

Thomas Kaiser et al.

377

a function of random variable A, taking expectation over the statistics of A, and

tr represents the trace of a square matrix. The log function is understood as the
natural logarithm of its argument. Hence, the unit of the channel capacity is in
nats.
5.8.2. A statistical model for the multipath fading of
ultra-wideband indoor wireless systems
For UWB indoor wireless communication systems, its propagation model can be
often described by the real impulse response [110]:
Npath (t)

h(t, ) =

al (t) l (t) ,

(5.246)

l=1

where t and are the observation time and delay time of the impulse response,
respectively, Npath (t) is the number of multipath components at time t, al (t) and
l (t) stand for the random time-varying amplitude and arrival-time of the received
signal via the lth path at time t, respectively, and denotes the Dirac delta function.
It is clear that a UWB wireless channel can be characterized by the statistics of the
random variables (or processes) Npath (t), l (t), and al (t). The statistical models for
the narrowband indoor wireless channels have been well studied and some models
have been widely accepted, see, for example, the survey paper [111]. However,
these models cannot be applied directly to UWB wireless channels, since a UWB
signal with its GHz bandwidth can resolve paths with a path delay dierence on the
order of one nanosecond. Hence, the number of scatterers within one resolvable
path is in the order of 2 or 3 (when considering a typical oce environment and
employing a ray-tracing model) [112]. With this phenomenon being considered,
several models for UWB channels have been proposed recently, for example, see
[110, 113]. In [113], the so-called POCA-NAZU model is presented to describe the
small-scale multipath amplitude fading for the received UWB signals. In [110],
a stochastic tapped-delay-line (STDL) propagation model for the UWB indoor
channel was proposed, as shown in (5.246). The parameters of the STDL model
are extracted from the measured impulse responses. It is shown there that the bestfit distribution of the small-scale magnitude statistics is Nakagami distribution,
diering from a Rayleigh distribution. Since the model in [110] is extracted based
on experimental data, we will adopt this model as the basis of our analysis in this
section. Note that now a standard channel model for UWB indoor communication
systems is drafting [114].
Let X(t) denote the transmitted UWB signal, which is propagated through an
indoor multipath fading channel, and let Y (t) denote the received signal. Then the
relationship between X(t) and Y (t) can be modeled by
Npath (t)

Y (t) =

l=1

al (t)X t l (t) + N(t).

(5.247)

378

Spatial aspects of UWB

Let us denote al as
al = l l ,

(5.248)

where l := sign(al ), the sign of al , and l := |al |, the magnitude of al . It has

been shown in [110] that the statistics of l in (5.247) can be approximated by a
Nakagami distribution, the general form of whose pdf is as follows:

2mm x2m1 mx2 /l

m e
pl (x) = (m)l

when x 0,
when x < 0,

1
m ,
2

(5.249)

where denotes the Gamma function, l = E (a2l ), and m = [E (a2l )]2 / Var[a2l ].
In general, m should be a function of l. For example, in [110], m is modeled as a
random variable, whose pdf is not specified. For that reason, we assume a constant
m (see also the comment on page 16). In this section, we substitute m with another
parameter by simply defining = 2 m. Hence, it is clear that 1. By so doing
the pdf of |al | in (5.249) can be rewritten as



/2

pl (x) = 2l

1
2
x1 e(x /2l )
(/2)

when x 0,

(5.250)

when x < 0.

It is also known that the power of the amplitude fading is exponentially decreasing with the excess delay. Therefore, it is reasonable to assume that l varies
with l according to the following model:
l = rl1 ,

(5.251)

where r < 1 is a constant. According to the data reported in [110], r takes value
between 0.93 and 0.97. Of course, the value of r is determined by the scenario of
the communications.
For the model regarding l , no discussion has been addressed in the literature,
to the best of the authors knowledge. It seems justifiable to assume that
1
1
pl (x) = (x 1) + (x + 1),
(5.252)
2
2
that is, l takes the signs +1 and 1 with equal probability.
It is shown in [110] that the excess delay of lth delay bin, l , is lognormally
distributed. However, we will assume that l is deterministic in this section to make
our analysis tractable, which is also reasonable based on the following argument.
Recalling the exposition in [115] on the characterization of a wideband frequency-selective fading channels, a tapped-delay-line channel is also obtained as
follows:
Y (t) =

L

l=1

al (t)X t (l 1) + N(t),

(5.253)

Thomas Kaiser et al.

379

where = 1/WX , with WX being the bandwidth of the transmitted signal X, and
L =  Tm WX  +1  WX / fc  +1, with Tm and fc being the multipath spread and
the coherence bandwidth of the channel, respectively. In channel capacity calculation, we can assume that the transmitted signal will occupy the whole bandwidth
of the channel frequency band. Hence, we will also use WX to denote the bandwidth of the channel. In [115], model (5.253) is derived from frequency domain.
It has a direct interpretation in time domain, that is, the output of the receiver
consists of the signals coming from several delay bins. Denote the resolution of the
path delay for each delay bin as
. Then it is easy to see from this interpretation
that L =  Tm /
 + 1. Thus we can see that
1/WX = .
Even though model (5.253) is derived from wideband channels, it applies to
UWB channels as well if the random nature in the excess delay is ignored. On the
other hand, if the fingers of UWB RAKE receiver are large enough, it is reasonable
to take l in model (5.247) as deterministic since we can simply choose l as (l 1)
times of the width of one delay bin.
Notice that another important dierence between models (5.247) and (5.253)
is that the statistics of al (t) for the two cases is dierent in general. In UWB case,
since the bandwidth of a UWB signal is very large (up to several GHz), a UWB receiver can resolve paths with a path delay dierence on the order of one nanosecond or even less than that. Thus the central limit theorem cannot apply to UWB
case and al for UWB channels will have its own distribution. Therefore, we will
adopt (5.253) with al having the distribution composed by (5.250) and (5.252) as
the basic model of our study.
We presume that the communication is carried out using bursts. The burst
duration is assumed to be short enough so that the channel can be regarded as
essentially fixed during a burst, but long enough that the standard informationtheoretic assumption of infinitely long code block lengths is a reasonable idealization. In this quasistatic scenario, it is meaningful to associate a channel capacity
with a given realization of the channel fading.
We further assume that the receiver possesses a complete knowledge of the
instantaneous channel parameters, that is, the realizations of al (t) and their statistics. However, we assume that the transmitter is not aware of the information
about channel parameters. Thus no power control can be applied to the transmitted signals.
Now we are ready to discuss the channel capacity.
5.8.3. The case of single-transmit and single-receive antenna
(SISO)frequency-flat channels
First we study the case of single-transmit and single-receive antenna and for frequency flat channels. Some simplification procedures are needed to make the analysis tractable. Our main assumptions are as follows.
Assumption 1. The channel is frequency flat, that is, L = 1.

380

Spatial aspects of UWB

Assumption 2. The number is assumed to be constant.

Assumption 3. The noise N(t) is zero-mean Gaussian and white with variance N2 .
Assumption 4. The power of the transmitted signal is bounded by PX , that is,
E (X 2 (t)) PX for all t.
Based on Assumption 1, (5.253) is simplified to
Y (t) = a(t)X(t) + N(t).

(5.254)

For notational convenience, we define = 1 since only 1 is of concern.

In Sections 5.7.3 and 5.7.4, we will discuss ergodic capacity and hence assume
that the fading process is ergodic.
Based on Assumption 2 and the result presented in the preceding subsection,
the random variable |a(t)| will be of distribution defined by pdf (5.250).
The channel capacity for an AWGN channel with a given fading amplitude a
is given by
'

a2 PX
.
N2

C |a = WX log 1 +

(5.255)

So the ergodic channel capacity, denoted as Ce , turns out to be



Ce = Ea C |a =

WX
(/2)

log 1 +
0

u /21 u
u
e du,

(5.256)

where
:=

N2

=
,
2 PX
2 SNR

SNR :=

PX
.
N2

(5.257)

SNR defined in (5.257) can be considered as the ratio of signal power (at the receiver side) to noise power. Let us define
J(; ) :=

log 1 +
0

u /21 u
u
e du.

(5.258)

Then we have
Ce =

WX
J(; ).
(/2)

(5.259)

The function J(; ) can be calculated recursively in terms of the following

Thomas Kaiser et al.

381

equations:

J(; ) = e

J(1; ) = erfi

(2)/2

 

  

1 , +
1 J( 2; ),
2
2
2


E 2 log 2 log 2 2 F2 [1, 1], 2,

3
,
2

1

J(2; ) = e (0, ),
(5.260)
where E 0.5772 is Euler constant, and 2 F2 ([1 , 2 ], [3 , 4 ], z), (, z), and
erfi(z) denote the generalized hypergeometric function, incomplete Gamma function, and imaginary error function, respectively, whose definition can be seen in
[116].
5.8.4. The case of multiple-transmit and multiple-receive antennas
(MIMO)frequency-flat channels
In this case, the input-output relation (channel model) can be described by
Y(t) = A(t)X(t) + N(t),

(5.261)

where A(t) = [anm (t)]mY mX is a random matrix characterizing the amplitude

fading of the channel, X(t) RmX and Y(t) RmY are the vectors of transmitted signals and received signals, respectively, mX and mY denote the numbers of
transmit and receive antennas, respectively, and N(t) RmY is the noise vector.
Assumption 5. It is assumed that all anm , n = 1, . . . , mY , m = 1, . . . , mX , are independent.
Assumption 6. The noise N is zero-mean Gaussian with covariance matrix N2 ImY .
Assumption 7. The power of the transmitted signal is bounded by PX , that is,
E (XT (t)X(t)) PX for all t.
Assumption 8. The receiver possesses a complete knowledge of the instantaneous
channel parameters, while the transmitter is not aware of the information about
the channel parameters.
According to the discussion in Section 5.7.2, the magnitude of each element
of matrix A, say |anm |, has the following pdf:

2
(2/)/2 (/2)

p|anm | (x) =

x1 ex

2 /2

when x 0,
when x < 0.

(5.262)

382

Spatial aspects of UWB

The mutual information between X and Y for a given A is


I X; Y | A = H Y | A H Y | X, A = H Y | A H (N).

(5.263)

It is well known that if X is constrained to have covariance Q, the choice of X that

maximizes I(X; Y | A) is Gaussian variable with covariance Q. Thus the channel
capacity for a given fading matrix A turns out to be [117, 105, 106]


C |A = WX log det ImY +

1
AQAT ,
N2

(5.264)

where AT represents the transpose of matrix A. Let us define (Q)=EA [log det(ImY+
(1/N2 )AQAT )]. Then the ergodic channel capacity is given by
Ce = WX max (Q).

(5.265)

tr(Q)PX

Similar to the argument as that in [105, 106], we can show (for the details, see
[118]) that the optimal Q should be Qopt = (PX /mX )I even for the distribution of
A defined in (5.262). Thus we have
!

'

Ce = WX EA log det ImY +

PX
AAT
mX N2

("

(5.266)

To use (5.266) to calculate the channel capacity, we need to know the distribution of the matrix AAT . Unfortunately, this distribution is known only when A
possesses some special distribution (typically normal distribution) if both mX > 1
and mY > 1, see, for example, [119]. Therefore, we will consider some special cases
in the following.
(i) Single-transmit and multiple-receive antennas (SIMO). In this case, mX = 1 and
mY > 1. By using the property of matrix determinant, we can obtain the analytic
expression for the ergodic channel capacity as follows:


WX
 J mY ; .
Ce = 
(/2)mY

(5.267)

(ii) Multiple-transmit and single-receive antennas (MISO). In this case, mX > 1

and mY = 1. The ergodic channel capacity is given by


WX
 J mX ; mX .
Ce = 
(/2)mX

(5.268)

Remark 5.8.1. Notice that when = 2, the fading model for each elements of A reduces to Rayleigh distribution, which corresponds to the classic narrowband wireless communication channel. So we expect that the results obtained for this specific
also recover the results obtained in [105, 106]. This is indeed the case. Substituting = 2 into (5.267) and (5.268), respectively, readily reveals that (5.267) and
(5.268) indeed reduce to [106, equations (9) and (10)].

Thomas Kaiser et al.

383

(iii) Limiting case: infinitely many transmit and receive antennas. In this case, we
suppose that mX , mY , and at the same time max{mX , mY }/ min{mX ,
mY } , which is a constant. Following the similar development of [106, equation (13)], we can show that
Ce
WX
=
lim
mmin mmin
2A2

 +

'

P
log 1 + X2
N

( 



+
1

d,

(5.269)

where mmin = min{mX , mY }, = A2 , = ( 1)2 , and



A2 := E Ai j E Ai j

2

=
1

0
1
2 2 (( + 1)/2)
= 1
,
2

(/2)

if mX mY ,

(5.270)

if mX < mY .

Specifically, if mX = mY , the above formula gives (cf. [105, 106])



Ce
1 + 4 SNR1 1
1
= log(SNR1 ) 1 +
+ 2 tanh1 
,
lim
mX WX mX
2 SNR1
1 + 4 SNR1
(5.271)
where SNR1 is defined as SNR1 := A2 PX /N2 .
Equations (5.269) and (5.271) show that when both the numbers of transmit
and receive antennas are large and they are in the same order, the ergodic channel
capacity of the MIMO UWB wireless systems is proportional to the numbers of either
transmit or receive antennas.
The detailed derivations for (5.267), (5.268), and (5.269) can be found in
[118].
5.8.5. Capacity versus outage probability frequency-flat channels
The results that we have obtained in the previous subsections apply to the case
where the fading matrix is ergodic and there are no constraints on the decoding
delay on the receiver. In practical communication systems, we often run into the
case where the fading matrix is generated or chosen randomly at the beginning of
the transmission, while no significant channel variability occurs during the whole
transmission. In this case, the fading matrix is clearly not ergodic. We suppose
that the fading matrix still has the distribution defined in the previous subsections. In this case, it is more important to investigate the channel capacity in the
sense of outage probability. An outage is defined as the event that the communication channel does not support a target data rate. Thus, according to [120], outage
probability, denoted by Pout (R), is defined as follows. With a given rate R, we associate a set R in the space of fading matrix A. The set is the largest possible set for
which C , the capacity of the compound channel with parameter A R , satisfies
/ R }. Thus it
C R. The outage probability is then defined as Pout (R) = Pr{A

384

Spatial aspects of UWB

is clear that
)

Pout (R) = Pr A
/ R = Pr C |A < R = Pr C |A R ,

(5.272)

that is, the outage probability can be actually viewed as the cumulative distribution
function of the conditional Shannon capacity. Notice that the last equality of the
above equation follows from the fact that C(X; Y | A) is a continuous function of
continuous random variable A.
Based on the above discussion, we can evaluate the outage probability for
frequency-flat UWB channels for the following three cases.
(i) SISO case. In this case, we suppose that Assumptions 1, 2, 3, and 4 hold. Let us
define = a2 . It is easy to obtain the cdf of as follows:
P (x) = Pr{ x} =

x
0

/2

1
1

x/21 ex/2 dx =
,
x ,
(/2)
(/2) 2 2
(5.273)

where (, z) is the incomplete Gamma function, whose definition can be seen in

[116]. Therefore, from (5.255), it follows that
#

Pout (R) = Pr


N2  R/WX 
1

e
1
, eR/WX 1 .
=
PX
(/2) 2

(5.274)

(ii) SIMO case. In this case, we suppose that Assumptions 1, 2, 4, 5, and 6 hold.
Similarly we can obtain the outage probability as




1

 mY , eR/WX 1 .
Pout (R) = 
2
(/2)mY

(5.275)

(iii) MISO case. In this case, we suppose that Assumptions 1, 2, 3, 5, 7, and 8 hold.
We have




1

 mX , mX eR/WX 1 .
Pout (R) = 
2
(/2)mX

(5.276)

5.8.6. Numerical resultsfrequency-flat channels

In this subsection, we will investigate the variation of channel capacity with respect to various kinds of parameters. Suppose WX = 1. So the unit of the channel
capacity is nats/s/Hz.
Figure 5.68 depicts the variation of channel capacity Ce with respect to the
number for SISO case. We can see from this figure that even though the channel
capacity increases with , the quantities increased is not large compared to the base
case ( = 1), especially when 10. For example, for the case of SNR = 0 dB, when
is increased from 1 to 10, Ce increases (0.6695 0.5335)/0.5335 = 25.5%, while

Thomas Kaiser et al.

385

Channel capacity Ce

2.5
2
1.5
1
0.5
0
0

10

15

20

25

30

SNR = 10 dB
SNR = 0 dB
SNR = 10 dB

Channel capacity Ce

Figure 5.68. Variation of ergodic channel capacity Ce (in nats/s/Hz) with for SISO case.

1
0.8
0.6
0.4
0.2
0
0

10

15

20

mY
=2
=4
=6
Figure 5.69. Variation of ergodic channel capacity Ce (in nats/s/Hz) with the number of receive antennas mY for SIMO case (SNR = 10 dB).

when is increased from 10 to 30, Ce increases only (0.6850 0.6695)/0.5335 =

2.31%.
Figure 5.69 shows the relationship between channel capacity and the number
of receive antennas for SIMO case. It can be seen from this figure that the capacity
increases with mY almost logarithmically. This phenomenon is similar to the corresponding one in the case of narrowband Rayleigh fading channels (cf. [105, 106,
Example 3]).
Figure 5.70 shows the relationship between channel capacity and the number
of transmit antennas for MISO case. It is interesting to observe from this figure
that in the case of large SNR the capacity increases with mX rapidly when mX
is small (mX 6), however, the increase is very slow when mX becomes large
(mX > 6). This phenomenon is dierent from the one in the case of narrowband
Rayleigh fading channel, see [105, 106, Example 4], where it is found that Ce does

386

Spatial aspects of UWB

0.0955
Channel capacity Ce

0.095
0.0945
0.094
0.0935
0.093
0.0925
0.092
0.0915

10

15

10

15

mX
=2
=4
=6
(a)
2.4
Channel capacity Ce

2.35
2.3
2.25
2.2
2.15
2.1
2.05
2

5
mX
=2
=4
=6
(b)

Figure 5.70. Variation of ergodic channel capacity Ce (in nats/s/Hz) with the number of transmit
antennas mX for MISO case. (a) SNR = 10 dB. (b) SNR = +10 dB.

not change with mX when mX 2. An important phenomenon can also be observed by comparing Figures 5.70(a) and 5.70(b), that is, when the signal-to-noise
ratio is low, the benefit obtained by distributing the available power to dierent
transmit antennas is very limited as far as average capacity is concerned.
From Figures 5.69 and 5.70, we can see that increasing the number of receiver
antennas can obtain more benefit in channel capacity than increasing the number
of transmit antennas. Principally, the channel capacity could be increased indefinitely by employing a large number of receive antennas, but it appears to increase
only logarithmically in this number; while employing 3 5 transmit antennas

Thomas Kaiser et al.

387

can approach the best advantage of the multiple transmit antenna systems (for
the case of single receive antenna and under the assumption of frequency nonselectiveness). The reason for this phenomenon is twofold. First, the power is constrained to be a constant, for dierent mX , among all the transmit antennas, while
no such constraint is applied to receive antennas. Second, it is assumed that the
receiver possesses the full knowledge about the channel state.
5.8.7. SISO casefrequency-selective channels
Now we begin to investigate frequency-selective UWB channels. We first study
SISO case. Equation (5.253) is the model we will consider in this subsection. Let
SX ( f ) stand for the power spectral density (PSD) functions of the process X(t).
Again we need some simplifying assumptions, which are as follows.
Assumption 9. The number is assumed to be constant and the same for all delay
bins.
Assumption 10. The amplitude fading al (t), l = 1, . . . , L, are assumed to be independent to each other and has the same distribution, as described by (5.250) and
(5.252).
Assumption 11. The noise N(t) is zero-mean white Gaussian with double-sided
PSD being N0 .
Assumption 12. The power of the transmitted signal is bounded by PX , that is,


E X 2 (t) = 2

 WX /2
WX /2

SX ( f )df PX .

(5.277)

Based on the assumption made in the end of Section 5.7.2, the dependence of
the fading al (t) on time can be neglected for our analysis. Hence, we will simply
denote al (t) as al . For notational convenience, we use the variable  to denote the
L-tuple (a1 , . . . , aL ), that is,  = (a1 , . . . , aL ). Define
(t) =

al X t (l 1)

(5.278)

l=1

and denote the PSD of as S ( f ). Now (5.253) can be written as

Y (t) = (t) + N(t).

(5.279)

To apply the classical Shannon capacity formula to model (5.279), we can divide
the whole frequency band [WX /2, +WX /2] of S ( f ) to many small bands with
bandwidth f . From frequency f to f + f , the spectrum of can be considered

388

Spatial aspects of UWB

as flat with its value being approximated by S ( f ). Therefore, for this subchannel,
the mutual information I| , conditioned upon , is


I| = f log 1 +

S ( f ) f
N0 f



S ( f )
.
= f log 1 +

N0

(5.280)

Thus, taking the limit f df in (5.280), we obtain the conditional mutual information, denoted as I| , between X and Y as follows:
I | =

 WX /2

WX /2

log 1 +

S ( f )
df .
N0

(5.281)

So the conditional channel capacity C | is given by the maximization of I| subject to the power constraint (5.277), that is,
C | =  W

X /2

WX /2

Define H( f ) =

+L

l=1 al

 WX /2

max
SX ( f )df PX /2

WX /2

log 1 +

S ( f )
df .
N0

(5.282)

e j 2 f (l1) . Then we have



2

S ( f ) = SX ( f )H( f ) .

(5.283)

From (5.282) and (5.283), it is clear that the capacity is achieved when X(t) is such
that, among other conditions, the equality for power constraint holds.
If the information about , and hence about H( f ), is available to the transmitter, the result of [117, Section 8.3] can be used to obtain the optimal SX ( f ) to
achieve C | , which is as follows:

N0



H( f )2
SX ( f ) =

when f F ,

(5.284)

when f
/ F ,

where F is the range of f for which N0 / |H( f )|2 and is a constant which
satisfies the following equation:
'


f F [WX /2, +WX /2]

N
P
 0 2 df = X .
H( f )
2

(5.285)

The above solution for the optimal SX ( f ) is actually the frequency-domain version
of water-filling algorithm.
In our case, since the transmitter is unaware of the fading information , the
above algorithm does not apply. Now let us consider the design for optimal SX ( f )
 W /2
which should maximize WX X /2 log(1+SX ( f )|H( f )|2 /N0 )df . Since no information
on |H( f )| is available to the transmitter, it is reasonable to assume, at the transmitter side, that |H( f )| is a constant. Then applying formula (5.284) to the case
of the presumed |H( f )| and noticing the power constraint yield that the optimal

Thomas Kaiser et al.

389

SX ( f ) is uniform:

PX

when f

SX ( f ) = 2WX

WX WX
,
,
2
2

(5.286)

otherwise.

Substituting (5.286) into (5.282) gives

C | =

'

 WX /2
WX /2

2 (

log 1 +

PX H( f )
2WX N0

df .

(5.287)

Define = PX /2WX N0 . Simple manipulation will yield

W
C | = X


0

log 1 +

"2


al cos (l 1)u

l=1

"2

al sin (l 1)u du.

l=1

(5.288)
For the general model of frequency-selective channel (5.278)-(5.279), it seems
dicult to obtain the closed-form expression for C | . In the following, we will
consider the case of L = 2. When L > 2, we can use Monte Carlo method to
evaluate C | .
The case of L = 2. To simplify notation, let us define C | = C | /WX . To obtain
the analytical expression for the capacity, we need further assume that r = 1 here,
that is, 1 = 2 := . Since only two clusters are involved, the errors in the capacity introduced by this simplification can be neglected for the moment. Applying
L = 2 to (5.288), we have


C | = log

1 + 1 + 2 +



1 + 1 + 2
2

2

42 1 2

(5.289)

where 1 and 2 are defined as 1 = a21 and 2 = a22 , respectively.

By a long and tedious computation, we can show that
!

1 

( 1)!

 1 e1 (x)/2
PC | (x) = 1
2
2 (/2) ( + 1)/2
l=0


4
2
'

1
2 (/2)


 1 (x)
0 (x)

l

e/2

l1 (x)
l!


"

/2

1 ex 2 ex ex 1

 (

4 1 ex 2 ex ex 1
1
F
, ; + 1;
2 2 2
2

d,
(5.290)

390

Spatial aspects of UWB

where 0 (x) = (ex 1)/, 1 (x) := 2(ex ex/2 )/, and F(1 , 2 ; 3 ; z) denotes the
hypergeometric function, whose definition can be found in [116].
The detailed derivation for (5.290) can be found in [118].
Having the expression for PC | (x) at hand, the outage probability is simply
given by


Pout (R) = PC| (R) = PC |

R
,
WX

(5.291)

where R is the communication rate supportable by the channel with a probability

of 1 Pout (R), or in other words, with an outage probability of Pout (R).
5.8.8. MIMO casefrequency-selective channels
In this case, the input-output relation can be generally described by the following
equation:
Y(t) =

Al X t (l 1) + N(t),

(5.292)

l=1

where X(t) and Y(t) are mX - and mY -dimensional vectors, respectively. Let SX ( f )
denote the power spectral density matrix of the process X(t).
In addition to Assumption 6, we further need the following assumptions.
Assumption 13. The amplitude fading matrices Al , l = 1, . . . , L, are assumed to
be independent, and all the entries of Al , l = 1, . . . , L, are also assumed to be
independent. Suppose Al = [al,mn ]mY mX with al,mn = l,mn l,mn , l = 1, . . . , L,
m = 1, . . . , mY , n = 1, . . . , mX . Then l,mn and l,mn assume distributions as specified by (5.252) and (5.250), respectively, with l being governed by the model
(5.251).
Assumption 14. The power of the transmitted signal is bounded by PX , that is,


E XT (t)X(t) = 2

 WX /2
WX /2

tr SX ( f ) df PX ,

(5.293)

where WX is the bandwidth of the transmitted signal X.

= (A1 , . . . , AL ). Based on the same idea as that in the derivation of
Define 
denoted as
(5.288), we can show that the channel capacity conditioned upon ,
C | , is given by
W
C | = X


0

log det ImY

L

+
Al e j(l1)u
mX l=1

"!

L

l=1

The details for the derivation of (5.294) are given in [118].

ATl ej(l1)u

"

du.
(5.294)

Thomas Kaiser et al.

391

Since it is extremely dicult, if possible, to obtain the closed-form expression

for the integral in (5.294), we use Monte Carlo method instead to evaluate the
distribution of C | , denoted as PC| (x). In the simulation, it is preferable to first
find the distribution of C | := C | /WX .
Having PC | (x), the outage probability of the channel capacity can be obtained through the equation


Pout (R) = PC| (R) = PC |

R
.
WX

(5.295)

5.8.9. Numerical resultsfrequency-selective channels

In this subsection, we will investigate the capacity results for dierent system configurations, which depend on several factors such as the number of transmit and
receive antennas (mX and mY , respectively), frequency-selectivity (characterized
by L), and the ratio between signal power and noise power (), and so forth.
In all the simulations except for the case stated otherwise, it is assumed that
= 4, 1 = 1, and r = 0.95. In Monte Carlo method, we first generate 20 000
random channel realizations, then use formula (5.294) to calculate the channel
capacity for a given realization of the channel, and finally find the cdf of the channel capacity. As aforementioned, the parameter is chosen to be a constant in this
section. Of course, the case for the dependence of on l can be also included in
the simulation if its exact model is given.
All results, except those for the case of SISO and L = 2, will be obtained based
on Monte Carlo simulations, while for the case of SISO and L = 2, we calculate
the capacity via two approaches, one is analytic method where parameter r is approximated by 1, and another is Monte Carlo method where r equals 0.95.
We first compare the results obtained by Monte Carlo method to those obtained through analytic method for the case of SISO and L = 2. The result is plotted in Figure 5.71, which shows that the dierence between these two methods is
negligible.
Figure 5.72 shows the eect of frequency-selectivity on the channel capacity,
where the case of mX = mY = 1 is investigated. The curve labelled with L = 1
corresponds to the case of frequency-flat channel. It can be observed from this figure that the channel capacity for a given outage probability increases with L and
the fluctuation (or variance) of the channel capacity decreases with L. Comparing
Figures 5.72(a) and 5.72(b), we can see that this phenomenon is displayed more
clearly in the scenario of high signal-to-noise ratio. The reason that the channel capacity increases with L is very simple: the power of the received signal is increased
with the increase of L, while the reason that the fluctuation of the channel capacity
decreases with L could be understood in such a way: due to more and more echoes
received, the channel behaves more and more deterministically.
In Figure 5.72, we have used uniform power spectrum distribution (PSD) for
the transmitted signal since it is assumed that the transmitter is not aware of the
information about channel parameters. On the other hand, if the transmitter has

Spatial aspects of UWB

Outage probabillity Pout

392
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

= 20 dB
= 15 dB
= 10 dB
= 5 dB
= 0 dB

0.2 0.4 0.6 0.8

1.2 1.4 1.6

R/WX
Analytic method
Monte Carlo method

Outage probabillity Pout

(a)
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

= 0 dB
= 5 dB
= 10 dB
= 15 dB
= 20 dB

R/WX
Analytic method
Monte Carlo method
(b)
Figure 5.71. Outage probability Pout versus transmission rate R/WX (in nats/s/Hz) for dierent .
Comparison between analytic method and Monte Carlo method (L = 2, mX = mY = 1). (a) For the
case of low SNR. (b) For the case of high SNR.

the knowledge about channel parameters, optimal PSD can be employed. It is interesting to compare the capacity for the two cases: uniform and optimal PSD
schemes. The result is plotted in Figure 5.73. From this figure we can see that when
the signal-to-noise ratio is low, optimal PSD scheme can increase the channel capacity considerably compared to uniform PSD scheme, while when the ratio is
high, the benefit of optimal PSD scheme can be hardly seen compared to uniform one. This phenomenon is demonstrated more clearly in Figure 5.74, where
the ergodic capacity is used for illustration. Let Cuni and Copt denote the ergodic
capacity of the system for uniform and optimal PSD schemes respectively. We use
Cuni /Copt as an index for the eciency of uniform PSD scheme relative to optimal one. We can see from this figure that the eciency of uniform PSD scheme is

Outage probabillity Pout

Thomas Kaiser et al.

393
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

0.2

0.4

0.6

0.8

R/WX
L=1
L=2
L=3
L=4

L=5
L=6
L=7
L = 50

Outage probabillity Pout

(a)
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

R/WX
L=1
L=2
L=3
L=4

L=5
L=6
L=7
L = 50
(b)

Figure 5.72. Outage probability Pout versus transmission rate R/WX (in nats/s/Hz) for dierent L.
Eect of frequency-selectivity on the channel capacity (mX = mY = 1). (a) = 10 dB. (b) =
+10 dB.

lower than 0.61 when the signal-to-noise ratio is low ( 20 dB). This says that
the transmission rate can be increased roughly more than 1.6 times if the optimal
PSD scheme instead of the uniform one is adopted for the aforementioned case.
But when +10 dB, Cuni /Copt almost approaches one.
From the results shown in Figures 5.73 and 5.74, we can say that one should
consider measures, such as frequency domain water-filling algorithm, to take full
advantage of frequency-selectivity if the signal-to-noise ratio of the UWB systems
is low.

Spatial aspects of UWB

Outage probabillity Pout

394
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

0.2

0.4

0.6

0.8

1.2

R/WX
Opt. PSD, L = 2
Uni. PSD, L = 2
Opt. PSD, L = 7

Uni. PSD, L = 7
Opt. PSD, L = 50
Uni. PSD, L = 50

Outage probabillity Pout

(a)

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

1.5

2.5

3.5

4.5

5.5

R/WX
Opt. PSD, L = 2
Uni. PSD, L = 2
Opt. PSD, L = 7

Uni. PSD, L = 7
Opt. PSD, L = 50
Uni. PSD, L = 50

(b)
Figure 5.73. Outage probability Pout versus transmission rate R/WX (in nats/s/Hz) for dierent L.
Comparison between optimal PSD and uniform PSD (mX = mY = 1). (a) = 10 dB. (b) = +10 dB.

