You are on page 1of 149

EURASIP Journal on Wireless Communications and Networking

Reconfigurable Radio for Future


Generation Wireless Systems
Guest Editors: Frederik Petr, Ahmet Kondoz,
Stefan Kaiser, and Ashish Pandharipande
Reconfigurable Radio for Future
Generation Wireless Systems
EURASIP Journal on
Wireless Communications and Networking
Reconfigurable Radio for Future
Generation Wireless Systems


Guest Editors: Frederik Petr, Ahmet Kondoz,
Stefan Kaiser, and Ashish Pandharipande
EURASIP Journal on
Wireless Communications and Networking
Copyright 2005 Hindawi Publishing Corporation. All rights reserved.
This is a special issue published in volume 2005 of EURASIP Journal on Wireless Communications and Networking. All articles are
open access articles distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
Editor-in-Chief
Phillip Regalia, Institut National des Telecommunications, France
Associate Editors
Thushara Abhayapala, Australia Fary Ghassemlooy, UK Eric Moulines, France
Farid Ahmed, USA Alfred Hanssen, Norway Sayandev Mukherjee, USA
Alagan Anpalagan, Canada Stefan Kaiser, Germany A. Nallanathan, Singapore
Anthony C. Boucouvalas, UK G. K. Karagiannidis, Greece Kamesh Namuduri, USA
Jonathon Chambers, UK Hyung-Myung Kim, Korea Athina Petropulu, USA
Biao Chen, USA Chi Chung Ko, Singapore H. Vincent Poor, USA
Pascal Chevalier, France Richard J. Kozick, USA Brian Sadler, USA
Chia-Chin Chong, Korea Bhaskar Krishnamachari, USA Ivan Stojmenovic, Canada
Soura Dasgupta, USA Vincent Lau, Hong Kong Lee Swindlehurst, USA
PetarM. Djuric, USA Dave Laurenson, Scotland Sergios Theodoridis, Greece
Abraham Fapojuwo, Canada Tho Le-Ngoc, Canada Lang Tong, USA
Michael Gastpar, USA Tongtong Li, USA Luc Vandendorpe, Belgium
Alex B. Gershman, Canada Wei (Wayne) Li, USA Yang Xiao, USA
Wolfgang Gerstacker, Germany Steve McLaughlin, UK Lawrence Yeung, Hong Kong
David Gesbert, France Marc Moonen, Belgium Weihua Zhuang, Canada
Contents
Editorial, Frederik Petr, Ahmet Kondoz, Stefan Kaiser, and Ashish Pandharipande
Volume 2005 (2005), Issue 3, Pages 271-274
Software-Defined RadioBasics and Evolution to Cognitive Radio, Friedrich K. Jondral
Volume 2005 (2005), Issue 3, Pages 275-283
Flexible Radio: A Framework for Optimized Multimodal Operation via Dynamic Signal Design,
Ioannis Dagres, Andreas Zalonis, Nikos Dimitriou, Konstantinos Nikitopoulos,
and Andreas Polydoros
Volume 2005 (2005), Issue 3, Pages 284-297
Adaptive Transmitter Optimization in Multiuser Multiantenna Systems: Theoretical Limits, Effect of
Delays, and Performance Enhancements, Dragan Samardzija, Narayan Mandayam,
and Dmitry Chizhik
Volume 2005 (2005), Issue 3, Pages 298-307
Flexible Transmission Scheme for 4G Wireless Systems with Multiple Antennas, Franois Horlin,
Frederik Petr, Eduardo Lopez-Estraviz, Frederik Naessens, and Liesbet Van der Perre
Volume 2005 (2005), Issue 3, Pages 308-322
Reconfigurable Signal Processing and Hardware Architecture for Broadband Wireless
Communications, Ying-Chang Liang, Sayed Naveen, Santosh K. Pilakkat, and Ashok K. Marath
Volume 2005 (2005), Issue 3, Pages 323-332
Modular Software-Defined Radio, Arnd-Ragnar Rhiemeier
Volume 2005 (2005), Issue 3, Pages 333-342
Adaptive Mobile Positioning in WCDMA Networks, B. Dong and Xiaodong Wang
Volume 2005 (2005), Issue 3, Pages 343-353
Flexible Frequency Discrimination Subsystems for Reconfigurable Radio Front Ends,
Bruce E. Carey-Smith, Paul A. Warr, Phill R. Rogers, Mark A. Beach, and Geoffrey S. Hilton
Volume 2005 (2005), Issue 3, Pages 354-363
Flexible Analog Front Ends of Reconfigurable Radios Based on Sampling and Reconstruction with
Internal Filtering, Yefim S. Poberezhskiy and Gennady Y. Poberezhskiy
Volume 2005 (2005), Issue 3, Pages 364-381
A Reconfigurable Spiral Antenna for Adaptive MIMO Systems, Bedri A. Cetiner, J. Y. Qian, G. P. Li,
and F. De Flaviis
Volume 2005 (2005), Issue 3, Pages 382-389
Multimode Communication Protocols Enabling Reconfigurable Radios, Lars Berlemann, Ralf Pabst,
and Bernhard Walke
Volume 2005 (2005), Issue 3, Pages 390-400
Towards a Fraud-Prevention Framework for Software Defined Radio Mobile Devices,
Alessandro Brawerman and John A. Copeland
Volume 2005 (2005), Issue 3, Pages 401-412
EURASIP Journal on Wireless Communications and Networking 2005:3, 271274
c 2005 Hindawi Publishing Corporation
Editorial
Frederik Petr e
Wireless Research, Interuniversity Micro-Electronics Center (IMEC), Kapeldreef 75, 3001 Leuven, Belgium
Email: frederik.petre@imec.be
Ahmet Kondoz
Centre for Communication Systems Research (CCSR), University of Surrey, Guildford GU2 7XH, UK
Email: a.kondoz@surrey.ac.uk
Stefan Kaiser
DoCoMo Communications Laboratories Europe GmbH, Landsberger Str. 312, 80687 Munich, Germany
Email: kaiser@docomolab-euro.com
Ashish Pandharipande
Communication and Networking Lab, Samsung Advanced Institute of Technology, P.O. Box 111, Suwon 440-600, Korea
Email: p.ashish@samsung.com
1. BACKGROUND
Future-generation wireless systems aim to support a spec-
trum of services over a variety of networks in a way transpar-
ent to the user. Flexibility and adaptivity are key ingredients
of such future-generation wireless systems in order to deliver
optimal quality of service (QoS) for dierent applications
over diverse communication environments. Rather than re-
lying on the traditional horizontal communication model,
consisting of a single wireless access system, these future 4G
systems will employ a vertical communication model, which
integrates dierent existing and new evolving wireless access
systems on a common IP-based platform, to complement
each other for dierent service requirements and radio envi-
ronments. To enable seamless and transparent interworking
between these dierent wireless access systems, or communi-
cation modes, through horizontal (intrasystem) and vertical
(intersystem) handovers, multimode functionality is needed
to support the dierent existing air interfaces and the newly
emerging ones.
It is expected that multimode capabilities will be ulti-
mately focussed on the terminal side to target a larger mar-
ket base. New challenges then appear in terms of minimiz-
ing the terminal cost, size, and power consumption, while
at the same time maximizing its exibility with respect to
communication standards as well as its adaptivity with re-
spect to varying user requirements and changing communi-
cation conditions. The conventional approach to the design
of a multimode terminal is the provision of a custom base-
band processor for every communication mode. However,
with the growing number of standards and communication
modes, this approach is becoming increasingly infeasible and
economically unacceptable. A more ecient approach to-
wards this design is to adopt a recongurable (as opposed to
xed) radio concept, such that the terminal can adapt to the
best-suited communication mode under the control of a QoS
manager. A high degree of exibility is required not only for
the digital baseband processing but also for the analog radio
frequency (RF) front end, which should accept a large range
of carrier frequencies, possess a exible bandwidth, and deal
with a wide variety of operational conditions. Likewise, the
same high degree of exibility is called for not only at the
physical layer but also at the medium access control (MAC)
(and possibly higher) layer(s), to be compatible with the pro-
tocols of the dierent standards.
2. OVERVIEWOF THE SPECIAL ISSUE
This special issue, which has been conceptualized within the
framework of the IST-FP6 Network of Excellence in Wire-
less COMmunications (NEWCOM), and, more specically,
within the context of NEWCOM Project D on Flexible Ra-
dio, contains 3 invited papers and 9 regular papers.
The rst (invited) paper Software-dened radio
Basics and evolution to cognitive radio, by F. K. Jondral,
reviews the basic concepts and terminology of software-
dened radio (SDR) and discusses its future evolution
272 EURASIP Journal on Wireless Communications and Networking
towards cognitive radio. The author further emphasizes the
importance of standardization and introduces the so-called
software communications architecture (SCA) as an exam-
ple framework that allows an object-oriented development
of SDRs.
2.1. Flexible baseband processing
The second (invited) paper Flexible radio: A framework
for optimized multimodal operation via dynamic signal de-
sign, by I. Dagres et al., introduces a general framework
for the study and design of exible/recongurable radio sys-
tems, with a special focus on the baseband portion of the
physical layer and its interactions with procedures taking
place in the higher layers. Furthermore, the authors describe
specic tools and fundamentals that underpin such exible
transceiver architectures to provide multistandard capabili-
ties, channel adaptivity, and user/service personalization.
The third (invited) paper Adaptive transmitter opti-
mization in multiuser multiantenna systems: Theoretical
limits, eect of delays, and performance enhancements, by
D. Samardzija et al., considers optimum linear precoders for
multiantenna, multiuser systems. Optimality is considered
in terms of maximizing the sum rate capacity subject to an
average transmitter power constraint. Performance limits of
the proposed schemes under channel prediction and delayed
feedback are presented.
The fourth paper Flexible MIMO transmission scheme
for 4G wireless systems with multiple antennas, by F. Hor-
lin et al., presents a generic transmission scheme that allows
to instantiate combinations of OFDM and cyclic-prexed
single-carrier modulation schemes with DS-CDMA. Addi-
tionally, space-division multiplexing (SDM) and orthogonal
space-time block coding (STBC) have been integrated in the
generic transmission scheme. For each resulting mode, the
optimal linear MMSE multiuser receiver has been derived. A
mode selection strategy has also been proposed that trades
o eciently the communication performance in a typical
suburban dynamic outdoor environment with the complex-
ity and PAPR at the mobile terminal.
The fth paper Recongurable signal processing and
hardware architecture for broadband wireless communica-
tions, by Y.-C. Liang et al., proposes a exible baseband
transceiver, which can be recongured to any type of cyclic-
prex-based communication scheme. In addition, the au-
thors introduce a corresponding recongurable hardware
architecture, and identify the common blocks that can be
reused across the dierent communication schemes. Finally,
they recognize that the major challenge is to have an ecient
systemconguration and management function that will ini-
tiate and control the reconguration based on user require-
ments and channel conditions.
The sixth paper Modular software-dened radio, by A.-
R. Rhiemeier, proposes a model of signal processing software
including irregular, connected, directed, acyclic graphs with
random node weights and random edges. Several approaches
for mapping such software to a given hardware are discussed.
Taking into account previous ndings as well as new re-
sults fromsystemsimulations presented, the paper concludes
on the utility of pipelining as a general design guideline for
modular software-dened radio.
The seventh paper Adaptive mobile positioning in
WCDMA networks, by B. Dong and X. Wang, introduces
a technique for mobile tracking in wideband code-division
multiple-access (WCDMA) systems employing multiple re-
ceive antennas. To achieve a high estimation accuracy, the al-
gorithm utilizes the time dierence of arrival (TDOA) mea-
surements in the forward link pilot channel, the angle of ar-
rival (AOA) measurements in the reverse-link pilot channel,
as well as the received signal strength. The proposed algo-
rithm jointly tracks the unknown system parameters as well
as the mobile position and velocity.
2.2. Flexible analog RF front ends
The eighth paper Flexible frequency discrimination sub-
systems for recongurable radio front ends, by B. Carey-
Smith et al., surveys recent advances in exible, frequency-
selective, circuit components (including bandpass and band-
stop lters, and narrowband tunable antennas) applicable to
software-dened radio front ends. In this perspective, the au-
thors discuss the ltering requirements in the SDR context
and advocate the use of intelligent, adaptive control to pro-
vide environment-aware frequency discrimination.
The ninth paper Flexible analog front ends of recon-
gurable radios based on sampling and reconstruction with
internal ltering, by Y. Poberezhskiy and G. Poberezhskiy,
pursues several ways to overcome the challenges of practi-
cal realization and implementation of novel sampling and
reconstruction techniques with internal ltering. In this per-
spective, the impact of these novel techniques on the analog
front-end architectures and capabilities of software-dened
radios is discussed.
The tenth paper A recongurable spiral antenna for
adaptive MIMO systems, by B. Cetiner et al., studies the de-
sign of spiral antennas that are recongurable in the sense
that they can alter antenna characteristics through structural
change. In their work, the authors propose a recongurable
spiral antenna architecture based on RF-MEMS technology.
The presented technology allows monolithic integration of
RF-MEMS with antenna structures on any microwave lami-
nate substrate, with the capability to change the impedance
and radiation characteristics of the antenna. As a reference
model, the design, fabrication, and characterization of con-
ventional single-arm rectangular spiral antennas radiating
circularly polarized elds along their axes are presented in
the paper.
2.3. Flexible MAC and higher-layer protocols
The eleventh paper Multimode communication protocols
enabling recongurable radios, by L. Berlemann et al., pro-
poses a generic protocol stack, comprising common pro-
tocol functionality for recongurable wireless communica-
tion systems. More specically, the proposed generic proto-
col stack contains parameterizable modules of basic proto-
col functions that reside in the data link layer and the net-
work layer of the open systems interconnect (OSI) model.
Editorial 273
It is demonstrated that the presented parameterizable mod-
ules can be regarded as a toolbox for the timely and cost-
ecient development of future communication protocols.
The twelfth paper Towards a fraud-prevention frame-
work for software dened radio devices, by A. Brawerman
and J. Copeland, considers a framework for security en-
hancement in mobile SDR devices through the introduction
of new hardware units and protocols. The presented frame-
work oers enhanced security by incorporating features like
monitoring against malicious attacks and viruses, authen-
tication, critical information-protection, and anticloning.
Proofs and experimental results are also given to validate the
presented fraud-prevention framework.
ACKNOWLEDGMENTS
Many people deserve our gratitude for helping us to put to-
gether this special issue. First of all, we wish to express our
gratitude to the Editor-in-Chief, Phil Regalia, for giving us
the opportunity and the support to realize this special is-
sue within the context of the IST FP6 Network of Excellence
in Wireless COMmunications (NEWCOM). Naturally, we
would like to thank the authors of the regular papers for their
valuable and timely contributions. We are also grateful to the
authors of the three invited papers: Friedrich Jondral, An-
dreas Polydoros and his coauthors, and Narayan Mandayam
and his coauthors. Fianlly, our appreciation goes to the many
obliging reviewers, without themour decision making would
have been impossible.
Frederik Petre
Ahmet Kondoz
Stefan Kaiser
Ashish Pandharipande
Frederik Petr e was born in Tienen, Bel-
gium, on December 12, 1974. He received
the E.E. degree and the Ph.D. degree in
applied sciences from the Katholieke Uni-
versiteit Leuven (KULeuven), Leuven, Bel-
gium, in July 1997 and December 2003,
respectively. In September 1997, he joined
the Design Technology for Integrated In-
formation and Communication Systems
(DESICS) Division at the Interuniversity
Micro-Electronics Center (IMEC), Leuven, Belgium. Within the
Digital Broadband Terminals (DBATE) Group of DESICS, he
rst performed predoctoral research on wireline transceiver de-
sign for twisted pair, coaxial cable, and power-line communi-
cations. During the fall of 1998, he visited the Information
Systems Laboratory (ISL), Stanford University, California, USA,
working on OFDM-based power-line communications. In Jan-
uary 1999, he joined the Wireless Systems (WISE) Group of
DESICS as a Ph.D. researcher, funded by the Institute for Sci-
entic and Technological Research in Flanders (IWT). Since Jan-
uary 2004, he has been a Senior Scientist within the Wireless
Research Group of DESICS. He is investigating the baseband sig-
nal processing algorithms and architectures for future wireless
communication systems, like third-generation (3G) and fourth-
generation (4G) cellular networks, and wireless local area net-
works (WLANs). His main research interests are in modulation
theory, multiple access schemes, channel estimation and equaliza-
tion, smart antenna, and MIMO techniques. He is a Member of the
ProRISC Technical Program Committee and the IEEE Benelux Sec-
tion on Communications and Vehicular Technology (CVT). He is
a Member of the Executive Board and Project Leader of the Flexible
Radio project of the Network of Excellence in Wireless COMmuni-
cations (NEWCOM), established under the sixth framework of the
European Commission.
Ahmet Kondoz after receiving his Ph.D. de-
gree in 1987 from the University of Surrey,
he became a lecturer in 1988, reader in 1995,
and in 1996 he was promoted to Professor
in multimedia communication systems. He
took part in the formation of the Centre for
Communication Systems Research (CCSR)
and led the multimedia communication ac-
tivities within CCSR. He is the Founding
Director of I-Lab, the new multidisciplinary
media lab. Professor Kondozs current research interests are in
the areas of digital signal, image/video, speech/audio processing
and coding, wireless multimedia communications, error resilient
media transmission, immersive/virtual/augmented environments,
and the related human factors issues including human computer
interaction/interface. He has published more than 250 journal and
conference papers, two books, and 7 patents. Professor Kondoz has
been involved in ETSI, ITU, INMARSAT, and NATO standardiza-
tions of low-bit-rate speech communication activities. He is the
Managing Director of MulSys Limited, a UniS spin-out company
marketing worlds rst secure voice product over the GSM voice
channel which has been pioneered by Professor Kondozs team in
the I-Lab. Professor Kondoz has been awarded several prizes, the
most signicant of which are the Royal Television Societies Com-
munications Innovation Award and the IEE Benefactors Premium
Award.
Stefan Kaiser received the Dipl.-Ing. and
Ph.D. degrees in electrical engineering from
the University of Kaiserslautern, Germany,
in 1993 and 1998, respectively. From 1993
to 2005, he was with the Institute of Com-
munications and Navigation of the Ger-
man Aerospace Center (DLR), Oberpfaf-
fenhofen, Germany, where he was Head of
the Mobile Radio Transmission Group. He
worked on multicarrier communications
(OFDM and MC-CDMA) for future wireless systems and digital
terrestrial video broadcasting (DVB-T). In 1998, he was a Visiting
Researcher at the Telecommunications Research Laboratories (TR-
Labs), Edmonton, Canada, working in the area of wireless com-
munications. Since 2005, he has been with DoCoMo Communica-
tions Laboratories Europe GmbH, Munich, Germany, where he is
Head of the Wireless Solution Laboratory. His research interests in-
clude among others multicarrier communications, multiple-access
schemes, and space-time processing for future mobile radio appli-
cations (B3G, 4G).
274 EURASIP Journal on Wireless Communications and Networking
Ashish Pandharipande was born in India
in 1977. He received the B.Eng. degree in
electronics and communications engineer-
ing from the College of Engineering, Osma-
nia University, Osmania, India, in 1998. He
pursued his graduate education at the Uni-
versity of Iowa, Iowa City, where he received
the M.S. degrees in electrical and computer
engineering and mathematics in 2000 and
2001, respectively, and the Ph.D. degree in
electrical and computer engineering in 2002. He was with the
Electrical and Computer Engineering Department, University of
Florida, Gainesville, as a Postdoctoral Researcher in 2003. He is cur-
rently a Senior Researcher in the technical research sta at Samsung
Advanced Institute of Technology, Korea. His research interests are
in the areas of multicarrier (OFDM) and MIMO wireless commu-
nications, recongurable and cognitive wireless systems, multirate
signal processing, and signal processing techniques in communica-
tions.
EURASIP Journal on Wireless Communications and Networking 2005:3, 275283
c 2005 Friedrich K. Jondral
Software-Dened RadioBasics and Evolution
to Cognitive Radio
Friedrich K. Jondral
Universit at Karlsruhe (TH), Institut f ur Nachrichtentechnik, D-76128 Karlsruhe, Germany
Email: fj@int.uni-karlsruhe.de
Received 24 February 2005; Revised 4 April 2005
We provide a brief overview over the development of software-dened or recongurable radio systems. The need for software-
dened radios is underlined and the most important notions used for such recongurable transceivers are thoroughly dened.
The role of standards in radio development is emphasized and the usage of transmission mode parameters in the construction
of software-dened radios is described. The software communications architecture is introduced as an example for a framework
that allows an object-oriented development of software-dened radios. Cognitive radios are introduced as the next step in radio
systems evolution. The need for cognitive radios is exemplied by a comparison of present and advanced spectrum management
strategies.
Keywords and phrases: software-dened radio, recongurable transceiver, mobile communication standards, cognitive radio,
advanced spectrum management.
1. INTRODUCTION
Recongurability in radio development is not such a new
technique as one might think. Already during the 1980s re-
congurable receivers were developed for radio intelligence
in the short wave range. These receivers included interesting
features like automatic recognition of the modulation mode
of a received signal or bit stream analysis. Recongurability
became familiar to many radio developers with the publica-
tion of the special issue on software radios of the IEEE Com-
munication Magazine in April 1995.
We refer to a transceiver as a software radio (SR) if its
communication functions are realized as programs running
on a suitable processor. Based on the same hardware, dier-
ent transmitter/receiver algorithms, which usually describe
transmission standards, are implemented in software. An SR
transceiver comprises all the layers of a communication sys-
tem. The discussion in this paper, however, mainly concerns
the physical layer (PHY).
The baseband signal processing of a digital radio (DR) is
invariably implemented on a digital processor. An ideal SR
directly samples the antenna output. A software-dened ra-
dio (SDR) is a practical version of an SR: the received signals
are sampled after a suitable band selection lter. One remark
This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
concerning the relation between SRs and SDRs is necessary at
this point: it is often argued that an SDR is a presently realiz-
able version of an SR since state-of-the-art analog-to-digital
(A/D) converters that can be employed in SRs are not avail-
able today. This argument, although it is correct, may lead to
the completely wrong conclusion that an SR which directly
digitizes the antenna output should be a major goal of future
developments. Fact is that the digitization of an unnecessary
huge bandwidth lled with many dierent signals of which
only a small part is determined for reception is neither tech-
nologically nor commercially desirable.
1
However, there is no
reason for a receiver to extremely oversample the desired sig-
nals while respecting extraordinary dynamic range require-
ments for the undesired in-band signals at the same time.
Furthermore, the largest portion of the generated digital in-
formation, which stems from all undesired in-band signals,
is ltered out in the rst digital signal processing step.
A cognitive radio (CR) is an SDR that additionally senses
its environment, tracks changes, and reacts upon its ndings.
A CR is an autonomous unit in a communications environ-
ment that frequently exchanges information with the net-
works it is able to access as well as with other CRs. From our
point of view, a CR is a rened SDR while this again repre-
sents a rened DR.
1
This is not an argument against the employment of multichannel or
wideband receivers.
276 EURASIP Journal on Wireless Communications and Networking
R
e
c
e
i
v
e
T
r
a
n
s
m
i
t
Radio
frequency
(RF)
Analog-to-digital
conversion
(A/D)
Baseband
processing
Data
processing
Control
(parameterization)
Radio front end
T
o
u
s
e
r
F
r
o
m
u
s
e
r
Figure 1: SDR transceiver.
According to its operational area an SDR can be
(i) a multiband system which is supporting more than one
frequency band used by a wireless standard (e.g., GSM
900, GSM 1800, GSM 1900),
(ii) a multistandard system that is supporting more than
one standard. Multistandard systems can work within
one standard family (e.g., UTRA-FDD, UTRA-TDD
for UMTS) or across dierent networks (e.g., DECT,
GSM, UMTS, WLAN),
(iii) a multiservice system which provides dierent services
(e.g., telephony, data, video streaming),
(iv) a multichannel system that supports two or more in-
dependent transmission and reception channels at the
same time.
Our present discussion is on multimode systems which are
combinations of multiband and multistandard systems.
The SDR approach allows dierent levels of recongura-
tion within a transceiver.
(i) Commissioning: the conguration of the system is
done once at the time of product shipping, when the
costumer has asked for a dedicated mode (standard or
band). This is not a true reconguration.
(ii) Reconguration with downtime: reconguration is only
done a few times during product lifetime, for example,
when the network infrastructure changes. The recon-
guration will take some time, where the transceiver is
switched o. This may include the exchange of com-
ponents.
(iii) Reconguration on a per call basis: reconguration is
a highly dynamic process that works on a per call de-
cision. That means no downtime is acceptable. Only
parts of the whole system (e.g., front-end, digital base-
band processing) can be rebooted.
(iv) Reconguration per timeslot: reconguration can even
be done during a call.
Figure 1 shows an SDR transceiver that diers from a
conventional transceiver only by the fact that it can be recon-
gured via a control bus supplying the processing units with
the parameters which describe the desired standard. Such
a conguration, called a parameter-controlled (PaC) SDR,
guarantees that the transmission can be changed instanta-
neously if necessary (e.g., for interstandard handover).
The rest of this paper is organized as follows. In Section 2
we take a look at the most important wireless transmis-
sion standards currently used in Europe and specify their
main parameters. Section 3 provides an overview of design
approaches for mobile SDR terminals, especially over PaC-
SDRs. In Section 4 the software communications architec-
ture (SCA), as it is used in the US Joint Tactical Radio System
(JTRS), is introduced. The notion of cognitive radio (CR)
is discussed in Section 5 and the need for a modied spec-
trum management in at least some major portions of the
electromagnetic spectrum is underlined in Section 6. Finally,
in Section 7 we propose the development of technology cen-
tric CRs as a rst step towards terminals that may sense their
environment and react upon their ndings. Conclusions are
drawn in Section 8.
2. MOBILE COMMUNICATIONSTANDARDS
Standards are used to publicly establish transmission meth-
ods that serve specic applications employable for mass mar-
kets. The presently most important mobile communication
standards used in Europe are briey described in the follow-
ing paragraphs.
Personal area networks
Bluetooth is a short distance network connecting portable
devices, for example, it enables links between computers,
mobile phones or connectivity to the internet.
Cordless phone
DECT (digital enhanced cordless telecommunications) pro-
vides a cordless connection of handsets to the xed telephone
system for in-house applications. Its channel access mode is
FDMA/TDMA and it uses TDD. The modulation mode of
DECT is Gaussian minimum shift keying (GMSK) with a
bandwidth (B) time (T) product of BT = 0.5. The transmis-
sion is protected only by a cyclic redundancy check (CRC).
Wireless local area networks
Today, IEEE 802.11b installations are the most widely used
in Europe. Also, IEEE 802.11a systems are in operation. If
IEEE.11a is to be implemented into an SDR, it should be
recognized that its modulation mode is OFDM. It should
be pointed out here that there are major eorts towards the
development of joint UMTS/WLAN systems which use the
SDR approach.
Cellular systems
GSM (global system for mobile communication) is presently
the most successful mobile communication standard world-
wide. Channel access is done via FDMA/TDMA and GSM
uses FDD/TDD. The modulation mode of GSM is GMSK
with a bandwidth time product of BT = 0.3. Error correction
coding is done by applying CRC as well as a convolutional
code. GSM was originally planned to be a voice communi-
cation system, but with its enhancements HSCSD, GPRS, or
EDGE, it served more and more as a data system, too. In Eu-
rope, GSM systems are operating in the 900 MHz (GSM 900)
SDRBasics and Evolution to Cognitive Radio 277
800 900 1000 1100 1200 1300 1400 1500 1600
8
9
0
9
1
5
9
3
5
9
6
0
1600 1700 1800 1900 2000 2100 2200 2300 2400 2500
1
7
1
0
1
7
8
5
1
8
0
5
1
8
8
0
1
9
2
0
1
9
8
0
2
0
1
0
2
0
2
5
2
1
1
0
2
1
7
0
2
4
8
3
.
5
5100 5200 5300 5400 5500 5600 5700 5800 5900
5
1
5
0
5
3
5
0
5
4
7
0
5
7
2
5
GSM
DECT
UTRA-TDD
UTRA-FDD
MSS
ISM
WLAN
f (MHz)
Figure 2: Mobile spectrum in Europe.
as well as in the 1800 MHz (GSM 1800) bands. The North
American equivalent of GSM is IS-136. Also, GSM 1900 as
well as IS-95, a second-generation CDMA system, are widely
used in the US. UMTS (universal mobile telecommunication
system) is the European version of the third-generation fam-
ily of standards within IMT-2000. One of the dierences with
respect to second-generation systems is that third-generation
systems are mainly developed for data (multimedia) trans-
mission. UMTS applies two air interfaces: UTRA-FDD and
UTRA-TDD according to the duplex modes used. The chan-
nel access mode is CDMA. CRC, convolutional codes, as well
as turbo codes [1] are employed for error protection. The
basic data modulation is QPSK. Furthermore, it should be
mentioned that one mobile user within an UTRA-FDD cell
can occupy up to seven channels (one control and six trans-
port channels) simultaneously.
Figure 2 gives an overview over the present spectrum al-
location for mobile communications in Europe. Besides the
spectra of the standards mentioned above, also the spectra
allocated to mobile satellite system (MSS) as well as to indus-
trial, scientic, and medical (ISM) applications are specied.
The arrows within some of the bands indicate whether uplink
(mobile to base station) or downlink (base station to mobile)
trac is supported.
In connection with mobile communications, some addi-
tional groups of standards have to be discussed.
Professional mobile radio
PMR standards are developed for police, reghters, and
other administrative applications. The main dierence to cel-
lular systems is that they allow direct handheld to hand-
held communication. The main PMR systems in Europe are
TETRA (recommended by ETSI) and TETRAPOL.
Location and navigation
One important feature of mobile terminals is their ability to
determine their own location as well as to track location in-
formation. Today many location-dependent services rely on
the global positioning system (GPS). Currently the European
satellite location and navigation system Galileo is under de-
velopment.
Digital broadcast
There is a possibility that digital broadcast systems may be
used as downstreaming media within future mobile commu-
nication infrastructures. The main developments in Europe
in this area are digital audio broadcast (DAB) and digital
video broadcast (DVB).
To have a sound basis for the description of a PaC-SDR
that can be switched between dierent standards, the most
important parameters of selected air interfaces are summa-
rized in Table 1.
3. MOBILE SDR TERMINALS
The general structure of a PaC-SDR terminal was already
given in Figure 1. Nowwe are going to look into the PaC-SDR
transceiver structure in a more detailed way. The main pro-
cessing modules of an SDR terminal are the radio front-end,
the baseband processing, and the data processing. Since a lot
of information about baseband processing can be found in
the literature [2, 3] and since data processing is out of the
scope of this paper, we are going to focus on the front-end
here.
The receiver branch transforms the analog RF antenna
signal into its digital complex baseband representation.
278 EURASIP Journal on Wireless Communications and Networking
Table 1: Parameters of selected air interfaces.
Bluetooth DECT GSM UTRA-FDD
Frequency range 2.4 GHz (ISM band) 1900 MHz 900, 1800, 1900 MHz 2 GHz
Channel bandwidth 1 MHz 1728 kHz 200 kHz 5 MHz
Access mode TDMA FDMA/TDMA FDMA/TDMA
Direct sequence (DS)
CDMA
Duplex mode TDD TDD FDD FDD
Users per carrier frequency 8 maximum 12 8
Modulation
FH sync. to master station,
GMSK GMSK QPSK GFSK with modulation
index between 0.28 and 0.35
Error correction code No (CRC) CRC, convolutional CRC, convolutional, turbo
Bit (chip) rate 1 Mbps 1152 kbps 270.833 kbps 3.840 Mchip/s
Number of bits (chips)/burst
(slot)
625 480 (DECT P32) 156.25 2560
Frame duration 10 ms 4.615 ms 10 ms
Number of bursts
(slots)/frame

24 8 15
Burst (slot) duration 0.625 ms 0.417 ms 0.577 ms 0.667 ms
Maximum cell radius 510 m (1 mW Tx power) 300 m 36 km (10 km) Few km
Spreading sequences
User specic OVSF codes,
call specic scrambling
Spreading factor
2
k
(k = 2, 3, . . . , 8), 512
for downlink only
Bit (chip) pulse shaping
Gauss (BT = 0.5) Gauss (BT = 0.5) Gauss (BT = 0.3)
Root-raised cosine,
lter roll-o factor 0.22
Net data rate 1 Mbps 26 kbps 13 kbps 8 kbps to 2 Mbps
Evolutionary concepts UWB GPRS, HSCSD, EDGE HSDPA
Comparable systems PHS, PACS, WACS IS-136, PDC UMTS-TDD. Cdma2000
TETRA IEEE 802.11a GPS DVB-T
Frequency range 400 MHz 5.5 GHz 1200, 1500 MHz VHF, UHF
Channel bandwidth 25 kHz 20 MHz 7 (VHF) or 8 MHz (UHF)
Access mode TDMA FDMA/TDMA
Direct sequence spread
spectrum
FDMA
Duplex mode FDD/TDD Half duplex
Users/carrier
4

frequency
Modulation /4-DQPSK
OFDM with subcarrier
BPSK, QPSK
OFDM with subcarrier
modulation modulation
BPSK/QPSK/16QAM/64QAM QPSK/16QAM/64QAM
Error correction code CRC, Reed-Muller, RCPC Convolutional Reed-Solomon, convolutional
Bit (chip) rate 36 kbps 6/9/12/18/24/36/48/54 Mbps 50 bps
9.143 Msamples/s for an
8 MHz channel
Number of bits (chips)
510 (255 symbols)
52 modulated symbols per

2k mode: 2048 + guard int.


per burst (slot) OFDM symbol 8k mode: 8192 + guard int.
Frame duration 56.67 ms Packets of several 100 s 15 s (7500 bit) 68 OFDM symbols
Number of bursts
4 Variable 5 subframes 68
(slots) per frame
Burst (slot) duration 14.167 ms
1 OFDM symbol of 3.3 s +
30 s
2k mode: 224 s + guard time
0.8 s guard time 8k mode: 896 s + guard time
Maximum cell radius Some 10 m
Spreading sequences Gold or PRN code
Spreading factor 1023 or 10 230
Bit (chip) pulse Root-raised cosine,

Rectangular, other ltering
shaping lter roll-o factor 0.35 possible
Net data rate Up to 28.8 kbps Up to 25 Mbps 49.8131.67 Mbps
Evolutionary concepts IEEE 802.11n Galileo
Comparable systems TETRAPOL HiperLAN/2 GLONASS DAB
SDRBasics and Evolution to Cognitive Radio 279
RF

RF

A/D
A/D
I
/
Q
b
a
l
a
n
c
i
n
g
S
a
m
p
l
i
n
g
r
a
t
e
a
d
a
p
t
a
t
i
o
n
Inphase
component
Quadrature
component
Parameter control
Figure 3: SDR/CR receiver front-end.
Figure 3 shows how it works: coming from the antenna, the
RF signal is rst bandpass ltered and then amplied. Fol-
lowing a two-way signal splitter, the next step is an analog
mixing with the locally generated RF frequency in the in-
phase (I) path and with the same frequency phase shifted by
/2 in the quadrature (Q) path. Afterwards, the I and Q
components of the signal are lowpass ltered and A/D con-
verted. The sampling rate of the A/D converters should be
xed for all signals and has to be chosen in such a way that the
conditions of Shannons sampling theorem are fullled for
the broadest signal to be processed. Before the sampling rate
can be adapted to the signals standard, the impairments of
the two-branch signal processing that come from the analog
mixers and lters as well as from the A/D converters them-
selves have to be corrected [4].
The reason for the Sampling rate adaptation is that the
signal processor should work at the minimum possible rate.
For a given standard, this minimum sampling rate depends
on f
c
= 1/T
c
, the symbol or chip rate, respectively. Usually a
sampling rate of f
s
= 4 f
c
is sucient for the subsequent sig-
nal processing where, after the precise synchronization, the
sampling rate may be reduced once more by a factor of 4.
If the fraction of the sampling rates at the adaptors output
and input is rational (or may be suciently close approxi-
mated by a rational number), the sampling rate adaptation
can be implemented by an increasing of the sampling rate
followed by an interpolation lowpass lter and a decreasing
of the sampling rate. If the interpolation lowpass is imple-
mented by an FIR lter, the impulse response usually be-
comes quite long. The solution is to take the up and down
sampling into account within the lter process. Since the up-
sampled signal is usually generated by the insertion of zeros,
the processing of these zeros can be omitted within the l-
ter. This leads to the polyphase structure of Figure 4. Because
dierent input/output ratios have to be realized for dier-
ent standards, the number of lter coecients that must be
stored may become large. If necessary, a direct computation
of the lter coecients can be more ecient than their ad-
vance storage [5]. After the sampling rate adaptation, the sig-
nal is processed within the complex baseband unit (demod-
ulation and decoding). The SDR data processing within the
higher protocol layers [6] is not considered in the present pa-
per.
z
1
z
1
z
1





+ + +
g
j
g
J+j
g
2J+j
g
(L1)J+j
Figure 4: Polyphase lter for sampling rate adaptation.
The SDR transmitter branch consists of the procedures
inverse to that of the receiver branch. That is, the signal to be
transmitted is generated as a complex baseband signal, from
which, for example, the real part is taken to be shifted to the
(transmission) RF.
For SDRs, recongurability means that the radio is able
to process signals of dierent standards or even signals that
are not standardized but exist in specic applications. One
method to implement recongurability is parameterization
of standards. We look at a communication standard as a set
of documents that comprehensively describe all functions of
a radio system in such a way that a manufacturer can de-
velop terminals or infrastructure equipment on this basis.
Standardization is one necessary condition to make a com-
munication system successful on the market, as exemplied
by GSM. Standardization pertains to all kinds of communi-
cation systems, that is, especially to personal, local, cellular,
or global wireless networks. Of course, a standard has to con-
tain precise descriptions of all the functions of the system.
Especially for a mobile system, both the air interface and the
protocol stack have to be specied. Parameterization means
that every standard is looked upon as one member of a family
of standards [7]. The signal processing structure of the fam-
ily is then developed in such a way that this structure may be
switched by parameters to realize the dierent standards.
When developing an SDR, one has to pay attention to the
fact that there are substantial dierences between the second-
generation FDMA/TDMA standards (GSM or IS-136), the
third-generation CDMA standards (UMTS or cdma2000),
and the OFDM-modulated WLAN standards (IEEE 802.11a
or HiperLAN/2) (cf. Table 1). Within UMTS, spreading at
the transmitter and despreading at the receiver have to be
realized. IFFT and FFT operations are necessary for WLAN
transceivers. Aside from such fundamental dierences, sim-
ilarities among communication standards are predominant.
For example, when looking at the signal processing chains,
we remark that the error correction codes of all the second-
generation standards are very similar: a combination of a
block code for the most important bits and a convolutional
code for the larger part of the voice bits is applied. Channel
coding for data transmission is done by a powerful convolu-
tional code. UTRA, as a third-generation air interface, oers
net data rates of up to 2 Mbps and guarantees BERs, of up
to 10
6
for specic applications. To reach these BERs turbo
codes are employed for data transmission. Of course, within
an SDR all these procedures have to be integrated into a gen-
eral encoding/decoding structure. Also a common modula-
tor/demodulator structure has to be specied. Solutions to
these tasks are given, for example, in [2, 3, 7].
280 EURASIP Journal on Wireless Communications and Networking
4. THE SOFTWARE COMMUNICATIONS
ARCHITECTURE
The Joint Tactical Radio System (JTRS) represents the fu-
ture (mobile) communications infrastructure of the US joint
forces. Introducing JTRS stands for an essential step towards
the unication of radio communication systems, the trans-
parency of services, and the exchangeability of components.
The development of the JTRS is accompanied and supervised
by the US forces Joint Program Oce (JPO).
Development, production, and delivery continue to be
the tasks of competing industrial communications software
and hardware suppliers. An important new aspect added by
the JTRS set-up is that the suppliers are guided to aim for
a most perfect interchangeability of components due to the
supervision function of the JPO. The tool used by the JPO is
the software communications architecture (SCA) [8], an open
framework that prescribes the developing engineers how the
hardware or software blocks have to act together within the
JTRS. The communication devices emerging from this phi-
losophy are clearly SDRs.
A major group of suppliers and developers of communi-
cation software and hardware founded the SDR Forum [9] to
promote their interests. The importance of the SDR Forum,
however, reaches well beyond the application of SDRs in the
JTRS. This is underlined by the SDR Forum membership of
European and Asian industrial and research institutions that
usually are mainly interested in the evolution of commercial
mobile communication networks.
The SCAdescribes howwaveforms are to be implemented
onto appropriate hardware devices. A waveform is dened
by the determination of the lower three layers (network,
data link, physical) of the ISO/OSI model. Therefore, wave-
form is a synonym of standard or air interface. Based on
the waveform denition, a transmission method is com-
pletely determined. The denition of a waveform, there-
fore, lays down the modulation, coding, access, and duplex
modes as well as the protocol structure of the transmission
method.
The SCA denes the software structure of an SDR that
may be usable within the JTRS. The underlying hardware
as well as the software is described in object-oriented terms.
Moreover, the structures of application program interfaces
(APIs) and of the security environment are described. Each
component has to be documented in a generally accessible
form.
The JTRS operating environment (OE) dened in the
SCA consists of three main components:
(i) a real-time operating system,
(ii) a real-time request broker,
(iii) the SCA core framework.
When developing an SCA compliant radio device the
supplier gets the operating system and the CORBA middle-
ware from the commercial market. The core framework as
well as the waveform is developed by him or he also gets it
from the market or (in future) it may be contributed by the
JPO.
The SCA is the description of an open architecture with
distributed components. It strictly separates applications
(waveforms) from the processing platform (hardware, oper-
ating system, object request broker, core framework). It seg-
ments the application functions and denes common inter-
faces for the management and the employment of software
components. It denes common services and makes use of
APIs to support the portability of hardware and software
components and of applications.
The connections between the applications and the core
framework within the SCA are given by the APIs. Standard-
ized APIs are essential in assuring the portability of applica-
tions as well as for the exchangeability of devices. APIs guar-
antee that application and service programs may commu-
nicate with one another, independent of the operating sys-
temand the programming language used. APIs are waveform
specic since uniform APIs for all waveforms would be inef-
cient for implementations with bounded resources. There-
fore, the goal is to have a standard set of APIs for each wave-
form. The single APIs are essentially given by the layers of the
ISO/OSI model.
(i) A PHY API supports initialization and conguration
of the system in non-real-time. In real-time it takes care of
the transformation of symbols (or bits) to RF in the trans-
mitter branch. In the receiver branch it transforms RF signals
to symbols (bits).
(ii) A MAC API supports all the MAC functions of the
ISO/OSI layer model (e.g., timeslot control in TDMA or FEC
control).
(iii) An LLC API makes available an interface for the
waveforms link layer performance (according to the ISO/OSI
layer model: data link services) on component level.
(iv) A network API makes available an interface for the
waveforms network performance on component level.
(v) A security API serves for the integration of data secu-
rity procedures (INFOSEC, TRANSEC).
(vi) An input/output API supports the input and output
of audio, video, or other data.
The security relevant SCAaspects are written down in the
SCA security supplement [8]. The SCA security functions and
algorithms are of course dened with respect to the military
security requirements of JTRS.
5. USER CENTRIC ANDTECHNOLOGY CENTRIC
COGNITIVE RADIOPROPERTIES
The description of CR given by Mitola and Maguire in their
seminal paper [10] mainly focuses on the radio knowledge
representation language (RKRL). CRis looked upon as a small
part of the physical world using and providing information
over very dierent time scales. Equipped with various sen-
sors, a CR acquires knowledge from its environment. Em-
ploying software agents, it accesses data bases and contacts
other sources of information. In this context, CR seems to
become the indispensable electronic aid of its owner. Read-
ing [10] leads to the impression that a CR must be a complex
device that helps to overcome all problems of everyday life, all
the same whether they are recognized by the CRs owner or
SDRBasics and Evolution to Cognitive Radio 281
not. Of course, these visions as well as the recognition cycle for
CRs in [11] are strongly intended to stimulate new research
and development. For a more pragmatic point of view, how-
ever, we approach CR in a dierent way.
The properties of CRs may be divided into two groups:
(i) user centric properties that comprise support func-
tions like nding the address of an appropriate restau-
rant or a movie theater, recommendation of a travel
route, or supervision of appointments,
(ii) technology centric properties like spectrum monitor-
ing, localization, and tracking, awareness of processing
capabilities for the partitioning or the scheduling of
processes, information gathering, and knowledge pro-
cessing.
From our point of view, many of the user centric proper-
ties can be implemented by using queries to data bases. This
type of intelligence can be kept in the networks and activated
by calls. In transceiver development, much more dicult de-
sign choices need to be made to realize the wanted technol-
ogy centric properties of a CR. Therefore, we concentrate on
the latter in the following sections.
6. THE NEEDFOR ADVANCEDSPECTRUM
MANAGEMENT
Today, spectrumis regulated by governmental agencies. Spec-
trum is assigned to users or licensed to them on a long-
term basis normally for huge regions like countries. Doing
this, resources are wasted, because large-frequency regions
are used very sporadically. The vision is to assign appropriate
resources to end users only as long as they are needed for a
geographically bounded region, that is, a personal, local, re-
gional, or global cell. The spectrum access is then organized
by the network, that is, by the users. First examples for self-
regulation in mobile radio communications are to be found
in the ISM (24002483.5 MHz) and in the WLAN (5150
5350 MHz and 54705725 MHz) bands.
Future advanced spectrum management will comprise
[12] the following.
(i) Spectrum reallocation: the reallocation of bandwidth
from government or other long-standing users to new
services such as mobile communications, broadband
internet access, and video distribution.
(ii) Spectrum leases: the relaxation of the technical and
commercial limitations on existing licensees to use
their spectrum for new or hybrid (e.g., satellite and
terrestrial) services and granting most mobile radio li-
censees the right to lease their spectrum to third par-
ties.
(iii) Spectrum sharing: the allocation of an unprecedented
amount of spectrum that could be used for unlicensed
or shared services.
If we look upon the users behavior in an FDMA/TDMA
system over the time/frequency plane (cf. Figure 5), we
may nd out that a considerable part of the area remains
f
0
f
u
F
r
e
q
u
e
n
c
y
0
Time
Figure 5: FDMA/TDMA signals over the time/frequency plane,
spectrum pool.
unused [12, 13]. This unused area marks the pool from
which frequencies can be allocated to secondary users (SUs),
for example, in a hotspot. In the following we denote the
FDMA/TDMA users as primary users (PUs). In order to
make the implementation of the SUs system into the PUs
system feasible, two main assumptions should be fullled:
(i) the PUs system is not disturbed by the SUs system,
(ii) the PUs system remains unchanged (i.e., all signal
processing that has to be done to avoid disturbances
of the PUs communications must be implemented in
the SUs system).
Now we assume that the transmission method within the
SUs system is OFDM. Figure 6 gives a brief overview over an
OFDM transmitter: the sequential data stream is converted
to a parallel stream, the vectors of which are interpreted as
signals in the frequency domain. By applying an inverse fast
Fourier transform (IFFT), these data are transformed into
the time domain and sent over the air on a set of orthogonal
carriers with separation f on the frequency axis. If some
carriers should not be used, it is necessary to transmit zeros
on these carriers. This is the strategy to protect the PUs sys-
tem from disturbances originating from the SUs system. In
order to make the SUs system work, the following problems
have to be solved.
(i) The reliable detection of upcoming PUs signals
within an extremely short time interval. (This means that
the detection has to be performed with a very high detection
probability ensuring a moderate false alarm probability.)
(ii) The consideration of hidden stations.
(iii) The signaling of the present transmission situation
in the PUs system to all stations of the SUs system such that
these do not use the frequencies occupied by the PUs.
The solutions to these problems have recently been found
[13]. The keywords for these solutions are distributed detec-
tion, boosting of the detection results and combining them
in the hotspots access point to an occupancy vector, and dis-
tributing the occupancy vector to all mobile stations in the
hotspot.
282 EURASIP Journal on Wireless Communications and Networking
Bit
sequence
1
m
.
.
.
1
m
.
.
.
1
m
.
.
.
Cod
Cod
Cod
S
e
r
i
a
l
-
t
o
-
p
a
r
a
l
l
e
l
c
o
n
v
e
r
s
i
o
n
I
D
F
T
(
I
F
F
T
)
P
a
r
a
l
l
e
l
-
t
o
-
s
e
r
i
a
l
c
o
n
v
e
r
s
i
o
n
D/A
RF-
Mod.
(a)
f
f
3
f
2
f
1
f
0
f
1
f
2
f
3
f

(b)
Figure 6: OFDM. (a) Transmitter. (b) Spectrum.
The central point in our present discussion is that the SUs
systems transceivers have in some sense to act like CRs. They
have to sense their spectral neighborhood for PUs signals
and to react upon their ndings.
7. TECHNOLOGY CENTRIC COGNITIVE RADIO
In a more advanced spectrum sharing system, CRs have to
apply more advanced algorithms. If a portion of the spec-
trum may be accessed by any access mode, the following
procedure becomes imaginable: starting from the transmis-
sion demand of its user, the CR decides about the data rate,
the transmission mode, and therefore about the bandwidth
of the transmission. Afterwards it has to nd an appropri-
ate resource for its transmission. This presumes that the CR
knows where it is (self-location), what it is able to do (self-
awareness), and where the reachable base stations are. To get
more information about possible interferences it should, for
example, be able to detect signals active in adjacent frequency
bands and to recognize their transmission standards [14].
Summing up, a CR should have implemented the follow-
ing technologies (possibly among others):
(i) location sensors (e.g., GPS or Galileo);
(ii) equipment to monitor its spectral environment in an
intelligent
2
way;
2
Intelligent means that searching for usable frequency bands is not done
by just scanning the whole spectrum.
Control
SDR
core
Spectrum
monitoring
Localization
Information and knowledge processing
Figure 7: Technology centric cognitive radio.
(iii) in order to track the locations or the spectral environ-
ments developments, learning and reasoning algorithms have
to be implemented;
(iv) when complying with a communications etiquette, it
has to listen before talk as well as to prevent the disturbance
of hidden stations;
(v) in order to be fair it has to compromise its own de-
mands with the demands of other users, most probably in
making decisions in a competitive environment using the re-
sults of game theory [15];
(vi) it has to keep its owner informed via a highly sophis-
ticated man-machine interface.
A rst block diagram of a technology centric CR is given
in Figure 7. One of the most important decisions that have to
be made in an open access environment is whether a control
channel is to be implemented or not. The most challenging
development is that of the information and knowledge pro-
cessing.
8. CONCLUSIONS
Standardization of a transmission mode is necessary to en-
sure its success on the market. From standards we can learn
about the main parameters of a system and, by comparing
dierent standards, we may conclude about similarities and
dissimilarities within their signal processing chains. Keep-
ing this knowledge in mind, we are able to construct PaC-
SDRs. A far more general setup is given by the SCA which is
a framework for the recongurablity of transceivers and for
the portability of waveforms from one hardware platform to
another. Starting from SDRs, the next step in the evolution
of intelligent transmission devices leads to CRs that may be
looked upon as a small part of the physical world using and
providing information over very dierent time scales. Since
this approach seems to be very futuristic, we take a look at
the urgent problem of ecient spectrum usage. In order to
introduce advanced spectrum management procedures (e.g.,
spectrum pooling), the employment of CRs that at least are
able to monitor their electromagnetic environments and to
track their own locations is necessary. Therefore, the devel-
opment of technology centric CRs is proposed here as a rst
step towards general CRs.
SDRBasics and Evolution to Cognitive Radio 283
ACKNOWLEDGMENT
The author gratefully acknowledges the inuence that the 6th
European Frameworks Integrated Project End-to-End Recon-
gurability (E
2
R) as well as the Software-Dened Radio Fo-
rum have on his present work.
REFERENCES
[1] C. Heegard and S. B. Wicker, Turbo Coding, Kluwer Academic,
Boston, Mass, USA, 1999.
[2] F. K. Jondral, Parametrizationa technique for SDR imple-
mentation, in Software Dened RadioEnabling Technolo-
gies, W. Tuttlebee, Ed., pp. 232256, John Wiley & Sons, Lon-
don, UK, 2002.
[3] A. Wiesler and F. K. Jondral, A software radio for second-and
third-generation mobile systems, IEEE Trans. Veh. Technol.,
vol. 51, no. 4, pp. 738748, 2002.
[4] P. Rykaczewski, D. Pienkowski, R. Circa, and B. Steinke, Sig-
nal path optimization in software dened radio systems,
IEEE Trans. Microwave Theory Tech., vol. 53, no. 3, pp. 1056
1064, 2005.
[5] T. Hentschel and G. Fettweis, Sample rate conversion for
software radio, IEEE Commun. Mag., vol. 38, no. 8, pp. 142
150, 2000.
[6] M. Sieber, Design of a generic protocol stack for an adaptive
terminal, in Proc. 1st Karlsruhe Workshop on Software Radios,
pp. 3134, Institut f ur Nachrichtentechnik, Universit at Karl-
sruhe (TH), Karlsruhe, Germany, March 2000.
[7] F. K. Jondral, R. Machauer, and A. Wiesler, Software Radio
Adaptivit at durch Parametrisierung, J. Schlembach Fachverlag,
Weil der Stadt, Germany, 2002.
[8] Software communications architecture specication,
jtrs-5000sca v3.0, Joint Tactical Radio System (JTRS)
Joint Program Oce, August 2004, available online on
http://jtrs.army.mil.
[9] Software Dened Radio Forum, available online on
http://www.sdrforum.org.
[10] J. Mitola III and G. Q. Maguire, Cognitive radio: making
software radios more personal, IEEE Pers. Commun., vol. 6,
no. 4, pp. 1318, 1999.
[11] J. Mitola III, Cognitive radio: an integrated agent architec-
ture for software dened radio, Ph.D. dissertation, Computer
Communication System Laboratory, Department of Telein-
formatics, Royal Institute of Technology (KTH), Stockholm,
Sweden, May 2000.
[12] G. Staple and K. Werbach, The end of spectrum scarcity,
IEEE Spectr., vol. 41, no. 3, pp. 4852, 2004.
[13] T. A. Weiss and F. K. Jondral, Spectrum pooling: an inno-
vative strategy for the enhancement of spectrum eciency,
IEEE Commun. Mag., vol. 42, no. 3, pp. 814, 2004.
[14] M.

Oner and F. K. Jondral, Air interface recognition for a
software radio system exploiting cyclostationarity, in Proc.
15th IEEE International Symposium on Personal, Indoor and
Mobile Radio Communications (PIMRC 04), vol. 3, pp. 1947
1951, Barcelona, Spain, September 2004.
[15] J. Neel, J. Reed, and R. Gilles, The role of game theory in
the analysis of software radio networks, in Proc. Software De-
ned Radio Forum Technical Conference and Product Exhibi-
tion (SDR 02), vol. 2, pp. NP302, San Diego, Calif, USA,
November 2002.
[16] J. Mitola III, Software Radio Architecture: Object-Oriented Ap-
proaches to Wireless Systems Engineering, John Wiley & Sons,
New York, NY, USA, 2000.
[17] J. Mitola III and Z. Zvonar, Eds., Software Radio Technologies:
Selected Readings, John Wiley & Sons, New York, NY, USA,
2000.
[18] J. Mitola III and W. Tuttlebee, Eds., Software Dened Radio:
Origins, Drivers and International Perspectives, John Wiley &
Sons, Chichester, UK, 2002.
[19] W. Tuttlebee, Ed., Software Dened Radio: Enabling Technolo-
gies, John Wiley & Sons, Chichester, UK, 2002.
[20] M. Dillinger, K. Madani, and N. Alonistioti, Eds., Software De-
ned Radio: Architectures, Systems and Functions, John Wiley
& Sons, Chichester, UK, 2003.
[21] J. Reed, Software Radioa Modern Approach to Radio Engi-
neering, Prentice-Hall, Upper Saddle River, NJ, USA, 2002.
[22] H. Harada and R. Prasad, Simulation and Software Radio for
Mobile Communications, Artech House, Boston, Mass, USA,
2002.
[23] S. Haykin, Cognitive radio: brain-empowered wireless com-
munications, IEEE J. Select. Areas Commun., vol. 23, no. 2,
pp. 201220, 2005.
Friedrich K. Jondral received a Diploma
and a Doctoral degree in mathematics from
the Technische Universit at Braunschweig,
Germany, in 1975 and 1979, respectively.
During the winter semester 1977/78, he was
a Visiting Researcher in the Department
of Mathematics, Nagoya University, Japan.
From 1979 to 1992, Dr. Jondral was an em-
ployee of AEG-Telefunken (now European
Aeronautic Defence and Space Company
(EADS)), Ulm, Germany, where he held various research and devel-
opment, as well as management positions. Since 1993, Dr. Jondral
has been Full Professor and Head of the Institut f ur Nachrichten-
technik at the Universit at Karlsruhe (TH), Germany. There, from
2000 to 2002, he served as the Dean of the Department of Electri-
cal Engineering and Information Technology. During the summer
semester of 2004, Dr. Jondral was a Visiting Faculty in the Mobile
and Portable Radio Research Group of Virginia Tech, Blacksburg,
Va. His current research interests are in the elds of ultra-wide-
band communications, software-dened and cognitive radio, sig-
nal analysis, pattern recognition, network capacity optimization,
and dynamic channel allocation. Dr. Jondral is a Senior Member
of the IEEE; he currently serves as an Associate Editor of the IEEE
Communications Letters and as a Member of the Software-Dened
Radio Forums Board of Directors.
EURASIP Journal on Wireless Communications and Networking 2005:3, 284297
c 2005 Ioannis Dagres et al.
Flexible Radio: A Framework for Optimized Multimodal
Operation via Dynamic Signal Design
Ioannis Dagres
Institute of Accelerating Systems & Applications (IASA), National Kapodistrian University of Athens (NKUA),
P.O. Box 17214, 10024 Athens, Greece
Email: jdagres@phys.uoa.gr
Andreas Zalonis
Institute of Accelerating Systems & Applications (IASA), National Kapodistrian University of Athens (NKUA),
P.O. Box 17214, 10024 Athens, Greece
Email: azalonis@phys.uoa.gr
Nikos Dimitriou
Institute of Accelerating Systems & Applications (IASA), National Kapodistrian University of Athens (NKUA),
P.O. Box 17214, 10024 Athens, Greece
Email: nikodim@phys.uoa.gr
Konstantinos Nikitopoulos
Institute of Accelerating Systems & Applications (IASA), National Kapodistrian University of Athens (NKUA),
P.O. Box 17214, 10024 Athens, Greece
Email: cnikit@cc.uoa.gr
Andreas Polydoros
Institute of Accelerating Systems & Applications (IASA), National Kapodistrian University of Athens (NKUA),
P.O. Box 17214, 10024 Athens, Greece
Email: polydoros@phys.uoa.gr
Received 16 March 2005; Revised 19 April 2005
The increasing need for multimodal terminals that adjust their conguration on the y in order to meet the required quality of
service (QoS), under various channel/system scenarios, creates the need for exible architectures that are capable of performing
such actions. The paper focuses on the concept of exible/recongurable radio systems and especially on the elements of exibility
residing in the PHYsical layer (PHY). It introduces the various ways in which a recongurable transceiver can be used to provide
multistandard capabilities, channel adaptivity, and user/service personalization. It describes specic tools developed within two
IST projects aiming at such exible transceiver architectures. Finally, a specic example of a mode-selection algorithmic architec-
ture is presented which incorporates all the proposed tools and, therefore, illustrates a baseband exibility mechanism.
Keywords and phrases: exible radio, recongurable transceivers, adaptivity, MIMO, OFDM.
1. INTRODUCTION
The emergence of speech-based mobile communications
in the mid 80s and their exponential growth during the
90s have paved the way for the rapid development of new
wireless standards, capable of delivering much more ad-
vanced services to the customer. These services are and
This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
will be based on much higher bit rates than those pro-
vided by GSM, GPRS, and UMTS. The new services (video
streaming, video broadcasting, high-speed Internet, etc.)
will demand much higher bit rates/bandwidths and will
have strict QoS requirements, such as the received BER
and the end-to-end delay. The new and emerging stan-
dards (WiFi, WiMax, DVB-T, S-DMB, IEEE 802.20) will
have to compete with the ones based on wired commu-
nications and overcome the barriers posed by the wireless
medium to provide seamless coverage and uninterrupted
communication.
Flexible Radio Framework for Optimized Multimodal Operation 285
Another issue that is emerging pertains to the equipment
that will be required to handle the plethora of the new stan-
dards. It will be highly unlikely that the user will have avail-
able a separate terminal for each of the introduced standards.
There will be the case that the use of a specic standard will
be dictated by factors such as the user location (inside build-
ings, in a busy district, or in a suburb), the user speed (pedes-
trian, driving, in a high-speed train), and the required quality
(delay sensitivity, frame error rate, etc.). There might also be
cases in which it would be preferred that a service was de-
livered using a number of dierent standards (e.g., WiFi for
video, UMTS for voice), based on some criteria related to the
terminal capabilities (say, power consumption) and the net-
work capacity constraints. Therefore, the user equipment has
to follow the rapid development of new wireless standards by
providing enough exibility and agility to be easily upgrade-
able (with perhaps the modication/addition of specic soft-
ware code but no other intervention in hardware).
We note that exibility in the terminal concerns both the
analog/front-end (RF/IF) as well as digital (baseband) parts.
The paper will focus on the issues pertaining to the base-
band exibility and will discuss its interactions with the pro-
cedures taking place in the upper layers.
2. DEFINITIONS OF RADIOFLEXIBILITY
The notion of exibility in a radio context may be dened
as an umbrella concept, encompassing a set of nonoverlap-
ping (in a conceptual sense) postulates or properties (each of
which must be dened individually and clearly for the overall
denition to be complete) such as adaptivity, recongurabil-
ity, modularity, scalability, and so on. The presence of any
subset of such features would suce to attribute the quali-
fying term exible to any particular radio system [1]. These
features are termed nonoverlapping in the sense that the
occurrence of any particular one does not predicate or force
the occurrence of any other. For example, an adaptive sys-
tem may or may not be recongurable, and so on. Additional
concepts can be also added, such as ease of use or seam-
lessly operating from the users standpoint, as long as these
attributes can be quantied and identied in a straightfor-
ward way, adding a new and independent dimension of ex-
ibility. Recongurability, for instance, which is a popular di-
mension of exibility, can be dened as the ability to rear-
range various modules at a structural or architectural level
by means of a nonquantiable
1
change in its conguration.
Adaptivity, on the other hand, can be dened as the radio sys-
tem response to changes by properly altering the numerical
value of a set of parameters [2, 3]. Thus, adaptive transmitted
(Tx) power or adaptive bit loading in OFDM naturally fall in
the latter category, whereas dynamically switching between,
say, a turbo-coded and a convolutional-coded system in re-
sponse to some stimulus (or information) seems to t better
the code-recongurability label, simply because that type of
1
Nonquantiable here means that it cannot be represented by a nu-
merical change in a parametric set.
change implies a circuit-design change, not just a numeric
parameter change. Furthermore, the collection of adaptive
and recongurable transmitted-signal changes in response to
some channel-state-information feedback may be termed dy-
namic signal design (DSD). Clearly, certain potential changes
may fall in a grey area between denitions.
2
A primitive example of exibility is the multiband oper-
ation of current mobile terminals, although this kind of ex-
ibility driven by the operator is not of great research interest
from the physical-layer point of view. A more sophisticated
version of such a exible transceiver would be the one that
has the intelligence to autonomously identify the incumbent
system conguration and also has the further ability to ad-
just its circumstances and select its appropriate mode of op-
eration accordingly. Software radio, for example, is meant to
exploit recongurability and modularity to achieve exibil-
ity. Other approaches may encompass other dimensions of
exibility, such as adaptivity in radio resource management
techniques.
3. FLEXIBILITY SCENARIOS
In response to the demand for increasingly exible radio
systems from industry (operators, service providers, equip-
ment manufacturers, chip manufacturers, system integra-
tors, etc.), government (military communication and signal-
intelligence systems), as well as various user demands, the
eld has grown rapidly over the last twenty years or so (per-
haps more in certain quarters), and has intrigued and acti-
vated R&D Departments, academia, research centers, as well
as funding agencies. It is now a rapidly growing eld of in-
quiry, development, prototyping, and even elding. Because
of the enormity of the subject matter, it is hard to draw solid
boundaries that exclusively envelop the scientic topic, but
it is clear that such terms as SR, SDR, recongurable radio,
cognitive/intelligent/smart radio, and so on are at the cen-
ter of this activity. Similar arguments would include work
on exible air-interface waveforms and/or generalized (and
properly parameterized) descriptions and receptions thereof.
Furthermore, an upward look (from the physical-layer bot-
tom of the communication-model pyramid) reveals an ever-
expanding role of research on networks that include recon-
gurable topologies, exible medium-access mechanisms,
interlayer optimization issues, agile spectrum allocation [4],
and so on. In a sense, ad hoc radio networks t the concept,
as they do not require any rigid or xed infrastructure. Simi-
larly, looking down at the platform/circuit level [5], we see
intense activity on exible and malleable platforms and de-
signs that are best suited for accommodating such exibility.
In other words, every component of the telecommunication
2
This terminology is to a certain degree arbitrary and not universally
agreed upon; for instance, some authors call a radio system recongurable
because it is adaptive, meaning that it adapts to external changes. On
the other hand, the term adaptive has a clear meaning in the signal-
processing-algorithms literature (e.g., an adaptive equalizer is the one whose
coecient values change slowly as a function of the observation), and the
denition proposed here conforms to that understanding.
286 EURASIP Journal on Wireless Communications and Networking
and radio universe can be seen as currently participating in
the radio-exibility R&D work, making the eld exciting as
well as dicult to describe completely.
Among the many factors that seem to motivate the
eld, the most obvious seems to be the need for multistan-
dard, multimode operation, in view of the extreme pro-
liferation of dierent, mutually incompatible radio stan-
dards around the globe (witness the analog-to-digital-to-
wideband-to-multicarrier evolution of air interfaces in the
various cellular-system generations). The obvious desire for
having a single-end device handling this multitude in a com-
patible way is then at the root of the push for exibility. This
would incorporate the desire for legacy-proof functional-
ity, that is, the ability to handle existing systems in a single
unied terminal (or single infrastructure access point), re-
gardless of whether this radio system is equipped with all the
related information prestored in memory or whether this is
software-downloaded to a generically architected terminal;
see [6] for details. In a similar manner, future-proof sys-
tems would employ exibility in order to accommodate yet-
unknown systems and standards with a relative ease (say, by a
mere resetting of the values of a known set of parameters), al-
though this is obviously a harder goal to achieve that legacy-
proofness. Similarly, economies of scale dictate that radio
transceivers employ reusable modules to the degree possible
(hence the modularity feature). Of course, truly optimized
designs for specic needs and circumstances, lead to point
solutions, so that exibility of the modular and/or generic
waveform-design sort may imply some performance loss. In
other words, the benet of exibility may come at some cost,
but hopefully the tradeo is still favorable to exible designs.
There are many possible ways to exploit the wide use of a
single exible recongurable baseband transceiver, either on
the user side or on the network side. One scenario could be
the idea of location-based reconguration for either multi-
service ability or seamless roaming. A exible user terminal
can be capable of reconguring itself to whichever standard
prevails (if there are more than one that can be received) or
exists (if it is the only one) at each point in space and time,
either to be able to receive the ever-available (but possibly
dierent) service or to receive seamlessly the same service.
Additionally, the network side can make use of the future-
proof reconguration capabilities of its exible base stations
for soft infrastructure upgrading. Each base station can be
easily upgradeable to each current and future standard. An-
other interesting scenario involves the combined reception
of the same service via more than one standard in the same
terminal. This can be envisaged either in terms of standard
selection diversity, according to which a exible terminal
will be able to download the same service via dierent air-
interface standards and always sequentially (in time) select
the optimum signal (to be processed through the same ex-
ible baseband chain) or, in terms of service segmentation
and standard multiplexing, meaning that a exible termi-
nal will be able to collect frames belonging to the same ser-
vice via dierent standards, thus achieving throughput maxi-
mization for that service, or receive dierent services (via dif-
ferent standards) simultaneously. Finally, another exibility
scenario could involve the case of peer-to-peer communica-
tion whereby two exible terminals could have the advan-
tage of reconguring to a specic PHY (according to condi-
tions, optimization criteria) and establish a peer-to-peer ad
hoc connection.
The aforementioned scenarios of exibility point to the
fact that the elements of wireless communications equip-
ment (on board both future terminals and base station sites)
will have to fulll much more complicated requirements than
the current ones, both in terms of multistandard capabilities
as well as in terms of intelligence features to control those
capabilities. For example, a exible terminal on either of the
aforementioned scenarios must be able to sense its environ-
ment and location and then alter its transmission and recep-
tion parameters (frequency band, power, frequency, modula-
tion, and other parameters) so as to dynamically adapt to the
chosen standard/mode. This could in theory allow a multidi-
mensional reuse of spectrum in space, frequency, and time,
overcoming the various spectrum usage limitations that have
slowed broadband wireless development and thus lead to one
vision of cognitive radio [7], according to which radio nodes
become radio-domain-aware intelligent agents that dene
optimum ways to provide the required QoS to the user.
It is obvious that the advantageous operation of a truly
exible baseband/RF/IF platform will eventually include the
use of sophisticated MAC and RRM functionalities. These
will have to regulate the admission of newusers in the system,
the allocation of a mode/standard to each, the conditions of
a vertical handover (from one standard to another), and the
scheduling mechanisms for packet-based services. The cri-
teria for assigning resources from a specic mode to a user
will depend on various parameters related to the wireless
channel (path loss, shadowing, fast fading) and to the spe-
cic requirements imposed by the terminal capabilities (min-
imization of power consumption and transmitted power),
the generated interference, the user mobility, and the service
requirements. That cross-layer interaction will lead to the ul-
timate goal of increasing the multiuser capacity and coverage
while the power requirements of all exible terminals will be
kept to a minimum required level.
4. FLEXIBLE TRANSCEIVER ARCHITECTURE
AT THE PHY-DYNAMIC SIGNAL DESIGN
4.1. Transmission schemes and techniques
Research exploration of the next generation of wireless sys-
tems involves the further development of technologies like
OFDM, CDMA, MC-CDMA, and others, along with the use
of multiple antennas at the transmitter and the receiver. Each
of these techniques has its special benets in a specic envi-
ronment: for example, OFDM is used successfully in WLAN
systems (IEEE 802.11a), whereas CDMA is used successfully
in cellular 2G (IS-95) and 3G (UMTS) systems. The selection
of a particular one relies on the operational environment of
each particular system. In OFDM, the available signal band-
width is split into a large number of subcarriers, orthog-
onal to each other, allowing spectral overlapping without
Flexible Radio Framework for Optimized Multimodal Operation 287
Outer
code
Inner
code
Tx1
Tx2
TxM
t
Rx1
RxM
r
MIMO channel
SISO channel
.
.
.
Inner
decoder
Outer
decoder
CSI
Figure 1: MIMO code design procedure.
interference. The transmission is divided into parallel sub-
channels whose bandwidth is narrow enough to make them
eectively frequency at. Acyclic prex is used to combat ISI,
in order to avoid (or simplify) the equalizer [8].
The combination of OFDM and CDMA, known as
MC-CDMA [9], has gained attention as a powerful trans-
mission technique. The two most frequently investigated
types are multicarrier CDMA (MC-CDMA) which employs
frequency-domain spreading and multicarrier DS-CDMA
(MC-DS-CDMA) which uses time-domain spreading of the
individual subcarrier signals [9, 10]. As discussed in [9],
MC-CDMA using DS spread subcarrier signals can be fur-
ther divided into multitone DS-CDMA, orthogonal MC-DS-
CDMA, and MC-DS-CDMA using no subcarrier overlap-
ping. In [11, 12], it is shown that the above three types of
MC-DS-CDMAschemes with appropriate frequency spacing
between two adjacent subcarriers can be unied in the family
of generalized MC-DS-CDMA schemes.
Multiple antennas with transmit and receive diversity
techniques have been introduced to improve communication
reliability via the diversity gain [13]. Coding gain can also
be achieved by appropriately designing the transmitted sig-
nals, resulting in the introduction of space-time codes (STC).
Combined schemes have already been proposed in the lit-
erature. MIMO-OFDM has gained a lot of attention in re-
cent years and intensive research has already been performed.
Generalized MC-DS-CDMA with both time- and frequency-
domain spreading is proposed in [11, 12] and eorts on
MIMO MC-CDMA can be found in [14, 15, 16, 17, 18].
4.2. Dynamic signal design
Flexible systems do not just incorporate all possible point so-
lutions for delivering high QoS under various scenarios, but
possess the ability to make changes not only on the algorith-
mic but also on the structural level in order to meet their
goals. Thus, the DSD goal is to bring the classic design proce-
dure of the PHY layer into the intelligence of the transceiver
and initiate new system architectural approaches, capable of
creating the tools for on-the-y reconguration. The mod-
ule responsible for all optimization actions is herein called
supervisor, also known as controller and the like.
The dierence between adaptive modulation and cod-
ing (AMC) and dynamic signal design (DSD) is that AMC
is a design approach with a main focus on developing algo-
rithms for numerical parameter changes (constellation size,
Tx power, coding parameters), based on appropriate feed-
back information, in order to approach the capacity of the
underlying channel. The type of channel code in AMC is pre-
determined for various reasons, such as known performance
of a given code in a given channel, compatibility with a given
protocol, xed system complexity, and so on. Due to the va-
riety of channel models, system architectures, and standards,
there is a large number of AMC point solutions that will suc-
ceed in the aforementioned capacity goal.
In a typical communication system design, the algorith-
mic choice of most important functional blocks of the PHY
layer is made once at design time, based on a predetermined
and restricted set of channel/system scenarios. For example,
the channel waveform is selected based on the channel (fast
fading, frequency selective) and the system characteristics
(multi/single-user, MIMO). On the other hand, truly exi-
ble transceivers should not be restricted to one specic sce-
nario of operation, so that the choice of channel waveform,
for instance, must be broad enough to adapt either para-
metrically or structurally to dierent channel/system condi-
tions. One good example of such a exible waveform would
be fully parametric MC-CDMA, which can adjust its spread-
ing factor, the number of subcarriers, the constellation size,
and so on. Similarly, MIMO systems that are able to change
the number of active antennas or the STC, on top of a exi-
ble modulation method like MC-CDMA, can provide a large
number of degrees of freedom to code designers.
With respect to the latter point, we note that STC de-
sign has relied heavily on the pioneering work of Tarokh
et al. in [19], where design principles were rst established.
Recent overall code design approaches divide coding into
inner and outer parts (see Figure 1), in order to produce
easily implementable solutions [20, 21]. Inner codes are
the so-called ST codes, whereas outer codes are the clas-
sic SISO channel codes. Each entity tries to exploit a dif-
ferent aspect of channel properties in order to improve the
overall system performance. Inner codes usually try to get
288 EURASIP Journal on Wireless Communications and Networking
Table 1: Flexible design tools and inputs.
Physical-layer
exibility
Modulation (a exible
scheme like MC-CDMA)
Space-time coding Channel coding
Tools
Adjustable FFT size,
spreading code length,
constellation size (bit
loading), Tx power per
carrier (power loading)
Adjustable number of Tx/Rx
antennas used, exible ST
coding scheme as opposed to
(diversity/multiplexing/coding/SNR
gain)
Flexible FEC codes (e.g.,
turbo, convolutional, LDPC)
with adjustable coding rate,
block size,
code polynomial
Inputs
Number of users
sharing the same BW,
channel type
(indoor/outdoor)
Channel variation in time
(Doppler), Rx antenna
correlation factor, feedback
dealy, goodness of channel
estimation
Eective channel
parameters (including
STC eects)
diversity/multiplexing/SNR gain, while outer codes try to
get diversity/coding gain. The best choice of an inner/outer
code pair relies on channel characteristics, complexity, and
feedback-requirement (CSI) considerations.
There are several forms of diversity that a system can of-
fer, such as time, frequency, and space. The ability to change
the number of antennas, subcarriers, spreading factor and
the ST code provides great control for the purpose of reach-
ing the diversity oered by the current working environment.
There are many STCs presented in the literature which ex-
ploit one form of diversity in a given system/environment.
All these point solutions must be taken into account in order
to design a system architecture that eciently incorporates
most of them.
Outer channel codes must also be chosen so as to ob-
tain the best possible overall system performance. In some
cases, the diversity gain of the cascade coding can be analyti-
cally derived, based on the properties of both coding options
[20]. Even in these idealized scenarios, however, individually
maximizing the diversity gain of both codes does not im-
prove performance. This means that, in order to maximize
the overall performance of the system, a careful tradeo is
necessary between multiplexing gain, coding gain, and SNR
gain.
Newchannel estimation methods must also be developed
in order to estimate not only the channel gain values but also
other related inputs (see Table 1). For example, the types of
diversity that can be exploited by the receiver or the corre-
lation factor between multiple antennas are important in-
puts for choosing the best coding option. Another input is
the channel rate of change (Doppler), normalized to the sys-
tem bandwidth, in order to evaluate the feedback delay. In
most current AMC techniques, this kind of input informa-
tion has not been employed, since the channel characteristics
have not been considered as system design variables.
5. FLEXIBILITY TOOLS
The paper is based on techniques developed in two IST
projects, WIND-FLEX and Stingray. The main goal of
WIND-FLEX was the development of exible (in the
sense of Section 2) architectures for indoor, high-bit-rate
wireless modems. OFDM was the signal modulation of
choice [22], along with a powerful turbo-coded scheme.
The Stingray Project targeted a Hiperman-compatible [23]
MIMO-OFDM system for Fixed Wireless Access (FWA) ap-
plications. It relied on a exible architecture that exploited
the channel state information (CSI) provided by a feedback
channel from the receiver to the transmitter, driven by the
needs of the supported service.
In the following sections, the key algorithmic choices
of both projects are presented, which can be incorporated
in a single design able to operate in a variety of environ-
ments and system congurations. Since a exible transceiver
must operate under starkly dierent channel scenarios, the
transmission-mode-selection algorithm must rely solely on
instantaneous channel measurements and not on the aver-
age behavior of a specic channel model. This imposes the
restriction of low channel dynamics in order to have the ben-
et of feedback information. On both designs, a maximum
of one bit per carrier is allowed for feedback information,
along with the mode selection number. The simplicity of this
feedback information makes both designs robust to channel
estimation errors or feedback delay.
5.1. AMC in WIND-FLEX
The WIND-FLEX (WF) system was placed in the 17 GHz
band, and has been measured to experience high frequency
selectivity within the 50 MHz channel widths. The result
is strong performance degradation due to few subcarri-
ers experiencing deep spectral nulls. Even with a power-
ful coding scheme such as turbo codes, performance degra-
dation is unacceptable. The channel is fairly static for a
large number of OFDM symbols, allowing for ecient de-
sign of adaptive modulation algorithms in order to deal
with this performance degradation. In order to keep imple-
mentation complexity at a minimum, and also to minimize
the required channel feedback trac, two design constraints
have been adopted: same constellation size for all subcarri-
ers, as well as same power for all within an OFDM sym-
bol, although both these parameters are adjustable (adap-
tive).
Flexible Radio Framework for Optimized Multimodal Operation 289
Target BER
Required uncoded
BER LUT
Mode Tx power
evaluation
Target throughput
(i.e., code (type,rate),
constellation)
Estimated channel
gains (frequency domain)
Estimated noise PSD
Tx power
needed
Figure 2: Simplied block diagram of algorithm 1.
Two algorithms have been proposed in order to optimize
the performance. The rst algorithm (Figure 2) evaluates the
required Tx power for a specic code, constellation, and
channel realization to achieve the target BER. If the required
power is greater than the maximum available/allowable Tx
power, a renegotiation of the target QoS (lowering the re-
quirements) takes place. This approach exhibits low com-
plexity and limited feedback information requirements. The
relationship of the uncoded versus the coded BER perfor-
mance in an OFDM system have been given in [24] for turbo
codes and can be easily extended to convolutional codes. An
implementation of this algorithm is described in [25].
The large SNR variation across the subcarriers of OFDM
degrades system performance even when a strong outer code
is used. To counter, the technique of Weak Subcarrier ex-
cision (WSCE) is introduced as a way to exclude a certain
number of subcarriers from transmission. The second pro-
posed algorithm employs WSCE along with the appropriate
selection of code/constellation size. This is called the coded
weak subcarrier excision (CWSCE) method.
In WIND-FLEX channel scenarios performance im-
proved when using a xed number of excised subcarriers.
The bandwidth penalty introduced by this method was com-
pensated by the ability to use higher code rates. In Figure 3,
bit error rate (BER) simulation curves are shown for the un-
coded performance of xed WSCE and are compared with
the bit loading algorithm presented in [26] for the NLOS
channel scenario. {Rate 1} and {Rate 2} are the system
throughputs when using 4-QAM with 10% and 20% WSCE,
respectively. The BER performance without bit loading or
WSCE is also plotted for a 4-QAM constellation.
There is a clear improvement by just using a xed WSCE
scheme, and there is a marginal loss in comparison to the
nearly optimum bit-loading algorithm. Based on the average
SNR across the subcarriers, semianalytic computation of the
average and outage capacity for the eective channel is possi-
ble in order to evaluate a performance upper bound of a sys-
tememploying such WSCE plus uniformpower loading. The
use of an outer code helps to come close to this bound. We
note that the average capacity of an OFDM system without
10
1
10
2
10
3
B
E
R
2 4 6 8 10 12 14 16 18
SNR/bit
4-QAM without bit loading
Bit-loading rate-2 case
WSCE (10%) rate-1 case
Bit-loading rate-1 case
WSCE (20%) rate-2 case
Figure 3: Uncoded performance for WIND-FLEX NLOS channel.
power-loading techniques is
C
E
= E
_
_
_
1
N
N
_
k=1
log
2
_
1 + SNR
k
_
_
_
_
bits/carrier, (1)
where the expectation operator is over the stochastic chan-
nel. For a system employing WSCE, the summation is over
the used carriers along with appropriate transmit energy nor-
malization. These capacity results are based on the qua-
sistatic assumption. For each burst, it is also assumed that
a suciently large number of bits are transmitted, so that
the standard innite time horizon of information theory is
meaningful. In Figure 4, the system average capacity (SAC)
and the 1% system outage capacity (SOC) of the WF system
employing various WSCE scenarios are presented. Here, the
denitions are as follows.
(i) SAC (system average capacity). This is equivalent to
the mean or ergodic capacity [27] applied to the ef-
fective channel. It serves as an upper bound of systems
with boundless complexity or latency that use a spe-
cic inner code.
(ii) SOC (system outage capacity). This is the 1% outage
capacity of the STC-eective channel.
(iii) AC and OC. This is the average capacity and outage
capacity of the actual sample-path channel.
The capacity of an AWGN channel is also plotted as
an upper bound for a given SNR. At low SNR regions, the
capacity of a system employing as high as 30% WSCE is
higher than a system using all carriers without power load-
ing. At high SNR, the capacity loss asymptotically approaches
the bandwidth percentage loss of WSCE. The capacity using
adaptive WSCE is also plotted. In some channel realizations,
290 EURASIP Journal on Wireless Communications and Networking
7
6
5
4
3
2
1
0
C
a
p
a
c
i
t
y
(
b
i
t
s
/
s
/
H
z
)
0 2 4 6 8 10 12 14 16 18 20
Average channel SNR
0% WSCE SOC
10% WSCE SAC
10% WSCE SOC
20% WSCE SAC
20% WSCE SOC
30% WSCE SAC
30% WSCE SOC
Optimum selection SOC
AWGN
Optimum selection SAC
0%WSCE SAC
Figure 4: System average capacity and system 1% outage capacity
of dierent WSCE options.
in the low-to-medium SNR region, a 30% to 50% WSCE is
needed. This result motivates the design of the second algo-
rithm. The impact of CWSCE is the ability to choose between
dierent code rates for the same target rate, a feature absent
from the rst algorithm. Assume an ordering of the dierent
pairs {code rate-constellation size} based on the SNR neces-
sary to achieve a certain BER performance. It is obvious that
this ordering also applies to the throughput of each pair (a
system will not include pairs that need more power to pro-
vide lower throughput). For each of these pairs, the xed per-
centage of excised carriers is computed so that they all pro-
vide the same nal (target) throughput.
The block diagram of CWSCE algorithm is given in
Figure 5. The respective denitions are as follows:
(i) x
i
, i = 1, . . . ,l, is one of the system-supported constel-
lations;
(ii) y
i
, i = 1, . . . , M, is one of the supported outer chan-
nel codes. These can be totally dierent codes like
turbo, convolutional, LDPC, or the codes resulting
from puncturing one mother code, or both;
(iii) z
i
, i = 1, . . . ,n, are the resulting WSCE percentages for
the n competitive triplets;
(iv) Pos(z
i
) are the positions of the z
i
% of weakest gains.
(v) H is the vector of the estimated channel gains in the
frequency domain;
(vi)