To investigate the eect of the two numbers mX and mY on the capacity, we

study three cases: case (i)mY = 1, but mX is a variable larger than one (MISO);
case (ii)mX = 1, but mY is a variable larger than one (SIMO); and case (iii)
both mX and mY are variables larger than one, but we fix mX = mY .
The results for MISO case are depicted in Figure 5.75. From this figure, it can
be seen that for a given , the outage probability decreases with the increase of the
number of transmit antennas when R/WX is lower than some value, but increases
instead when R/WX is higher than this value. Notice that the corresponding outage
probability is so high when R/WX is larger than this value that to transfer information at this rate is of little practical interest.

Thomas Kaiser et al.

395
1

Cuni /Copt

0.9
0.8
0.7
0.6
0.5
0.4
30

20

10

10

20

30

L=2
L=7

L = 15
L = 50

Figure 5.74. Eciency of uniform PSD relative to optimal PSD (mX = mY = 1).

The results for SIMO case are demonstrated in Figure 5.76, which shows that,
for a given outage probability Pout , the transmission rate R increases with log mY
almost linearly.
Figure 5.77 shows the results for MIMO case, from which we can observe that,
for a given outage probability Pout , the transmission rate R increases with mX (or
mY ) almost linearly.
5.8.10. LOS effect on UWB channel capacity
In a recent paper [121], Gesbert investigated the problem that whether or not
multipaths for MIMO wireless communication systems are beneficial as channel
capacity and coverage are concerned. This concern is motivated by the observation: for a MIMO systems, multipaths can increase the diversity but degrade SNR,
while the line-of-sight (LOS) transmission, corresponding to an extreme case for
a lower multipaths, can strengthen SNR but provide no diversity. Therefore, it
seems that a compromising region for the benefit of multipaths should exist. This
is indeed the case for narrowband communications, as shown in [121]. In this
subsection, we aim to investigate if it is also true for UWB communications.
We will consider model (5.292) for two extreme cases: LOS and non-LOS
(NLOS) case. For NLOS case, we will assume that all Al (l = 1, 2, . . . , L) are of
the distribution as specified in Assumption 14. For LOS case, we will assume that
all Al (l = 2, . . . , L) are of the distribution as specified in Assumption 14, but A1
is of the following form A1 = vr vtT , where vr and vt are mX - and mY -dimensional
vectors, respectively, each entry is +1 or 1 with equal probability, and all the elements in vr and vt are independent. The above model for A1 is borrowed from
narrowband LOS channel model, see [121]. Note that the value of each entry in
the above chosen A1 will satisfy E ([A1 ]2i j ) = 1 for all i and j, which makes the
comparison between LOS and NLOS be justfiable.
The numerical results are depicted in Figures 5.78, 5.79, 5.80, 5.81, 5.82, and
5.83. It can be seen from Figures 5.78, 5.79, and 5.80 that for a moderate L (L = 15),

Spatial aspects of UWB

Outage probabillity Pout

396
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

mx = 16

mx = 2
mx = 1

mx = 3, 4,
5, 6, 7, 8

3.5

4.5

4.85

4.9

R/WX

Outage probabillity Pout

(a)
1 mx = 1
0.99
0.98 mx = 2
0.97
0.96 mx = 3
0.95
0.94 mx = 4
0.93
0.92
mx = 5, 6, 7, 8, 16
0.91
0.9
4.7
4.75

4.8
R/WX

(b)
Figure 5.75. Outage probability Pout versus transmission rate R/WX (in nats/s/Hz) for dierent mX .
Eect of the number of transmit antennas on the channel capacity ( = +10 dB, L = 15, mY = 1,
MISO). (a) Panorama. (b) Detail.

the dierence in capacity between LOS and NLOS cases can be perceived only
when both the numbers of the transmit and receive antennas are high. From Figures 5.81, 5.82, and 5.83, it can be observed that when L is small, NLOS scenario
is always favorable for MIMO, but LOS scenario is favorable for SIMO and MISO
when communication rate is low. When the communication rate is high, NLOS
scenario is favorable for SIMO and MISO, but notice that the outage probability
for communication in this region of the rate is so high that communication at this
rate is almost discarded.
5.8.11. Summary
In this section, we have investigated the channel capacity for UWB wireless systems. The results can be roughly summarized for two cases: frequency-flat channels and frequency-selective channels.

Thomas Kaiser et al.

397

R/WX

7.5
7
6.5
6
5.5
5
4.5
4
3.5
0

0.5

1.5

2.5

log(mY )
Pout = 0.05
Pout = 0.1
Pout = 0.2

Pout = 0.3
Pout = 0.4
Pout = 0.5

Figure 5.76. The transmission rate R/WX (in nats/s/Hz) versus mY for dierent outage probability
Pout ( = +10 dB, L = 15, mX = 1, SIMO).

60

R/WX

50
40
30
20
10
2

10

12

14

16

mX ( = m Y )
Pout = 0.05
Pout = 0.2
Pout = 0.5
Figure 5.77. The transmission rate R/WX (in nats/s/Hz) versus mX (or mY ) for dierent outage probability Pout ( = +10 dB, L = 15, mX = mY ).

For frequency-flat UWB channels, the analytic expression for the ergodic
channel capacity of UWB wireless communication systems is presented for four
special cases: SISO, SIMO, MISO, and MIMO (in the situation of infinitely many
transmit and receive antennas). Formulae on the outage probability about the
channel capacity are also presented for cases SISO, SIMO, and MISO. It is shown
that increasing the number of receive antennas can obtain more benefit in channel
capacity than increasing the number of transmit antennas. Principally, the channel capacity could be increased indefinitely by employing a large number of receive
antennas, but it appears to increase only logarithmically in this number for SIMO
case; while employing 3 5 transmit antennas can approach the best advantage

Spatial aspects of UWB

Outage probabillity Pout

398
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

10 12

14 16

18

20

R/WX
LOS, mXY = 1
NLOS, mXY = 1
LOS, mXY = 2
NLOS, mXY = 2
LOS, mXY = 3

NLOS, mXY = 3
LOS, mXY = 4
NLOS, mXY = 4
LOS, mXY = 5
NLOS, mXY = 5

Outage probabillity Pout

(a)

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
25

30

35

40

45

50

55

60

65

R/WX
LOS, mXY = 6
NLOS, mXY = 6
LOS, mXY = 7
NLOS, mXY = 7

LOS, mXY = 8
NLOS, mXY = 8
LOS, mXY = 16
NLOS, mXY = 16

(b)
Figure 5.78. Outage probability Pout versus transmission rate R/WX (in nats/s/Hz) for dierent number of antennas. Comparison between LOS and NLOS (MIMO, = +10 dB, L = 15).

of the multiple transmit antenna systems (irrespective of all other parameters considered herein) as far as channel capacity is concerned for MISO case. We have also
observed that when the signal-to-noise ratio is low, the benefit in average capacity
obtained by distributing the available power to dierent transmit antennas is very
limited.
For frequency-selective UWB channels, an analytic expression for the capacity
of UWB wireless communication systems for SISO case and when STDL consists

Outage probabillity Pout

Thomas Kaiser et al.

399
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

3.5

4.5

5.5

6.5

7.5

R/WX
LOS, mY = 1
NLOS, mY = 1
LOS, mY = 2
NLOS, mY = 2
LOS, mY = 3
NLOS, mY = 3
LOS, mY = 4
NLOS, mY = 4
LOS, mY = 5

NLOS, mY = 5
LOS, mY = 6
NLOS, mY = 6
LOS, mY = 7
NLOS, mY = 7
LOS, mY = 8
NLOS, mY = 8
LOS, mY = 16
NLOS, mY = 16

Outage probabillity Pout

Figure 5.79. Outage probability Pout versus transmission rate R/WX (in nats/s/Hz) for dierent number of antennas. Comparison between LOS and NLOS (SIMO, = +10 dB, L = 15).

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
3.4

mx = 2

mx = 16

mx = 1

mx = 3, 4,
5, 6, 7, 8

3.6

3.8

4.2

4.4 4.6

4.8

R/WX
LOS
NLOS
Figure 5.80. Outage probability Pout versus transmission rate R/WX (in nats/s/Hz) for dierent number of antennas. Comparison between LOS and NLOS (MISO, = +10 dB, L = 15).

of only two taps and the fading power ratio is approximated by one has been obtained, while the closed form for the channel capacity for other cases seems to
be extremely dicult to obtain. Therefore, Monte Carlo simulation method is resorted to obtain the capacity for these cases. The numerical results show that the
dierence between the results obtained by the analytic method and the Monte
Carlo method for the case of SISO and L = 2 is negligible. The following observations can be drawn from our results.

Spatial aspects of UWB

Outage probabillity Pout

400
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

10

15

20

25

R/WX
LOS, L = 1
NLOS, L = 1
LOS, L = 2
NLOS, L = 2
LOS, L = 3
NLOS, L = 3

LOS, L = 4
NLOS, L = 4
LOS, L = 4
NLOS, L = 5
LOS, L = 6
NLOS, L = 6
(a)

Outage probabillity Pout

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

26

28

30

32

34

36

R/WX
LOS, L = 7
NLOS, L = 7
LOS, L = 8
NLOS, L = 8
LOS, L = 9
NLOS, L = 9
LOS, L = 10

NLOS, L = 10
LOS, L = 11
NLOS, L = 11
LOS, L = 12
NLOS, L = 12
LOS, L = 50
NLOS, L = 50
(b)

Figure 5.81. Outage probability Pout versus transmission rate R/WX (in nats/s/Hz) for dierent L.
Comparison between LOS and NLOS (MIMO, = +10 dB, mX = mY = 8).

Outage probabillity Pout

Thomas Kaiser et al.

401

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
3.5

4.5

5.5

6.5

7.5

R/WX
LOS, L = 1
NLOS, L = 1
LOS, L = 2
NLOS, L = 2
LOS, L = 3
NLOS, L = 3
LOS, L = 4
NLOS, L = 4
LOS, L = 5
NLOS, L = 5
LOS, L = 6
NLOS, L = 6
LOS, L = 7

NLOS, L = 7
LOS, L = 8
NLOS, L = 8
LOS, L = 9
NLOS, L = 9
LOS, L = 10
NLOS, L = 10
LOS, L = 11
NLOS, L = 11
LOS, L = 12
NLOS, L = 12
LOS, L = 50
NLOS, L = 50

Figure 5.82. Outage probability Pout versus transmission rate R/WX (in nats/s/Hz) for dierent L.
Comparison between LOS and NLOS (SIMO, = +10 dB, mY = 8).

(i) When the signal-to-noise ratio is low, the outage probability decreases remarkably with L.
(ii) When the signal-to-noise ratio is lower than some value, say 20 dB, using
optimal power spectrum distribution at the transmitter side can increase the reliable transmission rate considerably, compared to the uniform power spectrum
distribution scheme, while when the signal-to-noise ratio is higher than some
value, say 10 dB, the benefit of optimal power spectrum distribution is very limited. Therefore, one should consider measures, such as frequency-domain waterfilling algorithm, to take full advantage of frequency selectivity if the signal-tonoise ratio of the UWB systems is low.
(iii) In MISO case, the outage probability decreases with the increase of the
number of transmit antennas when the communication rate is lower than some
value, but increases when the rate is higher than this value (but communication
at a rate higher than this value is often inadmissible due to too high outage probability). Therefore, if the required transmission rate is indeed too high and the
available power is limited, it is better to concentrate all power in one antenna,
rather than to distribute the power in multiple antennas, to transmit information
signals.

Spatial aspects of UWB

Outage probabillity Pout

402
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1.5

2.5

3.5

4.5

5.5

R/WX
LOS, L = 1
NLOS, L = 1
LOS, L = 2
NLOS, L = 2
LOS, L = 3
NLOS, L = 3
LOS, L = 4
NLOS, L = 4
LOS, L = 5
NLOS, L = 5
LOS, L = 6
NLOS, L = 6
LOS, L = 7

NLOS, L = 7
LOS, L = 8
NLOS, L = 8
LOS, L = 9
NLOS, L = 9
LOS, L = 10
NLOS, L = 10
LOS, L = 11
NLOS, L = 11
LOS, L = 12
NLOS, L = 12
LOS, L = 50
NLOS, L = 50

Figure 5.83. Outage probability Pout versus transmission rate R/WX (in nats/s/Hz) for dierent L.
Comparison between LOS and NLOS (MISO, = +10 dB, mX = 8).

(iv) In SIMO case, the communication rate supportable by the channel with a
given outage probability increases approximately logarithmically with the number
of receive antennas.
(v) In MIMO case with equal number of transmit and receive antennas, the
communication rate supportable by the channel with a given outage probability
increases approximately linearly with the number of transmit or receive antennas.
(vi) The eect of LOS transmission on UWB systems are also investigated. It
is shown that
(1) for a moderate number of received pulses, the dierence in capacity between LOS and NLOS cases can be perceived only when both the numbers of the transmit and receive antennas are high;
(2) when L is small, NLOS scenario is always favorable for MIMO, but LOS
scenario is favorable for SIMO and MISO when the communication rate
is low.
Acknowledgments
The author thanks S. Bories and X. Nie for carrying out the channel measurements at ENSTA, and A. Durantini and D. Cassioli for providing channel measurements carried out by RadioLabs and authorising their publication in this book.

Thomas Kaiser et al.

403

This work has been partly funded by the European Commission within IST project
ULTRAWAVES (FP5).
Bibliography
[1] J. Kunisch and J. Pamp, Measurement results and modeling aspects for the UWB radio channel, in Proceedings of IEEE Conference on Ultra Wideband Systems and Technologies (UWBST
02), pp. 1923, Baltimore, Md, USA, May 2002.
[2] Federal Communications Commission (FCC), First Report and Order 02-48, February 2002.
[3] A. Saleh and R. Valenzuela, A statistical model for indoor multipath propagation, IEEE Journal
on Selected Areas in Communications, vol. 5, no. 2, pp. 128137, 1987.
[4] J. Kunisch and J. Pamp, UWB radio channel modeling considerations, in Proceedings of International Conference on Electromagnetics in Advanced Applications (ICEAA 03), Torino, Italy,
September 2003.
[5] J. Romme and B. Kull, On the relation between bandwidth and robustness of indoor UWB
communication, in Proceedings of IEEE Conference on Ultra Wideband Systems and Technologies
(UWBST 03), pp. 255259, Reston, Va, USA, November 2003.
[6] Federal Communications Commission (FCC), First report and order revision of part 15 of the
commissions rules regarding ultra-wideband transmission systems, Washington, DC, USA, ET
Docket 98-153, 2002.
[7] C. A. Balanis, Antenna Theory: Analysis and Design, John Wiley & Sons, New York, NY, USA,
1997.
[8] C. E. Baum, General Properties of Antennas, Sensor and Simulation Notes, Note 330, Air Force
Research Laboratory, Directed Energy Directorate, Kirtland, New Mexico, 1991.
[9] J. D. Kraus, Antennas, Electrical Engineering Series, McGraw-Hill, New York, NY, USA, 2nd edition, 1988.
[10] W. Wiesbeck and E. Heidrich, Wide-band multiport antenna characterization by polarimetric
RCS measurements, IEEE Transactions on Antennas and Propagation, vol. 46, no. 3, pp. 341350,
1998.
[11] C. Waldschmidt, W. Sorgel, and W. Wiesbeck, Technology: Antenna Design, to appear in JASP
Book on Smart Antennas.
[12] W. Sorgel and W. Wiesbeck, Influence of the antennas on the ultra-wideband transmission,
EURASIP Journal on Applied Signal Processing, vol. 2005, no. 3, pp. 296305, 2005.
[13] W. Sorgel, F. Pivit, and W. Wiesbeck, Comparison of time domain and frequency domain measurement procedures for ultra wideband antennas domain measurement procedures for ultra
wideband antennas, in Proceedings of 25th Annual Meeting & Symposium on Antenna Measurement Techniques Association (AMTA 03), Irvine, Calif, USA, October 2003.
[14] L. Bowen, E. Farr, and W. Prather, A collapsible impulse radiating antenna, in Proceedings of
Ultra-Wideband, Short-Pulse Electromagnetics Conference, Edinburgh, Scotland, UK, June 2000.
[15] W. Sorgel, C. Waldschmidt, and W. Wiesbeck, Transient responses of a Vivaldi antenna and a
logarithmic periodic dipole array for ultra wideband communication, in Proceedings of IEEE
Antennas and Propagation Society International Symposium, vol. 3, pp. 592595, Columbus,
Ohio, USA, June 2003.
[16] R. S. Carson, Radio Communications Concepts: Analog, John Wiley & Sons, New York, NY, USA,
1990.
[17] S. G. Glisic, Advanced Wireless Communications: 4G Technologies, John Wiley & Sons, New York,
NY, USA, 2004.
[18] M. Z. Win and R. A. Scholtz, Impulse radio: how it works, IEEE Communications Letters, vol. 2,
no. 2, pp. 3638, 1998.
[19] The UWB Multiband Coalition, available at: http://www.multibandofdm.org/.
[20] G. R. Aiello and G. D. Rogerson, Ultra-wideband wireless systems, IEEE Microwave, vol. 4,
no. 2, pp. 3647, 2003.

404

Spatial aspects of UWB

[21] S. Roy, J. R. Foerster, V. S. Somayazulu, and D. G. Leeper, Ultrawideband radio design: the
promise of high-speed, short-range wireless connectivity, Proceedings of IEEE, vol. 92, no. 2, pp.
295311, 2004.
[22] A. Sibille and S. Bories, Spatial diversity for UWB communications, in Proceedings of 5th European Personal Mobile Communications Conference (EPMCC 03), pp. 367370, Glasgow, Scotland, UK, April 2003.
[23] IEEE, IEEE standard for information technologytelecommunications and information exchange between systemslocal and metropolitan area networksspecific requirements part
15.3: wireless Medium Access Control (MAC) and Physical Layer (PHY) specifications for high
rate Wireless Personal Area Networks (WPANs), IEEE 802.15.3-2003, 2003, pp. 1315.
[24] S. N. Makarov, Antenna and EM Modeling with MATLAB, John Wiley & Sons, New York, NY,
USA, 2002.
[25] D. Cassioli, M. Z. Win, and A. F. Molisch, The ultra-wide bandwidth indoor channel: from
statistical model to simulations, IEEE Journal on Selected Areas in Communications, vol. 20,
no. 6, pp. 12471257, 2002.
[26] A. F. Molisch, J. R. Foerster, and M. Pendergrass, Channel models for ultrawideband personal
area networks, IEEE Wireless Communications, vol. 10, no. 6, pp. 1421, 2003.
[27] V. Hovinen, M. Hamalainen, and T. Patsi, Ultra wideband indoor radio channel models: preliminary results, in Proceedings of IEEE Conference on Ultra-Wideband Systems and Technologies
(UWBST 02), pp. 7579, Baltimore, Md, USA, May 2002.
[28] B. Vucetic and J. Yuan, Space-Time Coding, John Wiley & Sons, New York, NY, USA, 2003.
[29] A. Hottinen, O. Tirkkonen, and R. Wichman, Multi-antenna Transceiver Techniques for 3G and
Beyond, John Wiley & Sons, New York, NY, USA, 2003.
[30] W. C.-Y. Lee and Y. S. Yeh, Polarization diversity system for mobile radio, IEEE Transactions
on Communications, vol. 20, no. 5, pp. 912923, 1972.
[31] R. G. Vaughan, Polarization diversity in mobile communications, IEEE Transransactions on
Vehicular Technology, vol. 39, no. 3, pp. 177186, 1990.
[32] B. Lindmark and M. Nilsson, On the available diversity gain from dierent dual-polarized antennas, IEEE Journal on Selected Areas in Communications, vol. 19, no. 2, pp. 287294, 2001.
[33] C. Degen and W. Keusgen, Performance evaluation of MIMO systems using dual-polarized antennas, in Proceedings of 10th International Conference on Telecommunications (ICT 03), vol. 2,
pp. 15201525, Papeete, Tahiti, French Polynesia, FebruaryMarch 2003.
[34] S. M. Alamouti, A simple transmit diversity technique for wireless communications, IEEE Journal on Selected Areas in Communications, vol. 16, no. 8, pp. 14511458, 1998.
[35] R. van Nee and R. Prasad, OFDM for Wireless Multimedia Communications, Artech House, Norwood, Mass, USA, 2000.
[36] S. Hara and R. Prasad, Multicarrier Techniques for 4G Mobile Communications, Artech House,
Norwood, Mass, USA, 2003.
[37] A. Saleh and R. Valenzuela, A statistical model for indoor multipath propagation, IEEE Journal
on Selected Areas in Communications, vol. 5, no. 2, pp. 128137, 1987.
[38] IEEE 802.15 WPAN High Rate Alternative PHY Task Group 3a (TG3a), August 2005, available at:
http://www.ieee802.org/15/pub/TG3a.html.
[39] Automatic selection of receiving channels, US patent no. 1747218, February 1930.
[40] W. C. Jakes, Microwave Mobile Communications, John Wiley & Sons, New York, NY, USA, 1974.
[41] M. Weisenhorn and W. Hirt, Performance of binary antipodal signaling over indoor UWB
MIMO channel, in Proceedings of IEEE International Conference on Communications (ICC 03),
vol. 4, pp. 28722878, Anchorage, Alaska, USA, May 2003.
[42] C. Roblin, S. Bories, and A. Sibille, Characterization tools of antennas in the time domain, in
Proceedings of International Workshop on Ultra Wideband Systems (IWUWBS 03), Oulu, Finland,
June 2003.
[43] D. Cassioli, M. Z. Win, and A. F. Molisch, The ultra-wide bandwidth indoor channel: from
statistical model to simulations, IEEE Journal on Selected Areas in Communications, vol. 20,
no. 6, pp. 12471257, 2002.

Thomas Kaiser et al.

405

[44] K. Kunisch and J. Pamp, Measurement results and modeling aspects for the UWB radio channel, in Proceedings of COST 273 5th Management Committee Meeting (MCM 02), Lisbon, Portugal, September 2002, Document TD(02)105.
[45] V. Hovinen, M. Hamalainen, and T. Patsi, Ultra wideband indoor radio channel models: preliminary results, in Proceedings of IEEE Conference on Ultra Wideband Systems and Technologies
(UWBST 02), pp. 7579, Baltimore, Md, USA, May 2002.
[46] D. Cassioli and A. Durantini, A time-domain propagation model of the UWB indoor channel in the FCC-compliant band 3.6-6 GHz based on PN-sequence channel measurements, in
Proceedings of IEEE 59th Vehicular Technology Conference (VTC 04), vol. 1, pp. 213217, Milan,
Italy, May 2004.
[47] R. J.-M. Cramer, R. A. Scholtz, and M. Z. Win, Evaluation of an ultra-wide-band propagation
channel, IEEE Transactions on Antennas and Propagation, vol. 50, no. 5, pp. 561570, 2002.
[48] K. Haneda and J.-I. Takada, An application of SAGE algorithm for UWB propagation channel estimation, in Proceedings of COST 273 8th Management Committee Meeting (MCM 03),
Prague, Czech Republic, September 2003, Document TD(03)182.
[49] C. Prettie, D. Cheung, L. Rusch, and M. Ho, Spatial correlation of UWB signals in a home
environment, in Proceedings of IEEE Conference on Ultra Wideband Systems and Technologies
(UWBST 02), pp. 6569, Baltimore, Md, USA, May 2002.
[50] A. Durantini, W. Ciccognani, and D. Cassioli, UWB propagation measurements by PNsequence channel sounding, in Proceedings of IEEE International Conference on Communications
(ICC 04), vol. 6, pp. 34143418, Paris, France, June 2004.
[51] S. Bories, C. Roblin, and A. Sibille, Ultra-wideband monocone antenna for UWB channel measurements, in Proceedings of International Workshop on Ultra Wideband Systems (IWUWBS 03),
Oulu, Finland, June 2003.
[52] L. Yang and G. B. Giannakis, Space-time coding for impulse radio, in Proceedings of IEEE
Conference on Ultra Wideband Systems and Technologies (UWBST 02), pp. 235239, Baltimore,
Md, USA, May 2002.
[53] A. Sibille, Time-domain diversity in ultra-wideband MIMO communications, in Proceedings
of COST 273 5th Management Committee Meeting (MCM 02), Lisbon, Portugal, September
2002, Document TD(02)105.
[54] E. Zollinger, Extremely wideband antennas, in Digital Mobile Radio Towards Future Generation Systems, DG XIII, Ed., chapter 3, pp. 8082, European Commission EUR 18957, Brussels,
Belgium, 1999, COST 231 final report.
[55] V. Murino, A. Trucco, and A. Tesei, Beam pattern formulation and analysis for wide-band
beamforming systems using sparse arrays, Signal Processing, vol. 56, no. 2, pp. 177183, 1997.
[56] M. G. M. Hussain, Principles of space-time array processing for ultrawide-band impulse radar
and radio communications, IEEE Transransactions on Vehicular Technology, vol. 51, no. 3, pp.
393403, 2002.
[57] F. J. Harris, On the use of windows for harmonic analysis with the discrete Fourier transform,
Proceedings of IEEE, vol. 66, no. 1, pp. 5183, 1978.
[58] B. Steinberg, Principles of Aperture and Array System Design, John Wiley & Sons, New York, NY,
USA, 1976.
[59] A. Papoulis, Signal Analysis, McGraw-Hill, New York, NY, USA, 1985.
[60] S. Ries and T. Kaiser, On the Influence of Prefiltering on the Sidelobe Level for Broadband
Impulse Beamforming, 2004, submitted for publication.
[61] E. J. Bond, X. Li, S. C. Hagness, and B. D. Van Veen, Microwave imaging via space-time beamforming for early detection of breast cancer, IEEE Antennas and Propagation Magazine, vol. 51,
no. 8, pp. 16901705, 2003.
[62] M. Ghavami and R. Kohno, Recursive fan filters for a broad-band partially adaptive antenna,
IEEE Transactions on Communications, vol. 48, no. 2, pp. 185188, 2000.
[63] M. Ghavami and R. Kohno, Improvement in the DOA estimation of broadband signals using
partially IIR beamformers, IEEE Transactions on Communications, vol. 50, no. 6, pp. 897901,
2002.

406

Spatial aspects of UWB

[64] R. G. Pridham and R. A. Mucci, Digital interpolation beamforming for low-pass and bandpass
signals, Proceedings of IEEE, vol. 67, no. 6, pp. 904919, 1979.
[65] D. Rathjen, G. Bodecker, and M. Siegel, Omnidirectional beamforming for linear antennas by
means of interpolated signals, IEEE Journal of Oceanic Engineering, vol. 10, no. 3, pp. 260268,
1985.
[66] T. Blu, P. Thevenaz, and M. Unser, MOMS: maximal-order interpolation of minimal support,
IEEE Transactions on Image Processing, vol. 10, no. 7, pp. 10691080, 2001.
[67] S. Ries, On the reconstruction of signals by a finite number of samples, Signal Processing,
vol. 23, no. 1, pp. 4568, 1991.
[68] J. R. Higgins, Sampling Theory in Fourier Analysis and Signal Analysis: Foundations Vol. 1, Clarendon Press, Oxford, UK, 1996.
[69] T. Blu, P. Thevenaz, and M. Unser, Linear interpolation revitalized, IEEE Transactions on Image
Processing, vol. 13, no. 5, pp. 710719, 2004.
[70] C. Buchholz and T. Kaiser, Is DoA estimation feasible for dense multipath UWB indoor channels? in Proceedings of 16th International Conference on Wireless Communications, pp. 420425,
Calgary, Alberta, Canada, July 2004.
[71] A. F. Molisch, J. R. Foerster, and M. Pendergrass, Channel models for ultrawideband personal
area networks, IEEE Wireless Communications, vol. 10, no. 6, pp. 1421, 2003.
[72] A. I. Zayed, Advances in Shannons Sampling Theory, CRC Press, Boca Raton, Fla, USA, 1993.
[73] P. L. Butzer and R. J. Nessel, Fourier Analysis and Approximation Vol. 1, Birkhauser, Basel,
Switzerland, 1971.
[74] J.-C. Guey, M. P. Fitz, M. R. Bell, and W.-Y. Kuo, Signal design for transmitter diversity wireless
communication systems over Rayleigh fading channels, IEEE Transactions on Communications,
vol. 47, no. 4, pp. 527537, 1999.
[75] V. Tarokh, N. Seshadri, and A. R. Calderbank, Space-time codes for high data rate wireless
communication: performance criterion and code construction, IEEE Transactions on Information Theory, vol. 44, no. 2, pp. 744765, 1998.
[76] S. M. Alamouti, A simple transmit diversity technique for wireless communications, IEEE Journal on Selected Areas in Communications, vol. 16, no. 8, pp. 14511458, 1998.
[77] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, Space-time block codes from orthogonal designs, IEEE Transactions on Information Theory, vol. 45, no. 5, pp. 14561467, 1999.
[78] B. M. Hochwald and T. L. Marzetta, Unitary space-time modulation for multiple-antenna communications in Rayleigh flat fading, IEEE Transactions on Information Theory, vol. 46, no. 2, pp.
543564, 2000.
[79] D. Agrawal, V. Tarokh, A. Naguib, and N. Seshadri, Space-time coded OFDM for high datarate wireless communication over wideband channels, in Proceedings of 48th IEEE Vehicular
Technology Conference (VTC 98), vol. 3, pp. 22322236, Ottawa, Ontario, Canada, May 1998.
[80] R. S. Blum, Y. G. Li, J. H. Winters, and Q. Yan, Improved space-time coding for MIMO-OFDM
wireless communications, IEEE Transactions on Communications, vol. 49, no. 11, pp. 1873
1878, 2001.
[81] H. Bolcskei and A. J. Paulraj, Space-frequency coded broadband OFDM systems, in Proceedings of IEEE Wireless Communications and Networking Conference (WCNC 00), vol. 1, pp. 16,
Chicago, Ill, USA, September 2000.
[82] W. Su, Z. Safar, M. Olfat, and K. J. Ray Liu, Obtaining full-diversity space-frequency codes
from space-time codes via mapping, IEEE Transactions on Signal Processing, vol. 51, no. 11, pp.
29052916, 2003, Special issue on MIMO systems.
[83] W. Su, Z. Safar, and K. J. Ray Liu, Systematic design of space-frequency codes with full rate
and full diversity, in Proceedings of IEEE Wireless Communications and Networking Conference
(WCNC 04), vol. 3, pp. 14361441, Atlanta, Ga, USA, March 2004.
[84] A. H. Tewfik and E. Saberinia, High bit rate ultra-wideband OFDM, in Proceedings of IEEE
Global Telecommunications Conference (GLOBECOM 02), vol. 3, pp. 22602264, Taipei, Taiwan,
November 2002.

Thomas Kaiser et al.