N
0
is the estimated power spectral density of the noise.
(vii) RUB
i
, i = 1, . . . ,n, is the required uncoded BER for
constellation x
i
and code y
i
;
(viii) PTx
i
, i = 1, . . . ,n, is the required Tx power for the ith
triplet.
The algorithm calculates the triplet that needs the min-
imum Tx power for a given target BER. If the mini-
mum required power is greater than the maximum avail-
able/allowable Tx power, it renegotiates the QoS. Transmit-
power adaptation is usually avoided, although it can be han-
dled with the same algorithm. The triplet selection will still
be the one that needs the minimum Tx power. The extra
computation load is mainly due to the channel-tap sorting.
Proper exploitation of the channel correlation in frequency
(coherence bandwidth) can reduce this complexity overhead.
Instead of sorting all the channel taps, one can sort groups
of highly correlated taps. These groups can be restricted to
have an equal number of taps. There are many sorting algo-
rithms in the literature with dierent performance-versus-
complexity characteristics that can be employed, depending
on implementation limitations.
Simulation results using algorithm 1 for adaptive
transmission-power minimization are presented in Figure 6.
The performance gain of the proposed algorithm is shown
for 4-QAM, the code rates 1/2 and 2/3. Performance is plot-
ted for no adaptation, as well as for algorithm 1 in an NLOS
scenario. The performance over a at (AWGN) channel is
also shown for comparison reasons, since it represents the
coded performance limit (given that these codes are designed
to work for AWGN channels). The main simulation system
parameters are based on the WIND-FLEX platform. It uses a
parallel-concatenated turbo code with variable rate via three
puncture patterns (1/2, 2/3, 3/4) [28]. The recursive system-
atic code polynomial used is (13, 15)
oct
. Perfect channel esti-
mation and zero phase noise are also assumed.
In addition to the transmission power gain, the adaptive
schemes practically guarantee the desired QoS for every chan-
nel realization. Note that in the absence of adaptation, users
experiencing bad channel conditions will never get the re-
quested QoS, whereas users with a good channel would
correspondingly end up spending too much power versus
what would be needed for the requested QoS. By adopting
these algorithms, one computes (for every channel realiza-
tion) the exact needed power for the requested QoS, and thus
can either transmit with minimum power or negotiate for a
lower QoS when channel conditions do not allow transmis-
sion. An average 2 dB additional gain is achieved by using the
second algorithm versus the rst one.
5.2. Adaptive STC in Stingray
As mentioned, Stingray is a Hiperman-compatible 2
2 MIMO-OFDM adaptive system. The adjustment rate,
namely, the rate at which the system is allowed to change the
Tx parameters, is chosen to be once per frame (one frame =
178 OFDM symbols) and the adjustable sets of the Tx pa-
rameters are
(1) the selected Tx antenna per subcarrier, called trans-
mission selection diversity (TSD),
(2) the {outer code rate, QAM size} set.
The antenna selection rule in TSD is to choose, for ev-
ery carrier k, to transmit from the Tx antenna T(k) with the
Flexible Radio Framework for Optimized Multimodal Operation 291
List of supported
channel codes
Competitive
triplet evaluation
List of supported
constellations
WSCE
Channel/noise
estimator
Required uncoded
BER LUT
Mode Tx power
evaluation
_
_
_
_
_
(x
1
, y
1
, z
1
)
.
.
.
(x
n
, y
n
, z
n
)
_
_
_
_
_ [(x
1
, y
1
), . . . , (x
n
, y
n
)]
.
.
.
.
.
.
[Pos(z
1
), . . . , Pos(z
n
)] _
_
_
_
_
(x
1
, RUB
1
)
.
.
.
(x
n
, RUB
n
)
_
_
_
_
_

N
0

H
Target
throuhput
_
_
_
_
_
PTx
1
.
.
.
PTx
n
_
_
_
_
_
Target BER
Figure 5: Simplied block diagram of algorithm 2.
10
2
10
3
10
4
10
5
10
6
B
E
R
2 4 6 8 10 12 14
SNR/bit
NLOS rate 2/3
NLOS rate 1/2
NLOS alg. 1 rate 2/3
NLOS alg. 1 rate 1/2
AWGN rate 2/3
AWGN rate 1/2
Figure 6: Simulation results using algorithm 1: max-log map, 4 it-
erations, NLOS, 4-QAM, rate = 1/2 and 2/3.
best performance from a maximum-ratio combining (MRC)
perspective. For the second set of parameters, the optimiza-
tion procedure is to choose the set that maximizes the system
throughput (bit rate), given a QoS constraint (BER).
In order to identify performance bounds, TSD is com-
pared with two other rate-1 STC techniques, beamforming
and Alamouti. Beamforming is the optimal solution [29] for
energy allocation in an N
T
1 system with perfect channel
knowledge at the transmitter side, whereby the same symbol
is transmitted from both antennas multiplied by an appro-
priate weight factor in order to get the maximum achiev-
able gain for each subcarrier. Alamoutis STBC is a blind
technique [30], where for each OFDM symbol period two
OFDM signals are simultaneously transmitted from the two
antennas.
Each of the three STC schemes can be treated as an ordi-
nary OFDMSISOsystemproducing (ideally) N independent
Gaussian channels [31]. This is the eective SISO-OFDM
channel. For the Stingray system (2 2), the corresponding
eective SNR (ESNR) per carrier is as follows:
For TSD, ESNR
k
=
_

H
T(k),0
k

2
+

H
T(k),1
k

2
_
E
s
N
0
, (2)
for Alamouti, ESNR
k
=
_

H
0,0
k

2
+

H
0,1
k

2
+

H
1,0
k

2
+

H
1,1
k

2
_
E
s
2N
0
, (3)
for beamforming, ESNR
k
=

max
k
E
s
N
0
, (4)
where
max
k
is the square of the maximum eigenvalue of the
2 2 channel matrix
_
H
00
k
H
10
k
H
01
k
H
11
k
_
, H
i, j
k
is the frequency re-
sponse of the channel between the Tx antenna i and Rx an-
tenna j at subcarrier k = 0, 1, . . . , N 1, and N
0
is the one-
sided power spectral density of the noise in each subcarrier.
In Figure 7, BER simulation curves are presented for all
inner code schemes and 4-QAM constellation. Both perfect
and estimated CSI scenarios are presented. The channel es-
timation procedure uses the preamble structure described in
[32].
For all simulations, path delays and the power of chan-
nel taps have been selected according to the SUI-4 model
for intermediate environment conditions [33]. The average
channel SNR is employed in order to compare adaptive sys-
tems that utilize CSI. Note that this average channel SNR is
independent of the employed STC. Having normalized each
Tx-Rx path to unit average energy, the channel SNR is equal
to one over the power of the noise component of any one of
the receivers. Alamouti is the most sensitive scheme to esti-
mation errors. This is expected, since the errors in all four
channel taps are involved in the decoding procedure. Based
on the ESNR, a semianalytic computation of the average and
292 EURASIP Journal on Wireless Communications and Networking
10
1
10
2
10
3
B
E
R
0 2 4 6 8 10
Average channel SNR/bit
BF-PCSI
TSD-PCSI
ALA-PCSI
BF-ECSI
TSD-ECSI
ALA-ECSI
Figure 7: STCs BER performance for perfect/estimated CSI
(PCSI/ECSI) and 4-QAM constellation.
outage capacity for the eective channel is possible in order
to evaluate a performance upper bound of these inner codes.
In Figure 8, the average capacity and the 1% outage ca-
pacity of the three competing systems are presented. For
comparison reasons, the average and outage capacity of the
2 2 and 1 1 systems with no channel knowledge at the
transmitter and perfect knowledge at the receiver are also
presented. It is clear that all three systems have the same slope
of capacity versus SNR. This is expected, since the rate of all
three systems is one. A system exploiting all the multiplexing
gain oered by the 2 2 channel may be expected to have a
slope similar to the capacity of the real channel (AC, OC). It
is also evident that the cost of not targeting full multiplexing
is a throughput loss compared to that achievable by MIMO
channels. On the other hand, the goal of high throughput in-
curs the price of either enhanced feedback requirements or
higher complexity. Comparing the three candidate schemes,
we conclude that beamforming is a high-complexity solution
with considerable feedback requirements, whereas Alamouti
has low complexity with no feedback requirement. TSD has
lower complexity than Alamouti, whereas in comparison
with beamforming, it has a minimal feedback requirement.
The gain over Alamouti is approximately 1.2 dB, while the
loss compared to beamforming is another 1.2 dB.
For all schemes, frequency selectivity across the OFDM
tones is limited due to the MIMO diversity gain. That is
one of the main reasons why bit loading and WSCE gave
marginal performance gain. The metric for selecting the sec-
ond set of parameters was the eective average SNR at the
receiver (meaning the average SNR at the demodulator af-
ter the ST decoding). The system performance simulation
curves based on the SNR at the demodulator (Figure 9) were
the basis for the construction of the Tx mode table (TMT),
7
6
5
4
3
2
1
0
B
i
t
s
/
c
a
r
r
i
e
r
0 5 10 15
Average channel SNR
2 2 TSD SAC
2 2 TSD SOC
2 2 ALA SAC
2 2 ALA SOC
2 2 BF SAC
2 2 BF SOC
1 1 AC
1 1 OC
2 2 AC
2 2 OC
Figure 8: System average capacity and system 1% outage capacity
of dierent STC options.
10
1
10
2
10
3
10
4
10
5
B
E
R
2 4 6 8 10 12 14 16 18 20
ESNR
4-QAM, 1/2
4-QAM, 2/3
4-QAM, 3/4
16-QAM, 1/2
16-QAM, 2/3
16-QAM, 3/4
64-QAM, 1/2
64-QAM, 2/3
64-QAM, 3/4
Figure 9: TSD-turbo system performance results.
which consists of SNR regions and code-rate/constellation
size sets for all the QoS operation modes (BER) that will be
supported by the system. The selected inner code is TSD and
the outer code is the same used in the WF system. Since per-
fect channel and noise-power knowledge are assumed, ESNR
is in fact the real prevailing SNR. This turns out to be a
good performance metric, since the outer (turbo) code per-
formance is very close to that achieved on an AWGN channel
Flexible Radio Framework for Optimized Multimodal Operation 293
Table 2: Transmission mode table in the case of perfect channel
SNR estimation.
Thr/put
BER
4-QAM
1/2
4-QAM
2/3
4-QAM
3/4
16-QAM
1/2
10
3
> 3.6 > 5.6 > 6.6 > 8.6
10
4
> 4.2 > 6.4 > 7.6 > 9.2
10
5
> 4.7 > 7 > 8.4 > 9.8
10
6
> 5 > 7.6 > 8.9 > 10.7
Thr/put
BER
16-QAM
2/3
16-QAM
3/4
64-QAM
2/3
64-QAM
3/4
10
3
> 11 > 12.2 > 15.9 > 17.3
10
4
> 11.7 > 12.9 > 16.5 > 17.9
10
5
> 12.3 > 13.6 > 16.9 > 18.6
10
6
> 13.1 > 14.5 > 17.5 > 19.8
with equivalent SNR. Ideally, an estimation process should
be included for assessing system performance as a function
of the actual measured channel, which would then be the in-
put to the optimization. Using this procedure in Stingray, the
related SNR uctuation resulted in marginal performance
degradation.
Based on those curves, and assuming perfect channel-
SNR estimation at the receiver, the derived TMT is presented
in Table 2.
By use of this table, the average system throughput (ST)
for various BER requirements is presented in Figure 10. The
system outage capacity (1%) is a good measure of through-
put evaluation of the system and is also plotted in the same
gure. The average capacity is also plotted, in order to show
the dierence from the performance upper bound.
The system throughput is very close to the 1% outage ca-
pacity, but it is 5 to 7 dB away from the performance limit,
depending on the BER level. Since the system is adaptive,
probably the 1% outage is not a suitable performance tar-
get for this system. The SNR gain achieved by going from
one BER level to the next is about 0.8 dB. This marginal gain
is expected due to the performance behavior of turbo codes
(very steep performance curves at BER regions of interest).
5.3. Flexible algorithms for phase noise and residual
frequency offset estimation
Omnipresent nuisances such as phase noise (PHN) and
residual frequency osets (RFO), which are the result of a
nonideal synchronization process, compromise the orthogo-
nality between the subcarriers of the OFDM systems (both
SISO and MIMO). The resulting eect is a Common Er-
ror (CE) for all the subcarriers of the same OFDM sym-
bol plus ICI. Typical systems adopt CE compensation algo-
rithms, while the ICI is treated as an additive, Gaussian, un-
correlated per subcarrier noise parameter [34]. The phase-
impairment-correction schemes developed in Stingray and
WF can be implemented either by the use of pilot symbols or
by decision-directed methods. They are transparent to the se-
lection of the Space-Time coding scheme, and they are easily
adaptable to any number of Tx/Rx antennas, down to the
7
6
5
4
3
2
1
0
B
i
t
s
/
c
a
r
r
i
e
r
0 5 10 15
Channel SNR
SAC
SOC
ST, QOS = 10
3
ST, QOS = 10
4
ST, QOS = 10
5
ST, QOS = 10
6
Figure 10: TSD-turbo system throughput (perfect CSI-SNR esti-
mation).
1 1 (SISO) case. In [35, 36] it is shown that the quality
of the CE estimate, which is typically characterized by the
Variance of the estimation error (VEE), aects drastically
the performance of the ST-OFDM schemes. In [34, 35, 36]
it is shown that the VEE is a function of the number and
the position of the subcarriers used for estimation purposes,
of the corresponding channel taps and of the pilot modu-
lation method (when pilot-assisted modulation methods are
adopted). Figure 11 depicts the dependence of the symbol er-
ror rate of an Alamouti STC OFDMsystemwith tentative de-
cisions on the number of subcarriers assigned for estimation
purposes. It is clear that this system is very sensitive to the
estimation error, and therefore to the selection of the corre-
sponding pilot number.
Additionally, the working range of the decision-directed
approaches is mainly dictated by the mean CE and the SNR,
which should be such that most of the received symbols are
within the bounds of correct decisions (i.e., the resulting er-
ror from the tentative decisions should be really small). This
may be dicult to ensure, especially when transmitting high-
order QAM constellations. An improved supervisor has to
take into account the eect of the residual CE error on the
overall system performance for selecting the optimal triplet,
by inserting its eect into the overall calculations.
Two approaches can be followed for the system optimiza-
tion. When the systemprotocol forces a xed number of pilot
symbols loaded on xed subcarriers (as in Hiperman), the
corresponding performance loss is calculated and the possi-
ble triplets are decided. It is noted that an enhanced super-
visor device could decide on the use of adaptive pilot modu-
lation in order to minimize estimation errors by maximizing
the received energy, since the pilot modulation may signif-
icantly aect the system performance. Figure 12 depicts the
eect of the pilot modulation method for the 2 2 Alamouti
294 EURASIP Journal on Wireless Communications and Networking
10
0
10
1
10
2
10
3
10
4
S
E
R
6 7 8 9 10 11 12 13 14 15
E
b
/N
0
(dB)
No PHN, no RFO
L
0
= 256
L
0
= 8
Figure 11: Eect of number of subcarriers used for estimation pur-
poses on decision-directed, 2 2 ST-OFDM system, Alamouti en-
coded, loaded with 16-QAM.
ST-OFDM system including 8 pilots, 256 subcarriers, and as-
suming independent compensation per receiver antenna for
a realization of an SUI-4 channel. Three modulation meth-
ods are considered: randomly generated pilots (RGPs), or-
thogonal generated pilots (OGPs), and xed pilot pattern
(FPP), where the same pilots are transmitted from any Tx
antenna. Thus, the selection of the pilot modulation scheme
is another parameter to be decided, since its aects system
performance in a signicant way.
On the other hand, when the system protocol allows for
a variable number of pilot symbols, the optimization proce-
dure becomes more complex. After a training period of some
OFDM symbols, the mean CE can be roughly estimated.
Using this estimate and taking into account that the whole
OFDM symbol is loaded with the same QAM constellation,
it can be decided whether a specically chosen constellation
is robust to the CE, so that the decision directed methods
(based on tentative decisions) are reliable. For the constella-
tions where the pilot-symbol use is necessary, the supervisor
has to select appropriately the position and the number of
pilot symbols.
6. TOWARDS A FLEXIBLE ARCHITECTURE
As already mentioned, a exible transceiver must be
equipped with the appropriate robust solutions for all possi-
ble widely ranging environments/system congurations. To
target the universally best possible performance translates to
high complexity. A rst step towards a generic exible ar-
chitecture should be one that eciently incorporates simple
tools in order to deliver not necessarily the best possible, but
an acceptable performance under disparate system/channel
environments.
10
0
10
1
10
2
10
3
10
4
S
E
R
5 6 7 8 9 10 11 12 13 14
E
b
/N
0
(dB)
No PHN, no RFO
RGP
OGP
FPP
Figure 12: Eect of pilot modulation on 2 2 ST-OFDM system,
Alamouti encoded, loaded with 16-QAM (L
0
= 8; 2 estimators).
The aforementioned CWSCE and TSD methods do be-
long to this category of exible (partial) solutions. The ca-
pacity penalty for their use (compared to the optimal solu-
tions) has been shown herein to be small. Both require com-
mon feedback information (1 bit/carrier) and can be incor-
porated appropriately in a system able to work under a va-
riety of antenna congurations, when such limited feedback
information is available. When feedback information is not
available, CWSCE has the appropriate modules for mode se-
lection (algorithm 1) for the SISO case, while Alamouti can
be the choice for the MIMO case. Both STC schemes trans-
form the MIMO channel into an inner SISO one, allowing
for the use of AMC (mode selection) techniques designed for
SISO systems. In the Stingray system, as already explained,
the average ESNR at the demodulator is a sucient met-
ric for choosing the Tx mode, whereas in WIND-FLEX the
uncoded BER is, respectively, used. Employing TMT tables
with the required uncoded BER and code-rate/constellation-
size sets for all the QoS operation modes in MIMO sys-
tems will increase the complexity, but it will permit seam-
less incorporation of both systems into one single common
architecture. The uncoded performance of the eective chan-
nel is thus the only metric that need be used for choosing
the Tx mode and can be computed for a variety of STC op-
tions. Furthermore, the fully parametric PHN and RFO algo-
rithms mentioned above are transparent to the selection of
the ST coding scheme and can provide the appropriate in-
formation about their performance under dierent environ-
ments/modes.
The overall block diagram of a proposed architecture
for the mode selection algorithm is given in Figure 13. It is
meant to be able to work for all systems employing one or
two antennas at the Tx/Rx.
Flexible Radio Framework for Optimized Multimodal Operation 295
List of supported
channel codes
Competitive
triplet evaluation
List of supported
constellations
Required uncoded
BER LUT
WSCE
& PCE
Eective channel/
noise estimator
WSCE/TSD/ALA
PHN/RFO
estimator
Mode Tx power
evaluation
[VEE(z
1
), . . . , VEE(z
l
)]
_
_
_
_
_
(x
1
, y
1
, z
1
)
.
.
.
(x
n
, y
n
, z
n
)
_
_
_
_
_
[(x
1
, y
1
), . . . , (x
n
, y
n
)]
.
.
.
.
.
.
[Pos(z
1
), . . . , Pos(z
n
)]
_
_
_
_
_
(x
1
, RUB
1
)
.
.
.
(x
n
, RUB
n
)
_
_
_
_
_
H
EF
N
EF
0
H
EF
H
EF
N
EF
0
_
_
_
_
_
PN(x
1
)
.
.
.
PN(x
l
)
_
_
_
_
_
Target
throughput
WSCE
(on/o)
_
_
_
_
_
PTx
1
.
.
.
PTx
n
_
_
_
_
_
Target BER
Operation mode
Figure 13: Block diagram of proposed algorithm for mode selection.
The related parameters are dened as follows:
(i) PN(x
i
), i = 1, . . . , l, is the number of needed pilots for
a specic PHN/RFO performance, when the operation
mode enables variable number of pilots;
(ii)

H
EF
is the vector of the estimated eective channel
gains in the frequency domain (STC dependent);
(iii) PCE : pilot carrier excision (an enhancement of the
WSCE module which provides the pilot positions for a
given number of used pilots).
Here, WSCE is active only when the system is 1 1. On
all other Tx-Rx antenna choices, all subcarriers are assumed
on. When only a xed number of pilot symbols are permit-
ted (e.g., when a specic protocol is used), the PHN/RFO es-
timator provides the VEE for each constellation choice to the
Tx power evaluation module. In a peer-to-peer communica-
tion system, where two exible terminals could have the pos-
sibility of reconguring to a specic PHY, the number of pi-
lots can be allowed to change and the optimum solution de-
pends on the constellation size. The competitive-triplet eval-
uation must take this variable pilot number into account.
The supervisor module is responsible for this optimization
procedure. The best choice depends not only on the chan-
nel/system characteristics but also on the selected optimiza-
tion criteria such as maximizing the throughput, minimizing
the Tx power, and so on.
7. CONCLUSIONS
The scientic eld of radio exibility is growing in impor-
tance and appeal. Although still in fairly nascent form for
commercial use, exible radio possesses attractive features
and attributes that require further research. The present pa-
per presents the exibility concept, denition, and related
scenarios while, in parallel, explores in some depth the tool
of dynamic signal design for instantiating some of these at-
tributes in a specic application environment. Two design
approaches are presented (based on the WF and Stingray
projects) and the key algorithmic choices of both are pre-
sented and incorporated into one exible design capable of
successfully operating in a variety of environments and sys-
tem congurations. It is evident that physical-layer exi-
bility requires not only novel system architectures but also
new algorithms that eciently utilize existing and/or new
modulation/coding techniques that can be adjusted to var-
ious environment and system scenarios, in order to oer
QoS close to that delivered by corresponding point-optimal
solutions.
ACKNOWLEDGMENT
The work presented in this paper has been supported by
STINGRAY (IST-2000-30173) and WIND-FLEX (IST-1999-
10025) projects that have been partly funded by the Euro-
pean Commission.
REFERENCES
[1] A. Polydros, J. Rautio, G. Razzano, et al., WIND-FLEX: de-
veloping a novel testbed for exploring exible radio concepts
in an indoor environment, IEEE Commun. Mag., vol. 41,
no. 7, pp. 116122, 2003.
[2] A. J. Goldsmith and S.-G. Chua, Variable-rate variable-
power MQAM for fading channels, IEEE Trans. Commun.,
vol. 45, no. 10, pp. 12181230, 1997.
[3] T. Keller and L. Hanzo, Adaptive modulation techniques
for duplex OFDM transmission, IEEE Trans. Veh. Technol.,
vol. 49, no. 5, pp. 18931906, 2000.
[4] F. K. Jondral, Flexible spectrum allocation and cognitive ra-
dio, in Proc. 11th National Symposium of Radio Science (URSI
05), pp. 2732, Poznan, Poland, April 2005.
296 EURASIP Journal on Wireless Communications and Networking
[5] G. Masera, Flexible hardware for wireless communications:
a case of study, in Proc. 11th National Symposium of Radio
Science (URSI 05), Poznan, Poland, April 2005.
[6] E. Buracchini, The software radio concept, IEEE Commun.
Mag., vol. 38, no. 9, pp. 138143, 2000.
[7] J. Mitola and G. Maguire, Cognitive radio: making software
radios more personal, IEEE Personal Communications Maga-
zine, vol. 6, no. 4, pp. 1318, 1999.
[8] R. van Nee and R. Prasad, OFDM Wireless Multimedia Com-
munications, Artech House, Boston, Mass, USA, 2000.
[9] R. Prasad and S. Hara, Overview of multicarrier CDMA,
IEEE Commun. Mag., vol. 35, no. 12, pp. 126133, 1997.
[10] L. Hanzo, M. Munster, B. J. Choi, and T. Keller, OFDM
and MC-CDMA for Broadband Multi-User Communications,
WLANs and Broadcasting, John Wiley & Sons, New York, NY,
USA, 2003.
[11] L.-L. Yang and L. Hanzo, Performance of generalized mul-
ticarrier DS-CDMA over Nakagami-mfading channels, IEEE
Trans. Commun., vol. 50, no. 6, pp. 956966, 2002.
[12] L.-L. Yang and L. Hanzo, Multicarrier DS-CDMA: a multiple
access scheme for ubiquitous broadband wireless communi-
cations, IEEE Commun. Mag., vol. 41, no. 10, pp. 116124,
2003.
[13] A. J. Paulraj, D. A. Gore, R. U. Nabar, and H. Bolcskei, An
overview of MIMO communicationsa key to gigabit wire-
less, Proc. IEEE, vol. 92, no. 2, pp. 198218, 2004.
[14] M. Juntti, M. Vehkapera, J. Leinonen, et al., MIMO MC-
CDMA communications for future cellular systems, IEEE
Commun. Mag., vol. 43, no. 2, pp. 118124, 2005.
[15] M. Vehkaper a, D. Tujkovic, Z. Li, and M. Juntti, Layered
space-frequency coding and receiver design for MIMO MC-
CDMA, in Proc. International Conference on Communications
(ICC 04), vol. 5, pp. 30053009, Paris, France, June 2004.
[16] J. Tang and X. Zhang, Link-adaptation-enhanced dynamic
channel allocation for MIMO-OFDM wireless networks, in
Proc. IEEE Wireless Communications and Networking Confer-
ence (WCNC 05), New Orleans, La, USA, March 2005.
[17] J.-H. Deng, J.-Y. Wu, and T.-S. Lee, Space-path spreading for
high rate MIMOMC-CDMAsystems with transmit diversity,
in Proc. IEEE Global Telecommunications Conference (GLOBE-
COM 03), vol. 6, pp. 33973401, San Francisco, Calif, USA,
December 2003.
[18] Y. Kim, T. Kim, C. Kang, and D. Hong, Union bound of max-
imum likelihood detection in downlink MIMO MC-CDMA
systems, in Proc. International Conference on Communica-
tions (ICC 04), vol. 6, pp. 32803283, Paris, France, June
2004.
[19] V. Tarokh, N. Seshadri, and A. R. Calderbank, Space-time
codes for high data rate wireless communication: perfor-
mance criterion and code construction, IEEE Trans. Inform.
Theory, vol. 44, no. 2, pp. 744765, 1998.
[20] Y. Gong and K. B. Letaief, Concatenated space-time block
coding with trellis coded modulation in fading channels,
IEEE Transactions on Wireless Communications, vol. 1, no. 4,
pp. 580590, 2002.
[21] B. Hochwald and S. ten Brink, Achieving near-capacity on
a multiple-antenna channel, IEEE Trans. Commun., vol. 51,
no. 3, pp. 389399, 2003.
[22] L. J. Cimini, Analysis and simulation of a digital mobile chan-
nel using orthogonal frequency division multiplexing, IEEE
Trans. Commun., vol. 33, no. 7, pp. 665675, 1985.
[23] ETSI, Broadband Radio Access Networks (BRAN)
HIPERMAN, OFDM Physical (PHY) layer, draft, 2002.
[24] I. Dagres and A. Polydoros, Dynamic transceivers: adaptiv-
ity and recongurability at the signal-design level, in Proc.
Software Dened Radio Technical Conference and Product Ex-
position (SDR 03), Orlando, Fla, USA, November 2003.
[25] K. Babionitakis, I. Dagres, K. Nakos, and D. Reisis, A VLSI
architecture for minimizing the transmission power in OFDM
transceivers, in Proc. 10th IEEE International Conference on
Electronics, Circuits and Systems (ICECS 03), vol. 1, pp. 308
311, Sharjah, United Arab Emirates, December 2003.
[26] P. Chow, J. M. Cio, and J. A. C. Bingham, A practical dis-
crete multitone transceiver loading algorithm fordata trans-
mission over spectrally shaped channels, IEEE Trans. Com-
mun., vol. 43, no. 234, pp. 773775, 1995.
[27] G. Foschini and M. Gans, On the limits of wireless commu-
nications in a fading environment when using multiple anten-
nas, Wireless Personal Communications, vol. 6, no. 3, pp. 311
335, 1998.
[28] M. Benedix, J. Ertel, and A. Finger, Turbo coding for an
OFDM-based wireless LAN at 17 GHz, in Proc. 7th Inter-
national OFDM-Workshop, pp. 137141, Hamburg, Germany,
September 2002.
[29] R. Knopp and G. Caire, Power control and beamforming
for systems with multiple transmit and receive antennas,
IEEE Transactions on Wireless Communications, vol. 1, no. 4,
pp. 638648, 2002.
[30] S. Alamouti, A simple transmit diversity technique for wire-
less communications, IEEE J. Select. Areas Commun., vol. 16,
no. 8, pp. 14511458, 1998.
[31] I. Dagres, A. Zalonis, and A. Polydoros, An ecient adap-
tive space time coding scheme for MIMO-OFDM systems, in
Proc. 14th IST Mobile and Wireless Telecommunications Sum-
mit, Dresden, Germany, June 2005, to appear.
[32] I. Harjula, S. Boumard, and A. M. Gallardo, WP4-Training
symbol design for Stingray demonstrator, 2002, (document
Stingray 4VTTN009d.doc in IST-2000-30173 STINGRAY).
[33] V. Erceg, K. V. S. Hari, M. S. Smith, and D. S. Baum, Channel
Models for Fixed Wireless Applications, 2001, IEEE 802.16.3
Task Group Contributions, Doc. IEEE 802.16.3.c-01/29r4.
[34] K. Nikitopoulos and A. Polydoros, Decision-directed com-
pensation of phase noise and residual frequency oset in a
Space-Time OFDM receiver, IEEE Commun. Lett., vol. 8,
no. 9, pp. 573575, 2004.
[35] K. Nikitopoulos and A. Polydoros, Phase-impairment ef-
fects and compensation algorithms for OFDM systems, IEEE
Trans. Commun., vol. 53, no. 4, pp. 698707, 2005.
[36] K. Nikitopoulos and A. Polydoros, Phase noise and resid-
ual frequency oset compensation in space-time OFDM sys-
tems, in Proc. IST Mobile and Wireless Telecommunications
Summit, pp. 7782, Aveiro, Portugal, June 2003.
Ioannis Dagres holds a B.S. degree in com-
puter engineering and an M.S. degree in sig-
nal processing from the Technical Univer-
sity of Patras, Greece, in 1997 and 1999, re-
spectively. He is currently working towards
his Ph.D. degree at the same university.
Since 1999, he has participated in several
EU projects such as ADAMAS (ADAptive
Multicarrier Access System), WIND-FLEX
(Wireless INDoor FLEXible Modem Archi-
tecture), and STINGRAY (Space Time CodING for Adaptive Re-
congurAble Systems) as a Research Associate. His research inter-
ests include the topics of signal processing for communications
and, in particular, adaptive signal design for modern communica-
tion systems.
Flexible Radio Framework for Optimized Multimodal Operation 297
Andreas Zalonis received the B.S. degree
in physics and the M.S. degree in telecom-
munications and electronics, both from the
National Kapodistrian University of Athens
(NKUA), Greece, in 2000 and 2002, re-
spectively. He is currently working towards
his Ph.D. degree in fourth-generation dig-
ital communications systems at the same
university. Since 2001, he has participated
in several national and European research
projects in the area of wireless communications. Since 2003 he
works as a Research Associate in the Institute of Accelerating Sys-
tems and Applications (IASA), NKUA.
Nikos Dimitriou holds a Diploma degree
in electrical and computer engineering from
the National Technical University of Athens,
Greece (1996), an M.S. degree with distinc-
tion in mobile and satellite communica-
tions from the University of Surrey, United
Kingdom (1997), and a Ph.D. degree from
the same University (2000). During that
period his work was focused on radio re-
source management issues and techniques
for next-generation mobile multimedia communication systems
(UMTS, W-CDMA). He participated in projects related to trac
modeling and network dimensioning for UMTS and GPRS (led
by Ericsson and One2One), worked as an MSc tutor and super-
vised various MSc projects. Since 2001, he has been working in
the National Kapodistrian University of Athens and specically
in the Institute of Accelerating Systems and Applications (IASA),
as a Senior Research Associate, coordinating the involvement of
the institute in several EU-IST projects such as ADAMAS (ADAp-
tive Multicarrier Access System), WIND-FLEX (Wireless INDoor
FLEXible Modem Architecture), SATIN (SATellite UMTS Ip Net-
work), STINGRAY (Space Time CodING for Adaptive Recong-
urAble Systems), and in the Network of Excellence in Wireless
Communications (NEWCOM). His research is focused on cross-
layer optimization techniques for next-generation exible wireless
systems.
Konstantinos Nikitopoulos was born in
Athens, Greece, in 1974. He received the
B.S. degree in physics and the M.S. de-
gree in electronics and telecommunications
from the National Kapodistrian University
of Athens (NKUA), Athens, Greece, in 1997
and 1999, respectively. Since 2005, he holds
a Ph.D. degree from the same university.
Since 1999, he has been a Research Associate
at the Institute of Accelerating Systems and
Applications, Athens, Greece, and has participated in several EUre-
search projects. Since October 2003, he has been with the General
Secretariat for Research and Technology of the Hellenic Ministry
of Development, Athens, Greece. Since October 2004, he has been
also a part-time Instructor in the Electrical Engineering Depart-
ment, the Technological Educational Institute of Athens, Athens,
Greece. His research interests include the topics of signal processing
for communications and, in particular, in synchronization for mul-
ticarrier systems. Dr. Nikitopoulos is a National Delegate of Greece
to the Joint Board on Communication Satellite Programmes of the
European Space Agency (ESA).
Andreas Polydoros was born in Athens,
Greece, in 1954. He received the Diploma
in electrical engineering from the National
Technical University of Athens in 1977, the
M.S.E.E. degree from the State University of
New York at Bualo in 1979, and the Ph.D.
degree in electrical engineering from the
University of Southern California in 1982.
He was a faculty member in the Department
of Electrical Engineering-Systems and the
Communication Sciences Institute (CSI), the University of South-
ern California, in 19821997, and has been a Professor since 1992.
He codirected CSI in 19911993. Since 1997, he has been a Pro-
fessor and Director of the Electronics Laboratory, Division of Ap-
plied Physics, Department of Physics, University of Athens, Greece.
His general areas of scientic interest are statistical communication
theory and signal processing with applications to spread-spectrum
and multicarrier systems, signal detection and classication in un-
certain environments (he is the coinventor of per-survivor process-
ing, granted a US patent in 1995), and multiuser radio networks.
He cofounded and currently heads the Technology Advisory Board
of Trellis Ware Technologies, and also serves on numerous scientic
boards of academic nonprot organizations in Greece and abroad.
Professor Polydoros is the recipient of a 1986 US National Science
Foundation Presidential Young Investigator Award. He became a
Fellow of the IEEE in 1995, cited for contributions to spread-
spectrum systems and networks.
EURASIP Journal on Wireless Communications and Networking 2005:3, 298307
c 2005 Dragan Samardzija et al.
Adaptive Transmitter Optimization in Multiuser
Multiantenna Systems: Theoretical Limits, Effect
of Delays, and Performance Enhancements
Dragan Samardzija
Wireless Research Laboratory, Bell Labs, Lucent Technologies, 791 Holmdel-Keyport Road, Holmdel, NJ 07733, USA
Email: dragan@bell-labs.com
Narayan Mandayam
Wireless Information Network Laboratory (WINLAB), Rutgers University, 73 Brett Road, Piscataway, NJ 08854-8060, USA
Email: narayan@winlab.rutgers.edu
Dmitry Chizhik
Wireless Research Laboratory, Bell Labs, Lucent Technologies, 791 Holmdel-Keyport Road, Holmdel, NJ 07733, USA
Email: chizhik@bell-labs.com
Received 6 March 2005; Revised 21 April 2005
The advances in programmable and recongurable radios have rendered feasible transmitter optimization schemes that can greatly
improve the performance of multiple-antenna multiuser systems. Recongurable radio platforms are particularly suitable for
implementation of transmitter optimization at the base station. We consider the downlink of a wireless system with multiple
transmit antennas at the base station and a number of mobile terminals (i.e., users) each with a single receive antenna. Under
an average transmit power constraint, we consider the maximum achievable sum data rates in the case of (1) zero-forcing (ZF)
spatial prelter, (2) modied zero-forcing (MZF) spatial prelter, and (3) triangularization spatial prelter coupled with dirty-
paper coding (DPC) transmission scheme. We show that the triangularization with DPC approaches the closed-loop MIMO rates
(upper bound) for higher SNRs. Further, the MZF solution performs very well for lower SNRs, while for higher SNRs, the rates
for the ZF solution converge to the MZF rates. An important impediment that degrades the performance of such transmitter
optimization schemes is the delay in channel state information (CSI). We characterize the fundamental limits of performance in
the presence of delayed CSI and then propose performance enhancements using a linear MMSE predictor of the CSI that can be
used in conjunction with transmitter optimization in multiple-antenna multiuser systems.
Keywords and phrases: transmitter beamforming, dirty-paper coding, correlated channels, channel state information, MMSE
prediction.
1. INTRODUCTION
For a wide range of emerging wireless data services, the
application of multiple antennas appears to be one of the
most promising solutions leading to even higher data rates
and/or the ability to support greater number of users.
Multiple-transmit multiple-receive antenna systems rep-
resent an implementation of the MIMO (multiple-input
multiple-output) concept in wireless communications [1]
This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
that can provide high-capacity (i.e., spectral eciency) wire-
less communications in rich scattering environments. It has
been shown that the theoretical capacity (approximately) in-
creases linearly as the number of antennas is increased [1, 2].
With the advent of exible and programmable radio
technology, transmitter optimization techniques used in
conjunction with MIMOprocessing can provide even greater
gains in systems with multiple users. Recongurable ra-
dio platforms are particularly suitable for implementation
of transmitter optimization at the base station. Such opti-
mization techniques have great potential to enhance perfor-
mance on the downlink of multiuser wireless systems. From
an information-theoretic model, the downlink corresponds
Transmitter Optimization in Multiuser Multiantenna Systems 299
to the case of a broadcast channel [3]. Recent studies that
have also focussed on multiple-antenna systems with mul-
tiple users include [4, 5, 6, 7, 8, 9, 10] and the references
therein.
In this paper, we study multiple-antenna transmitter op-
timization (i.e., spatial preltering) schemes that are based
on linear preprocessing and transmit power optimization
(keeping the average transmit power conserved). Specically,
we consider the downlink of a wireless system with multiple
transmit antennas at the base station and a number of mo-
bile terminals (i.e., users) each with a single receive antenna.
We consider the maximum achievable sum data rates in the
case of (1) zero-forcing spatial prelter, (2) modied zero-
forcing spatial prelter, and (3) triangularization spatial pre-
lter coupled with dirty-paper coding transmission scheme
[11]. We study the relationship between the above schemes
as well as the impact of the number of antennas on perfor-
mance.
After characterizing the fundamental performance lim-
its, we then study the performance of the above transmitter
optimization schemes with respect to delayed channel state
information (CSI). The delay in CSI may be attributed to the
delay in feeding back this information from the mobiles to
the base station or alternately to the delays in the ability to
reprogram/recongure the transmitter prelter. Without ex-
plicitly characterizing the source and the nature of such de-
lays, we show how the performance of the above transmitter
optimization schemes is degraded by the CSI delay. In or-
der to alleviate this problem, we exploit correlations in the
channel by designing a linear MMSE predictor of the chan-
nel state. We then show how the application of the MMSE
predictor can improve performance of transmitter optimiza-
tion schemes under delayed CSI.
The paper is organized as follows. In Section 2 we de-
scribe the system model. In Section 3, we describe the var-
ious transmitter optimization schemes including their fun-
damental performance limits as well as the eect of delayed
CSI. In Section 4, a formal channel model capturing chan-
nel correlations and a linear MMSE predictor of the channel
state which is used to overcome the eect of delayed CSI are
presented.
2. SYSTEMMODEL
In the following we introduce the system model. We use a
MIMO model [1] that corresponds to a system presented in
Figure 1. It consists of M transmit antennas and N mobile
terminals (each with a single receive antenna). In other words
each mobile terminal presents a MISO channel as seen from
the base station.
In Figure 1, x
n
is the information bearing signal intended
for mobile terminal n and y
n
is the received signal at the cor-
responding terminal (for n = 1, . . . , N). The received vector
y = [y
1
, . . . , y
N
]
T
is
y = HSx + n, y C
N
, x C
N
, n C
N
,
S C
MN
, H C
NM
,
(1)
MIMO
channel
H
Mobile 1
y
1
Mobile 2
y
2
Mobile N
y
N
1
2
M
Transmitter
TX
transform
S
x
1
x
2
x
N
.
.
.
.
.
.
.
.
.
Figure 1: System model consisting of M transmit antennas and N
mobile terminals.
where x = [x
1
, . . . , x
N
]
T
is the transmitted vector (E[xx
H
] =
P
av
I
NN
), n is AWGN (E[nn
H
] = N
0
I
NN
), H is the MIMO
channel response matrix, and S is a transformation (spatial
preltering) performed at the transmitter. Note that the vec-
tors x and y have the same dimensionality. Further, h
nm
is
the nth row and mth column element of the matrix H corre-
sponding to a channel between mobile terminal n and trans-
mit antenna m. If not stated otherwise, we will assume that
N M.
Application of the spatial preltering results in the com-
posite MIMO channel G given as
G = HS, G C
NN
, (2)
where g
nm
is the nth row and mth column element of the
composite MIMO channel response matrix G. The signal re-
ceived at the nth mobile terminal is
y
n
= g
nn
x
n
. , .
Desired signal for user n
+
N
_
i=1, i=n
g
ni
x
i
. , .
Interference
+n
n
. (3)
In the above representation, the interference is the signal that
is intended for mobile terminals other than terminal n. As
said earlier, the matrix S is a spatial prelter at the transmit-
ter. It is determined based on optimization criteria that we
address in the next section and has to satisfy the constraint
trace
_
SS
H
_
N (4)
which keeps the average transmit power conserved. We rep-
resent the matrix S as
S = AP, A C
MN
, P C
NN
, (5)
where A is a linear transformation and P is a diagonal ma-
trix. P is determined such that the transmit power remains
conserved.
300 EURASIP Journal on Wireless Communications and Networking
3. TRANSMITTER OPTIMIZATIONSCHEMES
Considering dierent forms of the matrix A, we study the
following transmitter optimization schemes.
(1) Zero-forcing (ZF) spatial preltering scheme where A
is represented by
A = H
H
_
HH
H
_
1
. (6)
As can be seen, for N M, the above linear transformation
is zeroing the interference between the signals dedicated to
dierent mobile terminals, that is, HA = I
NN
. x
n
are as-
sumed to be circularly symmetric complex random variables
having Gaussian distribution N
C
(0, P
av
). Consequently, the
maximum achievable data rate (capacity) for mobile termi-
nal n is
R
ZF
n
= log
2
_
1 +
P
av
|p
nn
|
2
N
0
_
, (7)
where p
nn
is the nth diagonal element of the matrix P dened
in (5). In (6) it is assumed that HH
H
is invertible, that is, the
rows of H are linearly independent.
(2) Modied zero-forcing (MZF) spatial preltering
scheme that assumes
A = H
H
_
HH
H
+
N
0
P
av
I
_
1
. (8)
In the case of the above transformation, in addition to the
knowledge of the channel H, the transmitter has to know the
noise variance N
0
. x
n
are assumed to be circularly symmet-
ric complex random variables having Gaussian distribution
N
C
(0, P
av
). The maximum achievable data rate (capacity) for
mobile terminal n now becomes
R
MZF
n
= log
2
_
1 +
P
av
|g
nn
|
2
P
av

N
i=1, i=n
|g
ni
|
2
+ N
0
_
. (9)
While the transformation in (8) appears to be similar in form
to an MMSE linear receiver, the important dierence is that
the transformation is performed at the transmitter. Using the
virtual uplink approach for transmitter beamforming (intro-
duced in [7, 8]), we present the following proposition.
Proposition 1. If the nth diagonal element of P is selected as
p
nn
=
1
_
a
H
n
a
n
(n = 1, . . . , N), (10)
where a
n
is the nth column vector of the matrix A, the constraint
in (4) is satised with equality. Consequently, the achievable
downlink rate R
MZF
n
for mobile n is identical to its correspond-
ing virtual uplink rate when an optimal uplink linear MMSE
receiver is applied.
See Appendix A for a denition of the corresponding vir-
tual uplink and a proof of the above proposition.
(3) Triangularization spatial preltering with dirty-paper
coding (DPC) where the matrix A assumes the form
A = H
H
R
1
, (11)
where H = (QR)
H
and Qis unitary and R is upper triangular
(see [12] for details on QR factorization). In general, R
1
is a
pseudoinverse of R. The composite MIMO channel G in (2)
becomes G = L = HS, a lower triangular matrix. It permits
application of dirty-paper coding designed for single-input
single-output (SISO) systems. We refer the reader to [4, 5, 6,
13, 14, 15, 16] for further details on the DPC schemes.
By applying the transformation in (11), the signal in-
tended for terminal 1 is received without interference. The
signal at terminal 2 suers from the interference arising from
the signal dedicated to terminal 1. In general, the signal at
terminal n suers from the interference arising from the sig-
nals dedicated to terminals 1 to n 1. In other words,
y
1
= g
11
x
1
+ n
1
,
y
2
= g
22
x
2
+ g
21
x
1
+ n
2
,
.
.
.
y
n
= g
nn
x
n
+
n1
_
i=1
g
ni
x
i
+ n
n
,
.
.
.
y
N
= g
NN
x
N
+
N1
_
i=1
g
Ni
x
i
+ n
N
.
(12)
Since the interference is known at the transmitter, DPC can
be applied to mitigate the interference (the details are given
in Appendix B). Based on the results in [13], the achievable
rate for mobile terminal n is
R
DPC
n
= log
2
_
1 +
P
av
|g
nn
|
2
N
0
_
= log
2
_
1 +
P
av
|r
nn
p
nn
|
2
N
0
_
,
(13)
where r
nn
is the nth diagonal element of the matrix R dened
in (11). Note that DPC is applied just in the case of the lin-
ear transformation in (11), with corresponding rate given by
(13).
Note that trace(AA
H
) = N, thereby satisfying the con-
straint in (4). Consequently, we can select P = I
NN
and
present the following proposition.
Proposition 2. For high SNR (P
av
N
0
) and P = I
NN
, the
achievable sum rate of the triangularization with DPC scheme
is equal to the rate of the equivalent (open loop) MIMO system.
In other words, for P
av
N
0
,
N
_
n=1
R
DPC
n
= log
2
_
det
_
I
NN
+
P
av
N
0
HH
H
__
. (14)
Transmitter Optimization in Multiuser Multiantenna Systems 301
Proof. Starting from the right-side term in (14) and with
HH
H
= R
H
R, for P
av
N
0
,
log
2
_
det
_
I
NN
+
P
av
N
0
R
H
R
__
log
2
_
det
_
P
av
N
0
R
H
R
__
= log
2
_
P
av
N
0

r
11

2

P
av
N
0

r
NN

2
_
=
N
_
i=1
log
2
_
P
av
N
0

r
ii

2
_

N
_
i=1
log
2
_
1 +
P
av
N
0

r
ii

2
_
=
N
_
n=1
R
DPC
n
(15)
which concludes the proof.
The ZF and MZF schemes should be viewed as trans-
mitter beamforming techniques using conventional channel
coding to approach the achievable rates [7, 8]. The trian-
gularization with DPC scheme is necessarily coupled with a
nonconventional coding, that is, the DPC scheme.
Once the matrix A is selected, the elements of the diag-
onal matrix P are determined such that the transmit power
remains conserved and the sum rate is maximized. The con-
straint on the transmit power is
trace
_
APP
H
A
H
_
N. (16)
The elements of the matrix P are selected such that
diag(P) =
_
p
11
, . . . , p
NN
_
T
= arg max
trace(APP
H
A
H
)N
N
_
i=1
R
n
.
(17)
3.1. Fundamental limits
To evaluate the performance of the above schemes, we con-
sider the following baseline solutions.
(1) No preltering solution where each mobile terminal
is served by one transmit antenna dedicated to that mobile.
This is equivalent to S = I. A transmit antenna is assigned
to a particular terminal corresponding to the best channel
(maximumchannel magnitude) among all available transmit
antennas and that terminal.
(2) Equal resource TDMA and coherent beamforming
(denoted as TDMA-CBF) is a solution where signals for dif-
ferent terminals are sent in dierent (isolated) time slots. In
this case, there is no interference, and each terminal is us-
ing 1/N of the overall resources. When serving a particular
mobile, ideal coherent beamforming is applied using all M
transmit antennas.
(3) Closed-loop MIMO (using the water-pouring opti-
mization on eigenmodes) is a solution that is used as an up-
per bound on the achievable sum rates. In the following, it
is denoted as CL-MIMO. This solution would require that
multiple terminals act as a joint multiple-antenna receiver.
0 5 10 15 20 25
SNR (dB)
0
1
2
3
4
5
6
7
8
9
A
v
e
r
a
g
e
u
s
e
r
r
a
t
e
(
b
i
t
s
/
s
y
m
b
o
l
)
No preltering
TDMA-CBF
ZF
MZF
DPC
CL-MIMO
Figure 2: Average rate per user versus SNR(M = 3, N = 3, Rayleigh
channel).
This solution is not practical because the terminals are nor-
mally individual entities in the network and they do not co-
operate when receiving signals on the downlink.
In Figure 2, we present average rates per user versus
SNR = 10 log (P
av
/N
0
) for a system consisting of M = 3
transmit antennas and N = 3 terminals. The channel is
Rayleigh, that is, the elements of the matrix H are complex
independent and identically distributed Gaussian random
variables with distribution N
C
(0, 1). From the gure we ob-
serve the following. The triangularization with DPC scheme
is approaching the closed-loop MIMO rates for higher SNR.
The MZF solution is performing very well for lower SNRs
(approaching CL-MIMO and DPC rates), while for higher
SNRs, the rates for the ZF scheme are converging to the MZF
rates. The TDMA-CBF rates are increasing with SNR, but still
signicantly lower than the rates of the proposed optimiza-
tion schemes. The solution where no preltering is applied
clearly exhibits properties of an interference limited system
(i.e., after a certain SNR, the rates are not increasing). Corre-
sponding cumulative distribution functions (cdf) of the sum
rates normalized by the number of users are given in Figure 3
for SNR = 10 dB (see more on the capacity-versus-outage
approach in [17]).
In Figure 4, we present the behavior of the average rates
per user versus number of transmit antennas. The average
rates are observed for SNR = 10 dB, N = 3, and variable
number of transmit antennas (M = 3, 6, 12, 24). The rates in-
crease with the number of transmit antennas and the dier-
ence between the rates for dierent schemes becomes smaller.
As the number of transmit antennas increases, while keeping
the number of users N xed, the spatial channels (i.e., rows of
the matrix H) are getting less cross-correlated (approaching
302 EURASIP Journal on Wireless Communications and Networking
0 1 2 3 4 5 6
Rate (bits/symbol)
ZF
TDMA-CBF
No preltering
MZF
DPC
CL-MIMO
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
P
r
o
b
a
b
i
l
i
t
y
Figure 3: CDF of rates, SNR = 10 dB, per user (M = 3, N = 3,
Rayleigh channel).
1 2 3 4 5 6 7 8
M/N
2
3
4
5
6
7
8
A
v
e
r
a
g
e
u
s
e
r
r
a
t
e
(
b
i
t
s
/
s
y
m
b
o
l
)
ZF
MZF
DPC
Figure 4: Average rate per user versus M/N (SNR = 10 dB, N =
3, variable number of transmit antennas M = 3, 6, 12, 24, Rayleigh
channel).
orthogonality for M ). It can be shown that for orthog-
onal channels, all three schemes perform identically.
We now illustrate a case when the number of available
terminals N
t
(i.e., users) is equal to or greater than the num-
ber of transmit antennas M. Out of N
t
terminals, the trans-
mitter will select N = M terminals and perform the above
transmitter optimization schemes for the selected set. There
are N
t
!/((N
t
M)!M!) possible sets. Between the transmit
3 4 5 6 7 8 9 10 11 12
Number of available terminals
2
2.5
3
3.5
4
4.5
5
5.5
A
v
e
r
a
g
e
u
s
e
r
r
a
t
e
(
b
i
t
s
/
s
y
m
b
o
l
)
ZF
MZF
DPC
Figure 5: Average rate per user versus number of available terminals
(SNR = 10 dB, M = 3, Rayleigh channel).
antennas and each terminal, there is (1 M)-dimensional
spatial channel. For each set of the terminals there is a matrix
channel H
j
C
MM
where each row corresponds to a dier-
ent spatial channel of the corresponding terminal in the set.
The selected terminals are the ones corresponding to the set
J = arg
j
min
_
_
_ H
H
j
_
H
j
H
H
j
_
1
_
_
_, (18)
where is the Frobenius norm. The above criterion will
favor the terminals whose spatial channels have low cross-
correlation. In Figure 5, we present the average rates per user
(the average sum rates divided by N = M) versus number of
available terminals. The increase in the rates with the number
of available terminals is a result of multiuser diversity (i.e.,
having more terminals allows the transmitter to select more
favorable channels).
3.2. Effect of CSI delay
As a motivation for the analysis presented in the follow-
ing sections, we now present the eects of imperfect chan-
nel state knowledge. In practical communication systems the
channel state H has to be estimated at the receivers, and
then fed to the transmitter. Specically, mobile terminal n
feeds back the estimate of the nth row of the matrix H, for
n = 1, . . . , N. In the case of a time-varying channel, this prac-
tical procedure results in noisy and delayed (temporally mis-
matched) estimates being available to the transmitter to per-
form the optimization. As said earlier, the MIMO channel
is time varying. Let H
i1
and H
i
correspond to consecutive
block-faded channel responses. The temporal characteristic
of the channel is described using the correlation
k =
E
_
h
(i1)nm
h

inm
_

, (19)
Transmitter Optimization in Multiuser Multiantenna Systems 303
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Correlation
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
5.5
A
v
e
r
a
g
e
u
s
e
r
r
a
t
e
(
b
i
t
s
/
s
y
m
b
o
l
)
No preltering
ZF
MZF
Figure 6: Average rate per user versus temporal channel correlation
k (SNR = 10 dB, M = 3 (solid lines), M = 6 (dashed lines), N = 3,
Rayleigh channel).
where = E[h
inm
h

inm
], and h
inm
is a stationary random pro-
cess (for m = 1, . . . , M and n = 1, . . . , N, denoting transmit-
and receive-antenna indices, respectively). Low values of the
correlation k correspond to higher mismatch between H
i1
and H
i
. Note that the above channel is modeled as a rst-
order discrete Markov process. In the case of the Jakes model,
k = J
0
(2 f
d
), where f
d
is the maximum Doppler frequency
and is the time dierence between h
(i1)nm
and h
inm
. In ad-
dition, the above simplied model assumes that there is no
spatial correlation.
We assume that the mobile terminals feed back H
i1
which is used at the base station to perform the transmitter
optimization for the ith block. In other words, the downlink
transmitter is ignoring the fact that H
i
= H
i1
. In Figure 6,
we present the average rate per user versus the temporal
channel correlation k in (19). From these results we note the
very high sensitivity of the schemes to the channel mismatch.
In this particular case, the performance of the ZF and MZF
schemes becomes worse than when there is no preltering.
See also [18] for a related study of channel mismatch and
achievable data rates for single-user MIMO systems. Note
that the above example and the model in (19) is a simpli-
cation that we only use to illustrate the schemes sensitivity
to imperfect knowledge of the channel state. In the following
section, we introduce a detailed channel model incorporat-
ing correlations in the channel state information.
4. CHANNEL STATE PREDICTIONFOR
PERFORMANCE ENHANCEMENT
In the following, we rst address the temporal aspects of the
channel H. For each mobile terminal, there is a (1 M)-
dimensional channel between its receive antenna and M
transmit antennas at the base station. The MISO channel
h
n
= [h
n1
h
nM
] for mobile terminal n (n = 1, . . . , N) cor-
responds to the nth row of the channel matrix H, and we
assume that it is independent of other channels (i.e., rows of
the channel matrix). The temporal evolution of the MISO
channel h
n
may be represented as [19, 20]
h
n
(t) = [1 1]D
n
N
n
, D
n
C
Nf Nf
, N
n
C
Nf M
,
(20)
where N
n
is an (N
f
M)- dimensional matrix with elements
corresponding to complex i.i.d. random variables with dis-
tribution N
C
(0, 1/N
f
). D
n
is an N
f
N
f
diagonal Doppler
shift matrix with diagonal elements
d
ii
= e
ji t
(21)
representing the Doppler shifts that aect N
f
plane waves
and

i
=
2

v
n
cos
_

i
_
, for i = 1, . . . , N
f
, (22)
where v
n
is the velocity of mobile terminal n and the angle of
arrival of the ith plane wave at the terminal is
i
(generated
as U[0 2]).
It can be shown that the model in (20) strictly conforms
to the Jakes model for N
f
. This model assumes that
at the mobile terminal the plane waves are coming from all
directions with equal probability. Further, note that each di-
agonal element of D
n
corresponds to one Doppler shift. D
n
and N
n
are independently generated. With minor modica-
tions, the above model can be modied to capture the spatial
correlations as well (see [21]).
We assume that the transmitter has a set of previous
channel responses (for mobile terminal n) h
n
(t) where t =
kT
ch
and k = 0, 1, . . . , (L 1). The time interval T
ch
may correspond to a period when a new CSI is sent from
the mobile terminal to the base station. Knowing that the
wireless channel has correlations, based on previous channel
responses the transmitter may perform a prediction of the
channel response h
n
() at the time moment . In this paper
we assume that the prediction is linear and that it minimizes
the mean square error between true and predicted channel
state. The MMSE predictor W
n
is
W
n
= min
T
arg E

T
H
h
un
h
n
()
H

2
, (23)
where h
un
is a vector dened as
h
un
=
_
h
n
(0)h
n
_
T
ch
_
h
n
_
(L 1)T
ch
__
T
. (24)
In other words, the vector is constructed by stacking up the
previous channel responses available to the transmitter. We
dene the following matrices:
U
n
= E
_
h
un
h
H
un
_
,
V
n
= E
_
h
un
h
n
()
_
.
(25)
304 EURASIP Journal on Wireless Communications and Networking
0 1 2 3 4 5 6 7 8
CSI delay (ms)
0.5
1
1.5
2
2.5
3
3.5
A
v
e
r
a
g
e
u
s
e
r
r
a
t
e
(
b
i
t
s
/
s
y
m
b
o
l
)
No preltering
ZF
MZF
Figure 7: Average rate per user versus CSI delay, with MMSE pre-
diction (dashed lines) and without MMSE prediction (solid lines)
(SNR = 10 dB, M = 3, N = 3, channel based on model in (20),
f
c
= 2 GHz, v = 30 kmph).
It can be shown that the linear MMSE predictor W
n
is [22]
W
n
= U
1
n
V
n
. (26)
The above predictor exploits the correlations of the MISO
channel. Note that dierent linear predictors are needed for
dierent mobile terminals.
A practical implementation of the above prediction can
use sample estimates of U
n
and V
n
as

U
n
=
1
N
w
1
_
i=Nw
h
un
_
iT
ch
_
h
un
_
iT
ch
_
H
,

V
n
=
1
N
w
1
_
i=Nw
h
un
_
iT
ch
_
h
n
_
+ iT
ch
_
.
(27)
Underlying assumption in using the above estimates is that
the channel is stationary over the integration window N
w
T
ch
.
Further, if the update of the CSI is performed at discrete time
moments kT
ch
(k = 0, 1, . . .), the update period T
ch
should
be such that
T
ch
<
1
2 f
doppler
. (28)
In Figure 7, we present the average rates per user versus
the delay of the CSI. The system consists of M = 3 trans-
mit antennas and N = 3 terminals. The channel is modeled
based on (20) (assuming that the carrier frequency is 2 GHz
and the velocity of each mobile terminal is 30 kmph and set-
ting the number of plane waves N
f
= 100). Because the ideal
channel state H(t + ) is not available at the transmitter, we
assume that H(t) is used instead to perform the transmitter
0 20 40 60 80 100 120
Velocity (km/h)
0.5
1
1.5
2
2.5
3
3.5
A
v
e
r
a
g
e
u
s
e
r
r
a
t
e
(
b
i
t
s
/
s
y
m
b
o
l
)
No preltering
ZF
MZF
Figure 8: Average rate per user versus terminal velocity, with
MMSE prediction (dashed lines) and without MMSE prediction
(solid lines) (SNR = 10 dB, M = 3, N = 3, channel based on model
in (20), f
c
= 2 GHz, = 2 milliseconds).
optimization at the moment t + . Figure 7 presents average
rates for the ZF and MZF schemes, for SNR = 10 dB. Results
depicted by the solid lines correspond to the application of
the delayed CSI H(t) instead of the true channel state H(t+).
The dashed lines depict results when the MMSE predicted
channel state H
MMSE
(t +) is used instead of the true channel
state H(t + ). Without any particular eort to optimally se-
lect the implementation parameters, in this particular exam-
ple, we use L = 10 previous channel responses to construct
the vectors in (24). Further, the length of the integration win-
dow in (27) is selected to be N
w
= 100. The results clearly
point to improvements in the performance of the schemes
when the MMSE channel state prediction is used. The re-
sults suggest that the temporal correlations in the channel
alone are signicant enough to support the application of
the MMSE prediction. The presence of spatial correlations in
the channel model will further improve the benets of such
channel state prediction schemes used in conjunction with
transmitter optimization.
For the above assumptions, in Figure 8 we present the av-
erage rates per user versus the terminal velocity with the CSI
delay = 2 milliseconds. From the results, we can see that
the prediction scheme signicantly extends the gains of the
transmitter optimization even for higher terminal velocities.
5. CONCLUSION
The advances in programmable and recongurable radios
have rendered feasible transmitter optimization schemes that
can greatly improve the performance of multiple-antenna
multiuser systems. In this paper, we presented a study on
multiple-antenna transmitter optimization schemes for mul-
tiuser systems that are based on linear transformations and
Transmitter Optimization in Multiuser Multiantenna Systems 305
transmit power optimization. We considered the maximum
achievable sum data rates in the case of the zero-forcing,
the modied zero-forcing, and the triangularization spatial
preltering coupled with the dirty-paper coding transmis-
sion scheme. We showed that the triangularization with DPC
approaches the closed-loop MIMO rates (upper bound) for
higher SNR. Further, the MZF solution performed very well
for lower SNRs (approaching closed-loop MIMO and DPC
rates), while for higher SNRs, the rates for the ZF scheme
converged to that of the MZF rates. A key impediment to the
successful deployment of transmitter optimization schemes
is the delay in the channel state information (CSI) that is
used to accomplish this. We characterized the degradation in
the performance of transmitter optimization schemes with
respect to the delayed CSI. A linear MMSE predictor of the
channel state was introduced which then improved the per-
formance in all cases. The results have suggested that the tem-
poral correlations in the channel alone are signicant enough
to support the application of the MMSE prediction. In the
presence of additional spatial correlations, the usefulness of
such prediction schemes will be even greater.
APPENDICES
A. DEFINITIONOF THE VIRTUAL LINK AND
PROOF OF PROPOSITION1
We now describe the corresponding virtual uplink for the
system in Figure 1. Let x
n
be the uplink information-bearing
signal transmitted from mobile terminal n (n = 1, . . . , N)
and let y
m
be the received signal at the mth base sta-
tion antenna (m = 1, . . . , M). x
n
are assumed to be cir-
cularly symmetric complex random variables having Gaus-
sian distribution N
C
(0, P
av
). Further, the received vector y =
[ y
1
, . . . , y
M
]
T
is
y =

H x + n = H
H
x + n, y C
M
, x C
N
,
n C
M
,

H C
MN
,
(A.1)
where x = [ x
1
, . . . , x
N
]
T
is the transmitted vector (E[ x x
H
] =
P
av
I
NN
), n is AWGN (E[ n n
H
] = N
0
I
MM
), and

H = H
H
is
the uplink MIMO channel response matrix.
It is well known that the MMSE receiver is the opti-
mal linear receiver for the uplink (multiple-access channel)
[23, 24]. It maximizes the received SINR (and rate) for
each user. The decision statistic is obtained after the receiver
MMSE ltering as
x
dec
= W
H
y, (A.2)
where the MMSE receiver is
W=
__
HH
H
+
N
0
P
av
I
_
1
H
_
H
= H
H
_
HH
H
+
N
0
P
av
I
_
1
.
(A.3)
Proof of Proposition 1. Note that W = A in (8), for the MZF
transmitter spatial preltering. We normalize the column
vectors of the matrix W in (A.3) as
W
nor
= WP, (A.4)
where P is dened in (10). In other words, the nth diagonal
element of P is selected as
p
nn
=
1
_
w
H
n
w
n
(n = 1, . . . , N), (A.5)
where w
n
is the nth column vector of the matrix W (where
w
n
= a
n
, which is the column vector of A for n = 1, . . . , N).
It is well known that any normalization of the columns of the
MMSE receiver in (A.3) does not change the SINRs. In other
words, the SINR for the nth uplink user (n = 1, . . . , N) is
SINR
UL
n
=
P
av

w
H
n

h
n

2
P
av

N
i=1, i=n

w
H
n

h
i

2
+ N
0
w
H
n
w
n
=
P
av

w
H
n

h
n

2
/
_
w
H
n
w
n
_
P
av

N
i=1, i=n

w
H
n

h
i

2
/
_
w
H
n
w
n
_
+ N
0
,
(A.6)
where

h
n
is the nth column vector of the matrix

H. Note
that

h
H
n
= h
n
which is the nth row vector of the downlink
MIMO channel H. The corresponding downlink SINR when
the MZF spatial perltering is used (with P dened in (10))
is
SINR
MZF
n
=
P
av

h
n
a
n

2
/
_
a
H
n
a
n
_
P
av

N
i=1, i=n

h
i
a
i

2
/
_
a
H
n
a
n
_
+ N
0
. (A.7)
As said earlier, w
n
= a
n
and

h
H
n
= h
n
. Thus, SINR
MZF
n
=
SINR
UL
n
for n = 1, . . . , N leading to identical rates, which
concludes the proof.
B. SPATIAL PREFILTERINGWITHDPC
One practical, but suboptimal, single-dimensional DPC so-
lution is described in [14, 15]. Starting from that solution we
introduce the DPC scheme.
The transmitted signal in (1) intended for terminal n is
x
n
= f
mod
_
x
n
I
n
_
, (B.1)
where x
n
is the information-bearing signal for terminal n and
f
mod
() is a modulo operation (i.e., a uniform scalar quan-
tizer). For a real variable x, f
mod
(x) is dened as
f
mod
(x) =
_
(x + Z) mod(2Z)
_
Z (B.2)
and in the case of a complex variable a + jb, f
mod
(a + jb) =
f
mod
(a) + j f
mod
(b). The constant Z is selected such that
E[x
n
x

n
] = P
av
. Further, from (12), I
n
is the normalized in-
terference at terminal n:
I
n
=
n1
_
i=1
g
ni
x
i
g
nn
, (B.3)
306 EURASIP Journal on Wireless Communications and Networking
assuming that g
nn
= 0. Note that I
n
is only known at
the transmitter. At terminal n, the following operation is
performed:
f
mod
_
y
n
g
nn
_
= x
n
+ n

n
, (B.4)
where n

n
is a wrapped-around AWGN (due to the nonlin-
ear operation f
mod
()). For high SNR and with x
n
being uni-
formly distributed over the single-dimensional region, the
achievable rate is approximately 1.53 dB away from the rate
in (13) [14, 15].
To further approach the rate in (13), based on [14], the
following modications of the suboptimal scheme in (B.1)
are needed. The transmitted signal intended for terminal n is
now
x
n
= f
k
_
x
n

n
I
n
+ d
n
_
, (B.5)
where f
k
() is a modulo operation over a k-dimensional re-
gion.
n
is a parameter to be optimized (0 <
n
1) and
d
n
is a dither (uniformly distributed pseudonoise over the k-
dimensional region). At terminal n, the following operation
is performed:
f
k
_
y
n
g
nn
_
= x
n
+
_
1
n
_
u
n
+
n
n

n
, (B.6)
where n

n
is a wrapped-around AWGN (due to the nonlin-
ear operation f
k
()) and u
n
is uniformly distributed over the
k-dimensional region. For k and x
n
being uniformly
distributed over the k-dimensional region, the rate in (13)
can be achieved [14]. Further details on selecting
n
and d
n
are beyond the scope of this paper. We refer the reader to [14]
and references therein.
ACKNOWLEDGMENT
This work is supported in part by the National Science Foun-
dation under Grant no. FMF 0429724.
REFERENCES
[1] G. J. Foschini, Layered space-time architecture for wireless
communication in a fading environment when using multiple
antennas, Bell Labs Technical Journal, vol. 1, no. 2, pp. 4159,
1996.
[2] G. J. Foschini and M. J. Gans, On limits of wireless commu-
nications in a fading environment when using multiple anten-
nas, Wireless Personal Communications, vol. 6, no. 3, pp. 311
335, 1998.
[3] T. M. Cover and J. A. Thomas, Elements of Information Theory,
John Wiley & Sons, New York, NY, USA, 1st edition, 1991.
[4] G. Caire and S. Shamai, On the achievable throughput of
a multiantenna Gaussian broadcast channel, IEEE Trans. In-
form. Theory, vol. 49, no. 7, pp. 16911706, 2003.
[5] D. N. C. Tse and P. Viswanath, On the capacity region of
the vector Gaussian broadcast channel, in Proc. IEEE Interna-
tional Symposium on Information Theory (ISIT 03), pp. 342
342, Yokohoma, Japan, JuneJuly 2003.
[6] S. Vishwanath, G. Kramer, S. Shamai , S. Jafar, and A. Gold-
smith, Capacity bounds for Gaussian vector broadcast chan-
nels, in Proc. DIMACS Workshop on Signal Processing for
Wireless Transmission, vol. 62, pp. 107122, Rutgers Univer-
sity, Piscataway, NJ, USA, October 2002.
[7] E. Rashid-Farrohi, L. Tassiulas, and K. J. R. Liu, Joint op-
timal power control and beamforming in wireless networks
using antenna arrays, IEEE Trans. Commun., vol. 46, no. 10,
pp. 13131324, 1998.
[8] E. Visotsky and U. Madhow, Optimum beamforming using
transmit antenna arrays, in Proc. 49th IEEE Vehicular Tech-
nology Conference (VTC 99), vol. 1, pp. 851856, Houston,
Tex, USA, May 1999.
[9] E. Visotsky and U. Madhow, Space-time transmit precoding
with imperfect feedback, IEEE Trans. Inform. Theory, vol. 47,
no. 6, pp. 26322639, 2001.
[10] S. Thoen, L. Van der Perre, M. Engels, and H. De Man,
Adaptive loading for OFDM/SDMA-based wireless net-
works, IEEE Trans. Commun., vol. 50, no. 11, pp. 17981810,
2002.
[11] D. Samardzija and N. Mandayam, Multiple antenna trans-
mitter optimization schemes for multiuser systems, in Proc.
58th IEEE Vehicular Technology Conference (VTC 03), vol. 1,
pp. 399403, Orlando, Fla, USA, October 2003.
[12] G. Strang, Linear Algebra and Its Applications, Harcourt Brace
Jovanovich, San Diego, Calif, USA, 3rd edition, 1988.
[13] M. H. M. Costa, Writing on dirty paper (Corresp.), IEEE
Trans. Inform. Theory, vol. 29, no. 3, pp. 439441, 1983.
[14] G. J. Foschini and A. H. Diaz, Dirty paper coding: perturb-
ing o the innite dimensional lattice limit, in Proc. DI-
MACS Workshop on Signal Processing for Wireless Transmis-
sion, vol. 62, pp. 141160, Rutgers University, Piscataway, NJ,
USA, October 2002.
[15] U. Erez, R. Zamir, and S. Shamai, Additive noise channels
with side information at the transmitter, in Proc. 21st IEEE
Convention of Electrical and ELectronic Engineers in Israel, pp.
373376, Tel-Aviv, Israel, April 2000.
[16] A. S. Cohen and A. Lapidoth, Generalized writing on dirty
paper, in Proc. IEEE International Symposium on Information
Theory (ISIT 02), pp. 227227, Lausanne, Switzerland, June
July 2002.
[17] E. Biglieri, J. Proakis, and S. Shamai, Fading channels:
information-theoretic and communications aspects, IEEE
Trans. Inform. Theory, vol. 44, no. 6, pp. 26192692, 1998.
[18] D. Samardzija and N. Mandayam, Pilot-assisted estimation
of MIMO fading channel response and achievable data rates,
IEEE Trans. Signal Processing, vol. 51, no. 11, pp. 28822890,
2003, Special Issue on MIMO.
[19] H. Xu, D. Chizhik, H. Huang, and R. Valenzuela, A wave-
based wideband MIMO channel modeling technique, in
Proc. 13th IEEE International Symposium on Personal, Indoor
and Mobile Radio Communications (PIMRC 02), vol. 4, pp.
16261630, Lisbon, Portugal, September 2002.
[20] D. Chizhik, Slowing the time-uctuating MIMO channel by
beam-forming, IEEE Transactions on Wireless Communica-
tions, vol. 3, no. 5, pp. 15541565, 2004.
[21] D. Samardzija and N. Mandayam, Downlink multiple an-
tenna transmitter optimization on spatially and temporally
correlated channels with delayed channel state information,
in Proc. Conference on Information Sciences and Systems (CISS
04), Princeton University, Princeton, NJ, USA, March 2004.
[22] S. Haykin, Adaptive Filter Theory, Prentice-Hall, Englewood
Clis, NJ, USA, 2nd edition, 1991.
[23] U. Madhow and M. L. Honig, MMSE interference sup-
pression for direct-sequence spread-spectrum CDMA, IEEE
Trans. Commun., vol. 42, no. 12, pp. 31783188, 1994.
[24] S. Verd u, Multiuser Detection, Cambridge University Press,
Cambridge, UK, 1998.
Transmitter Optimization in Multiuser Multiantenna Systems 307
Dragan Samardzija was born in Kikinda,
Serbia and Montenegro, in 1972. He re-
ceived the B.S. degree in electrical engineer-
ing and computer science in 1996 from the
University of Novi Sad, Serbia and Mon-
tenegro, and the M.S. and Ph.D. degrees in
electrical engineering from the Wireless In-
formation Network Laboratory (WINLAB),
Rutgers University, in 2000 and 2004, re-
spectively. Since 2000 he has been with the
Wireless Research Laboratory, Bell Labs, Lucent Technologies,
where he is involved in research in the eld of MIMO wireless sys-
tems. His research interests include detection, estimation, and in-
formation theory for MIMO wireless systems, interference cancel-
lation, and multiuser detection for multiple-access systems. He has
also been focusing on implementation aspects of various commu-
nication architectures and platforms.
Narayan Mandayam received the B.Tech. (with honors) degree in
1989 from the Indian Institute of Technology, Kharagpur, and the
M.S. and Ph.D. degrees in 1991 and 1994 from Rice University, all
in electrical engineering. Since 1994 he has been at Rutgers Uni-
versity where he is currently a Professor of electrical and computer
engineering and also an Associate Director at the Wireless Infor-
mation Network Laboratory (WINLAB). He was a Visiting Fac-
ulty Fellow in the Department of Electrical Engineering, Princeton
University, in Fall 2002 and a Visiting Faculty at the Indian Insti-
tute of Science in Spring 2003. His research interests are in various
aspects of wireless data transmission including system modeling
and performance, signal processing, and radio resource manage-
ment, with emphasis on open access techniques for spectrum shar-
ing. Dr. Mandayam is a recipient of the Institute Silver Medal from
the Indian Institute of Technology, Kharagpur, in 1989, and the
US National Science Foundation CAREER Award in 1998. He has
served as an Editor for the IEEE Journals Communication Letters
and Transactions on Wireless Communications. He is a coauthor
with C. Comaniciu and H. V. Poor of the book Wireless Networks:
Multiuser Detection in Cross-Layer Design, Springer, NY, 2005.
Dmitry Chizhik is a member of techni-
cal sta in the Wireless Research Labora-
tory, Bell Labs, Lucent Technologies. He re-
ceived a Ph.D. degree in electrophysics from
the Polytechnic University, Brooklyn, NY,
in 1991. His thesis work has been in ul-
trasonics and nondestructive evaluation. He
joined the Naval Undersea Warfare Center,
New London, Conn, where he did research
on scattering from ocean oor, geoacous-
tic modeling, and shallow water acoustic propagation. In 1996 he
joined Bell Laboratories, working on radio propagation modeling
and measurements, using deterministic and statistical techniques.
His recent work has been in measurement, modeling, and channel
estimation of MIMO channels. The results are used both for deter-
mination of channel-imposed bounds on channel capacity, system
performance, as well as for optimal antenna array design. His re-
search interests are in acoustic and electromagnetic wave propaga-
tion, signal processing, and communications.
EURASIP Journal on Wireless Communications and Networking 2005:3, 308322
c 2005 Francois Horlin et al.
Flexible Transmission Scheme for 4GWireless
Systems with Multiple Antennas
Francois Horlin
Wireless Research, Interuniversity Micro-Electronics Center (IMEC), Kapeldreef 75, B-3001 Leuven, Belgium
Email: francois.horlin@imec.be
Frederik Petr e
Wireless Research, Interuniversity Micro-Electronics Center (IMEC), Kapeldreef 75, B-3001 Leuven, Belgium
Email: petre@imec.be
Eduardo Lopez-Estraviz
Wireless Research, Interuniversity Micro-Electronics Center (IMEC), Kapeldreef 75, B-3001 Leuven, Belgium
Email: lopezest@imec.be
Frederik Naessens
Wireless Research, Interuniversity Micro-Electronics Center (IMEC), Kapeldreef 75, B-3001 Leuven, Belgium
Email: naessen@imec.be
Liesbet Van der Perre
Wireless Research, Interuniversity Micro-Electronics Center (IMEC), Kapeldreef 75, B-3001 Leuven, Belgium
Email: vdperre@imec.be
Received 15 October 2004; Revised 11 May 2005
New air interfaces are currently being developed to meet the high requirements of the emerging wireless communication systems.
In this context, the combinations of the multicarrier (MC) and spread-spectrum (SS) technologies are promising candidates. In
this paper, we propose a generic transmission scheme that allows to instantiate all the combinations of orthogonal frequency-
division multiplexing (OFDM) and cyclic-prexed single-carrier (SC) modulations with direct-sequence code-division multiple
access (DS-CDMA). The generic transmission scheme is extended to integrate the space-division multiplexing (SDM) and the
orthogonal space-time block coding (STBC). Based on a generalized matrix model, the linear frequency-domain minimum mean
square error (MMSE) joint detector is derived. A mode selection strategy for up- and downlink is advised that eciently trades
o the cost of the mobile terminal and the achieved performance of a high-mobility cellular system. It is demonstrated that
an adaptive transceiver that supports the proposed communication modes is necessary to track the changing communication
conditions.
Keywords and phrases: code-division multiple access, OFDM, cyclic-prex single carrier, space-division multiplexing, space-time
block coding, joint detection.
1. INTRODUCTION
Because of the limited frequency bandwidth, on the one
hand, and the potential limited power of terminal stations,
on the other hand, spectral and power eciency of fu-
ture systems should be as high as possible. New air inter-
faces need to be developed to meet the new system require-
This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
ments. Combinations of the multicarrier (MC) and spread-
spectrum (SS) modulations, named multicarrier spread-
spectrum techniques, could be interesting candidates. They
might benet from the main advantages of both MC and SS
schemes such as high spectral eciency, multiple-access ca-
pabilities, narrowband interference rejection, simple one-tap
equalization, and so forth.
Cellular systems of the third generation are based
on the recently emerged direct-sequence code-division
multiple-access (DS-CDMA) technique [1]. Intrinsically,
DS-CDMA has interesting networking abilities. First, the
Flexible Transmission Scheme for 4G Wireless Systems 309
communicating users do not need to be time synchronized
in the uplink. Second, soft handover is supported between
two cells making use of dierent codes at the base stations.
However, the system suers from intersymbol interference
(ISI) and multiuser interference (MUI) caused by multipath
propagation, leading to a high loss of performance.
The use of the orthogonal frequency-division multiplex-
ing (OFDM) modulation is widely envisaged for wireless
local area networks [2]. At the cost of the addition of a
cyclic prex, the time dispersive channel is seen in the fre-
quency domain as a set of parallel independent at sub-
channels and can be equalized at a low-complexity. An al-
ternative approach to OFDM, that benets from the same
low complexity equalization property, is single-carrier block
transmission (SCBT), also known as single-carrier (SC) with
frequency-domain equalization. Since the SCBT technique
benets from a lower peak-to-average power ratio (PAPR),
[3] encourages the use of SCBT in the uplink and OFDM in
the downlink in order to reduce the constraints on the analog
front end and the processing complexity at the terminal.
There are potential benets in combining OFDM (or
SCBT) and DS-CDMA. Basically the frequency-selective
channel is rst equalized in the frequency domain using the
OFDM modulation technique. DS-CDMA is applied on top
of the equalized channel, keeping the interesting orthogo-
nality properties of the codes. The DS-CDMA signals are
either spread across the OFDM carriers (referred to as in-
trablock spreading) leading to multicarrier CDMA (MC-
CDMA) [4, 5, 6, 7], or across the OFDM blocks (referred to
as interblock spreading) leading to multicarrier block-spread
CDMA (MCBS-CDMA) [8, 9, 10, 11]. The SCBT counter-
parts named here single-carrier CDMA (SC-CDMA) and
single-carrier block-spread CDMA (SCBS-CDMA) have also
been proposed in [12, 13, 14, 15], respectively. The dier-
ent avors to mix the MC and SS modulations complement
each other and allow to make an ecient tradeo between
the spectral and power eciency according to the user re-
quirements, channel propagation characteristics (time and
frequency selectivity), and terminal resources. For example,
it has been recently proposed, in [16] for the downlink, and
in [17] for the uplink, to perform a two-dimensional spread-
ing (combination of MC-CDMA with MCBS-CDMA) in or-
der to cope with the time and frequency selectivity of the
channels. The chips of a given symbol are mapped on adja-
cent channels where the fading coecients are almost con-
stant so that the orthogonality properties of the codes are
preserved and a low-complexity single-user detector can be
used.
To meet the data rate and quality-of-service require-
ments of future broadband cellular systems, their spectral
eciency and link reliability should be considerably im-
proved, which cannot be realized by using traditional single-
antenna communication techniques. To achieve these goals,
multiple-input multiple-output (MIMO) systems, which de-
ploy multiple antennas at both ends of the wireless link, ex-
ploit the extra spatial dimension, besides the time, frequency,
and code dimensions, which allows to signicantly increase
the spectral eciency and/or to signicantly improve the
link reliability relative to single-antenna systems [18, 19, 20].
MIMO systems explicitly rely on the fact that the channels
created by the additional spatial dimension are independent
from each other. This is approximately true in rich scattering
environments. However, when the channels are correlated,
the gain obtained by the use of multiple antennas is lim-
ited. Until very recently, the main focus of MIMO research
was on single-user communications over narrowband chan-
nels, thereby neglecting the multiple-access aspects and the
frequency-selective fading channel eects, respectively.
First, if the multiantenna propagation channels are suf-
ciently uncorrelated, MIMO systems can create N
min
par-
allel spatial pipes, which allows to realize an N
min
-fold ca-
pacity increase, where N
min
= min{N
T
, N
R
} (N
T
and N
R
denote the number of transmit and receive antennas, resp.)
is called the spatial multiplexing gain [18, 19, 20]. Speci-
cally, space-division multiplexing (SDM) techniques exploit
this spatial multiplexing gain, by simultaneously transmit-
ting N
min
independent information streams at the same fre-
quency [21, 22]. In [23], SDM is combined with SC-CDMA
to increase the data rate of multiple users in a broadband cel-
lular network.
Second, MIMO systems can also create N
T
N
R
indepen-
dently fading channels between the transmitter and the re-
ceiver, which allows to realize an N
T
N
R
-fold diversity in-
crease, where N
T
N
R
is called the multiantenna diversity gain.
Specically, space-time coding (STC) techniques exploit di-
versity and coding gains, by encoding the transmitted signals
not only over the temporal domain but also over the spa-
tial domain [24, 25, 26]. Space-time block coding (STBC)
techniques, introduced in [25] for N
T
= 2 transmit anten-
nas, and later generalized in [26] for any number of trans-
mit antennas, are particularly appealing because they facili-
tate maximum-likelihood (ML) detection with simple linear
processing. However, these STBC techniques have originally
been designed for frequency-at fading channels exploiting
only multiantenna diversity of order N
T
N
R
. Therefore, time-
reversal (TR) STBC techniques, originally proposed in [27]
for single-carrier serial transmission, have been combined
with SCBT in [28, 29] for signaling over frequency-selective
fading channels. In [30, 31], the TR-STBC technique of [28]
is combined with SC-CDMA to improve the performance
of multiple users in a broadband cellular network. Although
this technique enables low-complexity chip equalization in
the frequency domain, it does not preserve the orthogonality
among users, and hence, still suers from multiuser inter-
ference. The space-time coded multiuser transceiver of [32],
which combines the TR-STBC technique of [29] with SCBS-
CDMA, preserves the orthogonality among users as well
as transmit streams, regardless of the underlying multipath
channels. This allows for deterministic ML user separation
through low-complexity code-matched ltering as well as de-
terministic ML transmit stream separation through linear
processing. Another alternative to remove MUI determinis-
tically in a space-time coded multiuser setup [33, 34] com-
bines generalized multicarrier (GMC) CDMA, originally de-
veloped in [35], with the STBC techniques of [26] but imple-
mented on a per-carrier basis.
310 EURASIP Journal on Wireless Communications and Networking
Intrablock
spreading
Interblock
spreading
IFFT redundancy
s
m
[ j]
m
s
m
[ j]
N
inter c
m
[n]
x
m
[n]
F
H
Q
x
m
[n]
T
u
m
[n]
Figure 1: Single-antenna transmitter model.
We propose a transmission chain composed of generic
blocks and able to instantiate all the communication modes
combining OFDM/SCBT and DS-CDMA as special cases. In
contrast with the transceiver proposed in [35, 36, 37] that
relies on a sharing of the set of carriers to retain the orthog-
onality between the users, our transmission scheme relies on
orthogonal CDMA, and thus inherits the nice advantages of
CDMA related to universal frequency reuse in a cellular net-
work, like increased capacity and simplied network plan-
ning. Furthermore, the focus is especially put on the com-
munication modes emerging in the standards. We demon-
strate the rewarding synergy between existing and evolving
MIMO communication techniques and our generic trans-
mission technique, which allows to increase the spectral ef-
ciency and to improve the link reliability of multiple users
in a broadband cellular network. Considering realistic prop-
agation channels, we also advise a strategy for mode selec-
tion according to the communication conditions, making an
ecient tradeo between the desired performance and the
required computational complexity.
The paper is organized as follows. In Section 2, the
generic transmission scheme is presented that can capture
the standard emerging communication modes as special
cases. It is shown how the MC and SC modes can be instan-
tiated. Section 3 is devoted to the extension of the transmis-
sion scheme introduced in Section 2 to multiple-antenna sys-
tems. Both SDM and STBC multiple-antenna techniques are
considered. A low-complexity minimum mean square error
(MMSE) receiver is designed in Section 4 based on a gener-
alized matrix model. Finally, a strategy for communication
mode selection is proposed in Section 5, based on the evalu-
ation of the cost and performance of each mode in a highly
mobile cellular environment.
In the sequel, we use single- and double-underlined let-
ters for the vectors and matrices, respectively. Matrix I
N
is the
identity matrix of size N and matrix 0
MN
is a matrix of zeros
of size MN. The operators ()

, ()
T
, and ()
H
denote, re-
spectively, the complex conjugate, transpose, and transpose
conjugate of a vector or a matrix. The operator is the Kro-
necker product between two vectors or matrices. We index
the transmitted block sequence by i, the MIMO coded block
sequence and the intrablock chip block sequence by j, and
the interblock chip block sequence by n.
2. SINGLE-ANTENNA GENERIC
TRANSMISSIONSCHEME
The transmission scheme for the mth user (m = 1, . . . , M)
is depicted in Figure 1. Since we focus on a single-user
transmission, the transmission scheme applies to both the
up- and downlink. In the uplink, the dierent users signals
are multiplexed at the receiver, after propagation through
their respective multipath channels. In the downlink, the dif-
ferent user signals are multiplexed at the transmitter, before
the inverse fast Fourier transform (IFFT) operation.
The transmission scheme comprises four basic opera-
tions: intrablock spreading, interblock spreading, IFFT, and
adding transmit redundancy. The symbols s
m
[ j] are rst
serial-to-parallel converted into blocks of B symbols, lead-
ing to the sequence s
m
[ j] := [s
m
[ jB] s
m
[( j + 1)B 1]]
T
.
The blocks s
m
[ j] are linearly precoded with a QB (Q B)
matrix,
m
, which possibly introduces some redundancy and
spreads the symbols in s
m
[ j] with length-Q codes as follows:
s
m
[ j] :=
m
s
m
[ j]. (1)
We refer to this rst operation as intrablock spreading, since
the information symbols s
m
[ j] are spread within a single pre-
coded block s
m
[ j]. The precoded block sequence s
m
[ j] is
then block spread with the elements c
m
[n] of a length-N
inter
code sequence, leading to N
inter
successive chip blocks:
x
m
[n] := s
m
[ j]c
m
_
n jN
inter
_
, (2)
where j = n/N
inter
. We refer to this second operation as
interblock spreading, since the information symbols s
m
[ j] are
spread along N
inter
dierent chip blocks. The third operation
involves the transformation of the frequency-domain chip
block sequence x
m
[n] into the time-domain chip block se-
quence:
x
m
[n] := F
H
Q
x
m
[n], (3)
where F
H
Q
is the QQIFFTmatrix. Finally, the KQ(K Q)
transmit matrix T possibly adds some transmit redundancy
to the time-domain chip blocks:
u
m
[n] := T x
m
[n]. (4)
With K = Q + L denoting the total block length T = T
cp
:=
[I
T
cp
, I
T
Q
]
T
, where I
cp
consists of the last L rows of I
Q
, T
adds redundancy in the form of a length-L cyclic prex (cp).
The chip block sequence u
m
[n] is parallel-to-serial converted
into the scalar sequence [u
m
[nK] u
m
[(n + 1)K 1]]
T
:=
u
m
[n], and transmitted over the air at a rate 1/T
c
(T
c
stands
for the chip duration).
In the following, we will detail how our generic transmis-
sion scheme instantiates dierent communication modes,
and, thus, supports dierent emerging communication stan-
dards. We distinguish between the MC modes, on the one
hand, and the SC modes, on the other hand. In Figure 2, the
principle of CDMA spreading in two possible dimensions is
illustrated.
Flexible Transmission Scheme for 4G Wireless Systems 311
Code
Spreading
factor
Frame block
Subchannel
(MC carrier frequency or SC time instant)

(a)
Code
Spreading
factor
Frame block
Subchannel
(MC carrier frequency or SC time instant)
(b)
Figure 2: (a) Intrablock spreading pattern (MC/SC-CDMA). (b) Interblock spreading pattern (MCBS/SCBS-CDMA).
2.1. Instantiation of the multicarrier modes
The MCmodes always comprise the IFFT operation, and add
transmit redundancy in the form of a cyclic prex (T = T
cp
).
The MC modes comprise MC-CDMA and MCBS-CDMA as
particular instantiations of the generic transmission scheme.
2.1.1. MC-CDMA
As we have indicated in the introduction, MC-CDMA rst
performs classical DS-CDMA symbol spreading, followed by
OFDM modulation, such that the information symbols are
spread across the dierent subcarriers located at dierent fre-
quencies and characterized by a dierent fading coecient
[4, 5, 6]. With Q = BN
intra
and N
intra
the intrablock spreading
code length, the Q B intrablock spreading matrix
m
=
m
spreads the chips across the subcarriers, where the mth users
Q B spreading matrix
m
is dened as

m
:= I
B
a
m
, (5)
with a
m
:= [a
m
[0] a
m
[N
intra
1]]
T
the mth users N
intra

1 code vector. The interblock spreading operation is dis-


carded by setting N
inter
= 1. Since intrablock spread-
ing does not preserve the orthogonality among users in a
frequency-selective channel, MC-CDMA requires advanced
multiuser detection for uplink reception in the base station,
and frequency-domain chip equalization for downlink re-
ception in the mobile station. MC-CDMAhas been proposed
as a candidate air interface for future broadband cellular sys-
tems [7].
2.1.2. MCBS-CDMA
The MCBS-CDMA transmission scheme is the only MC
mode that comprises the interblock spreading operation
N
inter
> 1. Since the CDMA spreading is applied on each
carrier independently, which can be seen as a constant fad-
ing channel if the propagation environment is static, MCBS-
CDMA retains the orthogonality among users in both up-
and downlink [11]. Hence, it has the potential to convert a
dicult multiuser detection problem into an equivalent set
of simpler and independent single-user equalization prob-
lems. However, it will be shown in Section 5 that the perfor-
mance of MCBS-CDMA is highly degraded under medium-
to-high mobility conditions since the orthogonality between
the users is lost in that case. In case no channel state in-
formation (CSI) is available at the transmitter, it performs
linear precoding to robustify the transmitted signal against
frequency-selective fading. In case CSI is available at the
transmitter, it allows to optimize the transmit spectrum of
each user separately through adaptive power and bit load-
ing. Note that classical MC-DS-CDMA can be seen as a spe-
cial case of MCBS-CDMA, because it does not include linear
precoding, but, instead, only relies on bandwidth-consuming
forward error correction (FEC) coding to enable frequency
diversity [8, 10, 38].
2.2. Instantiation of the single-carrier modes
The SC modes employ a fast Fourier transform (FFT) as part
of the intrablock spreading operation to annihilate the IFFT
operation. For implementation purposes, however, the IFFT
is simply removed (and not compensated by an FFT), in
order to minimize the implementation complexity. The SC
modes rely on cyclic prexing (T = T
cp
) to make the chan-
nel appear circulant. The SC modes comprise SC-CDMA
and SCBS-CDMA as particular instantiations of the generic
transmission scheme.
2.2.1. SC-CDMA
The SC-CDMA transmission scheme, which combines SCBT
with DS-CDMA, can be interpreted as the SC counterpart
of MC-CDMA [12, 13]. This mode is captured through our
general transmission scheme, by setting Q = BN
intra
. The
intrablock spreading matrix
m
= F
Q

m
, with
m
de-
ned in (5), performs symbol spreading on the B informa-
tion symbols, followed by an FFT operation to compensate
312 EURASIP Journal on Wireless Communications and Networking
d
m
1
[i]
d
m
N
T
[i]
MIMO
coding
.
.
.
.
.
.

m
N
inter
N
inter
c
m
[n]
c
m
[n]
F
H
Q
F
H
Q
T
T
u
m
1
[n]
u
m
N
T
[n]
s
m
1
[ j] s
m
1
[ j] x
m
1
[n]
x
m
1
[n]
s
m
N
T
[ j] s
m
N
T
[ j] x
m
N
T
[n]
x
m
N
T
[n]
Figure 3: MIMO transmitter model.
for the subsequent IFFT operation. Like MC-CDMA, SC-
CDMA does not preserve the orthogonality among the users
in a frequency-selective channel. It requires advanced mul-
tiuser detection for uplink reception at the base station and
chip equalization for downlink reception at the mobile ter-
minal. On the contrary to MC-CDMA, each user symbol is
spread amongst multiple subchannels of the same power at-
tenuation. The interblock spreading operation is left out by
setting N
inter
= 1.
2.2.2. SCBS-CDMA
The SCBS-CDMA transceiver can be considered as the SC
counterpart of MCBS-CDMA. It is the only SC mode that
entails the interblock spreading operation (N
inter
> 1). The
intrablock spreading matrix
m
= F
Q
only performs an FFT
operation to compensate for the subsequent IFFT operation.
If the propagation channels are static, SCBS-CDMA retains
the orthogonality among users in both the up- and downlink,
even after propagation through a frequency-selective chan-
nel (like MCBS-CDMA). It also converts a dicult multiuser
detection problem into an equivalent set of simpler and in-
dependent single-user equalization problems. The orthogo-
nality property is however lost in time-selective channels, as
studied numerically in Section 5.
3. EXTENSIONTOMULTIPLE ANTENNAS
The generic transmission model is extended in Figure 3 to
include two types of MIMO techniques. We assume N
T
an-
tennas at the transmit side and N
R
antennas at the receive
side. The information symbols d
m
nT
[i], which are assumed
independent and of variance equal to
2
d
, are rst serial-to-
parallel converted into blocks of B symbols, leading to the
block sequence d
m
nT
[i] := [d
m
nT
[iB] d
m
nT
[(i + 1)B 1]]
T
for
n
T
= 1, . . . , N
T
. A MIMO coding operation is performed
across the dierent transmit antenna streams, that results
into the N
T
antenna sequences s
m
nT
[ j] input to the generic
transmission scheme. The overall rate increase obtained by
the use of multiple antennas is either 1 or N
T
depending on
the MIMO technique that is selected.
3.1. Space-division multiplexing
In this section, we combine our generic transmission scheme
with SDM, which allows to instantiate all combinations of
SDM with OFDM/SCBT and CDMA as special cases. The
SDM technique is implemented by sending independent
streams on each transmit antenna n
T
, as expressed in
s
m
nT
[ j] = d
m
nT
[i], (6)
where j = i.
3.2. Space-time block coding
In this section, we combine our generic transmission scheme
with STBC, which allows to instantiate all combinations of
STBC with OFDM/SCBT and CDMA as special cases. For
conciseness, we limit ourselves to the case of N
T
= 2 transmit
antennas (Alamouti scheme [25, 27]). The STBC coding is
implemented by coding the two antenna streams across two
time instants, as expressed in
_
s
m
1
[ j]
s
m
2
[ j]
_
=
_
d
m
1
[i]
d
m
2
[i]
_
,
_
s
m
1
[ j + 1]
s
m
2
[ j + 1]
_
=
_
d
m
1
[i]

d
m
2
[i]

_
,
(7)
where i = j/2 and
:=
NT

B
with
NT
:=
_
0 1
1 0
_
. (8)
In the case of the MC modes, the STBC coding is applied in
the frequency domain on a per-carrier basis so that

B
:= I
B
. (9)
In the case of the SC modes, the STBC coding is applied in
the time domain by further permuting the vector elements
so that

B
:= F
T
B
F
B
(10)
is a B B permutation matrix implementing a time reversal.
It is easily checked that the transmitted block at time in-
stant j + 1 from one antenna is the time-reversed conjugate
of the transmitted symbol at time instant j from the other
antenna (with possible permutation and sign change). As we
will show later, this property allows the deterministic trans-
mit stream separation at the receiver, regardless of the under-
lying frequency-selective channels.
4. RECEIVER DESIGN
4.1. Cyclostationarization of the channels
Adopting a discrete-time baseband equivalent model, the
chip-sampled received signal at antenna n
R
(n
R
= 1, . . . , N
R
),
v
nR
[n], is the superposition of a channel-distorted version of
the MN
T
transmitted user signals, which can be written as
v
nR
[n] =
M
_
m=1
NT
_
nT=1
L
m
_
l=0
h
m
nR,nT
[l]u
m
nT
[n l] + w
nR
[n], (11)
Flexible Transmission Scheme for 4G Wireless Systems 313
where h
m
nR,nT
[l] is the chip-sampled nite impulse response
(FIR) channel of order L
m
that models the frequency-
selective multipath propagation between the mth users an-
tenna n
T
and the base station antenna n
R
, including the ef-
fect of transmit/receive lters and the remaining asynchro-
nism of the quasisynchronous users, and w
nR
[n] is additive
white Gaussian noise (AWGN) at the base station antenna n
R
with variance
2
w
. Furthermore, the maximum channel or-
der L, that is, L = max
m
{L
m
}, can be well approximated by
L (
max,a
+
max,s
)/T
c
+ 1, where
max,a
is the maximum
asynchronism between the nearest and the farthest user of
the cell, and
max,s
is the maximum excess delay within the
given propagation environment [35].
Assuming perfect time and frequency synchronization,
the sequence v
nR
[n] is serial-to-parallel converted into the
sequence v
nR
[n] := [v
nR
[nK] v
nR
[(n + 1)K 1]]
T
. From
the scalar input/output relationship in (11), we can derive
the corresponding block input/output relationship:
v
nR
[n] =
M
_
m=1
NT
_
nT=1
_
H
m
nR,nT
[0] u
m
nT
[n]
+ H
m
nR,nT
[1] u
m
nT
[n 1]
_
+ w
nR
[n],
(12)
where w
nR
[n] := [w
nR
[nK] w
nR
[(n + 1)K 1]]
T
is the
corresponding noise block sequence, H
m
nR,nT
[0] is a K K
lower triangular Toeplitz matrix with entries [H
m
nR,nT
[0]]
p,q
=
h
m
nR,nT
[p q] for p q [0, L
m
] and [H
m
nR,nT
[0]]
p,q
= 0
else, and H
m
nR,nT
[1] is a K K upper triangular Toeplitz
matrix with entries [H
m
nR,nT
[1]]
p,q
= h
m
nR,nT
[K + p q] for
K + p q [0, L
m
] and [H
m
nR,nT
[1]]
p,q
= 0 else (see, e.g.,
[35] for a detailed derivation of the single-user case). The
delay-dispersive nature of multipath propagation gives rise
to so-called interblock interference (IBI) between successive
blocks, which is modeled by the second term in (12).
The Q K receive matrix R removes the redundancy
from the chip blocks, that is, y
nR
[n] := R v
nR
[n]. With
R = R
cp
= [0
QL
, I
Q
], R again discards the length-L cyclic
prex. The purpose of the transmit/receive pair is twofold.
First, it allows for simple block-by-block processing by re-
moving the IBI, that is, RH
m
nR,nT
[1]T = 0
QQ
, provided the
CP length to be at least the maximum channel order L. Sec-
ond, it enables low-complexity frequency-domain processing
by making the linear channel convolution to appear circu-
lant to the received block. This results in a simplied block
input/output relationship in the time domain:
y
nR
[n] =
M
_
m=1
NT
_
nT=1

H
m
nR,nT
x
m
nT
[n] + z
nR
[n], (13)
where

H
m
nR,nT
= R H
m
nR,nT
[0] T is a circulant channel ma-
trix, and z
nR
[n] = R w
nR
[n] is the corresponding noise
block sequence. Note that circulant matrices can be diago-
nalized by FFT operations, that is,

H
m
nR,nT
= F
H
Q

m
nR,nT
F
Q
,
where
m
nR,nT
is a diagonal matrix composed of the frequency-
domain channel response between the mth users antenna n
T
and the base station antenna n
R
.
4.2. Matrix model
Based on (13), a generalized matrix model is developed that
relates the vector of transmitted users symbols to the vector
of received samples. It encompasses all the combinations of
OFDM/SCBT with CDMAconsidered in this paper as special
cases. Based on this model, a multiuser joint detector opti-
mized according to the MMSE criterion will be derived and
its complexity will be reduced by exploiting the cyclostation-
arity properties of the matrices.
The generalized input/output matrix model that relates
the MIMO-coded symbol vector dened as
s[ j] :=
_
s
1
[ j]
T
s
M
[ j]
T
_
T
(14)
with
s
m
[ j] :=
_
s
m
1
[ j]
T
s
m
NT
[ j]
T
_
T
(15)
for m = 1, . . . , M to the received and noise vectors dened as
y[ j] :=
_
y
1
[ j]
T
y
NR
[ j]
T
_
T
,
z[ j] :=
_
z
1
[ j]
T
z
NR
[ j]
T
_
T
(16)
with
y
nR
[ j] :=
_
_
y
nR
_
jN
inter
__
T

_
y
nR
_
( j + 1)N
inter
1
__
T
_
T
,
z
nR
[ j] :=
_
_
z
nR
_
jN
inter
__
T

_
z
nR
_
( j + 1)N
inter
1
__
T
_
T
,
(17)
for n
R
= 1, . . . , N
R
, is given by
y[ j] = C F
H
s[ j] + z[ j], (18)
where the channel matrix is dened as
:=
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_

1
1
0
QQ
.
.
.
.
.
.
.
.
.
0
QQ

M
1
.
.
.
.
.
.
.
.
.

1
NR
0
QQ
.
.
.
.
.
.
.
.
.
0
QQ

M
NR
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
(19)
with

m
nR
:=
_

m
nR,1

m
nR,NT
_
(20)
314 EURASIP Journal on Wireless Communications and Networking
for m = 1, . . . , M and n
R
= 1, . . . , N
R
, and the intrablock
spreading, Fourier, and interblock spreading matrices are de-
ned, respectively, as
:=
_
_
_
_
_
I
NT

1
0
NTQNTB
.
.
.
.
.
.
.
.
.
0
NTQNTB
I
NT

M
_
_
_
_
_
,
F := I
NRM
F
Q
,
C := I
NR

_
C
1
C
M
_
,
(21)
in which
C
m
:= c
m
I
Q
, (22)
with c
m
:= [c
m
[0] c
m
[N
inter
1]]
T
. Note that the model
(18) only holds for static channels. In case of time-selective
channels, the channel matrices

H
m
nR,nT
cannot be diagonalized
anymore and are dierent from one chip block to the next
one.
4.2.1. Space-division multiplexing
Taking (6) into account, the model (18) is instantiated to the
SDM input/output matrix model
y
sdm
[i] = C
sdm

_
F
sdm
_
H

sdm

sdm

sdm


d[i]
+ z
sdm
[i],
(23)
where the vector of transmitted symbols is dened as

d[i] :=
_

d
1
[i]
T


d
M
[i]
T
_
T
, (24)
with

d
m
[i] :=
_
d
m
1
[i]
T
d
m
NT
[i]
T
_
T
(25)
for m = 1, . . . , M, and the received and noise vectors are de-
ned as
y
sdm
[i] := y[ j],
z
sdm
[i] := z[ j].
(26)
The MIMO encoding, intrablock spreading, channel,
Fourier, and interblock spreading matrices are, respectively,
given by

sdm
:= I
MNTB
,

sdm
:= ,

sdm
:= ,
F
sdm
:= F,
C
sdm
:= C.
(27)
4.2.2. Space-time block coding
Taking (7) into account, the model (18) is instantiated to the
STBC input/output matrix model
y
stbc
[i] = C
stbc

_
F
stbc
_
H

stbc

stbc

stbc


d[i]
+ z
stbc
[i],
(28)
where the vector of transmitted symbols is given in (25)
assuming that N
T
= 2, the received and noise vectors are
dened as
y
stbc
[i] :=
_
y[ j]
y[ j + 1]

_
,
z
stbc
[i] :=
_
z[ j]
z[ j + 1]

_
.
(29)
The MIMO encoding, intrablock spreading, channel,
Fourier, and interblock spreading matrices are, respectively,
given by

stbc
:=
_
_
I
MNTB
I
M

_
_
,

stbc
:=
_
_
0
MNTQMNTB
0
MNTQMNTB

_
_
,

stbc
:=
_
_
0
NRMQMNTQ
0
NRMQMNTQ

_
_
,
F
stbc
:=
_
_
F 0
NRMQNRMQ
0
NRMQNRMQ
F

_
_
,
C
stbc
:=
_
_
C 0
NRNinterQNRMQ
0
NRNinterQNRMQ
C

_
_
.
(30)
4.3. Multiuser joint detector
In order to detect the transmitted symbol block of the pth
user

d
p
[i] based on the received sequence of blocks within
the received vector y
mode
[i] (mode stands for sdm or
stbc), a rst solution consists of using a single-user receiver
that inverts successively the channel and all the operations
performed at the transmitter. The single-user receiver relies
implicitly on the fact that CDMA spreading has been applied
on top of a channel equalized in the frequency domain. After
CDMA despreading, each user stream is handled indepen-
dently. However the single-user receiver fails in the uplink
where multiple channels have to be inverted at the same time.
The optimal solution is to jointly detect the transmitted
symbol blocks of the dierent users within the transmitted
vector

d[i] based on the received sequence of blocks within
the received vector y
mode
[i]. The optimum linear joint detec-
tor according to the MMSE criterion is computed in [39]. At
the output of the MMSE multiuser detector, the estimate of
Flexible Transmission Scheme for 4G Wireless Systems 315
the transmitted vector is

d[i] =
_

2
w

2
d
I
MNTB
+ G
H
mode
G
mode
_
1
G
H
mode
y
mode
[i],
(31)
where
G
mode
:= C
mode

_
F
mode
_
H

mode

mode

mode
. (32)
The MMSE linear joint detector consists of two main opera-
tions [39, 40].
(i) First, a lter matched to the composite impulse re-
sponses multiplies the received vector in order to minimize
the impact of the white noise. The matched lter consists
of the CDMA interblock despreading, the FFT operator to
move to the frequency domain, the maximum ratio combin-
ing (MRC) of the dierent received antenna channels, the
CDMA intrablock despreading, the IFFT to go back to the
time domain in case of the SC modes, and the STBC decod-
ing.
(ii) Second, the output of the matched lter is still multi-
plied with the inverse of the composite impulse response au-
tocorrelation matrix of size MN
T
B that mitigates the remain-
ing intersymbol, interuser, and interantenna interference.
In the case of interblock spreading (MCBS and SCBS-
CDMA), the spreading matrix C has the property that
C
H
C = I
NRQM
if the CDMA codes are chosen orthogo-
nal. When the propagation channels are static, the dierent
user streams are perfectly separated at the output of the in-
terblock despreading operation and further treated indepen-
dently. The MMSE multiuser joint detector exactly reduces
to independent single-user detectors. In case of time-varying
propagation channels, model (18) is not valid anymore and
the multiuser MMSE joint detector cannot be simplied to
single-user detectors.
In the case of intrablock spreading (MC- and SC-
CDMA), however, the linear MMSE receiver is dierent from
the single-user receiver, and suers from a higher computa-
tional complexity. Fortunately, both the initialization com-
plexity, which is required to compute the MMSE receiver, and
the data processing complexity can be signicantly reduced
for MC- and SC-CDMA, by exploiting the initial cyclosta-
tionarity property of the channels. Based on a few permuta-
tions and on the properties of the block circulant matrices
given in [12], it can be shown that the initial inversion of the
square autocorrelation matrix of size MN
T
B can be replaced
by the inversion of B square autocorrelation matrices of size
MN
T
.
5. MODE SELECTIONSTRATEGY
In this section, a cost and performance comparison between
the dierent communication modes is made, which can serve
as an input for an ecient mode selection strategy.
5.1. Complexity
To evaluate the complexity of the dierent receivers, we dis-
tinguish between the initialization phase, when the receiver is
calculated, and the data processing phase, when the received
data is actually processed. The rate of the former is related
to the channels fading rate, whereas the latter is executed
continuously at the symbol block rate. The complexity will
be described in terms of complex multiply/accumulate cycle
(MAC) operations. It is assumed that 2N
3
complex MAC op-
erations are required to invert a matrix of size N, and that
N log
2
(N) complex MAC operations are required to com-
pute an FFT of size N.
5.1.1. Initialization complexity
The complexity required to compute the MMSE receiver is
given in Table 1 for the base station and for the terminal,
respectively. It has been assumed that the inversion of the
MMSE equalizer block diagonal inner matrix is dominant.
Since the size of each block is equal to the number of inter-
fering users times the number of interfering antennas, the
computation of the equalizer is more complex in case of
intrablock spreading (number of interfering users equal to
the number of users) than in case of interblock spreading
(number of interfering users equal to one), and in case of
SDM (number of interfering antennas equal to the number
of transmit antennas) than in case of STBC (number of in-
terfering antennas equal to one).
5.1.2. Data processing complexity
The complexity needed during the data processing phase to
transmit and receive each users transmitted complex symbol
is further given in the Table 2. The computational eort is al-
most equally shared between the transmitter and the receiver
in the case of the MC-based modes. In the case of the SC-
based modes, all the computational eort has been moved
to the receiver. From a terminal complexity point of view, it
is clearly advantageous to use the SC-based modes in uplink
and the MC-based modes in downlink. Both the transmis-
sion and the reception are less complex in case of interblock
spreading than in case of intrablock spreading, because the
(I) FFT operators can be executed at a lower rate in the for-
mer case (namely, before the spreading at the transmitter and
after the despreading at the receiver). Since STBC combines
N
T
successive symbol blocks, it is on the overall more com-
plex than the SDM scheme.
5.2. Peak-to-average power ratio
The PAPR is illustrated in Figure 4 for each mode as a func-
tion of the spreading factor. The user signals are spread by pe-
riodic Walsh-Hadamard codes for spreading, which are over-
laid with an aperiodic Gold code for scrambling. QPSKmod-
ulation is assumed with Q = 128 subchannels, and a cyclic
prex length of L = 32. The PAPR is dened as the maxi-
mum instantaneous peak power over all combinations of the
transmitted symbols divided by the average transmit power.
While the MC-based communication modes feature a
very large PAPR, the SC-based communication modes have
316 EURASIP Journal on Wireless Communications and Networking
Table 1: Initialization complexity, M users, N
T
transmit antennas, Q carriers/subchannels, and B complex symbols per block.
Base station Mobile terminal
SDM STBC SDM STBC
MC/SCBS-CDMA 2QM(N
T
)
3
2MQ 2Q(N
T
)
3
2Q
MC/SC-CDMA 2B(MN
T
)
3
2B(M)
3
2B(MN
T
)
3
2B(M)
3
Table 2: Data processing complexity, M users (M = 1 at the terminal), N
T
transmit antennas, N
R
receive antennas, Q carriers/subchannels,
B complex symbols per block.
Transmitter Receiver
SDM STBC SDM STBC
MCBS-CDMA log
2
Q N
T
log
2
Q
N
R
N
T
log
2
Q + N
R
N
R
log
2
Q + N
T
N
R
SCBS-CDMA 0 0
_
N
R
N
T
+ 1
_
log
2
Q + N
R
_
N
R
+ 1
_
log
2
Q + N
T
N
R
MC-CDMA
Q
MB
log
2
Q N
T
Q
MB
log
2
Q
N
R
N
T
Q
MB
log
2
Q + N
R
Q
B
N
R
Q
MB
log
2
Q + N
T
N
R
Q
B
SC-CDMA 0 0
_
N
R
N
T
Q
MB
+ 1
_
log
2
B + N
R
Q
B
_
N
R
Q
MB
+ 1
_
log
2
B + N
T
N
R
Q
B
7 6 5 4 3 2 1 0
log
2
(spreading factor)
25
20
15
10
5
0
P
A
P
R
(
d
B
)
SC-CDMA
SC-BS-CDMA
MC-CDMA
MC-BS-CDMA
Figure 4: PAPR as a function of the spreading factor, QPSK, 128
carriers.
a PAPR close to 0 dB. MC-CDMA has a PAPR improving for
an increasing spreading factor due to the fact that the total
number of possible chip combinations at the output of the
spreading operation is decreasing. The use of the SC-based
communication modes is encouraged in the uplink to re-
duce the power amplier backo at the mobile terminal and,
thus, to increase its power eciency (even if the power am-
plier backo is not directly settled based on the low prob-
ability power peak values in the MC-based systems, it is ex-
pected that the variation in instantaneous transmitted power
is much lower in the SC-based systems).
5.3. Goodput performance
We consider a mobile cellular system which operates in an
outdoor suburban macrocell propagation environment. The
channel model is largely inspired from the 3 GPP TR25.996
geometrical spatial channel model [41]. The radius of the
cell is equal to 3 km. The mobile terminals are moving at a
speed ranging from 0 (static) to 250 km/h (highly mobile).
The system operates at a carrier frequency of 2 GHz, with a
system bandwidth of 5 MHz. Linear antenna arrays are as-
sumed at the mobile terminals and at the base station. An
antenna spacing equal to 0.5 wavelength is selected at the mo-
bile terminals, which results in a correlation equal to 0.2 be-
tween the antennas, and an antenna spacing equal to 5 wave-
lengths is selected at the base station, which results in a cor-
relation smaller than 0.1 between the antennas. We assume
a six-sector antenna at the base station and omnidirectional
antennas at the mobile terminals. The propagation environ-
ment is characterized by a specular multipath composed of
6 paths (each path represents the reection on a scatterer)
and 10 subpaths per path (each subpath represents the re-
ection on a specic part of the scatterer). For each path, the
excess delay standard deviation is equal to 238 nanoseconds,
the angle spread standard deviation at the base station and
that at the mobile terminals are equal to 6 and 68 degrees, re-
spectively. For each subpath, the angle spread standard vari-
ation at the base station and that at the mobile terminals are
equal to 2 and 5 degrees, respectively. The computation of the
path loss is based on the COST231 Hata urban propagation
model.
Monte Carlo simulations have been performed to aver-
age the bit error rate (BER) over 500 stochastic channel re-
alizations, and to compute the corresponding goodput, de-
ned as the throughput oered to the user assuming a re-
transmission of the erroneous packets until they are correctly
Flexible Transmission Scheme for 4G Wireless Systems 317
received. The information bandwidth is spread by the
spreading factor equal to 8. The user signals are spread by pe-
riodic Walsh-Hadamard codes for spreading, which are over-
laid with an aperiodic Gold code for scrambling. QPSK, 16-
QAM, or 64-QAM modulation is used with Q = 128 sub-
channels, and a CP length of L = 32. We assume a packet size
of 512 complex symbols (4 blocks of 128 symbols in case of
MCBS/SCBS-CDMA or 32 blocks of 16 symbols in case of
MC/SC-CDMA). Convolutional channel coding in conjunc-
tion with frequency-domain interleaving is employed ac-
cording to the IEEE 802.11a/g standard. The code rate varies
from 1/2, 2/3, to 3/4. At the receiver, soft-decision Viterbi
decoding is used.
We distinguish between the uplink and the downlink. In
the uplink, transmit power control is applied such that the
received symbol energy is constant for all users. The power
transmitted by each terminal depends on the actual channel
experienced by it. The BER (or the goodput) is determined
as a function of the received bit energy, or, equivalently, as
a function of the transmit power averaged over the dier-
ent channel realizations. In the downlink, no transmit power
control is applied. For a constant transmit power at the base
station, the received symbol energy at each terminal depends
on the channel under consideration. The BER (or the good-
put) is determined as a function of the transmit power, or,
equivalently, as a function of the received bit energy averaged
over the channel realizations. For a given received symbol en-
ergy, the required transmit powers for the dierent commu-
nication modes appear to be very similar.
Rather than comparing all possible modes for the two
link directions, we only consider the relevant modes for each
direction. For the uplink, the SC modes demonstrate two
pronounced advantages compared to MC modes [3]. First,
the SC modes exhibit a smaller PAPR than the MC modes,
which leads to increased terminal power eciency. Second,
the SC modes allow to move the IFFT at the transmitter to
the receiver, which results in reduced terminal complexity.
For the downlink, the MC modes are the preferred modula-
tion schemes, since they only incur a single FFT operation at
the receiver side which also leads to reduced terminal com-
plexity.
Figure 5 illustrates the users goodput in the downlink
of a MCBS-CDMA-based static communication system for
dierent combinations of the constellation size and channel
coding rate. The signal is received at the terminal through 2
antennas. A typical users load of 5 users has been assumed.
Looking to those combinations that give the same asymptotic
goodput (16-QAM and CR 3/4 compared to 64-QAM and
CR 1/2), it is always preferable to combine a high constel-
lation size with a low code rate. The same conclusion holds
in order to select the combination constellation-coding rate
that maximizes the goodput for each SNR value. The enve-
lope is obtained by progressively employing QPSK, 16-QAM,
64-QAMconstellations and a code rate equal to 1/2, and then
by increasing the code rate progressively to the values 2/3,
3/4 while keeping the constellation size xed to 64-QAM.
The gain in communication robustness provided by the use
of more channel coding exceeds the loss in communication
30 25 20 15 10 5 0
E
b
/N
0
(dB)
2
1.5
1
0.5
0
G
o
o
d
p
u
t
(
M
b
p
s
)
QPSK, 1/2
QPSK, 2/3
QPSK, 3/4
16-QAM, 1/2
16-QAM, 2/3
16-QAM, 3/4
64-QAM, 1/2
64-QAM, 2/3
64-QAM, 3/4
Figure 5: Tradeo between channel coding rate and constellation,
static downlink, MCBS-CDMA, using MRC.
robustness due to the use of a higher constellation. As it will
be shown later, the conclusion may be reversed in a mobile
environment due to better robustness of lower-order map-
pings to changing channel conditions.
Figures 6 and 7 illustrate the gain obtained by the use of
dierent multiple-antenna techniques in the downlink of an
MCBS-CDMA-based static system and in the uplink of an
SCBS-CDMA-based static system, respectively. Four dier-
ent systemcongurations are considered (11, 12, 21, 2
2). A typical users load is assumed (number of users equal to
5). Two optimal combinations of the constellation and chan-
nel coding rate are considered (16-QAM with coding rate
1/2, 64-QAM with coding rate 3/4). Exploiting the spatial
diversity at the mobile terminal or at the base station by the
use of two antennas (1 2 or 2 1 congurations), MRC
reception, or STBC transmission enables a signicant gain in
transmit power (up to 12 dB gain is achieved in the down-
link, up to 7 dB gain is achieved in the uplink). However, the
additional gain obtained by combining STBC transmission
with MRC reception (2 2 conguration) is relatively small
(less than 2 dB). SDM suers from a high 8 dB loss in the
goodput regions that can also be reached by a single-antenna
system (this loss can be reduced if more complex nonlinear
receivers are considered). One can only achieve an increase of
capacity by the use of multiple antennas at very high signal-
to-noise ratio (SNR) values. As a result, we advise to use one
directive antenna at the base station to exploit the very small
angle spread and increase the cell capacity, and two diversity
antennas at the mobile terminal to improve the link reliabil-
ity. The receive MRC technique is performed in the down-
link, while the STBC coding scheme is applied in the up-
link.
318 EURASIP Journal on Wireless Communications and Networking
35 30 25 20 15 10 5 0
E
b
/N
0
(dB)
3
2.5
2
1.5
1
0.5
0
G
o
o
d
p
u
t
(
M
b
p
s
)
SISO 1 1
MRC 1 2
STBC 2 1
STBC 2 2
SDM 2 2
Figure 6: Multiple-antenna gain in the downlink, static environ-
ment, MCBS-CDMA; dashed curves: 16-QAM and coding rate 1/2,
solid curves: 64-QAM and coding rate 3/4.
35 30 25 20 15 10 5 0
E
b
/N
0
(dB)
3
2.5
2
1.5
1
0.5
0
G
o
o
d
p
u
t
(
M
b
p
s
)
SISO 1 1
MRC 1 2
STBC 2 1
STBC 2 2
SDM 2 2
Figure 7: Multiple-antenna gain in the uplink, static environment,
SCBS-CDMA; dashed curves: 16-QAM and coding rate 1/2, solid
curves: 64-QAM and coding rate 3/4.
Figures 8 and 9 illustrate the system performance sensi-
tivity to the users load assuming static channels. Interblock
spreading (MCBS-CDMA and SCBS-CDMA) is compared
to intrablock spreading (MC-CDMA and SC-CDMA). The
number of users ranges from1 to 8. Again the same two com-
binations of the constellation and channel coding rate have
been selected (16-QAM with coding rate 1/2, 64-QAM with
coding rate 3/4). In case of interblock spreading, the MMSE
multiuser receiver reduces to an equivalent but simpler
20 15 10 5 0
E
b
/N
0
(dB)
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
G
o
o
d
p
u
t
(
M
b
p
s
)
IA 1 user
IA 5 users
IA 8 users
IE x users
Figure 8: Impact of the user load on the downlink, static environ-
ment, MCBS-CDMA (IE stands for interblock spreading) and MC-
CDMA (IA stands for intrablock spreading), using MRC; dashed
curves: 16-QAM and coding rate 1/2, solid curves: 64-QAM and
coding rate 3/4.
20 15 10 5 0
E
b
/N
0
(dB)
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
G
o
o
d
p
u
t
(
M
b
p
s
)
IA 1 user
IA 5 users
IA 8 users
IE x users
Figure 9: Impact of the user load on the uplink, static environment,
SCBS-CDMA (IE stands for interblock spreading) and SC-CDMA
(IA stands for intrablock spreading), using STBC; dashed curves:
16-QAMand coding rate 1/2, solid curves: 64-QAMand coding rate
3/4.
single-user receiver, which performs channel-independent
block despreading followed by MMSE single-user equaliza-
tion. MCBS-CDMA and SCBS-CDMA are MUI-free trans-
mission schemes, such that their users goodput remains un-
aected by the users load. In case of intrablock spreading,
the MMSE multiuser receiver outperforms the single-user
detector. In the downlink, the performance of the MMSE
multiuser joint detector is slightly decreasing for an in-
creasing number of users and converges to the one of the
Flexible Transmission Scheme for 4G Wireless Systems 319
250 200 150 100 50 0
Speed (km/h)
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
G
o
o
d
p
u
t
(
M
b
p
s
)
16-QAM, 1/2, IA, SNR = 8 dB
16-QAM, 1/2, IE, SNR = 8 dB
64-QAM, 3/4, IA, SNR = 16 dB
64-QAM, 3/4, IE, SNR = 16 dB
Figure 10: Impact of the terminal speed on the downlink, MCBS-
CDMA (IE stands for interblock spreading) and MC-CDMA (IA
stands for intrablock spreading), using MRC.
single-detector at full user load. MMSE multiuser reception
is especially needed in the uplink, since a single-user receiver
cannot get rid of the MUI, and features a BER curve at-
tening already at low SNRs. The impact of the users load is
much more pronounced in the SC-CDMA uplink than in the
MC-CDMA downlink since the signals propagate through
dierent channels, which is more dicult to compensate for.
In the downlink, MC-CDMA always outperforms MCBS-
CDMA since it benets from the frequency diversity oered
by the CDMA spreading. In the uplink, the performance of
SCBS-CDMAis equivalent to the one of SC-CDMAfor a typ-
ical users load, worse for a small users load and better for a
high users load.
Figures 10 and 11 compare the eect of the terminal
speed on the users goodput in case of intrablock and in-
terblock spreading. A typical users load of 5 users has
been assumed, which makes MC-CDMA (SC-CDMA) and
MCBS-CDMA (SCBS-CDMA) perform equally well in static
conditions. For each combination of the constellation and
coding rate, the SNR value corresponding to the maximum
slope in the goodput curves has been chosen (realistic work-
ing point in the curves illustrated in Figures 8 and 9). Since
the symbol block duration is higher in case of interblock
spreading than in case of intrablock spreading, the perfor-
mance of MCBS-CDMA (SCBS-CDMA) is signicantly re-
duced, while the performance of MC-CDMA (SC-CDMA)
remains acceptable when the speed of the mobile terminals
increases. Since the orthogonality between the users is lost
when the speed increases, the impact is more severe in the
uplink than in the downlink. If a 10-percent performance
loss is acceptable, interblock spreading should be used up to
60 km/h in the downlink or 10 km/h in the uplink for detec-
tion complexity reasons, while intrablock spreading should
250 200 150 100 50 0
Speed (km/h)
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
G
o
o
d
p
u
t
(
M
b
p
s
)
16-QAM, 1/2, IA, SNR = 8 dB
16-QAM, 1/2, IE, SNR = 8 dB
64-QAM, 3/4, IA, SNR = 16 dB
64-QAM, 3/4, IE, SNR = 16 dB
Figure 11: Impact of the terminal speed on the uplink, SCBS-
CDMA (IE stands for interblock spreading) and SC-CDMA (IA
stands for intrablock spreading), using STBC.
be used for higher speeds. This conclusion is in line with
the two-dimensional spreading strategy proposed in [16, 17]
for a MC-based system, that prioritizes the spreading in the
time domain rather than in the frequency domain, for the
sake of complexity and performance. It is also interesting to
note that the goodput achieved with high constellations at
high speed is smaller than the goodput achieved with low
constellations.
6. CONCLUSIONS
A generic transmission scheme has been designed that al-
lows to instantiate all the combinations of OFDM and
cyclic-prexed SC modulations with DS-CDMA. The SDM
and STBC multiple-antenna techniques have been inte-
grated in the generic transmission scheme. For each resulting
mode, the optimal linear MMSE multiuser receiver has been
derived.
A mode selection strategy has also been proposed that
trades o eciently the communication performance in a
typical suburban dynamic outdoor environment with the
complexity and PAPR at the mobile terminal.
(i) A hybrid modulation scheme (MC in the downlink,
SC in the uplink) should be used in order to minimize the
mobile terminal PAPR and data processing power.
(ii) Under low-to-medium mobility conditions, it is bet-
ter to use a high constellation and a low channel coding rate
to achieve the maximum goodput for a given SNR value.
However, the lowest constellation order should be selected
under high mobility conditions.
(iii) One directive antenna should be used at the base
station to increase the cell capacity and multiple antennas
should be used at the mobile terminal in combination with
320 EURASIP Journal on Wireless Communications and Networking
diversity techniques like MRC reception and STBC transmis-
sion to improve the link reliability.
(iv) Since interblock and intrablock spreading perform
equally well in typical user loads, interblock spreading should
be used at low terminal speeds to minimize the data pro-
cessing complexity while intrablock spreading should only be
used at high terminal speeds.
It has been demonstrated that an adaptive transceiver is
interesting to support dierent communication modes and
to eciently track the changing communication conditions.
REFERENCES
[1] T. Ojanper a and R. Prasad, Wideband CDMA for Third Gen-
eration Mobile Communications, Artech House, Norwood,
Mass, USA, 1998.
[2] R. van Nee, G. Awater, M. Morikura, H. Takanashi, M. Web-
ster, and K. W. Halford, New high-rate wireless LAN stan-
dards, IEEE Commun. Mag., vol. 37, no. 12, pp. 8288, 1999.
[3] D. Falconer, S. L. Ariyavisitakul, A. Benyamin-Seeyar, and
B. Eidson, Frequency domain equalization for single-carrier
broadband wireless systems, IEEE Commun. Mag., vol. 40,
no. 4, pp. 5866, 2002.
[4] N. Yee, J.-P. Linnartz, and G. Fettweis, Multicarrier CDMA
in indoor wireless radio networks, in Proc. IEEE International
Symposium on Personal, Indoor and Mobile Radio Communi-
cations (PIMRC 93), vol. 1, pp. 109113, Yokohama, Japan,
September 1993.
[5] K. Fazel, Performance of CDMA/OFDM for mobile commu-
nication system, in Proc. 2nd IEEE International Conference
on Universal Personal Communications (ICUPC 93), vol. 2,
pp. 975979, Ottawa, Ontario, Canada, October 1993.
[6] K. Fazel, S. Kaiser, and M. Schnell, A exible and high
performance cellular mobile communications system based
on multi-carrier SSMA, Kluwer Journal of Wireless Personal
Communications, vol. 2, no. 1/2, pp. 121144, 1995.
[7] S. Kaiser, OFDM code-division multiplexing in fading chan-
nels, IEEE Trans. Commun., vol. 50, no. 8, pp. 12661273,
2002.
[8] V. M. DaSilva and E. S. Sousa, Multicarrier orthogonal
CDMA signals for quasi-synchronous communication sys-
tems, IEEE J. Select. Areas Commun., vol. 12, no. 5, pp. 842
852, 1994.
[9] S. Kondo and L. B. Milstein, Performance of multicarrier
DS-CDMA systems, IEEE Trans. Commun., vol. 44, no. 2,
pp. 238246, 1996.
[10] L.-L. Yang and L. Hanzo, Multicarrier DS-CDMA: a multiple
access scheme for ubiquitous broadband wireless communi-
cations, IEEE Commun. Mag., vol. 41, no. 10, pp. 116124,
2003.
[11] F. Petr e, G. Leus, M. Moonen, and H. De Man, Multi-
carrier block-spread CDMA for broadband cellular down-
link, EURASIP Journal on Applied Signal Processing, vol. 2004,
no. 10, pp. 15681584, 2004, Special Issue on Multi-Carrier
Communications and Signal Processing .
[12] M. Vollmer, M. Haardt, and J. Gotze, Comparative study of
joint-detection techniques for TD-CDMA based mobile ra-
dio systems, IEEE J. Select. Areas Commun., vol. 19, no. 8,
pp. 14611475, 2001.
[13] K. L. Baum, T. A. Thomas, F. W. Vook, and V. Nangia,
Cyclic-prex CDMA: an improved transmission method for
broadband DS-CDMA cellular systems, in Proc. IEEE Wire-
less Communications and Networking Conference (WCNC 02),
vol. 1, pp. 183188, Orlando, Fla, USA, March 2002.
[14] S. Zhou, G. B. Giannakis, and C. Le Martret, Chip-
interleaved block-spread code division multiple access, IEEE
Trans. Commun., vol. 50, no. 2, pp. 235248, 2002.
[15] F. Petr e, G. Leus, L. Deneire, and M. Moonen, Downlink
frequency-domain chip equalization for single-carrier block
transmission DS-CDMA with known symbol padding, in
Proc. IEEE Global Telecommunications Conference (GLOBE-
COM 02), vol. 1, pp. 453457, Taipei, Taiwan, November
2002.
[16] H. Atarashi, N. Maeda, S. Abeta, and M. Sawahashi, Broad-
band packet wireless access based on VSF-OFCDM and
MC/DS-CDMA, in Proc. 13th IEEE International Sympo-
sium on Personal, Indoor and Mobile Radio Communications
(PIMRC 02), vol. 3, pp. 992997, Lisbon, Portugal, Septem-
ber 2002.
[17] N. Chapalain, D. Mottier, and D. Castelain, An OFCDM up-
link transmission system with channel estimation based on
spread pilots, in Proc. IST Mobile and Wireless Communica-
tions Summit, Lyon, France, June 2004.
[18] G. J. Foschini and M. J. Gans, On limits of wireless commu-
nications in a fading environment when using multiple an-
tennas, Kluwer Journal of Wireless Personal Communications,
vol. 6, no. 3, pp. 311335, 1998.
[19] G. G. Raleigh and J. M. Cio, Spatio-temporal coding
for wireless communication, IEEE Trans. Commun., vol. 46,
no. 3, pp. 357366, 1998.
[20] D. Gesbert, H. Bolcskei, D. A. Gore, and A. J. Paulraj, Out-
door MIMO wireless channels: models and performance pre-
diction, IEEE Trans. Commun., vol. 50, no. 12, pp. 1926
1934, 2002.
[21] A. J. Paulraj and T. Kailath, Increasing capacity in wireless
broadcast systems using distributed transmission/directional
reception (DTDR), U.S. Patent 5345599, Stanford University,
Stanford, Calif, USA, 1994.
[22] G. J. Foschini, Layered space-time architecture for wireless
communication in a fading environment when using multiple
antennas, Bell Labs Technical Journal, vol. 1, no. 2, pp. 4159,
1996.
[23] F. W. Vook, T. A. Thomas, and K. L. Baum, Transmit array
processing for cyclic-prex CDMA, in Proc. 56th IEEE Ve-
hicular Technology Conference (VTC 02), vol. 1, pp. 270274,
Vancouver, British Columbia, Canada, September 2002.
[24] V. Tarokh, N. Seshadri, and A. R. Calderbank, Space-time
codes for high data rate wireless communication: perfor-
mance criterion and code construction, IEEE Trans. Inform.
Theory, vol. 44, no. 2, pp. 744765, 1998.
[25] S. M. Alamouti, A simple transmit diversity technique for
wireless communications, IEEE J. Select. Areas Commun.,
vol. 16, no. 8, pp. 14511458, 1998.
[26] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, Space-time
block codes from orthogonal designs, IEEE Trans. Inform.
Theory, vol. 45, no. 5, pp. 14561467, 1999.
[27] E. Lindskog and A. J. Paulraj, A transmit diversity scheme for
channels with intersymbol interference, in Proc. IEEE Inter-
national Conference on Communications (ICC 00), vol. 1, pp.
307311, New Orleans, La, USA, June 2000.
[28] N. Al-Dhahir, Single-carrier frequency-domain equalization
for space-time block-coded transmissions over frequency-
selective fading channels, IEEE Commun. Lett., vol. 5, no. 7,
pp. 304306, 2001.
[29] S. Zhou and G. B. Giannakis, Single-carrier space-time
block-coded transmissions over frequency-selective fading
channels, IEEE Trans. Inform. Theory, vol. 49, no. 1, pp. 164
179, 2003.
[30] S. Barbarossa and F. Cerquetti, Simple space-time coded
Flexible Transmission Scheme for 4G Wireless Systems 321
SS-CDMA systems capable of perfect MUI/ISI elimination,
IEEE Commun. Lett., vol. 5, no. 12, pp. 471473, 2001.
[31] F. W. Vook, T. A. Thomas, and K. L. Baum, Cyclic-prex
CDMA with antenna diversity, in Proc. 55th IEEE Vehicu-
lar Technology Conference (VTC 02), vol. 2, pp. 10021006,
Birmingham, Ala, USA, May 2002.
[32] F. Petr e, G. Leus, L. Deneire, M. Engels, M. Moonen, and
H. De Man, Space-time block coding for single-carrier
block transmission DS-CDMA downlink, IEEE J. Select. Ar-
eas Commun., vol. 21, no. 3, pp. 350361, 2003, Special Issue
on MIMO Systems and Applications .
[33] Z. Liu and G. B. Giannakis, Space-time block-coded multi-
ple access through frequency-selective fading channels, IEEE
Trans. Commun., vol. 49, no. 6, pp. 10331044, 2001.
[34] A. Stamoulis, Z. liu, and G. B. Giannakis, Space-time block-
coded OFDMA with linear precoding for multirate services,
IEEE Trans. Signal Processing, vol. 50, no. 1, pp. 119129, 2002.
[35] Z. Wang and G. B. Giannakis, Wireless multicarrier commu-
nications, IEEE Signal Processing Mag., vol. 17, no. 3, pp. 29
48, 2000.
[36] G. B. Giannakis, P. A. Anghel, and Z. Wang, Wideband gen-
eralized multicarrier CDMA over frequency-selective wire-
less channels, in Proc. IEEE Int. Conf. Acoustics, Speech, Sig-
nal Processing (ICASSP 00), vol. 5, pp. 25012504, Istanbul,
Turkey, June 2000.
[37] G. B. Giannakis, Z. Wang, A. Scaglione, and S. Bar-
barossa, AMOUR-generalized multicarrier transceivers for
blind CDMA regardless of multipath, IEEE Trans. Commun.,
vol. 48, no. 12, pp. 20642076, 2000.
[38] Q. Chen, E. S. Sousa, and S. Pasupathy, Performance of a
coded multi-carrier DS-CDMA system in multipath fading
channels, Kluwer Journal of Wireless Personal Communica-
tions, vol. 2, no. 1-2, pp. 167183, 1995.
[39] A. Klein, G. K. Kaleh, and P. W. Baier, Zero forcing and mini-
mum mean-square-error equalization for multiuser detection
in code-division multiple-access channels, IEEE Trans. Veh.
Technol., vol. 45, no. 2, pp. 276287, 1996.
[40] L. Vandendorpe, Performance analysis of IIR and FIR lin-
ear and decision-feedback MIMO equalizers for transmulti-
plexers, in Proc. IEEE International Conference on Commu-
nications (ICC 97), vol. 2, pp. 657661, Montreal, Quebec,
Canada, June 1997.
[41] 3GPP Technical specication group, TR25.996 v6.1.0, Spatial
channel model for Multiple Input Multiple Output (MIMO)
simulations, September 2003.
Francois Horlin was born in Bruxelles, Bel-
gium, in 1975. He received the Electri-
cal Engineering degree and the Ph.D. de-
gree from the Universit e catholique de Lou-
vain (UCL), Louvain-la-Neuve, Belgium, in
1998 and 2002, respectively. In September
2002, he joined the Interuniversity Micro-
Electronics Center (IMEC), Leuven, Bel-
gium, as a Senior Researcher. He is currently
the Head of the research activities on digital
signal processing for broadband communications. His elds of in-
terest are in digital signal processing, mainly for high-bit-rate mul-
tiuser communications. During his Ph.D. thesis, he proposed a new
solution for jointly optimizing the transmitter and the receiver in a
multiuser multi-input multi-output (MIMO) type of communica-
tion systems. He is now working on the integration of a multistan-
dard mobile terminal. In this context, he is responsible for the de-
velopment of the functionality of a fourth-generation (4G) cellular
system, targeting high capacity in very high mobility conditions.
Frederik Petr e was born in Tienen, Bel-
gium, on December 12, 1974. He received
the Electrical Engineering degree and the
Ph.D. degree in applied sciences from the
Katholieke Universiteit Leuven (KULeu-
ven), Leuven, Belgium, in July 1997 and
December 2003, respectively. In September
1997, he joined the Design Technology for
Integrated Information and Communica-
tion Systems (DESICS) Division at the In-
teruniversity Micro-Electronics Center (IMEC) in Leuven, Bel-
gium. Within the Digital Broadband Terminals (DBATE) Group
of DESICS, he rst performed predoctoral research on wireline
transceiver design for twisted pair, coaxial cable, and powerline
communications. During the fall of 1998, he visited the Informa-
tion Systems Laboratory (ISL) at Stanford University, Calif, USA,
working on OFDM-based powerline communications. In January
1999, he joined the Wireless Systems (WISE) Group of DESICS
as a Ph.D. researcher, funded by the Institute for Scientic and
Technological Research in Flanders (IWT). Since January 2004, he
has been a Senior Scientist within the Wireless Research Group of
DESICS. He is investigating the baseband signal processing algo-
rithms and architectures for future wireless communication sys-
tems, like third-generation (3G) and fourth-generation (4G) cel-
lular networks, and wireless local area networks (WLANs). His
main research interests are in modulation theory, multiple-access
schemes, channel estimation and equalization, smart antenna, and
MIMO techniques. He is a Member of the ProRISC Technical Pro-
gram Committee and the IEEE Benelux Section on Communica-
tions and Vehicular Technology (CVT). He is a Member of the Ex-
ecutive Board and Project Leader of the Flexible Radio Project of
the Network of Excellence in Wireless Communications (NEW-
COM), established under the sixth framework of the European
Commission.
Eduardo Lopez-Estraviz was born in Fer-
rol, Spain, in 1977. He received the
Telecommunications Engineering degree
from the Universidad de Vigo (UVI), Vigo,
Spain, in 1996 and 2002, respectively. In
June 2002, he joined the Interuniversity
Micro-Electronics Center (IMEC), Leuven,
Belgium, as a researcher. His elds of inter-
est are in digital signal processing for broad-
band communications. He is now working
on the integration of a multistandard mobile terminal. In this
context, he is working on developing functionality for a fourth-
generation (4G) cellular system, targeting high capacity in very
high mobility conditions.
Frederik Naessens was born in Roeselare,
Belgium, in 1979. He received the Engi-
neering degree from the Catholic School for
Higher Education Bruges-Ostend (KHBO)
Academy in Ostend, Belgium, in 2001. Af-
ter this studies, he joined the Interuniversity
Micro-Electronics Center (IMEC), Leuven,
Belgium, as a development engineer. Cur-
rently he is also a part-time student follow-
ing an M.S. degree in ICT at the University
of Kent, United Kingdom. He has been involved in a demonstra-
tor setup for GPS and GLONASS satellite navigation systems. He
is now working on developing functionality for future cellular sys-
tems (4G).
322 EURASIP Journal on Wireless Communications and Networking
Liesbet Van der Perre received the M.S. de-
gree and the Ph.D. degree in electrical engi-
neering from the KU Leuven, Belgium, in
1992 and 1997, respectively. Her work in
the past focused on system design and dig-
ital modems for high-speed wireless com-
munications. She was a System Architect in
IMECs OFDM ASICs development and a
Project Leader for the turbo codec. Cur-
rently, she is leading the Software-Dened
Radio Baseband Project, and she is the Scientic Director of Wire-
less Research in IMEC.
EURASIP Journal on Wireless Communications and Networking 2005:3, 323332
c 2005 Ying-Chang Liang et al.
Recongurable Signal Processing and Hardware
Architecture for Broadband Wireless Communications
Ying-Chang Liang
Institute for InfoComm Research (I2R), 21 Heng Mui Keng Terrace, Singapore 119613
Email: ycliang@i2r.a-star.edu.sg
Sayed Naveen
Institute for InfoComm Research (I2R), 21 Heng Mui Keng Terrace, Singapore 119613
Email: naveen@i2r.a-star.edu.sg
Santosh K. Pilakkat
Institute for InfoComm Research (I2R), 21 Heng Mui Keng Terrace, Singapore 119613
Email: pilakkat@i2r.a-star.edu.sg
Ashok K. Marath
Institute for InfoComm Research (I2R), 21 Heng Mui Keng Terrace, Singapore 119613
Email: ashok@i2r.a-star.edu.sg
Received 1 October 2004; Revised 22 April 2005
This paper proposes a broadband wireless transceiver which can be recongured to any type of cyclic-prex (CP) -based com-
munication systems, including orthogonal frequency-division multiplexing (OFDM), single-carrier cyclic-prex (SCCP) system,
multicarrier (MC) code-division multiple access (MC-CDMA), MC direct-sequence CDMA (MC-DS-CDMA), CP-based CDMA
(CP-CDMA), and CP-based direct-sequence CDMA (CP-DS-CDMA). A hardware platform is proposed and the reusable com-
mon blocks in such a transceiver are identied. The emphasis is on the equalizer design for mobile receivers. It is found that
after block despreading operation, MC-DS-CDMA and CP-DS-CDMA have the same equalization blocks as OFDM and SCCP
systems, respectively, therefore hardware and software sharing is possible for these systems. An attempt has also been made to
map the functional recongurable transceiver onto the proposed hardware platform. The dierent functional entities which will
be required to perform the reconguration and realize the transceiver are explained.
Keywords and phrases: recongurable signal processing, broadband communications, software-dened radio, cyclic prex,
frequency-domain equalization.
1. INTRODUCTION
A number of wireless standards govern personal wireless
communications, to name a few, GSM and CDMA for 2G
cellular networks; WCDMA, CDMA-2000, and TDS-CDMA
for 3G cellular networks; IEEE 802.11a/b/g for wireless lo-
cal area networks; IEEE 802.16 for wireless wide area net-
works (WiMAX); and IEEE 802.15 for personal area net-
works (PAN). In order to satisfy the need of customers mo-
bility, the designed radio transceivers should not be tied to
any specic network. Software-dened radios (SDRs) [1] or
This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
cognitive radios (CRs) [2] are the well-suited implementa-
tions of such network independent radios.
Dierent frame format and modulation schemes have
been proposed for dierent networks. For example, orthog-
onal frequency-division multiplexing (OFDM) [3] has been
adopted in IEEE 802.11a and IEEE 802.11g, and IEEE 802.16.
Single-carrier cyclic-prex (SCCP) system is selected for
IEEE 802.16 as well. Cyclic-prex (CP) -based CDMA sys-
tems are considered for beyond 3G systems. For example,
CP-CDMA and CP-based direct-sequence CDMA (CP-DS-
CDMA) are candidate schemes for enhanced versions of 3G
DS-CDMA systems [4, 5, 6, 7]; multicarrier CDMA (MC-
CDMA) and multicarrier direct-sequence CDMA (MC-DS-
CDMA) [8] are being extensively studied for potential adop-
tion in beyond 3G (B3G) or fourth-generation (4G) cellular
324 EURASIP Journal on Wireless Communications and Networking
systems [10, 21]. The common feature of these systems is
that they are block-based transmission systems, and a CP
portion is inserted to each data block in order to suppress
the interblock interference (IBI) and to simplify the receiver
design. SDR structures and implementations have been ex-
plored in [11, 12, 13, 14, 15, 16, 17, 18, 19] in dierent ways.
In this paper, we propose a universal mobile transceiver
which is congurable to any type of CP-based systems. We
present a hardware platform for the proposed transceiver
and identify the reusable common blocks. In particular, a
study of the equalizer design at the mobile terminals is car-
ried out for each system. Single-user environments are con-
sidered for OFDM and SCCP systems. For CDMA mul-
tiuser systems, we target a receiver architecture for the down-
link only. It is found that after block despreading operation,
MC-DS-CDMA and CP-DS-CDMA have the same equal-
ization blocks as OFDM and SCCP systems, respectively,
showing that hardware and software sharing is possible for
these systems. More specically, denoting N and G as the
block size and processing gain, respectively, these two sys-
tems perform signal detection of N symbols over G consec-
utive time blocks. However, MC-CDMA and CP-CDMA re-
quire only one time block to perform signal detection. The
receiver complexity of MC-CDMA and CP-CDMA can be
much higher if near ML detection is required [5] due to the
requirement of N-dimensional signal detection within each
time block.
In [11], a recongurable architecture for multicarrier
based CDMA systems is proposed. Specically, the systems
discussed there include MC-CDMA, MC-DS-CDMA, and
MT-CDMA of [8] only. The architecture proposed in our pa-
per is more generic in the sense that it can be recongured
to support both multicarrier systems and single-carrier sys-
tems, including SCCP, CP-CDMA, and CP-DS-CDMA. Fur-
thermore, we propose a hardware platform to support the re-
congurable architecture. An attempt has been made to map
the functional recongurable transceiver onto the proposed
hardware platform. The dierent functional entities which
will be required to perform the reconguration and realize
the transceiver are explained.
This paper is organized as follows. In Section 2, the
system models are described for various CP-based sys-
tems. In Section 3, the linear MMSE receivers are specied
for dierent systems. Section 4 proposes the recongurable
transceiver and details the congurations for each system.
We also analyze the complexity of the equalizer for each sys-
tem. The hardware architecture for the proposed transceiver
is proposed in Section 5. Finally, conclusions are drawn in
Section 6.
2. SYSTEMDESCRIPTION
In this section, we review the input-output relations for the
CP-based systems, which serve as the basis for the design of
recongurable signal processing and hardware architecture.
The transmission is on a block-by-block basis, with each
block consisting of the CP sub-block of P symbols and the
data sub-block of N symbols. Thanks to the use of CP, the
channel matrix is a circular matrix, which can be represented
as H = W
H
W, where W is the N-point DFT matrix, and
= diag{
1
, . . . ,
N
} is a diagonal matrix, the elements of
which are the channels frequency-domain responses in each
subcarrier.
For multicarrier systems, the data symbol sub-block is
pretransformed using the N-point IDFT matrix W
H
; while
for single-carrier systems, the data symbol sub-block is sent
out directly. At the receiver side, both single-carrier and mul-
ticarrier systems rst pretransform the received data block
with the DFT matrix W, the output of which is then utilized
to recover the transmitted signals via either linear or nonlin-
ear receivers.
For the multiuser CDMA case, spreading and despread-
ing are added in the transmitter and receiver, respectively.
Since the transceiver design for the mobile terminal is of in-
terest, for the CDMA multiuser case, we concentrate on the
receiver architecture for the downlink only. For OFDM and
SCCP systems, single-user environments are considered.
2.1. OFDM
For OFDMsystems, the input-output relation at the nth time
block after FFT can be expressed as [3]
y(n) = s(n) + u(n), (1)
where s(n) is the N 1 input signal vector, y(n) is the N 1
FFT output of the received signal vector, and u(n) is the FFT
output of the AWGN vector, which is of dimension N 1.
2.2. SCCP systems
For SCCP systems, the input-output relation at the nth block
after FFT can be expressed as [20]
y(n) = Ws(n) + u(n), (2)
where s(n), y(n), and u(n) are dened as in OFDM systems.
Next, we look at four CDMA-based systems: MC-DS-
CDMA, MC-CDMA, CP-CDMA, and CP-DS-CDMA. Sup-
pose all users have the same processing gain G. We introduce
the following common notations for all systems: T for to-
tal number of users; D(n) for long scrambling codes at the
nth block, where D(n) = diag{d(n; 0), . . . , d(n; N 1)}; c
i
for short codes of user i, c
i
= [c
i
(0), . . . , c
i
(G 1)]
T
, with
c
H
i
c
j
= 1 for i = j, and c
H
i
c
j
= 0 for i = j.
2.3. MC-DS-CDMA
MC-DS-CDMA performs time-domain spreading [8] as fol-
lows.
(1) N symbols from each user-specic spreading codes
are taken.
(2) The chip signals corresponding to the same symbol
index from all users are summed and transmitted over the
same subcarrier but through dierent times blocks. Thus it
takes G consecutive blocks to transmit out the whole chip
sequences of the N symbols.
Recongurable Signal Processing for Wireless Communications 325
Mathematically, the nth received block after FFT
operation can be written as
y(n) = D(n)
T1
_
i=0
c
i
(n)s
i
+ u(n), (3)
where
y(n) =
_
y(n; 0), . . . , y(n; N 1)
_
T
,
= diag
_