407

[85] J. Balakrishnan, A. Batra, and A. Dabak, A multi-band OFDM system for UWB communication, in Proceedings of IEEE Conference on Ultra Wideband Systems and Technologies, pp. 354
358, Reston, Va, USA, November 2003.
[86] The UWB Multi-Band Coalition, Internet: www.uwbmultiband.org.
[87] IEEE 802.15WPAN High Rate Alternative PHY Task Group 3a (TG3a), Internet: http://www.
ieee802.org/15/pub/TG3a.html.
[88] R. A. Scholtz, Multiple access with time-hopping impulse modulation, in Proceedings of IEEE
Military Communications Conference (MILCOM 93), vol. 2, pp. 447450, Boston, Mass, USA,
October 1993.
[89] M. Z. Win and R. A. Scholtz, Impulse radio: how it works, IEEE Communications Letters, vol. 2,
no. 2, pp. 3638, 1998.
[90] M. L. Welborn, System considerations for ultra-wideband wireless networks, in Proceedings
of IEEE Radio and Wireless Conference (RAWCON 01), pp. 58, Waltham, Mass, USA, August
2001.
[91] J. R. Foerster, The performance of a direct-sequence spread ultrawideband system in the presence of multipath, narrowband interference, and multiuser interference, in Proceedings of IEEE
Conference on Ultra Wideband Systems and Technologies (UWBST 02), pp. 8791, Baltimore,
Md, USA, May 2002.
[92] N. Boubaker and K. B. Letaief, Ultra wideband DSSS for multiple access communications using
antipodal signaling, in Proceedings of IEEE International Conference on Communications (ICC
03), vol. 3, pp. 21972201, Anchorage, Alaska, USA, May 2003.
[93] R. A. Horn and C. R. Johnson, Matrix Analysis, Cambridge University Press, New York, NY,
USA, 1985.
[94] L. Yang and G. B. Giannakis, Space-time coding for impulse radio, in Proceedings of IEEE
Conference on Ultra Wideband Systems and Technologies (UWBST 02), pp. 235239, Baltimore,
Md, USA, May 2002.
[95] M. Nakagami, The m-distributionA general formula of intensity distribution of rapid fading, in Statistical Methods in Radio Wave Propagation, W. G. Homan, Ed., pp. 336, Pergamon
Press, Oxford, UK, 1960.
[96] M. Z. Win and R. A. Scholtz, Ultra-wide bandwidth time-hopping spread-spectrum impulse
radio for wireless multiple-access communications, IEEE Transactions on Communications,
vol. 48, no. 4, pp. 679689, 2000.
[97] J. Proakis, Digital Communications, McGraw-Hill, New York, NY, USA, 4th edition, 2001.
[98] M. K. Simon and M.-S. Alouini, Digital Communication over Fading Channels: A Unified Approach to Performance Analysis , John Wiley & Sons, New York, NY, USA, 2000.
[99] W. P. Siriwongpairat, M. Olfat, and K. J. Ray Liu, Performance analysis and comparison of
time-hopping and direct-sequence UWB-MIMO systems, EURASIP Journal on Applied Signal
Processing, vol. 2005, no. 3, pp. 328345, 2005, Special issue on UWB-State of the art.
[100] S. S. Ghassemzadeh, L. J. Greenstein, T. Sveinsson, and V. Tarokh, A multipath intensity profile model for residential environments, in Proceedings of IEEE Wireless Communications and
Networking Conference (WCNC 03), vol. 1, pp. 150155, New Orleans, La, USA, March 2003.
[101] J. Zhang, R. A. Kennedy, and T. D. Abhayapala, New results on the capacity of M-ary PPM
ultra-wideband systems, in Proceedings of International Workshop on Ultra WideBand Systems
(IWUWBS 03), Oulu, Finland, June 2003.
[102] L. Zhao and A. M. Haimovich, The capacity of an UWB multiple-access communications system, in Proceedings of IEEE International Conference on Communications (ICC 02), vol. 3, pp.
19641968, New York, NY, USA, AprilMay 2002.
[103] L. Zhao and A. M. Haimovich, Capacity of M-ary PPM ultra-wideband communications over
AWGN channels, in Proceedings of IEEE VTS 54th Vehicular Technology Conference (VTC 01),
vol. 2, pp. 11911195, Atlantic City, NJ, USA, October 2001.
[104] G. J. Foschini and M. J. Gans, On limits of wireless communications in a fading environment
when using multiple antennas, Wireless Personal Communications, vol. 6, no. 3, pp. 311335,
1998.

408

Spatial aspects of UWB

[105] I. E. Telatar, Capacity of multi-antenna Gaussian channels, European Transactions on Telecommunications, vol. 10, no. 6, pp. 585595, 1999.
[106] I. E. Telatar, Capacity of multi-antenna Gaussian channels, Tech. Rep., AT&T Bell Laboratories,
Murray Hill, NJ, USA, 1995.
[107] M. J. Gans, N. Amitay, Y. S. Yeh, et al., Outdoor BLAST measurement system at 2.44 GHz:
calibration and initial results, IEEE Journal on Selected Areas in Communications, vol. 20, no. 3,
pp. 570583, 2002.
[108] A. F. Molisch, M. Steinbauer, M. Toeltsch, E. Bonek, and R. S. Thoma, Capacity of MIMO
systems based on measured wireless channels, IEEE Journal Selected Areas in Communications,
vol. 20, no. 3, pp. 561569, 2002.
[109] A. van Zelst and T. C. W. Schenk, Implementation of a MIMO OFDM-based wireless LAN
system, IEEE Transactions on Signal Processing, vol. 52, no. 2, pp. 483494, 2004.
[110] D. Cassioli, M. Z. Win, and A. F. Molisch, The ultra-wide bandwidth indoor channel: from
statistical model to simulations, IEEE Journal on Selected Areas in Communications, vol. 20,
no. 6, pp. 12471257, 2002.
[111] H. Hashemi, The indoor radio propagation channel, Proceedings of IEEE, vol. 81, no. 7, pp.
943968, 1993.
[112] M. Z. Win, R. A. Scholtz, and M. A. Barnes, Ultra-wide bandwidth signal propagation for
indoor wireless communications, in Proceedings of IEEE International Conference on Communications (ICC 97), vol. 1, pp. 5660, Montreal, Quebec, Canada, June 1997.
[113] H. Zhang, T. Udagawa, T. Arita, and M. Nakagawa, A statistical model for the small-scale multipath fading characteristics of ultra wideband indoor channel, in Proceedings of IEEE Conference
on Ultra Wideband Systems and Technologies (UWBST 02), pp. 8185, Baltimore, Md, USA, May
2002.
[114] J. R. Foerster, M. Pendergrass, and A. F. Molisch, A channel model for ultrawideband indoor
communication, Tech. Rep. TR-2003-73, Mitsubishi Electric Research Laboratories (MERL),
Cambridge, Mass, USA, 2003, available at: http://www.merl.com/reports/docs/TR2003-73.pdf.
[115] J. G. Proakis, Digital Communications, McGraw-Hill, Singapore, 4th edition, 2001.
[116] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Products, 5th Corrected and
Enlarged Edition, Prepared by A. Jerey, Academic Press, New York, NY, USA, 1980.
[117] R. G. Gallager, Information Theory and Reliable Communication, John Wiley & Sons, New York,
NY, USA, 1968.
[118] F. Zheng and T. Kaiser, On the evaluation of channel capacity of multi-antenna UWB indoor
wireless systems, in Proceedings of IEEE 8th International Symposium on Spread Spectrum Techniques and Applications (ISSSTA 04), pp. 525529, Sydney, Australia, AugustSeptember 2004.
[119] R. J. Muirhead, Aspects of Multivariate Statistical Theory, John Wiley & Sons, New York, NY,
USA, 1982.
[120] E. Biglieri, J. Proakis, and S. Shamai, Fading channels: information-theoretic and communications aspects, IEEE Transactions on Information Theory, vol. 44, no. 6, pp. 26192692, 1998.
[121] D. Gesbert, Multipath: curse or blessing? A system performance analysis of MIMO wireless
systems, in Proceedings of IEEE International Zurich Seminar on Communications (IZS 04), pp.
1417, Zurich, Switzerland, February 2004.
[122] A. V. Oppenheim and R. W. Schafer, Discrete-Time Signal Processing, Prentice-Hall, Englewood
Clis, NJ, USA, 1989.
Thomas Kaiser: Department of Communication Systems, Faculty of Engineering,
University of Duisburg-Essen, 47048 Duisburg, Germany
Email: [email protected]
Jurgen Kunisch: IMST GmbH, 47475 Kamp-Lintfort, Germany
Email: [email protected]
Werner Sorgel: Institut fur Hochstfrequenztechnik und Elektronik (IHE), Universitat Karlsruhe (TH),
76131 Karlsruhe, Germany
Email: [email protected]

Thomas Kaiser et al.

409

Christian Waldschmidt: Institut fur Hochstfrequenztechnik und Elektronik (IHE),

Universitat Karlsruhe (TH), 76131 Karlsruhe, Germany
Email: [email protected]
Werner Wiesbeck: Institut fur Hochstfrequenztechnik und Elektronik (IHE),
Universitat Karlsruhe (TH), 76131 Karlsruhe, Germany
Email: [email protected]
F. Argenti: Department of Electronics and Telecommunications, University of Florence,
50139 Florence, Italy
Email: [email protected]
T. Bianchi: Department of Electronics and Telecommunications, University of Florence,
50139 Florence, Italy
Email: [email protected]
L. Mucchi: Department of Electronics and Telecommunications, University of Florence,
50139 Florence, Italy
Email: [email protected]
L. S. Ronga: Consorzio Nazionale Interuniversitario per le Telecomunicazioni (CNIT),
Universit`a degli Studi di Firenze, 50139 Firenze, Italy
Email: [email protected]

Alain Sibille: Ecole

Nationale Superieure de Techniques Avancees (ENSTA), 32 Boulevard Victor,
75739 Paris Cedex 15, France
Email: [email protected]
Sigmar Ries: Fachhochschule Sudwestfalen, Hochschule fur Technik und Wirtschaft,
Frauenstuhlweg 31, 58644 Iserlohn, Germany
Email: [email protected]
Christiane Senger: Department of Communication Systems, Faculty of Engineering,
University of Duisburg-Essen, 47048 Duisburg, Germany
Email: [email protected]
W. Pam Siriwongpairat: Department of Electrical and Computer Engineering and Institute for
Systems Research, University of Maryland, College Park, MD 20742, USA
Email: [email protected]
Masoud Olfat: Department of Electrical and Computer Engineering and Institute for
Systems Research, University of Maryland, College Park, MD 20742, USA
Email: [email protected]
Weifeng Su: Department of Electrical and Computer Engineering and Institute for Systems Research,
University of Maryland, College Park, MD 20742, USA
Email: [email protected]
K. J. Ray Liu: Department of Electrical and Computer Engineering and Institute for Systems Research,
University of Maryland, College Park, MD 20742, USA
Email: [email protected]
Feng Zheng: Department of Electronic and Computer Engineering, University of Limerick, Limerick,
Ireland
Email: [email protected]

UWB ranging
Domenico Porcino, Jurgen Sachs,
Rudolf Zetik, and Andy Ward

6.1. Introduction
Ultra wideband technology has been identified as one of the most promising techniques to enhance a mobile terminal or a sensor with accurate ranging and tracking capabilities.
In simple words, the basic idea behind most accurate positioning systems under development today is to determine the time that a radio wave takes to propagate from the transmitter to the receiver and then convert that measurement into
a distance to enable the estimation of the range between the two devices. The initial estimate of this range distance is often called the pseudorange. By calculating
the pseudorange from the querying device to multiple devices at known locations,
(technique often called trilateration) it is possible to identify with high accuracy
the position of a device itself (positioning). Finally, it is also possible to keep calculating these range estimates over time and follow the device over a given time
window while it is moving inside the area covered by the ranging system (tracking).
The system just described only gives a simplistic view of what needs to be
done to tackle a substantial number of technical challenges. Complex mathematical algorithms will be used in support of the ranging and tracking calculations. In
fact, in the first place, it is necessary to be able to measure with a good accuracy
a very low-power signal in any potential condition of operation (severe multipath
included). This signal might additionally be composed of the intentional combination of multiple signals transmitted by a number of parallel antennas or nearby
transmitters. Then it is necessary to distinguish inside the received signal a sufficient number of characteristics to allow an informed estimate of the distance
between the receiver and the transmitter(s). Finally it might be necessary to combine the information from multiple (and possible redundant) sources to be able to
provide the application layer with usable information on the position of the node
and an estimate of the reliability of such information.

412

UWB ranging

The obstacles in the detection and processing of the radio-wave signals are
many, including significant constraints on the maximum power allowed over air
(regulatory constraints), constraints on the radio architecture (standards and cost
constraints), constraints on the maximum processing power (technology constraints), and time needed to derive the information (quality-of-service constraints).
Ranging systems based on UWB technology face all these challenges that need to
be appropriately addressed. But the step forward oered with respect to narrowband technologies is significant and comes from the same definition of ultra-wideband signals. In fact, the transmission of a substantial bandwidth (a minimum of
500 MHz to comply with FCC definitions and regulations) gives UWB receivers
the possibility to achieve an extremely narrow (i.e., precise) temporal resolution
with no precedence in the history of radio communications in any band below
10 GHz.
Making a correct use of such property has allowed the development of practical systems which are oering today ranging resolutions in the order of few tens
of centimeters and coverage of areas as large as 60 meters with a single set of UWB
beacons/base stations.
This chapter oers an introduction to this important and unique feature of
UWB systems, guiding the reader to an understanding of some of the abovementioned challenges and potential solutions that will drive future UWB positioning systems.
6.2. UWB location system techniques, architectures, and analysis
In general, location systems operate by sensing some physical relationship (e.g.,
range or bearing) between a set of reference points and the object to be located, or
some marker attached to that object.
Some location sensors give position location information, that is to say, they
report the location of an object as a point in some coordinate space. Other location
sensors give proximity information, that is, some indication of how near an object
is to another object or some point in space.
6.2.1. Positioning methods in theory
Triangulation systems use measurements of the bearings of objects from known
points to find their locations. Similarly, trilateration (or multilateration) schemes
find object positions by determining their ranges from known points. Hybrid triangulation/trilateration schemes using both bearing and range information are
also possible.
The number of independent bearings or ranges required to fix the position
of an object depends on the number of degrees of freedom of the solution. For
example, in a 2D triangulation system, at least two independent bearings must
be found between an object and known points in the environment to calculate
the objects position, because the tags position is represented by coordinates in

Domenico Porcino et al.

413

Triangulation

Trilateration
Shadow
position
solution

1
Unique position
solution

r1

r2

2
True position
solution
Figure 6.1. Triangulation and trilateration in 2D using two measurements.

two orthogonal dimensions. Similarly, in a 3D triangulation system, at least three

independent bearings must be found. In general, multilateration systems require
an additional measurement to fully resolve ambiguities in the position calculation,
as shown in Figure 6.1.
6.2.2. Practical positioning systems
It is possible to implement practical multilateration systems using UWB technology. In these systems, ranges are determined between the object (or a tag marker
placed on it) and the known points by measuring the time of flight of an UWB signal between a transmitter and a receiver (with either the transmitter being placed
at the object to be tracked, and receivers at known points, or vice versa). The measured signal times-of-flight can be converted into ranges by multiplication by the
(known) speed of light.
Time-of-arrival (TOA) systems measure the signal time of flight between
transmitter and receiver directly. A practical problem arises in these systems: how
should the transmitter and receiver be synchronized in order for the receiver to
measure this time of flight? Note that because the speed of light is very high, this
synchronization must be very tight if accurate ranges are to be measured, and often it is required that the tracked objects have unimpeded mobility (precluding the
use of timing cables of known length for synchronization).1
One solution to this problem involves bidirectional ranging . A signal is sent
from a transceiver at one point to a transceiver at another point, which then sends a
signal in response back to the first transceiver. By measuring the round-trip time of
the signal, the first transceiver can determine the distance to the second transceiver
and back, that is, twice the range between the transceivers. In this arrangement, the
relative timing of signal emission and detection occurs only at the first transceiver,
and only local synchronization between transmitter and receiver is requiredthis
is, of course, a relatively simple proposition.
Once a number of ranges from an object to known points have been determined, a nonlinear model can be fitted to the data to estimate the objects location.
1

TOA systems that use lower-velocity signals can be very simple. See, for example, ultrasonic
systems [1].

414

UWB ranging

The nonlinear model for a 2D TOA system would be



li =

2

xi u

+ yi v

2

+ i

(i = 1, . . . , n),

(6.1)

where (xi , yi ) is the position of the ith known point, li is the measured range of the
object to the ith known point, (u, v) is the estimate of the objects location, and i
is a term which can account for the errors (residuals) in the fitted model.
In some circumstances, it might not be desirable to equip the object, or its
tag marker, with both an UWB transmitter and receiver, and bidirectional ranging
cannot be used. Time-dierence-of-arrival (TDOA) systems are suitable in these
situations. A TDOA system uses a mobile transmitter and a set of synchronized
receivers placed at known points, or a mobile receiver and set of synchronized
transmitters placed at known points. In either case, the mobile and infrastructure
components of the system are unsynchronized, but the infrastructure components
are synchronizedbecause these components tend to be static a variety of wired
or wireless techniques can be used to achieve synchronization.
TDOA systems are unable to directly measure the time of flight of a signal
from a transmitter to a receiver, because these two system components are unsynchronized, and there exists an unknown oset between the clocks of the mobile
unit and the surrounding infrastructure. Rather than convert the time-of-arrival
of a signal at a receiver directly into a transmitter-receiver range, the receiver in a
TDOA system converts the time-of-arrival into a pseudorange (by multiplication
by the speed of light). The pseudorange encapsulates both the range and clock
oset information.
To compute the unknown position of an object in the TDOA system, a nonlinear model is fitted to multiple pseudoranges and the positions of the known
points. The nonlinear model for a 2D TDOA system would be
pi =



2

xi u

+ yi v

2

+ c t + i

(i = 1, . . . , n),

(6.2)

where (xi , yi ) is the position of the ith known point, pi is the pseudorange between
the object and the ith known point, (u, v) is the estimate of the objects location, c
is the speed-of-light, t is the unknown transmitter-receiver clock oset, and i is a
term which accounts for the errors in the fitted model. Note that the TDOA system
must estimate an additional unknown parameter (the clock oset) as compared
to a TOA system, and therefore a TDOA system must measure at least one more
pseudorange than the equivalent TOA system measures ranges.
Practical triangulation sensors, or angle-of-arrival (AOA) systems, can also be
implemented using UWB technology. Bearings can be determined by employing
arrays of receivers and measuring dierences in the signals received by each antenna (such as signal phase or time-of-signal-arrival). By using 2D antenna arrays,
a single receiver unit can determine bearings of the signal transmitter in both azimuth and elevation.

Domenico Porcino et al.

415

It is of course possible to develop hybrid TOA/AOA and TDOA/AOA systems,

which capture and make use of both range (or pseudorange) and angular information to determine the position of an UWB transmitter or receiver.
6.2.3. Proximity versus positioning
Proximity location sensors are naturally simpler than positioning sensors, because
they essentially need to determine only a single quantity (the range between two
objects), whereas a 2D position location sensor must determine at least two quantities (because the solution has two degrees of freedom), and a 3D position location sensor must determine at least three. The methods used in UWB proximity
sensing systems are often similar to those employed in TOA position sensorsthat
is, two-way ranging between cooperating devices.
However, the information reported by a proximity sensor is often less useful
than that reported by a positioning system. Consider, for example, a service deployed in an oce which used location information to send a document from a
users laptop to a nearby printer. It might be the case that the closest printer to the
user was in the room immediately above them on an adjacent floor of the building.
A proximity-based solution would suggest this printer as being the most suitable,
but this choice might be very inconvenient to the user. In contrast, a positionbased solution might determine the coordinates of the user and candidate printers, and software could relate this information to the known architecture of the
building to select the most appropriate printer.
6.2.4. Symmetric versus asymmetric architectures
Symmetric location systems architectures are those in which the entities involved
in the location-finding process have identical (or very similar) capabilities. Asymmetric architectures typically involve a mobile component and an infrastructure
component with dierent characteristics.
The bidirectional-ranging TOA system described above has a symmetric architecture, because both the mobile and known system components are capable
of transmission and reception of UWB signals. An architecture of this type might
be most suitable when UWB-enabled communications devices (such as those in
PAN and LAN environments) are augmented with location-finding capabilityin
this situation, all system components will already be capable of bidirectional UWB
communications.
In TDOA and AOA systems, the mobile and infrastructure components of the
system have dierent capabilities and levels of functionality, and system designers are free to exploit this asymmetry to improve the overall performance of the
system. For example, UWB receivers are often more complex and power-hungry
than their transmitter counterparts (as can be seen also from commercial solutions
discussed within the IEEE 802.15 working groups). In an UWB TDOA system,
it might therefore be advantageous to make the mobile component (which must

UWB ranging
Fraction of readings with error
less than abscissa

416
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Error (m)

Figure 6.2. A cumulative probability graph describing 3D location system accuracy.

meet size and power constraints) the transmitter, and the infrastructure component (where form factor is less important, and mains power may be available) the
receiver.
6.2.5. Analysis location system accuracy
Developers of location sensor systems often describe their systems as being accurate to x cm, but statements like this provide only a rough indication of how well
a system will work in practice. For example, it is very unlikely that the system will
always be accurate to within x centimeters, because most sensor systems generate
a small proportion of very large outlying readings.
Typically, quoted figures refer to either the root-mean-square (RMS) error of
the system , or the accuracy level that 50% of the readings will meet. However,
neither of these measures give the system designer a good idea of what the outlying error distribution looks like, and it is frequently these errors that determine
how eective a sensor system is in practice. Consider, for example, a system that
requires the location of an object to be found within one meter of its true position for successful operation. A location sensor with an RMS accuracy of 75 cm
might seem quite appropriate for this task. However, if only 80% of all readings
lie within one meter of the true position (a figure that could be entirely consistent
with the quoted accuracy), then one in five attempts to use the system will fail. This
level of service might be adequate for some applications, but would be unlikely to
be sucient for, say, a safety-critical system.
An informative way of describing location system accuracy is via the use of a
cumulative probability graph showing the fraction of readings having an error less
than or equal to some value.
A typical cumulative probability graph (often called cumulative distribution
function of errors or cdf) for an UWB location system can be seen in Figure 6.2.
For example, it can be seen that 50% of the 3D readings produced by the system
lie within 14 cm of the true 3D position, and 95% lie within 39 cm. Comparing
data in this graphical form, designers of dierent systems can therefore use the

Domenico Porcino et al.

417
Locus of points which are equidistant
from the four receivers

RX 1

RX 2

TX

RX 4

RX 3

Figure 6.3. Pathological geometric dilution of precision in a TDOA system.

same graph to determine whether or not the sensor is suitable for their needs and
applications.
When assessing a sensor system that gives 3D information about the position
of an object, it is also important to consider whether quoted accuracy figures refer
to the full 3D position, or only to a 2D component of that position. For example,
a 3D system may be very accurate in two axes, but substantially less accurate in the
third (e.g., the global positioning system [2], GPS, is very accurate in horizontal
position, but relatively inaccurate in height). Manufacturers may quote performance in the accurate plane only.
6.2.6. Geometric dilution of precision
A further complication in the analysis and use of location systems is the geometric
dilution of precision. This eect relates the fundamental ranging or bearing accuracy of the components of a location sensor to its positioning accuracy at some
point in space.
Consider a TDOA system with a mobile transmitter and four receivers as
shown in Figure 6.3. As can be seen, all receivers are equidistant from the transmitter, and so all will report the same pseudorange. However, this condition (that
all reported pseudoranges are equal) exists at all points along the dashed line, and
variations in the value of the pseudorange might be because of changes in the
transmitter-receiver distance or variations in the unknown clock oset, so the location system will be unable to determine where along this line the transmitter lies.
Therefore, the location system will be very inaccurate in the Z dimension with the
transmitter in this X-Y position, because the particular geometry of the transmitters and receivers leads to magnification (in this case, infinite magnification!) of
ranging errors.
Given a particular location system architecture and deployment, it is possible
to calculate in advance the geometric dilution of precision at any point in space
the geometric dilution of precision depends only on the relative positions of the
transmitter and receivers, and the chosen location system architecture. For example, in the example given above, it is not necessary to consider the technology

418

UWB ranging
10

90 cm
70 cm
50 cm

5
30 cm
10 cm
0

10

10

10

Figure 6.4. Simulated 2D location accuracy contours for a hybrid TDOA/AOA system. The simulated
receivers covering the 20 m 20 m space (marked as big dots) are placed at coordinates (5 m, 0 m)
and (5 m, 0 m). The area within a particular location accuracy contour represents the region in which
the 2D accuracy of the system will be better than the value marked on the contour.

involved in measurements, or the measurement accuracy, to identify the inability

of the TDOA system to determine the z-coordinate of a transmitter somewhere on
the dashed line.
The graphs in Figure 6.4 show the expected 2D accuracy of a hybrid TDOA/
AOA system within a 20 m 20 m space.2 Note that in practice, the positioning
accuracy tends to be much worse than might be expected from the ranging (or
bearing) accuracy.
6.3. Comparison of UWB and alternative radio-based systems
Radio-based positioning systems have several general advantages over alternative
systems using optical, acoustic, inertial, or magnetic technology. These advantages
include non-line-of-sight operation, compact and physically robust implementation, and minimal disruption to existing environments.
Radio location systems have been extremely successful in the wide area, and
systems like GPS and LORAN [3] have been used for many years for applications
such as navigation, asset tracking, and surveying. Unfortunately, the traditional
techniques of radio location are dicult to use in indoor environments, because
multipath signals are inevitably presentradio frequency (RF) energy sent from a
transmitter bounces o walls, windows, supporting structures, equipment, and so
on, causing receivers to detect a confusing superposition of signals. Furthermore,
electromagnetic radiation travels at very high speed, and any system attempting to
2The simulation assumes that the TDOA/AOA system can measure ranges accurate to +/ 5 cm

and bearings accurate to +/ 1 .

Domenico Porcino et al.

419

determine the propagation time of a signal from a transmitter to a receiver must

be able to make measurements with extremely high time resolution.
Having said this, several in-building radio location systems have been developed.
Assisted GPS (A-GPS) is a technique that enables GPS receivers to detect satellite signals even when the line of sight to the satellites is obscured by the structure
of a building [4]. Whereas most GPS receivers are self-contained, an A-GPS unit
receives aiding information from another source (e.g., the cellphone network operator) to allow it to lock on to the severely attenuated GPS signals found inside
buildings. Several US cellphone carriers have selected this method for use in enhanced 911 services, but accuracy could be significantly worse than the 10 m accuracy (95%, 2D) of GPS outdoors, and the receiver/aiding-channel combination
is relatively complex and power-hungry. This technique is one of the methods that
might be used for cellular device location [5].
Signal strength systems that estimate transmitter-receiver distances from the
received strength of a reference signal have been developed to allow the location
of wireless LAN nodes within buildings. Unfortunately, received signal strength
varies not only with distance, but also with composition of the media through
which the signal has passed (which may be air, concrete, metal, etc.), and the relative orientations of the transmitting and receiving antennae. Some systems are set
up to give a simple indication of proximity to other nodes, others use a fixed infrastructure of base stations to provide a positioning capability, but reported accuracy
is around 10 m (95%, 2D) [6]. Some other systems, such as Ekahau positioning
system [7], use standard IEEE 802.11 access points, but require extensive surveys
of the building to get enough statistics of the received signal strength (RSSI) to be
able to fingerprint the position of the tag to a claimed accuracy of 1 m.
Systems based on narrowband RF detection. Examples include the following.
(i) The WhereNet location system, a TDOA system operating in the 2.4 GHz
industrial, scientific, and medical (ISM) band [8]. Active tags are attached to the objects to be located, and their signals are detected by a
set of receivers placed at known points. The manufacturers suggest that
this system has an accuracy of a few meters in in-building environments.
(ii) The PinPoint system, which measures the round-trip time of flight of
a spread-spectrum radio signal sent in the ISM bands (2.4 GHz and
5.8 GHz) from a fixed infrastructure of transceivers placed around a
building to active transponders placed on equipment [9]. Again, this
type of system is accurate to within a few meters indoors.
(iii) Bluetooth location systems, which use timing of RF signals between cooperating nodes, and which are being commercialized by companies like
Bluesoft [10]. Some modifications to the Bluetooth radio and protocols
are required to implement this functionality. Accuracy is similar to the
other ISM band systems described above.
Other RF technologies can be used to provide some indication of the proximity of an object to some point in space.

420

UWB ranging

Passive RFID systems (e.g., [11]) infer the presence of nearby objects using
small, passive transponder tags attached to those objects. A reader unit energizes
transponders in its vicinity via electromagnetic coupling, causing the tags to respond with a radio message containing their unique identifier. The tags are cheap
and virtually indestructible, but they must be placed very close to a reader antenna (within a few meters) to be detected, and no precise information regarding
the tags location in relation to the reader can be deduced.
Active RFID systems (e.g., [12]) use battery-powered tags which transmit signals coded with each tags unique identifier. Because they do not rely on electromagnetic coupling to provide power to the tag, they operate over much longer
ranges (up to a few hundred meters) than passive RFID systems, but the tags are
much more expensive.
6.3.1. UWB and indoor location systems
As noted above, one of the major limiting factors in the use of RF positioning
systems within buildings is the presence of multipath signals. UWB location systems have a great advantage in this regard over other RF positioning technologies,
because they are able to make use of bandwidths in the multi-GHz range, rather
than, say, the 80 MHz bandwidth available to systems operating in the 2.4 GHz
ISM band.
As the bandwidth of a transmitter-receiver link increases, so does the ability
of the receiver to distinguish between one incoming signal and a second arriving
a short time afterwards. Because multipath signals must travel a longer distance
from the transmitter to the receiver than the direct path signal, they will arrive at
the receiver later. Therefore, the increased bandwidth available to UWB systems
allows them to better distinguish between line-of-sight signals and those which
have reached a receiver after one reflection (or several reflections) from surfaces
in the environment. Since only the direct path signal conveys useful information
for the purposes of location, the increased system bandwidth results in improved
location accuracy.
6.3.2. UWB location systems outdoors
UWB location systems may also have advantages over other RF-based positioning techniques in outdoor environments. In the areas around buildings (the socalled urban canyon), an UWB location system could supplement other widearea techniques such as GPS which have diculty in those environments due to
RF shadowing. And although systems like GPS can provide very high (centimeter
or even millimeter) accuracy positioning through the use of carrier-phase techniques, these methods can require continuous tracking and might require a long
time to achieve their first position fix from a cold start. An UWB system can often
provide similar tracking accuracy with lower power consumption at the mobile
device and better dynamic response.

Domenico Porcino et al.

421

6.3.3. Disadvantages of UWB location systems

When compared to other RF-based positioning technologies, UWB location systems do have some disadvantages.
Although they can make use of very wide operating bandwidth, the transmitted signal powers permitted by regulation are typically very low, limiting the
operating range of UWB location systems compared to traditional RF location
technologies.
Potential interference concerns have also lead to regulation (in the US, at
least), which shifts the permitted operating frequencies of UWB systems away
from the congested 03 GHz band. However, higher frequencies propagate
through construction materials less easily than lower frequencies, and so the eect
of this aspect of the regulations is to further decrease the ability of UWB location
systems to operate over long ranges indoors.
6.4. A typical RF link budget for UWB positioning systems
The radio frequency (RF) characteristics of the link between an UWB transmitter and an UWB receiver have a very significant influence on key parameters of a
correspondent ranging system. Important system parameters such as the area of
coverage, the needed receiver sensitivity, the noise figure of the receiver, the gains
on the antennas, the path dynamics are all connected and interrelated with the
final quality (accuracy, reliability) and convenience (maximum range, minimum
infrastructure) that the positioning system will be able to oer.
Looking at the positioning system as any wideband radio link between two
UWB devices, we can start a simple analysis considering the constraints under
which our positioning system will need to operate. Table 6.1 will guide the reader
through this simple exercise. In particular, we will need to comply with
(i) international recognized definitions of UWB systems,
(ii) power emission restrictions,
(iii) feasibility of the corresponding UWB parts of the system.
As for the first constraint, we will adhere to the only ocial definition of UWB
systems which is widely recognized (i.e., the UWB definition given in the FCC First
Report and Order [13]). Considering that definition and the likely developments
in other region of the world, we assume a bandwidth for the system larger than
500 MHz. Given that the positioning system is designed with range resolution as
an important target, we set for this link budget a goal of 1800 MHz bandwidth for
the UWB system, with a center frequency around 4 GHz.
The second constraint (power limit) is very important to ensure adequate coverage at feasible receiver sensitivity. To keep within the FCC recommendation, we
fix the average EIRP level to 41.2 dBm/MHz, which, integrated over the 1.8 GHz
chosen for transmission, allows us to send over air about 8.74 dBm (which is
about 0.14 mW).
Other parameters, which are usually very significant on the communications
side, such as the throughput or the coding of the system, are a secondary element

422

UWB ranging

in our link budget as this is considered an auxiliary function to the positioning

system rather than a founding block. For sake of simplicity, we limit our desired
throughput to a modest 60 kbps (just slightly faster than PC analog modems).
Very important are the considerations on the path characteristics. One of the
targets to keep in mind is the maximization of the link distance, with a realistic
set of propagation characteristics (typically far from ideal line of sight). We select
a target of 70 m distance between the transmitter and the receiving units. We then
distinguish three separate elements in the path characterization: a close-in path
loss factor P0 , a far-field path loss at distance d(PL ), correlated to the decay exponent of the signal attenuation, and a lognormal shadowing contribution (PS ). The
formulae applied for the close-in and far-field pass loss are reported


1
,
4d0
d
PL = 10 n log .
d0
P0 = 20 log

(6.3)

The parameters selected for the propagation part of this simple exercise are
in line with current models and reported indoor measurements for UWB systems.
The path loss exponent we consider is n = 2.5, and the standard deviation of the
lognormal shadowing distribution is about 2 dB (i.e., lower than the narrowband
equivalent). The resulting signal loss in the air channel will be
PLtot = P0 + PL + Ps = 44.48 dB + 46.12 dB + 4 dB = 94.6 dB.