0
, . . . ,
N1
_
,
D(n) = diag
_
d(n; 0), . . . , d(n; N 1)
_
,
s
i
=
_
s
i
(0), . . . , s
i
(T 1)
_
T
,
u(n) =
_
u(n; 0), . . . , u(n; N 1)
_
T
.
(4)
Note that u(n) is the FFT output of the AWGN vector. On
the other hand, gathering the received signals of the kth sub-
carrier from 0th block to (G 1)th block yields
` y(k) =
k
M
_
i=0
s
i
(k)
`
D(k)c
i
+ u(k), (5)
where
` y(k) =
_
y(0; k), . . . , y(G 1; k)
_
T
,
`
D(k) = diag
_
d(0; k), . . . , d(G 1; k)
_
,
u(k) =
_
u(0; k), . . . , u(G 1; k)
_
T
.
(6)
Despreading ` y(k) using
`
D(k)c
i
, we obtain
z
i
(k) = c
H
i
`
D
H
(k)` y(k)
=
k
s
i
(k) + u(k).
(7)
Thus we have
z
i
= s
i
+ u
i
, (8)
where z
i
= [z
i
(0), . . . , z
i
(N 1)]
T
, and u
i
is dened accord-
ingly. Therefore, after block despreading, an MC-DS-CDMA
is equivalent to an OFDM, and it does not experience any
MAI and ISI.
2.4. MC-CDMA
MC-CDMAperforms spreading over frequency domain. De-
note T as the total number of users and M = N/G as a posi-
tive integer. The nth received block after FFT can be written
as
y(n) = D(n)Cs(n) + u(n), (9)
where
y(n) =
_
y(n; 0), . . . , y(n; N 1)
_
T
,
D(n) = diag
_
d(n; 0), . . . , d(n; N 1)
_
,
C = diag
_

C, . . . ,

C
_
,
s(n) =
_
s
T
1
(n), . . . , s
T
M
(n)
_
T
,
(10)
and u(n) is the FFT output of the AWGN vector. Further,

C =
_
c
0
, . . . , c
T1
_
,
s
i
(n) =
_
s
0
(n; i), . . . , s
T1
(n; i)
_
T
.
(11)
Dividing y(n) into M nonoverlapping short-column vectors,
each with G elements, (9) can then be decoupled as
y
i
(n) =
i
D
i
(n)

C s
i
(n) + u
i
(n) (12)
for i = 1, . . . , M, where
i
and D
i
(n) are the ith sub-blocks
of and D(n). From (12), MC-CDMA experiences MAI, but
not ISI.
2.5. CP-CDMA
CP-CDMA is a single-carrier dual of MC-CDMA. The M =
N/G symbols of each user are rst spread out with user-
specic spreading codes, then the chip sequence for all users
are summed up; the total chip signal of size N is then passed
to CP inserter, which adds a CP. Using the duality between
CP-CDMA and MC-CDMA, from (12), the nth received
block of CP-CDMA after FFT can be written as [5]
y(n) = WD(n)Cs(n) + u(n). (13)
From the above, it is seen that CP-CDMA experiences both
MAI and ISI.
2.6. CP-DS-CDMA
CP-DS-CDMA is the single-carrier dual of MC-DS-CDMA.
It performs block spreading as follows: the N symbols of each
user are rst spread out with its own spreading codes, then
the chip sequence for all users are summed up; the total chip
signal corresponding to dierent chip indices is transmitted
over dierent time block, thus it takes G blocks to send out
the whole chip sequence of the N symbols.
The nth received block after FFT can be written as [5]
y(n) = WD(n)
T1
_
i=0
c
i
(n)s
i
+ u(n), (14)
where n = 0, . . . , G 1.
Simplication is available if the same set of long scram-
bling codes are chosen for blocks from 0 to (G1), or if long
scrambling codes are not used (D(n) = I). We consider the
326 EURASIP Journal on Wireless Communications and Networking
case when long scrambling codes are not used, and collect
the kth subcarriers received signals from the 0th to (G1)th
blocks, we obtain
` y(k) =
k
T1
_
i=0
W(k, :)s
i
c
i
+ u(k), (15)
where W(k, :) denotes the kth row of matrix W,
` y(k) =
_
y(0; k), . . . , y(G 1; k)
_
T
,
c
i
=
_
c
i
(0), . . . , c
i
(G 1)
_
T
,
u(k) =
_
u(0; k), . . . , u(G 1; k)
_
T
.
(16)
Using c
i
to perform despreading, we obtain
z
i
(k) = c
H
i
` y(k)
=
k
W(k, :)s
i
+ u(k).
(17)
Thus we have
z
i
= Ws
i
+ u
i
, (18)
which is the single-carrier duality of the MC-DS-CDMA
model (8). From (18), it is seen that CP-DS-CDMA does not
experience MAI, but it does have ISI.
3. LINEAR EQUALIZERS
We rst look at the generic multiple-input multiple-output
(MIMO) channel model. Suppose the MIMO channel is de-
scribed by the following input-output relation:
x = Hs + n, (19)
where s, x, and n denote the transmitted signal vector, re-
ceived signal vector, and received noise vector, respectively;
H is the channel matrix, representing the responses from the
transmit antennas to the receive antennas. Without loss of
generality, we assume that s, x, and n are all N 1 vectors,
and H is an N N matrix. Assume that E[ss
H
] =
2
s
I,
E[nn
H
] =
2
n
I, and E[sn
H
] = 0. Denote P as the linear
equalizer for the channel (19), which generates the output
z = P
H
x. (20)
Then, the minimum mean square error (MMSE) equalizer is
given by
P =
2
s
_

2
s
HH
H
+
2
n
I
_
1
H. (21)
Now, we are ready to specify the linear receivers for each of
the CP-based systems.
3.1. MC-DS-CDMA/OFDM
An MC-DS-CDMA after block despreading is equivalent to
an OFDM. Thus both can employ the same one-tap equalizer
to retrieve the transmitted signals: (
2
s
[
2
s
||
2
+
2
n
I]
1
)
H
.
Note that MC-DS-CDMA performs symbol-level equaliza-
tion.
3.2. MC-CDMA
For MC-CDMA, the signal separation is done for each sub-
block. The channel matrix for the ith sub-block of the nth
block is
H
i
=
i
D
i
(n)

C. (22)
Thus the MMSE equalizer is given by
P
i
=
2
s
_

2
s

2
+
2
n
I
_
1

i
D
i
(n)

C (23)
for i = 1, . . . , Q, which can realized through a one-tap equal-
izer, (
2
s
[
2
s
||
2
+
2
n
I]
1
)
H
, followed by multiple despread-
ers, [D
i
(n)

C]
H
, i = 1, . . . , Q. Note that MC-CDMA performs
chip-level equalization.
3.3. CP-CDMA
For CP-CDMA, the channel matrix for the nth block is given
by
H = WD(n)C. (24)
Thus the MMSE equalizer is given by
P =
2
s
_

2
s
||
2
+
2
n
I
_
1
WD(n)C (25)
which is realized through a one-tap equalizer, (
2
s
[
2
s
||
2
+

2
n
I]
1
)
H
, followed by an IDFT operator W
H
, followed by a
despreader [D(n)C]
H
. Note that CP-CDMA performs chip-
level equalization.
3.4. CP-DS-CDMA/SCCP
A CP-DS-CDMA performs block despreading using G con-
secutive blocks, and after block despreading, it is equivalent
to a SCCP system both have the following channel matrix:
H = W. (26)
Thus the MMSE equalizer is given by
P =
2
s
_

2
s
||
2
+
2
n
I
_
1
W (27)
which is realized through a one-tap equalizer, (
2
s
[
2
s
||
2
+

2
n
I]
1
)
H
, followed by a IDFT operator W
H
. Note that
chip-level equalization is performed for CP-DS-CDMA.
Recongurable Signal Processing for Wireless Communications 327
Source coding
Scrambling
Channel
coding
Interleaving
Pilot insertion
Modem
Tx
Burst formation
Tx FE
processing
Channel
Receiver FE
processing
Modem
Rx
Constellation
demapping
Descrambling
Viterbi/turbo
decoding
Deinterleaving
Figure 1: Block diagram of the proposed transceiver congurable for CP-based systems.
Block interleaving
mapping
Serial
to
parallel
Code
spreading
IFFT CP insertion
Parallel
to
serial
Figure 2: Functional block of the recongurable Modem Tx.

Serial
to
parallel
CP removal
FFT
Block
despread
One-tap
equalizer
IFFT
Despreading
Synchronization
Figure 3: Functional block of the recongurable Modem Rx.
4. PROPOSEDTRANSCEIVER
Based on the input-output relations and the detailed equal-
izers presented in the previous section for each CP-based sys-
tem, in this section, a universal transceiver is proposed which
is congurable to any type of these systems.
4.1. Transceiver overview
The proposed transceiver is illustrated in Figure 1, which
contains source coding, scrambling, channel coding, inter-
leaving, and pilot insertion blocks, as well as Modem Tx
and burst formation block at the transmitter side; and re-
ceiver front end (FE) and Modem Rx as well as constellation
demapper, descrambling, decoding, and deinterleaving at the
receiver side. The data symbols are passed to the recong-
urable Modem Tx block where they are modulated accord-
ing to the required specication. The burst formation and
Tx front-end format the processed symbols to proper fram-
ing structure with suitable preambles and they are modulated
onto a carrier and transmitted. After undergoing the channel
and additive noise distortion, the received signal is down-
converted in the RX FE processing and fed to the cong-
urable Modem Rx, which does the synchronization (frame,
code, and frequency), CP removal, FFT, channel estimation,
channel equalization, and phase compensation. The detected
symbols are further processed in the subsequent block to
generate the information bits.
In the recongurable transceiver, though each block has
to be recongured to meet the particular standards require-
ment, here we restrict our discussion to the Modem Tx and
Modem Rx blocks.
4.2. Systemcongurations
The Modem Tx and Modem Rx are shown in Figures 2 and
3, respectively, and are realized as parameterized functions.
We dene a ag parameter i(A) for block A; if i(A) = 1, the
block will be functioning; and if i(A) = 0, it will be unity.
We also use CS to denote code spreading block; BI for
block interleaving block; IFFT for IFFT block; FFT for
FFT block; BDI for block despreading and deinterleaving
block; OTE for one-tap equalizer; and DS for despread-
ing block.
Denote G and N as the processing gain and the size of
each block. Note that G = 1 for non-CDMA systems. The
Modem Tx takes in a set of M data symbols, where M = N
for OFDM, SCCP, MC-DS-CDMA, and CP-DS-CDMA, and
M = N/G for MC-CDMA and CP-CDMA.
The ag parameters and selection of M for Modem Tx
are shown in Table 1 for dierent systems. For non-CDMA
systems, the CS block is a unity function. The BI block works
for MC-DS-CDMA and CP-DS-CDMA, which translates the
spread output of size NG into G serial blocks, each with size
N. The BI block may also work for MC-CDMA to achieve
almost equal diversity for each transmitted symbol [9]. The
IFFT processing block in Modem Tx is switched o in the
case of single-carrier systems (SCCP, CP-CDMA, and CP-
DS-CDMA). A CP insertion block is added to remove IBI
and to translate the channel matrix into a circular matrix.
Finally, parallel-to-serial conversion is done for signal trans-
mission.
At Modem Rx, a common synchronization (SYN) block
is applied to each system. For-single-user case, this SYN
328 EURASIP Journal on Wireless Communications and Networking
Table 1: Parameter congurations for Modem Tx.
Schemes M i(CS) i(BI) i(IFFT)
OFDM N 0 0 1
SCCP N 0 0 0
MC-DS-CDMA N 1 1 1
MC-CDMA N/G 1 0/1 1
CP-CDMA N/G 1 0 0
CP-DS-CDMA N 1 1 1
Table 2: Parameter congurations for Modem Rx.
Schemes i(BDI) i(OTE) i(IFFT) i(DS)
OFDM 0 1 0 0
SCCP 0 1 1 0
MC-DS-CDMA 1 1 0 0
MC-CDMA 0 1 0 1
CP-CDMA 0 1 1 1
CP-DS-CDMA 1 1 1 0
block implements time and frequency synchronization as
well as phase noise estimation. Code acquisition is needed
for CDMA case. Channel estimation block is also common
for each system, which can be implemented using preamble
blocks or common pilot channels or dedicated pilot chan-
nels, based on the systems to be congured to. The FFT
is performed in Modem Rx for all systems. The other pa-
rameters are system dependent, which are summarized in
Table 2.
(i) For OFDM, only OTE is needed; for MC-DS-CDMA,
both BDI and OTE are required.
(ii) For SCCP, OTE and IFFE blocks function; for CP-DS-
CDMA, BDI, OTE, and IFFT blocks are required.
(iii) For MC-CDMA, both OTE and DS blocks are needed;
for CP-CDMA, OTE, IFFT, and DS blocks are all re-
quired.
4.3. Complexity issues
The equalizer complexities of dierent systems are com-
pared. For non-CDMA systems, such as OFDM and SCCP,
the block size N is usually small, say 64 in IEEE 802.11a. For
CDMA systems, the block size N can be much larger, say at
least 256. Thus we only compare the receiver complexities of
CDMA-based systems.
Table 3 illustrates the required operations per block for
each CDMA-based system. Note that for both BDI and DS
blocks, the number of despreaders of size Gis considered. We
use the number of complex multiplications (NCM) per block
as the complexity metric. As the despreading operation may
only involve addition operations since the codes are usually
BPSK or QPSK modulated, this complexity is ignored. Treat
the required NCM for an N-dimensional OTE as 3N, and
the NCM for an N-point FFT (IFFT) as N log
2
(N). Then the
required NCMs per block for each CDMA system are given
Table 3: Equalizer complexity per block for CDMA-based systems.
Schemes FFT BDI OTE IFFT DS
MC-DS-CDMA 1 N/G 1/G 0 0
MC-CDMA 1 0 1 0 N/G
CP-CDMA 1 0 1 1 N/G
CP-DS-CDMA 1 N/G 1/G 1/G 0
as follows:
(i) MC-DS-CDMA: N log
2
(N) + 3(N/G);
(ii) MC-CDMA: N log
2
(N) + 3N;
(iii) CP-CDMA: 2N log
2
(N) + 3N;
(iv) CP-DS-CDMA: N log
2
(N) + 3(N/G) + (N/G) log
2
(N).
It is seen that MC-CDMA and CP-CDMA have relatively
higher complexity than the other two. However, we point out
here that the lower complexity of MC-DS-CDMA and CP-
DS-CDMA is achieved with the assumption that the wireless
channel is static within the G consecutive blocks, which may
not be the case for fast-fading environments. In those cases,
MC-CDMA and CP-CDMA could be a better choice. For-
tunately, the complexity increment for these two systems is
around 2 times only.
As we mentioned earlier, we target the recongurable re-
ceiver architecture for the downlink only. Now, we quan-
tify the gain of doing so with respect to the multimode re-
ceiver which gathers the implementation for each system.
Adding the NCM for OFDM, which is N log
2
(N) + 3N, and
the NCM for SCCP, which is 2N log
2
(N) + 3N, with those
for CDMA systems, the required NCM per data block for
the simple architecture supporting the six modes is given by
(8N + (N/G)) log
2
(N) + 12N + 6(N/G). With the recong-
urable receiver, the required blocks are FFT, BDI, OTE, IFFT,
and DS. Again, ignoring the complexity for BDI and DS, then
the total NCM required by the recongurable receiver per
data block is 2N log
2
(N) +3N. For large processing gain, the
required NCM for the recongurable receiver is about 25%
of that for the multimode receiver.
Finally, we list out the typical values of block size N for
dierent standards. In 802.11a/g, N = 64, and in the propos-
als for 802.11n, N = 64 or N = 128. N = 256 for 802.16a/e
OFDM mode, and N is as high as 1024 [21] for B3G systems.
For CDMA-based systems, when the chip rate is xed, the
processing gain is usually variable depending on the users
data rate [21].
5. HARDWARE ARCHITECTURE
The proposed transceiver architecture is mapped onto the
hardware architecture shown in Figure 4. In this gure,
generic high-level functional subsystems and interfaces of the
software-dened radio (SDR) transceiver system are shown.
The main entities of the architecture are the general-purpose
processor (GPP) and SDR modem subsystem.
The GPP is an x86-based processor with a PCI-based in-
terface for the SDR modem subsystem. The processing re-
sources available on the SDR modem subsystem are Ti DSP
Recongurable Signal Processing for Wireless Communications 329



Waveform
driver1
(802.11)
Waveform
driver2
(WCDMA)
Waveform
driver < n >
SDR modem
driver
SDR modem subsystem
DSP
SDR executive
RTOS
Platform
manager
Host interface
manager
L
1
/ L
2
engine1
L
1
/ L
2
engine < n >
Baseband
engine1
Baseband
engine < n >
RTR Engine
RF front end
RF front end
System &
conguration
manager
User
interface
Cong. database
Upper-layer protocols (TCP/IP)
and application software
Kernel/OS
General-purpose processor
Firmware engine
Figure 4: Transceiver architecture blocks and interfaces.
TMS320C6416 running at a clock speed of 600 MHz, a Xil-
inx FPGA (XCV2V6000), and the RF front end. Along with
the analog RF circuitry, the RF front end also has the ADC
(analog devices AD6544), DAC (AD9777), and some recon-
gurable logic for digital front-end processing.
The logical functions performed in these resources will
be dependent on the type of system (OFDM, MC-DS-
CDMA, SCCP, CP-DS-CDMA, MC-CDMA, CP-CDMA)
that is being realized.
A set of logical entities should be realized on the hard-
ware platform comprised of GPP and SDR modem subsys-
tem. The specic functions performed in these resources will
be dependent on the type of system being realized in the
platform. The transient functions in the architecture which
are conguration dependent are shown in dotted outlines in
Figure 4. The functions that are always present irrespective of
the congured system are shown in solid outlines.
SDR modem subsystem as shown in Figure 4 covers the
RF and digital signal processing functions in the system. This
subsystem interfaces to the GPP as a device (specically as a
PCI device).
Burst formation, transmitter front-end (FE) processing
and receiver FE processing functions are realized in the RF
front end. In the receive path, the RF signals down-converts
(5 GHz band signals for 802.11a WLAN) to a 70 MHz IF and
quadrature is digitized into I and Q for further processing by
baseband run-time recongurable (RTR) engine in the re-
ceive section. In the transmit section, the RF subsystem up-
converts the baseband I and Q signals to the required fre-
quency (5 GHz band for 802.11a WLAN) for transmission.
The RF front end is congured to the required congura-
tion during the initial conguration process. The parameters
for conguration are provided by the respective waveform
driver.
The platform has a run-time recongurable (RTR)
programmable logic hardware engine. This is used in
computation-intensive front-end signal processing of Mo-
dem Tx shown in Figure 2 and Modem Rx of Figure 3. This
subsystem consists of recongurable hardware and a man-
agement rmware module that runs on the embedded Ti
DSP processor, managing the resources on the hardware.
The major functions implemented in the RTR engine are
channelization, synchronization and timing recovery, chan-
nel estimation and equalization, demodulation, despreading,
and de-interleaving. These are typically computation inten-
sive processing suitable for parallel hardware implementa-
tion. Corresponding reverse processing is done in the uplink.
The rest of the functions such as source/channel cod-
ing, scrambling, pilot insertion, and so forth and their cor-
responding reverse in the receiver path are performed using
the processing resources of DSP. Again the exact functions
performed by these engines depend on the system applica-
tions that are congured.
The rmware engine, which is implemented in the DSP,
represents the run-time recongurable rmware subsystem.
This typically implements some of the Layer-1 control pro-
cedures like scheduling and Layer-1 management and MAC
processing (for example in IEEE 802.11a). These functions
also depend on the waveform application being recongured
to as the processing requirements for dierent CP-based sys-
tems are dierent.
An SDR executive provides the platform services needed
to recongure and activate waveform applications, such as
802.11a for OFDM system. It is independent of any other ap-
plication running on the platform. The SDR executive con-
sists of the RTOS, a basic set of resource and conguration
management processes and communication and data path
management functions. The executive interfaces with the
330 EURASIP Journal on Wireless Communications and Networking
system and conguration manager (SCM) on GPP through
the SDR Modem driver and waveform driver executed on the
GPP. The SDRexecutive incorporates a command interpreter
to receive and process the commands from the drivers. It also
manages the communication channels between the GPP and
SDR Modem.
To enable instantiation of dierent waveform applica-
tions on the same hardware, the SDR executive supports dy-
namic loading of COFF modules and reconguration of FP-
GAs. The SDR executive will be stored on a nonvolatile stor-
age and will be loaded at boot time. This module receives
rmware and bit map information from the system and con-
guration manager and congures on the platform subsys-
tems.
System and conguration manager (SCM) is the central
management function of the architecture. The SCM uses the
GPP resources to perform high-level functions like system
management, conguration/reconguration management of
the platform, conguration storage, user interface, and so
forth. This module performs its functions in conjunction
with other platformfunctions, specically the SDRexecutive,
kernel/OS, user interface, and conguration database.
The major functions performed by the SCM are
(a) SDR Modem control (reset, platform loading, initial-
ization, etc.);
(b) waveform application database management (add/
delete waveform applications);
(c) waveform application management (loading, activa-
tion, deactivation, unloading);
(d) diagnostic support (diagnostic commands, error and
message logging, etc.).
The SCM uses the SDR Modem driver to communicate
with the SDR executive to download platform components
as needed.
SDR Modem driver, a kernel driver, interfaces the SDR
modem subsystem platform to the system and conguration
manager process and acts as the root device driver for the
SDR Modem hardware. The driver will be loaded when the
SDR Modem device is detected and basically interfaces with
the SDR executive to pass command and response between
the SCM and SDR executive. This also provides facility to
transfer the COFF les and FPGA bit maps to congure the
DSP and FPGA, respectively.
Kernel/OS block in Figure 4 is the operating-system ker-
nel on the GPP. The operating system used is a general-
purpose OS, LINUX, with real-time extensions, or a dedi-
cated real-time OS like LynxOS could also be used. The real-
time extensions are necessary as part of the waveform appli-
cations, depending on the application, may be implemented
inside the kernel.
Waveform drivern, are function-specic drivers in the
host OS. These are shown in dotted outline to indicate that
they are not always present. A driver will be loaded into
the host memory only when corresponding waveform ap-
plication is being instantiated; and it is unloaded when the
application is unloaded. These are technology specic and
some of them may be modied versions of standard drivers.
For example in the case of OFDM, this will be a modi-
ed, dynamically loadable version of the standard wireless
LAN 802.11a drivers. As for CDMA application, this may
be a specic driver developed for the purpose, incorporat-
ing the Layer 2 and higher protocols interfacing with Layer-
1 functions inside the SDR Modem. These driver modules
are loaded and unloaded by the SCM, whenever the cor-
responding waveform application is created and destroyed.
Upper-layer protocols and application software will make use
of the communication capabilities provided by the waveform
drivers and interface with them.
A conguration database holds all the conguration and
status information including all the system images for each
of the waveformapplications that are supported. At least part
of it will be nonvolatile (e.g., disk based). The SCM manages
the conguration database, it maintains the current status of
the system, and will estimate, along with the SDR executive,
the resource availability for the selected application. If suf-
cient resources are available, the waveform application will
be installed and feedback will be given to the user. If resources
are not available, it will abort the loading and free up all the
resources reserved for this waveform application. The load-
ing process includes, in addition to loading of SDR Modem
rmware modules and FPGA bit maps, any host driver mod-
ules. A loaded waveform application can be activated only
upon explicit activation by the SCM. This involves activa-
tion of RF subsystem, hardware and rmware components
(loaded and congured during the loading process), as well
as loading of the driver on the host side. The SCM imple-
ments error logging and diagnostic command support for
the development and testing support. This includes ability to
log module-wise diagnostic messages into nonvolatile stor-
age.
Finally, the user interface exposes the user features pro-
vided by the SCM to the user.
6. CONCLUSIONS
In this paper, we have proposed a broadband mobile
transceiver and a hardware architecture which can be con-
gured to any cyclic-prex (CP) -based system. We have
identied the reusable common blocks and studied the re-
ceiver complexity of the equalization block for dierent sys-
tems. It is found that after the despreading operation, MC-
DS-CDMA and CP-DS-CDMA have the same equalization
blocks as OFDM and SCCP systems, respectively, showing
that both hardware and software sharing is possible for those
systems. We also noticed that although MC-CDMA and CP-
CDMA require only one block to perform signal detection,
the receiver complexity is higher than the other two CDMA
systems. However, MC-DS-CDMA and CP-DS-CDMA rely
on the assumption that the wireless channel is static within
the G blocks. For fast-fading environments, MC-CDMA and
CP-CDMA may still be the better choice to compromise the
fast fading and complexity issues. Further, functionality of
the dierent functions in the proposed hardware architecture
is elaborated. It is seen that though dierent functions can be
Recongurable Signal Processing for Wireless Communications 331
realized on a recongurable hardware, the major challenge is
to have an ecient system conguration and management
function which will initiate and control the reconguration
as per waveform application requirements.
REFERENCES
[1] J. Mitola III, The software radio architecture, IEEE Com-
mun. Mag., vol. 33, no. 5, pp. 2638, 1995.
[2] J. Mitola III and G. Q. Maguire Jr., Cognitive radio: making
software radios more personal, IEEE Pers. Commun., vol. 6,
no. 4, pp. 1318, 1999.
[3] L. J. Cimini Jr., Analysis and simulation of a digital mobile
channel using orthogonal frequency division multiplexing,
IEEE Trans. Commun., vol. 33, no. 7, pp. 665675, 1985.
[4] Y.-C. Liang, A. S. Madhukumar, and F. Chin, Single carrier
cyclic prex based CDMA systems, Tech. Rep., Centre for
Wireless Communications (Now Institute for InfoComm Re-
search), Singapore, Singapore, 2001.
[5] Y.-C. Liang, Block-Iterative GDFE (BI-GDFE) for CP-
CDMA, in Proc. 61st IEEE Semiannual Vehicular Technology
Conference (VTC 05), Stockholm, Sweden, MayJune 2005.
[6] K. L. Baum, T. A. Thomas, F. W. Vook, and V. Nangia,
Cyclic-prex CDMA: an improved transmission method for
broadband DS-CDMA cellular systems, in Proc. IEEE Wire-
less Communications and Networking Conference (WCNC 02),
vol. 1, pp. 183188, Orlando, Fla, USA, March 2002.
[7] S. Zhou, P. Xia, G. Leus, and G. B Giannakis, Chip-
interleaved block-spread CDMA versus DS-CDMA for cel-
lular downlink: a comparative study, IEEE Transactions on
Wireless Communications, vol. 3, no. 1, pp. 176190, 2004.
[8] S. Hara and R. Prasad, Overview of multicarrier CDMA,
IEEE Commun. Mag., vol. 35, no. 12, pp. 126133, 1997.
[9] X. Cai, S. Zhou, and G. B. Giannakis, Group-orthogonal
multicarrier CDMA, IEEE Trans. Commun., vol. 52, no. 1,
pp. 9099, 2004.
[10] A. Jamalipour, T. Wada, and T. Yamazato, Atutorial on multi-
ple access technologies for beyond 3G mobile networks, IEEE
Commun. Mag., vol. 43, no. 2, pp. 110117, 2005.
[11] K.-C. Chen and S.-T. Wu, A programmable architecture for
OFDM-CDMA, IEEE Commun. Mag., vol. 37, no. 11, pp. 76
82, 1999.
[12] S.-T. Wu and K.-C. Chen, Programmable multiuser synchro-
nization for OFDM-CDMA, in Proc. 53rd IEEE Vehicular
Technology Conference (VTC 01), vol. 2, pp. 830834, Rhodes,
Greece, May 2001.
[13] J. Helmschmidt, E. Schuler, S. Rao, S. Rossi, S. di Matteo, and
R. Bonitz, Recongurable signal processing in wireless termi-
nals [mobile applications], in Proc. Design, Automation and
Test in Europe Conference and Exhibition (DATE 03), pp. 244
249, Munich, Germany, March 2003.
[14] J. R. Moorman, Implementation of a 3G W-CDMA software
radio, in Proc. IEEE International Conference on Communi-
cations (ICC 03), vol. 4, pp. 24942499, Anchorage, Alaska,
USA, May 2003.
[15] I. P. Seskar and N. B. Mandayam, A software radio architec-
ture for linear multiuser detection, IEEE J. Select. Areas Com-
mun., vol. 17, no. 5, pp. 814823, 1999.
[16] J. Palicot and C. Roland, FFT: a basic function for a recon-
gurable receiver, in Proc. 10th International Conference on
Telecommunications (ICT 03), vol. 1, pp. 898902, Papeete,
Tahiti, FebruaryMarch 2003.
[17] P. Burns and M. C. Reed, A complexity cost function for the
signal processing in a WCDMA base station for dimension-
ing of a software dened radio, in Proc. 8th IEEE Interna-
tional Symposium on Spread Spectrum Techniques and Appli-
cations (ISSSTA 04), pp. 854858, Sydney, Australia, August
September 2004.
[18] A. Gupta, A. Forenza, and R. W. Heath Jr., Rapid MIMO-
OFDM software dened radio system prototyping, in Proc.
IEEE Workshop on Signal Processing Systems (SIPS 04), pp.
182187, Austin, Tex, USA, October 2004.
[19] S. Chuprun, J. Kleider, and C. Bergstrom, Emerging software
dened radio architectures supporting wireless high data rate
OFDM, in Proc. IEEE Radio and Wireless Conference (RAW-
CON 99), pp. 117120, Denver, Colo, USA, August 1999.
[20] D. Falconer, S. L. Ariyavisitakul, A. Benyamin-Seeyar, and
B. Eidson, Frequency domain equalization for single-carrier
broadband wireless systems, IEEE Commun. Mag., vol. 40,
no. 4, pp. 5866, 2002.
[21] N. Maeda, Y. Kishiyama, H. Atarashi, and M. Sawahashi,
Variable spreading factor-OFCDM with two dimensional
spreading that prioritizes time domain spreading for forward
link broadband wireless access, in Proc. 57th IEEE Semian-
nual Vehicular Technology Conference (VTC 03), vol. 1, pp.
127132, Jeju, Korea, April 2003.
Ying-Chang Liang received the Ph.D. de-
gree in electrical engineering from Jilin
University, China, in 1993. He is now a
Lead Scientist in the Institute for Info-
comm Research (I2R), Singapore. He also
holds Adjunct Associate Professor positions
in Nanyang Technological University and
the National University of Singapore, Sin-
gapore. From December 2002 to December
2003, he was a Visiting Scholar in the De-
partment of Electrical Engineering, Stanford University. From Au-
gust 1997 to November 2001, he was a Senior Member of techni-
cal sta in the Centre for Wireless Communications, Singapore.
His research interests include space-time wireless communications,
recongurable signal processing and cognitive radio, smart an-
tennas, ultra-wideband communications, and capacity-achieving
schemes for MIMO fading channels. He received the Best Paper
Award from the 50th IEEE Vehicular Technology Conference. He
was also the corecipient of the 1997 National Natural Science Award
from China. He has been an Associate Editor for IEEE Transactions
on Wireless Communications since 2002. He was the Publication
Chair for 2001 IEEE Workshop on Statistical Signal Processing and
is a Senior Member of IEEE.
SayedNaveen is an Associate Scientist in the
Communications & Devices Division, In-
stitute for Infocomm Research (I2R), Sin-
gapore. He received his B.E. degree in
electronics and communication engineer-
ing from the University of Madras, India,
in 1992, and the M.Eng. degree in commu-
nication systems from the Regional Engi-
neering College, Trichy, India, in 1995. After
the M.S. degree, he joined the Central Re-
search Laboratory, Bharat Electronics Limited, Bangalore, India, as
a member of research sta, where he was working on power ampli-
er linearization and wireless transceivers for defense communi-
cations. In 1999, he joined the WCDMA base station development
team, Sanyo LSI Technology India, Bangalore, India, as a Senior En-
gineer of signal processing. In 2000, he joined I2R (formerly Cen-
tre for Wireless Communications). His research interests include
recongurable signal processing and architectures for broadband
wireless communications and software-dened radio.
332 EURASIP Journal on Wireless Communications and Networking
SantoshK. Pilakkat received his B.Tech. de-
gree in electronics from the Regional Engi-
neering College, Calicut, India, in 1987. He
is currently a Principal Research Engineer
and Project Manager at the Institute for In-
focomm Research, A-STAR, Singapore. He
has been involved in research and develop-
ment of wireline and wireless communica-
tion systems in various capacities through
out his career. His current areas of interest
include embedded real-time systems, software-dened radios, and
recongurable system architectures.
Ashok K. Marath is the Manager of the
Embedded Systems Department, Institute
for Infocomm Research (I2R), Singapore.
He graduated from the National Institute
of Technology, Calicut, in 1986, and com-
pleted his M.S. degree in the National Uni-
versity of Singapore in 1999. Starting his ca-
reer as an RF design engineer with SAMEER
(Society for Applied Microwave Electronics
Engineering &Research), Bombay, India, he
has more than 15 years of research and development experience in
wireless communications in RF design and physical layer process-
ing. He has been active in software radio research for the past 3
years and he is currently involved in a software-radio-based multi-
mode terminal development project. Prior to this, he was involved
in projects dealing with RF and system-level issues in various wire-
less systems such as WATM, RF ID, PHS, and WCDMA. Prior to
joining I2R, he was a design engineer with PCI Ltd, Singapore, in-
volved in the design of GSM hand phone. From 1987 to 1993 he
was with transmission R&D of Indian Telephone Industries, Ban-
galore, designing RF and baseband subsystems for digital transmis-
sion equipments. He has got extensive experience in system-level
issues of various communication systems. He holds 2 patents and
has got few pending applications.
EURASIP Journal on Wireless Communications and Networking 2005:3, 333342
c 2005 Arnd-Ragnar Rhiemeier
Modular Software-Dened Radio
Arnd-Ragnar Rhiemeier
Institut f ur Nachrichtentechnik, Universit at Karlsruhe (TH), 76128 Karlsruhe, Germany
Email: rhiemeier@int.uni-karlsruhe.de
Received 1 October 2004; Revised 14 February 2005
In view of the technical and commercial boundary conditions for software-dened radio (SDR), it is suggestive to reconsider the
concept anew from an unconventional point of view. The organizational principles of signal processing (rather than the signal
processing algorithms themselves) are the main focus of this work on modular software-dened radio. Modularity and exibility
are just two key characteristics of the SDR environment which extend smoothly into the modeling of hardware and software. In
particular, the proposed model of signal processing software includes irregular, connected, directed, acyclic graphs with random
node weights and random edges. Several approaches for mapping such software to a given hardware are discussed. Taking into
account previous ndings as well as new results from system simulations presented here, the paper nally concludes with the
utility of pipelining as a general design guideline for modular software-dened radio.
Keywords and phrases: exible digital baseband signal processing, rmware support for reconguration, computing resource
allocation, multiprocessing, modeling of SDR software.
1. INTRODUCTION
Software-dened and hardware recongurable radio systems
have attracted more and more attention recently because
they are expected to be among the key techniques to serv-
ing future wireless communication market needs. In con-
trast to the strong convergence tendency in wired networks,
a growing number of standards and communication modes
can be observed in wireless access networks. Presumably, this
trend will prevail, eventually due to the natural diversity in
service requirements and radio environments. The better the
match between the physical channel (the properties of which
are determined in part by the user mobility) and the signal
processing in the transceiver, the easier to achieve the opti-
mal quality of service (QoS) on the physical layer. Further-
more, the ongoing introduction of UMTS in Europe shows
that diversity in standards is not only a technical challenge; if
market response falls short of business expectations (based
on a particular communication standard), or if users de-
mand shifts to a dierent wireless access technology (and
thus to a dierent sort of underlying signal processing), it
would be benecial for any manufacturing company to be
able to respond quickly to such situations. Software-dened
and recongurable radio systems have the potential to allow
short time-to-market product designs under these commer-
cial conditions.
This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
The present paper puts an emphasis on the physical-
layer signal processing because its capabilities represent the
fundamental limits for higher layers in delivering their ser-
vices. Therefore, mastering all aspects of physical-layer sig-
nal processing in software-dened and hardware recong-
urable radios is vital for delivering best end-to-end service
to the end user, by means of a single communication de-
vice. Modular software-dened radio (Mod-SDR) strives for
casting light on one important aspect which has been largely
neglected hitherto: the design guidelines which govern the
coordinated interplay of signal processing software modules
embedded in logical structures of some arbitrary wireless
communication standards.
1.1. Related work
A great number of important contributions on software-
dened radio [1, 2, 3] and recongurability [4, 5, 6, 7, 8] can
be found in the literature. However, many authors narrow
down their research interest to one particular aspect of the
signal processing chain: sample rate adaptation [9, 10, 11],
RF front-end design [12, 13, 14, 15, 16], A/D conversion [17,
18], or channel coding [19, 20, 21], just to name a few exam-
ples. Notably, work related to signal processing in the digital
baseband is centered around algorithms [22, 23]. However,
structural properties of signal processing software (including
an abstract way for representation) and the principles of or-
ganizing the execution of multiple algorithms in a distributed
multiprocessing hardware system have not been studied in-
tensively in the context of software radio. One major contri-
bution of Mitola [24] attempts to reexamine software radio
334 EURASIP Journal on Wireless Communications and Networking
from a truly unorthodox point of view, but his ndings still
pertain to algorithms and eventually do not reach beyond the
Turings theory of computing. Nevertheless, his contribution
hints at the fact that SDR requires an understanding which
is radically dierent from classical communications and its
BER curves.
1.2. Motivation for modular software-dened radio
The motivation for introducing a novel view of exible radio
system design is twofold. First, the fact that a radio terminal
accomplishes its signal processing by software or recong-
urable hardware rather than by dedicated hardware does not
render the rules of real-time computing obsolete, but so far
this aspect has not been considered systematically in the con-
text of software-dened radio. Second, any BER produced by
some software implementation has to be at least as good as
the BER of its equivalent ASICimplementation. Hence, it ap-
pears to be questionable to present BER curves as a measure
of quality for the design of a software-dened radio.
The goal of modular software-dened radio is to estab-
lish general guidelines for designing and operating exible
signal processing systems. In order to make these guidelines
general, they need to be independent of particular commu-
nication standards, independent of algorithmic implementa-
tion details, and independent of technological advances in
microelectronics. The model for SDR software will reect
these important aspects.
1.3. Compatibility with the real world
In the same way as the ideal software radio concept [1, 25]
has evolved into some compromise approaches usually sum-
marized under the termof software-dened radio, the advent
of recongurable, distributed signal processing hardware in
radio devices [26, 27] can be seen as another step in this evo-
lution towards implementations which are both technologi-
cally feasible and economically attractive.
Critics generally claim that SDRs will always be noto-
riously power-inecient and inherently overpriced, hence
never prove competitive against carefully designed ASICs.
This may be true indeed if the exibility of SDR is uncon-
ditionally passed on to the end user in the form of future
upgradability. Therefore, it is more reasonable to predict
that the exibility of SDR is likely to stay under the imme-
diate control of manufacturers, all the more so to support a
sustainable business model. Actually, time-to-market is the
master argument in favor of software-dened radio tech-
niques. The present paper shares this view suggesting to per-
ceive software-dened and recongurable radios as wireless
communications embedded real-time systems which are tun-
able to end-user needs or network operator needs, but not
more. Modular software-dened radio provides those em-
bedded systems with design guidelines and the core of a QoS
manager for the physical layer.
2. MODELINGOF HARDWARE ANDSOFTWARE
One fundamental assumption is that SDR will become real-
ity soonest in the form of some distributed multiprocessing
hardware architecture, providing sucient computing power
at a much lower electric power consumption than that of a
single comparable general-purpose processor. Furthermore,
all hardware resources of a Mod-SDR device such as pro-
cessors, buses, memories, and interfaces are administered by
a nonpreemptive operating system. Supplied by the termi-
nal manufacturer, this piece of rmware is protected from
any direct manipulation on the part of the end user. How-
ever, by means of a physical layer API, both user applications
and the network may address transmission mode requests to
the QoS manager which belongs to the core framework ser-
vices/applications running immediately on top of the oper-
ating system (see SCA, [28, Figure 2-1]). A request will cer-
tainly include an abstract representation of the signal pro-
cessing software which needs to be executed to realize the
transmission mode. modular software-dened radio makes
use of directed acyclic graphs to represent signal processing
software.
The main task of the QoS manager consists of the map-
ping of signal processing software to the available hardware
while respecting the real-time requirements which are de-
ned by the air interface of the requested communication
standard [29]. In return for requests, the QoS manager ac-
cepts or rejects transmission modes, based on the availability
of resources and on a decision process which includes parti-
tioning of the graph and scheduling of all software modules.
Partitioning is the process of uniquely assigning each module
to a processor, while scheduling means determining individ-
ual trigger instants for each module.
Throughout this work, a software module is dened to be
the smallest entity of executable machine code, which can-
not be preempted by the operating system. Software mod-
ules are self-contained, independent entities, and there is no
data exchange across module boundaries other than input
data from predecessor modules and output data to successor
modules. In principle, any module may be connected to any
other module, the only requirement being data type compat-
ibility in between all modules.
2.1. Details of the software model
For the Mod-SDR software model to be independent of pro-
cessor types (DSP, FPGA, ASSP, specic coprocessor, and
ASIC) and of technological advances in microelectronics, the
processing runtime of software modules is taken into ac-
count as the main behavioral attribute of signal processing
software. At the same time, algorithmic details are abstracted
away in this manner. Given the framework of directed acyclic
graphs, the nodes of a graph represent software modules car-
rying some signal processing runtime as the node weight.
Due to the vast variety of unpredictable inuences on the
processing runtime of software modules in an SDR envi-
ronment [29], node weights are subject to random variation
within a stochastic linear resource runtime model [30]. The
basic assumption of linearity originates from the idea that
the more processing runtime is needed, the more data is to
be produced at the output of a software module:
p
m
= c r
m
, nodes m : 1 m M, (1)
Modular Software-Dened Radio 335
where p
m
is the processing runtime of some node m and r
m
is ms output memory resource demand. Formally, R
+
is the constant of proportionality translating a memory de-
mand into a runtime, hence its unit [] = (bit/s)
1
. It can
be interpreted as the absolute speed of a processor when
executing the signal processing machine code behind node
m. The constant factor c is unitless, and its value is drawn
from a random experiment, for each m anew. The charac-
teristics of the SDR environment as well as all shortcomings
of a strictly linear resource runtime model are modeled by
the real-valued random variable C. A complete description
of this variable is given by its probability density function
(pdf) f
C
(c). Throughout this paper, all realizations c stem
from independent identically distributed random variables
C for all nodes m. The pdf employed in this work has a
rectangularly windowed Gaussian shape with the parameters

C
= 1.0,
C,e
= E{(C
C
)
2
} = 0.25, and windowed in-
terval [0.5; 1.5]. The choice of a Gaussian is certainly arbi-
trary, but there are strong hints that the actual shape of the
pdf has much less inuence on the performance of the QoS
manager than the eective relative spread
P,e
/
P,e
of pro-
cessing runtimes [31].
The structural properties of Mod-SDR software are cap-
tured in directed edges m, n between a pair of nodes m and
n of a graph. In order to be independent of any particular
communication standard, those graphs are not only random
in their node runtimes, but also in their directed edges.
The random graph generator which is used for computer
simulations produces irregular, connected, directed, acyclic
graphs with a xed number of nodes M = 40. The gener-
ation process starts out from a chain of two nodes indexed
m = 1 (referred to as the source node) and m = 2 (referred
to as the target node). Subsequently, nodes are added in an
iterative process by placing them somewhere relative to the
existing graph. For every node to be placed, one predeces-
sor and one successor are selected at random and with equal
probability from the existing nodes of the graph. Should a
new node shortcut the edge between adjacent existing nodes,
then that edge is removed with probability of 0.5. Connect-
edness of the graph is enforced in a simple way by exempting
m = 1 from the node selection procedure, whereas the prop-
erty of noncyclicity needs to be veried throughout the en-
tire graph generation process. Finally, some additional graph
properties, which are related to the data input and output
behavior of nodes, need to be determined. First, the output-
to-input data ratio of all nodes be 1.0. Second, all edges out-
going from any demultiplexing node convey the full output
data volume. Third, all edges incoming to any multiplexing
node convey only one Kth of the predecessors output data
volumes, where K is the number of incoming edges. These
rules make sure that edge weights remain in the same order
of magnitude throughout the entire graph. Figure 1 shows
one realization of a random graph as an example.
2.2. Details of the hardware model
A symmetric multiprocessor architecture (see Figure 2)
serves as the model of a modular, easy-to-extend multi-
processing hardware system for Mod-SDR. The architecture
8
10 25
20
32
40
3
26
19
35
37
16
1
5
34
24 14
30
12
15
33
17
2
29
39
11 21 38
4
18
7
13
9
28
22
31
23
27 36
6
Figure 1: Irregular, connected, directed, acyclic graph.
Shared
memory
i/o
memory
M
1
P
1
M
2
P
2
B B
B

Figure 2: Symmetric multiprocessor architecture.
generally includes L N identical processors P
l
with associ-
ated (distributed) memory M
l
, a shared memory and an in-
put/output (i/o) memory for interfacing of the physical layer
signal processing subsystem to the outside world, that is, to
other processing subsystems or to the analog front end of the
Mod-SDR transceiver.
All of these hardware resources are connected by B N
separate data buses. It is assumed that processors are actively
involved in bus transfers, that is, no useful signal processing
code of a module can be executed by a processor while this
processor is exchanging data with the shared memory or with
the i/o memory via some bus. This is certainly a conservative
assumption, which can also be interpreted as a worst case on
one hand. On the other hand, however, it is very unlikely
that simultaneous signal processing and bus communica-
tions can be achieved in the general case of Mod-SDRgraphs.
Even if a processor architecture supports communication la-
tency hiding behind useful core computations, those mech-
anisms would require coordination on the code program-
ing level across module boundaries. However, this intermod-
ule control ow contradicts the Mod-SDR self-containment
paradigm, where any module may indeed be linked to any
other module (logically, by the directed edge of a graph), in
order to accomplish a useful signal processing task, but with-
out mutual knowledge of their respective machine code in-
ternals.
Naturally, bus access is exclusive. Conicting access tim-
ings on the part of processors need to be arbitrated by the
scheduler, which is built into the QoS manager. The bus
speed is given relative to the signal processing speed of the
processors, in the form of a relative bus speed R
+
. The
unitless factor describes how much faster a processor can
transfer an amount of data over the bus rather than produce
336 EURASIP Journal on Wireless Communications and Networking
the same amount of data as the output of a module (by reg-
ular signal processing). Large values of the relative bus speed
( 1) represent fast buses, whereas < 1 represents slow
buses. Partitioning always entails a cut aecting a certain sub-
set of edges in the graph. The logical data owalong cut edges
then translates into an asynchronous, physical data ow be-
tween processors via the shared memory. How this is orga-
nized using pairs of bus transfer nodes is described in detail
in [32]. The basic idea is that bus transfers require runtime in
the same way as regular signal processing nodes, but paired
for intermediate results to be written to and read from the
shared memory. Therefore, the resulting runtime model for
bus transfer nodes is similar to (1), but includes in the place
of c:
p
m,n
=
1
r
m
, edges m, n aected by the cut.
(2)
These runtimes do not depend on the SDR environment,
but on the deterministic capabilities of the Mod-SDR device.
Their values p
m,n
reappear later as edge weights that repre-
sent potential link cost for all edges m, n. Actual costs are
only incurred by those edges aected by the partitioning cut.
The focus of the present work is on fundamental design
and operating principles in Mod-SDR systems. Before tack-
ling the general case of L N processors, the case L = 2
should be well understood rst. With L = 2 processors, only
B 2 is reasonable. In this paper, the case of B = 1 bus is
studied.
2.3. Proposed measure of quality
It is the declared goal of Mod-SDR to go beyond designing
for some particular set of standards or transmission modes.
Therefore, concrete real-time requirements such as deadline
periods (which are clearly determined by the air interfaces of
any standard) are missing. Instead, the more general require-
ment for maximum speedup of a multiprocessing hardware
system is considered. The speedup s R
+
is dened as the
factor by which the multiprocessor Mod-SDR implementa-
tion terminates faster than the same software implementa-
tion on a single processor running at the same speed of .
The advantage of speedup is evident; it is a relative measure
of quality for the design of (and the way of operating) a mul-
tiprocessor computing system. The actual value of is not of
importance because it cancels out in the speedup s.
3. MAPPINGAPPROACHES
Although the Mod-SDR software model includes random
graphs, statistics in this approach should not be confused
with any statistical structure of computational demand
[24]. The way of building and using a population of SDR ter-
minals involves a randomprocess, but transmission mode re-
quests are realizations of this random process. Therefore, the
QoS manager has to deal with realizations of random graphs.
Per request, the total number of nodes and all associated
node weights are arbitrary, but xed. Likewise, the logical
structures captured in the set of directed edges remain xed
over their entire utilization period. Potentially, QoS parame-
ters may be negotiated during connection setup, but once a
transmission mode has been accepted, the (static) scheduling
situation is deterministic, and so is the demand for comput-
ing power [29].
All SDR design techniques related to mapping structured
signal processing software to hardware may be roughly classi-
ed by the number of graph copies involved in the partition-
ing and scheduling process; some approaches use but a sin-
gle copy, while others imply multiple identical copies of the
graph. In principle, techniques of both classes are applicable
to signal processing for circuit-switched services as well as
for packet-switched services. However, ordinary protocol re-
quirements in packet-oriented networks (e.g., stop-and-wait
ARQ) may oftentimes force the QoS manager to operate on
a single copy of the given graph. As a matter of principle,
single-copy variants are less demanding in terms of program
memory and data memory.
The following partitioning approaches are employed in
Mod-SDR system simulations to be discussed in a later sec-
tion of the this paper.
(i) Implicit partitioning by the Hu algorithm [29]
eventually, a scheduling algorithm.
(ii) Kernighan-Lin (KL) algorithm [30]a method for
local search of the design space.
(iii) Spectral partitioning [33]an application of alge-
braic graph theory.
These approaches primarily operate on a single copy of
the graph which is given by some transmission mode request.
The originally implied scheduling idea is that radio signals
are processed on a frame-by-frame basis, only one frame per
real-time period, and all computations relate to a single radio
frame only. An accurate pseudocode of the static scheduler is
given in [34].
However, pipelined scheduling [33, 35] in combination
with these partitioning approaches revokes the memory ad-
vantage of a single graph copy, because pipelining involves
the processing of radio signals related to several dierent
radio frames within the time of one real-time period. The
number of dierent radio frames involved in these computa-
tions is called the depth of the pipeline. Indeed, the program
memory remains unaected by pipelining, but not the data
memory; multiple intermediate computation results from
several dierent frames must be kept in memory, which re-
sults in a much higher demand than before.
Aradical alternative to the above partitioning approaches
is graph duplication; one complete copy of the graph is as-
signed to the rst processor, another copy is assigned to the
second processor. No partitioning algorithm is needed, the
workload on both processors is perfectly balanced by con-
struction, and bus access is reduced to the necessary i/o data
transfers. The associated scheduling scheme has been named
graph duplication pipelining (GDP) [35]. Despite its obvi-
ous advantages, GDP suers from both increased data mem-
ory and programs memory demand due to the duplication
process. An alternative pipelining approach that returns to
operating on a single copy of the graph is half-frame pipelin-
ing (HFP) [36, 37]. HFP is primarily based on a scheduling
Modular Software-Dened Radio 337
10
1
10
0
10
1
10
2
Relative bus speed
0.6
0.7
0.8
0.9
1
F
r
a
c
t
i
o
n
o
f
m
a
x
s
p
e
e
d
u
p
Figure 3: Implicit Hu partitioning, no pipelining.
idea where the QoS manager supports the following kind
of time-interleaved frame processing; while one processor is
still busy processing the second half of some frame indexed
i, the other processor already starts processing its successor
frame indexed (i + 1). In contrast to GDP, this approach re-
quires load-balanced partitioning of the graph under one ad-
ditional side condition, namely maintaining a vertical cut rel-
ative to the source node and the target node. Both the KL al-
gorithm and a modied variant of spectral partitioning have
been tested for this purpose. It can be shown [38] that the
KL algorithm is a simple and ecient partitioning support
for HFP.
In principle, all of these mapping approaches are eligible
for application by the QoS manager of a Mod-SDR, inde-
pendent of the service type. It remains to be shown which
approach achieves a high probability of good speedup in the
highly unpredictable SDR environment.
4. SYSTEMSIMULATIONS OF MOD-SDR
The three partitioning approaches have been discussed ex-
tensively elsewhere, but the comparison of performance was
merely based on one particular sample graph. The node
weights did stem from random experiments, but the struc-
ture of the graph remained xed throughout all simula-
tion runs. Here, in contrast, the random graph model of
Section 2.1 (also including random edges) is applied to im-
prove the expressiveness of Mod-SDR system simulations.
Those new results are discussed in the following.
All gures show speedup measurements (dots) as a func-
tion of the relative bus speed . These results are given as
a fraction of the maximum speedup s which is theoreti-
cally achievable [33] by perfect parallelization. Hence, on
one hand, a fractional value of 1.0 represents the upper per-
formance limit for any Mod-SDR realization. On the other
hand, the fractional value of 0.5 represents a reasonable lower
limit, because belows = 0.5, the two-processor systemwould
eectively work slower than a single-processor system, thus
rendering any distributed processing approach meaningless
in principle. At a sample size of 2000 realizations per , the
observed measurement range is often so densely populated
by dots that the latter amalgamate into vertical lines. There-
fore, in addition to the individual speedup measurements,
the contour lines of the 5%, the 50% (median), and the 95%
quantile are estimated and overlayed to the gures. A con-
tour line represents the maximum speedup achieved by the
given quantile of Mod-SDR realizations (95%: topmost, 5%:
bottommost, median: in between) as a function of .
4.1. Circuit-switched services
Figure 3 shows the speedup results for implicit Hu partition-
ing and nonpipelined scheduling. Obviously, the faster the
bus, the better the speedup. Since Hus algorithm is eventu-
ally a pure scheduling algorithm, it does not take into ac-
count any link cost while partitioning graphs implicitly. A
naive working assumption could be that speedup approaches
the limit of 1.0, if the bus is somewhat fast enough, be-
cause link cost approaches zero as . Figure 3 dis-
proves this assumption. The behavior observed over the en-
tire range can be explained in part by the occurrence of
PHYSICAL WAIT IDLE and LOGICAL WAIT IDLE condi-
tions [30]. The former arises after the arbitration of concur-
rent bus access requests, whereas the latter originates from
logical interdependencies of nodes in the graph; although the
bus is fully accessible, one processor is forced to remain idle
waiting for some intermediate results to be produced by the
other processor.
Both conditions cause idle times in the processors, and
thus reduce speedup. While concurrent bus access requests
become more and more unlikely as bus speed increases, the
LOGICAL WAIT IDLE condition continues to prevail in all
schedules independent of . Another reason for speedup loss
against the limit can be identied in a special property of Hus
algorithm; the approach strictly aims at maximum paral-
lelization. However, the random graph model produces real-
izations with a degree of inherent parallelism

d, 0.3

d 0.8
[36]. As a consequence, if the graph cannot be parallelized
due to its given structural properties (small

d value below

d = 0.5), the Hu algorithm systematically fails to generate


high speedup. Load imbalance between the two processors
(equal to the dierence of aggregate runtimes between the
two partitions) is the resulting eect of this failure.
Pipelining eliminates LOGICAL WAIT IDLE conditions
by deliberately constructing a dense schedule in a rst step.
All resulting anticausal data dependencies between the parti-
tions are resolved in a second step by rescheduling bus trans-
fers of intermediate results across the boundaries of real-
time processing periods. In this way, a radio frame pipeline
of some depth is created (cf. Section 3). Figure 4 shows the
speedup results for implicit Hu partitioning under pipelin-
ing. Indeed, the overall speedup behavior has improved; all
contour lines indicate higher speedup for the same quantile
of realizations, and the speedup spread for fast buses is re-
duced. Nevertheless, implicit partitioning pursuant to Hus
algorithm continues to suer from its systematic drawbacks
mentioned above: strong dependency on graph structure and
complete insensitivity to link cost.
338 EURASIP Journal on Wireless Communications and Networking
10
1
10
0
10
1
10
2
Relative bus speed
0.6
0.7
0.8
0.9
1
F
r
a
c
t
i
o
n
o
f
m
a
x
s
p
e
e
d
u
p
Figure 4: Implicit Hu partitioning, pipelining.
10
1
10
0
10
1
10
2
Relative bus speed
0.6
0.7
0.8
0.9
1
F
r
a
c
t
i
o
n
o
f
m
a
x
s
p
e
e
d
u
p
Figure 5: KL partitioning, no pipelining.
The KL algorithm has been introduced to Mod-SDR [30]
as a remedy for this situation. Figure 5 shows its performance
without pipelining. Astonishingly enough at a rst glance,
although the approach explicitly considers link cost (and it
is even capable of trading load balance for link cost), the
KL algorithm shows no systematic superiority compared to
Hus algorithm under these operating conditions. Speedup
degradation in the low range is a bit more graceful than
in Figure 3, but at high bus speeds, the KL algorithm is eas-
ily outperformed by an approach as simple as Hus. This can
be explained by the fact that the partitioning approach by
Kernighan and Lin indeed considers link cost, but tacitly as-
sumes that bus transfers are nonconicting at all times. Fur-
thermore, its partitioning cut has an arbitrary orientation
relative to the source node and the target node. As a con-
sequence, a large number of LOGICAL WAIT IDLE condi-
tions still occurs causing a large spread over the [0.7; 0.9]
range of speedup values.
As before, processor idle times associated with LOGI-
CAL WAIT IDLE conditions can be completely eliminated
10
1
10
0
10
1
10
2
Relative bus speed
0.6
0.7
0.8
0.9
1
F
r
a
c
t
i
o
n
o
f
m
a
x
s
p
e
e
d
u
p
Figure 6: KL partitioning, pipelining.
by pipelining. Figure 6 shows the resulting performance
of the KL algorithm. The contour lines reveal the highest
speedup and the lowest speedup spread observed so far. Nat-
urally, the speedup increases as the relative bus speed in-
creases, because link cost tends to be reduced and bus con-
icts become less and less likely. Nevertheless, the results of
this gure prove that there are better partitions than Hus for
all . It can be concluded that the approach of Kernighan and
Lin successfully eects a good compromise between maxi-
mum load balance and minimum link cost.
Since the KL algorithm is based on a local search of
the design space (taking Hus solution as a starting cong-
uration), it may terminate in local optimum points. Global
search methods, in contrast, should be able to avoid local
optima and nally produce a better overall speedup. Spec-
tral partitioning is a global search method, because it assesses
the properties of the graph as a whole by operating on the
matrix W of node weights and edge weights [33], and it is
based on eigenvector computation for minimizing the cost
of the partitioning cut [39]. Whats more, Ws diagonal el-
ements w
m,m
= p
m
are the nodes processing runtimes ac-
cording to (1) and its o-diagonal elements w
m,n
= 2 p
m,n
amount to the potential runtimes of bus transfer node pairs,
where p
m,n
is from (2). The weight matrix W is real-valued
and symmetric, and spectral partitioning deliberately ex-
ploits this property [40, 41].
Figure 7 shows the performance results for spectral par-
titioning and nonpipelined scheduling. Unfortunately, these
results are much worse than those of Kernighan and Lin; the
95% contour line just achieves 0.8 at high bus speeds, and
the speedup spread remains large in the [0.5; 0.8] interval.
Clearly, such a behavior is completely unacceptable in prac-
tice; a fractional value of 0.5 means that the SDR implemen-
tation on the two-processor systemmeets real-time deadlines
in the exact same way as on a single-processor system. Con-
sequently, because the investment into the second processor
does not pay o at all in the form of speedup, it must be con-
cluded that the two-processor system is either ill-designed in
its hardware or ill-conditioned in its operations.
Modular Software-Dened Radio 339
10
1
10
0
10
1
10
2
Relative bus speed
0.6
0.7
0.8
0.9
1
F
r
a
c
t
i
o
n
o
f
m
a
x
s
p
e
e
d
u
p
Figure 7: Spectral partitioning, no pipelining.
Previous gures have shown that, as a matter of fact, bet-
ter contour lines and narrower spread are possible using the
same hardware and nonpipelined scheduling. Therefore, the
reason for the inferior speedup behavior of spectral parti-
tioning needs to be identied. Figure 8 shows its speedup re-
sults under partitioning. Obviously, only a small part of all
realizations experience an improvement in speedup due to
the elimination of LOGICAL WAIT IDLE conditions. No-
tably, the 5% contour line remains at the same speedup level
for high bus speeds. These results back previous ndings on
spectral partitioning; the approach in its current form [33]
does not generate well-balanced partitions. Load imbalance,
as mentioned before in the context of Hus algorithm, is a
genuine feature of partitioning, not of scheduling. Failure
to generate load-balanced partitions cannot be compensated
for by any scheduling technique.
To sum up, the KL algorithm is able to outperform Hus
algorithm (but not systematically) in the low-to-midrange
region ( < 5), if frame-by-frame signal processing is the de-
sired way of operating the software-dened physical layer of
a Mod-SDR. True systematic superiority of the KL algorithm
in speedup can only be observed under pipelining. A big dis-
advantage, however, is the depth of the pipeline; it depends
on the graph structure, its actual value is not predictable, and
it is quite a large integer number. (Histograms not shown
graphically here: mean value around 20 at M = 40 nodes
per graph, spread over a window of [5; 35], independent of
for implicit Hu and spectral partitioning, depending on
for KL.) Memory demand increases linearly with the depth
of the pipeline, and therefore pipelined operation in combi-
nation with the above approaches does not lead to workable
Mod-SDR implementations.
In retrospect, frame-by-frame signal processing must be
considered inadequate for circuit-switched services. There is
simply no need to restrict partitioning algorithms to operat-
ing on a single graph copy anyway under these conditions.
If the restriction is dropped, the QoS manager can approach
partitioning and scheduling in a dierent, much simpler way.
First of all, it turns out [35] that GDP (cf. Section 3) is op-
timal regarding delay; the depth of the pipeline is exactly 2
10
1
10
0
10
1
10
2
Relative bus speed
0.6
0.7
0.8
0.9
1
F
r
a
c
t
i
o
n
o
f
m
a
x
s
p
e
e
d
u
p
Figure 8: Spectral partitioning, pipelining.
(or 3, if continuous RF transmission is to be automatically
supported by the physical layer processing subsystem [38]).
Second, GDP is optimal regarding speedup; it reaches the
speedup limit s, or the fractional value of 1.0 in Figures 4, 6,
and 8, independent of . In view of the previously discussed
diculties in approaching the limit, GDP is certainly the best
design choice for circuit-switched services. If GDPs mem-
ory demand is still an issue, the QoS manager could easily
resort to HFP. Its speedup performance for circuit-switched
services (see [38, Figure 2]) is suboptimal, but totally com-
parable to that of the KL algorithm under pipelining (see
Figure 6), however, at a constant pipeline depth of 2 and at
less program and data memory demand than GDPs.
4.2. Packet-switched services
As mentioned in the beginning of Section 3, packet-oriented
networks may require the QoS manager to operate on a sin-
gle copy of the graph. Then the graph contains the complete
set of computational tasks necessary for processing a single
packet, but subsets of these tasks may be repetitive in nature.
Taking the IEEE 802.11a wireless LAN standard as an exam-
ple, it is easy to identify such tasks, even when looking at a
single packet only: intercarrier/intraconstellation interleav-
ing, constellation mapping, and IFFT [42]. All of these need
to be repeated for every single OFDM symbol alike, just op-
erating on dierent data within the packet. In contrast, non-
repetitive computations (per single packet) include scram-
bling and channel coding of IEEE 802.11a.
The speedup which could be expected under the con-
ditions of the random graph model and a completely
nonrepetitive task graph would be identical to that of Figures
3, 5, and 7. However, if a dominant subset of tasks were in fact
repetitive, then pipelining approaches such as HFP and GDP
could result in better speedup, when applied to the subset.
The following results of HFP and GDP for packet pro-
cessing are conditional on the assumption that there are ex-
actly N
F
frames per packet which need to be processed iden-
tically. Furthermore, both processors are considered to be
exclusively reserved for physical layer signal processing as
soon as a packet has arrived. That is to say, one processor
340 EURASIP Journal on Wireless Communications and Networking
N
F
= 2
N
F
= 3
N
F
= 4
N
F
= 5
N
F
= 6
N
F
= 7
10
1
10
0
10
1
10
2

0.5
0.6
0.7
0.8
0.9
1
F
r
a
c
t
i
o
n
o
f
m
a
x
s
p
e
e
d
u
p
Upper bounds
Figure 9: Half-frame pipelining, packet processing. Sample size
2000 per (, N
F
).
cannot nish the remainder of some higher-layer computa-
tional task, while the other processor already starts process-
ing the radio signals of a physical layer packet.
Figures 9 and 10 (adopted from [38]) show the speedup
performance of HFP and GDP, respectively. For reasons of
legibility, only the contour line triplets are drawn, parameter-
ized by the number of frames per packet N
F
: 2 N
F
7. In
comparison to circuit-switched processing, GDP is no longer
optimal, since the lling and the emptying of the pipeline
cause idle times on the processors. A detailed discussion of
HFP and GDP can be found in [38]. However, the crucial
point in the above gures lies in the dashed lines representing
upper bounds on HFP speedup, but lower bounds on GDP
speedup. Therefore, it can be concluded that GDP systemat-
ically outperforms HFP in packet processing.
Even for reasonable values of and low numbers of
frames per packet (or task repetitions in parts of a graph),
GDP closely approaches the speedup limit s. So far, only
transmissions with a constant N
F
have been examined. How-
ever, IP trac in real WLANs consists of a mix of packet sizes,
and hence physical layer packets contain dierent numbers
of radio frames. To gain more insight into this matter, packet
size statistics of some tangible system have to be known. For
the example of IEEE 802.11a, it has been shown [37] that
small N
F
(values of 10 and below) occur in the great majority
of packets, so smart signal processing of small-sized packets
is indeed an important issue in established WLAN standards.
Given its superior speedup performance, GDP should be the
rst choice for packet-oriented signal processing.
5. SUMMARY ANDCONCLUSION
As a starting point, some technical and commercial bound-
ary conditions of SDR have been briey reviewed. It follows
fromthis account that certain design issues, which are related
N
F
= 3
N
F
= 5
N
F
= 7
10
1
10
0
10
1
10
2

0.5
0.6
0.7
0.8
0.9
1
F
r
a
c
t
i
o
n
o
f
m
a
x
s
p
e
e
d
u
p
N
F
= 6
N
F
= 4
N
F
= 2
Lower bounds
Figure 10: Graph duplication pipelining, packet processing. Sam-
ple size 2000 per (, N
F
).
to real-time multiprocessing and modularity in exible sig-
nal processing software, have been neglected at large in the
existing SDR literature. Modular software-dened radio ad-
dresses these issues by looking into the organizational princi-
ples of signal processing rather than into the signal process-
ing itself. Therefore, a novel way of modeling SDR software
had to be introduced. Several techniques for mapping such
software to hardware have been briey reviewed. Mod-SDR
system simulations presented in Section 4 allow to draw the
following conclusions.
(i) With respect to circuit-switched services, frame-by-
frame signal processing has proven inadequate. Pipelining
methods such as GDP and HFP are to be preferred a priori.
GDP is optimal regarding speedup and delay. HFP is subop-
timal, but requires less memory than GDP. Therefore, HFP
can only prove competitive against GDP if memory is a se-
rious design issue. HFP can merely establish a compromise
between the achievable speedup and dynamic power dissipa-
tion of the bus.
(ii) With respect to packet-switched services, frame-by-
frame signal processing generally retains its right to exist.
Pipelining is a viable alternative only if repetitive signal pro-
cessing tasks can be identied. If so, GDP should be used.
As for the repetitive task in isolation, HFP is systematically
outperformed by GDP. Even frame-by-frame signal process-
ing (which is independent of the number of repetitions) may
show higher speedup than HFP.
Here, pipelining has been employed as a technique for
software execution. However, additional work on Mod-SDR
[34] provides strong hints that hardware subsystem pipelin-
ing also helps reducing dynamic power dissipation in CMOS
hardware, at the same time keeping speedup high. Therefore,
whenever signal processing in wireless communications is
repetitive in nature, the insertion of pipelining is the prefer-
able design guideline for Mod-SDR systems.
Modular Software-Dened Radio 341
Future research directions include the improvement of
spectral partitioning for direct comparison with the (non-
pipelined) KL approach and a more comprehensive study of
terminal behavior in packet-oriented networks. Further on, a
suitable extension of the current hardware model to hetero-
geneous multiprocessor systems and interconnect topologies
other than a bus would advance the design theory of modular
software-dened radio.
REFERENCES
[1] J. Mitola, The software radio architecture, IEEE Commun.
Mag., vol. 33, no. 5, pp. 2638, 1995.
[2] IEEE J. Select. Areas Commun., vol. 17, no. 4, 1999, Special
Issue on Software Radio.
[3] IEEE Commun. Mag., vol. 37, no. 2, 1999, Special Issue on
Software Radio.
[4] A. Ivers and D. Smith, A practical approach to the imple-
mentation of multiple radio congurations utilizing recong-
urable hardware and software building blocks, in Proc. IEEE
Military Communications Conference (MILCOM 97), vol. 3,
pp. 13271332, IEEE, Monterey, Calif, USA, November 1997.
[5] A. Kountouris, C. Moy, L. Rambaud, and P. Le Corre, A
recongurable radio case study: a software based multi-
standard transceiver for UMTS, GSM, EDGE and Bluetooth,
in Proc. IEEE Vehicular Technology Conference (VTC 01),
vol. 2, pp. 11961200, Atlantic City, NJ, USA, October 2001.
[6] O. Faust, B. Sputh, D. Nathan, S. Rezgui, A. Weisensee, and
A. Allen, A single-chip supervised partial self-recongurable
architecture for software dened radio, in Proc. 17th In-
ternational Symposium on Parallel and Distributed Processing
(IPDPS 03), pp. 191191, IEEE, Nice, France, April 2003.
[7] H. Miranda, P. Pinto, and S. Silva, A self-recongurable re-
ceiver architecture for software radio systems, in Proc. IEEE
Radio and Wireless Conference (RAWCON 03), pp. 241244,
IEEE, Boston, Mass, USA, August 2003.
[8] A. Pacici, C. Vendetti, F. Frescura, and S. Cacopardi, A re-
congurable channel codec coprocessor for software radio
multimedia applications, in Proc. International Symposium
on Circuits and Systems (ISCAS 03), vol. 2, pp. II-41II-44,
IEEE, Bangkok, Thailand, May 2003.
[9] T. Hentschel and G. Fettweis, Sample rate conversion for
software radio, IEEE Commun. Mag., vol. 38, no. 8, pp. 142
150, 2000.
[10] W. Abu-Al-Saud and G. Stuber, Ecient sample rate conver-
sion for software radio systems, in Proc. IEEE Global Telecom-
munications Conference (GLOBECOM 02), vol. 1, pp. 559
563, IEEE, Taipeh, Taiwan, Republic of China, November
2002.
[11] W. Abu-Al-Saud and G. Stuber, Modied CIC lter for sam-
ple rate conversion in software radio systems, IEEE Signal
Processing Lett., vol. 10, no. 5, pp. 152154, 2003.
[12] J. Ming, H. Y. Weng, and S. Bai, An ecient IF architecture
for dual-mode GSM/W-CDMA receiver of a software radio,
in Proc. IEEE International Workshop on Mobile Multimedia
Communications (MoMuC 99), pp. 2124, IEEE, San Diego,
Calif, USA, November 1999.
[13] J. Dodley, R. Erving, and C. Rice, In-building software radio
architecture, design and analysis, in Proc. IEEE 11th Interna-
tional Symposium on Personal, Indoor, and Mobile Radio Com-
munications (PIMRC 00), vol. 1, pp. 479483, IEEE, London,
UK, September 2000.
[14] W. Schacherbauer, A. Springer, T. Ostertag, C. Ruppel, and R.
Weigel, A exible multiband frontend for software radios us-
ing high IF and active interference cancellation, in Proc. IEEE
MTT-S International Microwave Symposium Digest (IMS 01),
vol. 2, pp. 10851088, IEEE, Phoenix, Ariz, USA, May 2001.
[15] A. Wiesler, Parametergesteuertes Software Radio f ur Mobil-
funksysteme, Ph.D. dissertation, Forschungsberichte aus dem
Institut f ur Nachrichtentechnik, Universit at Karlsruhe (TH),
Karlsruhe, Germany, May 2001.
[16] M. Beach, J. MacLeod, and P. Warr, Radio frequency trans-
lation for software dened radios, in Software Dened Radio:
Enabling Technologies, W. Tuttlebee, Ed., pp. 2578, John Wi-
ley & Sons, London, UK, 2002.
[17] P. B. Kennington and L. Astier, Power consumption of A/D
converters for software radio applications, IEEE Trans. Veh.
Technol., vol. 49, no. 2, pp. 643650, 2000.
[18] J. Singh, High speed analog-to-digital converter for software
radio applications, in Proc. IEEE 11th International Sympo-
sium on Personal, Indoor, and Mobile Radio Communications
(PIMRC 00), vol. 1, pp. 3942, IEEE, London, UK, Septem-
ber 2000.
[19] G. Ahlquist, M. Rice, and B. Nelson, Error control coding in
software radios: an FPGA approach, IEEE Personal Commu-
nications, vol. 6, no. 4, pp. 3539, 1999.
[20] M. Valenti, An ecient software radio implementation of the
UMTS turbo codec, in Proc. IEEE 12th International Sympo-
sium on Personal, Indoor, and Mobile Radio Communications
(PIMRC 01), vol. 2, pp. G108G113, IEEE, San Diego, Calif,
USA, September 2001.
[21] V. Thara and M. Siddiqi, Power eciency of software radio
based turbo codec, in Proc. IEEE Region 10 Conference on
Computers, Communications, Control, and Power Engineering
(TENCON 02), vol. 2, pp. 10601063, IEEE, Beijing, China,
October 2002.
[22] A. Wiesler and F. K. Jondral, Asoftware radio for 2nd and 3rd
generation systems, IEEE Trans. Veh. Technol., vol. 51, no. 4,
pp. 738748, 2002.
[23] F. K. Jondral, Parametrizationa technique for SDR imple-
mentation, in Software Dened Radio: Enabling Technologies,
W. Tuttlebee, Ed., pp. 232256, John Wiley & Sons, London,
UK, 2002.
[24] J. Mitola, Software radio architecture: a mathematical per-
spective, IEEE J. Select. Areas Commun., vol. 17, no. 4,
pp. 514538, 1999.
[25] J. Mitola, Software radiossurvey, critical evaluation and
future directions, in Proc. National Telesystems Conference
(NTC 92), pp. 13/1513/23, IEEE, Washington, DC, USA,
May 1992.
[26] C. Dick, Reinventing the signal processor, Xcell Journal,
vol. 45, pp. 7275, Spring 2003.
[27] P. Galicki, FPGAs have the multiprocessing I/O infrastruc-
ture to meet 3G base station design goals, Xcell Journal,
vol. 45, pp. 8084, Spring 2003.
[28] Software communications architecture specication,
jtrs-5000sca v2.2.1, Joint Tactical Radio System (JTRS)
Joint Program Oce, April 2004, [Online] available:
http://jtrs.army.mil.
[29] A.-R. Rhiemeier and F. K. Jondral, Mathematical modeling
of the software radio design problem, IEICE Transactions on
Communications, vol. E86-B, no. 12, pp. 34563467, 2003,
Special Issue on Software Dened Radio Technology and Its
Applications.
[30] A.-R. Rhiemeier and F. K. Jondral, Asoftware partitioning al-
gorithm for modular software dened radio, in Proc. 6th In-
ternational Symposium on Wireless Personal Multimedia Com-
munications (WPMC 03), pp. 4246, Yokosuka, Japan, Octo-
ber 2003.
[31] A.-R. Rhiemeier and F. K. Jondral, On the design of modu-
lar software dened radio systems, in Proc. IEE Colloquium
342 EURASIP Journal on Wireless Communications and Networking
on DSP Enabled Radio, Institute for System Level Integration
(ISLI), Alba Campus, Livingston, Scotland, UK, September
2003.
[32] A.-R. Rhiemeier and F. K. Jondral, Enhanced resource uti-
lization in software dened radio terminals, in Interna-
tionales Wissenschaftliches Kolloquium (IWK 03), Technische
Universit at, Ilmenau, Germany, September 2003.
[33] U. Berthold, A.-R. Rhiemeier, and F. K. Jondral, Spectral par-
titioning for modular software dened radio, in IEEE 59th
Vehicular Technology Conference (VTC 04), vol. 2, pp. 1218
1222, Milano, Italy, May 2004.
[34] A.-R. Rhiemeier and F. K. Jondral, Software partitioning and
hardware architecture for modular SDR systems, in Proc.
Software Dened Radio Technical Conference and Product Ex-
position (SDR 03), vol. 2, pp. 915, SDR Forum, Orlando, Fla,
USA, November 2003.
[35] U. Berthold, A.-R. Rhiemeier, and F. K. Jondral, A pipelin-
ing approach to operating modular software dened radio, in
Proc. IEEE/Sarno Symposiumon Advances in Wired and Wire-
less Communication (SARNOFF 04), pp. 201204, Princeton,
NJ, USA, April 2004.
[36] A.-R. Rhiemeier, A comparison of scheduling approaches in
modular software dened radio, in Proc. 3rd Karlsruhe Work-
shop on Software Radios (WSR04), pp. 3338, Karlsruhe, Ger-
many, March 2004, also appeared as reprint in: Frequenz: Jour-
nal of Telecommunications, vol. 58, no. 5/6, pp. 115120, 2004.
[37] A.-R. Rhiemeier, T. Weiss, and F. K. Jondral, Half-frame
pipelining for modular software dened radio, in Proc. IEEE
15th International Symposium on Personal, Indoor, and Mobile
Radio Communications (PIMRC 04), vol. 3, pp. 16641668,
IEEE, Barcelona, Spain, September 2004.
[38] A.-R. Rhiemeier, T. Weiss, and F. K. Jondral, A simple and ef-
cient solution to half-frame pipelining for modular software
dened radio, in Proc. Software Dened Radio Technical Con-
ference and Product Exposition (SDR 04), vol. A, pp. 119125,
SDR Forum, Phoenix, Ariz, USA, November 2004.
[39] C. H. Q. Ding, X. He, H. Zha, M. Gu, and H. D. Simon, A
min-max cut algorithm for graph partitioning and data clus-
tering, in Proc. IEEE International Conference on Data Min-
ing (ICDM 01), pp. 107114, San Jose, Calif, USA, November
2001.
[40] M. Fiedler, Algebraic connectivity of graphs, Czechoslovak
Mathematical Journal, vol. 23, no. 98, pp. 298305, 1973.
[41] M. Fiedler, A property of eigenvectors of non-negative
symmetric matrices and its application to graph theory,
Czechoslovak Mathematical Journal, vol. 25, no. 100, pp. 619
633, 1975.
[42] ANSI/IEEE Std 802.11, Wireless LAN MAC and PHY speci-
cations, 1999 Edition, and IEEE 802.11a-1999, High-speed
Physical Layer in the 5 GHz Band.
Arnd-Ragnar Rhiemeier received a rst
degree in electrical engineering from the
Ruhr-Universit at Bochum, Germany, in
1995, then continued his studies at the
Universit at Karlsruhe (TH), Germany. Sup-
ported by a grant from the German Aca-
demic Exchange Service (DAAD), he spent
two terms at the National Institute of Ap-
plied Sciences, Lyon, France, in 1996 and
1997, working in the eld of pattern recog-
nition. In 1998, he resumed his graduate studies at the Institut
f ur Nachrichtentechnik, Universit at Karlsruhe (TH). In 1999, he
completed his nal project at the Center for Communications and
Signal Processing Research (CCSPR), New Jersey Institute of Tech-
nology, Newark, NJ, USA, and received a Dipl.-Ing. degree in elec-
trical engineering from the Universit at Karlsruhe (TH), Germany.
Subsequently, he committed himself to a teaching and research as-
sistantship position at the Institut f ur Nachrichtentechnik. In 2004,
he received a Ph.D. degree summa cum laude in telecommuni-
cations for his work on software-dened radio architectures and
algorithms. His current professional interests include design ow
methodologies and the productization of advanced concepts in
communications such as MIMO and software-dened radio. He
has been an IEEE Member for 11 years and served as the Chairman
of the IEEE Student Branch Karlsruhe for two years.
EURASIP Journal on Wireless Communications and Networking 2005:3, 343353
c 2005 B. Dong and X. Wang
Adaptive Mobile Positioning in WCDMA Networks
B. Dong
Department of Electrical and Computer Engineering, Queens University, Kingston, ON, Canada K7L 3N6
Xiaodong Wang
Department of Electrical Engineering, Columbia University, New York, NY 10027-4712, USA
Email: wangx@ee.columbia.edu
Received 6 November 2004; Revised 14 March 2005
We propose a new technique for mobile tracking in wideband code-division multiple-access (WCDMA) systems employing multi-
ple receive antennas. To achieve a high estimation accuracy, the algorithm utilizes the time dierence of arrival (TDOA) measure-
ments in the forward link pilot channel, the angle of arrival (AOA) measurements in the reverse-link pilot channel, as well as the
received signal strength. The mobility dynamic is modelled by a rst-order autoregressive (AR) vector process with an additional
discrete state variable as the motion oset, which evolves according to a discrete-time Markov chain. It is assumed that the param-
eters in this model are unknown and must be jointly estimated by the tracking algorithm. By viewing a nonlinear dynamic system
such as a jump-Markov model, we develop an ecient auxiliary particle ltering algorithm to track both the discrete and contin-
uous state variables of this system as well as the associated system parameters. Simulation results are provided to demonstrate the
excellent performance of the proposed adaptive mobile positioning algorithm in WCDMA networks.
Keywords and phrases: mobility tracking, Bayesian inference, jump-Markov model, auxiliary particle lter.
1. INTRODUCTION
Mobile positioning [1, 2, 3, 4], that is, estimating the location
of a mobile user in wireless networks, has recently received
signicant attention due to its various potential applications
in location-based services, such as location-based billing, in-
telligent transportation systems [5], and the enhanced-911
(E-911) wireless emergence services [6]. In addition to fa-
cilitating these location-based services, the mobility infor-
mation can also be used by a number of control and man-
agement functionalities in a cellular system, such as mobile
location indication, hando assistance [3], transmit power
control, and admission control.
Various mobile positioning schemes have been proposed
in the literature. Typically, they are based on the measure-
ments of received signal strength [7], time of arrival (TOA)
or time dierence of arrival (TDOA) [8], and angle of arrival
(AOA) [4]. In [4], a hybrid TDOA/AOA method is proposed
and the mobile user location is calculated using a two-step
least-square estimator. Although this scheme oers a higher
location accuracy than the pure TDOA scheme, there is still
This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
a gap between its performance and the optimal performance
since it is based on a linear approximation of the highly non-
linear mobility model. Moreover, that work deals with the
static scenario only and does not address mobility tracking in
a dynamic environment. In [2, 9, 10], the extended Kalman
lter (EKF) is used to track the user mobility. It is well known
that the EKF is based on linearization of the underlying non-
linear dynamic system and often diverges when the system
exhibits strong nonlinearity.
On the other hand, the recently emerged sequential
Monte-Carlo (SMC) methods [11, 12] are powerful tools for
online Bayesian inference of nonlinear dynamic systems. The
SMC can be loosely dened as a class of methods for solv-
ing online estimation problems in dynamic systems, by re-
cursively generating Monte-Carlo samples of the state vari-
ables or some other latent variables. In [3], an SMC algo-
rithm for mobility tracking and hando in wireless cellular
networks is developed. In [8], several SMC algorithms for
positioning, navigation, and tracking are developed, where
the mobility model is simpler than the one used in [3]. Note
that in both works, the trial sampling density is based only
on the prior distribution and does not make use of the mea-
surement information, which renders the algorithms less ef-
cient. Moreover, the model parameters are assumed to be
perfectly known, which is not realistic for practical mobile
positioning systems.
344 EURASIP Journal on Wireless Communications and Networking
In this paper, we propose to employ a more ecient SMC
method, the auxiliary particle lter, to jointly estimate both
the mobility information (location, velocity, acceleration,
and the state sequence of commands) and the unknown sys-
tem parameters. We assume the mobility estimation is based
on TDOA measurements at the mobile station (MS) and
AOA measurements as well as the received signal strength
measurements in the neighbor base stations (BSs). All these
measurements are available in WCDMA networks. The re-
mainder of this paper is organized as follows. In Section 2,
we describe the nonlinear dynamic system model under con-
sideration, and present the mathematical formulation for the
problemof mobility tracking in a WCDMAwireless network.
In Section 3, we briey introduce some background materi-
als on sequential Monte-Carlo techniques. The new mobil-
ity tracking algorithms are developed in Section 4. Section 5
provides the simulation results; and Section 6 contains the
conclusions.
2. SYSTEMDESCRIPTIONS
2.1. Mobility model
Assume that a mobile of interest moves on a two-
dimensional plane, and the motion state x
k
[x
k
, v
x,k
,
r
x,k
, y
k
, v
y,k
, r
y,k
]
T
corresponds to the observation measure-
ments at t
k
= t
0
+ t k, where t is the sampling time in-
terval; x
k
and y
k
are, respectively, the horizontal and vertical
Cartesian coordinates of the mobile position at time instance
k; v
x,k
and v
y,k
are the corresponding velocities; r
x,k
and r
y,k
are the corresponding accelerators. The discrete-time mov-
ing equation can be expressed as [2, 3]
_
_
_
_
_
x
k
v
x,k
y
k
v
y,k
_
_
_
_
_
=
_
_
_
_
_
1 t 0 0
0 1 0 0
0 0 1 t
0 0 0 1
_
_
_
_
_
_
_
_
_
_
x
k1
v
x,k1
y
k1
v
y,k1
_
_
_
_
_
+
_
_
_
_
_
_
_
_
_
_
t
2
2
0
t 0
0
t
2
2
0 t
_
_
_
_
_
_
_
_
_
_
_
a
x,k1
a
y,k1
_
,
(1)
where a
k
[a
x,k
, a
y,k
]
T
is the driving acceleration vector at
time k. Note that in mobility tracking applications, the time
interval t between two consecutive update intervals is typi-
cally on the order of several hundred symbol intervals to al-
low for the measurements of TDOA, AOA, and RSS. Such
a relatively large time scale also makes it possible to employ
more sophisticated signal processing methods for more ac-
curate mobility tracking.
In practical cellular systems, a mobile user may have sud-
den and unexpected changes in acceleration caused by trac
lights and/or road turn; on the other hand, the acceleration
of the mobile may be highly correlated in time. In order to
incorporate the unexpected as well as the highly correlated
changes in acceleration, we model the motion of a user as a
dynamic system driven by a command s
k
[s
x,k
, s
y,k
]
T
and
a correlated random acceleration r
k
[r
x,k
, r
y,k
]
T
, that is,
a
k
= s
k
+ r
k
. Following [2, 3], the command s
k
is modelled
as a rst-order discrete-time Markov chain with nite state
S = {S
1
, S
2
, . . . , S
N
} and the transition probability matrix
A [a
i, j
], a
i, j
P(s
k
= S
j
| s
k1
= S
i
). It is assumed that
a
i, j
= p for i = j and a
i, j
= (1 p)/(N 1) for i = j, where
N is the total number of states. The correlated random ac-
celerator r
k
is modelled as the rst-order autoregressive (AR)
model, that is, r
k
= r
k1
+w
k
, where is the AR coecient,
0 < < 1, and w
k
is a Gaussian noise vector with covariance
matrix
2
w
I.
Based on the above discussion, the motion model can be
expressed as
_
_
_
_
_
_
_
_
_
x
k
v
x,k
r
x,k
y
k
v
y,k
r
y,k
_
_
_
_
_
_
_
_
_
. , .
xk
=
_
_
_
_
_
_
_
_
_
_
_
_
_
_
1 t
t
2
2
0 0 0
0 1 t 0 0 0
0 0 0 0 0
0 0 0 1 t
t
2
2
0 0 0 0 1 t
0 0 0 0 0
_
_
_
_
_
_
_
_
_
_
_
_
_
_
. , .
B
_
_
_
_
_
_
_
_
_
x
k1
v
x,k1
r
x,k1
y
k1
v
y,k1
r
y,k1
_
_
_
_
_
_
_
_
_
. , .
xk1
+
_
_
_
_
_
_
_
_
_
_
_
_
_
_
t
2
2
0
t 0
0 0
0
t
2
2
0 t
0 0
_
_
_
_
_
_
_
_
_
_
_
_
_
_
. , .
Cs
_
s
x,k
s
y,k
_
. , .
sk
+
_
_
_
_
_
_
_
_
_
_
_
_
_
_
t
2
2
0
t 0
1 0
0
t
2
2
0 t
0 1
_
_
_
_
_
_
_
_
_
_
_
_
_
_
. , .
Cw
_
w
x,k
w
y,k
_
. , .
wk
.
(2)
In short,
x
k
= Bx
k1
+ C
s
s
k
+ C
w
w
k
. (3)
2.2. Measurement model
Some new features in WCDMA systems (e.g., cdma2000)
such as network synchrony among the BSs, dedicated
reverse-link for each MS, adaptive antenna array for AOA es-
timation, and forward link common broadcasting channel,
make several measurements available in practice for mobile
tracking.
First of all, methods for determining the time dier-
ence of arrival (TDOA) from the spread-spectrum signal,
including the coarse timing acquisition with a sliding cor-
relator or matched lter, and ne timing acquisition with a
delay-locked loop (DLL) or tau-dither loop (TDL) [13, 14],
can be applied in WCDMA systems. Coarse timing acqui-
sition can achieve the accuracy within one chip duration
whereas ne synchronization by the DLL can achieve the
accuracy within fractional portion of chip duration. More-
over, in WCDMA systems, much higher chip rate is used
than that in IS-95 systems with shorter chip period, thereby
improving the precision of timing. Furthermore, with mul-
tiple antennas to collect the radio signal at the base sta-
tion (in particular, phaseinformation), we could apply array
Adaptive Mobile Positioning in WCDMA Networks 345
signal processing algorithms (e.g., MUSIC or ESPRIT) [15]
to estimate the angle of arrival (AOA). In addition, the re-
ceived signal strength indicator (RSSI) signal in WCDMA
systems contains the distance information between a mobile
and a given base station, which is quantied by the large-
scale path-loss model with lognormal shadowing [16]. Note
that by averaging the received pilot signal, the rapid uctu-
ation of multipath fading (i.e., small-scale fading eect) is
mitigated. Based on the above discussion, we consider the
measurements for mobility tracking to include RSS p
k,i
, AOA

k,i
at the BSs, and TDOA
k,i
fed back from MS. Denote
D
k,i
= [(x
k
a
i
)
2
+ (y
k
b
i
)
2
]
1/2
, where (a
i
, b
i
) is the po-
sition of the ith BS. We have
p
k,i
= p
0,i
10 log D
k,i
+ n
p
k,i
, i = 1, 2, 3,

k,i
=
1
c
_
D
k,i
D
k,1
_
+ n

k,i
, i = 2, 3,

k,i
= tan
(1)
_
y
k
b
i
x
k
a
i
_
+ n

k,i
, i = 1, 2, 3,
(4)
where i is the BS index; p
0,i
is a constant determined by
the wavelength and the antenna gain of the ith BS; n
p
k,i

N(0,
d
) is the logarithm of the shadowing component,
which is modelled as Gaussian distribution; c is the speed of
light and is the path-loss factor; n

k,i
N(0,

) is the mea-
surement noise of TDOA between the ith BS and the serving
BS; and n

k,i
N(0,

) is the estimation error of AOA at the


ith BS. The noise terms in (4) are assumed to be white both
in space and in time.
Denote the measurements at time instance k as z
k

[p
k,1
, p
k,2
, p
k,3
,
k,2
,
k,3
,
k,1
,
k,2
,
k,3
]
T
. Then, we have the
following measurement equation of the underlying dynamic
model:
z
k
= h
_
x
k
_
+ v
k
, (5)
where v
k
[n
p
k,1
, n
p
k,2
, n
p
k,3
n

k,1
, n

k,2
, n

k,1
, n

k,2
, n

k,3
] with co-
variance matrix Q = diag(
d
I,
t
I,

I); and h(x


k
)
[h
1
(x
k
), . . . , h
8
(x
k
)] where the form of each h
i
(), is given by
(4). Note that the availability of TDOAand AOAwill enhance
the mobility tracking accuracy. In practice, if in some served
mobiles such information is not available, mobility tracking
can still be performed based only on the RSS measurement,
with less accuracy.
2.3. Problemformulation
Based on the discussions above, the nonlinear dynamic sys-
tem under consideration can be represented by a jump-
Markov model as follows:
s
k
MC(, A), x
k
=Bx
k1
+ C
s
s
k
+ C
w
w
k
, z
k
=h(x
k
)+v
k
,
(6)
where MC(, A) denotes a rst-order Markov chain with
initial probability vector and transition matrix A. De-
note the observation sequence up to time k as Z
k

[z
1
, z
2
, . . . , z
k
], the corresponding discrete state sequence
S
k
[s
1
, s
2
, . . . , s
k
], and the continuous state sequence
X
k
[x
1
, x
2
, . . . , x
k
]. Let the model parameters be =
{, A,
w
,
d
,
t
,

}. Given the observations Z


k
up to time
k, our problem is to infer the current position and velocity.
This amounts to making inference with respect to
p
_
x
k
| Z
k
_
=
_
SkS
k
_

_
p
_
s
1
, . . . , s
k
, x
1
, . . . , x
k1
| Z
k
_
dx
1
, . . . , dx
k1

_
SkS
k
_

_
k

i=1
_
p
_
z
i
| x
i
_
p
_
x
i
| x
i1
, s
i
_
p
_
s
i
| s
i1
_
_
dx
1
, . . . , dx
k1
.
(7)
The above exact expression of p(x
k
| Z
k
) involves very high-
dimensional integrals and the dimensionality grows linearly
with time, which is prohibitive to compute in practice. In
what follows, we resort to the sequential Monte-Carlo tech-
niques to solve the above inference problem.
3. BACKGROUNDONSEQUENTIAL MONTE CARLO
Consider the following jump-Markov model:
x
k
= A
_
s
k
_
x
k1
+ B
_
s
k
_
v
k
,
z
k
= C
_
s
k
_
x
k
+ D
_
s
k
_

k
,
(8)
where v
k
i.i.d.
N
c
(0,
v
I),
k
i.i.d.
N
c
(0,

I), and s
k
is the
discrete hidden state evolving according to a discrete-time
Markov chain with initial probability vector and transi-
tion probability matrix A. Denote y
k
{x
k
, s
k
} and the
system parameters = {, A,
v
,

}. Suppose we want to
make an online inference about the unobserved states Y
k
=
(y
1
, y
2
, . . . , y
k
) from a set of available observations Z
k
=
(z
1
, z
2
, . . . , z
k
). Monte-Carlo methods approximate such in-
ference by drawing m random samples {Y
( j)
k
}
m
j=1
from the
posterior distribution p(Y
k
| Z
k
). Since sampling directly
from p(Y
k
| Z
k
) is often not feasible or computationally
too expensive, we can instead draw samples from some trial
346 EURASIP Journal on Wireless Communications and Networking
sampling density q(Y
k
| Z
k
), and calculate the target infer-
ence E
p
{(Y
k
) | Z
k
} using samples drawn from q() as
E
p
_

_
Y
k
_
| Z
k
_

=
1
W
k
m
_
j=1
w
( j)
k

_
Y
( j)
k
_
, (9)
where w
( j)
k
= p(Y
( j)
k
| Z
k
)/q(Y
( j)
k
| Z
k
), W
k
=

m
j=1
w
( j)
k
,
and the pair {Y
( j)
k
, w
( j)
k
}
m
j=1
is called a set of properly weighted
samples with respect to the distribution p(Y
k
| Z
k
) [17].
Suppose a set of properly weighted samples {Y
( j)
k1
,
w
( j)
k1
}
m
j=1
with respect to p(Y
k1
| Z
k1
) has been drawn
at time (k 1), the sequential Monte-Carlo (SMC) proce-
dure generates a new set of samples {Y
( j)
k
, w
( j)
k
}
m
j=1
properly
weighted with respect to p(Y
k
| Z
k
). In [18], it is shown that
the optimal trial distribution is p(y
k
| Y
( j)
k1
, Z
k
), which min-
imizes the conditional variance of the importance weights.
The SMC recursion at time k is as follows [17, 19].
For j = 1, . . . , m,
(i) draw a sample y
( j)
k
from the trial distribution
p
_
y
k
| Y
( j)
k1
, Z
k
_
p
_
z
k
| y
k
_
p
_
y
k
| y
( j)
k1
_
= p
_
z
k
| x
k
, s
k
_
p
_
x
k
| x
( j)
k1
, s
k
_
p
_
s
k
| s
( j)
k1
_
,
(10)
and let Y
( j)
k
= (Y
( j)
k1
, y
( j)
k
),
(ii) update the importance weight
w
( j)
k
w
( j)
k1
p
_
z
k
| Y
( j)
k1
, Z
k1
_
= w
( j)
k1
N
_
sk=1
p
_
s
k
| s
( j)
k1
_
_
p
_
z
k
| x
k
, s
k
, Z
k1
_
p
_
x
k
| x
( j)
k1
, s
k
_
dx
k
.
(11)
Apparently, it is dicult to use such an optima trial sampling
density because the importance weight update equation does
not admit a closed-form and involves a high-dimension inte-
gral for each sample stream [19]. To approximate the integral
in (11), we use
_
p
_
z
k
| x
k
, s
k
, Z
k1
_
p
_
x
k
| x
( j)
k1
, s
k
_
dx
k

_
p
_
z
k
| x
k
, s
k
, Z
k1
_

_
x
k
=
( j)
k
_
x
( j)
k1
, s
k
__
dx
k
= p
_
z
k
| Z
k1
,
( j)
k
_
x
( j)
k1
, s
k
__
,
(12)
where
k
(x
( j)
k1
, s
k
) is the mean of p(x
k
| x
( j)
k1
, s
k
). Using (12),
the importance weight update is approximated by
w
( j)
k
w
( j)
k1
N
_
sk=1
p
_
z
k
|
( j)
k
_
x
( j)
k1
, s
k
_
, Z
k1
_
p
_
s
k
| s
( j)
k1
_
. , .

_
x
( j)
k1
,s
( j)
k1
,zk
_
.
(13)
To make the SMC procedure ecient in practice, it is
necessary to use a resampling procedure as suggested in
[17, 18]. Roughly speaking, the aim of resampling is to du-
plicate the sample streams with large importance weights
while eliminating the streams with small ones. In [19], it is
suggested that we resample {Y
( j)
k1
} according to the weights

( j)
k
w
( j)
k1
(x
( j)
k1
, s
( j)
k1
, z
k
). Since the term (x
( j)
k1
, s
( j)
k1
, z
k
)
is independent of s
( j)
k
and x
( j)
k
, we use it as the p.d.f. for
generating the auxiliary index
k
before we sample the state
variables (s
k
, x
k
). Such a scheme is termed as the auxiliary
particle lter [20], where some auxiliary variable is intro-
duced in the sampling space such that the trial distribu-
tion for the auxiliary variable can make use of the cur-
rent measurement z
k
. In order to utilize the observation in
the trial sampling density of s
k
, we sample s
k
according to
p(z
k
|
k
(x
(
( j)
k
)
k1
, s
k
))p(s
k
| s
(
( j)
k
)
k1
) and sample x
k
according to
p(x
k
| X
(
( j)
k
)
k1
, s
( j)
k
). The importance weights are then updated
according to
w
( j)
k

p
_
z
k
| x
( j)
k
_
p
_
x
k
| x
(
( j)
k
)
k1
, s
( j)
k
_
p
_
s
( j)
k
| s
(
( j)
k
)
k1
_
p
_
z
k
|
k
_
x
(
( j)
k
)
k1
, s
( j)
k
__
p
_
s
( j)
k
| s
(
( j)
k
)
k1
_
p
_
x
k
| x
(
( j)
k
)
k1
, s
( j)
k
_
=
p
_
z
k
| x
( j)
k
_
p
_
z
k
|
k
_
x
(
( j)
k
)
k1
, s
( j)
k
__
.
(14)
Considering the jump-Markov model (8), we have

k
(x
(
( j)
)
k1
, s
( j)
k
) = E{x
k
| s
k
, X
(
( j)
k
)
k1
} = A(s
k
)x
(
( j)
k
)
k1
. The
auxiliary particle lter algorithm at the kth recursion is
summarized in Algorithm 1.
If the system parameter is unknown, we need to aug-
ment the unknown parameter to the state variable y
k
as
Adaptive Mobile Positioning in WCDMA Networks 347
(i) For j = 1, . . . , m and s
k
= 1, . . . , N, calculate the trial
sampling density
( j)
k
w
( j)
k1
(x
( j)
k1
, s
( j)
k1
, z
k
).
(ii) For j = 1, . . . , m,
(a) draw the auxiliary index
( j)
k
with probability
( j)
k
,
(b) draw a sample s
k
from the trial distribution
p(z
k
|
k
(x
(
( j)
k
)
k1
, s
k
))p(s
k
| s
(
( j)
k
)
k1
) and let
S
( j)
k
= (S
(
( j)
k
)
k1
, s
( j)
k
),
(c) draw a sample x
k
from the trial distribution
p(x
k
| X
(
( j)
k
)
k1
, s
( j)
k
) and let X
( j)
k
= (X
(
( j)
k
)
k1
, x
( j)
k
),
(d) update the importance weight
w
( j)
k
p(z
k
| x
( j)
k
)/ p(z
k
|
k
(x
(
( j)
k
)
k1
, s
( j)
k
)).
Algorithm 1: The auxiliary particle lter algorithm at the kth re-
cursion.
the new state variable. Therefore, we have to sample from the
joint density
p
_
y
k
, | Y
( j)
k1
, Z
k
_
= p
_
y
k
| Y
( j)
k1
, z
k
,
_
p
_
| Y
( j)
k1
, Z
k1
_
p
_
z
k
| y
k
,
_
p
_
y
k
| y
( j)
k1
,
_
p
_
| Y
( j)
k1
, Z
k1
_
.
(15)
And the importance weights are updated according to
w
( j)
k
w
( j)
k1
p
_
z
k
| Y
( j)
k1
, Z
k1
,
( j)
_
w
( j)
k1
N
_
sk=1
p
_
z
k
|
k
_
x
( j)
k1
, s
k
_
,
( j)
_
p
_
s
k
| s
( j)
k1
,
( j)
_
.
(16)
For each sample stream j, the trial sampling density for the
state variable (s
k
, x
k
) and the importance weight update are
both based on the sampled unknown parameter
( j)
. At the
end of the kth iteration, we update the trial sampling density
p( | Y
( j)
k
, Z
k
) based on p( | Y
( j)
k1
, Z
k1
), y
( j)
k
and z
k
. The
auxiliary particle lter algorithm at the kth recursion for the
case of unknown parameters is summarized in Algorithm 2.
4. NEWMOBILITY TRACKINGALGORITHM
4.1. Online estimator with known parameters
We next outline the SMC algorithm for solving the prob-
lem of mobility tracking based on the jump-Markov model
given by (6). Let y
k
= (x
k
, s
k
), X
k
= (x
1
, x
2
, . . . , x
k
), S
k
=
(s
1
, . . . , s
k
), Y
k
= (y
1
, . . . , y
k
), and Z
k
= (z
1
, . . . , z
k
). The
aim of mobility tracking is to estimate the posterior distri-
bution of p(Y
k
| Z
k
). Using SMC, we can obtain a set of
Monte-Carlo samples of the unknown states {Y
( j)
k
, w
( j)
k
}
m
j=1
that are properly weighted with respect to the distribution
p(Y
k
| Z
k
). The MMSE estimator of the location and veloc-
ity at time k can then be approximated by
E
_
x
k
| Z
k
_

=
1
W
k
m
_
j=1
x
( j)
k
w
( j)
k
, k = 1, 2, . . . , (17)
(i) For j = 1, . . . , m,
(a) draw samples of the unknown parameter {
( j)
}
m
j=1
from p( | Y
( j)
k1
, Z
k1
),
(b) calculate the auxiliary variable sampling density

( j)
k
w
( j)
k1

N
s
k
=1
p(z
k
|
k
(x
( j)
k1
, s
k
),
( j)
)p(s
k
|
s
( j)
k1
,
( j)
).
(ii) For j = 1, . . . , m,
(a) draw the auxiliary index
( j)
k
with probability
( j)
k
,
(b) draw a sample s
( j)
k
from the trial distribution
p(z
k
|
k
(x
(
( j)
k
)
k1
, s
k
),
( j)
)p(s
k
| s
(
( j)
k
)
k1
,
( j)
),
(c) draw a sample x
( j)
k
from the trial distribution
p(x
k
| X
(
( j)
k
)
k1
, s
( j)
k
,
( j)
) and let y
( j)
k
= (s
( j)
k
, x
( j)
k
) and
let Y
( j)
k
= (Y
(
( j)
k
)
k1
, y
( j)
k
),
(d) update the importance weight
w
( j)
k
p(z
k
| x
( j)
k
,
( j)
)/ p(z
k
|
k
(x
(
( j)
k
)
k1
, s
( j)
k
),
( j)
),
(e) update the sampling density p( | Y
( j)
k
, Z
k
) based on
p( | Y
( j)
k1
, Z
k1
), y
( j)
k
and z
k
.
Algorithm 2: The auxiliary particle lter algorithm of the kth re-
cursion for the case of unknown parameters.
where W
k
=

m
j=1
w
( j)
k
. Following the auxiliary particle lter
framework discussed in Section 3, we choose the sampling
density for generating the auxiliary index
k
as
q
_

k
= j
_
w
( j)
k1
_
sS
p
_
z
k
|
k
_
x
( j)
k1
, s
__
p
_
s | s
( j)
k1
_
, j = 1, . . . , m.
(18)
Considering the motion equation (3) and the measurement
equation (5), we have
k
(x
( j)
k1
, s) = Bx
( j)
k1
+C
s
s. Next we draw
a sample of state s
k
from the trial distribution
q
_
s
k
= s
_
p
_
z
k
|
k
_
x
(
( j)
k
)
k1
, s
__
p
_
s | s
(
( j)
k
)
k1
_
=
_
h
_

k
_
x
(
( j)
k
)
k1
, s
__
, Q
_
a
s
(
( j)
k
)
k1
,s
,
(19)
where (, ) denotes the p.d.f. of a multivariate Gaussian
distribution with mean and covariance . The trial sam-
pling density for x
k
is given by
p
_
x
k
| x
(
( j)
k
)
k1
, s
( j)
k
_
=
_
Bx
(
( j)
k
)
k1
+ C
s
s
( j)
k
,
w
C
w
C
T
w
_
. (20)
And the importance weight is updated according to
w
( j)
k

p
_
z
k
| x
( j)
k
_
p
_
z
k
|
k
_
x
(
( j)
k
)
k1
, s
k
__
, (21)
where p(z
k
| x
k
) = (h(x
k
), Q). Finally, we summarize the
adaptive mobile positioning algorithm with known parame-
ters in Algorithm 3.
348 EURASIP Journal on Wireless Communications and Networking
(I) Initialization: for j = 1, . . . , m, draw the state vector x
( j)
0
from the multivariate Gaussian distribution N(x
0
, 10I)
and draw s
( j)
0
uniformly fromS; all importance weights
are initialized as w
( j)
0
= 1.
(II) For k = 1, 2, . . .,
(a) for j = 1, . . . , m, calculate the trial sampling
density for the auxiliary index according to (18),
(b) for j = 1, . . . , m,
(i) draw an auxiliary index
( j)
k
with the
probability q(
k
= j),
(ii) draw a sample s
( j)
k
according to (19),
(iii) draw a sample x
( j)
k
according to (20),
(iv) update the importance weight w
( j)
k
according
to (21),
(v) append y
( j)
k
= {x
( j)
k
, s
( j)
k
} to Y
(
( j)
)
k1
to form
Y
( j)
k
= {Y
(
( j)
)
k1
, y
( j)
k
}.
Algorithm 3: Adaptive mobile positioning algorithm with known
system parameters.
Complexity
The major computation involved in Algorithm 3 includes
evaluations of Gaussian densities (i.e., mN evaluations in
(18), mN evaluations in (19), and m evaluations in (21))),
and simple multiplications (i.e., mN multiplications in (18)
and mN multiplications in (19)). Note that Algorithm 3 is
well suited for parallel implementations.
4.2. Online estimator with unknown parameters
We next treat the problem of jointly tracking the state Y
k
and
the unknown parameters = {, A,
w
,
d
,
t
,

}. We rst
specify the priors for the unknown parameters. For the initial
probability vector and the ith row of the transition prob-
ability matrix A, we choose a Dirichlet distribution as their
priors:
D
_

1
,
2
, . . . ,
N
_
,
a
i
D
_

1
,
2
, . . . ,
N
_
, i = 1, . . . , N.
(22)
For the noise variances,
w
,
d
,
t
, and

, we use the inverse


chi-square priors:

w

2
_

0,w
,
0,w
_
,
d

2
_

0,d
,
0,d
_
,

t

2
_

0,t
,
0,t
_
,


2
_

0,
,
0,
_
.
(23)
Suppose that at time (k 1), we have m sample streams of
state Y
k1
and parameter , {Y
( j)
k1
,
( j)
k1
}
m
j=1
, and the asso-
ciated importance weights {w
( j)
k1
}
m
j=1
, representing an im-
portant sample approximation to the posterior distribution
p(Y
k1
, | Z
k1
) at time (k 1). Note that here the index k
on the parameter samples indicates that they are drawn from
the posterior distribution at time k rather than implying that
is time-varying. By applying Bayes theorem and consider-
ing the system equations (6), at time k, we sample the state
variable and the unknown parameter from
p
_
y
k
, | Y
( j)
k1
, Z
k
_
p
_
z
k
| y
k
,
_
p
_
y
k
| y
( j)
k1
, z
k
,
_
p
_
| Y
( j)
k1
, Z
k1
_
,
(24)
where p( | Y
( j)
k1
, Z
k1
) is the trial sampling density for the
unknown parameter at time (k 1) and can be decomposed
as
p
_
| Y
( j)
k1
, Z
k1
_
= p
_
, A,
w
,
v
,
t
,

| Y
( j)
k1
, Z
k1
_
= p
_
| s
( j)
0
_
_
N

i=1
p
_
a
i
| , S
( j)
k1
_
_
p
_

w
| X
( j)
k1
, Z
k
_
p
_

d
| X
( j)
k1
, Z
k1
_
p
_

t
| X
( j)
k1
, Z
k1
_
p
_

| X
( j)
k1
, Z
k1
_
.
(25)
Suppose we have updated the trial sampling density for at
the end of time (k 1). Based on the sampled parameters

( j)
k
= {
( j)
, A
( j)
,
( j)
w ,
( j)
d
,
( j)
t
,
( j)

} p( | Y
( j)
k1
, Z
k1
) at
time k, we draw samples of the auxiliary index
k
, the dis-
crete state s
k
, and the continuous state x
k
according to (18),
(19), and (20) and update the importance weight using (21).
In (18), (19), (20), and (21), the known system parameter
is replaced by
(
( j)
k
)
k
and the noise covariance matrix Qis sub-
stituted by Q
(
( j)
k
)
= diag(
(
( j)
k
)
d
I,
(
( j)
k
)
t
I,
(
( j)
k
)

I). The location


and velocity are estimated through (17) and the minimum
mean-squared error (MMSE) estimate of the unknown pa-
rameter at time k is given by

k
= (1/W
k
)

m
j=1

( j)
k
w
( j)
k
,
where W
k
=

m
j=1
w
( j)
k
. At the end of time k, we update the
trial sampling density for as follows.
At the end of time k, we update the trial sampling density
for the initial state probability vector as
p
_
| s
( j)
0
_
D
_

1
+
s
( j)
0
1
,
2
+
s
( j)
0
2
, . . . ,
N
+
s
( j)
0
N
_
.
(26)
Given the prior distribution of the ith row a
i
of the tran-
sition probability matrix A at the end of time (k 1), that is,
p(a
i
| , S
( j)
k1
) D(
(k1, j)
i,1
,
(k1, j)
i,2
, . . . ,
(k1, j)
i,N
), at time k,
the trial sampling density for a
i
is updated according to
p
_
a
i
| , S
( j)
k
_
p
_
s
( j)
k
| , S
( j)
k1
, a
i
_
p
_
a
i
| , S
( j)
k1
_
D
_
_
_
(k1, j)
i,1
+
s
( j)
k1
i

s
( j)
k
1
. , .

(k, j)
i,1
,
(k1, j)
i,2
+
s
( j)
k1
i

s
( j)
k
2
. , .

(k, j)
i,2
, . . . ,

(k1, j)
i,N
+
s
( j)
k1
i

s
( j)
k
N
. , .

(k, j)
i,N
_
_
_.
(27)
Adaptive Mobile Positioning in WCDMA Networks 349
And given the noise variance sampling density at time (k1),
p(
w
| X
( j)
k1
, Z
k1
)
2
(
k1,w
,
( j)
k1,w
), at time k, the trial
sampling density for
w
is updated according to
p
_

w
| Y
( j)
k
, Z
k
_
p
_
x
k
| x
( j)
k1
, s
( j)
k
,
w
_
p
_

w
| X
( j)
k1
, Z
k1
_

2
_

k1
+ 1,
( j)
k,w
_
,
(28)
where
( j)
k,w
= (
0,w
+ k 1)/(
0,w
+ k)
( j)
k1,w
+

2
i=1
(x
k,3i

x
k1,3i
)
2
/2(
0,w
+ k). Similarly, we have
p
_

d
| Y
( j)
k
, Z
k
_

2
_

k1,d
+ 1,
( j)
k,d
_
, (29)
p
_

t
| Y
( j)
k
, Z
k
_

2
_

k1,t
+ 1,
( j)
k,t
_
, (30)
p
_

| Y
( j)
k
, Z
k
_

2
_

k1,
+ 1,
( j)
k,
_
, (31)
where
( j)
k,d
= (
0,d
+ k 1)/(
0,d
+ k)
( j)
k1,d
+

3
i=1
(p
i,k

h
i
(x
( j)
k
))
2
/3(
0,d
+ k),
( j)
k,t
= (
0,t
+ k 1)/(
0,t
+ k)
( j)
k1,t
+

2
i=1
(
i,k
h
i+3
(x
( j)
k
))
2
/2(
0,t
+ k) and
( j)
k,
= (
0,
+ k
1)/(
0,
+ k)
( j)
k1,
+

3
i=1
(
k,i
h
i+5
(x
( j)
k
))
2
/3(
0,
+ k). Fi-
nally, we summarize the adaptive mobile positioning algo-
rithm with unknown system parameters Algorithm 4.
Complexity
Compared with the known parameter case, that is,
Algorithm 3, the additional computation in Algorithm 4 is
introduced by the updates of the trial densities of the un-
knowns and the draws of these parameters, which at it-
eration, involve 4m simple multiplications, as well as the
m(N + 1) samplings from the Dirichlet distribution and 4m
samplings from the inverse chi-square distribution. As noted
previously, since in mobility tracking applications the up-
date is performed at a time scale of several hundred symbols,
the above SMC-based tracking algorithm is feasible to imple-
ment in practice.
5. SIMULATION
Computer simulations are performed on a WCDMA
hexagon cellular network to assess the performance of the
proposed adaptive mobile positioning algorithms. The net-
work under investigation contains 64 BSs with cell radius
2 km. The mobile trajectories within the network are gener-
ated randomly according to the mobility model described in
Section 2.1 and xed for all simulations. On the other hand,
the pilot signals are generated randomly according to the ob-
servation model (5) for each simulation realization. Some
parameters used in the simulations are the sampling interval
t = 0.5 seconds; the correlation coecient of the random
accelerator in (3) is = 0.6; the variance of each random
variable in w
k
is
w
= 1; the standard deviation of lognor-
mal shadowing

d
= 5 dB. We consider two scenarios. In
scenario 1, the standard deviation of AOA

= 4/360, the
(I) Initialization: for j = 1, . . . , m, draw the samples of the
initial probability vector , the ith row a
i
of the
transition probability matrix, the noise variance
w
,
d
,

t
, and

according to their prior distributions in (22)


and (23), respectively. Draw the state vector x
( j)
0
from
the multivariate Gaussian distribution N(x
0
, 10I), and
draw s
( j)
0
uniformly fromS, all importance weights are
initialized as w
( j)
0
= 1.
(II) For k = 1, 2, . . .,
(a) for j = 1, 2, . . . , m, calculate the trial sampling
density for the auxiliary index according to (18),
where the actually unknown parameter is
replaced by
( j)
k1
,
(b) for j = 1, 2, . . . , m,
(i) draw an auxiliary index
( j)
k
with the
probability q(
k
= j),
(ii) draw a sample s
( j)
k
according to (19),
(iii) draw a sample x
( j)
k
according to (20),
(iv) update the importance weights w
( j)
k
according to (21),
(v) append y
( j)
k
= {x
( j)
k
, s
( j)
k
} and Y
(
( j)
k
)
k1
to form
Y
( j)
k
= {Y
(
( j)
k
)
k1
, y
( j)
k
},
(vi) update the trial sampling density for
according to (26), (28), (29), (30), and (31),
(vii) sample the unknown system parameters

( j)
k
= (
( j)
, A
( j)
,
( j)
w ,
( j)
d
,
( j)
t
,
( j)

) according
to (26), (28), (29), (30), and (31),
respectively.
Algorithm 4: Adaptive mobile positioning algorithm with un-
known system parameters.
standard deviation of TDOA

t
= 100/c; whereas in sce-
nario 2, the standard deviation of AOA

= 2/360, the
standard deviation of TDOA

t
= 50/c; where c = 3 10
8
m/s is the speed of light. In both scenarios, the base station
transmission power p
0,i
= 90 mW, the path-loss index = 3,
and the number of samples m = 250. All simulation results
are obtained based on M = 50 random realizations.
5.1. Performance comparison with existing techniques
We rst compare the performance of the extended Kalman
lter (EKF) mobility tracker [2], the standard particle l-
ter mobility tracker [3], and the proposed auxiliary parti-
cle lter (APF) mobility tracker (Algorithm 3) in terms of
the normalized mean-squared error (NMSE) assuming that
the system parameters are known. The NMSE is dened as
NMSE = (1/L)

L
k=1
(( x
k
x
k
)
2
+( y
k
y
k
)
2
)/(x
2
k
+y
2
k
), where
L is the observation window size. The NMSE results based
on the dierent observations (i.e., RSS only, RSS/AOA and
RSS/AOA/TDOA) for scenarios 1 and 2 are reported in Tables
1 and 2, respectively. It is seen that both the standard PF and
the APF signicantly outperform the EKF in the above two
scenarios under the same observations. In fact, the perfor-
mance gain varies from 510 dB for dierent scenarios and
observations. Moreover, by utilizing the current observations
in the trial sampling density, the APF demonstrates further
improvement over the standard PF (roughly 3 dB).
350 EURASIP Journal on Wireless Communications and Networking
Table 1: Performance comparisons between EKF, standard PF, and APF in terms of NMSE based on dierent observations for scenario 1.
Mobility tracker RSS RSS/AOA RSS/AOA/TDOA
EKF (known) 27.24 dB 29.47 dB 31.62 dB
Standard PF (known) 33.47 dB 41.75 dB 48.64 dB
APF (known) 36.63 dB 44.87 dB 52.21 dB
Standard PF (unknown) 31.47 dB 39.88 dB 45.17 dB
APF (unknown) 34.92 dB 43.33 dB 49.18 dB
Table 2: Performance comparisons between EKF, standard PF, and APF in terms of NMSE based on dierent observations for scenario 2.
Mobility tracker RSS RSS/AOA RSS/AOA/TDOA
EKF (known) 27.24 dB 31.01 dB 33.95 dB
Standard PF (known) 33.47 dB 43.51 dB 51.37 dB
APF (known) 36.63 dB 46.72 dB 55.37 dB
Standard PF (unknown) 31.47 dB 41.96 dB 47.71 dB
APF (unknown) 34.92 dB 45.21 dB 52.79 dB
True
Estimated RSS
Estimated RSS/AOA
Estimated RSS/TDOA/AOA
3000 3500 4000 4500 5000 5500 6000 6500 7000 7500
X
5000
5200
5400
5600
5800
6000
6200
6400
6600
Y
Figure 1: Estimated trajectories based on dierent observations for
scenario 1.
We also compare the APF mobility tracker (Algorithm 4)
with the standard PF mobility tracker assuming that the sys-
tem parameters are unknown. The NMSE results for scenar-
ios 1 and 2 are reported in Tables 1 and 2, respectively. It is
seen that performance penalty due to unknown system pa-
rameters is less than 3 dB whereas the APF is still 3-4 dB bet-
ter than the standard PF.
5.2. Tracking performance of the proposed algorithm
In Figure 1, we compare the trajectories estimated by the
APF algorithm (Algorithm 4) based on RSS only, RSS/AOA,
and RSS/AOA/TDOA, respectively for scenario 1. It is seen
that the online estimation algorithm based on the combined
observations RSS/AOA/TDOA achieves the best perfor-
RSS
RSS/AOA
RSS/TDOA/AOA
0 50 100 150 200 250 300
t
0
50
100
150
200
250
300
350
400
R
S
E
(
m
)
Figure 2: The root-squared error as a function of time for dierent
mobile positioning schemes for scenario 1.
mance and the one based on RSS only performs the
worst. We report the corresponding root-squared error
(RSE) as a function of time for scenarios 1 and 2 in
Figures 2 and 3, respectively. RSE is dened as RSE =
_
(( x
k
x
k
)
2
+ ( y
k
y
k
)
2
). It is observed that by incorporat-
ing the AOA measurements into the observation function,
the RSE is signicantly reduced. Further RSE reduction is
achieved by using additional TDOA measurements. Figures
4 and 5 show the empirical cumulative distribution function
(CDF) of root-squared error (RSE) based on dierent ob-
servations (i.e., RSS only, RSS/AOA, and RSS/TDOA/AOA)
measurements. It is seen that the estimated location based
on RSS only is most likely to have large deviation from the
Adaptive Mobile Positioning in WCDMA Networks 351
RSS
RSS/AOA
RSS/TDOA/AOA
0 50 100 150 200 250 300
t
0
50
100
150
200
250
300
350
R
S
E
(
m
)
Figure 3: The root-squared error as a function of time for dierent
mobile positioning schemes for scenario 2.
RSS
RSS/AOA
RSS/TDOA/AOA
0 200 400 600 800 1000 1200
Root-squared error (m)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
C
D
F
Figure 4: CDF of root-squared error based on dierent mobile po-
sitioning schemes for scenario 1.
actual location whereas that based on RSS/TDOA/AOA has
the smallest outage probability. By comparing the estimation
performance in scenarios 1 and 2, it is seen that the algorithm
achieves better performance for scenario 2 due to the smaller
measurement noise.
We also report the eect of the variance of TDOA
measurement on the estimation performance in Figure 6
in terms of root mean-squared error (RMSE) dened as
RMSE =
_
(1/L)