(6.4)

Considering two omnidirectional antennas with zero dBi gain, we can see that
the average power arriving to the receiver will be in the order of 103.35 dBm (i.e.,
a tiny 46.23 pW).
Considering a thermal noise floor, KTB, of 4.002 1021 Watts (or equivalently in log form = 174 dBm), our data rate of 60 kbps and a receiver noise
figure of 10 dB (adequate given the low-cost target in mind), we can calculate the
received noise power per bit PN as 116.2 dBm.
Adding the spectral characteristics of a simple binary antipodal modulation
(such as BPSK) with a BER target of 103 (Eb/N0 = 6.6 (1.59) dB = 8.19 dB)
and further implementation losses, as written in (6.5), we still manage to save (see
(6.5)) over 1 dB (LM ) of link margin from our original budget, achieving the target
link distance of 70 m
LM = PR PN PEB PIL = 103.357 dB (116.218) dB 8.19 dB 3 dB = 1.67 dB.
(6.5)
The corresponding UWB positioning system would have access to a spatial
resolution inversely proportional to the usable bandwidth resulting in a discrimination to the level of 17 cm, very good to provide accurate range information to
the applications using the positioning information.

Domenico Porcino et al.

423

Table 6.1. A typical RF link budget for an UWB positioning system.

TX system parameters

Value

Units

Throughput(Rb)
Bandwidth(B)
Average Tx power(Pt)

60
1800
8.7473

Kbps
MHz
dBm

0
4000

dBi
MHz

Link distance(d)
Path loss at 1 m(P0 )
Path loss exponent(n)

70
44.483
2.5

m
dB

Path loss at distance d(PL )

Lognormal shadowing standard deviation
Worst-case shadowing loss(PS )

46.1275
2
4

dB
dB
dB

dBi
dBm

TX antenna gain(GT)
Geometric center frequency(Fc)
Path characteristics

RX system parameters
RX antenna gain(Gr)
RX received power(Pr)

103.36

Noise-related factors
Average noise power per bit
Receiver noise figure

126.22

Average RX noise power per bit(PN)

Modulation minimum Eb/N0 (at BER = 103 )(PEB)
Implementation losses(PIL)

116.22

10

dBm
dB

8.19
3

dBm
dB
dB

Link margin at distance d(LM )

1.6708

dB

Temporal resolution

0.16667

Budget

Many are of course the potential tradeos that could be considered with
changes to this simple UWB positioning link budget for a low data rate system
with positioning capabilities. But the essence of this simple exercise is to show that
building a low cost, low data rate system with accurate positioning feature is feasible with a set of relatively standard ultra-wideband parameters.
6.5. Characteristics of a fine-grained UWB positioning system
The Ubisense location system [14] is one example of a fine-grained UWB indoor
location system. The system is primarily designed to track people and objects in
support of smart space applications these applications involve computing and
communications systems which react to the locations of objects in the environment to make work processes faster, safer, and easier. An example smart space application might be a corporate memory application which automatically records
who meets who, where and when, within an oce environment to allow employees

424

UWB ranging

(a)

(b)

Figure 6.5. Ubisense tag (a) and base station (b).

to more easily recall details of meetings at a future date. Of course, the fundamental UWB tracking technology used in the Ubisense system to support smart space
applications can also be used to solve other (more basic) tracking problems both
indoors and outdoors.
The Ubisense location technology (shown in Figure 6.5) uses small UWB
transmitter tags, which are attached to objects and carried by people. The signals from these tags are detected by a set of base stations placed at known points
around the area to be instrumentedbase stations are networked using the standard structured cabling found in many modern buildings. Base stations are capable of determining both pseudoranges and bearings to tags, and so the system
represents a hybrid TDOA/AOA location technology.
Many features of the Ubisense system result from the desire to support smart
space applications.
High accuracy. Smart space applications use a model of the environment built using location information, and make automatic decisions based on the relative and
absolute positions of objects in that model. If the applications model of the environment diers significantly from the users mental model (constructed using
their senses), then the decisions made by the application could dier from those
which the user would make, leading quickly to user disillusionment with the smart
space application. It is therefore important that the smart space applications sense
the world as well, or nearly as well, as the human user does. The Ubisense system
is designed to achieve 3D positioning accuracy of 15 cm (95% confidence level, averaged across the space covered by the system)the raw pseudoranges and angles
measured by the TDOA/AOA technology are accurate to around 5 cm and 1 .
Figure 6.6 shows the results of measurements taken using a prototype
Ubisense system in late 2003. Two base stations were used to cover a 10 m 10 m
test space, and a tag was placed at points on a 1 meter grid within that space
the tag positions were determined in three dimensions, although the projection
shows only the 2D (x-y) locations of the tags. The area over which readings are
distributed around each point on the unit meter grid gives an indication of the

Domenico Porcino et al.

425

9
8
7
6
5
4
3
2
1
0

5 4 3

Figure 6.6. Tracking results from prototype Ubisense system in a 10 m 10 m space.

accuracy of the system at each of those grid points, and illustrates the diering
magnification of errors due to geometric dilution of precision across the measurement space.
Flexible tracking. Many smart space applications demand real-time performance,
because they perform user-interface functions. An example of this type of smart
space application is a mobile desktop process in an oce environment, which
automatically moves a users personal desktop to any display which they approach.
Real-time performance demands high tracking rates, so that applications can respond instantly to changes in the environment. However, it would be wasteful of
tag battery power to demand that they transmit continuously at the highest rate
which might be appropriate for any smart space application, because that update
rate might only be required for a short period of time in specific circumstances.
The Ubisense system is therefore designed to allow applications to request (via the
location system) dierent tracking rates for individual tags. Smart space applications can change the activity rates for tags depending on the current state of the
environmentfor example, persons might be located only once every few seconds
if they were sitting down, but if they were moving, they might be located a few
times each second. In this way, smart space applications can balance their requirements for real-time location information with power management requirements.
Unlimited scalability. Smart space applications are often the most useful in large
environments where there are many people and many computing and communications devices which interact frequently. Examples of these types of environment might be large hospitals and oce buildings, which can cover tens or hundreds of thousands of square meters, and house thousands of people and items of
equipment. These characteristics demand that tracking systems intended to support smart space applications in these environments must be designed to scale in
terms of the area of coverage and the number of tracked objects. The Ubisense

426

UWB ranging

location technology achieves this scalability using cellular techniqueslarge areas are covered using a number of unit cells, each of which tracks objects over a
smaller area of space. In the Ubisense system, each unit cell uses a small number
of base stations (typically four) and can cover an area of up to around 1000 m2
(depending upon the degree of obstruction within the space). Because each cell
operates essentially independently from neighboring cells, and is capable of tracking many objects within that cell per second, the total area which can be covered
by the system is essentially unlimited, and as the number of cells involved in the
system increases, so does the aggregate tracking capacity of the whole system. A
control system links the cells and ensures that tags hand over from one cell to the
next as they move around the covered space to permit seamless, continuous tracking.
Hybrid TDOA/AOA architecture. Because many smart space applications are driven
by position data relating to people, it is important that the sensor system supporting those applications can locate people eectively. However, the microwave signals suitable for use in UWB systems do not propagate well through the human
body, which tends to shadow emissions from a body-worn tag, and a tag worn by
a person will therefore only see the limited set of base stations which happen
to lie in the unobstructed direction. Although it would be possible to compensate
for this shadowing by deploying more base station infrastructure, it is desirable
for reasons of cost and complexity to keep the level of system infrastructure to a
minimum. Therefore, it is important to ensure that the location system can track
a tag even when only a very few base stations can detect its signal. The hybrid
TDOA/AOA architecture employed by the Ubisense system enables a tags 3D position to be determined using information from only two base stations, making the
technology particularly well suited to the locations of people in cluttered indoor
environments.
6.6. Positioning techniques in harsh environments
One of the myths that surrounded the evolution of UWB technology to the commercial world was that this fantastic technology was multipath-free and would
allow the existence of joint data communications and positioning in almost any
environment. While it is undeniable that the temporal resolution of UWB signals
is much higher than the one oered by typical narrowband systems, it is clear to
any engineer that this resolution is finite and fading eects will be registered in
any relatively dense multipath area. From a positioning point of view, the presence of multipath might create diculties in distinguishing the leading trail of a
series of return signals from its reflections o walls or close-by objects. This will
render the estimation of the time of flight between transmitter and receiver (and
consequently the estimation of the relative distance) much more complicated than
in free-space propagating conditions. Fundamental importance in the dense multipath case has to be given to the exact propagation characteristics of the UWB
signals over air, which is usually simulated with the help of a channel model.

Domenico Porcino et al.

427

A radio channel is in fact a complex mathematical attempt to describe the

propagation phenomena trough air, physical objects, and people. The interaction
among electromagnetic waves and the real world is extremely dicult to predict
reliably and is clearly strongly dependent on the very precise details of the propagation scenario chosen to represent the typical case under analysis.
The radio-mobile channel has to physically represent the sum of all the eects
of loss and distortion that the signals undergo during their propagation from a
transmitter to a receiver. In the case of studies of UWB propagation through nonline-of-sight conditions, we will be interested in knowing how the UWB signals
will propagate through air and how this might aect our link budget.
The main eects that a radio wave encounters during its propagation can be
divided into the following.
(i) Long-term (path loss) characteristics. The long-term characteristics of
the radio-mobile channel describe how the mean signal will behave in
function of the distance at a given frequency. The loss is gradual with
received power decreasing almost as an exponential decay in logarithm
scale (represented by the decay facto n introduced in Section 6.4).
(ii) Medium-term (shadow fading) characteristics. The shadow fading shows
the time-varying factors, such as shadowing from buildings or big objects. It is added on top of the path loss and represented as a random
fluctuation with a lognormal distribution, with a standard deviation
dependent on propagation conditions.
(iii) Short-term (multipath or fast fading) characteristics. The short-term
characteristics (fast fading) describe the sudden variations on the received signal strength due to multipath and reflections coming o buildings and objects. This factor is superimposed on the long-term variation. These variations will typically happen for very small movements of
the mobile antenna: a few wavelengths are enough to cause fast fading
(as an example, at 4 GHz, the wavelength is just 7.5 cm).
These three eects will of course be summed together and will not easily be
discernible in environments where the propagation conditions are cluttered.
A more formal classification of environments based on the propagation conditions can be done with postprocessing and analysis of the path loss characteristics as registered in loco. In particular, from the path loss analysis, we can define
three dierent cases.
(i) Line of sight (LOS) is a case when a signal reaches the receiver directly
from the transmitter, that is, when there are no obstacles in the transmission path. In this case, the leading path is clearly distinguishable from
reflections and its energy is above 80% of any potential reflection.
(ii) OBS (obstructed line of sight), or soft (NLOS) soft non-line-of-sight is a
situation in which the transmission path is lightly obstructed but energy
is still recoverable from the direct path. In this case, the power coming
from reflections is dominant with respect to the power coming from the
direct path, (energy from reflected path energy from direct path).

428

UWB ranging

(iii) Hard NLOS is a situation of almost total absence of any penetration in a

direct path. In this case, the direct path between transmitter and receiver
is totally obstructed and the energy from reflected path is greater than
90% of total recovered energy.
The nice features of UWB systems are usually simulated or demonstrated in
environments where line of sight (or at worst soft line of sight) is guaranteed, but
many of the interesting applications for UWB as an asset tracking system will need
to be able to handle much more challenging conditions and combat the multipath. Severe multipath makes in fact the process of estimating the time of flight
extremely dicult and its eects need to be removed.
As in the case of narrowband technologies (see [15]) also for UWB systems,
there is a clear need to guarantee reasonable performance in harsh conditions and
it is therefore necessary to develop appropriate and reliable techniques for signal
detection in multipath.
Two dierent families of signal detection in dense multipath can be applied:
(i) techniques based on equalization,
(ii) techniques based on statistics.
Techniques based on equalization usually estimate the multipath delays introduced by the radio channel and compensate the received signal for them, trying
eectively to remove any spurious echoes from the received signal. A critical element in this receiver structure will be the channel sounder, which will determine
the coecients and weights of the reflected paths and consequent signal processing to apply to them to recover a reasonable estimate of the leading edge of the
direct signal.
Techniques based on statistics rely on a large number of measures (ideally
statistically independent) to derive statistical data on the temporal variation of
received signals. They can use all or only the part of collected data on which the
level of confidence is higher. Statistics can be of first order (averaging) or higher
orders (such as those used in MUSIC or ESPIRIT algorithms, [16, 17]).
A good example of an UWB positioning system designed to be reliable in extremely harsh conditions is the precision asset location (PAL) system developed
by Multispectral Solutions (MSSI) for the Oce of Naval Research (ONR) Naval
Total Asset Visibility (NTAV) Program. This governmental study [18] investigated
technologies for asset visibility and tracking on board of navy ships where hundreds of containers are transported at any time.
As shown in Figure 6.7, the conditions in the container ship are extremely
dicult for radio wave propagation, with floor-to-ceiling metal walls surrounding
every room. The experiments carried out by the University of Southern California
to characterize this environment [19] report very problematic propagation conditions. The severe attenuation and numerous reflections of these environments
cause deep nulls (3040 dB) in the received signals and very long delay spreads
(3 sec at the 20 dB points), over 10 times longer than what is recorded in any
oce conditions. In such harsh operating conditions, very numerous are the occurrences of total blockage of the signal, and this will cause severe strain to the
signal processing algorithms designed into the PAL signal recovery.

Domenico Porcino et al.

429

(a)

(b)

(c)
Figure 6.7. Harsh environments for UWB positioning systems (from [18, 19]).

The initial prototype PAL system consisted of a set of four passive UWB receivers, an UWB reference tag (positioned at a known location), and a collection
of UWB active tags. It transmitted short pulses over an instantaneous bandwidth
of approximately 400 MHz. The average energy sent over air was 5 nW, equivalent to 1.25 1011 Watts/MHz (79 dBm/MHz; i.e., well below FCC limits) in
the L-band (frequencies centered around 1.5 GHz). The following commercial developments of the PAL system into the Sapphire Precision Asset Location System
lead to a higher operation frequency of 6.16.6 GHz and FCC certification, but
still maintained a great resilience to harsh conditions.
Even in these challenging conditions, the UWB technology implemented in
the PAL system was reportedly able to measure positions in open space with an
RMS accuracy between 0.9 and 1.5 m. It also achieved 1 meter accuracy in the measurements of single-high stacked containers and about 3.5 meter accuracy with
double-high stacked containers (under heavy blockage).
6.7. UWB precise ranging with an experimental antenna-array system
Another example of an indoor location system is an experimental antenna-array
localizer developed at Technical University of Ilmenau. The system is designed to

430

UWB ranging

Tx

Target

y
Circle
centered
at Rx 1

[x(t), y(t)] = ?
Circle
centered
at Rx 2

s2 (t)

s1 (t)

[0, 0]
origin

Rx 1
D/2

Rx 2

D/2

Figure 6.8. Active positioning using two reference antennas at known positions.

investigate the precision of UWB ranging and positioning with colocated and spatially distributed antennas and to push the accuracy close to its limits in a controlled environment.
The localization techniques subjects of this section allow two dierent positioning approaches: an active and a passive approach. In the following pages, the
fundamentals of both methods will be considered in more detail and experimental
results carried out with UWB transmissions in the frequency range 0.8 to 5 GHz
will be shown.
6.7.1. Positioningactive and passive approaches
The active method presumes that the target device (which is the device to be
tracked) cooperates, that is, actively interacts with the positioning device. In order to cover this proactive role, the target has to include also a transmit or receive
unit and collaborate in the ranging measurements. The passive approach in contrast is based on a standalone positioning system exploiting simple wave scattering by the (unaware) target as in the case of a radar system. Both approaches are
dealing with trilateration principles, as briefly introduced at the beginning of this
chapter (see Figure 6.1 and (6.1), (6.2)).
Figures 6.8 and 6.9 represent the fundamental layout of the active and passive
positioning techniques based on time-delay measurements. In both arrangements,
simple uniform linear antenna arrays can be used if the up-down ambiguity (as
shown in Figure 6.1) is not in the focus of the investigations. In the case of necessity
to avoid this ambiguity, the linear antenna configuration has to be replaced by
a triangular one for 2D positioning (or a tetrahedron for 3D positioning). The
actual number of antennas used in a practical system usually exceeded the required
minimum, as shown in the figures. This redundancy allows working with an overdetermined system of equations resulting in solutions with a statistically improved
confidence. For simplicity, the figures are restricted to the 2D case.

431

Ellipse with focal

points Tx and Rx 1

s1 (t)

Rx 1

Target

Domenico Porcino et al.

s0 (t)
Tx
D

[x(t), y(t)] = ?
Ellipse with focal
points Tx and Rx 2
s2 (t)
Rx 2
x

Figure 6.9. Passive positioning by a three-element linear array.

Figure 6.8 represents the geometrical basics of the active approach. In this example it was assumed that the target object is carrying a radio transmitter and
the positioning unit (which would be the master in a master/slave configuration)
contains the receiver with a two-element antenna-array. The positioning unit estimates the times of flight a1 and a2 of the electromagnetic waves arriving from
the transmitter (target) to each of the antenna-array elements. Each of these time
intervals determines the radius of a circle centred at the corresponding receive antenna element. The target will be standing on both circles, and therefore will need
to be also at the interception point of the two. The radiuses of the two circles are
referred to as s1 (t) and s2 (t). The position of the target is determined as one of the
two intersections of the circles above described. Finding the position of the target,
that is, solving its cartesian coordinates [x(t),y(t)] will therefore be equivalent to
solving the equations



2
D 2
+ y(t)2 = s1 (t)2 = ca1 (t) ,
2



2
D 2
x(t) +
+ y(t)2 = s2 (t)2 = ca2 (t) .
2

x(t) +

(6.6)

Note that the indicated arrangement presumes an exact synchronization between the transmitter and the receivers. In a laboratory experiment, this hypothesis can be met by synchronizing transmitter and receiver units by wire (using a
cable connecting both units). In practical situations, the synchronization problem will need to be dealt with and consequently an additional receive channel is
required in order to measure the time-dierence of arrival. The corresponding
equations in such unsynchronized case will be slightly more complicated as mentioned in Section 6.2.
In the case of the passive approach, the target (at an unknown position) acts
as a scatterer and the positioning device works in a radar-type mode without any
cooperation from it. In these conditions, the round-trip time Tx-target-Rx is measured. This information will give access to the target locus. If two independent
monostatic radar systems (i.e., with identical receive and transmit antennas) are
used, the target locus can also be calculated from (6.6) replacing the time of flight
by half of the round-trip time.

432

UWB ranging

From the viewpoint of device complexity, it is easier to deal with multistatic

radar arrangement consisting of one transmit element and a number of receive
elements at dierent locations. Figure 6.9 demonstrates the simplest possible configuration.
Obviously, the measured round-tripe times p1 and p2 determine two ellipses; those focal points are determined by the locus of the transmit antenna and
the loci of the corresponding receive antennas. Thus, the target position results
from the intersection of both ellipses:
Tx-Rx1:


x(t) + D/2
a1 (t)

2

y(t)
b1 (t)

2
= 1,

(6.7)

= 1.

(6.8)

Tx-Rx2:


x(t) D/2
a2 (t)

2

y(t)
b2 (t)

2

The length of the main axis 2ai is given by the appropriate round-trip time
pi ,
2ai (t) = s0 (t) + si (t) = c pi (t)

(6.9)

and the length of the minor 2bi axis calculates from



D
2

2

+ bi2 (t) = a2i (t).

(6.10)

Each of the two mentioned approaches, the active and the passive ones, has its
specific advantages and disadvantages.
In the case of the active approach.
(i) the cooperation of the object to be localized is required,
(ii) an exact synchronization between transmitter and receiver is necessary
unless a time-dierence-of-arrival approach is applied. This would restrict the sophisticated demands on synchronization to the receivers
only,
(iii) the signal could be made more easily detectable compared with the passive case,
(iv) covering of larger areas is possible (depending on the regulatory constraints on max emitted power).
In the case of the passive approach, there exist the following.
(i) There is no need for object cooperation. That is, the objects to be localized do not have to be equipped beforehand and the system is working
in a noncooperative environment.

Domenico Porcino et al.

433

(ii) There are no synchronization problems since all system components can
be mutually wired (at least in an experimental setup).
(iii) The precision of position is limited by the size of the target unless its
scattering behavior is known and can be respected by solving the positioning equations.
(iv) The signal detection is quite more demanding due to interference with
scattering at unwanted and mostly dominant objects.
(v) Compared to the active approach, the covered area is restricted due to
higher path loss due to the double path length (transmitter target plus
target receivers) necessary before detection.
Generally speaking, the precision of the positioning estimate depends heavily
on the array geometry in connection with the actual locus of the target, the precision of the time delay estimator, and also on the design of the localizer unit provide
a stable ultra-wideband sounding signal. In the case of passive approach, the scattering behavior of the target will also play a significant part in the detection and
relative accuracy. Furthermore, appropriate antenna models and correction methods are needed if the required precision comes down to the antenna size or even
below. As can be seen, the large bandwidth and good temporal resolution of UWB
localization systems are an essential base for an accurate positioning calculation,
but are not sucient when the requirements call for extremely accurate devices.
6.7.2. Localization unit using ultra-wideband M-sequences
The theoretical calculation and system considerations done up to now are the
basis for the design of an accurate antenna-array localization system, as used in
our practical trials. The following section shortly considers the functioning of the
UWB electronics used in the experiments and some related aspects. Besides a wide
bandwidth, the UWB measuring system has to meet the following requirements:
(i) a high data-recording rate in order to be a match for the time variance
of the target to be positioned,
(ii) a multichannel arrangement capability in order to provide data for 2D
or 3D target localization,
(iii) a high degree of hardware configuration flexibility to adapt the system
performance to the actual requirements of the individual user.
There are dierent possibilities to generate excitation signals covering a wide
spectral band:
(i) chirp signals,
(ii) stepped sine wave,
(iii) short-impulse signals,
(iv) pseudorandom binary sequences and others.
However, not all of these signals are suitable to be used in the real-time localizing systems. For example, besides its cost, the network analyzer (stepped sine
wave signals) can easily meet the UWB bandwidth demand (even if not the instantaneous bandwidth demands of the FCC regulations). But, although suitable

434

UWB ranging

multichannel arrangements might exist, the real-time operation is prohibited by

the slow measurement rate of typical network analyzers. Other systems, for example, the short-impulse- based systems, oer real-time operation but they are often
subject to a higher susceptibility to jitter and drift.
The key to a powerful UWB localizer is the use of an appropriate stimulus
signal because the whole device hardware and signal processing depend upon it.
Some of the key points considering a suitable stimulus are the following.
(i) The stimulus must be an UWB signal that allows real-time operation.
Sequentially stepped narrowband signals are not suitable because they
prevent real-time operation due to the dead time during the system settling.
(ii) The stimulus must be generated in a stable manner by simple means up
to several GHz bandwidth.
(iii) The stimulus must be periodic in order to apply cost-eective subsampling methods for signal recording and to avoid a spectral bias error. It
should be allowed to work with a certain degree of subsampling without
loss of data due to the time variation of the channel.
(iv) The stimulus must have a low crest factor. The crest factor is defined
by the ratio between the RMS value and the peak value of a signal. Low
crest factor signals distribute their energy uniformly over the time. This
maximizes signals energy even at low peak voltages. However, impulse
compression (e.g., by correlation) is required in order to gain the impulse response on which the delay-time estimation is based. Impulse
compression further improves the SNR due to the correlation gain.
(v) The stimulus must follow the FCC rules concerning the covered frequency band and restrictions on power emission.
A family of signals that meets most of these strict requirements is formed by the
pseudorandom binary sequences, for example, M-sequence (maximum-length binary sequenceMLBS) . An M-sequence can be easily generated up to dozens of
GHz of bandwidth via a digital shift register, which is clocked by a stable singletone RF oscillator. Besides the advantage of having a reasonable correlation gain,
the M-sequence signals are characterized by small binary voltage amplitudes that
allow extremely fast digital switching in integrated circuit technology to meet the
demanding requirements on bandwidth and low jitter.
Figure 6.10 shows one channel of the UWB localizer, which was used for the
positioning experiments. It is working in the baseband covering the spectral band
from DC up to fc /2. Controlled by a single tone clock, a digital shift register generates the M-sequence. Since the M-sequence signal is periodical and the measurement scenario can be assumed to be locally stationary, it is possible to acquire the
MLBS signal by an undersampling approach. Here, the binary divider (2m ) determines the undersampling factor and provides the receiver sampling clock. The
measurement data are captured by a track-and-hold circuit (T&H), transformed
into the digital domain (ADC), optionally synchronously averaged (p is number
of averages), and finally stored (MEM) for oine processing or online processing
(DSP).

Domenico Porcino et al.

435
Enable
Clock rate fc

System
clock

Output
n-stage
shift register

2m
Sampling rate fs

IR

MEM
DSP

ADC

T&H

Input

Figure 6.10. Basic block diagram of the UWB RF interface.

As mentioned above, the impulse response (IR) results from an impulse compression, which is performed by the fast Hadamard-transform (FHT). The FHTalgorithm is very close to a fast Fourier transform algorithm except that it is based
on a pure summing of data samples, which promises very fast operation in a custom hardware implementation.
The architecture illustrated in Figure 6.10 represents only one channel of the
UWB localizer. For the position estimation (2D, or 3D), a multichannel arrangement is necessary. As seen from Figure 6.10, the circuits for signal generation and
signal capturing have a very simple structure. Since it is built from cost-eective
large-scale components or custom integrated circuits, minimizing the number of
components is not a fundamental target for the overall system. With this consideration, it looks possible to build a true multichannel and multiarray UWB locationing system at reasonable cost.
Figure 6.11 shows one possibility of a multichannel UWB localizer architecture. In this version, all receiving channels work continuously in parallel providing the shortest measurement time. The transmitter modules work sequentially
in time by activating only one shift-register output per measurement cycle. Thus,
the overall measurement time increases with the number of transmitter channels
and does not depend on the number of receiver channels. It should be noted that
usually even one single transmitter is sucient for most localizers applications.
In order to meet the FCC rules completely with the M-sequence approach,
the operational band has to shift to the range from 3 to 10 GHz approximately.
Figure 6.12 represents a possible schema of such accurate measurement device. It
is based on a homodyne concept, which allows for operation in arbitrary bands.
Additionally, the usable bandwidth doubles compared to the baseband concepts
shown earlier and it corresponds now to the clock rate of the shift register.
The system used for experiments of extremely accurate positioning was an
M-sequence multichannel UWB device [20, 21] operating in the frequency band
from DC up to 5 GHz . It was built from custom integrated circuits in SiGe:Ctechnology (shift register, binary divider, and T&H) that have been designed at
Ilmenau University of Technology in cooperation with MEODAT.
The extremely linear time axis and the superior jitter and drift behaviour
(compared to traditional sequential sampling oscilloscopes) are the result of the

436

UWB ranging
Digital control
gates
On/o
switched
output

n-stage
shift register

System
clock
2m

CIRF MEM
DSP

ADC

T&H

Permanently
working input

Figure 6.11. Multichannel UWB localizer.

Carrier fRF
IQ demodulator
Clock fc

Enable
n-stage
shift
register

m-stage
binary
divider

SUT

Sampler To processing
ADC

Sampler
To processing
ADC

/2

Figure 6.12. Extended UWB localizer for operation at arbitrary frequency bands. The scenario under
test (SUT) covers the antennas and the propagation path.

synchronous digital controlled sampling. The DSP module of the described experimental systems is based on standard o-shelf PCB products. The ADC was a
12-bit-video ADC and the sampling frequency was 17.68 MHz (depending on the
adjusted undersampling factor). An example of the complete RF electronics of a
device with 3 transmitter units and 4 receiver units is illustrated in Figure 6.13.
Initially, the UWB electronic components were developed for a minedetection radar. Due to its sucient large ambiguity range, it could also be used
for very accurate positioning purposes. Since the transmit and receive units are
jointly designed on one board, the system is better suited for passive positioning
tasks than for active one. Nevertheless under laboratory conditions, it can also be
applied to analyze the active approach by connecting the moving antennas via long
cables.
6.7.3. Experimental results
The experimental UWB system described in the previous section can demonstrate
the potential performance expressed in terms of the range resolution and precision
of the 2D/3D position estimation using both an active and a passive approach.

Domenico Porcino et al.

437

Figure 6.13. Experimental UWB localizer RF electronics (courtesy of MEODAT GmbH Ilmenau).

Range estimation. The accuracy of the range estimation is the most critical parameter of a positioning procedure. A first experiment demonstrated the excellent range resolution of an UWB system under idealized propagation conditions.
The expression describing the typical range resolution in transmit and back-scatter
mode, respectively, is given by
(i) transmit mode (active approach):
Rt

c
;
B

(6.11)

(ii) back-scatter mode (passive approach):

Rs

c
,
2B

(6.12)

where c represents the speed of propagation of radio waves (assumed

equal to the speed of light) and B the bandwidth of our system.
A positioning device having a bandwidth of about 5 GHz (as in the case of this example) provides a range resolution of Rt = 6 cm (active approach) and Rs = 3 cm
(passive approach) according to (6.11) and (6.12). But this range resolution can be
drastically improved by exploiting super-resolution techniques, on the condition
that there is some knowledge about the expected received signal. The following
example will show how accurate an UWB system can be made beyond the natural range resolution. In a simple experiment, the transmitter and receiver were
connected through a mechanical delay line allowing 200 m distance adjustments.
The standard deviation of the mechanical distance adjustment was estimated to
between 50 and 100 m. Thirty eight measurements were taken at dierent delays adjusted by the delay line (20 measurements with 200 m delay increments, 9
measurements with 2 mm delay increments, and 9 measurements with 2 cm delay
increments). Since the variable (perfect) delay line does not aect the time shape of
the propagating wave, the temporal pulse position can be estimated by means of a
high-resolution maximum-likelihood (ML) estimator as used for wideband channel sounding [22]. Beside the noise, the only unknown signal which degrades the
resolution was initiated by the cross-coupling within the RF electronics. Intentionally, this was not removed from the measurement signal with a prior calibration.

438

UWB ranging
800

Estimation error (m)

600
400
200
0
200
400
600

10

15
20
25
30
Measurement index

35

40

Figure 6.14. Error in the distance estimation with the experimental multiarray UWB system.

The estimation error resulting from this experiment is presented in Figure

6.14. It can be seen that for small delay steps (first 20 scans), the total positioning
estimation error is well below 200 mm. This is more than 300 times better than
predicted by (6.11). It is to be noted also that the total error covers the error of the
delay estimation as well as the mechanical adjustment error. The error increases at
larger delay increments since the cross-coupling gains more power on the overall
signal. By respecting the cross coupling signal, the estimation error can be reduced
and kept constant for dierent delay increments.
This experiment indicates that a quite high precision can be achieved if the
transmit signal is known by the processing unit and altered by the propagation.
This is close to the case of a real transmission in the active approach mode under line-of-sight conditions, provided that the impulse response function of the
involved antennas is independent from the radiation angle (which is not easy to
guarantee). In any other nonideal case, the resolution will degrade in function of
how much information is missing from the target. Some of the elements that will
make dicult the development of a very precise positioning system are
(i) non-line-of-sight propagation conditions (radiation through walls, edge
diraction, diraction at wires, tubes colons, etc),
(ii) back scattering from objects of limited size (passive approach) except for
purely specular scattering which only changes the sign of the amplitude
but not the signal shape,
(iii) angle-dependent radiation behavior of the antennas.
2D position estimation. The experimental multi-array UWB system introduced in
the previous section has also been used for further accuracy testing. The following
two examples demonstrate the UWB localizer ability to estimate 2D object position using both the passive and active positioning techniques. In both cases, the
objects to be tracked were mounted on a positioning unit [23] in order to have a

Domenico Porcino et al.