L
k=1
(( x
k
x
k
)
2
+ ( y
k
y
k
)
2
). It is seen that
the RMSE with RSS/TDOA/AOA monotonically increases
RSS
RSS/AOA
RSS/TDOA/AOA
0 200 400 600 800 1000 1200 1400
Root-squared error (m)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
C
D
F
Figure 5: CDF of root-squared error based on dierent mobile po-
sitioning schemes for scenario 2.
RSS/AOA, scenario 1
RSS/TDOA/AOA, scenario 1
RSS/AOA, scenario 2
RSS/AOA, scenario 2
10 20 30 40 50 60 70 80 90 100

t
5
10
15
20
25
30
35
40
45
50
R
M
S
E
(
m
)
Figure 6: RMSE as a function of
t

t
in Algorithm 4 using
dierent observations.
in both scenarios and the performance gain over that
of RSS/AOA diminishes as the variance of TDOA mea-
surements increases. When the TDOA measurement noise
variance is small, a large performance improvement by the
TDOA/AOA is achieved. However, when the AOA measure-
ment error increases above a certain level, the performance
improvements become negligible. The RMSE in scenario 2
is smaller than that in scenario 1 in both RSS/AOA and
RSS/TDOA/AOA location because of a better accuracy in
AOA and TDOA measurements.
352 EURASIP Journal on Wireless Communications and Networking
0 50 100 150 200 250 300 350 400 450 500
0
0.2
0.4
0.6
0.8
a
1
,
1
Iteration no.
0 50 100 150 200 250 300 350 400 450 500
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
a
2
,
3
Iteration no.
Figure 7: Parameter tracking performance of the transition proba-
bility matrix A as a function of the iteration number for scenario 1.
0 50 100 150 200 250 300 350 400 450 500
0
0.5
1
1.5
2

w
,
1
Iteration no.
0 50 100 150 200 250 300 350 400 450 500
0
0.2
0.4
0.6
0.8
1
1.2
1.4

w
,
2
Iteration no.
Figure 8: Parameter tracking performance of the motion variance

w
as a function of the iteration number for scenario 1.
We next illustrate the parameter tracking behavior of the
proposed adaptive mobile positioning algorithm with un-
known parameters in scenario 1. The estimates of the pa-
rameters a
1,1
and a
2,3
as a function of time index k for one
vehicle trajectory are plotted in Figure 7. We also plot the es-
timates of the noise variances
w,1
and
w,2
in Figure 8. It is
observed that although the initial estimates of the unknown
parameters are far from the actual value, after a short period
of time, the estimates of these unknown parameters converge
to the true values, demonstrating the excellent tracking per-
formance of the proposed algorithm.
6. CONCLUSIONS
We have considered the problem of mobile user position-
ing under the sequential Monte-Carlo Bayesian framework.
We have developed a new adaptive mobile positioning algo-
rithm based on the auxiliary particle lter algorithm. The al-
gorithm makes use of the measurements of time dierence of
arrival, angle of arrival as well as received signal strength, all
of which are available in practical WCDMA networks. The
proposed algorithm jointly tracks the unknown system pa-
rameters as well as the mobile position and velocity. Simu-
lation results show that the proposed algorithm has an ex-
cellent mobility tracking and parameter estimation perfor-
mance and it signicantly outperforms the existing mobility
estimation schemes.
ACKNOWLEDGMENTS
This work was supported in part by the US National Science
Foundation (NSF) under Grants DMS-0225692 and CCR-
0225826, and by the US Oce of Naval Research (ONR) un-
der Grant N00014-03-1-0039.
REFERENCES
[1] J. Caery and G. L. Stuber, Subscriber location in CDMA
cellular networks, IEEE Trans. Veh. Technol., vol. 47, no. 2,
pp. 406416, 1998.
[2] T. Liu, P. Bahl, and I. Chlamtac, Mobility modeling, loca-
tion tracking, and trajectory prediction in wireless ATM net-
works, IEEE J. Select. Areas Commun., vol. 16, no. 6, pp. 922
936, 1998.
[3] Z. Yang and X. Wang, Joint mobility tracking and hando in
cellular networks via sequential Monte Carlo ltering, IEEE
Trans. Signal Processing, vol. 51, no. 1, pp. 269281, 2003.
[4] L. Cong and W. Zhuang, Hybrid TDOA/AOAmobile user lo-
cation for wideband CDMA cellular systems, IEEE Transac-
tions on Wireless Communications, vol. 1, no. 3, pp. 439447,
2002.
[5] C.-D. Wann and Y.-M. Chen, Mobile location tracking with
velocity estimation, in Proc. 5th International Conference on
Intelligent Transportation Systems (ITS 02), pp. 566571, Sin-
gapore, Singapore, September 2002.
[6] J. H. Reed, K. J. Krizman, B. D. Woerner, and T. S. Rappaport,
An overview of the challenges and progress in meeting the E-
911 requirement for location service, IEEE Commun. Mag.,
vol. 36, no. 4, pp. 3037, 1998.
[7] M. Hellebrandt and R. Mathar, Location tracking of mobiles
in cellular radio networks, IEEE Trans. Veh. Technol., vol. 48,
no. 5, pp. 15581562, 1999.
[8] F. Gustafsson, F. Gunnarsson, N. Bergman, et al., Particle
lters for positioning, navigation, and tracking, IEEE Trans.
Signal Processing, vol. 50, no. 2, pp. 425437, 2002.
[9] M. McGuire and K. N. Plataniotis, Dynamic model-based l-
tering for mobile terminal location estimation, IEEE Trans.
Veh. Technol., vol. 52, no. 4, pp. 10121031, 2003.
[10] R. Togneri and L. Deng, Joint state and parameter estimation
for a target-directed nonlinear dynamic system model, IEEE
Trans. Signal Processing, vol. 51, no. 12, pp. 30613070, 2003.
[11] P. M. Djuric, J. H. Kotecha, J. Zhang, et al., Particle ltering,
IEEE Signal Processing Mag., vol. 20, no. 5, pp. 1938, 2003.
[12] A. Doucet, N. de Freitas, and N. Gordon, Sequential Monte
Carlo Method in Practice, Springer-Verlag, New York, NY,
USA, 2000.
Adaptive Mobile Positioning in WCDMA Networks 353
[13] R. E. Ziermer and R. L. Peterson, Digital Communications and
Spread Spectrum Systems, Macmillan, New York, NY, USA,
1985.
[14] A. J. Viterbi, CDMA: Principles of Spread Spectrum Commu-
nications, Addison-Wesley, Reading, Mass, USA, 4th edition,
1999.
[15] H. Krim and M. Viberg, Two decades of array signal process-
ing research: the parametric approach, IEEE Signal Processing
Mag., vol. 13, no. 4, pp. 6794, 1996.
[16] T. S. Rappaport, Wireless Communications: Principles and
Practice, Prentice-Hall, New York, NY, USA, 1996.
[17] X. Wang, R. Chen, and J. S. Liu, Monte Carlo Bayesian signal
processing for wireless communications, J. VLSI Signal Pro-
cessing, vol. 30, no. 13, pp. 89105, 2002.
[18] A. Doucet, S. J. Godsill, and C. Andrieu, On sequential
Monte Carlo sampling methods for Bayesian ltering, Statis-
tics and Computing, vol. 10, no. 3, pp. 197208, 2001.
[19] M. Davy, C. Andrieu, and A. Doucet, Ecient particle l-
tering for jump Markov systems. Application to time-varying
autoregressions, IEEE Trans. Signal Processing, vol. 51, no. 7,
pp. 17621770, 2003.
[20] M. Pitt and N. Shepard, Filtering via simulation: auxiliary
particle lters, Journal of the American Statistical Association,
vol. 94, no. 446, pp. 590599, 1999.
B. Dong received his M.S. degree in electrical engineering from
Queens University, Canada.
Xiaodong Wang received the B.S. degree
in electrical engineering and applied math-
ematics (with the highest honor) from
Shanghai Jiao Tong University, Shanghai,
China, in 1992; the M.S. degree in electri-
cal and computer engineering from Purdue
University in 1995; and the Ph.D. degree in
electrical engineering from Princeton Uni-
versity in 1998. From July 1998 to Decem-
ber 2001, he was an Assistant Professor in
the Department of Electrical Engineering, Texas A&M University.
In January 2002, he joined the faculty of the Department of Electri-
cal Engineering, Columbia University. Dr. Wangs research interests
fall in the general areas of computing, signal processing, and com-
munications. He has worked in the areas of digital communica-
tions, digital signal processing, parallel and distributed computing,
nanoelectronics and bioinformatics, and has published extensively
in these areas. Among his publications is a recent book entitled
Wireless Communication Systems: Advanced Techniques for Signal
Reception, published by Prentice Hall, Upper Saddle River, in 2003.
His current research interests include wireless communications,
Monte-Carlo-based statistical signal processing, and genomic sig-
nal processing. Dr. Wang received the 1999 NSF CAREER Award,
and the 2001 IEEE Communications Society and Information The-
ory Society Joint Paper Award. He currently serves as an Associate
Editor for the IEEE Transactions on Communications, the IEEE
Transactions on Wireless Communications, the IEEE Transactions
on Signal Processing, and the IEEE Transactions on Information
Theory.
EURASIP Journal on Wireless Communications and Networking 2005:3, 354363
c 2005 Bruce E. Carey-Smith et al.
Flexible Frequency Discrimination Subsystems
for Recongurable Radio Front Ends
Bruce E. Carey-Smith
Centre for Communications Research, University of Bristol, Bristol BS8 1UB, UK
Email: b.carey-smith@bristol.ac.uk
Paul A. Warr
Centre for Communications Research, University of Bristol, Bristol BS8 1UB, UK
Email: paul.a.warr@bristol.ac.uk
Phill R. Rogers
Centre for Communications Research, University of Bristol, Bristol BS8 1UB, UK
Email: phill.rogers@bristol.ac.uk
Mark A. Beach
Centre for Communications Research, University of Bristol, Bristol BS8 1UB, UK
Email: m.a.beach@bristol.ac.uk
Geoffrey S. Hilton
Centre for Communications Research, University of Bristol, Bristol BS8 1UB, UK
Email: geo.hilton@bristol.ac.uk
Received 8 October 2004; Revised 14 March 2005
The required exibility of the software-dened radio front end may currently be met with better overall performance by employing
tunable narrowband circuits rather than pursuing a truly wideband approach. A key component of narrowband transceivers is
appropriate ltering to reduce spurious spectral content in the transmitter and limit out-of-band interference in the receiver.
In this paper, recent advances in exible, frequency-selective, circuit components applicable to recongurable SDR front ends
are reviewed. The paper contains discussion regarding the ltering requirements in the SDR context and the use of intelligent,
adaptive control to provide environment-aware frequency discrimination. Wide tuning-range frequency-selective circuit elements
are surveyed including bandpass and bandstop lters and narrowband tunable antennas. The suitability of these elements to the
mobile wireless SDR environment is discussed.
Keywords andphrases: software-dened radio, tunable antenna, recongurable front end, MEMS, tunable bandpass lter, tunable
bandstop lter.
1. INTRODUCTION
The intention of software-dened radio (SDR) is to pro-
vide a exible radio platform capable of operating over a
continuously evolving set of communications standards and
modes. In contrast to the majority of currently available
This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
mobile telephones, which are predened to operate on a xed
number of standards, SDR must be capable of adapting to
both current and future mobile telecommunications stan-
dards [1]. The SDR concept imposes demanding require-
ments on the transceiver front end.
1
Current spectrum al-
locations and recent regulatory reforms (e.g., [2]) suggest
1
The front end is that part of the radio which performs channelisation
and up- and down-conversion. These functions may be performed in either
the analogue or digital domain.
Flexible Subsystems for Radio Front Ends 355
that transceiver operation from 600 MHz to 6 GHz will be
required to cover existing and emerging telecommunication
services.
2
The front end must not only be capable of oper-
ating across this entire band but must do so whilst attaining
performance comparable with xed frequency solutions. In
order to achieve the wide frequency coverage required from
the SDR front end, two approaches are possible: wideband
operation,
3
with all ltering being carried out in the dig-
ital domain; or exible narrowband operation, using tun-
able narrowband analogue circuits. In this paper, it is argued
that, when compared to wideband solutions, exible narrow-
band operation holds signicant advantages, and with cur-
rent technology would oer improved overall performance
in the SDR transceiver. A key component of narrowband
transceivers is appropriate ltering to reduce spurious spec-
tral content in the transmitter and limit out-of-band interfer-
ence in the receiver. The main focus of this paper is to review
techniques for providing tunable frequency discrimination
in the analogue front end of a narrowband recongurable
SDR receiver.
1.1. Practical SDR front ends
Practical implementations of the SDR front end must cur-
rently include analogue circuitry to perform frequency up-
and down-conversion, amplication, and ltering. In the
ideal implementation of SDR, maximum front-end exibil-
ity is achieved by the conversion of the analogue signals to
and from the digital domain as close to the antenna as pos-
sible [1] employing direct-digital conversion. However, per-
forming direct-digital conversion at RF and microwave fre-
quencies places currently unachievable demands on the con-
version hardware and digital processing circuitry. Commer-
cially available analogue-to-digital converters (ADCs) can
achieve analogue bandwidths in the order of 70 MHz with
greater than 90 dB spurious-free dynamic range (SFDR)
4
but
at the present rate of advance it will be some time before
direct-digital conversion is possible for the wireless commu-
nications environment [3]. The alternative to directly sam-
pling the RF signal is to retain the analogue-digital divide
at an intermediate frequency, away from the antenna, and
to realise the necessary front-end functionality using ana-
logue circuitry. Wideband analogue transceiver front ends
are being explored at both the component and system level
[4, 5].
1.2. Limitations of wideband front ends
Although having attractive features, a wideband front end
may not necessarily be the optimum solution for SDR. The
2
A more conservative goal, encompassing the majority of current stan-
dards, still requires a two-octave frequency coverage from GSM at 800 MHz
to the lower ISM band at 2.4 GHz.
3
For the purposes of this paper, the term wideband is used to describe
components whose operational bandwidth (frequency range over which
they are impedance matched) covers all the communication standards of
interest; multioctave for SDR.
4
For example, analog devices, AD6645.
major advantage of the wideband approach is the wide band-
width at the analogue-to-digital converter interface, allowing
concurrent operation at multiple frequencies. Both the re-
ceiver and transmitter are thus able to operate on multiple
standards simultaneously. However, design for wideband op-
eration leads to compromise in the transceiver performance,
caused by limitations in the front-end components. The non-
linearity of the active analogue front-end components lim-
its the distortion-free dynamic range of both the transmitter
and receiver. Wideband linearisation techniques for ampli-
ers and mixers have demonstrated some linearity improve-
ment [6, 7] and are necessary to limit in-band intermodu-
lation distortion. However, linearity improvements come at
a signicant cost to overall radio eciency. Alternatively, by
band-limiting the signal of interest, unwanted spectral prod-
ucts can be suppressed, leading to a signicant reduction in
the overall linearity requirements of the transceiver front end
[8].
Band limiting within the SDR environment must be ex-
ible since both the signals of interest and the unwanted dis-
tortion products and interference will vary as a function of
the users environment and choice of standard. In this pa-
per, techniques of introducing exible frequency discrimina-
tion into the transceiver front end are considered. Section 2
examines the dierent requirements for ltering within the
SDR transmitter and receiver. Tunable bandpass and band-
stop lters are discussed in Sections 3.1 and 3.2, respectively.
Bandpass lters oer general frequency discrimination to re-
duce the wideband interference level while bandstop lters
are particularly suited to removing high-level interferers and
providing selective isolation between the transmit and re-
ceive bands. Tunable, narrowband antennas are not normally
associated with frequency discrimination; however, they are
tuned transducers between guided and unguided electro-
magnetic waves and exhibit ltering behaviour. Narrowband
tunable antennas oer signicant advantages over wideband
antennas, one of which is the additional frequency discrim-
ination which they oer. These advantages are discussed in
Section 3.3.
2. FRONT-ENDREQUIREMENTS OF SDR
The design of the air interface of narrowband transceivers
concentrates on achieving compliance with a particular
standards receiver blocking mask and transmitter emission
mask. The latter must be met in order to comply with the
requirements of the current operating standard while the
former provides a guide as to the maximum signal levels
that will be encountered by the receiver. In conventional
(xed standard) radio, xed lters provide the necessary fre-
quency discrimination. An example of a typical transceiver
is shown in Figure 1. Some compromise in the performance
of individual ltering elements must be made to introduce
the exibility required for SDR. It may be possible, how-
ever, to recover some of this through combining their ca-
pabilities in an intelligent way. This idea is explored in
Section 2.2.
356 EURASIP Journal on Wireless Communications and Networking
Receiver
Transmitter
Band-select
lter
Image-reject
lter
LNA
Duplex
lter
Isolator
Harmonic
rejection lter
PA
Band-select
lter Preamplier
Mixer
Local
oscillator
Local
oscillator
Mixer
Frequency
down-conversion
Frequency
up-conversion
Figure 1: Typical transceiver front end.
2.1. Frequency discrimination requirements in the
transmitter and receiver
The high power amplication of transmitted signals leads
to harmonic (HD) and intermodulation distortion (IMD)
products which contaminate the spectrum of the emitted
signal [9]. Some IMD products appear in-band and can-
not be ltered out, placing fundamental linearity require-
ments on the power amplier (PA). Unnecessary IMD can
be avoided by harmonic ltering before the PA. This prevents
harmonic energy generated in the up-conversion and pream-
plication stages from producing additional IMD in the PA.
It also reduces the wideband noise requirements of these
stages. Harmonic ltering after the PA is usually manda-
tory in order to meet the emission mask for any given stan-
dard. Being well removed in frequency from the wanted sig-
nal, harmonic attenuation is usually straightforward using a
lowpass or bandpass lter. However, the multioctave air in-
terface of SDR demands that this ltering be tunable. The
greatest challenge for tunable lters in the transmitter, par-
ticularly post-PA, is in tolerating high power levels. There are
two factors to consider: the linearity of the tuning elements
and their absolute power handling capability. Both of these
factors must be considered when selecting the tuning mech-
anism for transmit lters.
In the receiver, the requirement for ltering is usually
more stringent. Signals in adjacent channels are controlled by
the current standard and can therefore be tolerated, assum-
ing a degree of linearity in the active front-end components.
However, out-of-band signals are not controlled; the only
guarantee is that they will fall below some specied maxi-
mum level, dened by the blocking mask. The required re-
jection is usually achieved using a bandpass lter.
In frequency division duplex (FDD) systems, the trans-
mitter can interfere detrimentally with the receiver since both
operate simultaneously. This can occur in two ways. Firstly,
the transmitter noise oor is generally much higher than
the sensitivity of the receiver and, given inadequate isolation,
the receiver will be desensitised. Secondly, leakage of the high
power transmit signal into the receiver can cause receiver
overload. In both cases, the SDR front end must provide
some form of exible frequency discrimination in the trans-
mit and receive paths. The use of separate, independently
tuned, transmit and receive narrowband antennas will lead
to some improvement in the isolation. Tunable notch lters
could be used to provide additional suppression.
2.2. Environment-aware frequency discrimination
For any given operating standard, the SDR transceiver must
select an appropriate ltering prole. These proles could
be predened for dierent standards, however, exibility
in the front-end circuitry raises the possibility of intelli-
gent, real-time adaptation of the transceivers frequency dis-
crimination prole to the current operating environment.
By constantly assessing the users signal environment, the
transceiver can respond with the appropriate level of sup-
pression on a frequency-by-frequency basis. The potential
advantage of this scheme is that the necessary ltering can be
provided using lower specication circuit blocks. Through
the intelligent and exible combination of lower perfor-
mance elements, ltering can be supplied where it is needed
rather than by unnecessary blanket coverage.
This concept has greater applicability to the receiver,
where the nature of the unwanted signals is changing in an
unpredictable way. The customary means of dealing with this
unpredictability is to provide a maximum amount of lter-
ing to match the blocking mask for a particular standard. The
blocking mask, however, assumes the worst case, where all
out-of-band signals are at the maximum interference level
(usually 0 dBm). In reality, troublesome interference will be
limited to specic frequency bands which will change as a
function of the environment. Wideband spectrum measure-
ments at a variety of high-communication trac locations
in a typical European city indicate that seldom do multiple
signals approach the typical wideband blocking specication
Flexible Subsystems for Radio Front Ends 357
0
10
20
30
40
50
60
70
80
90
100
R
e
c
e
i
v
e
d
s
i
g
n
a
l
p
o
w
e
r
(
d
B
m
)
4 8 12 16 20 24 28 32 36 40 44 48 52 56 60
10
2
Frequency (MHz)
Figure 2: Worst-case power prole.
simultaneously [10]. The worst-case power prole, com-
bined from measurements taken in 44 dierent locations, is
shown in Figure 2 where a zero dB gain, omnidirectional an-
tenna is assumed in order to calculate the absolute power lev-
els. Few signals exceed 20 dBm at the input to the receiver
and the majority of the spectrum is below 40 dBm.
Furthermore, at any given location the probability of re-
ceiving a number of high-level interfering signals reduces
quickly with their average signal strengths (Figure 3). For ex-
ample, the probability of receiving more than two interfering
signals above 25 dBm is less than 0.4%. The data used in
this analysis is the long-term cumulative maximum, so even
lower instantaneous probabilities can be expected. These re-
sults indicate that, by providing a high level of narrowband
suppression at a small number of frequencies, the wideband
interference level can be reduced signicantly from the typ-
ical receiver blocking specication. This, in turn, eases the
general ltering requirements, reducing the absolute stop-
band attenuation required of a front-end bandpass lter. This
may enable the use of a lower order lter or a more compact
implementation. The remaining sections of this paper look
at the performance of various narrowband tunable front-end
components, able to provide exible frequency discrimina-
tion.
3. FLEXIBLE FREQUENCY FRONT-ENDSUBSYSTEMS
3.1. Tunable bandpass lters
All practical transceivers employ bandpass ltering to pro-
vide wideband suppression of unwanted interference and
distortion products. In an SDR, this element must be exi-
ble. There are a number of lter tuning technologies appli-
cable for use in mobile telecommunication. Varactor-tuned
lters have been widely used due to their fast tuning speeds
and octave-frequency tuning range. However, their consider-
able loss and poor linearity have limited their use to IF sec-
tions of the transceiver. The most promising emerging tech-
nologies for RF are tunable ferroelectric lms and RF micro-
electromechanical systems (MEMS). Both technologies re-
sult in lters which have fast tuning speeds, are small, intro-
duce little distortion, and consume minimal power. Ferro-
electric lms such as BSTconsist of a thin lmof ferroelectric
1
0.1
0.01
0.001
50 45 40 35 30 25 20 15
Signal level L
P
r
o
b
a
b
i
l
i
t
y
N = 1
N = 2
N = 3
Figure 3: Results of spectrum measurements taken in a typical Eu-
ropean city. The graph shows the probability that there will be more
than N signals whose power is greater than the signal level L.
material placed on a nonferrous substrate, creating a two-
layered structure. The dielectric permittivity of the struc-
ture can be varied by applied external electric eld. Currently
these materials exhibit high loss and require large bias volt-
ages, however, greater than 30% tuning range has been doc-
umented and resonator Qs greater than 800 can be obtained
[11, 12].
The introduction and continuing improvement of mi-
croelectromechanical systems (MEMS) has led to the possi-
bility of MEMS-tuned lters for applications where size and
tuning speed are critical. Recently, considerable attention has
been given to their use at high frequencies. MEMS for RF
(RF MEMS) and microwave applications have been fabri-
cated and characterised for frequencies up to 40 GHz [13].
MEMS-tuned microwave lter design, although being a rel-
atively new research area is attracting considerable interest.
The majority of lumped-element MEMS-tuned lter designs
use xed air-core inductors and achieve tuning via adjustable
MEMS capacitors and are generally limited to frequencies be-
low 3 GHz. A group at Raytheon have developed a number
of MEMS-tuned lumped element lters using MEMS digi-
tal capacitor arrays to give continuous tuning in frequency
bands from 70 MHz to 2.8 GHz [14]. An octave tuning range
lumped element MEMS lter with concurrent bandwidth
tuning from 7 to 42% is presented in [15]. This is impres-
sive performance although the 4-bit MEMS capacitor arrays
lead to uneven coverage across the tuning band.
Numerous planar distributed designs have also been
published, for example, [16]. The majority of these designs
use some form of tunable capacitive loading to alter the res-
onant frequency of transmission-line resonators. The tun-
ing ranges tend to be more modest due to the reduction in
resonator Q as the capacitance is increased at lower tuning
frequencies. An alternative is to directly adjust the length of
the resonators. The use of MEMS switches to connect addi-
tional lengths of transmission-line to a hairpin-line lter has
358 EURASIP Journal on Wireless Communications and Networking
Hairpin resonator RF MEMS switch
MM 10
(a)
0
20
40
60
80
0 1 2 3 4
Frequency (GHz)
I
n
s
e
r
t
i
o
n
l
o
s
s
(
d
B
)
MEMS switches o
MEMS switches on
(b)
Figure 4: (a) Filter layout and (b) measured performance of a
MEMS-tuned coupled hair-pin lter.
been proposed [17]. Measured results show a tuning range
from 2.05 GHz to 2.25 GHz with a constant percentage band-
width of 4.5%. The lter layout and performance are shown
in Figure 4.
A limitation of distributed lter designs is that it is dif-
cult to alter the interresonator coupling and, because this
coupling dictates the overall lter Q, this leads to diculty
in bandwidth tuning. This limitation is overcome by us-
ing capacitively loaded dual-behaviour resonators which al-
low for independent control of the attenuation poles of the
lter [18]. The reported frequency tuning range is some-
what limited however, due to the frequency invariance of the
impedance inverters.
A wide-range, multioctave centre-frequency tunable l-
ter with concurrent bandwidth tuning capability has been
reported by the authors [19]. This design, suitable for
use with MEMS bistable contact and capacitive switches,
utilises switches and switched capacitors distributed at in-
tervals along pairs of coupled transmission lines (Figure 5)
to achieve both bandwidth and centre-frequency tuning. Be-
ing discrete in nature, there are a nite number of tuning
points. However, the distributed topology means that the
tuning range and resolution are limited only by the place-
ment density and electrical size of the resonators. Illustrative
performance is given in Figure 6.
l
s
l
s
l
s
C
c
C
c
C
c
Y
0e
, Y
0o
, v
pe
, v
po
1 2 n
(a)
LDCL pairs
Lumped
capacitors
(b)
Figure 5: (a) Schematic of a pair of lumped-distributed coupled
lines (LDCLs). (b) Partially cropped image of 3-element lter with
four pairs of LDCLs.
3.2. Tunable bandstop lters
Bandstop lters have found application in base-station in-
stallations where substantial rejection is required over a lim-
ited bandwidth to avoid cosite interference. They are prefer-
able to bandpass lters, particularly in the transmitter, due
to their low passband insertion loss and high attenuation in
the stopband. Bandstop lters may nd application in the re-
ceiver path where a high level of suppression is needed over a
limited bandwidth. Their low passband insertion loss means
they can be employed without greatly aecting the receiver
sensitivity.
The majority of tunable bandstop lters use some form
of variable capacitance to tune the electrical length in a ca-
pacitively coupled shunt stub design [20]. With some mod-
ication, these designs can yield tuning ranges of almost
an octave [21], although the use of lumped capacitive ele-
ments produces an asymmetric stopband response and in-
troduces spurious parasitic behaviour at frequencies above
the stopband. Discrete tuning of bandstop lters has re-
cently been demonstrated using MEMS switches to alter the
resonator properties also reaching almost an octave tuning
range [22]. The disadvantage of discrete tuning is that, to af-
fect a high tuning resolution, large numbers of tuning ele-
ments are needed.
Wide-range, continuous tuning can be attained by
employing a composite tuning mechanism, consisting of
varactor-loading and discrete transmission-line length ad-
justment using PINdiode switches [23] as shown in Figure 7.
This composite approach not only extends the relative
Flexible Subsystems for Radio Front Ends 359
0
10
20
30
40
50
60
70
I
n
s
e
r
t
i
o
n
l
o
s
s
(
d
B
)
0
10
20
30
R
e
t
u
r
n
l
o
s
s
(
d
B
)
200 300 400 500 700 1000 2000
Frequency (MHz)
(a)
0
10
20
30
40
50
60
70
I
n
s
e
r
t
i
o
n
l
o
s
s
(
d
B
)
300 350 400 450 500 550 600
Frequency (MHz)
(b)
Figure 6: Measured and modelled performance of LDCL lter at a
selection of tuning points. (a) Measured (solid) and modelled (bro-
ken line) centre-frequency tuning. (b) Measured bandwidth tuning
at a nominal centre frequency of 450 MHz.
centre-frequency tuning range beyond that achievable using
solely reactive loading, but also permits the independent tun-
ing of the pass- and stopbands. Independent placement of
both the pass- and stopbands is advantageous in that it re-
laxes the passband constraints by reducing the range of fre-
quencies over which minimum loss must be maintained. The
region of minimum loss can be tuned to the required fre-
quency and the loss of the lter at other frequencies becomes
less important. Measured results of the lter, showing a two-
octave tuning range and variable, relative passband position,
are given in Figure 8. The use of active semiconductor com-
ponents leads to moderate linearity performance. This may
be alleviated by employing MEMS counterparts throughout
the lter.
Port 1 Port 2
50
/4 at 1.75 GHz
G1
G9
G2 G2
.
.
.
.
.
.
.
.
.
.
.
.
9
0

9
0

Figure 7: Tunable notch lter schematic.


3.3. Narrowband tunable antennas
The performance of an SDR terminal hinges on its interface
with the radio channel: the antenna. Acceptable radiation ef-
ciency must be attained across the complete operational fre-
quency range. In addition, current market expectations place
constraints on the size of the SDR terminal. These require-
ments, wide operational bandwidth and small size, are in-
compatible. Reducing the size of an antenna results in either
its eciency decreasing (which is unacceptable for a termi-
nal antenna) or its bandwidth narrowing, for example, [24].
Conversely, the (instantaneous) operational bandwidth of an
antenna may be widened by increasing its size, for example,
[25].
The current practice is to design a passive antenna with
a wide operational bandwidth, and then conform its shape
to t inside an acceptable volume [5]. Analysis of such ele-
ments is normally based upon input response measurements.
Whilst this may seem a useful metric, it precludes the most
important characteristic of an antenna: its ability to radi-
ate. When radiation patterns are considered, they are usu-
ally simple 2D cuts,
5
for example, [26]; though most show
the trend of pattern degradation as a function of frequency
[27]. Comparison of channel capacity data based on anten-
nas whose only signicant dierence is polarisation and pat-
tern purity suggests that antennas with high polarisation and
pattern purities can yield higher channel capacities relative to
those that have poor polarisations and varying patterns [28].
An alternative to a large passive wideband structure is a
narrowband tunable antenna. Here the antenna can be made
arbitrarily small provided its instantaneous input response
(bandwidth) covers at least one channel of the standard. By
loading the antenna with a variable reactance, be that capac-
itive (e.g., using a varactor diode) or inductive, it is possible
5
Although this may seem a succinct way of characterising an antennas
radiation, it is insucient as it only accounts for one or two planes about
the structure; for an accurate characterisation, full 3D (co- and cross-polar)
patterns should be measured.
360 EURASIP Journal on Wireless Communications and Networking
0
10
20
30
I
n
s
e
r
t
i
o
n
l
o
s
s
(
d
B
)
0 0.5 1 1.5 2 2.5
Frequency (GHz)
6 pF, G9
1.6 pF, G7
0.9 pF, G5
0.6 pF, G4
0.5 pF, G3
(a)
0
5
10
15
20
25
30
I
n
s
e
r
t
i
o
n
l
o
s
s
(
d
B
)
0 0.5 1 1.5 2 2.5 3
Frequency (GHz)
2.7 pF, G1
1.2 pF, G3
0.7 pF, G6
(b)
Figure 8: Measured performance of a compositely tuned bandstop
lter showing (a) centre-frequency tuning and (b) decrease in lter
Qand the associated shift in passband frequency. (Approximate var-
actor capacitance is measured in pF and G denotes ground position
with reference to Figure 7.)
to vary, in a controlled fashion, the resonant frequency of
the antenna. Figure 9 shows the tuning range of an electri-
cally small antenna that is capacitively tuned using a varac-
tor diode. An example of the measured copolarisation pat-
tern for this antenna is shown in Figure 10. Because only
the resonant frequency of a single mode is varied, the radia-
tion characteristics remain relatively constant over the tuning
range, when compared with those of a passive wideband an-
tenna [29]. However, the use of a reactive element will intro-
duce losses (which are a function of bias voltage) and nonlin-
earities (due to the semiconductor junction). Prior research
0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
|
S
1
1
|
/
d
B
1.8 1.9 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8
Frequency/GHz 10
9
0 v
1 v
2 v
3 v
4 v
5 v
6 v
7 v
8 v
9 v
10 v
15 v
Figure 9: Measured S
11
response of the tunable narrowband an-
tenna with variation of varactor bias voltage.
has shown the eciency of a tunable antenna to be com-
parable to that of a passive wideband conformal antenna
[30], and the nonlinearities to be within thresholds of cur-
rent standards [31]. It is envisaged that with the development
of MEMS technologies, tunable devices with lower loss will
emerge.
Eectively, a tunable narrowband antenna is little more
than a tunable lter. Previous work has shown that the use
of such an antenna, whose instantaneous input bandwidth is
optimised for operation on a single channel, oers signicant
interference rejection relative to a passive wideband antenna
[32]. Given the congested nature of the spectrum this lter-
ing prior to the RF front-end is highly desirable.
3.4. Discussion
There remain signicant challenges in the design and pro-
duction of exible components for frequency discrimina-
tion in SDR transceivers. New technologies capable of mul-
tioctave tuning ranges have been demonstrated and emerg-
ing technologies promise lower loss and higher linearity per-
formance. However, for mobile applications, where size is a
primary consideration, the fabrication of such devices must
be addressed. Suitable lters and antennas must be mono-
lithic in order to simplify production and achieve appropri-
ate footprints while at the same time being environmentally
robust. For transmitter applications, power handling capa-
bilities must be addressed. High power RFMEMS-tuned l-
ters capable of 25 W have been investigated at VHF frequen-
cies [33], however, the tuning capabilities of these lters are
limited.
Flexible Subsystems for Radio Front Ends 361
2000 MHz
x
y
z
Figure 10: Typical measured copolarisation pattern of the tunable
narrowband antenna.
From a system implementation perspective, the SDR air
interface requirements will play a key role in determining the
optimum combination of exible ltering subsystems. The
majority of the frequency discrimination requirements for
the transceiver can be determined from known signal infor-
mation. For example, the suppression of transmit noise in
the receive band can be determined from a knowledge of
the transmitter noise prole and the receiver cochannel in-
terference rejection ratio for a given set of up- and down-
link channel parameters. For those requirements which deal
with signal information that is not known beforehand (i.e.,
receiver blocking), some relaxation in front-end ltering re-
quirements may be gained by dening them on a statisti-
cal basis (i.e., number of simultaneous interferers above a
given power level which must be tolerated by the receiver).
In this case, the wideband suppression level of the receiver
may be able to be reduced by employing notch lters to
deal with high-level interferers. Once the complete specica-
tions have been dened, the optimum combination of sub-
systems can be designed, based on their individual perfor-
mance.
A method of intelligent control is a critical prerequisite
for the use of exible subsystems in an SDR front end. In all
cases, it is necessary to have knowledge of the tuning rela-
tionship between the frequency response of the subsystems
and the signals used to control them. However, the major
diculty lies in determining the appropriate frequency pro-
le for each subsystem. Within the wanted channel, centre
frequency and bandwidth information can be used to tune
bandpass lters and antennas for optimumpassband loss and
radiation eciency. The appropriate receiver stopband pro-
les are not known a priori, and the most ecient prole
will assign bandstop lter suppression at the frequency of the
most problematic interference. Two control approaches ex-
ist: blind adaptation, where the lters are tuned to minimise
broadband detected power; and spectrummonitoring, where
a parallel receiver determines the oending signal charac-
teristics. The former suers from search algorithm latency;
complex algorithms are needed due to local minima in the
tuning space. The latter introduces the cost and complexity
of an additional receiver. The requirements of this receiver
are reduced signicantly since the only output required of
the spectrum monitor is the frequency of the largest signal,
with an accuracy relating to the bandwidth of the bandstop
lter.
4. SUMMARY
A wideband radio front end is attractive for SDR, however,
the resulting compromise in performance suggests that a nar-
rowband tunable approach is worthy of consideration. The
essential advantage of narrowband systems is the frequency
discrimination they oer, thereby reducing the linearity and
dynamic range requirements of the transceiver. This paper
has surveyed recent techniques to extend the tuning range
and overall exibility of three key elements of the narrow-
band radio front end; bandpass and bandstop lters and an-
tennas.
Multioctave centre-frequency tuning ranges have been
demonstrated in bandpass and bandstop lters with concur-
rent bandwidth tuning. However, miniaturisation and inte-
gration issues must be addressed to yield solutions suitable
for wireless terminals. Tunable narrowband antennas have
been shown not only to contribute helpful frequency dis-
crimination but also to display far superior performance for
a given size when compared to wide operational bandwidth
designs.
With a selection of tunable ltering components in the
front end, the potential exists for intelligent real-time adap-
tive control of the transceivers frequency discrimination pro-
le. Analysis of radio spectrum measurements suggests that,
for the receiver, a high-level of discrimination is needed only
at a few select frequencies at any given time. An adaptive l-
tering prole may allow the specications of the individual
ltering components to be relaxed.
REFERENCES
[1] J. Mitola, The software radio architecture, IEEE Commun.
Mag., vol. 33, no. 5, pp. 2638, 1995.
[2] M. Cave, Review of Radio Spectrum Management, In-
dependent review prepared for the department of trade
and industry and HM treasury, March 2002, available on
www.ofcom.org.uk .
[3] R. H. Walden, Performance trends for analog to digital con-
verters, IEEE Commun. Mag., vol. 37, no. 2, pp. 96101, 1999.
[4] H. Tsurumi and Y. Suzuki, Broadband RF stage architec-
ture for software-dened radio in handheld terminal appli-
cations, IEEE Commun. Mag., vol. 37, no. 2, pp. 9095,
1999.
[5] Z. D. Liu and P. S. Hall, Dual-band antenna for hand
held portable telephones, Electronics Letters, vol. 32, no. 7,
pp. 609610, 1996.
[6] P. A. Warr, M. A. Beach, and J. P. McGeehan, Gain-element
transfer response control for octave-band feedforward ampli-
ers, Electronics Letters, vol. 37, no. 3, pp. 146147, 2001.
[7] T. Nesimoglu, M. A. Beach, P. A. Warr, and J. R. MacLeod,
Linearised mixer using frequency retranslation, Electronics
Letters, vol. 37, no. 25, pp. 14931494, 2001.
362 EURASIP Journal on Wireless Communications and Networking
[8] J. R. MacLeod, M. A. Beach, P. A. Warr, and T. Nesimoglu,
Filter considerations in the design of a software dened ra-
dio, in Proc. IST Mobile Communications Summit (IST 01),
pp. 363368, Barcelona, Spain, September 2001.
[9] N. Pothecary, Feedforward Linear Power Ampliers, Artech
House, Norwood, Mass, USA, 1999, pp. 4952.
[10] G. T. Watkins, Flexible linearity prole feedforward low noise
ampliers for Software Dened Radio, Ph.D. thesis, University
of Bristol, Bristol, UK, 2003.
[11] I. Vendik, O. Vendik, V. Pleskachev, and M. Nikolski, Tun-
able microwave lters using ferroelectric materials, IEEE
Trans. Appl. Superconduct., vol. 13, no. 2, pp. 716719, 2003.
[12] B. Noren, Thin lm barium strontium titanate (BST) for
a new class of tunable RF components, Microwave Journal,
vol. 47, no. 5, pp. 210220, 2004.
[13] G. M. Rebeiz and J. B. Muldavin, RF MEMS switches and
switch circuits, IEEE Microwave, vol. 2, no. 4, pp. 5971,
2001.
[14] J. Brank, J. Yao, M. Eberly, A. Malczewski, K. Varian, and C.
Goldsmith, RF MEMS-based tunable lters, International
Journal of RF and Microwave Computer-Aided Engineering,
vol. 11, no. 5, pp. 276284, 2001.
[15] R. M. Young, J. D. Adam, C. R. Vale, et al., Low-loss bandpass
RF lter using MEMS capacitance switches to achieve a one-
octave tuning range and independently variable bandwidth,
in Proc. IEEE MTT-S International Microwave Symposium Di-
gest, vol. 3, pp. 17811784, Philadelphia, Pa, USA, June 2003.
[16] A. Abbaspour-Tamijani, L. Dussopt, and G. M. Rebeiz,
Miniature and tunable lters using MEMS capacitors, IEEE
Trans. Microwave Theory Tech., vol. 51, no. 7, pp. 18781885,
2003.
[17] J. R. MacLeod, T. Nesimoglu, M. A. Beach, and P. A. Warr,
Miniature distributed lters for software re-congurable ra-
dio applications, in Proc. IST Mobile Wireless Telecommunica-
tions Summit, pp. 159163, Thessaloniki, Greece, June 2002.
[18] E. Fourn, C. Quendo, E. Rius, et al., Bandwidth and cen-
tral frequency control on tunable bandpass lter by using
MEMS cantilevers, in Proc. IEEE MTT-S International Mi-
crowave Symposium Digest, vol. 1, pp. 523526, Philadelphia,
Pa, USA, June 2003.
[19] B. Carey-Smith, P. A. Warr, M. A. Beach, and T. Nesimoglu,
Wide tuning-range planar lters using lumped-distributed
coupled resonators, IEEE Trans. Microwave Theory Tech.,
vol. 53, no. 2, pp. 777785, 2005.
[20] I. C. Hunter and J. D. Rhodes, Electronically tunable mi-
crowave bandstop lters, IEEE Trans. Microwave Theory
Tech., vol. 82, no. 9, pp. 13541360, 1982.
[21] S. Toyoda, Quarter-wavelength coupled variable bandstop
and bandpass lters using varactor modes, IEEE Trans. Mi-
crowave Theory Tech., vol. 82, no. 9, pp. 13871389, 1982.
[22] G. Zheng and J. Papapolymerou, Monolithic recongurable
bandstop lter using RF MEMS switches, International Jour-
nal of RF and Microwave Computer-Aided Engineering, vol. 14,
no. 4, pp. 373382, 2004.
[23] B. Carey-Smith and P. A. Warr, Broadband congurable
bandstop lter with composite tuning mechanism, Electron-
ics Letters, vol. 40, no. 25, pp. 15871589, 2004.
[24] S. D. Rogers and C. M. Butler, Wide-band sleeve-cage
and sleeve-helical antennas, IEEE Trans. Antennas Propagat.,
vol. 50, no. 10, pp. 14091414, 2002.
[25] R. B. Waterhouse, Broadband stacked shorted patch, Elec-
tronics Letters, vol. 35, no. 2, pp. 98100, 1999.
[26] T. Huynh and K.-F. Lee, Single-layer single-patch wide-
band microstrip antenna, Electronics Letters, vol. 31, no. 16,
pp. 13101312, 1995.
[27] K.-L. Wong and W.-H. Hsu, Broadband triangular mi-
crostrip antenna with U-shaped slot, Electronics Letters,
vol. 33, no. 25, pp. 20852087, 1997.
[28] Radiocommunications Agency, Ref: AY4476BAntenna Ar-
ray Technology and MIMO Systems, March 2004, available
on http://www.ofcom.org.uk.
[29] P. R. Urwin-Wright, G. S. Hilton, I. J. Craddock, and P. N.
Fletcher, On the pattern control of an annular slot operating
in its DC mode, in Proc. 3rd Management Meeting of Cost
284, Budapest, Hungary, April 2003.
[30] P. R. Urwin-Wright, G. S. Hilton, I. J. Craddock, and P. N.
Fletcher, A tuneable electrically-small antenna operating in
the DC mode, in Proc. 5th European Personal Mobile Com-
municatons Conference, vol. 5, pp. 524528, Glasgow, UK,
April 2003.
[31] P. R. Urwin-Wright, G. S. Hilton, I. J. Craddock, and P. N.
Fletcher, A recongurable electrically-small antenna operat-
ing in the DC mode, in Proc. 57th IEEE Semiannual Vehicu-
lar Technology Conference (VTC 03), vol. 2, pp. 857861, Jeju,
Korea, April 2003.
[32] P. R. Rogers, An electrically-small tuneable annular slot an-
tenna for future mobile terminals, Ph.D. thesis, University of
Bristol, Bristol, UK, 2004.
[33] C. A. Hall, R. C. Luetzelschwab, R. D. Streeter, and J. H. Van-
patten, A 25 watt RF MEM-tuned VHF bandpass lter, in
Proc. IEEE MTT-S International Microwave SymposiumDigest,
vol. 1, pp. 503506, Philadelphia, Pa, USA, June 2003.
Bruce E. Carey-Smith received a B.E. de-
gree in electrical and electronic engineer-
ing from the University of Canterbury,
Christchurch, NZ, in 1995, and is currently
pursuing postgraduate study at the Uni-
versity of Bristol, Bristol, UK. From 1995
to 2002, he was with Tait Electronics Ltd.,
Christchurch, NZ, involved in the design of
RF circuits and systems for mobile radio ap-
plications. Subsequently he joined the Uni-
versity of Bristol, Bristol, UK, as a Research Associate in the Centre
for Communications Research where his current research interests
are in the areas of tunable microwave circuits and amplier lineari-
sation for software recongurable radio.
Paul A. Warr received his Ph.D. degree in
2001 from the University of Bristol, Bris-
tol, UK, for his work on octave-band lin-
ear receiver ampliers, his M.S. degree in
communications systems and signal pro-
cessing also from Bristol in 1996, and his
B.Eng. degree in electronics and commu-
nications from the University of Bath, UK,
in 1994. He is currently a Lecturer in ra-
dio frequency engineering at the University
of Bristol where his research covers the front-end aspects of soft-
ware (recongurable) radio and diversity-exploiting communica-
tion systems; responsive linear ampliers; exible lters; and linear
frequency translation. Funding sources for this research have in-
cluded UK DTI/EPSRC alongside CEC ACTS and IST programmes
and industrial collaborators. Dr. Warr is a Member of the Execu-
tive Committee of the IEE Professional Network on Communica-
tion Networks & Services. Prior appointments have included the
Marconi Company where he worked on secure, high-redundancy,
cross-platform communications.
Flexible Subsystems for Radio Front Ends 363
Phill R. Rogers obtained the M.Eng. degree
(rst class) in electrical and electronic en-
gineering from the University of Bristol in
2000. He obtained his Ph.D. degree in 2004
which was titled An electrically-small tune-
able annular slot antenna for future mo-
bile terminals. Since October 2003, he has
been employed as a Research Assistant at the
University of Bristol in a number of areas
including UWB antennas, MIMO antennas
and systems, and terminal antennas. He is currently involved in
several European projects and has authored and coauthored over
twenty publications.
Mark A. Beach received the Ph.D. degree
from the University of Bristol, Bristol, UK,
in 1989. In 1989, he joined the University of
Bristol as a member of academic sta where
he currently holds the post of Professor of
radio systems engineering. He has made
contributions to the European collaborative
projects, TSUNAMI, SATURN, ROMAN-
TIK, TRUST, and more recently SCOUT.
At present his interests are focused toward
multiple-input multiple-output (MIMO) channel characterization
and the design and optimization of space-time coded wireless ar-
chitectures for 3G and 4G wireless networks. His research interests
include smart antenna technology for wireless as well as analogue
RF circuitry for software denable radio (SDR). Professor Beach is
an active Member of the Institution of Electrical Engineers (IEE)
Professional Network on Antennas and Propagation as well as an
Editor of the IEEE Transactions on Wireless Communications.
Georey S. Hilton received the B.S. de-
gree from the University of Leeds, UK, in
1984, and the Ph.D. degree from the De-
partment of Electrical and Electronic En-
gineering, the University of Bristol, Bristol,
UK, in 1993. From1984, to 1986, he worked
as a design engineer at GEC-Marconi, be-
fore commencing research in the area of
microwave antennas, rst as a postgraduate
student and later as a member of research
sta at Bristol University. This work included design and analysis of
printed antenna elements and arrays, and also involved the devel-
opment of nite-dierence time-domain (FDTD) models for these
structures. In 1993, he joined the academic sta at the University
and currently holds the post of Senior Lecturer. Current research
interests include radiation pattern synthesis; array design and mod-
elling; the design of electrically small antennas, active antennas, and
integrated antenna transceivers, primarily for use in small portable
terminals as well as wideband applications, such as ground pene-
trating radar. This has led to publications (on both practical and
simulation work) in refereed journal and conference proceedings
in Europe, America, and Asia.
EURASIP Journal on Wireless Communications and Networking 2005:3, 364381
c 2005 Y. S. Poberezhskiy and G. Y. Poberezhskiy
Flexible Analog Front Ends of Recongurable
Radios Based on Sampling and Reconstruction
with Internal Filtering
YemS. Poberezhskiy
Rockwell Scientic Company, Thousand Oaks, CA 91360, USA
Email: ypoberezhskiy@rwsc.com
Gennady Y. Poberezhskiy
Raytheon Company, El Segundo, CA 90245, USA
Email: gennady@raytheon.com
Received 27 September 2004; Revised 4 April 2005
Bandpass sampling, reconstruction, and antialiasing ltering in analog front ends potentially provide the best performance of
software dened radios. However, conventional techniques used for these procedures limit recongurability and adaptivity of
the radios, complicate integrated circuit implementation, and preclude achieving potential performance. Novel sampling and
reconstruction techniques with internal ltering eliminate these drawbacks and provide many additional advantages. Several ways
to overcome the challenges of practical realization and implementation of these techniques are proposed and analyzed. The impact
of sampling and reconstruction with internal ltering on the analog front end architectures and capabilities of software dened
radios is discussed.
Keywords and phrases: software dened radios, recongurable and adaptive transceivers, sampling, analog signal reconstruction,
antialiasing ltering, A/D.
1. INTRODUCTION
Next generation of software dened radios (SDRs) should
be recongurable to support future wireless systems operat-
ing with dierent existing and evolving communication stan-
dards while providing a wide variety of services over vari-
ous networks. These SDRs should also be extremely adap-
tive to achieve high performance in dynamic communica-
tion environment and to accommodate varying user needs.
Modern radios, virtually all of which are digital, do not
meet these requirements. They contain large analog front
ends, that is, their analog and mixed-signal portions (AMPs)
[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]. The AMPs are
much less exible and have much lower scale of integration
than the radios digital portions (DPs). The AMPs are also
sources of many types of interference and signal distortion. It
can be stated that recongurability, adaptivity, performance,
and scale of integration of modern SDRs are limited by their
AMPs. Therefore, only radical changes in the design of the
AMPs allow development of really recongurable SDRs.
This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
It is shown in this paper that the changes in the AMP de-
sign have to be related rst of all to the methods of sampling,
reconstruction, and antialiasing ltering. It is also shown
that implementation of novel sampling and reconstruction
techniques with internal ltering [17, 18, 19, 20, 21, 22, 23]
will make the AMPs of SDRs almost as exible as their DPs
and signicantly improve performance of SDRs. To this end,
conventional architectures of the radio AMPs are briey ex-
amined in Section 2. It is shown that the architectures that
potentially can provide the best performance have the low-
est exibility and scale of integration. The main causes of
the conventional architectures drawbacks are determined.
In Section 3, novel sampling and reconstruction techniques
with internal ltering are described. The sampling technique
was obtained as a logical step in the development of inte-
grating sample-and-hold ampliers (SHAs) in [17, 18]. In
[19, 20], it was derived from the sampling theorem. The re-
construction technique with internal ltering was derived
from the sampling theorem in [21]. Initial analysis of both
techniques was performed in [22, 23]. Section 3 contains ex-
amination of their features and capabilities, which is more
detailed than that in [22, 23]. Challenges of these techniques
implementation and two methods of modication of sam-
pling circuits (SCs) with internal antialiasing ltering are
Flexible Analog Front Ends of Recongurable Radios 365
analyzed in Section 4. Since SCs and reconstruction circuits
(RCs) with internal ltering are inherently multichannel,
mitigation of the channel mismatch impact on the perfor-
mance of the SDRs is discussed in Section 5. Architectures
of the AMPs modied to accommodate sampling and recon-
struction with internal ltering are considered in Section 6.
2. CONVENTIONAL ARCHITECTURES OF
THE RADIOAMPS
2.1. AMPs of receivers
In digital receivers, the main purpose of AMPs is to create
conditions for signal digitization. Indeed, AMPs, regardless
of their architectures, carry out the following main func-
tions: antialiasing ltering, amplication of received sig-
nals to the level required for the analog-to-digital converter
(A/D), and conversion of the signals to the frequency most
convenient for sampling and quantization. Besides, they of-
ten provide band selection, image rejection, and some other
types of frequency selection to lower requirements for the
dynamic range of subsequent circuits. Most AMPs of mod-
ern receivers belong to one of three basic architectures: direct
conversion architecture, superheterodyne architecture with
baseband sampling, and superheterodyne architecture with
bandpass sampling. The examples of these architectures are
shown in Figure 1.
In a direct conversion (homodyne) architecture (see
Figure 1a), a radio frequency (RF) section performs prelimi-
nary ltering and amplication of the sumof a desired signal,
noise, and interference. Then, this sum is converted to the
baseband, forming its in-phase (I) and quadrature (Q) com-
ponents. A local oscillator (LO), which generates sine and
cosine components at radio frequency f
r
, is tunable within
the receiver frequency range. Lowpass lters (LPFs) provide
antialiasing ltering of the I and Q components while SHAs
and A/Ds carry out their sampling and quantization. Chan-
nel ltering is performed digitally in the receiver DP. For sim-
plicity, circuits providing frequency tuning, gain control, and
other auxiliary functions are not shown in Figure 1 and sub-
sequent gures. Although integrated circuit (IC) implemen-
tation of this architecture encounters many diculties, it is
simpler than that of the architectures shown in Figures 1b
and 1c.
In a superheterodyne architecture with baseband sam-
pling (see Figure 1b), the sum of a desired signal, noise, and
interference is converted to intermediate frequency (IF) f
0
af-
ter image rejection and preliminary amplication in the RF
section. Antialiasing ltering is performed at a xed IF. This
enables the use of bandpass lters with high selectivity, for
example, surface acoustic wave (SAW), crystal, mechanical,
and ceramic. Then, the sum is converted to the baseband and
its I and Q components are formed.
An example of a superheterodyne architecture with
bandpass sampling is shown in Figure 1c. In most cases,
such receivers have two frequency conversions. The 1st IF
is usually selected high enough to simplify image rejection
and reduce the number of spurious responses. The 2nd IF is
RF
section
LPF SHA A/D
I channel
cos 2 f
r
t
sin2 f
r
t
LPF SHA A/D
Q channel
To DP
(a)
RF
section
LPF
IF strip
IF
lter
SHA A/D
I channel
cos 2 f
0
t
sin2 f
0
t
LPF
LPF
LO
SHA A/D
Q channel
To DP
(b)
RF
section
LPF1
1st IF strip
2nd IF strip
1st IF
lter
2nd IF
lter
1st LO
2nd LO
LPF2
SHA A/D
To DP
(c)
Figure 1: Receiver AMP architectures: (a) direct conversion archi-
tecture, (b) superheterodyne architecture with baseband sampling,
and (c) superheterodyne architecture with bandpass sampling.
typically chosen to simplify antialiasing ltering and digitiza-
tion. Double frequency conversion also allows division of the
AMP gain between the 1st and 2nd IF strips. This architec-
ture performs real-valued bandpass sampling, representing
signals by the samples of their instantaneous values. In the
DP, these samples are converted to the samples of I and Q
components (complex-valued representation), to make digi-
tal signal processing more ecient.
The results of comparative analysis of the described ar-
chitectures are reected in Table 1. This analysis is not de-
tailed because each basic architecture has many modica-
tions. For example, superheterodyne architectures may have
366 EURASIP Journal on Wireless Communications and Networking
Table 1: Comparison of various AMP architectures of modern receivers.
Architecture Advantages Drawbacks
Direct conversion
receiver architecture
Absence of spectral images
caused by frequency conversion
Signicant phase and amplitude
imbalances between I and Q channels
Better adaptivity compared to
other modern architectures
High nonlinear distortions due to the
fall of substantial part of IMPs within
the signal spectrum
Better compatibility of AMP technology
with IC technology compared to other
architectures
LO leakage that creates interference to
other receivers and contributes to the
DC oset
Relatively low requirements for
SHA and A/D
Relatively low selectivity of
antialiasing ltering
Minimum cost, size, and weight
Direct current (DC) oset caused
by many factors
Flicker noise
Superheterodyne receiver
architecture with
baseband sampling
Radical reduction of LO leakage due
to the oset frequency conversion
High nonlinear distortions due to the
fall of substantial part of IMPs within
signal spectrum
High selectivity of antialiasing ltering
provided by SAW, crystal, mechanical, or
ceramic IF lters
Low adaptivity and recongurability of the
receiver AMP due to the use of SAW,
crystal, mechanical, or ceramic IF lters
Slight reduction of phase and amplitude imbalances
between I and Q channels compared to the direct
conversion architecture (due to conversion from a
constant IF to zero frequency)
Incompatibility of AMP technology with
IC technology due to the use of SAW,
crystal, mechanical, or ceramic IF lters
Reduction of icker noise due to
lesser gain at zero frequency
Still signicant phase and amplitude
imbalances between I and Q channels
Relatively low requirements for
SHA and A/D
Spurious responses due to
frequency conversions
Still signicant icker noise
Superheterodyne receiver
architecture with
bandpass sampling
Radical reduction of LO leakage
due to oset frequency conversion
Low adaptivity and recongurability of
the receiver AMP due to the use of SAW,
crystal, mechanical, or ceramic IF lters
High selectivity of antialiasing ltering
provided by SAW, crystal, mechanical,
or ceramic IF lters
Incompatibility of AMP technology with
IC technology due to the use of SAW,
crystal, mechanical, or ceramic IF lters
Exclusion of phase and amplitude
imbalances between I and Q channels
Still high nonlinear distortions due to
large input current of SHA
Exclusion of DC oset and icker
noise
Spurious responses due to
frequency conversions
Minimum IMPs falling within the
signal spectrum
Highest requirements for SHA and
A/D
dierent number of frequency conversions, and even the ar-
chitectures with a single conversion have dierent properties
depending on the parameters of their IF strips. For instance,
selection of a low IF in a single-conversion architecture en-
ables replacement of high-selectivity o-chip IF lters with
active lters. This increases exibility and scale of integra-
tion of an AMP at the expense of more complicated image
rejection.
Despite the absence of some details, Table 1 conclusively
shows that the superheterodyne architecture with bandpass
sampling has advantages that cannot be provided by other
architectures. Indeed, only bandpass sampling minimizes the
number of intermodulation products (IMPs) falling within
the signal spectrum. It also excludes phase and amplitude
imbalances between I and Q channels, DC oset, and icker
noise. The drawbacks of this architecture have the following
causes. Low adaptivity, recongurability, and scale of inte-
gration of the AMPs are caused by inexibility of the best IF
lters (e.g., SAW, crystal, mechanical, and ceramic) and in-
compatibility of their technology with IC technology. Inex-
ibility of these lters also does not allow avoiding spurious
responses. Two times higher sampling frequency required for
bandpass sampling raises requirements for SHA and A/D. At
present, track-and-hold ampliers (THAs) are usually used
Flexible Analog Front Ends of Recongurable Radios 367
as SHAs for bandpass sampling. A THA does not suppress
out-of-band noise and IMPs of all the stages between the an-
tialiasing lter and the THA capacitor. As a result of sam-
pling, these noise and IMPs fall within the signal spectrum.
The impact of this phenomenon is especially signicant in
receivers with bandpass sampling. THAs need large input
current because they utilize only a small fraction of signal
energy for sampling. The large input current requires a sig-
nicant AMP gain. This makes sampling close to the antenna
impossible. The large input current also increases nonlinear
distortions. Higher frequency of bandpass signals compared
to baseband ones further increases the required THA input
current and, consequently, nonlinear distortions. THAs are
very susceptible to jitter.
It is important to add that conventional sampling pro-
cedures have no theoretical basis. In contrast, sampling with
internal antialiasing ltering has been derived from the sam-
pling theorem. As shown in Section 3, it eliminates the draw-
backs of conventional sampling.
2.2. AMPs of transmitters
An AMP of a digital transmitter, regardless of its architecture,
has to perform reconstruction ltering, conversion of recon-
structed signals to the RF, and their amplication. Similar
to the receivers, modern transmitters have three basic AMP
architectures: direct up-conversion architecture, oset up-
conversion architecture with baseband reconstruction, and
oset up-conversion architecture with bandpass reconstruc-
tion. Simplied block diagrams of these architectures are
shown in Figure 2.
In a direct up-conversion architecture (see Figure 2a),
modulation and channel ltering are carried out in the trans-
mitter DP using complex-valued representation. The I and
Q outputs of the DP are converted to analog samples by
D/As. After baseband ltering and amplication of I and
Q components, an analog bandpass signal is formed di-
rectly at the transmitter RF. An LO, which generates cos 2 f
r
t
and sin2 f
r
t, is tunable within the transmitter frequency
range. The formed RF signal passes through a bandpass lter
(BPF) that lters out unwanted products of frequency up-
conversion, and enters a power amplier (PA). This archi-
tecture is the most exible and suitable for IC implementa-
tion among modern architectures. However, it cannot pro-
vide high performance. The baseband reconstruction causes
signicant amplitude and phase imbalances between the I
and Q channels, DC oset, and nonlinear distortions that re-
duce the accuracy of modulation. The DC oset also causes
the LO leakage through the antenna. Additional problem of
this architecture is that a voltage-controlled oscillator (VCO),
used as an LO, is sensitive to pulling from the PA output.
An AMP architecture with oset up-conversion and
baseband reconstruction (see Figure 2b) is not susceptible to
VCO pulling. It provides better reconstruction ltering than
the previous architecture due to the use of SAW, crystal, me-
chanical, or ceramic IF lters and allows slightly more accu-
rate formation of bandpass signals since it is performed at a
constant IF. If the IF is selected higher than the upper bound
D/A LPF
I channel
cos 2 f
r
t From
DP
D/A LPF
Q channel sin2 f
r
t
BPF PA
(a)
IF strip
IF
lter
BPF PA
D/A
From
DP
D/A
I channel
Q channel
LPF
LPF
LO
cos 2 f
0
t
sin2 f
0
t
(b)
From DP
D/A SHA
1st IF
lter
2nd IF
lter
BPF PA
1st IF strip
2nd IF strip
1st LO
2nd LO
(c)
Figure 2: Transmitter AMP architectures: (a) direct up-conversion
architecture, (b) oset up-conversion architecture with baseband
reconstruction, (c) oset up-conversion architecture with bandpass
reconstruction.
of the transmitter RF band, the BPF in the AMP can be re-
placed by an LPF. This architecture still has all the drawbacks
related to baseband reconstruction.
These drawbacks are eliminated in an oset up-
conversion architecture with bandpass reconstruction shown
in Figure 2c. Here, a bandpass IF signal is formed digitally
in the DP. This reduces nonlinear distortions and excludes
DC oset and amplitude and phase imbalances between I
and Q channels. As a result, modulation becomes more ac-
curate, and a spurious carrier is not present. However, like
in the case of receivers, these advantages are achieved at the
expense of reduced adaptivity of the AMP and incompatibil-
ity of its technology with IC technology caused by the most
eective IF reconstruction lters. Besides, the sample mode
368 EURASIP Journal on Wireless Communications and Networking
Table 2: Comparison of various AMP architectures of modern transmitters.
Architecture Advantages Drawbacks
Direct
up-conversion
transmitter
architecture
Better compatibility of AMP technology
with IC technology compared to other
modern architectures
Low accuracy of modulation due to
signicant phase and amplitude imbalances
between I and Q channels, DC oset, and
nonlinear distortions
Better adaptivity compared to
other modern architectures
LO leakage through the antenna caused
by DC oset and other factors
Pulling voltage-controlled LO from PA output
Oset
up-conversion
transmitter
architecture with
baseband
reconstruction
Insusceptibility to pulling the
voltage-controlled LO from the PA
output
Low accuracy of modulation due to
signicant phase and amplitude
imbalances between I and Q channels, DC
oset, and nonlinear distortion
High selectivity of reconstruction
ltering due to the use of SAW, crystal,
mechanical, or ceramic IF lters
Low adaptivity and recongurability of AMP
due to the use of SAW, crystal, mechanical, or
ceramic IF lters
Slightly better accuracy of modulation
due to forming a bandpass signal at a
constant IF
Incompatibility of AMP technology with
IC technology due to the use of SAW,
crystal, mechanical, or ceramic IF lters
Reduction of LO leakage
Oset
up-conversion
transmitter
architecture
with bandpass
reconstruction
The highest accuracy of modulation due to
radical reduction of phase and amplitude
imbalances between I and Q channels, DC
oset, and nonlinear distortion
Low adaptivity and recongurability
of AMP due to the use of SAW,
crystal, mechanical, or ceramic lters
Insusceptibility to pulling
voltage-controlled LO from PA
output
Incompatibility of AMP technology with
IC technology due to the use of SAW,
crystal, mechanical, or ceramic lters
High selectivity of reconstruction
ltering due to the use of SAW, crystal,
mechanical, or ceramic lters
Incomplete utilization of D/A output
power
Radical reduction of LO leakage High requirements for D/A
length t
s
in a conventional SHA at the D/A output should
meet the condition
t
s

1
_
2 f
0
_, (1)
where f
0
is a center frequency of the reconstructed signal,
which coincides with the 1st IF. Condition (1) can be by-
passed by using SHA with weighted integration. However,
they are not used. Condition (1) does not allow ecient uti-
lization of the D/A output current and, consequently, signal
reconstruction close to the antenna.
The results of the comparative analysis of the described
transmitter AMP architectures are reected in Table 2. Since
each basic architecture has many modications, this anal-
ysis is not detailed. However, it shows that the oset up-
conversion architecture with bandpass reconstruction pro-
vides the highest accuracy of modulation. As to the draw-
backs of this architecture, they can be eliminated by imple-
mentation of the proposed reconstruction technique with in-
ternal ltering (see Section 3).
3. SAMPLINGANDRECONSTRUCTIONWITH
INTERNAL FILTERING
3.1. General
As shown in Section 2, AMPs with bandpass sampling, re-
construction, and ltering provide the best performance of
both receivers and transmitters (see Figures 1c and 2c). At
the same time, conventional methods of sampling, recon-
struction, and ltering limit exibility of the AMPs, compli-
cate their IC implementation, and prevent achieving poten-
tial performance. Novel sampling and reconstruction tech-
niques with internal ltering [17, 18, 19, 20, 21, 22, 23] al-
low elimination of these drawbacks and provide additional
benets. These techniques have a strong theoretical founda-
tion because they are derived from the sampling theorem.
They can be used for both bandpass and baseband sam-
pling and reconstruction. However, this paper is mainly fo-
cused on bandpass applications of the proposed techniques
since the techniques are more benecial for these applica-
tions.
Flexible Analog Front Ends of Recongurable Radios 369
2 f
s
f
s
0 f
s
2 f
s
f
B

S
u
( f )

(a)
2 f
s
f
s
0 f
s
2 f
s
f

G
a
( f )

S
u1
( f )

(b)
Figure 3: Amplitude spectra and the desired AFR: (a) |S
u
( f )|,
(b) |S
u1
( f )| and |G
a
( f )| (dashed line).
3.2. Antialiasing and reconstruction ltering
To better describe operation of sampling and reconstruction
circuits (SCs and RCs) with internal ltering, we rst spec-
ify requirements for antialiasing and reconstruction lter-
ing. An antialiasing lter should minimally distort analog
signal u(t) intended for sampling and maximally suppress
noise and interference that can fall within the signal spec-
trum S
u
( f ) as a result of sampling.
When baseband sampling takes place, spectrum S
u
( f ) of
a complex-valued u(t), represented by its Iand Q compo-
nents, occupies the interval (see Figure 3a)
[0.5B, +0.5B], (2)
where B is a bandwidth of u(t). Sampling with frequency
f
s
causes replication of S
u
( f ) with period f
s
(see Figure 3b)
and mapping the whole f -axis for u(t) into the region
[0.5 f
s
, 0.5 f
s
[ for the sampled signal u(nT
s
), where T
s
= 1/ f
s
is a sampling period. Thus, antialiasing lter should cause
minimum distortion within interval (2) and suppress noise
and interference within the intervals
_
k f
s
0.5B, k f
s
+ 0.5B
_
, (3)
where replicas of S
u
( f ) are located in the spectrum S
u1
( f )
of u(nT
s
). In (3), k is any nonzero integer. In principle, noise
and interference within the gaps between all intervals (3) and
(2) can be suppressed in the DP. However, if these noise and
interference are comparable with or greater than u(t), weak-
ening them by an SC lowers requirements for the resolution
of an A/D and DP. A desired amplitude-frequency response
(AFR) |G
a
( f )| of an antialiasing lter is shown in Figure 3b
by the dashed line.
In the case of reconstruction, it is necessary to suppress
all the images of u(nT
s
) within intervals (3) and minimally
distort the image within interval (2). No suppression within
the gaps between intervals (3) and (2) is usually required.
When bandpass sampling takes place, spectrum S
u
( f ) of
real-valued bandpass u(t) occupies the intervals
_
f
0
0.5B, f
0
+ 0.5B
_

_
f
0
0.5B, f
0
+ 0.5B
_
, (4)
2 f
s
f
s
f
0
0 f
s
f
0
2 f
s
f
B

S
u
( f )

(a)
2 f
s
f
0
f
s
0 f
s
f
0
2 f
s
f

G
a
( f )

S
u1
( f )

(b)
Figure 4: Amplitude spectra and the desired AFR: (a) |S
u
( f )|,
(b) |S
u1
( f )| and |G
a
( f )| (dashed line).
where f
0
is a center frequency of S
u
( f ). A plot of S
u
( f ) is
shown in Figure 4a. For bandpass sampling and reconstruc-
tion, f
s
usually meets the condition
f
s
=
f
0
_
oor
__
f
0
/ f
s
_
+ 0.5
_
0.25
_. (5)
Selection of f
s
according to (5) minimizes aliasing and sim-
plies both digital forming of I and Q components at the
output of the receiver A/D and digital forming of a band-
pass signal at the input of the transmitter D/A. Therefore, f
s
that meets (5) is considered optimal. When f
s
is optimal, an
antialiasing lter should cause minimum distortion within
intervals (4) and suppress noise and interference within the
intervals
_

_
f
0
+ 0.5B + 0.5r f
s
_
,
_
f
0
0.5B + 0.5r f
s
__

_
f
0
0.5B + 0.5r f
s
, f
0
+ 0.5B + 0.5r f
s
_
,
(6)
where r is an integer, r [(0.52 f
0
/ f
s
), [, r = 0. Figure 4b
shows amplitude spectrum |S
u1
( f )| of u(nT
s
), and the de-
sired AFR |G
a
( f )| of an antialiasing lter for bandpass sam-
pling. Thus, a bandpass antialiasing lter has to suppress
noise and interference within intervals (6) and minimally
distort u(t) within intervals (4). Suppression of noise and in-
terference within the gaps between intervals (4) and (6) is not
mandatory, but it can be used to lower requirements for the
resolution of an A/D and DP.
Bandpass reconstruction requires only suppression of
u(nT
s
) images within intervals (6) and minimum distortion
within intervals (4).
3.3. Canonical sampling circuits
The block diagrams of two canonical SCs with internal an-
tialiasing ltering are shown in Figure 5. In Figure 5a, an in-
put signal u
i
(t) is fed into L parallel channels, whose out-
puts are in turn connected to an A/D by a multiplexer (Mx).
The spectrum of u
i
(t) is not limited by an antialiasing lter
and includes the spectrum of the signal u(t) that should be
sampled. The lth channel (l [1, L]) forms all samples with
370 EURASIP Journal on Wireless Communications and Networking
u
i
(t)
u(nT
s
)
1
.
.
.
L
. . .
. . .
WFG
Control
unit
.
.
.
Mx A/D
_
_
(a)
u
i
(t)
u(nT
s
)
1
.
.
.
L
. . .
. . .
WFG
Control
unit
.
.
.
Mx
_
_
A/D
A/D
(b)
Figure 5: Canonical SCs with internal antialiasing ltering: (a)
single-A/D version and (b) multiple-A/D version.
numbers l + kL, where k is any integer. The operational cy-
cle of each channel is equal to LT
s
, consists of three modes
(sample, hold, and clear), and is shifted by T
s
relative to the
operational cycle of the previous channel. The length of the
sample mode is equal to T
w
, where T
w
is the length of weight
function w
0
(t).
During the sample mode, u
i
(t) is multiplied by w
n
(t) =
w
0
(t nT
s
), and the product is integrated. As a result,
u
_
nT
s
_
=
_
nTs+0.5Tw
nTs0.5Tw
u
i
() w
n
() d. (7)
Equation (7) reects sampling, accumulation of the signal
energy with weight w
0
(t), and antialiasing ltering with im-
pulse response h(t) = w
0
(nT
s
+ 0.5T
w
t). Throughout the
hold mode with length T
h
, a channel is connected to the A/D
that quantizes the channel output. In the clear mode with
length T
c
, the channel is disconnected from the A/D, and the
capacitor of its integrator is discharged. It is reasonable to
allocate T
s
for both hold and clear modes: T
h
+ T
c
= T
s
. A
weight function generator (WFG) simultaneously generates
L1 copies w
n
(t) of w
0
(t) because, at any instant, L1 chan-
nels are in the sample mode, and one channel is in the hold or
clear mode. Each w
n
(t) is shifted relative to the previous one
10 9 8 7 6 5 4 3 2 1 0
5
0
5
t/T
s
u
i
(
t
)
(a)
10 9 8 7 6 5 4 3 2 1 0
1
0
1
t/T
s
w
n
(
t
)
(b)
10 9 8 7 6 5 4 3 2 1 0
1
0
1
t/T
s
w
n
(
t
)
(c)
10 9 8 7 6 5 4 3 2 1 0
1
0
1
t/T
s
w
n
(
t
)
(d)
10 9 8 7 6 5 4 3 2 1 0
1
0
1
t/T
s
w
n
(
t
)
(e)
10 9 8 7 6 5 4 3 2 1 0
1
0
1
t/T
s
w
n
(
t
)
(f)
Figure 6: Positions of samples and corresponding w
n
(t).
by T
s
. Positions of samples and corresponding w
n
(t) are illus-
trated by Figure 6 for L = 5. As follows from (7), w
0
(t) deter-
mines ltering properties of SCs. Examples of baseband and
bandpass weight functions w
0
(t) and the AFRs |G
a
( f )| of the
SCs with these w
0
(t) are shown in Figures 7 and 8, respec-
tively. Since an SC performs nite impulse response (FIR)
ltering with AFR |G
a
( f )|, which is the amplitude spectrum
of w
0
(t), it suppresses interference using zeros of its AFR.
When baseband sampling takes place, the distances between
the centers of adjacent intervals (2) and (3) are equal to f
s
(see Figure 3). To suppress all intervals (3), |G
a
( f )| should
Flexible Analog Front Ends of Recongurable Radios 371
2 1.5 1 0.5 0 0.5 1 1.5 2
1
0.8
0.6
0.4
0.2
0
t/T
s
w
0
(
t
)
(a)
4 3.5 3 2.5 2 1.5 1 0.5 0
0
20
40
60
80
f / f
s
A
F
R
|
G
a
(
f
)
|
(b)
Figure 7: Baseband SC (a) w
0
(t) and (b) AFR |G
a
( f )|, in dB.
have at least one zero within each of them. To achieve this,
condition T
w
1/ f
s
= T
s
is necessary. For bandpass sam-
pling, the distances between the centers of adjacent intervals
(4) and (6) are equal to 0.5 f
s
(see Figure 4). Consequently,
T
w
1/(0.5 f
s
) = 2T
s
is required. When T
h
+ T
c
= T
s
, the
length of the channel operational cycle is
LT
s
= T
w
+ T
h
+ T
c
3T
s
for bandpass u(t),
LT
s
= T
w
+ T
h
+ T
c
2T
s
for baseband u(t).
(8)
It follows from (8) that L 3 is required for bandpass sam-
pling and L 2 is necessary for baseband sampling. Only
bandpass sampling is considered in the rest of the paper.
In the SC shown in Figure 5b, the required speed of A/Ds
is lower and an analog Mx is replaced with a digital one. This
version is preferable when the maximum speed of the A/Ds
is lower than f
s
, or when L slower A/Ds cost less and/or con-
sume less power than faster one.
3.4. Canonical reconstruction circuits
The block diagrams of canonical RCs with internal ltering
are shown in Figure 9. In Figure 9a, a demultiplexer (DMx)
periodically (with period LT
s
) connects the output of a D/A
to each of L channels. The lth channel (l [1, L]) processes
samples with numbers l + kL, where k is any integer. Opera-
tional cycle duration of each channel is equal to LT
s
and de-
layed by T
s
relative to that of the previous channel. The cycle
consists of three modes: clear, sample, and multiply. In the
clear mode, the SHA capacitor is discharged. Then, during
the sample mode, this capacitor is connected to the D/A by
the DMx and charged. Throughout these modes, there is no
signal at the channel output because zero level enters the sec-
ond input of a multiplier from a WFG. The reasonable total
4 3 2 1 0 1 2 3 4
1
0.5
0
0.5
1
t/T
s
w
0
(
t
)
(a)
5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0
0
20
40
60
80
f / f
s
A
F
R
|
G
a
(
f
)
|
(b)
Figure 8: Bandpass SC (a) w
0
(t) and (b) AFR |G
a
( f )|, in dB.
length of the sample and clear modes is equal to T
s
. In the
subsequent multiply mode with duration T
w
, the signal from
the SHAis multiplied by the appropriate copy of w
0
(t) gener-
ated by the WFG, and the product enters an adder that sums
the output signals of all the channels. Since at any instant,
L 1 channels are in the multiply mode and one channel is
in the sample or clear mode, the WFG simultaneously gen-
erates L 1 copies of w
0
(t), each delayed by T
s
relative to
the previous one. The RC reconstructs an analog signal u(t)
according to the equation
u(t)

_
n=
u
_
nT
s
_
w
n
(t) =

_
n=
u
_
nT
s
_
w
0
_
t nT
s
_
.
(9)
It follows from (9) that the RC performs reconstruction l-
tering with transfer function determined by w
0
(t).
In the version of a canonical RCshown in Figure 9b, digi-
tal words are distributed by a digital DMx among L channels.
Presence of a D/A in each channel allows removal of SHAs.
Here, the channel operational cycle consists of two modes:
convert and multiply. In the rst mode, the D/A converts
digital words into samples u(nT
s
), which are multiplied by
w
n
(t) during the multiply mode. This version has the follow-
ing advantages: lower requirements for the speed of D/As,
replacement of an analog DMx by a digital one, and removal
of SHAs.
3.5. Advantages of the SCs and RCs and challenges
of their realization
Both SCs and RCs with internal ltering make AMPs highly
adaptive and easily recongurable because w
0
(t), which
determines their ltering properties, can be dynamically
372 EURASIP Journal on Wireless Communications and Networking
Digital
words
u(nT
s
)
D/A DMx
. . .
. . .
L
WFG
Control
unit

u(t)
1
SHA
SHA
.
.
.
.
.
.
(a)
Digital
words
DMx
. . .
. . .
L
WFG
Control
unit

u(t)
1
D/A
D/A
.
.
.
.
.
.
u(nT
s
)
(b)
Figure 9: Canonical RCs with internal reconstruction ltering:
(a) single-D/A version and (b) multiple-D/A version.
changed. Internal ltering performed by these structures al-
lows removal of conventional antialiasing and reconstruc-
tion lters or their replacement by wideband low-selectivity
lters realizable on a chip. This makes the AMP technol-
ogy uniform and compatible with the IC technology. The
RCs with internal ltering utilize the D/A output current
more eciently than conventional devices, then bandpass
reconstruction takes place. The SCs with internal antialiasing
ltering accumulate signal energy in their storage capacitors
during the sample mode. This accumulation lters out jitter
and reduces the charging current of the storage capacitors
by 2040 dB in most cases. Reduced jitter enables the devel-
opment of faster A/Ds. The decrease in the charging current
lowers both the required gain of an AMP and its nonlinear
distortions. The reduced AMP gain allows sampling close to
the antenna. Smaller charging current also lowers input volt-
age of the SCs. Indeed, although the same output voltage has
to be provided by an SC with internal antialiasing ltering
and a conventional SHA, the SC input voltage can be signi-
cantly lower when the integrator operational amplier has an
adequate gain. As mentioned in Section 2.1, a conventional
SHA does not suppress out-of-band noise and IMPs of all
the stages between the antialiasing lter and its capacitor. As
a result of sampling, these noise and IMPs fall within the sig-
nal spectrum. The SCs with internal antialiasing ltering op-
erate directly at the A/D input and reject out-of-band noise
and IMPs of all preceding stages. Thus, they perform more
eective antialiasing ltering than conventional structures.
At the same time, practical realization of the SCs and
RCs with internal ltering and their implementation in SDRs
present many technical challenges. Canonical structures of
the SCs and RCs (see Figures 5 and 9) are rather complex.
Therefore, their simplication is highly desirable. This sim-
plication is intended, rst of all, to reduce complexity and
number of multiplications.
4. SIMPLIFICATIONOF THE SCs ANDRCs
4.1. Approaches to the problem
Approaches to simplication of the SCs and RCs depend
on the ways of w
n
(t) generation and multiplications. Ana-
log generation of w
n
(t) implies that multiplications of u
i
(t)
in the SCs and u(nT
s
) in the RCs by w
n
(t) are performed
by analog multipliers. Since only simple w
n
(t) can be gen-
erated by analog circuits, and this generation is not exible
enough, digital generation is preferable. When w
n
(t) are gen-
erated digitally, they can be converted to the analog domain
in the WFG (see Figures 5 and 9) or sent to the multipliers in
digital form. In the rst case, multiplications in the SCs and
RCs are analog. In the second case, these multiplications can
be carried out by multiplying D/As.
Since digital w
n
(t) have unwanted spectral images, spec-
tral components of an input signal u
i
(t) in the SCs and a re-
constructed signal u(t) in the RCs corresponding to the un-
wanted images should be suppressed. The suppression can
be performed by a wideband lter with fairly low selectiv-
ity that allows IC implementation. Such a lter is sucient
because a required sampling rate of w
0
(t) representation is
much higher than that of the A/D used in a receiver and the
D/A used in a transmitter when bandpass sampling and re-
construction take place. In practice, some kind of prelter-
ing is performed in all types of receivers, and some kind of
postltering is performed in transmitters. Usually, these pre-
ltering and postltering can provide the required suppres-
sion. Since preltering and postltering automatically sup-
press stopbands (6) remote from passband (4), internal l-
tering performed by SCs and RCs should rst of all sup-
press stopbands (6) closest to the passband. Complexity of
the SCs and RCs, caused by high sampling rate of w
0
(t) rep-
resentation, can be compensated by its low resolution. The
goal is to lower the required resolution of w
0
(t) represen-
tation or to nd other means that can reduce multiplying
D/As (or analog multipliers) to a relatively small number of
switches.
Simplication of the SCs and RCs can be achieved by
proper selection of w
0
(t) and optimization of their architec-
tures for a given w
0
(t). Below, attention is mostly focused on
the SCs because their practical realization is more dicult
than that of RCs due to higher requirements for their dy-
namic range. Achieving a high dynamic range of multiplica-
tions in the SCs is still a challenging task, although low input
current (compared to conventional SHAs) makes it easier.
Brief information on w
0
(t) selection is provided in
Section 4.2, and two examples of the SC simplication are
described and analyzed in Sections 4.3, 4.4, and 4.5. It is
Flexible Analog Front Ends of Recongurable Radios 373
important to emphasize that possible simplications of the
SCs are not limited to these examples.
4.2. Selection of weight functions
Selection of w
0
(t) is application specic and requires multi-
ple tradeos. For example, w
0
(t) that maximizes the dynamic
range of an AMP and w
0
(t) that provides the best internal l-
tering are dierent. Indeed, w
0
(t) with rectangular envelope
maximizes the dynamic range due to its minimum peak fac-
tor and the most ecient accumulation of the signal energy,
but it provides relatively poor internal ltering. At the same
time, w
0
(t) that provides the best internal ltering for given L
and f
s
/B has high peak factor and relatively poor accumula-
tion of signal energy. When both features are desirable, w
0
(t)
has to be selected as a result of a certain tradeo, and this re-
sult can be dierent depending on specic requirements for
the radio. To provide the best antialiasing ltering for given
L and f
s
/B, w
0
(t) should be optimized using the least mean
square (LMS) or Chebyshev criterion [23]. Any w
0
(t), op-
timal according to one of these criteria, requires high accu-
racy of its representation and multiplications. This compli-
cates realization of the SCs. Suboptimal w
0
(t) that provide
eective antialiasing ltering with low accuracy of represen-
tation and multiplications are longer than optimal w
0
(t) and,
consequently, require larger L. An increase of f
s
/B simplies
antialiasing ltering and allows reduction of L or accuracy
of multiplications for a given quality of ltering [20]. Tech-
nology of the SCs and technical decisions regarding these
and other units of the SDRs also inuence selection of w
0
(t).
Due to the complexity of these multiple tradeos, there is
no mathematical algorithm for w
0
(t) selection, and heuristic
procedures combined with analysis and simulation are used
for this purpose.
In general, a bandpass w
0
(t) can be represented as
w
0
(t) = W
0
(t)c(t) for t
_
0.5T
w
, 0.5T
w
_
,
w
0
(t) = 0 for t /
_
0.5T
w
, 0.5T
w
_
,
(10)
where W
0
(t) is a baseband envelope, and c(t) is a periodic
carrier (with period T
0
= 1/ f
0
) that can be sinusoidal or
nonsinusoidal. To provide linear phase-frequency response,
W
0
(t) should be an even function, and c(t) should be an even
or odd function. Assuming that T
w
= kT
s
where k is a nat-
ural number, we can expand c(t) into Fourier series over the
time interval [0.5T
w
, 0.5T
w
]:
c(t) =

_
m=
c
m
e
jm2 f0t
, (11)
where m is any integer and c
m
are coecients of the Fourier
series. Taking into account (10) and (11), we can write that
within the interval [0.5T
w
, 0.5T
w
],
w
0
(t) = W
0
(t)

_
m=
c
m
e
jm2 f0t
=

_
m=
w
m0
(t), (12)
where w
m0
(t) are partial weight functions, whose envelopes
are equal to c
m
W
0
(t) and whose carriers are harmonics of f
0
.
The spectra of w
m0
(t) are partial transfer functions G
m
( f ). It
follows from (12) that when f
0
/ f
s
is high enough ( f
0
/ f
s
> 3
is usually sucient), the distances between adjacent harmon-
ics of f
0
are relatively large, and overlapping of G
m
( f ) does
not notably aect the suppression within those stopbands (6)
that are close to the passband. Since remote stopbands (6)
are suppressed by preltering or postltering, the simplest
c(t), which is a squarewave, can be used when f
0
/ f
s
is suf-
cient. Combining a squarewave c(t) with an appropriately
selected K-level W
0
(t) allows reducing the multiplying D/As
to a small number of switches. Note that, besides w
0
(t) with
K-level W
0
(t), there are other classes of w
0
(t) that allow us
to do this. If discontinuities in W
0
(t) and c(t) are properly
aligned and f
0
/ f
s
> 3, overlapping of G
m
( f ) can be insigni-
cant even if condition T
w
= kT
s
is not met.
The lower f
0
/ f
s
is, the more signicantly G
m
( f ) are over-
lapped. As a result, both W
0
(t) and c(t) inuence the ltering
properties of the SCs and RCs. When f
0
/ f
s
= 0.25, c(t) has
the greatest impact on their transfer functions. To reduce the
multiplying D/As to a small number of switches in this case,
c(t) should also be a several-level function.
4.3. Separate multiplying by W
0
(t) and c(t)
The following method of the SC realization can be obtained
using separate multiplication of u
i
(t) by the envelope W
0
(t)
and carrier c(t) of w
0
(t). The nth sample at the output of the
SC is as follows:
u
_
nT
s
_
=
_
0.5Tw+nTs
0.5Tw+nTs
u
i
(t)w
0
_
t nT
s
_
dt. (13)
Taking into account (10), we can write
w
0
_
t nT
s
_
= W
0
_
t nT
s
_
c
_
t nT
s
_
. (14)
When condition (5) is met, (14) can be rewritten as
w
0
_
t nT
s
_
= W
0
_
t nT
s
_
c
_
t (nmod4)
T
0
4
_
. (15)
Since c(t T
0
/2) = c(t),
c
_
t nT
s
_
=
_

_
c(t)(1)
n/2
if n is even,
c
_
t
T
0
4
_
(1)
(n1)/2
if n is odd.
(16)
Substituting (16) into (14), and (14) into (13), we obtain
u(nT
s
) =
_
0.5Tw+nTs
0.5Tw+nTs
u
i
(t)W
0
_
t nT
s
_

_
c(t)(1)
n/2
if n is even
c
_
t
T
0
4
_
(1)
(n1)/2
if n is odd
_

_
dt.
(17)
374 EURASIP Journal on Wireless Communications and Networking
From
WFEG
From
CU
_
Mx A/D
WFEG CU
. . .
_
. . .
. . .
C Channel 1
C Channel L
cos 2 f
0
t
sin2 f
0
t
u
i
(t)
+

From
CU
.
.
.
.
.
.
Figure 10: Modied SC.
In (16) and (17), corresponds to in (5). In particular,
when c(t) = cos 2 f
0
t, (17) can be rewritten as follows:
u
_
nT
s
_
=
_
0.5Tw+nTs
0.5Tw+nTs
u
i
(t)W
0
_
t nT
s
_

_
_
_
cos
_
2 f
0
t
_
(1)
n/2
if n is even
sin
_
2 f
0
t
_
(1)
(n1)/2
if n is odd
_
_
_
dt.
(18)
The algorithm described by (18) can be carried out by the
modied SC shown in Figure 10. Here, u
i
(t) enters two mul-
tipliers where it is multiplied by cos 2 f
0
t and sin2 f
0
t.
These multiplications are similar to the beginning of the
procedure used for baseband sampling of bandpass signals
(see Figures 1a and 1b). However, further processing is dif-
ferent. Instead of baseband ltering of the lowest spectral
image after each multiplier, the dierential outputs of both
multipliers enter L channels through 4-contact switches. The
switches are necessary because each sample in any channel is
shifted by L/2 relative to the previous one in this channel
when (5) is true. A control unit (CU) provides proper opera-
tion of the switches. This switching shifts the multiplier out-
put spectral image from zero frequency to f
s
/4. After passing
decoupling capacitor C, it is processed in the channel. Sim-
ilar to the canonical structure in Figure 5a, the operational
cycle of each channel is equal to LT
s
, consists of three modes
(sample, hold, and clear), and is shifted by T
s
relative to the
operational cycle of the previous channel. The dierence is
that the channel input signal is multiplied by the appropri-
ate copy W
n
(t) of W
0
(t) instead of w
n
(t) during the sample
mode. A weight function envelope generator (WFEG) forms
W
n
(t). Each W
n
(t) is shifted relative to the previous one by T
s
to be in phase with the operational cycle of the correspond-
ing channel.
At rst glance, the structure in Figure 10 looks even more
complex than the canonical one shown in Figure 5a. How-
ever, appropriate selection of W
0
(t) can signicantly simplify
it. For example, when W
0
(t) is a rectangular function, the
WFEG and the multipliers in the channels are unnecessary.
As shown in Figure 11, the modied SC contains only 2 mul-
tipliers for any L in this case. Complexity of the SC can also
From
CU
.
.
.
From
CU
_
.
.
. Mx A/D
CU
_
. . .
. . .
C
Channel 1
C
Channel L
cos 2 f
0
t
sin2 f
0
t
u
i
(t)
+

Figure 11: Modied SC with rectangular W


0
(t).
be lowered compared to the canonical structures when some
other W
0
(t) are used. Note that single-ended circuits are used
in Figures 10 and 11 only for simplicity of illustration. In
practical applications, dierential circuits are preferable.
4.4. Analysis of the modied SC
Many features of the canonical and modied SCs are the
same. Indeed, when the same w
0
(t) are used, their ltering
properties are identical and they accumulate equal amounts
of signal energy. Consequently, they provide the same reduc-
tion of the input current compared to conventional sampling
structures. They are equally adaptive and equally suitable for
IC implementation. However, there is still substantial dier-
ence between them. Acanonical SCis not sensitive to DCo-
set, while the outputs of the modied SCs are inuenced by
DC osets in the rst two multipliers. Besides, the number
and values of IMPs that fall within the signal spectrum are
higher in the modied SCs than in the canonical ones. In-
deed, multiplication of u
i
(t) by w
n
(t) in each channel of the
canonical SC creates a spectral image at the frequency f
s
/4
because w
n
(t) are centered around corresponding sampling
instants and f
s
meets (5), whereas the rst two multiplica-
tions in the modied SCs create baseband spectral images.
Below, this is proven analytically.
Assuming that the DC oset in the multiplier of the lth
channel in a canonical SC is U
l
, where l = [(n1) modL]+1,
we can rewrite (13) as
u
_
nT
s
_
=
_
0.5Tw+nTs
0.5Tw+nTs
_
u
i
(t)w
0
_
t nT
s
_
+ U
l
_
dt. (19)
It follows from (16) and (19) that
u
_
nT
s
_
=
_
0.5Tw+nTs
0.5Tw+nTs
u
i
(t)W
0
_
t nT
s
_

_
c(t)(1)
n/2
if n is even
c
_
t
T
0
4
_
(1)
(n1)/2
if n is odd
_

_
dt + U
l
T
w
.
(20)
Flexible Analog Front Ends of Recongurable Radios 375
Equation (20) can be rewritten as
u
_
nT
s
_
= (1)
oor [(n+0.50.5)/2]
_
0.5Tw+nTs
0.5Tw+nTs
u
i
(t)W
0
_
t nT
s
_

_
c(t) if n is even
c
_
t +
T
0
4
_
if n is odd
_

_
dt + U
l
T
w
,
(21)
where sign corresponds to in (5). It follows from
(21) that, at the output of a canonical SC, the compo-
nent of the discrete-time signal, caused by the DC oset,
is located at zero frequency, while its desired component
is located at the frequency f
s
/4, as indicated by coecient
(1)
oor [(n+0.50.5)/2]
. Thus, the DC oset can be easily l-
tered out in the receiver DP.
For the modied SC, we can write
u
_
nT
s
_
=
_
0.5Tw+nTs
0.5Tw+nTs
W
0
_
t nT
s
_

_
_
u
i
(t)c(t) + U
1
_
(1)
n/2
if n is even
_
u
i
(t)c
_
t
T
0
4
_
+ U
2
_
(1)
(n1)/2
if n is odd
_

_
dt,
(22)
where U
1
and U
2
are DC osets in the rst two multipliers.
Similar to (20), this equation can be rewritten as
u
_
nT
s
_
= (1)
oor [(n+0.50.5)/2]
_
0.5Tw+nTs
0.5Tw+nTs
W
0
_
t nT
s
_

_
u
i
(t)c(t) + U
1
if n is even
u
i
(t)c
_
t +
T
0
4
_
+ U
2
if n is odd
_

_
dt.
(23)
It follows from (23) that both signal and DC oset compo-
nents after sampling are located at the frequency f
s
/4, as indi-
cated by coecient (1)
oor [(n+0.50.5)/2]
. Therefore, the DC
component cannot be ltered out.
Thus, DC oset and increased number and values of
IMPs lower the performance of the modied SC compared
to the canonical one. However, their performance is still sig-
nicantly better than that of the conventional baseband sam-
pling. Indeed, the entire signal processing is performed at
zero frequency in the conventional procedure. Consequently,
besides multipliers, all subsequent analog stages contribute
to the increase in the DC oset and nonlinear distortion. In
addition, baseband antialiasing lters create signicant phase
imbalance between I and Q channels. In the modied SCs,
signal processing after 4-contact switches is performed at
f
s
/4, and subsequent analog stages do not increase nonlin-
ear distortions and DC oset. The phase mismatch among
channels of the modied SC is negligible because all clock
impulses are generated in the control unit using the same
reference oscillator, and proper design allows us to mini-
mize time skew. As follows from Section 4.2, cos 2 f
0
t and
0.1 0.08 0.06 0.04 0.02 0
f
0
/ f
s
= 1.25
f
0
/ f
s
= 0.75
f
0
/ f
s
= 0.25

S
E
R
(
d
B
)
20
40
60
80
Figure 12: SER() for various f
0
/ f
s
.
sin2 f
0
t in the rst two multipliers of the modied SC can
be replaced by squarewaves with frequency f
0
and time shift
of 0.25T
0
= 0.25/ f
0
relative to each other when f
0
/ f
s
> 3,
and sucient preltering is provided. This replacement fur-
ther simplies the modied SCs. Thus, the described modi-
cation of the SCs substantially simplies their realization at
the expense of slightly lower performance.
4.5. Use of orthogonality of WFGoutputs
As mentioned in Section 4.2, increase of f
s
/B makes inter-
nal ltering easier and may allow reduction of L. In addition
to reducing L, high-ratio f
s
/B makes possible reducing the
number of multipliers N for given L if f
0
/B is also high. This
possibility is discussed below.
When (5) is true, the carrier of w
n
(t) generated for nth
sample is rotated by /2 relative to the carrier of w
n+2m+1
(t)
generated for (n + 2m + 1)th sample, where m is any inte-
ger. Thus, if the envelope of w
0
(t) is rectangular, in some
cases, w
n
(t) and w
n+2m+1
(t) can be sent to the same multi-
plier of the SC or RC with internal ltering even when these
weight functions overlap. This property can be used to re-
duce N for a given L. For example, if T
w
/T
s
= 2 (L = 3) and
u(t) = U
0
cos(2 f
0
t +
0
), one multiplier can be used for all
3 channels and perform, ideally accurate sampling. However,
a pure sinewave cannot carry information. In the case of a
bandpass signal u(t) = U(t) cos[2 f
0
t + (t)], sampling er-
ror is unavoidable, and signal-to-error power ratio (SER) for
this error is
SER =
16
2

2
__
f
0
T
w
_
2
1
_
f
0
T
w
_
3
1
_
2
(24)
when B 1/T
w
. Here, corresponds to in (5), =
B
RMS
/ f
0
, and B
RMS
is root mean square bandwidth of u(t).
Figure 12 illustrates the dependence SER() for several values
376 EURASIP Journal on Wireless Communications and Networking
of f
0
/ f
s
. Since the spectrum of the error determined above is
generally wider than S
u
( f ), a part of this error can be ltered
out in the receiver DP. Therefore, (24) is a lower bound of the
actual SER.
This method of reducing the number of multipliers N
can also be used for L > 3 if the corresponding SER is su-
ciently small. In this case, the minimally required N is
N =
_

_
0.5T
w
T
s
+ 1 if
0.5T
w
T
s
is even,
0.5T
w
T
s
if
0.5T
w
T
s
is odd.
(25)
For N > 1, this method can complicate the channel mis-
match compensation in the receiver DP described in the next
section. It is important to mention that (24) can be used for
any N.
It follows from (24) and Figure 12 that the described
method can be used only with very high-ratios f
s
/B that cor-
respond exclusively to sigma-delta A/Ds.
5. CHANNEL MISMATCHMITIGATION
5.1. Approaches to the problem
The SCs and RCs with internal ltering are inherently mul-
tichannel. Therefore, the inuence of channel mismatch on
the performance of SDRs must be minimized. This is espe-
cially important for the SUs because in the receivers, u(t) is
a sum of a desired signal s(t) and a mixture of the noise and
interference n(t). Thus, u(t) = s(t) + n(t). When the average
power of n(t) is larger than that of s(t), the average power of
the error e(t) caused by the channel mismatch can be com-
parable with or even exceed the power of s(t).
There are three approaches to this problem. The rst of
them includes technical and technological measures that re-
duce this mismatch: placing all the channels on the same die,
simplifying w
0
(t), and correcting circuit design. The second
approach is based on preventing an overlap of the signal and
mismatch error spectra. In this case, the error spectrum can
be ltered out in the DP. The third approach is adaptive com-
pensation of the channel mismatch in the DP. The rst ap-
proach alone is sucient for many types of transmitters and
for receivers with limited dynamic range. In high-quality re-
ceivers, the measures related to this approach are necessary
but usually not sucient. Therefore, the second and third
approaches are considered below.
5.2. Separation of signal and error spectra
To determine the conditions that exclude any overlap be-
tween spectra S
u1
( f ) of u(nT
s
) and S
e
( f ) of e(t), we rst nd
S
e
( f ). The phase mismatch among channels can be made
negligible because all clock impulses are generated in the
control unit using the same reference oscillator, and proper
design minimizes time skew. Therefore, it is sucient to take
into account only the amplitude mismatch caused by the dif-
ferences among the channel gains g
1
, g
2
, . . . , g
L
. The average
gain is g
0
= (g
1
+g
2
+ +g
L
)/L, and the deection fromg
0
is
0.5 f
s
0.2 f
s
0.4 f
s
|C
2
|
|C
1
|
|C
m
|
|C
1
|
|C
2
| |C
2
|
|C
1
|
0 0.2 f
s
0.4 f
s
0.5 f
s
0.6 f
s
0.8 f
s
f
(a)
0.5 f
s
0.25 f
s

G
a
( f )

S
u
( f )

0 0.25 f
s
0.5 f
s
0.75 f
s
f
B B
t
(b)
0.5 f
s
0.25 f
s
0 0.25 f
s
0.5 f
s
0.75 f
s
f

G
d
( f )

S
u1
( f )

S
e
( f )

(c)
Figure 13: Amplitude spectra and AFRs: (a) spectral components
of d(t); (b) |S
u
( f )|solid line, |G
a
( f )|dashed line; (c) |S
u1
( f )|
and |S
e
( f )|solid line, |G
d
( f )|dashed line.
d
l
= g
l
g
0
in the lth channel. Since samples of u(t) are gen-
erated in turn by all channels, the deections d
1
, d
2
, . . . , d
L
,
d
1
, d
2
, . . . , d
L
, d
1
, d
2
, . . . appear at sampling instants t = nT
s
as a periodic function d(t) with period LT
s
:
d(t) =

_
k=
L
_
l=1
_
d
l

_
t (kL + l)T
s
__
, (26)
where (t) is the delta function. The spectrum of d(t) is
S
d
( f ) =

_
m=
_
C
m

_
f
m
L
f
s
__
, (27)
where coecients
C
m
=
1
LT
s
L
_
l=1
_
d
l
exp
_
j2ml
L
__
. (28)
As reected by (27) and (28), S
d
( f ) is a periodic function
with the period f
s
because d(t) is discrete with sampling
period T
s
. Therefore, it is sucient to consider S
d
( f ) only
within the interval [0.5 f
s
, 0.5 f
s
[. Since d(t) is real-valued,
S
d
( f ) is even. Since d(t) is periodic with period LT
s
, S
d
( f ) is
discrete with the harmonics located at frequencies mf
s
/L,
m = 1, 2, . . . , oor (L/2) within the interval [0.5 f
s
, 0.5 f
s
[.
The spectral components of d(t) are shown in Figure 13a for
L = 5. When (5) is true, the images of the spectrum S
u1
( f )
Flexible Analog Front Ends of Recongurable Radios 377
of u(nT
s
) occupy the following bands within the interval
[0.5 f
s
, 0.5 f
s
[:
_
0.25 f
s
0.5B, 0.25 f
s
+ 0.5B
_

_
0.25 f
s
0.5B, 0.25 f
s
+ 0.5B
_
,
(29)
where B is a bandwidth of u(t). Figure 13b shows |S
u
( f )| and
the AFR |G
a
( f )| of antialiasing ltering performed by the SC
for f
0
= 0.25 f
s
. SpectrumS
e
( f ) is a convolution of S
u
( f ) and
S
d
( f ). Taking (5) into account, we get
S
e
( f ) =

_
m=
_
C
m
_
S
u
_
f f
s
_
m
L
0.25
__
+ S
u
_
f f
s
_
m
L
+ 0.25
____
.
(30)
Since e(t) is a real-valued discrete function with sampling pe-
riod T
s
, |S
e
( f )| is an even periodic function with the period
f
s
that is unique within the interval [0.5 f
s
, 0.5 f
s
[.
It follows from (30) that if L is even, the error image cor-
responding to m = L/2 falls to the frequencies 0.25 f
s
, that
is, within the signal spectrum. Therefore, S
e
( f ) and S
u
( f )
cannot be separated. When L is odd, the situation is dier-
ent. The centers of the images caused by the channel mis-
match are located at frequencies (r + 0.5) f
s
/(2L), where
r = 0, 1, . . ., 0.5(L 1) 1, 0.5(L 1) + 1, . . ., L 1 within
the interval [0.5 f
s
, 0.5 f
s
[. The bandwidth of each image is
B
1
= B + 2B
t1
, where B
t1
is the image one-sided transition
band. The images of S
e
( f ) are created by coecients C
m
.
Since these coecients are dierent, the images have dier-
ent transition bands. However, we assume for simplicity that
transition bands of all images are equal to those of the most
powerful image. Mean values of u(t) and e(t) are equal to
zero. Denoting the standard deviation of u(t) and e(t) as
u
and
e
, respectively, we can state that
u

e
. The standard
deviation
e1
of the most powerful spectral image of e(t) al-
ways meets condition
e1

e
. It is reasonable to assume
B
t
/B
t1
=
u
/
e1
= M where B
t
is the antialiasing-lter one-
sided transition band. Thus, B
t1
= B
t
/M and M > 1. Taking
into account that B
t
0.5 f
s
B, we obtain
B
1
B + 2
_
_
0.5 f
s
B
_
M
_
. (31)
Since channel ltering in the receiver DP removes all the
spectral components of e(t) outside bands (29), only the part
of S
e
( f ) which falls within these bands degrades the receiver
performance. It follows from (29) and (30) that S
e
( f ) and
S
u
( f ) do not overlap if (B + B
1
) f
s
/L. Inequality (31) al-
lows us to rewrite this condition as follows:
f
s
B
2L
(M1)
(ML)
and M L. (32)
According to (32), f
s
/B 2L when M . In practice,
M 100. Table 3 shows the minimum values of f
s
/B re-
quired to lter out e(t) when L is odd. It follows from Table 3
Table 3: Minimum values of f
s
/B.
M L 3 5 7 9 11
100 f
s
/B 6.1 10.4 14.9 19.6 24.5
1000 f
s
/B 6.01 10.04 14.08 18.15 22.22
that it is relatively easy to avoid an overlap of S
e
( f ) and S
u
( f )
and exclude an impact of the SCchannel mismatch on the re-
ceiver performance when L = 3. For odd L > 3, signicant
increase of f
s
is required. Consequently, combining the SCs
and sigma-delta A/Ds almost automatically excludes this im-
pact if L is odd.
When L is odd, but (32) is not met, S
e
( f ) and S
u
( f ) over-
lap. However, the overlap can be lowered by increasing f
s
/B
and, when L 5, by reducing the S
d
( f ) harmonics adjacent
to 0.5 f
s
since they create the closest-to-the-signal images of
S
e
( f ). Changing the succession of channel switching can re-
duce the harmonics. The succession that makes d(t) close to
a sampled sinewave minimizes the overlap.
Figure 13c shows |S
u1
( f )| and |S
e
( f )| for the situation
when L is odd and condition (32) is met. Here, the error im-
ages adjacent to the signal are created by C
2
, and the more
distant images by C
1
. The AFR |G
d
( f )| of the DP channel
lter is shown by the dashed line.
5.3. Compensation of channel mismatch in DP
If, despite all the measures, the residual error caused by the
mismatch still degrades the receiver performance, it can be
adaptively compensated in the DP. This compensation can
be performed either during the operation mode simultane-
ously with signal processing or during a separate calibration
mode. In all cases, channel gains g
l
are estimated rst, and
then deections d
l
are compensated.
There are many methods of fast channel gain estima-
tion in calibration mode. For example, when all the copies
w
n
(t) of w
0
(t) are simultaneously applied to the SC multipli-
ers and a test signal is sent to the SC input, estimation time is
T
e
= T
w
+LT
s
= (2L 1)T
s
, assuming that T
s
is required for
the hold and clear modes in each channel. A sinewave with
frequency f
0
is the simplest test signal. The estimation can
also be done when w
n
(t) are delayed relative to each other by
T
s
, like in the operational mode. If (5) is true and the test sig-
nal is a sinewave with frequency f
0
and arbitrary initial phase,
T
e
= 2T
w
+ (L + 1)T
s
= (3L 1)T
s
because two consecutive
samples are required for each channel to estimate its gain.
When the phase shift between the sinewave and the carrier of
w
0
(t) is equal to 45

, the estimation time can be reduced to


T
e
= T
w
+ LT
s
= (2L 1)T
s
.
Channel mismatch compensation performed during the
operation mode requires much longer estimation because
u(t) is a stochastic process. The block diagram of a simplied
version of such a compensator is shown in Figure 14. Here, a
demultiplexer (DMx) distributes digital words resulting from
the SC samples among L digital channels. Each digital chan-
nel corresponds to the SC channel with the same number.
Averaging units (AU) calculate the mean magnitudes of sam-
ples in each channel. The mean magnitudes are processed
378 EURASIP Journal on Wireless Communications and Networking
From
A/D
DMx Mx DF
GS
.
.
.
AU 1
AU L
1
.
.
.
L
1
.
.
.
L
K
1
K
L
. . .
Figure 14: Digital channel mismatch compensator.
RF strip
SC A/D
To DP
Figure 15: Modied receiver AMP architecture with digitization at
the RF.
in a gain scaler (GS), which generates coecients K
l
that
compensate the channel mismatch. Amultiplexer (Mx) com-
bines the outputs of all the channels. A subsequent digital l-
ter (DF) provides the main frequency selection. In practice,
channel mismatch compensation during the operation mode
requires the most statistically ecient methods of g
l
estima-
tion, and the compensator should be designed together with
automatic gain control (AGC) of the receiver.
6. AMPS ARCHITECTURES BASEDONSAMPLING
ANDRECONSTRUCTIONWITHINTERNAL
FILTERING
6.1. General
It is shown in Section 2 that the SDR front ends with band-
pass sampling, reconstruction, and antialiasing ltering po-
tentially provide the best performance. At the same time,
conventional methods of sampling, reconstruction, and l-
tering limit exibility of the front ends, complicate their IC
implementation, and do not allow achieving their potential
parameters. It follows from Section 3 that implementation of
the novel sampling and reconstruction techniques with inter-
nal ltering can eliminate these drawbacks and provide some
additional benets. Sections 4 and 5 demonstrate that chal-
lenges of the proposed techniques realization can be over-
come. The impact of these techniques on the architectures of
the radios AMPs is discussed below.
6.2. Modied receiver AMPs
Implementation of sampling with internal antialiasing lter-
ing in digital receivers requires modication of their front
ends. Since accumulation of the signal energy in the storage
capacitors of the SCs signicantly reduces the required gain
of AMPs, and antialiasing ltering performed by the SCs is
exible, it is reasonable to consider the possibility of signal
digitization at the receiver RF. This leads to the simplest AMP
architecture shown in Figure 15. Here, an RF strip performs
preltering and all the required amplication, an SC carries
out antialiasing ltering and sampling, and an A/D quantizes
the output of the SC. All further processing is performed in
a DP.
When multiplication of u
i
(t) by w
0
(t) is performed in the
analog domain, the carrier c(t) of w
0
(t) is a sinewave, the en-
velope W
0
(t) of w
0
(t) is a smooth function, and the AMP has
sucient dynamic range, preltering in the RF strip is used
only to limit the receiver frequency range R. The same type
of preltering can be used when c(t) is nonsinusoidal and/or
W
0
(t) is not a smooth function, but R is narrower than half
an octave. Such prelters do not require any adjustment dur-
ing frequency tuning.
If the conditions above are not met, the prelter band-
width should be narrower than R. Nonsinusoidal c(t) and
nonsmooth W
0
(t) require the prelter bandwidth that does
not exceed half an octave. In practice, the prelter bandwidth
is determined as a result of a tradeo. Indeed, on the one
hand, reduction of the prelter bandwidth allows increasing
its transition band. This simplies IC implementation of the
prelter. On the other hand, increase in the prelter band-
width simplies its frequency tuning.
In any case, signal u(t) intended for digitization is only
a part of u
i
(t), and u(t) usually has wider spectrum than a
desired signal s(t) since channel ltering is performed in the
DP. Therefore, a reasonable algorithm of the automatic gain
control (AGC) is as follows. The RF strip gain should be reg-
ulated by a control signal generated at the output of a digital
channel lter and constraints generated at the input of the
SC and at the output of the D/A. These constraints prevent
clipping of u(t) caused by powerful interference, which is l-
tered out by the digital channel lter, and clipping of u
i
(t)
caused by powerful interference, which is ltered out by the
SC. To compensate level variations due to the constraints,
feed-forward automatic scaling is usually required in the DP
with xed-point calculations.
Reconguration or adaptation of the receiver at the same
f
0
usually can be achieved by varying only W
0
(t). Frequency
tuning requires shifting the AFR of the SC along the fre-
quency axis and, sometimes, adjusting the prelter AFR. The
AMP has to carry out only coarse frequency tuning. Fine
tuning with the required accuracy can be performed in the
receiver DP. The reasonable increment f of coarse tun-
ing is about 0.1B, where B is the bandwidth of u(t). Thus,
the number of dierent center frequencies f
0
within the fre-
quency range R is about 10 R/B. In most cases, coarse tun-
ing requires changing both c(t) and W
0
(t). Indeed, when
f
0
is changed, usually f
s
should also be changed to preserve
condition (5). This in turn necessitates changing W
0
(t) be-
cause certain relations between f
s
and T
w
are necessary to
suppress noise and interference within intervals (6). During
coarse tuning, W
0
(t) can remain unchanged only when pre-
vious and subsequent frequencies f
0
have the same optimal
f
s
and keeping unchanged W
0
(t) does not cause additional
discontinuities in w
0
(t). However, this happens rarely, and
frequency tuning in the AMP shown in Figure 15 is rela-
tively complex. The SCs described in Section 4.3 cannot be
used in this architecture due to possible leakage of the c(t)
generator.
Flexible Analog Front Ends of Recongurable Radios 379
RF
section
LPF
IF strip
Low-Q
lter
A/D SC
To DP
LO
Figure 16: Modied superheterodyne receiver AMP architecture
with sampling at the IF.
A superheterodyne architecture of the receiver AMP
modied to accommodate sampling with internal antialias-
ing ltering at the IF is shown in Figure 16. Compared to the
previous architecture, this one simplies both frequency tun-
ing and preltering. Here, an RF section performs image re-
jection and preliminary amplication of the sum of a desired
signal, noise, and interference. Preltering and further signal
amplication are carried out in an IF strip. This preltering
is performed by a lter with low quality factor (Q) that can
be implemented on a chip.
In principle, preltering is necessary only when c(t) is
nonsinusoidal and/or W
0
(t) is not a smooth function. Oth-
erwise, it can be excluded. However, as shown in Section 4.2,
use of a K-level W
0
(t) and a squarewave carrier c(t) radi-
cally simplies the SC due to reducing multiplying D/As to
a relatively small number of switches. Besides, it allows in-
creasing the receiver IF, which, in turn, simplies image re-
jection in the RF section. When the receiver frequency range
R is wide, a variable IF allows avoiding spurious responses. In
practice, two or three dierent f
0
s are sucient, and they can
be selected so that transitions from one f
0
to another require
minimum adjustment. For example, these transitions may
require changing only c(t). If these frequencies are within the
bandwidth of the low-Q lter, the latter does not require any
adjustment when the IF is changed. In both AMP architec-
tures shown in Figures 15 and 16, all measures that reduce
the inuence of the SCchannel mismatch on the receiver per-
formance (see Section 5) should be taken. Therefore, when
condition (32) is not met, digital channel mismatch compen-
sator has to be implemented in the receiver DP.
Despite the dierences, the AMP architectures in Figures
15 and 16 utilize the advantages of sampling with internal
antialiasing ltering (see Section 3.5). First of all, removal
of high-quality conventional antialiasing lters (e.g., SAW,
crystal, mechanical, ceramic) and implementation of the SCs
with variable w
0
(t) make these architectures realizable on a
chip, recongurable, and adaptive. Then, the proposed sam-
pling signicantly improves performance by adding to the
benets of bandpass sampling described in Section 2.1 the
following advantages. A variable IF allows avoiding spuri-
ous responses in a superheterodyne AMP. Nonlinear distor-
tions are radically reduced due to rejection of out-of-band
IMPs of all preceding stages and lower input current of the
SC caused by accumulation of the signal energy. This accu-
mulation also lters out jitter, improving performance and
speed of the A/D. Finally, the accumulation of signal energy
lowers the required AMP gain and allows sampling close to
the antenna.
D/A RC RF strip PA
From DP
Figure 17: Modied transmitter AMP architecture with recon-
struction at the RF.
IF strip
Low-Q
lter
From DP
D/A RC
LO
BPF PA
Figure 18: Modied oset up-conversion transmitter AMP archi-
tecture with reconstruction at the IF.
6.3. Modied transmitter AMPs
Similar to the case of receivers, implementation of recon-
struction with internal ltering in transmitters requires
modication of their AMPs. This modication aects only
the transmitter drive (exciter) and does not inuence the
transmitter PA. Signal reconstruction at the transmitter RF
leads to the simplest AMP architecture shown in Figure 17.
Here, digital words corresponding to the samples of a band-
pass signal are formed in a DP at the transmitter RF. Then,
they are converted to the analog samples by a D/A. An RC
reconstructs the bandpass analog signal and carries out main
analog ltering. A subsequent RF strip amplies the signal
to the level required at the input of a PA and performs post-
ltering. This postltering is absolutely necessary when c(t)
is nonsinusoidal and/or W
0
(t) is not a smooth function. Al-
though the AMP in Figure 17 looks simple, its implementa-
tion causes problems related to frequency tuning of the trans-
mitter and digital-to-analog conversion of bandpass signals
at the varying RF.
These problems are solved in the oset up-conversion
AMP architecture modied to accommodate bandpass re-
construction with internal ltering at the IF shown in
Figure 18. The fact that reconstruction, preliminary ampli-
cation, and postltering of bandpass analog signals are car-
ried out at the transmitter IF signicantly simplies realiza-
tion of this procedures. An RCperforms main reconstruction
ltering, while postltering is carried out by a low-Q IF lter
that can be placed on a chip.
Implementation of reconstruction with internal exible
ltering makes the transmitter AMPs easily recongurable
and highly adaptive and increases their scale of integration.
This reconstruction also reduces the required AMP gain due
to more ecient utilization of the D/A output current. As a
result, reconstruction can be performed closer to the antenna
than in conventional architectures.
7. CONCLUSIONS
In modern SDRs, analog front end architectures with band-
pass sampling, reconstruction, and antialiasing ltering can
380 EURASIP Journal on Wireless Communications and Networking
potentially provide the best performance of both receivers
and transmitters. However, conventional methods of per-
forming these procedures limit exibility, complicate IC im-
plementation, and do not allow achieving the potential per-
formance of the radios.
Novel sampling and reconstruction techniques with in-
ternal ltering derived from the sampling theorem elimi-
nate these problems. The techniques provide high exibility
because their ltering and other properties are determined
by weight functions w
0
(t) that can be dynamically changed.
Since technology of the SCs and RCs with internal ltering is
compatible with IC technology, they radically increase scale
of integration in the AMPs. The RCs provide more ecient
utilization of the D/Aoutput current than conventional tech-
niques. The SCs accumulate the input signal energy. This
accumulation lters out jitter, improving performance and
speed of A/Ds, and reduces the input current. The reduc-
tion of the input current lowers nonlinear distortions and
required gain of AMPs.
Technical challenges of the SCs and RCs practical real-
ization can be overcome by proper selection of w
0
(t) and
optimization of their architectures for a given w
0
(t). Selec-
tion of w
0
(t) requires multiple tradeos. Simplication of
the SCs and RCs is usually intended to reduce complexity
and/or number of multiplications. Minimum complexity of
multiplications is achieved when multiplying D/As or ana-
log multipliers can be replaced by a relatively small number
of switches. This can be accomplished, for instance, by us-
ing w
0
(t) with K-level envelope W
0
(t) and squarewave car-
rier c(t) when W
0
(t) and c(t) are properly synchronized and
f
0
/ f
s
is adequately high ( f
0
/ f
s
> 3 is usually sucient). When
f
0
/ f
s
is low, c(t) should also be a several-level function. There
are other classes of w
0
(t) that allow replacing multipliers by
a small number of switches.
Separate multiplications of the input signal u
i
(t) by
W
0
(t) and c(t) and use of only two multipliers for multiply-
ing by c(t) lead to a method that signicantly simplies the
SCs. Although this is achieved at the expense of slightly re-
duced performance compared to the canonical SCs, the sim-
plied SCs still provide signicantly better performance of
the radios than conventional sampling.
Increase of f
s
/B simplies antialiasing and reconstruc-
tion ltering and allows reduction of L in some cases. When
both f
s
/B and f
0
/B are suciently high, use of WFG outputs
orthogonality allows reduction of N for a given L. However,
this method is practical only for very high f
s
/B.
Since SCs and RCs with internal ltering are inherently
multichannel, the impact of channel mismatch on the per-
formance of SDRs should be minimized. There are three ap-
proaches to the problem. The rst of them includes all tech-
nical and technological measures that reduce the mismatch.
The second one is based on preventing an overlap of signal
and mismatch error spectra. This can be achieved only when
L is odd, and condition (32) is met. In this case, the error
spectrum can be ltered out in the DP. Combination of the
SCs with odd L and sigma-delta A/Ds almost automatically
excludes the overlap. When L is odd, but condition (32) is
not met, the overlap cannot be avoided. However, it can be
lowered by increasing f
s
/B and, when L 5, by reducing the
S
d
( f ) harmonics adjacent to 0.5 f
s
. The third approach is
based on adaptive compensation of the channel mismatch in
the DP.
In principle, sampling and reconstruction with internal
ltering can be carried out at the radios RFs. However, fre-
quency conversion to an IF signicantly simplies practical
realization of the modied SDRs.
Implementation of the SCs and RCs with internal lter-
ing in SDRs radically increases recongurability, adaptivity
and scale of integration of their front ends. Simultaneously, it
improves performance of the radios due to signicant reduc-
tion of nonlinear distortions, rejection of out-of-band noise
and IMPs of all stages preceding sampling, avoiding spurious
responses, and ltering out jitter. This implementation also
substantially reduces front ends of SDRs, enabling sampling
and reconstruction close to the antenna.
REFERENCES
[1] T. Anderson and J. W. Whikohart, A digital signal processing
HF receiver, in Proc. 3rd International Conference on Com-
munication Systems & Techniques, pp. 8993, London, UK,
February 1985.
[2] C. M. Rader, A simple method for sampling in-phase and
quadrature components, IEEE Trans. Aerosp. Electron. Syst.,
vol. 20, no. 6, pp. 821824, 1984.
[3] M. V. Zarubinskiy and Y. S. Poberezhskiy, Formation of read-
outs of quadrature components in digital receivers, Telecom-
munications and Radio Engineering, vol. 40/41, no. 2, pp. 115
118, 1986.
[4] Y. S. Poberezhskiy, Digital Radio Receivers, Radio & Commu-
nications, Moscow, Russia, 1987 (Russian).
[5] J. B.-Y. Tsui, Digital Microwave Receivers: Theory and Concepts,
Artech House, Norwood, Mass, USA, 1989.
[6] M. E. Frerking, Digital Signal Processing in Communication
Systems, Van Nostrand Reinhold, New York, NY, USA, 1994.
[7] W. E. Sabin and E. O. Schoenike, Eds., Single-Sideband Sys-
tems and Circuits, McGraw-Hill, New York, NY, USA, 2nd edi-
tion, 1995.
[8] J. Mitola III, The software radio architecture, IEEE Com-
mun. Mag., vol. 33, no. 5, pp. 2638, 1995.
[9] R. I. Lackey and D. W. Upmal, Speakeasy: the military soft-
ware radio, IEEE Commun. Mag., vol. 33, no. 5, pp. 5661,
1995.
[10] B. Razavi, Recent advances in RF integrated circuits, IEEE
Commun. Mag., vol. 35, no. 12, pp. 3643, 1997.
[11] H. Meyr, M. Moeneclaey, and S. A. Fechtel, Digital Commu-
nications Receivers, John Willey & Sons, New York, NY, USA,
1998.
[12] A. A. Abidi, CMOS wireless transceivers: the newwave, IEEE
Commun. Mag., vol. 37, no. 8, pp. 119124, 1999.
[13] J. Mitola III, Software Radio Architecture, John Willey & Sons,
New York, NY, USA, 2000.
[14] C. Chien, Digital Radio Systems on a Chip: a System Approach,
Kluwer Academic, Boston, Mass, USA, 2000.
[15] M. Helfenstein and G. S. Moschytz, Eds., Circuits and Sys-
tems for Wireless Communications, Kluwer Academic, Boston,
Mass, USA, 2000.
[16] Y. Sun, Ed., Wireless Communication Circuits and Systems, IEE,
London, UK, 2004.
[17] Y. S. Poberezhskiy and G. Y. Poberezhskiy, Sampling with
weighted integration for digital receivers, in Proc. Digest of
Flexible Analog Front Ends of Recongurable Radios 381
IEEE MTT-S Symposium on Technologies for Wireless Appli-
cations, pp. 163168, Vancouver, British Columbia, Canada,
February 1999.
[18] Y. S. Poberezhskiy and G. Y. Poberezhskiy, Sampling tech-
nique allowing exclusion of antialiasing lter, Electronics Let-
ters, vol. 36, no. 4, pp. 297298, 2000.
[19] Y. S. Poberezhskiy and G. Y. Poberezhskiy, Sample-and-hold
ampliers performing internal antialiasing ltering and their
applications in digital receivers, in Proc. IEEE Int. Symp. Cir-
cuits and Systems (ISCAS 00), vol. 3, pp. 439442, Geneva,
Switzerland, May 2000.
[20] Y. S. Poberezhskiy and G. Y. Poberezhskiy, Sampling algo-
rithm simplifying VLSI implementation of digital receivers,
IEEE Signal Processing Lett., vol. 8, no. 3, pp. 9092, 2001.
[21] Y. S. Poberezhskiy and G. Y. Poberezhskiy, Signal reconstruc-
tion technique allowing exclusion of antialiasing lter, Elec-
tronics Letters, vol. 37, no. 3, pp. 199200, 2001.
[22] Y. S. Poberezhskiy and G. Y. Poberezhskiy, Sampling and sig-
nal reconstruction structures performing internal antialias-
ing ltering, in Proc. 9th International Conference on Elec-
tronics, Circuits and Systems (ICECS 02), vol. 1, pp. 2124,
Dubrovnik, Croatia, September 2002.
[23] Y. S. Poberezhskiy and G. Y. Poberezhskiy, Sampling and sig-
nal reconstruction circuits performing internal antialiasing
ltering and their inuence on the design of digital receivers
and transmitters, IEEE Trans. Circuits Syst I: Regular Papers,
vol. 51, no. 1, pp. 118129, 2004, erratum ibid. vol. 51, no. 6,
p. 1234, 2004.
Yem S. Poberezhskiy received the
M.S.E.E. degree from the National Tech-
nical University Kharkov Polytechnic
Institute, Kharkov, Ukraine, and the Ph.D.
degree in radio communications from
Moscow Radio Communications R&D
Institute, Moscow, Russia, in 1971. He has
held responsible positions in both industry
and academia. Currently, he is a Senior
Scientist at Rockwell Scientic Company,
Thousand Oaks, Calif. He is an author of over 200 publications and
30 inventions. A book Digital Radio Receivers (Moscow: Radio &
Communications, 1987, in Russian) is among his publications. His
current major research interests include communication systems;
theory of signals, circuits, and systems; mixed-signal processing;
digital signal processing; modulation/demodulation; synchroniza-
tion; and architecture of digital receivers and transmitters.
Gennady Y. Poberezhskiy received the
M.S.E.E. degree (with the highest honors)
from Moscow Aviation Institute, Russia, in
1993. He has held systems engineering posi-
tions in a number of companies. Currently,
he is a Principal Engineer at Raytheon Space
and Airborne Systems, El Segundo, Calif.
He is an author of 20 publications. His cur-
rent research interests include communica-
tion systems, mixed-signal processing, digi-
tal signal processing, communication, and GPS receivers.
EURASIP Journal on Wireless Communications and Networking 2005:3, 382389
c 2005 Bedri A. Cetiner et al.
A Recongurable Spiral Antenna
for Adaptive MIMOSystems
Bedri A. Cetiner
Space Science Center, Morehead State University, Morehead, KY 40351, USA
Email: ba.cetiner@moreheadstate.edu
J. Y. Qian
Department of Electrical Engineering and Computer Science, University of California at Irvine, Irvine, CA 92697-2625, USA
Email: qianj@uci.edu
G. P. Li
Department of Electrical Engineering and Computer Science, University of California at Irvine, Irvine, CA 92697-2625, USA
Email: gpli@uci.edu
F. De Flaviis
Department of Electrical Engineering and Computer Science, University of California at Irvine, Irvine, CA 92697-2625, USA
Email: franco@uci.edu
Received 17 October 2004; Revised 15 March 2005
We present a recongurable spiral antenna for use in adaptive MIMO systems. The antenna is capable of changing the sense of
polarization of the radiated eld. It is fabricated by using an RF-MEMS technology compatible with microwave laminate substrates
developed within the authors group. The proposed antenna structure is built on a number of rectangular-shaped bent metallic
strips interconnected to each other with RF-MEMS actuators. Two senses of polarization, RHCP and LHCP, are achieved by
conguring the physical structure of the antenna, that is, by changing the winding sense of the spiral, through judicious activation
of MEM actuators. The fabrication process for the monolithic integration of MEM actuators with bent microstrip pixels on
RO4003-FR4 microwave laminate substrate is described. The measured and calculated radiation and impedance characteristics
of the antenna are given. The operating frequency of the presented antenna design can easily be adjusted to be compatible with
popular IEEE networking standards such as 802.11a.
Keywords and phrases: adaptive MIMOsystems, recongurable spiral antenna, radio-frequency microelectromechanical systems.
1. INTRODUCTION
Recongurable wireless communication systems, which can
dynamically adapt themselves to constantly changing envi-
ronmental propagation characteristics, will be the key for
the next-generation communication scenarios. A communi-
cation system, capable of changing its output dynamically
through recongurability features, allows optimal system-
level performance at all times, regardless of changing char-
acteristics of the communication environment.
There has recently been enormous research performed
on MIMO systems [1] with associated technologies such as
This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
smart antennas and adaptive coding and modulation tech-
niques, which have been proven to dramatically increase
the wireless channel capacity and improve the diversity. Al-
though in these studies considerable attention has been given
to the performance analysis of these systems in the context
of coding and signal processing architectures, the investiga-
tion of the antenna aspect is mainly limited to the impact
of the number of antenna elements with little consideration
on their radiation and polarization characteristics as well as
array geometry. Multiple antenna elements of these systems
are xed by the initial design and cannot change their prop-
erties, that is, radiation pattern, polarization, and operating
frequency.
We have recently developed a microwave-laminate-
compatible RF-MEMS technology [2, 3, 4] that allows
fabricating multifunction recongurable antennas (MRAs)
A Recongurable Spiral Antenna for Adaptive MIMO Systems 383
on microwave laminate substrates that best meet the an-
tenna performance characteristics. An MRA can dynami-
cally congure its structural architecture and thus alter its
performance properties, that is, polarization, radiation pat-
tern, and operating frequency. Therefore, an adaptive MIMO
system equipped with MRAs will not be constrained to em-
ploy a xed antenna design over varying channel condi-
tions. This is an additional degree of freedom in adapt-
able parameters of an adaptive MIMO system and permits
the selection of the best antenna properties and congura-
tion in conjunction with the adapted transmission scheme
with respect to the channel condition [5]. Thus the gap be-
tween theoretical MIMO performance and practice is mini-
mized.
Motivated by the features of next-generation wireless
MIMO communication systems, as stated above, we have
aimed at developing innovative antenna architectures, which
combine multiple functions in one single antenna. One such
application example, a spiral antenna capable of changing its
polarization state through microwave-laminate-compatible
RF-MEMS technology, is presented in this paper. In this
application example, we only recongure the polarization
property, but there would be no process limitation to accom-
modate recongurability features in operation frequency and
radiation characteristics of a more complex recongurable
antenna design.
2. ANTENNA STRUCTURE, OPERATIONAL
MECHANISMFABRICATION, ANDRESULTS
Spiral antennas are attractive for communication appli-
cations where broadband characteristics with respect to
both input impedance and radiation pattern are required.
There have been extensive investigations regarding radiation
characteristics of spiral antennas with dierent geometri-
cal shapes such as circular, rectangular, and eccentric [6, 7].
These antennas are mainly used to radiate circularly po-
larized wave forming either an axial beamnormal to the
plane of the spiralor tilted beam patterno-normal to
the plane of the spiral [8]. The single-arm spiral, which is
used in this work, has the advantage of not requiring a balun
circuit between the spiral and the feed line, which is needed
for multiarm spiral antennas.
2.1. Antenna structure and operational mechanism
While the microwave-laminate-substrate-compatible RF-
MEMS technology has been used in [2, 4] for monolithic in-
tegration of antenna elements with RF-MEMS switches, the
role of RF-MEMS switches, which are located on antenna
feed lines, has been limited to routing the RF signal feed-
ing the antennas. In this work, we integrate a number of RF-
MEMS actuators within the geometrical structure of the an-
tenna to construct a recongurable spiral antenna. In other
words, RF-MEMS actuators are used as part of the physical
structure of the antenna, owing to the monolithic integra-
tion capability of the processing technique, providing a large
degree of structural recongurability.
The proposed recongurable spiral antenna architec-
ture is built on a number of printed rectangular-shaped
metal strips interconnected by RF-MEMS actuators on a
microwave-laminate printed circuit board (PCB) substrate,
RO4003-FR4 (
r
= 3.38, tan = 0.002). Shown in Figure 1
are two adjacent strips interconnected by an RF-MEMS ac-
tuator, which is made of a metallic movable membrane, sus-
pended over a metal stub protruding from an adjacent strip,
xed to both ends of the strip through metallic posts. The op-
timized height of these posts was found to be 8 m for a good
tradeo between up-position switch coupling and actuation
voltage. Metal stubs are covered by silicon-nitride (SiN
x
) lm
to prevent metallic membrane from sticking onto the stub
upon contact. This lm also provides a capacitive contact
for actuator down state isolating RF signal from DC. A DC
bias voltage of approximately 50 Vapplied between the mem-
brane and the stub causes an electrostatic force that pulls
the suspended membrane on top of the stub (actuator down
state or actuator on) and the actuator connects the strips
(see Figure 1c); otherwise strips are disconnected (actuator
up state or actuator o) (see Figure 1b). Judicious activation
of interconnecting actuators, that is, by keeping some of the
actuators in the up position (zero bias) while activating the
rest of them by applying DC bias voltages, allows the recon-
gurable spiral to congure its architecture into single-arm
rectangular spirals with opposite winding sense of the spi-
ral, left or right senses (see Figure 2). Accordingly, right- and
left-hand circularly polarized (RHCP and LHCP) radiation is
achieved. In Figure 2, for the clarity of illustration, each con-
gured geometry is depicted separately and actuators in the
up state are shown without metallic membrane. The antenna
is fed by a single coaxial probe as shown in Figure 2c. The
supply voltage is connected to the proper locations on the
antenna segments through resistive bias lines so as to prevent
RF signal from being shorted by the DC power supply.
The proposed prototype antenna is aimed to radiate an
axial beam of RHCP and LHCP elds. It is known that a
single-arm rectangular spiral antenna with outermost arm
peripheral length (circumference) of C,
1
e
< C < 2
e
, (1)
excites only the rst radiation mode giving rise to an ax-
ial beam of circular polarization [8], where
e
=
0
/[(
r
+
1)/2]
1/2
is the eective wavelength of the current traveling on
the spiral. The strip number and size are optimized so that
circumference of the antenna, C = 42 mm = 1.04
e
, satis-
es (1), and minimum number of actuators with associated
bias circuitries are needed.
2.2. Fabrication and results
For reference, as a rst step, conventional single-arm rect-
angular spiral antennas radiating circularly polarized eld
along their axes have been designed, fabricated, and charac-
terized. We chose to use RO4003-FR4 (
r
= 3.38, tan =
0.002) microwave laminated substrate [9] to realize the an-
tennas due to its low cost and widespread use in wireless sys-
tems. The substrate is conductor-backed to ensure that the
384 EURASIP Journal on Wireless Communications and Networking
Aluminum
PCB copper
Substrate
SiN
x
w
1
w
2
A A

w
3
w
2
(a)
Aluminum
Electroplated copper
PCB copper
Substrate
SiNx
PCB
g
A A

t1
t2
(b)
Aluminum
Electroplated copper
PCB copper
Substrate
SiNx
h1
h2
(c)
Figure 1: RF-MEMS actuator interconnecting two adjacent metallic strips. (a) Top view; width of metal strip w
1
= 800 m; width of stub
w
2
= 100 m; width of membrane w
3
= 150 m. (b) Side view (up position); thickness of nitride t
1
= 0.2 m; thickness of membrane
t
2
= 0.5 m; air gap g = 7.8 m. (c) Side view (down position); thickness of electroplated copper h
1
= 8 m; thickness of PCB copper
h
2
= 16 m.
Coax feeding
point
50 V
0 V
50 V
0 V
0 V
50 V
(a)
Actuator
in the up
state
Actuator
in the down
state
50 V
50 V
0 V
0 V
50 V
0 V
x
y
z (coax feed)
(b)

r
= 3.38 H
Coaxial cable
(c)
Figure 2: Schematics of the single-arm rectangular spiral antennas, which are recongured from the proposed recongurable spiral archi-
tecture by judicious activation of the interconnecting RF-MEMS actuators for (a) left-hand circular polarization, (b) right-hand circular
polarization, and (c) side view of the antenna. The outermost dimensions of the antenna are 9 12 (mm), the spiral line width is 0.8 mm.
A Recongurable Spiral Antenna for Adaptive MIMO Systems 385
DC bias path
Coax inner
conductor
Spiral
segments
(a)
(b)
(c)
RF-MEMS
actuator
(d)
Figure 3: Fabrication sequence for monolithic integration of RF-MEMS actuators with rectangular-shaped strip segments in constructing
recongurable spiral antenna. (a) Antenna pattern, DC bias path, and via formation. (b) Dielectric layer deposition and sacricial layer
planarization. (c) Aluminum (Al) membrane deposition. (d) Final release.
Figure 4: Photograph of the microfabricated recongurable spiral
antenna.
antenna radiates broadside to the printed spiral surface. Sub-
strate thickness is chosen to be 7.6 mm, which is one of the
standard thicknesses for PCB family substrates, the closest
one to the quarter wavelength at a center design frequency
of 5 GHz. Theoretical characterization of the antenna struc-
ture is conducted by using Ansoft HFSS 8.5 full-wave anal-
ysis tool [10] based on nite element method which takes
into account the edge eects due to nite-size dielectric and
conducting plane of the antenna.
A brief fabrication sequence for monolithic integration
of RF-MEMS actuators with rectangular-shaped strip seg-
ments of the spiral antenna is given in Figure 3. Details of
the fabrication process can be found in [2, 3, 4], so here
we only briey explain the process. The fabrication begins
with RO4003 laminate with copper layers of 16 micron on
Calculated
Measured
3 3.5 4 4.5 5 5.5 6
Frequency (GHz)
30
25
20
15
10
5
0
R
e
t
u
r
n
l
o
s
s
(
d
B
)
Figure 5: Return loss of the antenna for RHCP radiation.
both sides. We rst form the segments of the antenna and
the planar part of the bias circuitry by wet-etching the cop-
per layer. Vertical vias for bias circuitry and coax feed are
created by standard PCB processes. After this step, a thin
layer of HDICP CVD SiN
x
is deposited and etched by re-
active ion etching such that the SiN
x
covers only the tips of
the metal stubs protruding from the antenna segments (see
Figure 1a). We continue fabricating RF-MEMS actuators fol-
lowing the process ow shown in Figures 3b and 3d without
aecting the antenna structure. A photograph of the micro-
fabricated antenna is shown in Figure 4, and Figure 5 shows
386 EURASIP Journal on Wireless Communications and Networking
Calculated E
R
Measured E
R
Calculated E
L
Measured E
L
0
45
90
135
180
225
270
315
10 20 30
20
10
(a)
Calculated E
R
Measured E
R
Calculated E
L
Measured E
L
0
45
90
135
180
225
270
315
10 20 30
20
10
(b)
Calculated E
R
Measured E
R
Calculated E
L
Measured E
L
0
45
90
135
180
225
270
315
10 20 30
20
10
(c)
Calculated E
R
Measured E
R
Calculated E
L
Measured E
L
0
45
90
135
180
225
270
315
10
20
30
20
10
(d)
Figure 6: Radiation patterns for the RHCP spiral antenna at 5 GHz in (a) = 0 plane, (b) = 45 plane, (c) = 90 plane, and (d) = 135
plane.
the return loss of the spiral antenna with counterclockwise
sense of winding corresponding to the RHCP radiation. The
simulated result is also validated by comparison with experi-
mental data in this gure. Due to the symmetry between two
antenna congurations, the RHCP and LHCP spirals exhib-
ited almost identical return loss with a VSWR of less than
two covering the frequency band of 4.35.4 GHz. The size
of the antenna geometry can be scaled to change the oper-
ational bandwidth to make it compatible with popular IEEE
networking standards such as IEEE 802.11a.
Measured and calculated radiation patterns at 5 GHz in
four dierent planes of = 0

, 45

, 90

, 135

are shown in
Figures 6 and 7 for RHCP and LHCP spirals, respectively.
As seen from these gures, the antennas radiate circularly
polarized wave slightly o-broadside to the plane of spiral,
forming an almost axial beam pattern. This slight tilt from
the z-axis is due to the asymmetry of the antenna struc-
ture with respect to the z-axis. The measured average half-
power beamwidth (HPBW) is approximately 105