439

Positioning
device
Test bed
filled with sand
Transmit
antenna

Receive antenna
array

(a)

(b)

(c)
Figure 6.15. (a) Experimental positioning device in the measurement laboratory and (b), (c) pictures
of the antennas used in the measurements ((c) the Vivaldi antenna array and (b) the biconical antenna).

precise reference for error determination. The maximum inaccuracy of the positioning system was quantified in 0.75 mm for all axis within an area of 2 meters
by 4 meters. Continuous sweeps can be carried out with a maximum displacement speed of about 130 mm/s in the x-direction and 170 mm/s in the y-direction,
respectively. The measurement laboratory with the positioning test bed and antenna arrays used for the experiments can be seen in Figure 6.15. The scanner was
mounted at the ceiling level above a sandbox since its original purpose was radar
imaging of buried objects.
The objects were moved along a predefined track (corner points in m: [0; 0],
[0; 2.5], [0.8; 2.5], [0.8; 1.5], [0.9; 1.5], and [0.9; 0]) with the maximum velocity allowed by the positioning unit. The data was collected and processed by the
UWB localizer described above. The measurement rate was reduced to about 11
scans per second which was sucient given the slow movements. The whole measurement lasted about 90 seconds. A linear antenna array containing four Vivaldi
antennas was used for receiving and a biconical omnidirectional antenna (Figure
6.15) was introduced as the transmit antenna (for both active and passive approaches).

440

UWB ranging

Passive approach. In the case of passive positioning approach, the object was a
metal box of the size 0.5 m 0.5 m 0.1 m. The radiation pattern of the box was
not taken into account. Thus, the positioning accuracy is comparable only with the
box size since its radiation pattern (and by that, the shape of the back-scattered
signals) is strongly angle-dependent. The result of the positioning using the ML
based estimator is illustrated in Figure 6.16(a). The solid line represents the actual
object track. The estimated positions are represented by dots. Obviously, there are
some dierences between the two tracks, but the accuracy was very good. The
main cause of positioning errors was the scattering behavior of the tracked object.
Active approach. In the case of active positioning, the metal box was replaced by
the biconical antenna, which was connected via a long cable with the positioning device. Thus problems of the transmitter-receiver synchronization were neglected.
The estimation results using the ML-based estimator are displayed in Figure
6.16(b). Here, the estimated positions follow the real track much better than it
was in the case of the passive approach. The main reason was that the radiation
center of the antenna used in the second experiment was well defined (note, this
will change by going into the third dimension), and thus the shape of the LOS
signal was not varying. The standard deviation of the position error of the second
experiment was about 1.5 cm in both x- and y-directions. The maximum error of
azimuth angle ranged about one degree.
3D position estimation. The Technical University of Ilmenau experimental UWB
multiarray test bed was also used for 3D positioning measurements. An example of
3D positioning is illustrated in Figure 6.17. Here, the transmit antenna was moved
by hand in the air writing the year 2003. The dotted line represents the estimated
positions. In the right part of the figure, the written track is zoomed out. To improve the understanding, the contours of the laboratory space (big cube) and the
sandbox (smaller cube) were added to the picture.
The technical description and practical experiments shown in this paragraph
provide a useful reference to understand the potential precision of UWB multiarray positioning systems. Even if some limits are registered in practical situations,
the test bed designed by the Technical University of Ilmenau and MEODAT clearly
shows that UWB multiarray active localization units can be the solution to the
problem of extreme accuracy, when resolution to the level of 1 cm is needed.
6.8. Systems integration and UWB positioning technology
When considering deployment of UWB location technology across a large area
(such as that shown in Figure 6.18), it is important to consider the system-level
issues associated with such deployments (as opposed to the technical issues related to individual ranging or angle measurements). If systems-level issues are not
dealt with adequately in the design of the location sensor, performance will suer,
especially as the system scales beyond the simplest configurations.

Domenico Porcino et al.

441

2.5

Moving object

Y direction (m)

2
1.5
1
Receiver antenna
array

0.5
Transmitter antenna
0
1

0.5

0
0.5
X direction (m)

(a)

2.5

Y direction (m)

2
1.5
Moving transmitter
array

1
0.5

Receiver antenna
array

0
1

0.5

0.5

X direction (m)
(b)
Figure 6.16. (a) Results of the 2D position estimation with a passive approach; (b) results of the same
estimation with an active approach.

6.8.1. The role of infrastructure

The organization of the location system, and in particular the role of infrastructure within it, governs many of the system-level design decisions. Two distinct
system organizations are frequently discussedinfrastructure-based and ad hoc.
Infrastructure-based location systems involve a static set of equipment whose purpose is to support the tracking of a population of mobile devices. In ad hoc systems, on the other hand, the devices to be located also perform the function of
locating other devicesthe devices interacting in the system do not necessarily
have to be mobile, but the location system has the property that if they are moved,
then the system will continue to operate correctly.

442

UWB ranging
2.5
Z-axis (m)

2
1.5
1
0.5
0
0.5

1
0
X-axis (m) 1

Y -axis (m)
(a)

Elements of antenna array

(b)
Figure 6.17. (a) The 3D position estimation with active approach; (b) a detailed view.

Each of these two system organizations has advantages and disadvantages.

There is clearly a cost involved in installing infrastructure within an environment.
However, this cost can be reduced by an appropriate design (quantified before the
system is installed), and a properly designed infrastructure which will have minimal maintenance requirements once installed. Fixed infrastructure also lets the
system designers give predictions and guarantees about system performance and
give users confidence that the system will be robust. Furthermore, in many environments where UWB location technology will be deployed, infrastructure is already accepted. Hospitals and oces, for example, have substantial infrastructure
in the form of electrical wiring, air conditioning, pipework, telephone cabling,
computer networking, and so on. In these target environments, there already exist
well-developed processes for managing infrastructure, and it is likely that similar
processes can and will be used to support installed location sensor technology.

Domenico Porcino et al.

443

Figure 6.18. Example of Ubisense smart localization in a complex oce infrastructure.

Of course, some environments, such as denied territory during warfare, may

not be amenable to the deployment of sensor infrastructure. Proponents of ad hoc
systems architectures point to applications in these environments as being well
suited to ad hoc techniques. On the other hand, systems with ad hoc architectures
can be more unpredictablethe system performance at any particular point in
time will depend on the current physical organization of the system (because of
eects like geometric dilution of precision, discussed earlier in this chapter). Consequently, ad hoc systems do not have an inherent cost of deployment, but may
involve a significant maintenance cost if a minimal level of performance must be
guaranteed over time.
6.8.2. Surveying techniques
All the positioning techniques discussed in this chapter involve determining the
location of a mobile object by measuring properties such as its range or bearing
from points in space which are already known. Surveying of sensors is therefore an
important step in the deployment of any location system, because a poor survey
will result in poor accuracy during operation.
Infrastructure-based location systems involved fixed sensor components,
which must be surveyed only once (to create survey maps such as those shown in
Figure 6.19). It will normally be possible to complete the survey using equipment
which would not normally be considered for the location task because of constraints of cost, size, power, and so on. For example, an infrastructure-based UWB
location system might be surveyed with a laser total station, a tripod-mounted
device that uses two-way laser ranging to measure the position of a point in 3D

444

UWB ranging

Figure 6.19. 3D mapping of building infrastructure.

space to an accuracy of one or two millimeters. A laser total station would be totally impractical as a sensing tool for many tracking applicationsalthough it is
very accurate, it is power-hungry, line of sight only, large, and fragilebut its use
(once) to survey the system enables the known points to be located very accurately.
It is also possible to survey UWB location systems using measurements of
tracked devices which have been placed at known points in spacein eect, selfsurveying the system by applying the nonlinear models discussed earlier in this
chapter in reverse. Of course, location systems with ad hoc architectures must rely
on these techniques, because they cannot depend on an external surveying mechanism to operate if a component of the system moves. Although these approaches
are theoretically attractive, the limited accuracy of UWB location systems (when
compared to technologies such as laser total stations) leads to worse survey accuracy, and hence worse operational accuracy.
Whichever survey route is taken, the surveying process must be robust to
the eects of geometric dilution of precision (as described earlier) and multipath,
which both introduce into the surveyed position systematic errors which cannot
be mitigated by simple averaging.
6.8.3. System scalability
A sensor system deployed across a large building such as a hospital could involve
very high aggregate data rates, potentially up to tens of thousands of events per
second. At the same time, several applications may be running simultaneously
on top of the common sensor infrastructurefor example, a hospital may want
applications for security, oce productivity, and intelligent building infrastructure, in addition to specific healthcare applications. Furthermore, many of those

Domenico Porcino et al.

445

applications, such as the smart space applications mentioned previously, could demand real-time performance.
Maintaining low latency in a system whilst processing and distributing large
volumes of location information represents a significant challenge. Centralized
approaches to data management, such as those used by traditional Geographic
Information System (GIS) databases are unlikely to be suciently scalable, and
distributed system techniques must be used to partition the data processing and
transfer tasks. These methods will likely be used in any case to support multiple
applications that share the same sensor system, and to integrate the sensor system and location management platform with other software systems in the target
environment.
6.8.4. Sensor independence
Whichever variety of UWB technology is used to implement a large-scale sensor
deployment, it is desirable to make the applications of the system and the particular sensors being used independent of one another . By separating these components of the system, it becomes possible to replace the underlying sensor system
without requiring significant (and costly) changes to the applications themselves.
Changes in the fundamental sensor system may become desirable as UWB technology matures and the regulatory landscape changes.
The Ubisense system described earlier, as an example, employs a locationaware middleware layer to isolate applications and sensors. One of the middlewares tasks is to convert the raw location information from the sensor system
(which will tend to have sensor-specific characteristics) into application-relevant
but sensor-independent spatial facts (such as The user is standing in front of the
workstation.). By writing code which only deals with sensor-independent spatial relationships, developers can ensure that their applications are future-proofed
against changes in the sensor technology further down the system stack.
Bibliography
[1] A. Harter, A. Hopper, P. Steggles, A. Ward, and P. Webster, The anatomy of a context-aware
application, in Proceedings of 5th ACM/IEEE International Conference on Mobile Computing and
Networking (MOBICOM 99), pp. 5968, Seattle, Wash, USA, August 1999, ACM Press.
[2] I. A. Getting, Perspective/navigationThe Global Positioning System, IEEE Spectrum, vol. 30,
no. 12, pp. 3638, 4347, 1993.
[3] G. J. Sonnenberg, Radar and Electronic Navigation, Butterworths, London, UK, 1998.
[4] G. M. Djuknic and R. E. Richton, Geolocation and assisted GPS, IEEE Computer, vol. 34, no. 2,
pp. 123125, 2001.
[5] D. Porcino, Location of third generation mobile devices: a comparison between terrestrial and
satellite positioning systems, in IEEE 53rd Vehicular Technology Conference (VTC 01), vol. 4, pp.
29702974, Rhodes, Greece, May 2001.
[6] P. Bahl and V. N. Padmanabhan, RADAR: an in-building RF-based user location and tracking
system, in Proceedings of 19th Annual Joint Conference of the IEEE Computer and Communications
Societies (INFOCOM 00), vol. 2, pp. 775784, Tel Aviv, Israel, March 2000.
[7] Ekahau Incorporated website, 2005, http://www.ekahau.com/.
[8] WhereNet website, 2005, http://www.wherenet.com/.
[9] C. Lanzl and J. Webb, Position Location Finds Applications. Wireless System Design, June 1998.

446

UWB ranging

[10] Bluesoft Incorporated website, 2001, http://www.bluesoft-inc.com/.

[11] R. Want and D. M. Russell, Ubiquitous electronic tagging, IEEE Distributed Systems Online,
vol. 1, no. 2, 2000.
[12] Wavetrend Technologies Limited website, 2005, http://www.wavetrend.net/.
[13] Federal Communications Commission (FCC), Revision of Part 15 of the commissions rules
regarding ultra-wideband transmission systems, First Report and Order, ET Docket 98-153, FCC
02-8, adopted/released February 14/April 22, 2002.
[14] Ubisense Limited website, 2005, http://www.ubisense.net/.
[15] D. Porcino and A. Yule, Positioning Technology for cellular mobile devices, in IEEE Vehicular
Technology Conference (VTC 01), Rhodes, Greece, May 2001.
[16] H. Saarnisaari, TLS-ESPRIT in a time delay estimation, in IEEE 47th Vehicular Technology Conference (VTC 97), vol. 3, pp. 16191623, Phoenix, Ariz, USA, May 1997.
[17] M. Feder and E. Weinstein, Parameter estimation of superimposed signals using the EM algorithm, IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. 36, no. 4, pp. 477489,
1988.
[18] R. J. Fontana and S. J. Gunderson, Ultra-wideband precision asset location system, in Proceedings of IEEE Conference on Ultra Wideband Systems and Technologies, pp. 147150, Baltimore, Md,
USA, May 2002.
[19] S. J. Gunderson, R. Scholtz, K. Chugg, et al., Shipboard Environment Characterization, Appendix A, Technical Report: Naval Total Asset Visibility Tests on the USS Curtiss, Port Hueneme, CA,
25 December6 October 2000.
[20] M. Roberg, J. Sachs, P. Rauschenbach, et al., 11 GHz SiGe circuits for ultra wideband radar, in
Proceedings of Bipolar/BiCMOS Circuits and Technology Meeting, pp. 7073, Minneapolis, Minn,
USA, September 2000.
[21] J. Sachs, P. Peyerl, and M. Roberg, A new UWB-principle for sensor-array application, in Proceedings of 16th IEEE Instrumentation and Measurement Technology Conference (IMTC 99), vol. 3,
pp. 13901395, Venice, Italy, May 1999.
[22] A. Richter, M. Landmann, and R. S. Thoma, A Gradient Based Method for Maximum Likelihood Channel Parameter Estimation from Multidimensional Channel Sounding Measurement,
XXVIIth URSI General Assembly, Maastricht, NL, pp. CD, August 2002.
[23] R. Zetik, J. Sachs, and B. Schneegast, Non-destructive testing with imaging radar: first experience with a laboratory equipment, in Proceedings of Internationales Wissenschaftliches Kolloquium
(IWK 98), vol. 3, pp. 317322, Ilmenau, Germany, September 1998.
Domenico Porcino: Philips Research Laboratories, Cross Oak Lane, Redhill, Surrey RH1 5HA, UK
Email: [email protected]
Jurgen Sachs: Technical University Ilmenau, 98684 Ilmenau, Germany
Email: [email protected]
Rudolf Zetik: Technical University Ilmenau, 98684 Ilmenau, Germany
Email: [email protected]
Andy Ward: Ubisense, St. Andrews House, 90 St. Andrews Road, Cambridge CB4 1DL, UK
Email: [email protected]

Regulation and standardization

Christian Politano, Walter Hirt, Nils Rinaldi,
Gordana Drakul, Romeo Giuliano, and Franco Mazzenga

7.1. Introduction
Ultra-wideband (UWB) radio techniques have been extensively described in the
previous chapters, and we emphasize in this section the most fundamental parameter that characterizes UWB radio, which is the capability of UWB devices to
operate over a very large frequency range (several GHz bandwidth).
Since typical UWB radio may use spectrum already assigned to operative primary or secondary radio services, UWB radio devicesdespite their extremely
low transmission output powermay be a potential source of interference for incumbent radio services, and should provide the ability to coexist with legacy radio
services.
In this section, we provide the reader with a comprehensive description of the
worldwide UWB regulation and standardization framework under completion in
order to elaborate novel and eective means of spectrum management based on
coexistence mechanisms, instead of using conventional frequency-sharing mechanisms.
The first section, dedicated to regulation aspects, is introducing the UWB regulation framework elaborated under ITU-R to allow UWB devices, in particular
UWB communication and positioning applications, to operate without requirement for individual right to use radio spectrum (license-exempt) and on a no
protection, no harmful interference basis. Then a detailed overview on the European regulation framework is provided to explain how prudent but constructive
regulation rules are elaborated in this region.
The second section presents a UWB standardization overview in the USA and
in Europe. It provides the global overview of IEEE working groups mandated for
wireless personal area network (WPAN) including high-data-rate (IEEE802.15.3a)
and low-data-rate (IEEE802.15.4a) communication applications, and introducing
the most popular UWB technology candidates identified during the year 2004.
The European standardization process is also presented in this section with an
overview of the ETSI-TG31 task group mandated by the ECC for the definition

448

Regulation and standardization

of harmonized standards for short-range devices (SRDs) using UWB technology.

The last section describes a methodology applicable to compatibility studies
for evaluating UWB interference risks with incumbent radio services. It provides
some results of compatibility studies conducted for a few coexistence scenario between UWB and UMTS, fixed services. The particular interest in this section is
the description of the UWB characteristics impacting compatibility studies (activity factors, trac characteristics, radio access modes), and how interference risks
may be reduced by applying these features, allowing incumbent radio receivers to
coexist with UWB devices.
The results of such compatibility studies have been used for the elaboration
of interference mitigation techniques (proposed for standardization) in order to
specify coexistence mechanisms to reduce UWB interferences with incumbent radio services.
7.2. Regulation
7.2.1. Introduction
The FCCs (Federal Communications Commission) introduction of rules in early
2002 in the USA to authorize ultra-wideband (UWB) radio devices to be deployed
and marketed as unlicensed devices has generated studies and considerations of
radio spectrum implications within many administrations around the world. As a
result, the ITU-R (International Telecommunication Union-Radiocommunication
Sector) created Task Group 1/8 (TG 1/8) to study UWB radio coexistence issues.
As a technology, UWB does not fit into a classical radio regulatory process,
which is based on the allocation of dedicated spectrum bands for the dierent radiocommunication services. Since UWB radio signals spectrally overlap with frequency bands occupied by other services, the incumbent spectrum users are concerned about possible interference eects from single UWB devices. In addition,
concerns exist about the cumulative eect of UWB signals (aggregated power) by
raising the background noise for the dierent radiocommunication services.
The regulatory process driven by the work program within ITU-R TG 1/8, initially intended to be completed by end of 2004, has been extended to 2005 with the
objective to complete an ITU recommendation for international UWB regulation
before the end of 2005.
Figure 7.1 provides an overview of the selected frequency bands for shortrange devices using UWB radio technologies. While the frequency band below
1 GHz will be only allocated to UWB devices deployed within the USA, the frequency band between 3.1 to 10.6 GHz is intended to become the common band
allocated to UWB radio devices on a worldwide basis. Thus, the band between
3.1 to 10.6 GHz is of most interest for short-range communication devices. In
the medium term, the main emerging application will be wireless personal area
networks (WPANs) oering high data rates of up to 480 Mbps, for which a
standardization process is already well engaged (IEEE802.15.3a). UWB radio is

Christian Politano et al.

449

Emitted signal power

WPAN 802.15.4

UWB spectrum
(USA only)
0.868
0.9

WLAN 802.11a

Bluetooth, Zigbee
WLAN 802.11 b
WPAN 802.15.4
Cordless phones
Microwave ovens

FCC Part 15 limits

UWB spectrum
(USA and international)
2.4

3.1
Frequency (GHz)

10.6

Figure 7.1. Proposed frequency bands for UWB radiocommunication devices.

also providing advantages for future WPAN enabling low-data-rate applications

featuring location-tracking capability; for this type of UWB devices, technical
proposals have been submitted and a selection process has started within the
IEEE803.15.4a Standardization Group. Furthermore, there exist research programs
considering broadband evolutions towards data rates of 1 Gbps and beyond.
In the light of the large application potential of UWB radio technology and
the broad interest from the wireless industry worldwide, intensive eorts towards
a globally compatible UWB regulation have started. In particular, regulatory work
has started to investigate all potentially aected radiocommunication services. Under this harmonization process, the initial studies by the FCC are being extended
within the ITU, the CEPT (European Conference of Postal and Telecommunications Administrations) in Europe, and by the Japanese as well as other Asian regulatory bodies.
In the following, a specific focus is provided on the regulatory status within
the ITU-R and the European regulation work within CEPT, where a specific UWB
Task Group 3 (TG3) has been created under a mandate issued by the European
Commission. This mandate orders the CEPT to undertake all necessary work to
enable identification of the most appropriate technical and operational criteria for
the harmonized introduction of UWB radio applications in the European Union.
7.2.2. Worldwide statusInternational Telecommunication Union
In July 2002, the Study Group 1 (SG1) of the Radiocommunication Sector of the
ITU (ITU-R) decided to form a Task Group 8, named ITU-R TG 1/8, to carry out
relevant studies to answer two questions (ITU-R 226/1 and ITU-R 227/1) adopted
by SG1 on ultra-wideband technology.1 These questions concern the proposed
introduction of UWB devices and the implications of compatibility with radiocommunication services:
1International Telecommunication Union (ITU), Report of the second meeting of Task Group

1/8, Geneva, 2731 October, 2003.

450

Regulation and standardization

(i) Question 226/1: spectrum management framework related to the introduction of ultra-wideband (UWB) devices;
(ii) Question 227/1: compatibility between ultra-wideband (UWB) devices
and radiocommunication services.

The work load within ITU-R TG 1/8 was divided among four working groups:
(i) WG-1: UWB characteristics (Questions 226/1 and 227/1);
(ii) WG-2: UWB compatibility (Question 227/1);
(iii) WG-3: spectrum management framework (Question 226/1);
(iv) WG-4: measurement techniques (Decision 1/95).
The deliverables expected from these working groups have been defined as follows
[1]:
(i) ITU-R recommendation on the characteristics of UWB;
(ii) ITU-R recommendation(s) addressing compatibility between UWB and
radiocommunication services;
(iii) ITU-R recommendation providing guidance to administrations on a
spectrum management framework for UWB;
(iv) ITU-R recommendation on measurement techniques for UWB.
For each ITU-R TG 1/8 meeting, the Chairman issues a report (e.g., see [1]) with
a number of annexes produced by the working groups listed above. Thus far, these
documents must be considered continued working documents of a technical nature from whichat this timeno final conclusions can (or should) be drawn.
However, from the material available thus far, it is evident that particularly the
UWB compatibility issues dealt with in WG-2 pose key working and discussion
areas. For example, similar to within CEPT/ECC-TG3, a major discussion point
in TG 1/8 is to what extent coexistence-related results based on relevant field measurements should be admitted into the technical documents to allow verification
and calibration of theoretical results obtained from corresponding (worst-case)
modelling studies. As a result of these ongoing debates and other types of work,
it has been recognized that the four originally planned meetings of ITU-R TG 1/8
until end of 2004 will not suce to complete the work as anticipated. At its third
meeting in June 2004 in Boston, USA, the leadership of TG 1/8 already indicated
that a total of at least up to seven TG 1/8 meetings will be necessary to produce
the final deliverables. Thus, final globally applicable recommendations for UWB
radio as worked out by ITU-R TG 1/8 should not be expected before the end of
2005.
7.2.3. European regulatory process
The precedence set in the USA by the FCC in early 2002 has stimulated some considerable activities and to some extent competitive pressure in Europe to establish
a similar ruling for the deployment and use of devices based on UWB radio technology (UWB-RT). The European regulatory process started in early 2000 inspired
by the FCCs proposal for regulating UWB devices as adopted on May 10, 2000. A
number of interest groups and industrial parties have been active in the last few

Christian Politano et al.

451

years to assess and verify potential interference issues between UWB devices and
existing radio services. In this context, two main organizations are particularly relevant in Europe:
(i) CEPT (European Conference of Postal and Telecommunications Administrations) focused on the harmonization of telecommunications
regulations across its member countries;
(ii) ETSI (European Telecommunications Standards Institute) created in
1988 by the CEPT as a new organization to which its telecommunication standardization activities were transferred.
However, as of this writing (late 2004), given the innovative and somewhat disruptive nature of UWB-RT, significant work in the areas of UWB device characterization, measurement techniques, compatibility, as well as frequency management and regulation is still pending. Whereas some experts still question the
interference potential of a single UWB signal source in a typical home or oce
environment, many others express great concerns should UWB devices proliferate and become truly ubiquitous. It is feared that the simultaneous operation
and aggregation of a large number of UWB transmitters within a confined area
may pose a significant risk of harmful interference to incumbent radiocommunication services, such as for navigation, rescue and communications, fixed wireless access services, and even radio astronomy. The industrial UWB community
spends great eorts to address these legitimate concerns by proposing and designing systems that prevent interference under all practical and reasonable conditions. However, only a large research eort and scrutinizing analysis and, in particular, measurements will be able to help establish the most appropriate set of
rules for this very low-power and yet disruptive spectral underlay technology. In
Europe, eorts are under way to assess and define the required protection measures for all potentially aected services, for example, fixed wireless access (FWA)
within the bands 3.5 GHz, 3.64.2 GHz, 4.45.0 GHz; cellular systems in the bands
915 MHz, 18002100 MHz; space scientific services with bands 406406.1 MHz,
14001427 MHz, 15441545 MHz, 52505460 MHz. In particular, relevant field
measurements that will help assess various pending issues on coexistence between
UWB radio devices and existing active and passive radio systems or services are
pending. Such measurement campaigns will allow to assess the validity of results
already obtained from corresponding theoretical (worst-case) modelling studies
involving single and multiple UWB radio devices and incumbent radio services.
The inclusion of results based on actual measurements into the regulatory decision process may help define a globally compatible European regulatory framework for UWB radio, which accommodates both the incumbent radio services and
the deployment of technically and economically viable UWB radio applications.
The UWB radio regulatory process in Europe is mainly driven by Task Group 3
(TG3) of the Electronic Communications Committee (ECC) within the European
Conference of Postal and Telecommunications Administrations (CEPT) and Task
Groups 31A and 31B (TG31A/B) of the European Telecommunications Standards
Institute (ETSI). Historically, CEPT was founded on 26 June, 1959, in Montreux,

452

Regulation and standardization

Switzerland, uniting 19 administrations from Western Europe; today, CEPT encompasses 46 European countries and one of the key achievements of CEPT was
the establishment of ETSI in 1988.2 CEPT and ETSI are both working on UWB
radio matters under respective mandates issued to them by the European Commission [2, 3].
In particular, the recent mandate to CEPT is an order to identify the conditions relating to the harmonized introduction in the European Union (EU) of
radio applications based on UWB-RT, that is, . . . pursuant to Art. 4 of the Radio
Spectrum Decision, CEPT is mandated to undertake all necessary work to identify the
most appropriate technical and operational criteria for the harmonized introduction
of UWB-based applications in the European Union [2]. Among other reasons, the
EC provides the following justification for this mandate: UWB technology may
provide a host of applications of benefit for various EU policies. However, its characteristic broad underlay over spectrum already used by other radio services may also
have an impact on the proper operation of radio services of significance for the successful implementation of EU policies. It is therefore important to establish conditions of the use of radio spectrum for UWB which will allow UWB to be introduced
on the market as commercial opportunities arise, while providing adequate protection to other radio services. As explained below, in response to the EC mandate,
the CEPT has reorganized their ongoing UWB work in a new task group (ECCTG3).
The earlier EC mandate forwarded to ETSI calls for harmonized standards covering ultra-wideband (UWB) applications and derives from the R&TTE Directive
which defines the essential requirements equipment must meet to be placed on
the market and to be put into service for its intended purpose [3]. The mandate
points out that . . . as a technology, UWB radio does not fit in the classical radio regulatory paradigm, which bases itself on a subdivision of the spectrum in bands and
allocated for specific usage(s). A stipulation of the mandate is to align the work
with equivalent activities within the ITU and ISO/IEC while due account should
be taken of (draft) regulations adopted in other economies, so as to ensure a global
market for UWB devices.
At the seventh CEPT/ECC meeting held in Vilnius, Lithuania, on 1519
March, 2004, a decision based on the European Commissions UWB mandate
issued to CEPT resulted in the final approval of engaging in the work required for
the harmonization of frequency bands for UWB applications. The plenary session decided consequently to establish a specific task group to report to the ECC,
called Task Group 3 (ECC-TG3), which will take charge of the respective developments. The ECCs working methods were also modified accordingly to reflect
the principle that all ECC documents that are meant to respond to Commission
mandates need to be approved by part of the ECC plenary session.
2ETSI has a memorandum of understanding with CEPT, which allows ETSI to (a) provide direct

input to most of the drafting groups (project teams) of the ECC, and (b) give an indication of the
balance of views among ETSI members to the working groups; this function is carried out by the ETSI
liaison ocer.

Christian Politano et al.

453

7.2.3.1. Radio spectrum management in Europe

In Europe, radio frequencies are administered on a national basis, where national
administrations cooperate via CEPT and in particular the Electronic Communications Committee (ECC), which is administered by the European Communications
Oce (ECO). The ECC publishes decisions and recommendations to be followed
by national administrations in making their national frequency regulations; however, these recommendations and decisions of CEPT are not always adopted in
their entirety in every country. The European Commission is in the process of implementing its Radio Spectrum Decision, which has the objective of providing
legal certainty to spectrum allocations which support EU policies. Within ETSI see
also http://portal.etsi.org/erm/kta/Spectrum/spectrum.asp, TC ERM is the technical committee responsible for EMC and Radio Spectrum Matters (RM). Within
TC ERM, the working group RM is the centre of expertise on radio spectrum matters and is mandated to coordinate and support the needs and interests of the entire ETSI membership on radio spectrum matters. In particular, liaison to CEPT
ECC according the MOU between ECC and ETSI is required. TC ERM has the
primary responsibility for coordination of ETSI positions on the ecient use of
the radio spectrum and spectrum allocations. TC ERM has created working group
ERM RM to carry out the detailed preparation work on its behalf. CEPT/ECC has
three working groups concerned with radio spectrum management:
(i) Frequency Management (FM), including the Short-Range Device Maintenance Group (SRDMG);
(ii) Radio Regulatory (RR);
(iii) Spectrum Engineering (SE).
In addition, CEPT/ECC has project teams to cover specific issues. In general,
project teams are established to draft proposed decision and recommendation
documents. These are formally adopted by a corresponding working group (i.e.,
FM, RR, or SE), a process which includes a period of public consultation.3 Based
on the Radio Spectrum Policy Decision (Decision no. 676/2002/EC of the European Parliament and of the Council of 7 March, 2002, on a regulatory framework
for radio spectrum policy in the European Community), the European Commission seeks to
(i) relate spectrum demands to EU policy initiatives, supported by a Radio
Spectrum Policy Group, a Committee of Member States, chaired by the
European Commission, with ETSI and CEPT as permanent observers;
(ii) introduce legal certainty for technical harmonization measures carried
out by CEPT in areas of EU Community policy; the Commission, in
consultation with a Radio Spectrum Committee (RSC), will place
mandates on the CEPT; the resulting spectrum allocations will be codified in Commission Decisions, applicable throughout the European
Community;
3CEPT/ECC is currently reorganizing the structure of its working groups; for the latest informa-

tion, please see the web site of the European Radiocommunications Oce (http://www.ero.dk).

454

Regulation and standardization

(iii) increase transparency and information on the use of spectrum by requiring Member States to publish spectrum tables and other relevant
information in a common format accessible to all interested parties;
(iv) support the CEPT on the promotion of European interests in international negotiations.
In support of this activity, the Commission has established two advisory groups:
(i) The Radio Spectrum Policy Group (RSPG; Commission Decision 2002/
622/EC), to advise the Commission on radio spectrum policy issues, and
where the Commission should take spectrum harmonization measures
in support of other policies;
(ii) The Radio Spectrum Committee (RSC), which advises the European
Commission on mandates to be given to the CEPT in support of harmonization measures and the measures that should be taken to provide
legal certainty within the Community.
Some of the key services that have been identified for study by CEPT are as follows:
(i) aeronautical radionavigation service;
(ii) amateur service;
(iii) amateur satellite service;
(iv) earth exploration satellite service (active);
(v) fixed service;
(vi) fixed satellite service;
(vii) land mobile service (including 5 GHz radio LAN);
(viii) mobile satellite service;
(ix) radio astronomy;
(x) radio determination service;
(xi) radio navigation service.
7.2.3.2. ECC-TG3
The first meeting of ECC-TG3, whose main function is to prepare all aspects of
CEPTs response to the European Commissions mandate requesting the harmonization of spectrum used by ultra-wideband (UWB) systems in Europe, took
place on 19 April, 2004, in Maisons-Alfort, France [4]. This first meeting of ECCTG3 aimed at setting the organization of the Task Group as well as the general
objectives for future meetings in the light of its mandate. While limited to CEPT
administrations, the meeting was also attended by representatives from the European Commission, the European Radiocommunications Oce (ERO), the Chairman of ETSI ERM4 (Electromagnetic Compatibility and Radio Spectrum Matters)
4ETSI ERM is a horizontal technical committee responsible for the standardization of Electro-

magnetic Compatibility (EMC) and Radio Spectrum Matters on behalf of all other technical bodies of
ETSI. The Radio Matters (RM) working group is responsible for cooperation with the European Radiocommunications Committee (ERC) to secure appropriate spectrum allocations in the CEPT countries
for standardized systems, in order to ensure coexistence between dierent communications systems
standardized by ETSI. ERM has a number of task groups established on a short-term basis to deal
with particular issues; they disband on the resolution of the task. Concerning UWB radio matters,
the current ERM structure includes the following task groups: Standardization of Land-Mobile Radio, including Ultra-Wideband for Telecommunications and Radio locations Activities (TG31A), and
Automotive, including Ultra-Wideband Radar (TG31B).

Christian Politano et al.

455

Work Group TG31A, and a representative from the UWB Cluster. The latter is an
industry-led interest group that includes representatives of UWB radio-related European RTD (Research and Technology Development) projects supported by the
European Commission; the leading European RTD project within the UWB Cluster is the Integrated Project PULSERS (Pervasive Ultra-wideband Low Spectral Energy Radio Systems). Members of PULSERS have been instrumental in supporting
the Commission and its Radio Spectrum Committee5 (RSC) during preparation of
the UWB mandate issued to CEPT. As a result, subsequent ECC-TG3 meetings
are open for participation to recognized CEPT observers, representatives of the
European industry, and telecommunication operators. It was also concluded that,
as the main deliverable of ECC-TG3, a Draft ECC Report on UWB would be developed, essentially in the form of a technical document, together with whatever
else is deemed necessary. A further task is the coordination of European positions
and contributions within the Task Group 1/8 (TG 1/8) dealing with UWB matters in the Radiocommunication Sector of the International Telecommunications
Union (ITU-R).
The first ECC-TG3 meeting considered the current state of European activities
on UWB. It was decided to make use of the documentation already developed by
the Project Teams SE21 and SE24 of the CEPT/ECC Working Group for Spectrum
Engineering (WGSE), to obtain a description of the work already achieved and
to highlight areas seen as mature and to identify where further work is needed.
In particular, the Chairman of ETSI ERM TG31A oered its continued support
to the activity of ECC-TG3 on UWB, indicating that an ETSI system reference
document (SRDoc) on UWB imaging systems (wall probing radars and ground
probing radar systems) is to be approved by ETSI ERM RM at its September 2004
meeting while the SRDoc on UWB communication applications has been published and submitted to CEPT. Moreover, ECC-TG3 had received a liaison from
the Chairman of WGFM Project Team FM22 (a project team covering monitoring issues of CEPTs Frequency Management Working Group), indicating that a
subject on ultra-wideband measurements was added to the work program of
FM22 on spectrum monitoring; FM22 oered its willingness to assist ECC-TG3 in
conducting measurements related to UWB.
The mandate to CEPT as issued by the European Commission requests to report on actual or planned real-life testing within the European Union (EU) and to
consider the possible benefits of experimental rights to use radio spectrum for
UWB applications. In this and other relevant areas of the mandate, the Commission encourages CEPT to make use of ongoing activities and know-how of
EU RTD projects on UWB radio (see footnote 2 in [2]). At this first ECC-TG3
meeting, a representative from the ERO presented a report on the second meeting of ITU-R TG 1/8 (UWB) held in Geneva, Switzerland, 2731 October, 2003,
underlining in particular the diculties within its Work Group 3 (WG3) on defining a UWB spectrum management framework. Finally, the Chairman of SE24
5European Radio Spectrum Committee; http://europa.eu.int/information society/topics/radio

spectrum/all about/rsc/index en.htm.

456

Regulation and standardization

oered a statement as to the FCC regulations on UWB, initially adopted 14 February, 2002, indicating that SE24 had studied in detail the impact of the FCC limits on radiocommunication services and concluded that CEPT could not endorse
these limits.6
The general objectives of ECC-TG3 on UWB as mandated are
(i) to develop the draft ECC report on UWB;
(ii) to develop the draft ECC responses to the European Commission mandate to CEPT to harmonize radio spectrum use for ultra-wideband systems in the European Union;
(iii) to coordinate European positions in preparation for ITU-R TG1/8 on
ultra-wideband issues.
The deliverables of ECC-TG3 in response to the EC mandate, which may be generally referred to as the Draft ECC Responses to the EC Mandate on UWB, will
be submitted by ECC-TG3 for consideration at the ECC plenary meetings in July
2004 (Draft First Report), November 2004 (Draft Interim Report), and March
2005 (Draft Final Report). ECC-TG3 liaises with and consolidates outputs from
other working groups (e.g., WGFM and WGSE). ECC-TG3 will be open to ETSI
members in accordance with the terms of the CEPT/ETSI MOU and the group will
consult with relevant European organizations, in particular ETSI. As a result of the
technical work on UWB initially carried out within SE24 and transferred to ECCTG3, a Draft ECC Report will be submitted to WGSE for preliminary adoption
at its October 2004 meeting and WGSE will handle the consultation process, relying on TG3 in case of substantial comment to the draft ECC report, and finally
adopt the report at its first meeting in 2005. This Draft ECC Report will serve as
the technical basis for the Draft ECC Responses to the EC Mandate on UWB.
From the minutes reported in [4], it is interesting to observe that the ECs representative expressed his wish to follow the work of ECC-TG3 and help whenever
necessary with the interpretation on the objectives of the mandate. He also insisted
on the need for rapid progress due to growing industrial demands for regulating
this (UWB) technology and explained that the tight and challenging schedule of
the mandate was aimed at providing viable solutions in a timely fashion. Moreover, the ECs counselor also underlined that a large amount of funding is invested
by the European Union in various research projects on UWB,7 indicating that the
Commission strongly encourages CEPT to use the resources available in these RTD
activities to validate, where possible, UWB compatibility studies by a campaign of
experimental testing.
ECC-TG3 constituted itself into three working groups:
(i) WG1: UWB Characteristics and Measurement Techniques;
(ii) WG2: UWB Compatibility;
(iii) WG3: Frequency Management and Regulatory Issues.
6See, for example, the FCC mask for communication applications in [5].
7EC supported RTD projects related to UWB radio technology are, for example, whyless.com

(http:/www.whyless.org), UCAN (http://www.ucan.biz/), ULTRAWAVES (http://www.ultrawaves.org),

and PULSERS (http://www.pulsers.net).

Christian Politano et al.

457

ECC-TG3/WG1: UWB Characteristics and Measurement Techniques. The general terms of reference (TORs) of WG1 are to provide the following deliverables,
for approval by ECC-TG3:
(i) prepare chapter on technical characteristics of UWB devices for insertion in the Draft ECC Report on UWB;
(ii) prepare chapter on operational characteristics of UWB applications
for insertion in the Draft ECC Report on UWB;
(iii) prepare draft contributions intended for WG-1 of ITU-R TG1/8 on
UWB characteristics;
(iv) prepare draft contributions intended for WG-4 of ITU-R TG1/8 on
UWB measurement techniques.
In the course of this mandate, ECC-TG3/WG1 should
(i) consider operational characteristics of UWB applications, that is, communication equipment (Type 1) and radar imaging systems (Type 2);
(ii) consider technical characteristics of devices using UWB technology;
(iii) consider UWB measurement techniques;
(iv) identify the technical parameters of UWB systems to be included in the
overall harmonized regulatory approach;
(v) consider where design guidelines for existing and new radio standards
could improve coexistence between UWB devices and radiocommunication devices.
ECC-TG3/WG2: UWB Compatibility. The general TORs of WG2 are to provide
the following deliverables, for approval by ECC-TG3:
(i) prepare the Draft ECC Report on UWB;
(ii) prepare draft contributions intended for WG-2 of ITU-R TG1/8 on
UWB compatibility.
In the course of this mandate, ECC-TG3/WG2 should
(i) clarify relevant methodologies for compatibility analysis between devices using UWB technology and radiocommunication services (single
interferer and aggregate interference), including the definition of reference UWB deployment scenarios dependent on the type of UWB applications;
(ii) clarify relevant propagation models for use in compatibility studies;
(iii) report on actual or planned real-life testing within the European Union;
(iv) define protection requirements for radiocommunication services;
(v) identify the conditions of use of radio spectrum by UWB required to
protect other radio services from harmful interference, including the potential impact of UWB out-of-band emissions on other services.
WG2 is also expected to coordinate the liaison with WGFM Project Team FM22 as
well as the UWB Cluster and take into consideration results from experimentation campaigns.8
8The first meeting of ECC-TG3, a contribution was presented on the UWB Cluster (TG3#1 03R0) for further consideration by WG2 of TG3. The UWB Cluster, initiated by EU project members,
proposed to contribute as an industry association providing contributions relying on UWB theoretical

458

Regulation and standardization

ECC-TG3/WG3: Frequency Management and Regulatory Issues. The general TORs

of WG3 are to provide the following deliverables, for approval by ECC-TG3:
(i) prepare the draft ECC responses to the European Commission mandate;
(ii) prepare draft contributions intended for WG-3 of ITU-R TG1/8 on
UWB spectrum management framework.
In the course of this mandate, ECC-TG3/WG3 should
(i) consider the existing and developing regulatory environment, in particular ongoing ITU activities, and the extent of convergence which is feasible with non-EU regulation, including the application across the EU
of ITU RR Article 4.4, but also the implications of UWB emissions in
frequency bands covered by ITU RR footnote 5.340;
(ii) consider the applicability and impact of existing relevant definitions in
the radio regulations to UWB;
(iii) define where necessary relevant specific terminology for UWB (e.g.,
UWB emission, center frequency, necessary bandwidth, fractional bandwidth, unwanted out-of-band emissions, etc.);
(iv) consider the possible benefits of experimental rights to use radio spectrum (or licenses) for UWB applications;
(v) consider the designation of one or more harmonized frequency band(s)
for generic or specific UWB uses, and alternatively develop dierent
(technical) options for possible introduction of UWB devices;
(vi) consider what could be the possible elements of a monitoring and review mechanism aimed at ensuring that regulation of radio spectrum
for UWB remains responsive to technical and societal developments,
and to actual or perceived changes in the risk of harmful interference
with other radio services.
In the course of its work, WG3 is also expected to provide recommendations to
ECC-TG3 on the overall approach to UWB regulation, possibly consisting of
(i) short-range device annex in ERC/Recommendation 70-03 covering
generic UWB communications equipment;
(ii) ECC recommendation or ECC decision for imaging systems including
ground penetrating radar.
Nearly all possible UWB applications, in particular UWB communication and positioning applications, are expected to be operated without requirement for an
individual right to use radio spectrum (license-exempt) and on a no protection, no harmful interference basis. The regulatory regime for radar imaging systems is however likely to dier; regarding regulatory enforcement matters, ECCTG3/WG3 is expected to primarily focus on this type of UWB applications. The
schedule outlined in the Commissions mandate to CEPT calls for deliverables
according to the following schedule: First Report from CEPT by 15 July, 2004;
Interim Report by 15 November, 2004; Final Report by April, 2005. In addition,
CEPT is requested to report on the progress of its work pursuant to the mandate
studies and on UWB experimentation campaigns already defined and supported by the EU projects
work plans or by individual member contributions.

Christian Politano et al.

459

at all meetings of the Radio Spectrum Committee taking place during the course
of the mandate.
7.2.3.3. ETSI-TG31A/B
The task of TG31A is to investigate and develop generic and/or specific ETSI radio
standards for short-range devices (SRDs) using UWB technology see also http://
portal.etsi.org/erm/ERMtg31A ToR.asp. Considered UWB applications include,
but are not limited to, high-speed short-range wireless communications links,
security and movement/location detectors, electronic fences, proximity alarms,
medical sensors, intelligent automotive systems, and ground penetrating radars.
TG31A will also identify spectrum requirements to ensure spectrum eciency
and compatibility with other radio services. Inputs will be prepared for TC-ERM
Working Group ERM-EMC on EMC aspects relevant to devices covered by TG31A.
The TG31A Group issued a technical report (TR 101 994 [6]) that covers the intended applications, proposed technical parameters, and radio spectrum requirements for UWB. The report includes necessary information to support the cooperation under the MOU between ETSI and the European Radiocommunications Committee (ERC) of CEPT for amending the ERC Recommendation 7003. Another report (TR 101 994-2) covers the technical characteristics of shortrange ground, wall, and through-wall probing radars using UWB technology. The
progress made within ETSI-TG31A in the last three years includes the discussion of
a so-called sloped power spectral density (PSD) emission mask for UWB communication devices [7]. However, note that the emission limits indicated by this
mask are not an ocial proposal nor are they a recommendation. Instead, they are
merely under discussion as a possible means to obtain additional out-of-band protection for existing radiocommunication services in Europe, when compared with
the corresponding FCC. The task of TG31B involves the area of UWB automotive
radar, to develop standards, system reference documents, or other technical documents for short-range devices as covered by the ECC SRD Recommendation 70-03
and the relevant ECC decisions in the field of automotive UWB radar applications.
TG31B also investigates spectrum parameters for use in automotive radar systems
in order to ensure the ecient use of the radio spectrum, considering global harmonization9 see also http://portal.etsi.org/erm/ERMtg31B ToR.asp.
7.3. Standardization
7.3.1. Introduction
In todays information society, new technologies aect our life significantly. Technology makes things quicker, easier, better, but also more complex. Standardization can be seen as an essential requirement for an open exchange of information
9Special state-of-the-art measurement equipment may be capable to measure power levels of
perhaps as low as 90 dBm in a bandwidth of 1 MHz.

460

Regulation and standardization

enabling interoperability between dierent services/technologies. Standards need

to be produced at a speed that is consistent with market demand; they have to be
considered by all involved parties, otherwise they may not be widely accepted.
Since standardization and innovation go hand in hand, standards in certain
way allow those innovative ideas to reach the market. And one innovative idea
that has emerged in the latest years is the use of ultra-wide band (UWB) for civil
applications. UWB targets a broad spectrum of radiocommunications and localization applications, and is foreseen to be integrated in low-power low-cost devices.
In the following sections, UWB standardization eorts across the world are
presented. Next section will discuss the standardization activities carried out by the
IEEE in the US. In particular, two task groups are of interest: IEEE802.15 TG3a and
IEEE802.15 TG4a. Section 7.3.3 discusses UWB standardization progress made
in Europe by the ETSI. Standardization eorts in Japan (ARIB) are presented
in Section 7.3.4, while international standardization activities are presented in
Section 7.3.5 (GRSC).
It is to be mentioned that the presented standardization activities are ongoing, and that the status of these may have become obsolete. The current chapter
presents the standardization status as of spring 2004.
7.3.2. IEEE
The Institute of Electrical and Electronics Engineers (IEEE) Standards Association
is a leading developer (based in the US) of global industrial standards in various industries. The IEEE created in May 1999 the IEEE802.15 Wireless Personal
Area Network (WPAN) Working Group (formerly the WPAN Study Group (SG),
formed in March 1998) [8]. The IEEE802.15 WG aims to provide, in the IEEE802
family, standards for low-complexity and low-power consumption wireless connectivity. As of today, there are several task groups (TGs) and study groups (SGs)
identified within IEEE802.15:
(i) TG1: 1 Mbps WPAN/Bluetooth,
(ii) TG1a: WPAN 15.1 Revisions to Bluetooth v1.2,
(iii) TG2: Recommended Practice for Coexistence in Unlicensed Bands,
(iv) TG3: 20+ Mbps High Rate WPAN for Multimedia and Digital Imaging,
(v) TG3a: 110+ Mbps Higher Rate Alternative PHY for 802.15.3,
(vi) TG3b: WPAN 15.3 Maintenance,
(vii) SG3c: mmWave,
(viii) TG4: 200 kbps for Interactive Toys, Sensor, and Automation Needs,
(ix) TG4a: WPAN Low Rate Alt PHY,
(x) TG4b: WPAN 15.4 Revisions and Enhancements,
(xi) TG5: WPAN Mesh Networking.
Two of the standards projects are of interest for the reader, the IEEE802.15.3a TG3a
and the IEEE802.15 TG4a, as they are both susceptible of adopting UWB as radio
technology. These two TGs activities will be presented in the next sub-sections.

Christian Politano et al.

461

7.3.2.1. IEEE802.15 TG3a

The IEEE802.15 High Rate Alternative PHY Task Group (TG3a) for Wireless Personal Area Networks (WPANs) [9] became a TG in December 2002, when its
Project Authorization Request (PAR) [10] was approved. It aims to provide a higher-speed PHY enhancement amendment to 802.15.3 for applications involving
imaging and multimedia. Hence, the medium access control (MAC) as defined
within 802.15.3 will be kept for 802.15.3a systems; only the PHY will change. The
802.15.3 MAC is based on a centralized network topology, that is, organized in
piconet fashion, with one piconet coordinator and devices synchronized to it; the
common air interface is shared by implementing time-division multiplex access
(TDMA).
In December 2002, the TG3a issued a Call for Intent/Call for Proposals (CFI/
CFP) [11], in order to open the discussion about the 802.15.3a standard proposers.
The TG3a also issued (between December 2002 and January 2003) three important documents fixing the rules for the standardization procedure: the technical
requirements [12], the PHY selection criteria [13], and the down selection voting
procedure [14].
The PHY selection criteria document defines among other things the throughput performances to be achieved (110 Mbps at 10 meters, 200 Mbps at 4 meters,
and 480 Mbps at undefined range), performance requirements with simultaneously operating piconets (SOPs); signal acquisition performance requirements;
link budget and sensitivity calculation; power management modes, power consumption requirements; antenna practicality.
A common channel model to be used for system simulations from all proposers has been established by the TG3a [15]. It is a channel model based on the
Saleh-Valenzuela channel model [16], with slight changes.
The first TG3a meeting where the first proposals were presented (25 proposers) was held in March 2003. The vast majority of the proposals were based
on UWB radio technology. These UWB proposals could be divided into two technical approaches: the mono- or dual-band approach and the multiband approach.
The down selection voting procedure during the March 2003 meeting eliminated several proposals. Alliances between proposers were also formed, and as
a result the number of proposals presented during the next meeting (held in May
2003) was reduced from 25 to 13. After further down selection, the number of proposals squeezed to 7 in July 2003, to finally end with 2 main proposals in September 2003. These two proposals are the multiband OFDM (orthogonal frequencydivision multiplex) proposal and the DS-CDMA (direct-sequence code-division
multiple access) proposal. Since September 2003, the standardization process has
been stalled, as none of the two proposals have reached the 75% of the votes required for a proposal to become a standard. These two proposals are presented in
the following sections.
Multiband OFDM physical layer proposal. The Multiband OFDM Alliance (MBOA
[17]) is a group of companies that has gained more and more momentum during
the down selection process. Its PHY proposal is currently the majority-holding

462

Regulation and standardization

Group A

Group B

Group C

Group D

Band Band Band Band

#1
#2
#3
#4

Band Band Band Band Band Band Band Band Band

#5
#6
#7
#8 #9 #10 #11 #12 #13

3432 3960 4488 5016

MHz MHz MHz MHz

5808 6336 6864 7392 7920 8448 8976 9504 10032

MHz MHz MHz MHz MHz MHz MHz MHz MHz

Figure 7.2. Spectrum division into groups, as proposed by MBOA.

Input
data

Scrambler

Convolutional
encoder

Puncturer

Bit
interleaver

Constellation
mapping

IFFT
Insert pilots
Add prefix/GI

DAC
exp( j2 fc t)

Time-frequency code

Figure 7.3. Block diagram of an example MBOA TX architecture.

proposal at the IEEE802.15.3a meetings, reaching roughly 2/3 of the votes (which
is not enough as 75% are required).
The basic principle of the MBOA proposal is to divide the spectrum into several subbands of 528 MHz, on which the OFDM symbols are transmitted. The
current proposal band plan divides the bands into several groups, as illustrated in
Figure 7.2.
Subbands of Group A are recommended for first generation devices; it includes 3 bands in the bandwidth 3.1 to 4.9 GHz. Subbands of Group B (4.9
6.0 GHz) are reserved for future use; subbands of Group C (6.08.1 GHz) are reserved for devices with improved SOP performance; subbands of Group D (8.1
10.6 MHz) are also reserved for future use.
Information is transmitted using OFDM symbols on each subband. OFDM
carriers are being eciently generated using a 128-point IFFT/FFT. By limiting
the constellation size of the modulation to QPSK (quaternary phase-shift keying),
the internal precision requirement is reduced.
The OFDM symbols consist of 128 QPSK modulated tones: 100 data tones,
used to transmit information; 12 pilot tones, used for carrier and phase tracking;
10 user-defined tones; 4 NULL tones. The modulation is performed through signal
mapping to tones and then IFFT operation. A 60.6-nanosecond prefix is added
for robustness against multipath. A 9.5-nanosecond guard interval is inserted in
order to provide sucient time for switching between bands. The symbols are
then mixed up to the right subband, according to the time-frequency code. In this
way, information is coded across all bands in order to exploit frequency diversity,
and to provide robustness against multipath and interference. Figure 7.3 depicts
an example of a MBOA transmitter (TX).
The raw bit rate (bit rate over the air) of the MBOA system is 640 Mbps. Channel coding and frequency spreading techniques will reduce the information bit rate
(bit rate at MAC-PHY interface), depending on channel conditions; this will yield
information bit rates between 55 Mbps and 480 Mbps.

Christian Politano et al.

463
High band
UNII-ISM

Low band

5
6
7
8
9
Frequency (GHz)

10

11

Figure 7.4. The two subbands as defined in the DS-CDMA proposal.

The MBOA claims that this OFDM-based solution is spectrally ecient, that
it has inherent robustness against narrowband interferers, and that it has excellent
robustness in multipath environment. Furthermore, by dividing the UWB band
into several subbands, sensitive subbands and tones can be turned o (in order to
comply with changing regulations, or for enhanced coexistence with other devices)
when needed. Hence, ability of this system to comply with worldwide regulations is
claimed by the MBOA; coexistence with current and future systems is also assured
this way. Finally, the MBOA proposal claims to oer multimode devices, hence
scalability.
DS-CDMA physical layer proposal. The DS-CDMA proposal was presented initially by Xtreme Spectrum Inc. [18], a startup company that was later bought by
Motorola [19]. Along the way, the proposal has gained support by other companies as well. It is the second standing proposal in the IEEE, collecting the remaining
third of the votes.
The DS-CDMA proposal envisions a dual-band approach: 3 spectral modes
of operations are proposed. One is the low-band mode, from 3.1 to 5.15 GHz;
it provides information bit rates between 28.5 and 400 Mbps, and is designed to
support lower-rate, longer-range devices. The second is the high-band mode spanning from 5.825 to 10.6 GHz; it accommodates information bit rates from 57 to
800 Mbps, and targets high-rate, short-range devices. Finally, the last mode is the
multiband mode, which combines the bandwidth of the previous two modes. It
can achieve up to 1.2 Gbps in information bit rate, and supports low-rate, longerrange, high-rate, short-range devices.
The proposed spectrum division avoids carefully the UNII and ISM bands
(unlicensed national information infrastructure and industrial, scientific, and
medical radio bands, spanning roughly from 5.25 to 5.85 GHz), as illustrated in
Figure 7.4. This spectrum division proposal is adjustable, in case future spectral
allocation requires it. The DS-CDMA proposal uses two types of modulations:
binary phase shift keying (BPSK, mandatory) and quaternary phase shift keying
(QPSK, optional). For spreading, CDMA is used via low cross-correlation MBOK
(M-ary biorthogonal keying) ternary code sets (1, 0, +1). There are 4 CDMA
code sets per frequency band. CDMA also provides a selection of logical channels
(4 logical channels per frequency band, that is, up to 8 piconets accommodated

464

Regulation and standardization

when both subbands are used). The chip rate is 1.368 GHz for the low-band mode,
and 2.736 GHz for the high-band mode. For both modes, the CDMA code length
is 24 chips/symbol.
The DS-CDMA proposal is claimed to handle SOP conditions well, by using
frequency-division multiplexing (FDM, via 2 defined subbands), code-division
multiplexing (CDM, for spreading), and time-division multiplexing (TDM, via
802.15.3 MAC implemented as TDMA protocol). It is also claimed to be capable
to be robust against narrowband interference via its CDMA codes, and via tunable
notch filters.
7.3.2.2. IEEE802.15 TG4a
The IEEE802.15 Low Rate Alternative PHY Task Group (TG4a) for Wireless Personal Area Networks (WPANs) became an ocial Task Group in March 2004 [20];
with its committee work tracing back to November 2002, when an interest group
was formed. TG4a has defined a project for an amendment to 802.15.4 for an alternative PHY. While the 802.15.3 MAC is targeting centralized network topology
with TDMA, the 802.15.4 MAC uses a distributed architecture, where clear channel assessment (CCA) is preferred for sharing the radio medium.
The main goals of WP4a are to provide communications and high precision
ranging/location capability (1 meter accuracy or better), high aggregate throughput, and ultra-low power; as well as adding scalability to data rates, longer range,
and lower power consumption and cost. These additional capabilities over the existing 802.15.4 standard are expected to enable significant new applications and
market opportunities.
TG4a meets every other month, during the 802.15 plenary or interim meetings. TG4a is very young; the Call for Application (CFA) has already been issued,
but the TG4a Call for Proposals (CFP) has not been issued yet (expected to be issued in May 2004). Other activities such as the definition of selection criteria and
of the technical requirements have been initialised.
UWB is one strong candidate for the alternative PHY to be selected within
802.15.4a, due to its inherent low-power low-complexity characteristics, along
with precise localization capabilities. In fact, UWB channel model activities have
already been started. But as of today no proposal has been presented yet, as the
CFP is still to be issued in its final version.
7.3.2.3. Other IEEE groups related to UWB
Other groups interested in UWB technology are the IEEE802.15 TG5 (WPAN
Mesh Networking), the IEEE802.15 SG3c (mmWave), and the IEEE802.19 (Coexistence Technical Advisory Group).
The IEEE802.15 TG5 is a Task Group whose charter is to define recommended
practices for implementing mesh networking of WPANs [21]. Mesh topologies can
be used to increase the reliability and range of network, with focus on special needs
of WPAN application such as very low energy consumption and very low cost. The
TG5 may in this respect be of interest for UWB, as UWBs PHY is well adapted to
mesh networking.

Christian Politano et al.

465

The IEEE802.15 SG3c mmWave Study Group was formed to explore the use
of the 60 GHz band for WPANs. This little-used band (as defined in the Federal
Communications Commission (FCC) Article 47 CFr 15.255) provides 5 GHz of
bandwidth and avoids interference with nearly all electronic devices, given the
high attenuation of these wavelengths by walls and floors, hence allowing more
WPANs to occupy the same building. The intention of the SG3c is to consider
mobile and networked applications, network protocols and performance, mobile
and ad hoc networking, mobile and ubiquitous systems, network security, internet computing, wireless networks, multimedia distribution, last-mile applications.
Again, UWB may become the PHY of choice for this Study Group, as the very large
available bandwidth would enable UWB operation.
The IEEE802.19 Coexistence Technical Advisory Group (TAG) develops and
maintains policies defining the responsibilities of 802 standards developers to address issues of coexistence with existing standards and other standards under development [22]. Due to the very large bandwidth of UWB, several coexistence scenarios for UWB devices have been submitted to the IEEE802.19 TAG. The following scenarios have been envisaged in order to study single interferer coexistence
issues
(i) Home usage:
(1) a UWB digital video camera streams video to a multimedia hub
in the home that also forwards the stream to an HDTV monitor,
while the same multimedia hub is streaming MP3 audio to a remote audio terminal;
(2) the multimedia hub in a home is located 1 meter away from a cellphone or WLAN base station; the user may be talking on the phone
or the hub is supporting web browsing or email while uploading
video or still images.
(ii) Enterprise usage:
(1) a UWB video projector is used in a conference room to allow attendees to stream their presentations from their laptops, but all attendees want to use the corporate network simultaneously using
802.11b/g/a/n.
(iii) Kiosk/hot-spot usage:
(1) a laptop computer equipped with 802.11b/g/a/n uploads photos
to a kiosk while a digital still camera (DSC)/digital video camera (DVC) with a UWB device also uploads photos to the same
kiosk;
(2) in an airport hot spot, several users upload their DSC/DVC to their
laptop computers, then forward these images via WLAN to a central server for storage.
(iv) Mobile usage:
(1) mobile phone incorporating UWB technology and GPS;
(2) mobile phone with either DSC or DVC capability that uploads
video or still images to kiosk or A/V hub while maintaining 3G
connection.

466

Regulation and standardization

7.3.3. ETSI
The European Telecommunication Standards Institute (ETSI) is responsible for
development of standards for radiocommunications systems and equipment [23].
Radio standards contain by their nature several requirements which relate to ecient use of radio spectrum, including compatibility between dierent radio services.
In order to ensure the planning of an ecient use of radio frequency spectrum, a memorandum of understanding (MOU) has been elaborated between

ecommunications

ETSI and the Conference

Europenne
des Postes et Tel
(CEPT)
Electronic Communications Committee (ECC), for cooperation in development
of radio technical specifications. In the development of harmonized standards
containing technical specifications for radio equipment, the provisions of the
ETSI-CEPT MOU will be applied, to the extent that the MOU applies to the particular equipment. Compliance within essential requirements (Radio & Telecommunications Terminal Equipment (R&TTE) Directive) will ensure free circulation
of radio products within the European Environment Agency (EEA).
In general, assumptions are necessary for spectrum management purposes
concerning the performance of transmitters, receivers and control functions in
the areas of signalling, code-domain consideration and frequency sharing, and so
forth.
In this sense, the European Commission (through the ECC Telecommunication Conformity Assessment and Market Surveillance Committee (TCAM)) has
given a mandate (M/329) to the European Standardization Organization on March
19th, 2003, with the goal to establish a set of harmonized standards covering
UWB applications under the R&TTE Directive. In response to that, the ETSI Electromagnetic compatibility and Radio Spectrum Matters Task Group 31A (ERM
TG31A) [24] started developing a set of harmonized standards for short-range
devices (SRDs) using UWB technology.
The ERM TG31B, on the other side, has the task to develop standards, system reference document or other technical documents for short-range devices as
covered by the ECC SRD Recommendation 70-03, and the relevant ECC decision
in field of automotive ultra-wideband radar applications in relation with ecient
use of radio spectrum, considering global harmonization.
ERM TG31A and B are described in more details in the following two sections.
7.3.3.1. ETSI ERM TG31A
In order to support cooperation between ETSI and ECC, the TG31A produced
a technical report as system reference document [25], which provides information on the intended applications, the technical parameters and radio spectrum
requirements for UWB communications equipment in the frequency range from
3.1 GHz to 10.6 GHz. In this document, it was indicated that there is a potential
for variety of UWB communication applications as well as an increasing number
of sensing and positioning applications.

Christian Politano et al.

467

In Annex A1 of the mentioned document, a variety of communication applications are described, like wirelessly distributed services such as phone, cable, and
computer networking throughout a building or home. UWB has potential in both
home and business markets, based on its low cost as well as high-speed data transmission capability (100 Mbps500+ Mbps), and could be well suited to for various
video distribution applications (DVD player/recorder to TV or transmitting images/video from digital camera/camcorder to TV) and, generally, new high-speed
consumer SRDs.
Particularly, UWB technology is a potential cable replacement for applications
where USB 2.0/Firewire resides, for example,
(i) USB 2.0 allows data rates up to 480 Mbps;
(ii) IEEE1394 [18] (Firewire 400 Mbps and 800 Mbps);
(iii) USB 1.1 at 12 Mbps.
Other identified UWB applications are low data rate with very low activity factors,
and are typically used in sensor networks and RFID (radio frequency identification) systems. For example, the FCC recently approved a tagging system for UWB
precision asset location systems.
Since technology is still advancing, it is very unlikely that only one single
type of signaling and modulation will be used for this large variety of applications. The UWB communication devices based on single impulse radio technology
(as shown in Figure 7.5) and on multiband OFDM pseudocarrier approach (illustrated in Figure 7.6) have been given as two examples of UWB implementations
[25]. Since in time of publishing the Technical Report (January 2004) consensus
was not reached by ETSI delegates for power limits within the band 3.1 GHz to
10.6 GHz, the values given in Annex B2 of the Technical Report are supported
by the UWB industry within ETSI as a starting point, representing rather general
emission (power spectral density) limits requested for UWB devices, with no relation to the particular modulation scheme.
TG31A is also in charge of providing a system reference document covering
ground probing radars and wall probing radars applications [26], in support to the
cooperation under the MOU between ETSI and ECC. This document provides information on the intended applications, technical parameters, and the radio spectrum requirements. The scope is limited to short-range devices, in which the system is in close proximity to the materials being investigated (e.g. earth materials,
highway pavement,). The system is designed to radiate a sequence of pulses into
the ground and capture a waveform corresponding to the received signal scattered
by objects in the ground. The pulse repetition frequency (PRF) of the transmitted pulse may vary widely dependent upon the application, with a range between
1kHz and 10 MHz operating in the band from 30 MHz to 12.4 GHz. It is proposed
that such kind of devices will be covered by a generic standard for unintentional
radiation with radiation limits up to 1 GHz specified in EN 55022 for Class A,
and new amended limits up to 6/18 GHz as proposed in CISPR/I/105/CDV and
CISPR/I/106/CDV.
Besides producing the Technical Reports mentioned above, TG31A is working
on drafting a harmonized standard: EN 302 065, which deals with UWB devices

Regulation and standardization

Normalized amplitude

468
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
0

0.5

1.5

2.5

Time (ns)
(a)
80

Power (dBm/Hz)

100
120
140
160
180
200

10

12

Frequency (GHz)
(b)
Figure 7.5. Pulse shape with corresponding spectrum.

for communication purposes operating between 3.1 GHz and 10.6 GHz. Part 1 of
this draft standard deals with technical characteristics, while Part 2 provides parameters deals with the R&TTE Directive and is currently concentrating on measurement methods (this work is done with the ERM Task Group 33, which is in
charge of measurement methods for discontinuous transmission).
7.3.3.2. ETSI ERM TG31B
TG31B has the task to develop standards, system reference document, or other
technical documents for short-range devices as covered by the ECC SRD Recommendation 70-03, and the relevant ECC decision in field of automotive ultrawideband radar applications in relation with ecient use of radio spectrum, considering global harmonization.

Christian Politano et al.

469

1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1

10

15

20

25

30

35

40

(a)
0

Magnitude (dB)

10
20
30
40
50
60

Frequency (GHz)
(b)
Figure 7.6. Amplitude and frequency spectrum characteristics for a multiband implementation.

Automotive short-range radar (SRR) systems have been identified by the EU

policy eSafety initiative as significant technology for the improvement of road
safety in Europe addressing overall transport policy goal of cutting road fatalities in Europe by half by 2010, detecting the location and movement of objects
near a vehicle, enabling features such as near collision avoidance, improved airbag
activation, and suspension systems that better respond to road conditions (see
Figure 7.7).
Work in this group has been concentrated on SRR equipment for road safety
and intelligent transport systems (automotive collision warning), temporarily operating in the 24 GHz band and permanently in the 77 GHz to 81 GHz band [27].
The industry has been working towards an early introduction of equipment due
to interest to reuse the same technology during the lifetime of the model lines
(usually 67 years in Europe), so it is expected that new car lines will already use

470

Regulation and standardization

Intersection

Blind spot

Collision
warning

Side-crash

Rear crash
Lane change
support Parking aid
Overtake
support

Park-aid
precrash

Low speed
back-up
driving

Rear collision
warning

ACC with
stop & go

ACC
77 GHz

Collision
avoidance

Side-crash
Blind spot

Pedestrain
protection

Intersection

Figure 7.7. Short-range radar functions for vehicle.

long-term frequency equipment. Possible deployment at 24 Ghz will be not only

time limited but also overall car park penetration is restricted to below 10% as an
outcome of CEPTs compatibility studies.
Another activity of TG31B is related to adaptive cruise control using automotive long-range radar (LRR) operating in the 76 GHz to 77 GHz band. Adaptive
cruise control today uses LRR operating with maximum bandwidth of 1 GHz. It
uses distance scanning, which requires an operating range of approximately 150 m
and is used at vehicle velocities above 30 km/h. One or multiple narrow lobes control or scan the driving path in front of the car to determine the distance to the
vehicle driving ahead for maintaining a constant minimum safety distance.
ETSI is also supporting activity foreseen by the International Telecommunications Union (ITU) and CEPT using unconventional channel arrangements much
larger than those foreseen so far, in frequencies up to 58 GHz. The goal is to carry
at least a fully transparent 1 Gbps signal, to support interfaces like Gigabit Ethernet or Fibre Channel, and to use only one radio equipment with expected goal of
covering at least 12 km in line-of-sight (LOS) scenario.
7.3.4. ARIB/MMAC
The Association of Radio Industries and Businesses (ARIB) is a Japanese standardization and radio spectrum regulation body [28]. Since 2002, standardization
activities on Ubiquitous homelinks: 1394 over UWB have been initiated in the
Multimedia Mobile Access Communication (MMAC) Wireless Home-Link AdHoc Group, combining spectrally ecient UWB impulse radio technology and
flexibility and mobility of a broadband wireless access solution in and around the
home.
7.3.5. Global radio standardization collaboration
The Global Radio Standardization Collaboration (GRSC) Group (initially RAST
[29]) was formed in November 1994 at the initiative of ETSI, for the purpose of

Christian Politano et al.

471

providing an informal multinational information exchange focused on radio standardization trends and developments in the delegates various regions. GRSC was
foreseen to facilitate assessing the potential for harmonization and to complement
the more formal processes of other bodies, and particularly the ITU, in the work of
developing international standards recommendations. It enables regular exchange
of information between work programs and others, in a number of agreed highinterest technical areas like UWB and software-defined radio, attracting a growing
number of participants from around the world.
7.4. Coexistence with radio systems
7.4.1. Introduction
To protect existing systems against UWB emissions, the FCC and successively ETSI
restricted the UWB operating bands for communications in the 3.110.6 GHz frequency range by regulating UWB power emission with frequency-power masks for
each specific UWB application/device [30].
Even considering FCC (or ETSI) restrictions, the evaluation of interference
caused by UWB devices on existing narrowband terminals is important to guarantee nonconflicting coexistence and to gain acceptance of UWB technology worldwide. In fact, as shown in [31], coexistence among UWB and fixed wireless systems
may be critical even when FCC (or ETSI) rules are fully respected. However, the
analysis in [31] was too simplistic. In particular, the calculation of UWB interference on fixed systems receivers was carried out considering UWB terminals as electromagnetic radiators transmitting at their maximum allowable power for 100%
of time; the UWB signal spectrum was assumed to be flat and, finally, no realistic
communication system architecture was considered for the UWB. Other results on
the coexistence between UWB and fixed wireless systems operating in the 35 GHz
band have been presented in the open literature as well as in regulatory forums
[3236]. Some of the assumptions in [31] have been relaxed in [33, 34, 36] and
dierent conclusions were reached. As an example, it could be noted that using
100% activity factors contrasts with the typical deployment scenarios envisaged
for UWB devices where it is estimated that in any case the aggregate percentage
of time that any single device transmits is to be considered not exceeding 10%
[33, 34, 36].
Coexistence analysis was further extended to other existing and licensed narrowband and wideband systems. In [37], UWB interference on GSM, UMTS, and
GPS was analyzed even though results were obtained by considering baseband
UWB signals and not including any realistic system architecture for the UWB.
In [32] UWB interference on UMTS base stations was assessed introducing the
equivalent cumulative path gain parameter. However, even in this case, only the
UWB single terminal case transmitting at fixed power level was considered. Finally, several coexistence studies have been reported in the current literature for
the IEEE802.11a,b and Bluetooth standards [3840].
In this section, coexistence between a UWB hot spot located inside a building and fixed wireless systems FWA and PP as well as UMTS and DCS 1800 is

472

Regulation and standardization

analyzed. In order to render results comparable with those presented in the current literature, the scenario in [31] is considered but UWB interference is evaluated
accounting for the most important system parameters such as the UWB network
architecture, power control, trac characteristics (e.g., the UWB device activity
factor), and the spectral characteristics of the UWB signals. The direct-sequence
(DS) and the time-hopping (TH) UWB signals are considered and their spectra
are used to evaluate the fraction of UWB power leaking in the band of the victim
receivers.
The procedure adopted for coexistence assessment can be divided in the following four steps:
(1) calculation of the maximum allowable UWB interference limits for each
of the considered victim receivers;
(2) definition of the UWB hot-spot interference scenario;
(3) analytical calculation of the average interference due to UWB devices on
the victim receivers accounting for several features of the UWB signals
and system;
(4) comparison of the average UWB interference with the limits obtained in
point 1.
The proposed procedure is quite general and provides a unique framework to
analyze coexistence of UWB with several classes of victim terminals.
The main aim is to demonstrate that by proper selection of few UWB system
features, victim receivers can coexist with UWB terminals without any dangerous
interference even in very critical operating conditions where the victim receiver is
interfered by several UWB terminals. The organization of this section closely follows the steps of the procedure used for coexistence assessment. In Section 7.4.2
the maximum allowable UWB interference for each one of the considered victim
receivers is evaluated. In Section 7.4.3 the interference scenario and the characteristics of the UWB system are detailed. In Section 7.4.4 the power spectral densities
of the DS and TH UWB signals are obtained and are used in Section 7.4.5 to calculate the fraction of UWB interference power leaking in the band of the victim
receiver. In the same paragraph, several results are provided for the two cases of
power and non-power controlled UWB devices. Finally, conclusions are drawn.
7.4.2. Calculation of UWB interference limits
Considering one generic victim receiver, the degradation due to UWB interference
can be expressed in terms of the signal to noise ratio reduction r in the receiver,
that is,
r=

IUWB + N
,
N

(7.1)

where IUWB is the interference due to UWB devices in the area on the victim
receiver and N, in the simplest case, is the thermal noise power which depends

Christian Politano et al.

473

on the receiver characteristics. Using standard formulas, N in (7.1) can be expressed as

N = 144 + 10 log10 (BRX ) + NF [dBW],

(7.2)

where BRX is the victim receiver bandwidth in MHz and NF is the corresponding noise figure.
Starting from the specifications on the maximum allowed degradation r for
each of the considered victim receivers using (7.1) and (7.2), the maximum allowable UWB interference power IUWB is calculated. This value is then used as
reference to assess coexistence between UWB and the selected victim terminal.
7.4.2.1. Fixed wireless systems interference specifications
To evaluate the maximum allowable UWB interference on the existing fixed wireless systems (FWA and PP), the ITU-R requirements in [41] have to be considered.
They report on the limits of the interference power, I, due to unwanted emissions
from sources other than fixed services or services sharing the same band on primary bases. In [41] the total interference I on a fixed service link is partitioned as
89 % for the intra service interference; 10% for the co-primary services interference. The remaining 1% is for the aggregation of interference due to radio services
sharing frequency allocations on a non-primary basis, unwanted emissions (i.e.,
out-of-band and spurious emissions such as energy spread from radio systems,
etc.) in non-shared bands and finally unwanted radiations (e.g., industrial scientific medical (ISM) applications).
The previous percentages only apply to the performance objectives [42, 43]
and degradation allowance refers to the aggregation of the whole secondary services transmitters and unwanted signals.
In order to ensure not conflicting coexistence, ITU-R defined long and short
term interference criteria among generic wireless devices. For our purposes only
the long-term criteria are of interest. In particular, while for co-primary sharing
a degradation of r = 0.5 dB, corresponding to I/N = 10 dB, is admitted, for
secondary service interference and unwanted emissions, the performance degradation shall not exceed r 0.05 dB corresponding to I/N 20 dB. The previous
values for I/N represent generic objectives and refer to the average interference
which is the case analyzed in this section. However, due to the pulsed characteristic of UWB signals, separate considerations would be needed for both average and
peak interference objectives in the victim receiver band.
It should be noted that the allocation of 99% of the interference margin to
intra-service and/or co-primary services (89%+10%) might also be too pessimistic
with todays technology and in more realistic conditions higher margins than 1%,
in particular I/N closer to 0 dB, could be tolerated by fixed links for secondary
services. As an example, I/N = 10 dB have been proposed in [33, 34, 36], as
more realistic estimates of interference margins against UWB for todays FWA and
PP technologies.

474

Regulation and standardization

To calculate the FWA or PP receiver noise power N in (7.2), the receiver characteristics of the considered FWA and PP systems need to be specified.
The considered FWA systems operate in the 3.54.2 GHz and 4.45.0 GHz
bands. They include two communicating devices: the FWA central station (CS)
and an FWA terminal station (TS). When the FWA receiver operates at 3.5 GHz,
with NF
= 5 dB, and BRX
= 50 MHz or as an alternative BRX
= 14 MHz. Using

(7.2), N = 122 dBW is obtained for 50 MHz at 3.5 GHz and N

= 127.5 dBW
for 14 MHz at 3.5 GHz. When the FWA systems transmits in the 4.45.0 GHz band
with NF
= 6 dB and BRX
= 50 MHz, N
= 121 dBW is obtained for BRX =
50 MHz. As an example, the corresponding UWB interference limit is IUWB =
142 dBW for NF = 5 dB at 3.5 GHz.
To evaluate the UWB interference IUWB on the FWA receiver, it is necessary to
account for the FWA-TS antenna radiation patterns envelopes in both horizontal
and vertical planes. Typically a 90 sectorial antenna with mainlobe gain of 16 dB
is deployed. The FWA receiver antenna characteristics are indicated in [44] and for
brevity are not repeated.
Finally, a representative PP link operating at 4.0 GHz is considered. Its receiver
noise figure is NF = 6 dB and its bandwidth is about 40 MHz [31]. Using (7.2), the
noise power of the PP receiver is N
= 122 dBW and the corresponding UWB
interference limit is IUWB = 142 dBW.
Also in the PP case, to evaluate the UWB interference IUWB on the PP receiver,
the antenna characteristics for the PP have been be specified. As reported in [45],
a parabolic approximation for the antenna mainlobe is assumed with a gain of
43.6 dB and a 3 dB beam width of 1 degree.
7.4.2.2. DCS1800, UMTS interference specifications
As indicated in [35], UWB interference is considered to be harmful when the
signal-to-noise ratio degradation is about r = 1 dB. Assuming NF = 9 dB for
the UMTS terminal [46] and using (7.2), N = 128 dBW, IUWB = 134 dBW
is obtained.
For the GSM and DCS-1800 the same performance specifications as for the
UMTS are considered. A target NF = 9 dB is assumed10 [48] thus obtaining IUWB =
148 dBW.
The limits on the allowable UWB interference for the victim systems considered in this section have been listed in Table 7.1. Also the DCS-1800 case with
NF = 5 dB has been included.
7.4.3. UWB interference scenario
In this section, the main characteristics of the UWB interference scenario used
to assess coexistence are detailed. The UWB system is located inside a commercial/industrial building and interferes with the victim receiver that can be inside or
outside the building.
10In [47] a noise figure goal of N = 5 dB is indicated for the DCS receiver.
F

Christian Politano et al.

475

Table 7.1. UWB allowable interference limits for FWA, PP, UMTS, DCS1800.
System

Parameters

IUWB

flow

fhigh

FWA
PP
UMTS

NF = 5 dB, r = 0.05 dB
NF = 6 dB, r = 0.05 dB
NF = 9 dB, r = 1 dB

142 dBW

3.475 GHz
3.980 GHz
2.165 GHz

3.525 GHz
4.020 GHz
2.170 GHz

NF = 9 dB, r = 1 dB
NF = 5 dB, r = 1 dB

148 dBW

1.805 GHz

1.8052 GHz

DCS-1800

142 dBW
134 dBW
152 dBW

The geometry of the selected interference scenario is shown in Figure 7.8. The
UWB devices communicate in accordance with a master-slave network architecture. We assume that any UWB device (master or slave) is equipped with an omnidirectional antenna with 0 dB gain. The UWB slaves are randomly located on
each building floor according to a uniform distribution while UWB masters are
positioned on a regular grid as shown in Figure 7.8(c). The floor dimensions are
120 m 60 m and the vertical distance between two adjacent floors is 3 m. The
building is 10 floors high. The evaluation of the UWB interference on the victim
receivers is detailed in the following sections. Both power and non-power controlled UWB devices are considered in the analysis including the spectral characteristics of the UWB signals used for multiple access. Only interference due to
UWB upstream communications is considered.
7.4.4. Calculation of the UWB power spectral density
In this section, a closed-form expression for the power spectral density (PSD) of
the DS and TH UWB signals is obtained. The PSD is then used to evaluate the
fraction of UWB interference power leaking in the band of the selected victim
receiver.
7.4.4.1. UWB signal models
Two dierent pulse amplitude modulated UWB signals are considered: the DSUWB and the TH UWB [37].
The baseband DS UWB signal sDS (t) can be written as
sDS (t) =

ck w t kTc ,

(7.3)

k=

where w(t) is the baseband pulse transmitted in each chip interval of duration
Tc ; ck is the amplitude of the modulated chip, that is, ck = bk/Nc  dk , where bk is
the transmitted symbol, Nc is the number of chips per symbol, and {dk } is the
code sequence assigned to the generic UWB slave. Finally,  x denotes the largest
integer lower than x. The code sequence {dk }, dk {1, 1}, has duration longer
than the symbol period T = Nc Tc and can be approximated as an ideal random
white sequence.

476

Regulation and standardization

120 m
60 m

UWB device
FWA CS

UMTS, DCS1800, GPS

(victim)

Dclearance
PP
(victim)

FWA TS
(victim)
(a)
FWA CS
PP
(victim)

10 m

3 m x ith
floor

15 m

50 m

UWB device

FWA TS
(victim)

UMTS, DCS1800, GPS

(victim)

(m)

(b)
30
20
10
0
10
20
30

60

dX

dY
dx

dYInt
40

20

0
20
(m)
dYext

Master
Slave

40

60

Floor

(c)
Figure 7.8. UWB hot-spot interference scenario. (a) Upper view. (b) Side view. (c) Floor view.

The TH UWB signal can be written as

sTH (t) =

bk pk (t kT),

(7.4)

k=

where bk is the transmitted symbol in the kth time interval of duration T; pk (t) is
the corresponding pulse waveform. The random symbol sequence {bk } is white,

Christian Politano et al.

477

with zero mean and variance b2 . The pk (t) is

N f 1

pk (t) =

w t nT f ck,n Tc ,

(7.5)

n=0

where T f is the frame period and N f = T/T f is the number of frames per period
and the sequence {ck,n }, k = , . . . , + and n = 0, 1, . . . , N f 1, is the code word
segment accounting for the pulse time shift for each frame in the kth symbol time
interval. The {ck,n } takes integer values in {0, 1, . . . , Nh 1}, where Nh = T f /Tc .
For the signals in (7.3) and (7.4), the Gaussian pulse w(t) is considered:
w(t) =

1 (t)2 /2 2
e
,
2

(7.6)

where the parameter in (7.6) is the standard deviation of the Gaussian pulse
waveform which can be related to the practical duration of the pulse w(t) and to
its bandwidth;11 is the mean value indicating the pulse peak position within the
chip time interval Tc .
7.4.4.2. Spectral analysis for the selected UWB signals
Considering {ck } in (7.3) as an ideal random bipolar zero-mean white sequence
with standard deviation c2 , the power spectrum of the baseband DS UWB signal
is [49]
SDS ( f ) =


c2 
W( f )2 ,
Tc

(7.7)

where W( f ) is the Fourier transform of the pulse waveform w(t), for example,
W( f ) = e2

2 f 22

e j2 f .

(7.8)

The power spectrum of the TH UWB signal in (7.4) can be conveniently approximated as (see [50])
!

"

 2

2
 sin f N f T f 2


2 
 ,


STH ( f ) = b W( f ) N f 1 B( f ) + B( f )


T
sin2 f T f

(7.9)

where B( f ) is


B( f ) =

sin2 f Nh Tc
1

 2
Nh
sin2 f Tc

11In the following, the UWB signal bandwidth W

UWB

corresponding Gaussian pulse signal.

(7.10)

is defined as the 10 dB bandwidth of the

478

Regulation and standardization

and W( f ) is given in (7.8). The above expression for the TH UWB spectrum is
independent on the specific code sequence assigned to each user. The dependence
of the TH UWB spectrum from the generic code sequence {ck,n } assigned to the
user has been averaged out assuming that the period of the code sequence {ck,n } is
much longer than the symbol period and can be approximated as a white sequence.
To comply with the FCC requirements, the transmitted UWB signal can be
obtained as harmonic modulation of the baseband signal in (7.3) or (7.4), that is,


sUWB (t) = s(t) cos 2 f0 t ,

(7.11)

where f0 is the UWB center frequency that can vary from 0 (e.g., baseband UWB
signal) up to several GHz. Using standard formulas [49], the spectrum of the UWB
signal in (7.11) is
SUWB ( f ) =




*
1)
SDS,TH f f0 + SDS,TH f + f0 .
2

(7.12)

When the signal in (7.11) is transmitted, it is linearly distorted by the antenna. As

outlined in [37] for baseband transmissions, the antenna acts as a derivative filter
on the transmitted signal. However, when the UWB center frequency f0 increases,
the derivative eect tends to disappear and a nondistorting behavior of the antenna
can be assumed. Starting from the previous considerations, antenna distortion is
modelled with a highpass filter with transfer function Ha ( f ) that, in the simplest
case, can be approximated as

j2A f ,

| f | < fco ,

j,

| f | fco ,

Ha ( f ) =

(7.13)

where fco is the cuto frequency and A is a normalization constant such that
Ha ( fco ) = j. For frequencies f fco , the antenna is modelled as all pass.
Due to antenna filtering, the spectrum of the transmitted UWB signal is


2

SUWB TX ( f ) = SUWB ( f ) Ha ( f ) ,

(7.14)

where SUWB ( f ) is given in (7.12).

Neglecting for the moment the attenuation due to propagation, the average
interference due to a single UWB terminal in the victim receiver band is


IUWB,single = 2

fhigh
flow

SUWB TX ( f )df ,

(7.15)

where flow 0 and fhigh flow are the two extremes of the victim band (uplink or
downlink).12 These values have been indicated in the last two columns of Table 7.1
for each one of the selected victim receivers.
12To evaluate the maximum allowed UWB interference, the frequency response of each victim
receiver was assumed as an ideal brick-wall filter with band equal to that assigned to the specific service,
for example, 5 MHz for the UMTS, and so on.

UWB power fraction on victim band,

Christian Politano et al.

479
Gaussian pulse

0
20
40
60
80
100

UWB central frequency, f0 (GHz)

FWA
PP

UMTS
DCS

Figure 7.9. versus the UWB center frequency as a function of the UWB center frequency; DS UWB
signal, Gaussian pulse, WUWB = 3 GHz; fco = 1 GHz.

Indicating with PT the power transmitted by the UWB device, the interfering
power in (7.15) can be conveniently rewritten as
IUWB single = PT ,

(7.16)

where 1 is the fraction of UWB power in the band of the victim receiver and
 fhigh
= 2 flow SUWB TX ( f )df , where SUWB TX ( f ) is the normalized UWB spectrum,

that is, SUWB TX ( f )df = 1. From (7.12) and (7.14), depends on f0 , on the
UWB spectrum shape, and on the antenna characteristics, for example, fco in this
case.
In Figure 7.9, is plotted as a function of UWB center frequency assuming a
Gaussian pulse with 10 dB bandwidth of 3 GHz.13 In Figure 7.10, the values of
obtained for dierent values of N f and Nh in (7.9) are plotted in the DS and TH
UWB cases. The FWA case is considered.
As shown in Figure 7.10, is practically independent on Nh and N f and the
more significative variations are on the order of 3 dB and have been obtained for
N f /Nh = 8. Still in Figure 7.10, the values of obtained for both DS UWB and TH
UWB signals have been superimposed. The dierence between the two curves is
negligible and for this reason, in the following, only for the DS UWB signals are
considered.
The dependence of on the UWB signal bandwidth has been analyzed and
results have been shown in Figure 7.11 for the FWA and UMTS cases. For brevity,
the corresponding to the other victim systems have not been indicated.
Finally, in Figure 7.12, for dierent values of the antenna cuto frequency
fco is plotted in the FWA and UMTS cases.
13The 10 dB UWB bandwidth is W

UWB =

2 fB with fB such that |W0 ( fB )|2 = 0.1.

480

Regulation and standardization

UWB power fraction on victim band,

TH-CDMA, FWA
10

20

30

40

3
4
5
6
UWB central frequency, f0 (GHz)
N f = 32, Nh = 8
N f = 32, Nh = 4
N f = 16, Nh = 8

N f = 16, Nh = 16
DS-CDMA

UWB power fraction on victim band,

Figure 7.10. versus the UWB center frequency for dierent values of N f and Nh ; FWA case; Gaussian
pulse, WUWB = 3 GHz; fco = 1 GHz, f0 = 4.5 GHz.

Gaussian pulse

0
20

FWA

40

UMTS
60
80
100

1
2
3
4
5
6
UWB central frequency, f0 (GHz)

BWUWB = 2 GHz
BWUWB = 4 GHz
Figure 7.11. Fraction of the UWB power in the receiver band as function of the UWB carrier frequency; UWB signal with variable Gaussian pulse bandwidth; fco = 1 GHz; FWA and UMTS victim
systems.

The increase of for lower values of f0 and for fco = 3 GHz in the UMTS case
is due to the derivative distortion introduced by the antenna which leads to a shift
of the UWB spectral components towards higher frequencies.

UWB power fraction on victim band,

Christian Politano et al.

481
Gaussian pulse

0
20

FWA
40
60

UMTS

80
100

1
2
3
4
5
6
UWB central frequency, f0 (GHz)

fco = 0 GHz
fco = 3 GHz
Figure 7.12. Fraction of the UWB power in the receiver band as function of the UWB carrier frequency; UWB signal with 3 GHz Gaussian pulse bandwidth; variation of antenna cuto frequency fco ;
FWA and UMTS systems.

7.4.5. Calculation of UWB average interference

Using (7.16), the aggregate UWB interference in the victim band can be calculated
as
IUWB =

m PTm A dm ,

(7.17)

m=1

where M is the number of UWB active terminals in the hot spot, PTm is the power
transmitted by the mth UWB slave. The m are binary random variables accounting for the activity of the mth user, for example, Prob(m = 1) = p and Prob(m =
0) = 1 p, where p is the activity factor of the UWB device.
The terms {A(dm )}, m = 1, 2, . . . , M, in (7.17) account for both indoor and
(possible) outdoor propagations and includes the antenna gain of the victim receiver. The A(d) can be expressed as
A(d) = GTXUWB GRX L(d)SINT SOUT ,

(7.18)

where GTXUWB and GRX indicate the antenna gains of the UWB transmitter and of
the victim receiver, respectively.
In the FWA and PP cases, the antenna gain of the victim receiver depends
on the relative orientation and on the distance between the generic UWB slave
inside the building and the FWA or PP receiver outside the building. The SINT ,
SOUT are two log-normal random variables accounting for shadowing eects for
both indoor and (possibly) outdoor propagations. Finally, L(d) is the overall path

482

Regulation and standardization

loss whose model is taken from [34], that is,



LdB (d) = L f s(dB) LB(dB) 10INT log10 dint



10OUT log10

(7.19)

dout + dint
,
dint

where L f s = (4/)2 is the free-space attenuation at distance d = 1m and LB is

the building penetration loss. The INT and OUT are the path loss exponents for
indoor and outdoor propagations, respectively.
Referring to the scenario in Figure 7.8 for the FWA and PP, dint is the distance
of the UWB slave from the external wall of the building which is in front of the
victim receiver; dout is the distance (in 3D) between the point indicating the projection of the UWB slave on the external wall in front of the victim receiver (see
Figure 7.8) and the victim receiver itself. The victim receiver antenna gain GRX
in (7.18) is evaluated along the direction connecting the projection point for the
UWB and the victim receiver.
When the UMTS and DCS terminals are considered L(d) in (7.18) only accounts for the internal path loss model, that is, LdB (d) = L1 10INT log10 (d)
and d is the distance between the generic UWB transmitter and the victim receiver.
The UWB interference in (7.17) depends on the (instantaneous) geometry of
the UWB terminals inside the building, on their state (active or not) and on the
transmitted power PT required to the UWB slave to be received with a specified
signal-to-noise ratio to its serving master. In the power controlled case, {PTm } depend on the interference among the UWB devices in the area which in turn depend
on the relative position of the UWB devices. The UWB interference power for each
floor is obtained after averaging (7.17), for example,
)

IUWB = M pE PT A(d) .

(7.20)

The calculation of E{A(d)} can be easily reformulated as the average with respect
to the spatial coordinates (x, y, z) of the generic UWB device. Assuming that UWB
devices have omnidirectional antennas with 0 dB gain, the average interference
produced by the UWB devices at floor height hF is


* )

* )

IUWB hF = M pE PT L x, y, hF GRX x, y, hF E SINT E SOUT .

(7.21)

Using standard formulas, E{S} = e0.5S for S = SINT and SOUT . As shown in
the following, a simplified expressions for (7.21) can be obtained in the nonpower control case. In the power control case, the exact calculation of E{PT
L(x, y, hF )GRX (x, y, hF )} in (7.21) seems to be dicult and it is necessary to resort to simulation. The total UWB interference to be compared with the allowable
limits in Table 7.1 is obtained summing (7.21) over the floors, for example,
IUWB =

hF

IUWB hF = M pI O

i hF ,

hF

(7.22)

Christian Politano et al.

483
Gaussian pulse, WUWB = 3 GHz @10 dB

Maximum allowable power (dBm)

0
10
20
30
40
50
60

2
4
6
8
10
UWB central frequency, f0 (GHz)

12

DS
TH
Figure 7.13. Maximum transmitting power for DS and TH UWB signals in accordance with the FCC
limitations as a function of the UWB signal center frequency; fco = 1 GHz; N f = 32, Nh = 4 for the
TH UWB case.

where I = E{SINT }, O = E{SOUT }, and i(hF ) is the coecient accounting for

the interference produced by the UWB devices located at floor height hF , that is,
i(hF ) = E{PT L(x, y, hF )GRX (x, y, hF )}. By definition, the i(hF ) is independent
of the UWB trac characteristics as well as of number of UWB users in the area
and of the shadowing characteristics. It accounts for the (average) geometry of the
link. To obtain (7.22), it was implicitly assumed that the number of UWB slaves
M for each floor is the same and floors are isolated.
7.4.5.1. Non-power control case
In the non-power control case, the UWB devices transmit at their maximum allowable power which is fixed in accordance with their spectral characteristics and
accounting for the spectral emission masks issued by FCC (or ETSI).
The maximum transmitted power PTmax respecting the FCC/ETSI restrictions
depends on the UWB center frequency f0 . In Figure 7.13, PTmax as a function of f0
is plotted for both the DS and TH UWB signals. In the non-power controlled case,
the average UWB interference per floor in (7.20) can be rewritten as


*

IUWB hF = UWB M pPTmax E L x, y, hF GRX x, y, hF I O .

(7.23)

The (7.23) can be inserted in (7.22) to obtain the total interference on the victim
receiver.
Results for the non-power controlled case. The additional scenario parameters
adopted for the evaluation of IUWB in the non-power control case are f0 = 4.5 GHz
corresponding to a PTmax = 6.2 dBm for the DS and TH UWB cases with 10 dB

Regulation and standardization

Aggregate interference power (dBW)

484

100 UWB device/floor, AF = 10%, 100 Mbps

125
130
135
140
145
150
155

4
6
8
Victim distance (km)

10

12

FWA
PP
Figure 7.14. Average UWB interference as a function of the distance of the victim receiver from the
building; non-power control case; PP and FWA.

bandwidth WUWB = 3 GHz; the standard deviations SINT = 4 dB and SOUT = 0 dB

(e.g., no shadowing) have been considered.
In Figure 7.14, the average UWB interference (7.23) is plotted as a function of
the distance of the victim receiver from the building. The behavior of the PP curves
in Figure 7.14 can be explained looking at the scenario in Figure 7.8. It can be observed that, for a fixed clearance X by varying the distance of the PP receiver from
the UWB hot spot, the antenna gain under which the PP sees the hot-spot increases
up to the maximum (43.6 dBi). This leads to an increase in the interference power
as also indicated in Figure 7.14. By further increasing the PP-UWB distance, the
attenuation due to propagation becomes dominant and interfering power restarts
to decrease. Thus, dierently from the FWA to reduce UWB interference, it should
be more convenient to locate the PP receiver in the close proximity of the UWB
hot spot.
From Figures 7.14 and 7.15, it can be observed that for a fixed distance between the hot spot and the victim receiver, when it is impossible to reduce the
UWB transmitting power to respect the interference limits in Table 7.1, the only
viable alternative is to reduce the (average) number of active UWB terminals indicated by the product M p in (7.22). This may lead to a significant reduction of the
UWB system capacity due to coexistence requirements.
7.4.5.2. Power control case
As shown in this section, the adoption of power control in the UWB allows to
obtain a consistent reduction of the (average) power transmitted by the UWB devices. In the first generation of UWB devices, power control is not included in
order to reduce the UWB modem complexity and thus costs. However, as shown

Aggregate interference power (dBW)

Christian Politano et al.

485
100 UWB device/floor, AF = 10%, 100 Mbps,
f0 = 4.5 GHz, dMIN = 0.5 m

140

150

160

170

180

20
40
60
80
Victim distance on Y axis (m)

100

UMTS
DCS 1800
Figure 7.15. Average UWB interference as a function of the x position of the victim receiver inside the
building; non-power control case; UMTS and DCS-1800.

in the following, the adoption of power control introduces additional degrees of

freedom for the UWB network designer and allows to safely improve UWB system
capacity without incurring in coexistence problems also in very critical scenario
such as the one considered in this section.
A UWB system with a master-slave architecture is considered. Masters are uniformly located over a regular rectangular grid while the power controlled UWB
slaves are randomly located in the floor area in accordance with a uniform distribution. Unless otherwise noted, in the following, 32 masters per floor are assumed.
To evaluate the power transmitted by each slave (e.g., PTm in (7.20)), computer
simulation is adopted. The principle of operation of the simulation program is
illustrated in the flow chart in Figure 7.16. At the beginning of each outer loop
iteration, a new scenario is generated. Scenario generation consists in randomly
placing the slaves in the area according to a uniform spatial distribution. Each
slave is connected to the master with the lowest indoor overall loss including shadowing. The inner loop is used to simulate an iterative power control procedure.
In each inner iteration, the power of each transmitter is increased or decreased
in accordance with the result of the comparison between the received power-tointerference ratio C/I, calculated for each slave in the area, and the target C/I, that
is, (C/I)Target .14 The power updating procedure is halted when the power transmitted by each active UWB device is practically constant or the maximum allowable
transmission power has been reached. At the end of the inner loop, all the relevant
parameters, such as the power level transmitted by each user, the number of terminals in outage and so forth, are collected. The powers {PTm } transmitted by the
14The target carrier to interference ratio is evaluated as (C/I)
Target = (Eb /N0 )Target PG, where

PG is the processing gain that varies with the UWB slave bit rate.

486

Regulation and standardization

Scenario initialization
Scenario
generation

Inner loop

Outer loop

Interference
calculation

Transmission power
updating (power
control)

Equilibrium

No

Yes
Snapshot
statistic
collection

No

Snapshots
end
Yes
End

Figure 7.16. Simulator flowchart; upstream simulation.

mth UWB slave at equilibrium are then used in (7.21) to evaluate the interference
on the victim receiver. In the (C/I) calculation, the eects of the interference due
to the narrowband systems on the UWB are assumed to be negligible. However,
they could be easily included in the simulation by suitably increasing the UWB receiver noise power by an amount depending on the spectral characteristics of the
UWB signal and of the narrowband systems. The simulations results presented in
this section have been obtained considering variable UWB device density as well
as dierent bit rates for the UWB slaves.
In Figure 7.17, the histogram on the UWB transmitted power is plotted in the
case of terminals operating at 2, 25, and 100 Mbps. From Figure 7.17, it can be
observed that due to power control action the UWB transmitting power is always
well below the maximum allowable limits indicated by FCC. As expected, in this
case, the total average UWB interference power rapidly decreases as also shown in
the following. In Figure 7.18, the aggregate UWB interference power is reported as
a function of the distance of the FWA and PP receivers from the UWB hot spot.
From Figure 7.18, it can be observed that in every practical case the UWB interference is well below the limits in Table 7.1. Results in Figure 7.18 were obtained
considering the largest values for corresponding to f0 = 4.5 GHz in the FWA
and PP cases. Results in Figure 7.18 can be easily recalculated for dierent UWB

Christian Politano et al.

487
Transmit power pdf

0.06

2 Mbps
0.05
0.04
25 Mbps

0.03
0.02

100 Mbps
0.01
0

85

80

75

70 65 60
(dBm/MHz)

55

50

Figure 7.17. Power distribution for power controlled devices transmitting at 2, 25, and 100 Mbps, 400
UWB devices/floor.

Aggregate interference power (dBW)

Device/floor @ f0 = 4.5 GHz, AF = 5%, 100 Mbps

145

PP

155

FWA

165

175

100 UWB
200 UWB

6
9
Victim distance (km)

12

300 UWB
400 UWB

Figure 7.18. Aggregate UWB interference power as a function of the victim receiver distance from the
UWB hot spot; FWA, PP victim systems; UWB carrier frequency f0 = 4.5 GHz.

center frequencies by scaling them with the factor ( f0 )/R , where R is the interference fraction at f0 = 4.5 GHz and ( f0 ) is the fraction at frequency f0 which
can be easily obtained from the curves in Figure 7.9. As expected, by changing f0 ,
it is possible to reduce the interference well below the limits in Table 7.1 also for
every distance between the UWB hot spot and the fixed victim device.
In Figure 7.19, the aggregate UWB interference for the UMTS and DCS-1800
cases is shown. The oscillating interference pattern in Figure 7.19 is due to the
position of the UWB masters on the regular grid in Figure 7.8. The discontinuity
in the plots is due to the building penetration loss (10 dB was assumed) when
passing from indoor to outdoor (or vice versa).

Regulation and standardization

Aggregate interference power (dBW)

488

100 Mbps, AF = 5%, f0 = 4.5 GHz

155
165

UMTS

175
185

DCS 1800

195
205

20
40
60
80
Victim distance on Y axis (m)

100

100 UWB
400 UWB
Figure 7.19. Aggregate UWB interference power as a function of the victim receiver distance from the
UWB hot spot; UMTS, DCS-1800; UWB center frequency f0 = 4.5 GHz.

Aggregate interference power (dBW)

400 UWB device/floor @ f0 = 4.5 GHz, AF = 5%

145

PP
160

175

190

FWA

12

Victim distance (km)

100 Mbps
25 Mbps
2 Mbps
Figure 7.20. Aggregate UWB interference power as a function of the victim receiver distance from the
UWB hot spot; FWA and PP systems; UWB carrier frequency f0 = 4.5 GHz; 400 UWB devices/floor;
variation of the bit rate.

From the results in Figures 7.18 and 7.19, it is evident that when the UWB
system operates within the allowed FCC band, the systems that could be impaired
from UWB are PP and FWA systems. As indicated in Figure 7.19, the selection of
f0 = 4.5 GHz leads to a significative reduction of the UWB interfering power for
UMTS and DCS also for very large UWB device densities. For this reason in the
following, only the FWA and PP systems are analyzed in detail.

Aggregate interference power (dBW)

Christian Politano et al.

135

489
400 UWB device/floor @ f0 = 4.5 GHz, AF = 5%

145

PP

155

165

175

FWA

12

Victim distance (km)

32 master/floor
64 master/floor
128 master/floor
Figure 7.21. Aggregate UWB interference power as a function of the victim receiver distance from the
UWB hot spot; FWA and PP systems; UWB carrier frequency f0 = 4.5 GHz; variation of the number
of masters; 400 UWB devices/floor.

Figure 7.20 evidences the reduction of UWB interference considering UWB

transmissions at bit rates lower than 100 Mbps. It can be observed that the bit rate
reduction allows to increase the receiver processing gain thus leading to a corresponding reduction in the UWB slave transmitter power and thus in the interference on the victim receivers.
Finally, in Figure 7.21, IUWB is reported for dierent number of UWB master
stations in the area. As expected, the increase in the number of master stations
can lead to a significative reduction of UWB interference. This may represent an
additional degree of freedom for the UWB system designer at the expense of an
increased complexity of the overall system architecture.
7.4.6. Summary
The assessment of the potential interference caused by UWB devices is of great importance to guarantee nonconflicting coexistence between UWB technology and
any other existing and future wireless systems as well as to gain acceptance of
UWB technology worldwide. Any study on this aspect needs to go beyond simplistic models and should adopt realistic scenario deployments for UWB. Similarly to many other wireless local area network systems, UWB terminals will be in
sleep mode for the large percentage of time, will not run continuously, and will
not emit constantly at the maximum allowed power. In this section, the potential
harmful interference of UWB on many categories of victim receivers was analyzed.
Both non-power controlled and power controlled UWB terminals were considered. UWB interference on the narrowband systems was evaluated including the

490

Regulation and standardization

spectral characteristics of the transmitted UWB signal including antenna distorting eects and dierent UWB carrier frequencies. Dierently from the results presented in the open literature, mainly focusing on the non-power controlled single
terminal case, it was shown that by considering some simple system features for
the UWB system such as its architecture, the activity factor, power control, and so
forth, even for very large UWB terminal densities, there is no practical risk for the
selected victim receivers especially when the carrier frequency of the UWB is selected in the 3.110.6 GHz band. The coexistence issues of power controlled UWB
with fixed wireless systems (FWA and PP) have been analyzed in detail when the
UWB operates in the 3.110.6 GHz band. Even in this case, there is no practical
risk for the FWA and PP operations. Further margins on UWB interference, not
even considered, still exist such as admission control techniques, antenna polarization mismatches, deep NLOS, and multiple through-wall indoor losses.
Bibliography
[1] International Telecommunication Union (ITU)Radiocommunication Study Groups: Chairman
Task Group 1/8, Report of the second meeting of Task Group 1/8 (Geneva, 2731 October
2003), ITU-R Document 1-8/59-E, 2 December 2003.
[2] Mandate forwarded to CEPT to harmonize radio spectrum use for ultra-wideband systems in the
European Union, DG INFSO/B4, 18 February 2004; http://europa.eu.int/information society/
policy/radio spectrum/docs/current/mandates/rscom0408 mandate uwb.pdf.
[3] Mandate forwarded to CEN/CENELEC/ETSI in the field of information technology and telecommunications, DG ENTR/G/3, 25 February 2003; http://portal.etsi.org/public-interest/Documents/
mandates/M329EN.pdf.
[4] Report of the First Meeting of ECC-TG3, Maisons-Alfort, France, 19 April 2004, Document
TG3#1 10R1, approved and issued 23 April 2004.
[5] Federal Communications Commission (FCC), Revision of Part 15 of the commissions rules regarding ultra-wideband transmission systems, First Report and Order, ET Docket 98-153, Federal Communication Commission 02-8, Washington, DC, USA, adopted on 14 February 2002,
and released on 22 April 2002.
[6] ETSI-TG31A Technical Report: Electromagnetic compatibility Radio spectrum Matters (ERM);
Short Range Devices (SRD), Technical characteristics for SRD communication equipment using
Ultra-Wideband technology (UWB), TR 101 994-1 V1.0.0 (2003-8); also available as document
SE24M21 42.
[7] International Telecommunication Union (ITU)Radiocommunication Study GroupsDocument
1-8/TEMP/XX-E17 June 2004: Annex 1 to TG 1/8 Chairmans Report (Working Document),
Characteristics of ultra-wideband technology (Questions ITU-R 226/1 and ITU-R 227/1).
[8] IEEE 802.15 Overview web site, http://standards.ieee.org/wireless/overview.html#802.15.
[9] IEEE 802.15 WPAN High Rate Alternative PHY Task Group 3a (TG3a) web site, http://ieee802.
org/15/pub/TG3a.html.
[10] IEEE 802.15.3a Project Authorization Request, December 2002, http://standards.ieee.org/board/
nes/projects/802-15-3a.pdf.
[11] Task group 3a Call For Intent and Proposals (CFI/CFP), December 2002, http://grouper.ieee.org/
groups/802/15/pub/TG3a CFP.html.
[12] TG3a Technical Requirements, December 2002, http://grouper.ieee.org/groups/802/15/pub/2003/
Jan03/03030r0P802-15 TG3a-Technical-Requirements.doc.
[13] P802.15.3a Alt PHY Selection Criteria, December 2002, http://grouper.ieee.org/groups/802/15/
pub/2003/Jan03/03031r6P802-15 TG3a-PHY-Selection-Criteria.doc.
[14] TG3a Down Selection Voting Procedure, January 2003, http://grouper.ieee.org/groups/802/
15/pub/2003/Jan03/03041r7P802-15 TG3a-Down-Selection-Voting-Procedure.doc.

Christian Politano et al.

491

[15] Channel Modeling Sub-committee Report Final, November 2002.

[16] A. A. Saleh and R. A. Valenzuela, A statistical model for indoor multipath propagation, IEEE
Journal on Selected Areas in Communications, vol. 5, no. 2, pp. 128137, 1987.
[17] MultiBand OFDM Alliance web site, http://www.multibandofdm.org.
[18] Xtreme Spectrum web site, http://www.xtremespectrum.com.
[19] Motorolas UWB web site, http://www.motorola.com/uwb/.
[20] IEEE 802.15 WPAN Low Rate Alternative PHY task group 4a (TG4a) web site, http://www.
ieee802.org/15/pub/TG4a.html.
[21] IEEE 802.15 WPANTM Task Group 5 (TG5) web site, http://www.ieee802.org/15/pub/TG5.html.
[22] IEEE 802.19 Coexistence Technical Advisory Group (TAG) web site, http://grouper.ieee.org/
groups/802/19/.
[23] European Telecommunications Standards Institute web site, http://www.etsi.org/.
[24] ERM TG31A: Ultra Wide Band for Short Range Devices, Terms of References, http://portal.
etsi.org/erm/ermtg31a tor.asp.
[25] ETSI TR 101 994-1 V1.1.1 (2004-01) Technical characteristics for SRD equipment using Ultra
Wide Band technology (UWB) Part 1: Communications applications.
[26] ETSI TR 101 994-2 V1.0.0 (2004-03) Technical characteristics for SRD equipment using Ultra
Wide Band technology (UWB) Part 2: Ground-and Wall- Probing radar applications; System
reference Document.
[27] ETSI TR 101 982 Electromagnetic compatibility and Radio spectrum Matters (ERM); Radio
equipment to be used in the 24 GHz band; System Reference Document for automotive collision warning Short Range Radar.
[28] Association of Radio Industries and Businesses web site, http://www.arib.or.jp/.
[29] Global Radio Standardization Collaboration (GRSC) web site, http://www.rast.etsi.org/.
[30] Federal Communications Commission (FCC), Revision of Part 15 of the Commissions Rules
Regarding Ultra-Wideband Transmission Systems, First Report & Order, ET Docket 98-153, Federal Communication Commission 02-48, Washington, DC, USA, adopted on 14 February 2002,
and released on 22 April 2002.
[31] R. Macchi (Siemens), Revision of UWB and FS coexistence report (Annex 5-1), in Proceedings
of Conference of European Post and Telecommunication Administrations SE24 Meeting 17, Mainz,
Germany, September 2002.
[32] J. Huschke, Aggregated UWB interference into the uplink of UMTS, in Proceedings of Conference of European Post and Telecommunication Administrations SE24 Meeting 17, Mainz, Germany,
September 2002.
[33] R. Giuliano, G. Guidoni, F. Mazzenga (Radiolabs), and D. Porcino (Philips), Study of the coexistence UWBFixed Wireless Access (FWA) systems (with power control variation analysis), in
Proceedings of Conference of European Post and Telecommunication Administrations SE24 Meeting
18, Copenhagen, Denmark, December 2002, IST ULTRAWAVES.
[34] J. Foerster and A. Stephens (Intel), Response to the Revision of UWB and FS coexistence report
(Annex 5-1) concerning aggregate UWB interference into FS systems, in Proceedings of Conference of European Post and Telecommunication Administrations SE24 Meeting 18, Copenhagen,
Denmark, December 2002.
[35] J. Huschke, Generic power spectral density limits for a single UWB interfer into the uplink of
UMTS, in Proceedings of Conference of European Post and Telecommunication Administrations
SE24 Meeting 18, Copenhagen, Denmark, December 2002.
[36] R. Giuliano, G. Guidoni, F. Mazzenga, and F. Vatalaro, On the coexistence of UWB with fixed
wireless access systems, in Proceedings of IEEE Personal Wireless Communications (PWC 03), pp.
101113, Venice, Italy, September 2003.
[37] M. Hamalainen, V. Hovinen, R. Tesi, J. H. J. Iinatti, and M. Latva-aho, On the UWB system
coexistence with GSM900, UMTS/WCDMA, and GPS, IEEE Journal On Selected Areas in Communications, vol. 20, no. 9, pp. 17121721, 2002.
[38] D. Landi and C. Fischer, UWB interference to IEEE 802.11a/b WLAN, Swisscom Innovations,
D4.2v1.0, 18/08/03.

492

Regulation and standardization

[39] M. Hamalainen, J. Saloranta, J.-P. Makela, I. Oppermann, and T. Patana, Ultra wideband signal
impact on IEEE802.11b and bluetooth performances, in Proceedings of 14th IEEE International
Symposium on Personal, Indoor, and Mobile Radio, Communications (PIMRC 2003), vol. 3, pp.
29432952, Beijing, China, September 2003.
[40] D. Landi and C. Fischer, UWB interference with bluetooth, Swisscom Innovations, D4.3v1.0,
20/08/03.
[41] ITU-R F.1094-1 (10/95), Maximum allowable error performance and availability degradations
to digital radio-relay systems arising from interference from emissions and radiations from other
sources.
[42] ITU-R F.1397, Error performance objectives for real digital radio links used in the international
portion of a 27500 km hypothetical reference path at or above the primary rate.
[43] ITU-R F.1491, Error performance objectives for real digital radio links used in the national portion of a 27500 km hypothetical reference path at or above the primary rate.
[44] ITU-R F.1336, Reference radiation patterns of omnidirectional, sectoral and other antennas in
point-to-multipoint systems for use in sharing studies in the frequency range from 1 to about 70
GHz.
[45] ETSI EN 300 833 V1.4.1, Fixed Radio Systems; Point-to-point antennas; Antennas for point-topoint fixed radio systems operating in the frequency band 3 GHz to 60 GHz, (2002-07).
[46] 3GPP TR 25.942 V6.1.0 (2003-09) Technical Specification Group Radio Access Networks, Radio
Frequency (RF) system scenarios, (Release 6).
[47] ETS(I) 300 190, (GSM 05.05 version 5.4. 1): Digital cellular communication system (Phase 2+);
Radio transmission and reception, European Telecommunications Standards Institute, Sophia
Antipolis, Cedex, France, August 1997.
[48] J. Janssens and M. Steyaert, CMOS Cellular Receiver Front-Ends: From Specification to Realization,
Kluwer Academic, Dordrecht, the Netherlands, 2002.
[49] J. G. Proakis, Digital Communications, McGraw-Hill, New York, NY, USA, 4th edition, 2001.
[50] D. Cassioli and F. Mazzenga, Spectral analysis of UWB multiple access schemes using random
scrambling, IEEE Transactions on Wireless Communications, vol. 3, no. 5, pp. 16371647, 2004.

Christian Politano: STMicroelectronics, 1228 Plan-les-Ouates, Geneva, Switzerland

Email: [email protected]
Walter Hirt: IBM Research GmbH, Zurich Research Laboratory, 8803 Ruschlikon, Switzerland
Email: [email protected]
Nils Rinaldi: STMicroelectronics, 1228 Plan-les-Ouates, Geneva, Switzerland
Email: [email protected]
Gordana Drakul: STMicroelectronics, 1228 Plan-les-Ouates, Geneva, Switzerland
Email: [email protected]
Romeo Giuliano: RadioLabs, Consorzio Universit`a Industria-Laboratori di Radiocomunicazioni,
Via del politecnico 1, 00133 Rome, Italy
Email: [email protected]
Franco Mazzenga: Dipartimento di Ingegneria Elettronica, Universit`a di Roma Tor Vergata,
via del politecnico 1, 00133 Rome, Italy
Email: [email protected]

Index

Symbols
(UWB)2 , 221, 222
HV Corr., 296
HV Ind., 296
H Pol., 296
V Pol., 296
K model, 69
(DS) CDMA, 221
3D ranging experiments, 440
3D3215, 185
A
active positioning, 430
active RFID, 420
Aloha, 219
amplitudes, 74, 8991
analogue to digital converter, 197
angle of arrival, 414
antenna, 84
antenna design, 284
antenna design and analysis, 96
antenna near-field eects, 95
applications, 9
around the body radio propagation
channels, 106
arrival rate, 69
arrival statistics, 68
assisted GPS, 419
asymmetric location systems, 415
ATMega8L, 195
ATMEL Inc., 195
automatic gain control, 195
AWGN, 55
B
beam steering, 276
beamforming, 330
beampattern, 330
beamwidth, 332
BFP 540F, 185
BFP 540F, 186
binary pulse-position modulation
(BPPM), 353
body phantom, 104
body proximity, 95

body-area network, 25
body-worn device, 99, 102
boresight, 84
broadband, 331
busy tones, 206
C
carrier sensing, 206, 207
carrier sensing multiple access, 206
CDMA, 25
channel estimation, 25
channel model, 255, 256, 258
cluster arrival process, 88, 90
clusters, 69, 77, 88
coding gain, 364
coexistence, 175
collision avoidance, 206
complex programmable logic device, 186, 193
complexity, 3
control channel, 207
CoolRunner II, 186
correlations, 304
cost function, 208, 237, 238, 240
coupling, 302
cumulative probability graph, 416
D
Data Delay Devices Inc., 185
DATA throughput, 235
de-embedding, 26, 30, 35, 36
deconvolving, 80
delay dispersion, 24, 67
delay diversity, 67
deterministic channel modeling, 21
deterministic/statistic modeling of
the channel, 293
diuse scattering, 73
digital beamformer, 342
digital beamforming, 336
digital interpolation beamformer, 337
digital interpolation beampattern, 342
digital-to-analog converter, 195
direct sequence UWB, 118, 172, 220
distance information, 205
diversity gain, 364

494
DOA estimation, 330
DREAM, 214, 216218
DSR, 233
E
early-late, 192
sampler, 192
early-late tracker, 183
early-time response, 45
elevation angles, 84, 85
energy capture metric, 130
energy-awareness, 205
equalization, 120
ETSI, 6
experimental M-sequence device, 435
exposed terminal, 206
F
fading, 324
far-field, 3133, 35
fast stepped frequency chirps, 13
FCC, 1, 4
FCC pulse, 334
feeding network, 271
fiber optic RF feed, 101
flooding, 212, 214, 215, 217, 219
forest areas, 72
found connection, 233, 234, 237, 240, 244, 246
fractional calculus, 43
free space, 40
frequency correlation bandwidths, 115
frequency distortion, 85
frequency regulators, 22
frequency-dependent distortion, 80
Friis equation, 39
G
Gamma-distributed, 74
Gaussian impulse, 334
generalized Rake, 56
geometric dilution of precision, 417
GO, 42
GPS, 217, 417
GPSR, 218
grating lobes, 332
greedy forwarding, 209211
greedy forwarding strategy, 209
group mobility models, 223, 224
GTD, 41
GTD/UTD, 40
H
half-plane, 40
hand-held devices, 95

Index
hemisphere, 81
hidden terminal, 206
human body, 97
human body model, 98
I
IEEE 802.15.3a, 8
IEEE 802.15.4, 8
IEEE standards, 87
implementation issues, 298
impulse radio, 144
impulse radio UWB, 10
impulse response, 255, 257, 259, 262
individual echo, 255, 256, 261, 263, 266, 267
Inertia, 227, 231, 246, 248
inertia, 223, 224, 226
Inertia mobility model, 239
inertia model, 224
Infineon Technologies, 190
infrastructure for ranging, 441
InGaP/GaAs, 189
interacting objects, 68, 88
intercluster decay time constant, 72
interpath arrival times, 71
interpolation, 336
interpolation error, 341
interpolation kernel, 338
interrogation, 49
intracluster decay time constant, 72
ISI, 55
J
jitter, 189
K
Kerberos, 227, 229231, 243, 246, 247
Kerberos mobility model, 223, 226
kernels with finite duration, 340
L
LAR, 214, 215, 217, 219, 222, 233, 237, 246
LLR, 218
local oscillator, 189, 190, 192, 193
location accuracy, 416
location information, 209, 212, 217
location-aided routing protocol, 222
location-aware, 219, 221, 222
location-aware routing protocols, 209
location-based, 249
locationing information, 212, 249
lognormal distribution, 75, 91
low-noise amplifier, 189, 190

Index
M
M-sequence, 434
mainlobe, 332
MAX1304, 195
MAX155, 195
MAXIM Inc., 195
MBOA, 9
measurement data, 93
measurement methodologies, 96
measurement techniques, 24
microcontroller, 183, 190, 195, 197199
microstrip line, 190
shorted, 186
slotline, 190
tapered, 190
MIMO, 353
minimum mean-square error, 122
MLSE, 55
MMIC, 189
mobility model metrics, 227
mobility models, 222
modified Hermit polynomial functions, 161
modulation, 157
monocycle, 150
multiband, 353
multiband OFDM, 118, 220
multiband UWB, 16, 283
multicarrier schemes, 14, 147
multilateration, 413
multipath, 302
multipath cluster realization, 261
multipath clusters, 262, 267, 269
multipath diversity, 353
multiple access, 167
multiple access in multiband UWB, 288
mutual orientation, 284
N
Nakagami, 353
Nakagami distribution, 74
narrow transition regions, 41
narrowband RF detection, 419
NAZU, 76
NBB-300, 189
network analyzer, 24
normalized eective height, 270
Notice of Inquiry, 5
number of clusters, 70, 71, 91
number of DATA packets, 235, 248
number of DATA packets transferred, 233
number of found connections, 248
number of received DATA packets, 240, 244
numerical methods, 98

495
O
OFDM, 25, 353
oce environment, 71, 72, 93
on-body measurement setup, 102
on-demand, 214
on-demand routing protocol, 214
on-o keying, 158
out-of-band, 207
out-of-band signaling, 206
outdoor location systems, 420
P
PAL ranging system, 428
passive positioning, 430
passive RFID, 420
path loss, 24
Path loss model
frequency dependence, 64
path loss model, 57, 60, 92
distance dependence, 62
free-space model, 60
multislope model, 64
multiwall model, 63
PEC, 43
per-path pulse distortion, 53
per-path pulse response, 49
per-path pulse waveform distortion, 40
percentage of found connections, 233, 235,
241, 246
perfectly conducting half-plane, 43
perfectly conducting wedge, 44
perimeter forwarding, 209, 211
personal area networks, 23
phantom paths, 80, 81
physics-based generalized multipath
model, 41
plane wave, 45
POCA, 76
Poisson process, 68, 90
polarization diversity, 284
portable base station, 102
positioning information, 213
power delay profile, 72, 256258
power spectral density, 22
power-aware, 207, 208, 249
PPM, 220
printed circuit board, 192
prolate spheroidal wave functions, 164
proximity, 412
proximity location sensors, 415
pseudograting lobes, 343
pseudorandom noise, 185
pseudorange, 414
PSK, 55

496
pulse amplitude modulation, 157
pulse diraction, 41
pulse position modulation, 159, 181, 184
pulse repetition frequency, 183, 185, 187, 195
pulse shape, 85, 150
pulse shape modulation, 160
pulse shaping, 175
PULSERS, 10
Q
QAM, 55
R
radiation pattern, 27, 30
radio channel, 26, 27, 29, 30, 33, 36
radio channel frequency transfer
function, 110
radio location systems, 418
Rake, 87
RAKE combining, 314
Rake receiver, 67, 128
Random Waypoints, 223, 224, 229, 231
ranging, 413
ranging capability, 208, 209, 221
ranging experiments with antenna-array
systems, 437
ranging in harsh environments, 426
ranging tags, 424
ray arrival process, 88, 89, 91
ray tracing, 21, 73
Rayleigh, 74
receive characteristic, 38
reciprocity, 27, 34, 36, 37
reference point group mobility, 223, 224
reference velocity group mobility, 225
reference velocity group mobility model, 223
reflection, 78
relative bandwidth, 22
residential environment, 93
RF antenna connection, 100
RF link budget for location systems, 421
RF Micro Devices Inc., 189
RF on fiber optic, 100
Rice distribution, 74
rms delay spread, 72
RMS location accuracy, 416
round trip time, 413
routing metrics, 207
routing protocols, 209
RPGM, 225, 226, 229231
RPQ, 216
RRC, 240, 242
RRP, 233

Index
RRQ, 214, 215, 233, 246
RVGM, 226, 229231
S
Saleh-Valenzuela, 70, 88
sampling mixer, 189
scalability of location techniques, 425
scattering center, 41
Schottky diode, 189, 190
self-positioning algorithm, 222
semi-integral, 43
semishadow, 44
sensor independence, 445
shadowing, 24, 59, 65
shadowing depth, 92
short-range transmission, 283
signal strength positioning systems, 419
sliding window algorithm, 130
small-size body-worn, 95
smart antenna, 303
smart space applications, 423
space-frequency, 353
space-time block-coding, 286
space-variant, 255, 259, 269
spatial diversity, 302, 353
spatial resolution, 88
spatial variance, 329
spatially resolved measurements, 71
spatiotemporal receive characteristic, 32
spatiotemporal transmit characteristic, 31
spectrum randomisation, 167
ST, 353
standard models for path loss
and shadowing, 67
stochastic channel models, 21
successive cancellation algorithm, 130
surveying techniques, 443
symmetric location systems, 415
synchronisation, 175
T
TD-UTD/GTD, 47
TEM, 84
testbed, 181200
TH codes, 220
TH-CDMA, 221
TH-IR UWB, 221
TH-UWB, 169
time dierence of arrival, 414
time hopping, 353
time of arrival, 413
time of flight, 413
time-delay beamformer, 330
time-hopping impulse radio UWB, 220

Index
time-hopping IR, 125
time-modulated, 3
transfer function, 28, 30, 33, 34, 77,
255, 259, 269
transient response, 270
transmission, 78
trilateration, 412
U
ultra-wideband, 282
uniform phase distribution, 77
uniform spaced linear array, 273
UWB, 1, 21, 353
UWB coexistence, 448
UWB propagation environments, 59
UWB swept time-delay correlation channel
sounder, 110
V
variable gain amplifier, 193, 195
virtual sink, 263
virtual source, 263, 265
Vivaldi, 84
voltage-controlled crystal oscillator, 183, 199
W
wavefront, 46
wavelets, 163
Weibull distribution, 76
Wilkinson power divider, 189, 190, 197
wireless body area networking, 94
X
XILINX Inc., 186
Z
zero forcing, 122

497