. The an-
tenna radiates almost entirely circularly polarized wave in the
z-direction with axial ratio value of 0.9 dB. The gain at this
direction is 5.3 dB. Variations of axial ratio and gain in the
z-direction with respect to frequency are shown in Figure 8.
The circular polarization bandwidth over which the axial ra-
tio is less than 3 dB is approximately 11%. Gain of the an-
tenna with average value of 4.9 dB shows small variation over
this bandwidth. The dierence in performance characteris-
tics between the RF-MEMS integrated spiral antenna and
conventional single-arm rectangular spiral antenna was ob-
served to be negligible.
A Recongurable Spiral Antenna for Adaptive MIMO Systems 387
Calculated E
R
Measured E
R
Calculated E
L
Measured E
L
0
45
90
135
180
225
270
315
10 20
30
20
10
(a)
Calculated E
R
Measured E
R
Calculated E
L
Measured E
L
0
45
90
135
180
225
270
315
10 20 30
20
10
(b)
Calculated E
R
Measured E
R
Calculated E
L
Measured E
L
0
45
90
135
180
225
270
315
10 20
30
20
10
(c)
Calculated E
R
Measured E
R
Calculated E
L
Measured E
L
0
45
90
135
180
225
270
315
10 20 30
20
10
(d)
Figure 7: Radiation patterns for the LHCP spiral antenna at 5 GHz in (a) = 0 plane, (b) = 45 plane, (c) = 90 plane, and (d) = 135
plane.
3. CONCLUSION
Motivated by the fact that the antenna properties (polariza-
tion, operating frequency, and radiation behavior) can be
used as additional degrees of freedom in adaptive MIMO
system parameters, we presented a recongurable antenna
architecture employing RF-MEMS as a vehicle to achieve po-
larization recongurability. The antenna builds on a num-
ber of rectangular-shaped metallic strips monolithically in-
terconnected with RF-MEMS actuators. Its architecture is
dynamically recongured into RHCP and LHCP single-arm
rectangular spirals with opposite sense of windings by acti-
vating some of the actuators while keeping the rest in the
o-state. RF-MEMS technology compatible with microwave
laminate substrates is the key enabling multifunctional
recongurable antenna systems with MEMS integration at
low cost with high system-level performances. The dening
feature of this technology is its capability of allowing mono-
lithic integration of RF-MEMS, with antenna structures on
any microwave laminate substrate that best meets the an-
tenna performance characteristics. Experimental impedance
and radiation characteristics of the proposed architecture are
in excellent agreement with theoretical results. The results
showed that the antennas radiate right-hand and left-hand
circularly polarized axial beam waves with good axial ratio
and gain values covering the design frequency bandwidth of
4.35.4 GHz. If desired, this bandwidth can be changed by
scaling the size of the antenna to make it compatible with
IEEE networking standards such as 802.11a. The presented
application example has been intended to demonstrate an
388 EURASIP Journal on Wireless Communications and Networking
Calculated axial ratio
Measured axial ratio
Calculated gain
Measured gain
4.6 4.8 5 5.2 5.4 5.6
Frequency (GHz)
0
3
6
9
12
A
x
i
a
l
r
a
t
i
o
a
n
d
g
a
i
n
(
d
B
)
Figure 8: Frequency responses of axial ratio and gain in the z-
direction for RHCP spiral antenna.
initial design and fabrication that will pave the way for novel
antennas into which multifunctional features are dynami-
cally combined by making use of large number of actua-
tors. Multifunction recongurable antenna is very promis-
ing in the establishment of the next-generation multifunc-
tion highly integrated recongurable communication archi-
tectures.
ACKNOWLEDGMENTS
This work was supported in part by the US Air Force under
Grant F04611-03-C-004. The authors would like to acknowl-
edge the stas of the Integrated Nanosystems Research Facil-
ity clean room, University of California, Irvine, for their help
in this work.
REFERENCES
[1] Special Issue on MIMO Systems and Applications, IEEE J. Se-
lect. Areas Commun., vol. 21, no. 3, April 2003.
[2] B. A. Cetiner, J. Y. Qian, H.-P. Chang, M. Bachman, G. P.
Li, and F. De Flaviis, Monolithic integration of RF MEMS
switches with a diversity antenna on PCB substrate, IEEE
Trans. Microwave Theory Tech., vol. 51, no. 1, pp. 332335,
2003.
[3] H.-P. Chang, J. Y. Qian, B. A. Cetiner, F. De Flaviis, M.
Bachman, and G. P. Li, RF MEMS switches fabricated on
microwave-laminate printed circuit boards, IEEE Electron
Device Lett., vol. 24, no. 4, pp. 227229, 2003.
[4] B. A. Cetiner, J. Y. Qian, H. P. Chang, M. Bachman, G. P.
Li, and F. De Flaviis, Microwave laminate PCB compatible
RF MEMS technology for wireless communication systems,
in Proc. IEEE Antennas and Propagation Society International
Symposium, vol. 1, pp. 387390, Columbus, Ohio, USA, June
2003.
[5] B. A. Cetiner, H. Jafarkhani, J. Y. Qian, H. J. Yoo, A. Grau,
and F. De Flaviis, Multifunctional recongurable MEMS in-
tegrated antennas for adaptive MIMO systems, IEEE Com-
mun. Mag., vol. 42, no. 12, pp. 6270, 2004.
[6] J. A. Kaiser, The archimedean two-wire spiral antenna, IRE
Transactions on Antennas and Propagation, vol. 8, pp. 312323,
1960.
[7] R. T. Gloutak Jr. and N. G. Alexopoulous, Two-arm eccentric
spiral antenna, IEEE Trans. Antennas Propagat., vol. 45, no. 4,
pp. 723730, 1997.
[8] H. Nakano, J. Eto, Y. Okabe, and J. Yamauchi, Tilted-
and axial-beam formation by a single-arm rectangular spi-
ral antenna with compact dielectric substrate and conducting
plane, IEEE Trans. Antennas Propagat., vol. 50, no. 1, pp. 17
24, 2002.
[9] Microwave Products Technical Information, Rogers, Chandler,
Ariz, USA, 2001.
[10] Ansoft HFSS 8.5, Ansoft Corporation, Pittsburgh, Pa, USA.
Bedri A. Cetiner was born in Gazi Magosa,
Cyprus, in 1969. He received the Ph.D. de-
gree in electronics and communications en-
gineering from the Yildiz Technical Uni-
versity, Istanbul, in 1999. From November
1999 to June 2000 he was with the Univer-
sity of California, Los Angeles, as a North-
ern Atlantic Treaty Organization (NATO)
Science Fellow. He then joined the Depart-
ment of Electrical Engineering and Com-
puter Science, the University of California (UCI), Irvine, where he
worked as a postdoctoral researcher and Research Specialist from
June 2000 to June 2004. He is currently an Assistant Professor at
the Space Science Center, Morehead State University. His research
interest is focused on the analysis and design of microwave circuits
and RF-MEMS microwave devices and systems. He is recently con-
centrating on the applications of RF-MEMS to space science sys-
tems and a new class of recongurable antennas for use in adaptive
multi-input multi-output (MIMO) systems. He is the principal in-
ventor of microwave-laminate-compatible RF-MEMS technology.
He is a Member of the IEEE Antennas and Propagation, Microwave
Theory and Techniques, and Communications Societies.
J. Y. Qian received a B.S. degree in opto-
electronics and an M.S. degree in optics
from the University of Science and Technol-
ogy of China, in 1994 and 1997, respectively,
and a Ph.D. degree in electrical engineer-
ing and computer science from the Univer-
sity of California (UCI), Irvine, in 2004. He
is currently working as a postgraduate re-
searcher at UCI. His research interests in-
clude application of RF-MEMS switches for
development of microwave devices, and small-size recongurable
antennas for smart wireless communication systems.
G. P. Li developed a silicon silicide molec-
ular beam epitaxy (MBE) system while be-
ing with the University of California at Los
Angeles (UCLA), and then in the area of sil-
icon bipolar developed very large-scale in-
tegration (VLSI) technology and process-
related device physics while being with the
IBM T. J. Watson Research Center. During
his tenure as a Sta Member and Manager
of the Technology Group, IBM, he coordi-
nated and conducted research eorts in technology development
of high-performance and scaled-dimension (0.5 and 0.25 m)
A Recongurable Spiral Antenna for Adaptive MIMO Systems 389
bipolar devices and integrated circuits (ICs), as well as research into
optical switches and optoelectronics for ultra-high-speed IC mea-
surements. In 1987, he chaired a committee for dening the IBM
semiconductor technology for beyond the year 2000. He also led
a research/development team in transferring semiconductor chip
technology to manufacturing in IBM. In 1988, he joined the Uni-
versity of California (UCI), Irvine, where he is currently a Profes-
sor of electrical and computer engineering and Director of the In-
tegrated Nanosystems Research Facility (INRF). He has authored
over 170 research papers involving semiconductor materials, de-
vices, technologies, polymer-based bio-MEMS systems, RF-MEMS,
and circuit systems. Dr. Li was the recipient of the 1987 Outstand-
ing Research Contribution Award presented by IBM and the 1997
and 2001 Outstanding Engineering Professor Award presented by
UCI.
F. De Flaviis research focused on the in-
tegration of novel materials and technolo-
gies in electromagnetic circuits and antenna
systems for the realization of smart mi-
crowave systems. He is also focused on re-
search on novel numerical techniques en-
abling faster codes for the analysis and de-
sign of microwave circuits and antennas.
His current research is focused on the syn-
thesis of novel low-loss ferroelectric mate-
rial operating at microwave frequency, which can be used as a phase
shifters design to be employed in scan-beam antennas systems. He
is also working on modeling MEMS devices to be used as analog
tunable capacitors at microwave frequency, for the realization of
tunable lters, tunable phase shifters, and smart matching cir-
cuits. Some of his research is also focused on the development of a
novel numerical technique in the time domain, which will allow re-
duction in the memory storage and faster computation. Dr. Franco
De Flaviis received his Ph.D. degree from the University of Califor-
nia at Los Angeles (UCLA) in 1997, and he became Assistant Pro-
fessor at the University of California (UCI), Irvine, in June 1998.
EURASIP Journal on Wireless Communications and Networking 2005:3, 390400
c 2005 Lars Berlemann et al.
Multimode Communication Protocols Enabling
Recongurable Radios
Lars Berlemann
Chair of Communication Networks, RWTH Aachen University, Kopernikusstrae 16, D-52074 Aachen, Germany
Email: ber@comnets.rwth-aachen.de
Ralf Pabst
Chair of Communication Networks, RWTH Aachen University, Kopernikusstrae 16, D-52074 Aachen, Germany
Email: pab@comnets.rwth-aachen.de
Bernhard Walke
Chair of Communication Networks, RWTH Aachen University, Kopernikusstrae 16, D-52074 Aachen, Germany
Email: walke@comnets.rwth-aachen.de
Received 24 September 2004; Revised 21 February 2005
This paper focuses on the realization and application of a generic protocol stack for recongurable wireless communication sys-
tems. This focus extends the eld of software-dened radios which usually concentrates on the physical layer. The generic protocol
stack comprises common protocol functionality and behavior which are extended through specic parts of the targeted radio
access technology. This paper considers parameterizable modules of basic protocol functions residing in the data link layer of the
ISO/OSI model. System-specic functionality of the protocol software is realized through adequate parameterization and com-
position of the generic modules. The generic protocol stack allows an ecient realization of recongurable protocol software and
enables a completely recongurable wireless communication system. It is a rst step from side-by-side realized, preinstalled modes
in a terminal towards a dynamic recongurable anymode terminal. The presented modules of the generic protocol stack can also
be regarded as a toolbox for the accelerated and cost-ecient development of future communication protocols.
Keywords and phrases: generic protocol stack, link layer functions, modular layer composition, recongurability, software-
dened radio.
1. INTRODUCTION
The radio access of future ubiquitous communication net-
works will be released from the constrains of cellular wireless
networks, as for instance universal mobile telecommunication
system (UMTS), or wireless local area networks (WLANs).
Wireless mobile broadband systems, providing a patchy cov-
erage in densely populated urban areas, will play an im-
portant role. For details on such a xed and planned relay-
based radio network, see [1, 2]. The addressed future wireless
network will have to combine several radio access technolo-
gies (RATs). Consequently, multimode capable terminals and
base stations are required to enable the seamless interwork-
ing between these RATs. Multimode architectures can already
be found in existing systems, like IEEE 802.16 [3] with dier-
ent modes of the physical layer (PHY).
This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
Software-dened radios (SDRs) [4, 5] are a promising ap-
proach towards these multimode devices. The recent tech-
nological progress allows an extension of the key issues in
research of SDRs from the signal processing of the physical
layer on the complete communication chain used for wire-
less communication. The current research eorts are target-
ing at the realization of cognitive radios [4, 6, 7]: self-aware,
frequency-agile radio systems that are able to identify un-
used radio spectrum. These cognitive radios require proto-
col recongurability to unfold their advantage of dynamic
spectrum usage. Therefore, this paper extends the focus of
SDRs on the protocol software used for reliable commu-
nication over the air. Especially, the data link layer (DLL)
and network layer corresponding to the ISO/OSI reference
model [8] are considered. This work supplements the re-
search of the integrated project E2R [9] dealing with end-to-
end recongurability. The approach taken in this work aims
at maximizing exibility by providing a framework that is
both general enough to accommodate a wide range of proto-
cols, yet ecient enough to ensure competitive performance.
Multimode Protocols for Recongurable Radios 391
Reconguration
management/functions
(Classic) single
protocol stack
Recongurable
protocol stack
Same
proprety System-specic
protocol stack
Specic functionality
Specic parts
Common functionality
Generic
protocol stack
RAT 1 RAT x
RAT 2 Dierent Protocol
functions
Protocol
architecture
Data
structures
Protocol
framework
Protocol
management
Figure 1: UML diagram of the generic protocol stack in the context of protocol recongurability.
Similar goals have been formulated and followed in the soft-
ware engineering domain. The x-Kernel [10] architecture
composes a protocol graph of protocol components together
into a system, but the approach does not permit dynamic re-
conguration of the protocol graph and does not specically
target wireless communications. The DARPAactive network-
ing program [11] tries to answer the key question of loca-
tion (and nature) of programmability with the perspective
to build a exible distributed computing system, again with
a focus on xed networks.
This paper consolidates previous publications [12, 13]
and deduces summarizing conclusions. After introducing the
idea of a generic protocol stack in Section 2, its application
for protocol recongurability in the context of a multimode
capable protocol architecture is outlined thereafter. The re-
alization of a generic protocol stack, based on fundamental
protocol functions that can be parameterized, is summarized
in Section 3. The composition of specic protocol layers of
adequately parameterized modules is shown in Section 4.
Section 5 introduces the composition of system-specic lay-
ers at the example of the UMTS radio link control (RLC) layer,
the transmission control protocol (TCP), and the IEEE 802.11
medium access control (MAC) layer, which dier in their
development history as well as in their layer classication
corresponding to the ISO/OSI reference model. A segmen-
tation/reassembly module and an automatic repeat request
(ARQ) protocol module are validated and evaluated analyt-
ically and through simulations in Section 4. These modules
are used with dierent parameterization for composing the
above-mentioned three specic protocol layers.
2. THE GENERIC PROTOCOL STACK
The rationale for approaching a generic protocol stack is that
all communication protocols have common functions. These
commonalities can be exploited to build an ecient mul-
timode capable wireless system. The aim is to gather these
common parts in a single generic stack and specialize this
generic part. Thereby, the particular requirements of the tar-
geted RAT are considered, as depicted in Figure 1. The tar-
geted advantages of this concept are runtime recongurabil-
ity and maintainability, code/resource sharing, protocol de-
velopment acceleration through reusability, and continuous
performance evaluation in the context of quality-of-service
(QoS) dimensioning.
The initial step towards a generic stack is a detailed,
layer-by-layer analysis of communication protocols to iden-
tify their similarities. The realization of generic parts is cru-
cial for the success of the proposed concept in the face of a
tradeo between genericity, that is, general usability, and im-
plementation eort. As depicted in Figure 1, the generic pro-
tocol stack comprises
(i) fundamental protocol functions;
(ii) a protocol architecture;
(iii) data structures;
(iv) a protocol framework;
(v) protocol management.
They form, together with RAT specic parts, a system-
specic protocol stack. An ecient multimode capa-
ble recongurable stack is realized in adding cross-stack
management-related functions. The cross-stack manage-
ment of the generic protocol stack for enabling protocol re-
congurability is introduced in detail in Section 3.
2.1. Two complementing approaches for realization
There are in general two possibilities from the software en-
gineering perspective for approaching the generic protocol
stack: (1) parameterizable functional modules and/or (2) in-
heritance, depending on the abstraction level of the identied
protocol commonalities. As introduced above, this paper fo-
cuses more on the modular approach while the inheritance-
based approach is considered in [14, 15]. Additionally, [16]
takes up the idea of a generic protocol stack in focusing on a
generic link layer for the cooperation of dierent access net-
works at the level of the DLL. However, the link layer proto-
cols are not the only protocols that have to be considered in a
multimode capable network but the complete protocol stack.
This implies, for instance, higher layer functions as the con-
trol and management of the radio resources as well as mobil-
ity.
392 EURASIP Journal on Wireless Communications and Networking
Full
recongurability
Protocol
recon-
guration
manager
Applications
Data
TCP/IP
RLC, LLC
MAC
PHY
Data
Channel (modem)
Based on
generic
protocol
stack
Figure 2: A recongurable protocol stack based on generic func-
tional modules in the context of a completely recongurable termi-
nal.
2.2. Identifying commonalities
The evolution of the digital cellular mobile radio networks
originated in the global system for mobile communication
(GSM) toward systems of the third generation, as for exam-
ple UMTS, has shown that in their standardization, devel-
opers have fallen back on well-proven functions and mecha-
nisms which are adopted to the specic requirements of their
application. The approach towards an ecient multimode
protocol stack, introduced in this paper, is based on these
protocol commonalities.
As the architecture of modern communication protocols
cannot be forced into the classical layered architecture of the
ISO/OSI reference model, it is rather dicult to identify sim-
ilarities and attribute these to specic layers. Therefore, this
paper deepens the level of examination and considers fun-
damental protocol functions as introduced above as one ba-
sis for a generic protocol stack, contrary to [14, 15] where
complete protocols are analyzed for genericity. The identied
protocol functions correspond mainly to the DLL as speci-
ed in the ISO/OSI reference model. Nevertheless, they can
be found in multiple layers of todays protocol stacks. The
functions of segmentation/reassembly or an ARQ protocol
used for error correction are an example for this. They are
located in the RLC as well as in the transport layer, namely in
the TCP.
3. ENABLINGRECONFIGURABILITY
The generic protocol stack, with its pool of generic functions
as introduced above, enables an ecient as well as exible
realization of recongurable protocol stack. Full, end-to-end
recongurability fromthe modempart up to the applications
requires a layer overlapping management and additional re-
conguration functions. Therefore, as this paper considers
communication protocols implemented in software, a proto-
col reconguration manager is introduced in Figure 2, which
accomplishes all recongurability-related tasks of the PHY,
MAC, RLC/DLL, and transport layer. These system-specic
layers are based on the pool of generic protocol functions and
mechanisms.
3.1. Protocol reconguration manager
The protocol reconguration manager has thereby the fol-
lowing tasks:
(i) management of the (permanently or temporally) par-
allel existing protocols and protocol stacks;
(ii) creation, destruction and/or reconguration of a sin-
gle protocol or complete protocol stack;
(iii) administration of the user data ow during the recon-
guration process, as for instance the redirection of
the user data from the old to the recongured protocol
stack;
(iv) cross-layer optimization through protocol conver-
gence, as introduced in the next section. This implies
for example the transfer of protocol or user data from
the old stack to the new one;
(v) support and enabling of reconguration functions of
the network, as for instance the support of a network-
initiated reconguration or an update of the network
information about the status of the terminal.
The reconguration of the protocol stack, administrated
through the protocol reconguration manager, has two char-
acteristics: (1) the creation of a new stack/layer consisting of
adequate parameterized modules of the generic stack and de-
struction of the existing one and (2) the reconguration of
the existing protocol stack in exchanging the parameteriza-
tion of the corresponding modules.
3.2. Protocol convergence in future wireless networks
The generic protocol stack enables the protocol convergence
in future wireless networks. The convergence of such multi-
mode protocol stacks has two dimensions: on the one hand
the convergence between two adjacent layers, in the following
referred to as vertical convergence, and on the other hand, the
convergence between layers located in the dierent modes of
the protocol stack which have the same functions, in the fol-
lowing referred to as horizontal convergence. The generic pro-
tocol stack, as introduced above, enables both the horizontal
as well as the vertical protocol convergence.
Figure 3 depicts the transition between two protocol
stacks of dierent-air interface modes of a wireless network.
The dierent PHY options of IEEE 802.16 could serve as an
example, see [3]. Presently, these PHY options are not en-
visaged to coexist in terminal or access equipment, although
they share a common MAC protocol with very little option-
specic extensions. The protocol stack is separated into the
user and control plane (u- and c-plane) on the one hand
and the management plane (m-plane) on the other hand.
The split between common and specic parts of the pro-
tocol is here exemplary depicted for the MAC layer, as ex-
plained above. This split may be also necessary in the PHY,
RLC, or higher layers, depending on the targeted protocol
architecture and functional exibility of the common part.
A cross-stack management logically connects the protocol
stacks of the dierent modes on the m-plane.
A seamless interworking and optimized transition be-
tween mode 1 and mode 2 have certain requirements to the
Multimode Protocols for Recongurable Radios 393
Protocol stack
mode 1
PHY 1
MAC 1 Specic
part
MAC
Common
part
RLC
u-plane + c-plane
m-plane
Protocol status
information
Transition
Cross-stack management including
modes convergence protocol
RLC
Common
part MAC
MAC 2
PHY 2
Specic
part
Protocol stack
mode 2
Figure 3: A multimode capable protocol architecture under consideration of an optimized protocol convergence. Protocol information is
transferred between two modes, here exemplary depicted on the level of the MAClayer. This reects, for instance, the case of IEEE 802.16 [3].
Layer or
sublayer
Manager
Interface
Control
Service access point
Interface
Functional
module
Functional
module
Functional
module
Data
Interface
Service access point
System specic
Generic
Figure 4: Composition of a protocol specic layer or sublayer on
the basis of generic and system-specic functional modules.
cross-stack management. The protocol data, as for instance
the protocol status information of the existing connections,
has to be transferred between the two modes. This horizon-
tal convergence is performed by a mode convergence proto-
col. Therefore, protocol functions of the c-plane, common
to the dierent modes, are necessary to access u-plane status
information. These common functions rely on a well-dened
interface towards the mode-specic part of the layer. For fur-
ther details on the proposed protocol architecture and corre-
sponding infrastructure, see [17].
4. MODULAR APPROACHTHE GENERIC PROTOCOL
STACK AS TOOLBOX OF PROTOCOL FUNCTIONS
Again, the generic protocol stack is the realization of the
common parts, as illustrated in Figure 1, and implements
its common functions on the basis of modules. These com-
mon protocol functions get their system-specic behavior
based on parameterization. Once specied, these modules
Toolbox of functional
modules
Segmentation
unit
Manager
SAP
Multiplexer
Multiplexer
Interface
SAP
PDU factory
ARQ unit
Figure 5: Toolbox of functional modules as part of the generic pro-
tocol stack.
can be repeatedly used with a dierent set of parameters cor-
responding to the specic communication system. The mod-
ules of generic protocol functions formtogether with system-
specic modules a complete protocol layer, as depicted in
Figure 4. The communication inside the layer is performed
with the help of generic data structures, that is, generic ser-
vice primitives and generic protocol data units (PDUs), which
are also considered as being a part of the generic stack, see
again Figure 1. The functional modules form a toolbox of
protocol functions as illustrated in Figure 5.
A unique manager as well as interfaces for the service ac-
cess points (SAPs) to the adjacent layers complete the fully
functional protocol layer as depicted in Figure 4. In detail,
the mentioned components have the following tasks.
(i) Functional module (generic or RAT specic): realizes a
certain fundamental functionality as black box. In case
of a generic module, a list of parameters for charac-
terizing the functionality is given and the underlying
functionality is hidden. The comprehensiveness of the
fullled function is limited to t straightforward into
a single module.
(ii) Manager: composes and administrates the layer dur-
ing runtime. This implies the composition, rearrange-
ment, parameterization, and data questioning of the
functional modules. Additionally, the manager admin-
istrates the layers internal communication, as for in-
stance the connection of the layers modules through
generic service primitives. It is the layers counterpart
394 EURASIP Journal on Wireless Communications and Networking
SAP
Segmentation unit
To RRC
Interface
Manager
To RRC/ PDCP/ BMC (radio bearers) SAP
Interface
Multiplexer
Acknow-
ledged
mode
Transpa-
rent mode
Unacknow-
ledged
mode
Segmentation unit ARQ unit
PDU factory PDU factory
Multiplexer
Interface
SAP
To MAC (logical channels)
Figure 6: UMTS RLC layer based on the functional modules of the generic protocol stack.
of the protocol reconguration manager as introduced
above in Figure 2 and it realizes the recongurability
of the layer.
(iii) Interface: translates the generic service primitives with
specic protocol information as payload to system-
specic ones and enables thus the vertical as well as
horizontal integration of the system-specic parts of
the layer.
(iv) Service access point (SAP): here, services of the layer are
performed for the adjacent layers. The layer may com-
municate via generic primitives without a translation
interface to an adjacent layer if the said layer has the
same modular composition. The interface is needed if
it is demanded that the layer appears as a classic layer
tting into an ordinary protocol stack.
(v) PDU factory (as functional module, later depicted in
Figures 6, 7, and 8): composes layer-specic protocol
frames and places them as payload in generic PDUs.
This approach enables the simulation and performance
evaluation on several levels. A single (sub-) layer as well as a
complete protocol stack can be composed out of the intro-
duced modules. To facilitate understanding, the parameteri-
zation itself is introduced later.
4.1. Generic protocol functions of the data link layer
Taking Section 2.2 into account, the following functions of
the DLL are considered as being part of the generic protocol
stack:
(i) error handling with the help of forward error correc-
tion (FEC) or ARQ protocols as for instance Send-
and-Wait ARQ, Go-back-N ARQ

, or Selective-Reject
ARQ;
(ii) ow control

;
(iii) segmentation, concatenation, and padding of protocol
data units (PDUs)

;
(iv) discarding of several-times received segments

;
(v) reordering of PDUs

;
(vi) multiplexing/demultiplexing of the data ow, as for in-
stance the mapping of dierent channels

;
(vii) dynamic scheduling;
(viii) ciphering;
(ix) header compression.
The asterisk marked functions are considered in
this paper while the other functions are target of the au-
thorsongoing work.
4.2. Parameterization of functional modules
Parameterization implies not only specic values, as for in-
stance the datagram size of a segmentation module, but also
a conguration of behavior and characteristics of a module,
as for example the concretion of an ARQ module as a Go-
back-N ARQ protocol with specied window sizes for trans-
mission and reception. This implies as well a conguration of
the modules interface to the outside. Thus, the parameteri-
zation of functional modules may mean (1) a specication of
certain variables, (2) the switching on/o of certain function-
ality/behavior, and (3) an extension of the modules interface
to the outside.
Taking the example of the segmentation/reassembly
module, the parameterization may imply among other
things:
(i) use of concatenation;
(ii) use of padding, that is, lling up of the PDU to reach a
certain size;
(iii) transmitter or/and receiver role;
(iv) buer size for SDUs concatenated in a single PDU;
(v) size of PDU after handling;
(vi) behavior in case of error, that is, interworking with
ARQ module.
Multimode Protocols for Recongurable Radios 395
To higher layer
Manager
Interface
Multiplexer
ARQ ARQ ARQ ARQ
IP address
Multiplexer Multiplexer
TCP port
UDP port
TCP PDU factory UDP PDU factory
Multiplexer
Segmentation unit
IP PDU factory
Interface
To data link layer
Figure 7: TCP, IP, and UDP layers based on the functional modules of the generic protocol stack. The TCP layer (gray dash-dotted line) is
considered here.
At the example of the ARQ module, the parameterization
means the specication of
(i) ARQ protocol characteristic, for instance Go-Back-N
ARQ or Selective-Reject ARQ;
(ii) transmitter or/and receiver role;
(iii) receive and transmission window sizes;
(iv) xed, variable (TCP) window length or open/shut
mechanism (LLC);
(v) timer value, after a packet is assumed to be lost;
(vi) connection service: inexistent (UMTS RLC), sepa-
rated for each direction (802.11CSMA/CA with
RTS/CTS), 2-way handshake (GSM LLC), or 3-way
handshake (TCP);
(vii) use of negative acknowledgments (NACKs).
5. COMPOSITIONOF SYSTEM-SPECIFIC LAYERS
As introduced above, the link layer functions are not limited
in their appearance to the DLL. To illustrate the applicability
of the modular approach based on the toolbox of protocol
functions of Figure 5, a composition of three exemplary pro-
tocol layers, all dierently localized in a protocol stack cor-
responding to the ISO/OSI reference model, is introduced in
the following: (1) a UMTS RLC layer in Figure 6, (2) a TCP,
IP, and UDP layers in Figure 7, and (3) an IEEE 802.11 MAC
layer in Figure 8. The consideration of Figure 7 is limited
in the following to the TCP layer, marked through the gray
dash-dotted rectangle. The medium access of the distributed
coordination function (DCF) of 802.11 may be regarded as
a Send-and-Wait ARQ, simply realized in the ARQ module
by a Go-Back-N ARQ with a window length of 1. The per-
formance of these three layers is evaluated in the subsequent
section.
6. SIMULATIVE EVALUATIONANDVALIDATIONOF
THE FUNCTIONAL MODULES
The parameterizable modules are implemented in the speci-
cation and description language (SDL), and evaluated with the
help of a modular object-oriented software and environment
for protocol simulation (MOSEPS) that provides basic traf-
c generators, a model of an erroneous transmission chan-
nel and statistical evaluation methods. This section intro-
duces the modular approach to protocol functions in focus-
ing on the segmentation in UMTS and the error correction
through ARQ protocols in TCP and 802.11. The adequate-
ness of the modules for being used in a multimode capable
protocol stack is shown. The modules are parameterized cor-
responding to a specic protocol layer and their simulative
behavior is compared to analytical models known from the
literature.
6.1. UMTS radio link control layer
In general, segmentation is needed in all cases where higher
layer PDUs, referred to as service data units (SDUs), need to
be separated into multiple PDUs to be further handled by
lower layers. This restriction concerning the size of a PDU
may result for instance from reasonable limitations of a PDU
transmitted with error correction in an ARQprotocol. It may
also be motivated through the capacity of a transport channel
oered by the physical layer, using the notation of UMTS.
396 EURASIP Journal on Wireless Communications and Networking
To LLC
SAP
SAP
To STA
MGMT
layer
Interface
Manager
SAP
To PHY
Multiplexer
Mobility unit
Interface
Mobility/data
Segmentation unit
ARQ unit
PDU factory
Interface
Figure 8: 802.11 MAC layer based on the functional modules of the generic protocol stack.
In case of multiple users sharing a common channel, the
segmentation can increase the channels eciency in hav-
ing a multiplexing gain. Concatenation increases thereby the
channel utilization as it is outlined in Figure 9. The channel
utilization as quotient of user payload and available channel
capacity is given through
payload
channel capacity
=
l
payload

l
payload
/packet size

packet size
(1)
in the case of no concatenation. The packet size of the
physical channel is assumed to be xed and the packet
length of the user data l
payload
determines if the segmented
higher layer SDU(s) t(s) into a single PDU. We assume
that an additional physical channel is established if the user
data requires it. Thus, additional channel capacity is pro-
vided and the available capacity is increased. In case of no
concatenation, the xed-sized packet is transmitted partly
empty over the physical channel. Consequently, the chan-
nels overall utilization, that is, the eectively used capac-
ity compared to the amount of provided capacity, is de-
creased and follows a zigzag behavior when an additional
physical channel is used as illustrated by the solid line in
Figure 9.
The introduced example of Figure 9 focuses on the seg-
mentation aspects of the UMTS RLC in the unacknowledged
mode (UM). The UM of the RLC has the responsibility to
concatenate SDUs to a PDU of a predened length, here
128 bytes. The simulative results with and without concate-
nation are given by the markers. The payload packet size, that
is, the user data, is increased up to 500 bytes and the channel
utilization as introduced above is evaluated.
In applying the segmentation in a communication proto-
col, the protocol overhead comes into play. Due to this over-
head, more capacity than transmitted user data is required as
observable in Figure 9 for l
payload
=384 bytes in the case of no
1st channel
2nd channel
3rd channel
4th channel
0 100 200 300 400 500
Payload packet size l
payload
(byte)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
P
a
y
l
o
a
d
/
c
h
a
n
n
e
l
c
a
p
a
c
i
t
y
Sim.: with concatenation
Sim.: without concatenation
Analyt.: with concatenation
Analyt.: without concatenation
Figure 9: Utilization of channels with a xed packet segmentation
size of 128 bytes.
concatenation. The same stands for the usage of concatena-
tion, as the observed channel utilization does not match the
ideal one. The number of SDUs prepared for transmission in
a single PDU is here limited so that for small l
payload
values, a
PDU is not completely lled. In summary, the parameteriz-
able segmentation/reassembly module adequately reects the
expected behavior and can be validly used in a multimode
capable protocol stack.
6.2. Transmission control protocol layer
As introduced above in Figure 7, a TCP layer can be com-
posed out of the functional modules of the DLL as being
Multimode Protocols for Recongurable Radios 397
0 1000 2000 3000 4000 5000
Payload data frame length (byte)
0
0.2
0.4
0.6
0.8
1
1.2
O
v
e
r
h
e
a
d
t
o
p
a
y
l
o
a
d
Sim.: BER = 10
5
Sim.: BER = 10
6
Analyt.: BER = 10
5
Analyt.: BER = 10
6
(a)
0 1000 2000 3000 4000 5000
Size of data per packet (byte)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
O
v
e
r
h
e
a
d
t
o
p
a
y
l
o
a
d
Sim.: w = 8
Sim.: w = 64
Analyt.: w = 8
Analyt.: w = 64
(b)
Figure 10: TCP layer Go-back-N ARQ validation and evaluation. The protocol overhead to payload ration in dependency on the frame
length of the payload data is depicted. The lines are analytic results corresponding to (2), while the markers indicate simulative evaluation.
(a) Varied BER = 10
5
or 10
6
and window length w = 8. (b) Varied window length w = 8 or 64 and BER = 10
6
.
part of the generic protocol stack. To validate the Go-Back-N
mechanisms of the TCP layers ARQ module, we measure the
protocol overhead in dependency on the payload packet size
in the case of erroneous transmissions. The focus is thereby
on the inuence of two eects: the bit error ratio (BER) of
the radio channel, that is, the wireless medium, and the size
of the send and receive window w.
With a packet length of l
packet
= l
header
+ l
payload
, where
TCP has xed header length of l
header
= 40 bytes, the packet
error ratio (PER) can be calculated to
PER = 1 (1 BER)
lpacket 8
. (2)
Based hereon the overhead to payload quotient for the
Go-Back-N ARQ can be derived and approximated [12] to
overhead
payload
=
l
header
l
payload
+

w/2
i=1

(w/2) i + 1

(1 PER)
i1
PER
w/2

l
packet
l
payload
,
(3)
where w is the length of the transmission window leading
to the analytical results as depicted in Figure 10. This gure
illustrates the overhead to payload ratio in dependency on
the frame length of the payload data for (a) a BER of 10
5
and 10
6
on the one hand and (b) window length of 8 and
64 on the other hand. Figure 10a shows the expected per-
formance corresponding to the Go-Back-N ARQ. The over-
head to payload ratio increases with increasing bit error ra-
tio and an optimal frame length for the payload data to
minimize the said ratio can be determined. From the cross-
protocol optimization perspective, this frame length may be
used as a dimensioning rule for segmentation. The same
stands for Figure 10b: there the overhead-to-payload ratio
increases with increasing window length, as the amount of
data which has to be retransmitted, in the case of an error
corresponding to the Go-Back-N ARQ, increases. The dier-
ence between analysis and simulation for large payload data
frames in Figures 10a and 10b is reasoned by the send/receive
buers of the implemented TCP layers. Data packets in these
buers have to be discarded when an error is detected and are
neglected in the approximation (3). In summary, the ARQ
module of the generic protocol stack fulls adequately its in-
tended purpose.
6.3. IEEE 802.11 mediumaccess control layer
In this section, the modular composition of an IEEE 802.11
MAC layer as illustrated in Figure 8 is validated and evalu-
ated. Therefore, the average throughput of the carrier sensing
multiple access with collision avoidance (CSMA/CA) -based
decentralized medium access by the DCF with and without
request to send/clear to send (RTS/CTS) is analyzed and sim-
ulated.
398 EURASIP Journal on Wireless Communications and Networking
10 20 30 40 50 60 70 80 90 100
Number of stations
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
T
h
r
o
u
g
h
p
u
t
(
M
b
p
s
)
Sim.: without RTS/CTS
Sim.: with RTS/CTS
Analyt.: without RTS/CTS
Analyt.: with RTS/CTS
(a)
10 20 30 40 50 60 70 80 90 100
Number of stations
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
T
h
r
o
u
g
h
p
u
t
(
M
b
p
s
)
Sim.: without RTS/CTS
Sim.: with RTS/CTS
Analyt.: without RTS/CTS
Analyt.: with RTS/CTS
(b)
Figure 11: 802.11 MAClayer evaluation of the DCF-based mediumaccess under utilization of the ARQmodule. The total systemthroughput
depending on the number of transmitting stations is depicted. The lines are analytic results corresponding to (5), while the markers indicate
the simulative evaluation. Throughput evaluation-channel capacity=1Mbps; (a) packet size=128 bytes, (b) packet size=4096 bytes.
Table 1: Time slot durations in microseconds and probabilities that
the medium is empty (e), successfully (s) allocated, or a collision (c)
occurs [18, 19]. The xed values result from the time length of an
RTS/CTS sequence.
Probability
Duration with Duration without
RTS/CTS RTS/CTS
p
e
= (1 )
n
T
e
= 1 T
e
= 1
p
s
= n(1 )
n1
T
s
= 636 + 8l
payload
T
e
= 636 + 8l
payload
p
c
= 1 p
e
p
s
T
e
= 170 T
s
= 234 + 8l
payload
The channel capacity is mainly wasted by two eects:
MAC header sending and collisions. One way to an analytical
approach for determination of the throughput is to calculate
the collision probability p and the access probability with
the help of a two-dimensional Markov chain for the mod-
eling of the backo window of the DCF [18, 19] resulting
into
p = 1 (1 )
n1
, = 2

1 + W
0
+ pW
0
1 (2p)
8
1 2p

1
,
(4)
where n is the number of stations and W
0
the minimum
backo window size, here we chose W
0
= 8. With the
help of the average time slot length T
average
on the basis of
Table 1, the average total system throughput t
saturation
can be
calculated to
t
saturation
=
P
s
L
payload
T
average
, T
average
= P
e
T
e
+ P
s
T
s
+ P
c
T
c
. (5)
We assume a channel rate of 1 Mbps. The slot length,
short interframe space (SIFS) and distributed coordination
function IFS (DIFS) are 1, 6, and 10 microseconds resulting
into the analytical as well as simulative results of Figure 11.
There, the overall system throughput, with and without
RTS/CTS, in dependency on the number of stations is de-
picted. For small packets, L
payload
= 128 bytes, Figure 11a,
the headers are the main cause for an inecient use of the
medium. For larger frames, L
payload
=4096 bytes, Figure 11b,
a collision wastes more time, as a transmitting station is only
able to notice an interfered frame after its ending. Therefore,
the RTS/CTS mechanism is introduced, to have just a small
RTS frame lost in case of a collision. The simulation agrees
mainly with the analytic determination of the throughput
of (5) and illustrates the superiority of the RTS/CTS-based
solution. As the ARQ module of the generic protocol stack
reects the expected behavior of RTS/CTS mechanism [19],
this module can be legitimately used in an 802.11 MAC layer.
7. CONCLUSION
The introduced concept of a generic protocol stack enables
protocol software for future multimode capable systems
under the consideration of protocol reconguration. The
generic protocol stack, as a collection of modular protocol
Multimode Protocols for Recongurable Radios 399
functions, takes up the usual advance of software engineering
in the eld of protocol development and evaluation: it has
fallen back on well-proven and known protocol functions
and behavior from the portfolio of the engineers experience.
A generic realization of these functions in the form of inde-
pendent modules results in a toolbox of protocol functions as
a construction kit for protocol development. Dimensioning
rules for an adequate support of QoS in wireless communica-
tion from the perspective of the protocols can be derived. In
taking the tradeo of genericity into account, these thought-
ful realized modules stimulate eciency through reusabil-
ity and maintainability as well as accelerate the development
process itself. However, the consideration of common pro-
tocol functions and protocol convergence during the devel-
opment of future protocols will increase by itself the grade
of genericity and advantage of this approach. The eciency
of protocol recongurability benets from the introduced
generic approach and implies a clearly identied eort of
protocol management. Thus, the introduced approach is a
rst step to an end-to-end recongurable wireless system.
ACKNOWLEDGMENTS
The authors would like to thank the German Research Foun-
dation (DFG) for funding the work contributed to this paper
in the form of a Research College (Graduiertenkolleg). This
work has been continously funded from third parties grants.
REFERENCES
[1] B. Walke, R. Pabst, and D. Schultz, A mobile broadband sys-
tem based on xed wireless routers, in Proc. International
Conference on Communication Technology (ICCT 03), vol. 2,
pp. 13101317, Beijing, China, April 2003.
[2] R. Pabst, B. Walke, D. Schultz, et al., Relay-based deploy-
ment concepts for wireless and mobile broadband radio,
IEEE Commun. Mag., vol. 42, no. 9, pp. 8089, 2004.
[3] IEEE, Standard for Local and metropolitan area networks,
Part 16: Air Interface for xed Broadband Wireless Access
SystemsAmendment 2: Medium Access Control Modica-
tions and Additional Physical Layer Specications for 2-11
GHz, April 2003.
[4] J. Mitola, The software radio architecture, IEEE Commun.
Mag., vol. 33, no. 5, pp. 2638, 1995.
[5] W. Tuttlebee, Software Dened Radio: Enabling Technologies,
Wiley Series in Software Radio, John Wiley & Sons, New York,
NY, USA, 2002.
[6] J. Mitola, Cognitive Radio, Ph.D. thesis, KTH, Stockholm,
Sweden, 2000, pp. 4590.
[7] P. M ah onen, Cognitive trends in making: future of net-
works, in Proc. IEEE 15th International Symposium on Per-
sonal, Indoor and Mobile Radio Communications (PIMRC04),
Barcelona, Spain, September 2004, invited paper.
[8] ITU-T Recommendation X.200, Information Technology
Open System InterconnectionBasic Reference Model:
The Basic Model, International Telecommunication Union
(ITU), Geneva, Switzerland, 1994.
[9] End-to-End Recongurability (E2R), IST-2003-507995 E2R,
http://www.e2r.motlabs.com.
[10] N. C. Hutchinson and L. L. Peterson, The X-Kernel: An ar-
chitecture for implementing network protocols, IEEE Trans.
Software Eng., vol. 17, no. 1, pp. 6476, 1991.
[11] J. M. Smith and S. M. Nettles, Active networking: one view of
the past, present, and future, IEEE Trans. Syst., Man, Cybern.
C, vol. 34, no. 1, pp. 418, 2004.
[12] L. Berlemann, A. Cassaigne, and B. Walke, Generic protocol
functions for design and simulative performance evaluation
of the link-layer for recongurable wireless systems, in Proc.
7th International Symposium on Wireless Personal Multimedia
Communications (WPMC 04), pp. 55, Abano Terme, Italy,
September 2004.
[13] L. Berlemann, A. Cassaigne, R. Pabst, and B. Walke, Modu-
lar link layer functions of a generic protocol stack for future
wireless networks, in Proc. Software Dened Radio Technical
Conference (SDR 04), pp. 66, Phoenix, Ariz, USA, November
2004.
[14] M. Siebert and B. Walke, Design of generic and adaptive pro-
tocol software (DGAPS), in Proc. International Conference
on 3rd Generation Wireless and Beyond (3Gwireless 01), San
Francisco, Calif, USA, June 2001.
[15] L. Berlemann, M. Siebert, and B. Walke, Software dened
protocols based on generic protocol functions for wired and
wireless networks, in Proc. Software Dened Radio Technical
Conference (SDR 03), Orlando, Fla, USA, November 2003.
[16] J. Sachs, A generic link layer for future generation wireless
networking, in Proc. IEEE International Conference on Com-
munications (ICC 03), Anchorage, Alaska, USA, May 2003.
[17] L. Berlemann, R. Pabst, M. Schinnenburg, and B. Walke, A
recongurable multi-mode protocol reference model facilitat-
ing modes convergence, in Proc. 11th European Wireless Con-
ference (EW 05), Nicosia, Cyprus, April 2005.
[18] G. Bianchi, Throughput evaluation of the IEEE 802.11 dis-
tributed coordination function, in Proc. 5th Workshop on Mo-
bile Multimedia Communications (MoMuc 98), Berlin, Ger-
many, October 1998.
[19] A. Hettich, Leistungsbewertung der Standards HIPERLAN/2
und IEEE 802.11 f ur drahtlose lokale Netze, Ph.D. disser-
tation, RWTH Aachen University, Aachen, Germany, 2001,
ABMT 23, http://www.comnets.rwth-aachen.de.
Lars Berlemann was born in Aachen, Ger-
many, in 1977. After having absolved a stu-
dent internship at Inneon Technologies,
San Jose, he received his Diploma degrees in
electrical engineering and economics from
Aachen University in 2002 and 2003, re-
spectively. Since 2002, he served as a Re-
search Assistant at the Chair of Communi-
cation Networks (with Professor B. Walke)
of RWTH Aachen University, where he is
working towards his Ph.D. degree. He is a scholarship holder of the
Research College (Graduiertenkolleg) Software for Mobile Com-
munication Systems of the German Research Foundation (DFG)
and received the Siemens Mobile Academic Award in 2003. He
is responsible for the organization of the scientic program of Eu-
ropean Wireless Conference 2005 and PIMRC 2005. His research
work has so far been taken up by various national and European
Union funded research projects, such as the Academic Network for
Wireless Internet Research in Europe (ANWIRE), Wireless World
Initiative New Radio (WINNER), and CoCoNet. His research in-
terests include cognitive radios, distributed QoS support in spec-
trum sharing scenarios, spectrum etiquette, and exible protocol
stacks. He is the author/coauthor of 18 reviewed scientic pub-
lications and he is a Student Member of the IEEE ComSoc and
ITG/VDE.
400 EURASIP Journal on Wireless Communications and Networking
Ralf Pabst was born in Cologne, Germany,
in 1974. After having absolved a student
internships at SIEMENS ICN and Voda-
fone Germany, he received his Diploma in
electrical engineering from Aachen Univer-
sity in 2001 and served since then as a Re-
search Assistant at the Chair of Communi-
cation Networks (with Professor B. Walke)
of Aachen University, where he is work-
ing towards his Ph.D. degree. He has been
working on various national and European Union funded research
projects, such as the German National Project Multifunk, and the
EU FP5 IST projects; Wireless Strategic Initiative (WSI), Wire-
less World Research Initiative (WWRI), Network of Excellence
in Wireless Applications and Technology (NEXWAY), and Aca-
demic Network for Wireless Internet Research in Europe (AN-
WIRE). He has been actively involved in the organization of the
WG4 in the Wireless World Research Forum (WWRF). His cur-
rent research activities are focused on the Wireless World Initia-
tive New Radio (WINNER) Project under the 6th Framework Pro-
gram (FP6) of the European Union. His research interests include
the performance evaluation of relay-based deployment concepts
for next-generation wireless networks, associated resource man-
agement protocols, and spectral coexistence of wireless standards.
He is the author/coauthor of 19 scientic publications, including 2
journal articles and 2 textbook chapters.
Bernhard Walke for the last 13 years is run-
ning the Chair for Communication Net-
works, RWTH Aachen University, Ger-
many, where about 35 researchers work un-
der his guidance on topics like air-interface
design, formal specication of protocols,
xed network planning, development of
tools for stochastic event-driven simulation,
and analytical performance evaluation of
services and protocols of XG wireless sys-
tems. During that time, he has supervised more than 650 M.S.
theses and 43 Ph.D. theses covering most aspects of xed and mo-
bile communication networks. He has published more than 120 re-
viewed conference papers, 25 journal papers, and seven textbooks
on architecture, trac performance evaluation, and design of fu-
ture communication systems. Prior to joining academia, Professor
Walke worked in various industry positions for AEG-Telefunken
(now part of EADS AG). Professor Walke holds Diploma and Doc-
torate degrees in electrical engineering, both from the University of
Stuttgart, Germany. He is the Cochair of the PIMRC 2005, Berlin,
Steering Board Member of the annual European Wireless Confer-
ence, Senior Member of IEEE ComSoc, and Member of ITG/VDE
and GI.
EURASIP Journal on Wireless Communications and Networking 2005:3, 401412
c 2005 A. Brawerman and J. A. Copeland
Towards a Fraud-Prevention Framework for
Software Dened Radio Mobile Devices
Alessandro Brawerman
School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA 30318, USA
Email: ale@ece.gatech.edu
John A. Copeland
School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA 30318, USA
Email: copeland@ece.gatech.edu
Received 29 September 2004; Revised 8 March 2005
The superior recongurability of software dened radio mobile devices has made it the most promising technology on the wireless
network and in the communication industry. Despite several advantages, there are still a lot to discuss regarding security, for
instance, the radio conguration data download, storage and installation, users privacy, and cloning. The objective of this paper
is to present a fraud-prevention framework for software dened radio mobile devices that enhances overall security through
the use of new pieces of hardware, modules, and protocols. The framework oers security monitoring against malicious attacks
and viruses, protects sensitive information, creates and protects an identity for the system, employs a secure protocol for radio
conguration download, and nally, establishes an anticloning scheme, which besides guaranteeing that no units can be cloned
over the air, also elevates the level of diculty to clone units if the attacker has physical access to the mobile device. Even if cloned
units exist, the anticloning scheme is able to identify and deny services to those units. Preliminary experiments and proofs that
analyze the correctness of the fraud-prevention framework are also presented.
Keywords and phrases: cellular frauds, cloning, security and privacy issues, security protocols, software dened radio mobile
devices.
1. INTRODUCTION
Software dened radio [1] allows multiple radio standards
to operate on common radio frequency hardware, thereby
ensuring compatibility among legacy, current, and evolving
wireless communication technologies.
A software dened radio mobile device (SDR-MD) is ca-
pable of having its operation changed by dynamically load-
ing radio reconguration data (R-CFG les) over the air.
With dierent R-CFGs, the device can operate using dierent
wireless communication technologies while having a single
transceiver. A typical SDR-MD can manage communication
via satellite, over dierent cellular technologies, VoIP (voice
over internet protocol), and operations over the internet.
One of the key issues in SDR wireless communication in-
volves security. According to the SDR Forum [2], some of
This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
the concerns are the R-CFG download, storage, and instal-
lation; users privacy, that is, protection of the users iden-
tity, location, and communication with other devices; and -
nally, SDR-MD cloning, that is, illegally using services that
are billed to someone elses device.
To address the SDR Forum concerns and greatly en-
hance the overall security of SDR-MDs, a fraud-prevention
framework is proposed. The proposed framework oers se-
curity monitoring against malicious attacks and viruses that
may aect the conguration data, protects sensitive informa-
tion through the use of protected storage, creates and pro-
tects an identity for the system, employs a secure protocol
for R-CFG download, and nally, establishes an anticloning
scheme which guarantees that no units can be cloned over
the air, and elevates the level of diculty to clone units if the
attacker has physical access to the SDR-MD. Even if cloned
units exist, the anticloning scheme is able to identify and
deny services to those units.
Preliminary practical experiments using java 2 micro-
edition (J2ME) [3] and proofs that analyze the correctness
of the fraud-prevention framework are also presented.
402 EURASIP Journal on Wireless Communications and Networking
2. BACKGROUND
Research work has been done for each of the SDR concerns
previously described; however, no published work has devel-
oped a solution that encompasses more than one of the con-
cerns at once. This section is divided according to the SDR
Forum concerns. For each subsection, some of the relevant
related research is presented.
2.1. R-CFGdownload, storage, and installation
In [4], the authors discuss a model for securing the R-CFG
download and installation that involves the use of secret de-
vice keys and signatures. All security operations take place
within tamper-proof hardware that also contains the pro-
grammable components of the transceiver. This approach
provides good security for the radio software that lies within
the tamper-proof hardware, but leads to some drawbacks
such as the use of nonstandard security methods, lack of a
means for third-party vendors to provide R-CFGs, and, most
important, lack of a means for securing radio software that
resides outside the tamper-proof hardware.
2.2. Users privacy
Some eorts, called privacy extension to Mobile IPv6, deal
with users privacy. The basic idea of these eorts is to re-
place the MAC address of a mobile device with a random
one, called a temporal mobile identier (TMI) [5] or pseu-
dorandom interface identier (PII) [6].
In those schemes, personal mobile location privacy con-
trol relies on either the home administration, the foreign ad-
ministration, or both. Moreover, the home administration is
required to share some secrets with the foreign administra-
tion to prevent eavesdroppers from having any knowledge
about the binding users temporal identiers and real iden-
tiers. These eorts cannot completely control mobile loca-
tion privacy by a mobile user since the administration can
associate any identier (PII or TMI) with the corresponding
real ID of the mobile device.
2.3. SDR-MDcloning
The advanced mobile phone system(AMPS) [7] is the analog
mobile phone system standard introduced in the Americas
during the early 1980s. Despite the fact that it was a great ad-
vance in its time, the AMPS presented several security aws,
and multiple copies of cloned mobile stations were created
with little diculty.
The global system for mobile communication (GSM) [8]
is a globally accepted standard for digital cellular communi-
cation. The GSM authentication framework relies on special
cryptographic codes to authenticate customers and bill them
appropriately. A personalized smart card, called a SIM card,
stores a secret key that is used to authenticate the customer;
knowledge of the key is sucient to make calls billed to that
customer.
The SIM card is easily removable so that the user can
use other cell phones. The drawback is that someone who
has physical access to the SIM card can copy the information
to another card, thereby cloning the authentication informa-
tion of the user.
Cloning the SIM card is a relevant aw, however a much
more serious aw was discovered. In [9] it is shown that the
cryptographic codes used for authentication are not strong
enough to resist attacks. To exploit this vulnerability, an in-
dividual would interact with the SIM card repeatedly to learn
the secret key and would then be able to clone the phone
without having to clone the SIM card. Although it was con-
sidered that the attacker had physical access to the SIM card,
it was mentioned that over-the-air attacks are possible, mak-
ing cloning on GSM cellphones a more serious threat.
The Universal Mobile Telecommunications System
(UMTS) [10] is an open air-interface standard for third-
generation wireless telecommunications. It provides higher
data rates and xes several security aws encountered in the
GSM standard. Despite several advantages that the UMTS
standard provides, it also stores vital information in the SIM
card. Thus, like the GSM, someone might be able to copy the
authentication information from one SIM card to another.
Another drawback concerns the KASUMI block cipher,
which is at the core of the integrity and condentiality mech-
anisms in the UMTS network. Hardware implementations
are required to use at most 10 000 gates and must achieve en-
cryption rates in the order of 2 Mbps (maximum data rate).
Thus, a considerable eort must be performed in order to
implement a high-performance hardware component that
carries out the operations of the KASUMI block cipher.
As a nal remark, UMTS devices are not capable of re-
conguring their radio parameters via software. Thus, dual
mode or tri-mode expensive cell phones are necessary to
guarantee backward compatibility with other standards.
Simpler schemes that only detect cloned units and do not
try to prevent cloning have also been proposed. They can be
found in [11, 12].
2.4. Trusted computing group
The trusted computing group (TCG) [13] is an industry
standards body comprising computer and device manufac-
turers, software vendors, and others with an interest in en-
hancing the security of the computing environment across
multiple platforms and devices.
The TCG claims that it will develop and promote open
industry standard specications for trusted computing hard-
ware building blocks and software interfaces across multiple
platforms, including personal computers (PCs), servers, per-
sonal digital assistants (PDAs), and digital phones.
So far the TCG has only presented specication for the
PC environment [14]. Some of the benets include more se-
cure local data storage, a lower risk of identity theft, and the
deployment of more secure systems and solutions based on
open industry standards.
Despite the fact that the TCG specication for the PC
does point out and solve several security aws, this speci-
cation would not achieve a satisfactory performance if em-
ployed by constrained SDR-MDs.
A Fraud-Prevention Framework for SDR Mobile Devices 403
Secure/unsecured internet
connection module
R-CFG/CFG
security module
To the outside world
SDR device manager
Environment
discoverer
Unsecured LDDA
module
Other
applications
Application layer
Software core
SDR framework
R-CFG manager
Secure LDDA
module
CFG
manager
Application
manager
OS layer
OS
R-CFG
Cong. le
Encrypted
Unencrypted
Storage
RAM
Protected
RAM
ROM CPU
Hardware layer
SDR
FPGA
Tamper-protected hardware package
Signalizes cloning
Security hardware layer
Embedded
devices
Figure 1: The preliminary design of the fraud-prevention framework.
3. THE FRAUD-PREVENTION
FRAMEWORK SPECIFICATION
The fraud-prevention framework is composed of new pieces
of hardware, new modules, and new protocols. Figure 1 de-
picts the preliminary design of the framework. The dashed
squares are the main contributions of this work.
Note that the SDR device manager (SDR-DM) is respon-
sible for managing all the communication with the outside
world and for requesting the services of each module when
needed. Also, the environment discoverer module is respon-
sible for detecting which wireless communication technolo-
gies are available in the current SDR-MDs environment. This
module is assumed to be present in the software core SDR
framework and is outside the scope of this work.
The R-CFG manager is responsible for managing the
R-CFG les currently stored in the device and the R-CFG
currently installed. It also informs the SDR-DM when a dif-
ferent R-CFG is needed. The CFG manager is responsible for
managing the conguration (CFG) le. The CFG le is pro-
vided by the wireless operator (WO) and is used to set the
devices phone number. Note that both the R-CFG and CFG
les are stored in an encrypted storage. Standard encryption
algorithms such as RC5 [15] and RSA [16] can be used to
provide the encryption storage. Other modules as well as ba-
sic denitions are discussed in separate subsections below.
3.1. Basic denitions
This section presents denitions, components, and entities
that participate in the fraud-prevention framework. The
nomenclature used to specify the framework is presented in
Table 1.
The entities that participate in the framework as well as
their responsibilities are dened in Table 2.
404 EURASIP Journal on Wireless Communications and Networking
Table 1: Basic denitions.
C A 48-bit random number (nonce)
K
Y
{C} C is cryptographically transformed, somehow, with a key Y
MD(Z) Hash of Z
[C]
Alice
C is transformed using the private key of Alice
{C}
Alice
C is transformed using the public key of Alice
Attestation
It is used to check integrity status of a certain component. It is dened as the function Att(X),
which results in the hash of component X
Attestation key pair (AK)
It is used to obtain the attestation credential. Composed by the 2048-bit attestation private key
(AK
priv
) and public key (AK
pub
)
Attestation credential (AC)
It is used to identify the SDR-MD. It is signed by the privacy credential authority (Privacy CA)
and it is presented whenever the user tries to use the network services. AC = [AK
pub
]
Privacy CA
Null AC
When the SDR-MD discovers it is a cloned unit, it sets its AC to null.
Every bit in the AC is equal to 0
Endorsement key (EK)
It is used to uniquely identify the SDR-MD. It is never disclosed by the device.
Its size is also 2048 bits
R-CFG It is used to congure the radio of the SDR-MD
Valid R-CFG An R-CFG that has been approved by the regulatory agency
Invalid R-CFG
An R-CFG that has not been approved by the regulatory agency or it has been modied
after been approved by the regulatory agency
CFG
It is used to set up the phone number of the SDR-MD. It is signed by the WO.
CFG = [Phone no.]
WO
Table 2: Entities and responsibilities.
Manufacturer (manuf.)
Produces the SDR-MD. Generates the R-CFGs. Generates the SDR-MDs EK and informs
the Privacy CA about the EK. Calculates and stores the Att(EK) in the SDR-MD
Installs the initial R-CFG and stores the Att(R-CFG) in the SDR-MD
Regulatory agency (RA)
Tests, approves, and licenses the R-CFG. Basically, the RA tests the R-CFG in the specic hardware to ensure
that the device does not cause interference or function out of its dened spectrum, as dened in [17]
WO
Sells the SDR-MD. Provides communication services. Generates the CFG
Authenticates the SDR-MD to use the network. Detects cloned SDR-MDs
Privacy CA Provides the SDR-MD with an AK pair, the AC, and the WO public key
SDR-MD Utilizes the network services. Downloads R-CFGs and CFGs les. Detects if it is a cloned or valid unit
3.2. The tamper-protected hardware package
The TPHP must be physically protected from tampering.
This includes physically binding it to the other physical parts
of the SDR-MD such that it cannot be easily disassembled
and transferred to other devices. These mechanisms are in-
tended to resist tampering. Tamper evidence measures are to
be employed. Such measures enable detection of tampering
upon physical inspection. The package must limit pin prob-
ing and EMR scanning. Similar tamper-protected hardware
is the trusted platform module of [13] and the Intel wireless
trusted platform processor [18].
The TPHP is composed of two tamper resistant chips
(TRCs): TRC1, which is read only, and TRC2, which is
read/write. The TRC1 contains the EK, the attestation en-
gines responsible for measuring, reporting, and comparing
integrity values, and a specialized hardware to generate 48-bit
random numbers. The TRC2 contains the attestation engine
responsible for storing integrity values and protected non-
volatile memory to store the necessary keys. Notice that the
TPHP comes from the manufacturer with the RAs public key
already stored.
The attestation engines are divided into the attesta-
tion measurement engine (AMEng), attestation store engine
(AS Eng), attestation report engine (AREng), and attestation
comparison engine (ACEng). Table 3 presents the functions
of each attestation engine. Figure 2 depicts the components
of the TPHP as it comes from the manufacturer.
3.3. The secure SDR R-CFGdownload protocol
To install only valid R-CFGs, a secure SDR R-CFG download
protocol is dened as part of the fraud-prevention frame-
work. The secure protocol employs the mutual authentica-
tion and R-CFG validation and verication steps described
by the R-CFG/CFG security module.
A Fraud-Prevention Framework for SDR Mobile Devices 405
Table 3: Attestation engines and functions.
AM Eng
Measures Att(EK), Att(R-CFG), and Att(CFG),
and writes the results into R0, R1 and R2
AS Eng
Stores the Att(EK) in register 0(R0), the Att(R-CFG)
in register 1(R1), and the Att(CFG) in register 2(R2)
AR Eng Reads and reports the values of the registers
AC Eng
Compares the values of R0, R1, and R2, reported by
the AR Eng, with the values measured by the AM Eng
Whenever a manufacturer generates a new R-CFG, it has
to send the R-CFG to be approved and licensed by the RA.
This is called R-CFG validation.
To perform R-CFG validation, the protocol employs a
public-private key mechanism. The manufacturer sends to
the RA a combination of a header, which contains manufac-
turer, model, serial number range, and possibly some other
information; the new R-CFG; and the hardware in which the
R-CFG is to be tested and used.
The RA installs the R-CFG in the specied device and
tests the devices behavior. If no malfunction is observed,
the RA approves the R-CFG and assigns it a license num-
ber. During the test, the RA computes h = MD(headerR-
CFG). The value h is then signed with the RAs private key,
[h]
RA
. Figure 3 depicts the signing step. The signed hash
value, [h]
RA
, is sent back to the manufacturer along with the
assigned license number.
Once the R-CFG has been licensed, signed, and placed on
a server, the SDR-MDs can contact the server at any time to
download the combination of header, R-CFG, and [h]
RA
.
After an SDR-MD has connected to the manufacturers
server, mutual authentication is performed. The mutual
authentication step avoids masquerade and replay attacks.
When using an unsecured connection, this is done by ex-
changing random challenges (nonces) or by certicates,
while when using a secure connection, the protocol that pro-
vides the secure connection is assumed to take care of the
mutual authentication.
After the mutual authentication step has been success-
fully completed, the SDR-MD requests and downloads the
new R-CFG. Upon download completion, R-CFG verica-
tion is necessary to guarantee that the R-CFG has been ap-
proved by the RA and properly signed. The verication step
also tests whether the R-CFG is appropriate for the device
(Figure 4).
However, to guarantee that the R-CFGhas not been mod-
ied after being approved and signed by the RA, the follow-
ing steps are performed:
(1) a new hash value h

= MD(headerR-CFG) is calcu-
lated;
(2) the received [h]
RA
is decrypted to obtain h;
(3) h and h

are compared: if h = h

, the received R-CFG


is accepted. However, if h = h

, the R-CFG is rejected.


Figure 4 also shows the data integrity check. If the new R-
CFG has passed all the tests, it is then installed and the value
of Att(R-CFG) is stored in R1.
The steps of the secure SDR R-CFG download protocol
when using an unsecured connection, such as HTTP, are de-
picted in Figure 5. Dashed arrows indicate communication
inside the SDR-MD.
Although the protocol is specied using an unsecured
connection, the R-CFG is still protected since it is encrypted
with the EK, thus only that specic device which has initi-
ated the connection can correctly decrypt and install the R-
CFG. Details on how to obtain a lightweight secure connec-
tion using the Light SSL (LSSL) protocol, specied in the se-
cure/unsecured internet connection module, can be found in
[19].
The SDR R-CFG download protocol initiates with the
SDR-MD contacting the manufacturers server and estab-
lishing an unsecured connection. Next, the SDR-MD sends
MD(EK) and a nonce C encrypted by the EK. The manufac-
turer maintains a database of all available EKs (M EKDB),
indexed by MD(EK). The database has all information that
the manufacturer needs about each SDR-MD it has pro-
duced.
When the manufacturer receives the MD(EK), it searches
in its M EKDB for that value. If it does not nd the MD(EK),
the manufacturer ends the connection. On the other hand, if
MD(EK) is in M EKDB, then the manufacturer obtains the
EK of that device and generates a new nonce C

. The C

is
then encrypted by the EK and sent, along with C, to the SDR-
MD.
Upon receiving C and C

, the SDR-MD authenticates the


manufacturer if the received C is equal to the one that the
SDR-MD has previously generated. If authentication fails,
the SDR-MD terminates the connection; otherwise, it ob-
tains C

, sends it back to the manufacturer, and requests the


necessary R-CFG.
The manufacturer then authenticates the device. If au-
thentication fails, the manufacturer terminates the connec-
tion; otherwise, it sends the requested R-CFG encrypted by
the EK. The SDR-MD receives the R-CFG, veries it, and
checks the R-CFG data integrity. If the R-CFG tests show
no negative results, the SDR-MD installs the R-CFG and ac-
knowledges the manufacturer. The connection is then re-
leased.
After releasing the connection, the SDR-MD installs the
R-CFG and stores the Att(R-CFG) value in R1. Whenever the
SDR-MD is booting up, the AM Eng calculates a new Att(R-
CFG) value, which is then passed to the AC Eng to be com-
pared with R1. If Att(R-CFG) = R1, the current radio cong-
uration is trusted. On the other hand, if Att(R-CFG) = R1,
the SDR-DM blocks the use of any service.
3.4. The anticloning scheme
One of the more dangerous threats in SDR wireless commu-
nication is cloning. SDR-MD cloning is considered a federal
crime. According to [20], telecommunication fraud losses
are estimated at more than a billion dollars yearly. A large
amount of this loss is due to cloning. Besides illegal billing,
cloned units increase the competition of shared resources,
which increases network congestion and degrades network
services. Furthermore, the impact of overload trac from
406 EURASIP Journal on Wireless Communications and Networking
Random
no.
generator
AR Eng
TRC1-read only TRC2-read/write
AC Eng
AM Eng EK
RA pubkey
Protected
storage
R2
R1
R0
Att(EK)
Att(R-CFG)
0-cloned
1-valid
AS Eng
Figure 2: The tamper-protected hardware package in an invalid state.
Header
Header
R-CFG
R-CFG
RAs private key
[h]
RA
h H
Figure 3: R-CFG validation.
Header
R-CFG
RAs public key
[h]
RA
h
h

H
=
?
Header is checked
R-CFG verication Data integrity check
Figure 4: R-CGF verication and data integrity check.
cloned units is unpredictable. Thus, the estimation of traf-
c patterns is imprecise for network planning.
The anticloning scheme, which is part of the proposed
fraud-prevention framework, is designed to provide a core
set of hardware and software technologies that provide the
basis for a wireless network environment free of cloned units.
Unlike other cloning detection schemes, the proposed
anticloning scheme not only detects cloned units, but also
elevates the level of diculty to clone a valid unit. Also, as
a new feature, the SDR-MD is aware of cloning, that is, an
SDR-MD is able to discover if it is a cloned unit and take
the necessary steps to block the use of the network services.
Another advantage is that the anticloning framework is in-
dependent of technology, working well for dierent wireless
technologies.
3.4.1. Entering a valid state
The SDR-MD comes from the manufacturer in an invalid
state, that is, it does not have the AC, therefore, it cannot
identify itself to the network. After obtaining the AC, the
SDR-MD enters a temporary state, that is, it is able to prove
its identity, however, it does not have a phone number yet, it
does not have the CFGle installed. After obtaining the CFG,
the SDR-MD nally reaches a valid state. It is able to identify
itself and use the network services.
Figure 6 depicts the transition states that the SDR-MD
has to go through in order to reach a valid state. Note that
anytime after the SDR-MD has reached the valid state, it may
need a newR-CFGle or a newCFGle. While obtaining any
of those les, the SDR-MD goes to a temporary state. With
the new data locally stored, the security checks are executed
and the SDR-MD goes back to the valid state.
To obtain a valid AC, the SDR-MD has to execute the at-
testation credential protocol (ACP) depicted in Figure 7. The
ACP is a communication process between the SDR-MD and
the Privacy CA and it is executed only one time per each EK.
Whenever the manufacturer generates a new EK, it in-
forms the Privacy CA, in a safe way, about that EK. The
Privacy CA, like the manufacturer, maintains a database of
all available EKs (CA EKDB), indexed by MD(EK). This
database has all information that the Privacy CA needs to
know about each SDR-MD produced and links each SDR-
MD to its AC.
The ACP steps are dened as follows. First, the SDR-
MD contacts the Privacy CA and sends the value R0 =
Att(EK). The Privacy CA looks for a matching MD(EK) in
the CA EKDB. If it nds a match, the Privacy CA obtains the
EK of that unit and acknowledges the unit. If no equivalent
MD(EK) is found, either the manufacturer failed to inform
the Privacy CA about this unit or this is an invalid EK. Thus,
the Privacy CA does not provide an AC to the unit.
A Fraud-Prevention Framework for SDR Mobile Devices 407
M
D
(
E
K
)
i
n
M
E
K
D
B
?
G
e
t
s
E
K
O
b
t
a
i
n
s
C
G
e
n
e
r
a
t
e
s
C

K
E
K
{
C

}
C

,
r
e
q
.
R
-
C
F
G
t
y
p
e
X
V
e
r
i

e
s
C

K
E
K
{
R
-
C
F
G
}
A
C
K
E
n
d
A
C
K
O
b
t
a
i
n
s
R
-
C
F
G
R
-
C
F
G
v
e
r
i
f
.
R
-
C
F
G
d
a
t
a
i
n
t
.
I
n
s
t
a
l
l
s
R
-
C
F
G
K
E
K
{
R
-
C
F
G
}
C

O
b
t
a
i
n
s
C
V
e
r
i

e
s
C
O
b
t
a
i
n
s
C

K
E
K
{
C

}
M
D
(
E
K
)
,
K
E
K
{
C
}
G
e
n
e
r
a
t
e
s
C
SDR-DM
S
D
R
-
D
M
SDR-TPHP
M
D
(
E
K
)
,
K
E
K
{
C
}
E
s
t
a
b
l
i
s
h
e
s
u
n
s
e
c
u
r
e
c
o
n
n
e
c
t
i
o
n
Manufacturers server SDR device
Figure 5: The secure SDR R-CFG download protocol.
Invalid
state
Gets AC
Temp
state
Gets CFG
Needs CFG
Gets
R-CFG
Needs
R-CFC
Temp
state
Valid
state
Figure 6: Transition states of an SDR-MD.
Second, the Privacy CA generates an AK pair and the unit
authenticates the Privacy CA. The unit generates a nonce C
and sends it to the Privacy CA encrypted by the EK. The Pri-
vacy CA obtains C and sends it back along with an encrypted
message containing the AK pair. Upon receiving the message,
the unit veries C, authenticating the Privacy CA.
Third, after authenticating the Privacy CA, the unit ob-
tains the AK pair and acknowledges the Privacy CA. The Pri-
vacy CA then generates the AC = [AK
pub
]
Privacy CA
and sends
it, encrypted by the AK
pub
to the unit. The unit receives the
AC, decrypts it, and stores it in its TPHP. After that, the con-
nection is nally released.
After obtaining the AC, the nal step to enter the valid
state is to have the SDR-MD executing the CFG update pro-
tocol (CUP) to obtain a valid CFG. This protocol is executed
whenever the unit needs a new phone number. Figure 8 de-
picts the CUP step by step.
After connecting to the WOs server, the unit sends its
AC and the value of R2 = Att(CFG) along with a nonce C
encrypted by the WOs public key. The WOs public key is
obtained a priori through a secure protocol. If this is a new
unit, the value of R2 is null.
Upon receiving the AC, the WO veries if the AC is null.
If the comparison is positive, the unit is a clone and the WO
terminates the connection. Otherwise, the CUP continues its
normal ow.
The WO uses the Privacy CAs public key and decrypts
the AC, obtaining the AK
pub
. The WO has a database (DB),
indexed by the AK
pub
, that contains information about each
SDR-MD in a valid state, such as phone number and user
name. Next, the WO looks for a matching AK
pub
in the DB.
If it nds a match, it veries MD(CFG) = R2. If the compar-
ison is negative, this is an invalid unit; either this is a cloned
unit or a masquerade attack is occurring, and countermea-
sures are taken.
On the other hand, if the comparison is positive, this is a
valid unit. The WO then obtains C and generates a nonce C

to authenticate the unit. C is concatenated with C

and sent
encrypted by the AK
pub
to the SDR-MD. If the AK
pub
is not
in the DB, this is a unit in the temporary state.
Upon receiving K
AKpub
{CC

} from the WO, the unit au-


thenticates the WO if the received C is equal to the one pre-
viously generated. If authentication fails, the SDR-MD ter-
minates the connection. Otherwise, it sends C

back to the
WO.
Next, the WO authenticates the unit by verifying C

.
If authentication fails, the WO terminates the connection.
Otherwise, the WO generates a new CFG and stores the
MD(CFG) value in the DB. The unit receives the CFG en-
crypted by its AK
pub
and decrypts it. The unit then stores the
CFG in the protected storage of TRC2 and installs the new
phone number.
Next, the AM Eng measures Att(CFG) and writes the
value in R2. The unit then sends this value encrypted by the
WOs public key to the WO. The WO veries the value and
acknowledges the unit if the comparison is positive. Other-
wise, it informs the unit that an error occurred during the
CFG installation step. This step is repeated in the case of
408 EURASIP Journal on Wireless Communications and Networking
A
t
t
(
E
K
)
i
n
E
K
D
B
?
A
t
t
(
E
K
)
O
b
t
a
i
n
s
E
K
C
o
n
n
e
c
t
i
o
n
e
s
t
a
b
l
i
s
h
e
d
A
C
K
G
e
n
e
r
a
t
e
s
A
K
p
a
i
r
K
E
K
{
C
}
O
b
t
a
i
n
s
C
K
E
K
{
C

(
A
K
p
a
i
r
)
}
A
C
K
G
e
n
e
r
a
t
e
s
A
C
A
C
A
C
K
C
o
n
n
e
c
t
i
o
n
r
e
l
e
a
s
e
d
Privacy CA
SDR-TPHP
S
D
R
-
M
D
SDR-DM
R
e
q
u
e
s
t
R
0
R
=
A
t
t
(
E
K
)
A
C
K
G
e
n
e
r
a
t
e
s
C
K
E
K
{
C
}
K
E
K
{
C

(
A
K
p
a
i
r
)
}
O
b
t
a
i
n
s
C
A
u
t
h
e
n
t
i
c
a
t
e
s
C
A
O
b
t
a
i
n
s
A
K
p
a
i
r
A
C
K
A
C
S
t
o
r
e
s
A
C
A
C
K
Figure 7: Attestation credential protocol.
WO
R
e
q
.
n
e
w
C
F
G
A
C
,
{
R
2

C
}
W
O
O
b
t
a
i
n
s
A
K
p
u
b
f
r
o
m
A
C
O
b
t
a
i
n
s
A
t
t
(
C
F
G
)
V
e
r
i

e
s
A
t
t
(
C
F
G
)
O
b
t
a
i
n
s
C
G
e
n
e
r
a
t
e
s
C
C
o
n
n
e
c
t
i
o
n
e
s
t
a
b
l
i
s
h
e
d
K
A
K
p
u
b
{
C

}
C

V
e
r
i

e
s
C

G
e
n
e
r
a
t
e
s
C
F
G
S
t
o
r
e
s
M
D
(
C
F
G
)
K
A
K
p
u
b
{
C
F
G
}
{
R
2
}
W
O
V
e
r
i

e
s
A
t
t
(
C
F
G
)
=
M
D
(
C
F
G
)
A
C
K
C
o
n
n
e
c
t
i
o
n
r
e
l
e
a
s
e
d
S
D
R
-
M
D
SDR-TPHP
SDR-DM
R
e
q
.
A
C
,
R
2
G
e
n
e
r
a
t
e
s
C
A
C
,
{
R
2

C
}
W
O
K
A
K
p
u
b
{
C

}
O
b
t
a
i
n
s
C
V
e
r
i

e
s
C
O
b
t
a
i
n
s
C

K
A
K
p
u
b
{
C
F
G
}
O
b
t
a
i
n
s
C
F
G
S
t
o
r
e
s
C
F
G
I
n
s
t
a
l
l
s
C
F
G
R
2
=
A
t
t
{
C
F
G
}
A
C
K
Figure 8: The CFG update protocol.
errors. After receiving an acknowledgment, the unit releases
the connection.
After obtaining the AC from the Privacy CA and the CFG
le from the WO, the SDR-MD nally reaches a valid state.
Therefore, the unit is ready to use all the services oered
by the WO. Figure 9 depicts the tamper-protected hardware
package when the SDR-MD is in the valid state.
Note that the clone signal, sent by the ACEng, propagates
outside the TPHP to the CPU and inside the TPHP to the
TRC2, where it sets the AC to null if the SDR-MD is a clone
unit.
3.4.2. Cloning-aware procedure
The cloning-aware procedure is implemented in both sides,
the WO and the SDR-MD, and is responsible for detecting
whether the SDR-MD is a valid unit or a cloned unit.
After the unit has connected to the WO and requested
a service, the cloning-aware procedure starts in the SDR-
MD side. New Att(EK) and Att(CFG) values are measured
by the AM Eng and sent to the AC Eng, which also receives
the current value of R0 and R2 from the AR Eng. The AC
Eng compares the values and signalizes 1 for a valid unit, if
Att(EK) = R0 and Att(CFG) = R2, or 0 for a cloned unit, if
Att(EK) = R0 or Att(CFG) = R2. In this fashion the SDR-
MD is aware of cloning. Figure 10 illustrates the procedure.
If the SDR-MD is a valid unit, the AC is sent and the WO
cloning-aware procedure begins.
In the WO side, the procedure works basically as an au-
thentication module. The WO obtains the AC and veries if
it is valid or null. If the AC is null, the WO terminates the
connection, since the unit is a clone. Otherwise, the WO ob-
tains the AK
pub
from the AC and looks for a match in the DB.
If there is no match, the service is denied. If there is a match,
the WO prepares to authenticate the unit. If the unit is cor-
rectly authenticated, the WO allows the use of the service.
On the other hand, if the unit is not authenticated, the WO
concludes that this unit is trying to use other units AC (mas-
querade attack) and denies the service. Figure 11 illustrates
the procedure.
A Fraud-Prevention Framework for SDR Mobile Devices 409
Random
no.
generator
0-cloned
1-valid
EK AM Eng
AC Eng
AR Eng
TRC1-read only
AS Eng
R0
R1
R2
Att(EK)
Att(R-CFG)
Att(CFG)
AK
priv
AK
pub
AK pair
WO
pub
RA pubkey
AC
A
n
d
AC
CFG
TRC2-read/write
Figure 9: The tamper-protected hardware package in a valid state.
WO
Connection
established
SDR-DM AM Eng AC Eng AR Eng
Req. service
AC
WO proc.
Starts proc.
Sends signal
Ends proc.
Att(EK)
Att(CFG)
Att(EK),
Att(CFG)
Compares
Reads R0
Reads R2
R0, R2
Figure 10: Cloning-aware procedure: SDR-MD side.
WO
C
o
n
n
e
c
t
i
o
n
e
s
t
a
b
l
i
s
h
e
d
A
C
O
b
t
a
i
n
s
A
K
p
u
b
f
r
o
m
A
C
C
h
e
c
k
s
A
K
p
u
b
i
n
D
B
G
e
n
e
r
a
t
e
s
C
C
K
A
K
p
r
i
v
{
C
}
V
e
r
i

e
s
C
K
A
K
p
r
i
v
{
C
}
O
b
t
a
i
n
s
C
C A
C
SDR-DM
R
e
q
.
A
C
S
D
R
-
M
D
SDR-TPHP
S
D
R
-
M
D
i
s
a
b
l
e
t
o
u
s
e
s
e
r
v
i
c
e
s
Figure 11: Cloning-aware procedure: WO side.
4. PRELIMINARY EXPERIMENTS
The experiments were executed using J2ME, which is a
lightweight java version, specically designed to be used with
constrained devices. The experiments set-up is depicted in
PDA
client
Wireless link
(11 Mbps)
End-to-end
security
Manufacturer
server with SSL
Figure 12: The experiment set-up.
Figure 12. An SDR-MD, in this case a Sharp Zaurus PDA SL-
5600 with CPU speed of 400 MHz, 32 MB SDRAM, Linux
OS, and J2ME support, connects through an 11 Mbps wire-
less link to a Pentium 4 2.6 GHz server with 256 MB RAM.
4.1. The secure SDR R-CFGdownload protocol
Two preliminary experiments involving the secure SDR R-
CFG download protocol and the secure R-CFG/CFG mod-
ule are described. In the rst experiment, the time the R-
CFG/CFG security module takes to identify invalid R-CFGs
and delete them is measured. The second experiment com-
pares the secure protocol execution when using an unsecured
connection : HTTP, a lightweight secure connection, LSSL
[19], and the SSL protocol [21].
The graph in Figure 13 shows the results of the rst ex-
periment. The MD5 algorithm is used to calculate the nger-
print and to perform the data integrity check. As expected,
the larger the R-CFG is, the longer it takes to perform the
security checks.
Figure 14 depicts the results of the second experiment.
Note that the secure protocol with unsecured connection
presents best performance, since it does not need to spend
time with the cipher suite handshake and other extra steps
needed by secure connections. In case secure connections are
necessary, the use of the LSSL is suggested since it presents
better performance than the SSL, as can be noticed in this
experiment.
4.2. Anticloning scheme
It is expected that the anticlone scheme will not add any
further delay on the obtainment of network services when
comparing with the GSM and UMTS techniques. Although
SDR mobile devices are constrained by nature, encryption
and decryption operations are only executed for small pieces
410 EURASIP Journal on Wireless Communications and Networking
128 256 512 1M
R-CFG (KB)
600
650
700
750
800
850
1000
1100
T
i
m
e
(
m
s
)
Figure 13: Time to identify invalid R-CFGs.
128 256 512 1M
R-CFG (KB)
10
25
40
55
70
85
100
115
130
145
160
175
190
T
i
m
e
(
m
s
)
Secure protocol with HTTP
Secure protocol with LSSL
Secure protocol with SSL
10
2
Figure 14: Comparing the secure protocol varying the connection
type.
of information such as the 2048-bit EK and AK pair, and
the 48-bit nonce C. Furthermore, the attestation engines and
the random number generator in the TPHP are specialized
pieces of hardware that can quickly execute data integrity
measurements and generate a 48-bit random number.
5. CORRECTNESS PROOFS
This section presents a list of possible attacks involving the
R-CFG les and how the secure SDR R-CFG protocol avoids
those attacks. It then continues with correctness proofs that
show that the fraud-prevention framework provides an envi-
ronment free of cloned units.
Table 4 illustrates common methods of attacks that fail
against the proposed protocol.
Next, the correctness proofs are presented. It begins with
three lemmas. The rst lemma shows that only an SDR-MD
with a valid EK is provided an AC. The second lemma shows
that an SDR-MD only obtains a new CFG when its identity
is successfully proved. Finally, the third lemma shows that
only valid CFGs, that is, CFGs that have been generated and
signed by the WO, can be installed by an SDR-MD.
The proofs continue with two nal theorems. The rst
theorem proves that there is no possibility to clone an SDR-
MD over the air. The second theorem guarantees that only a
valid SDR-MD can use the network services.
Lemma 1. The Privacy CA only attests the identity of SDR-
MDs that have valid EKs.
Proof. Since the Privacy CA has a database of valid EKs and
this database is assumed to be secured stored, any SDR-MD
that requests an AC and sends an invalid MD(EK) value, that
is, hash of an EK that is not generated by the manufacturer,
has the AC denied.
A replay attack is not possible since the ACP is executed
only once per each EK. Impersonation of the SDR-MD, that
is, masquerade attack, is noticed by the authentication step.
Lemma 2. No SDR-MD obtains a CFG le unless its identity
is successfully proved.
Proof. According to the CUP denition, only after being au-
thenticated by the WO, the SDR-MD is given a new CFG.
This eliminates the possibility of masquerade attacks and re-
play attacks.
Only after responding correctly to the challenge gener-
ated by the WO, the SDR-MD is given a new CFG. Therefore,
no SDR-MD obtains a new CFG le unless it has proved its
identity.
Lemma 3. Only valid CFG les are installed in each SDR-MD.
Proof. To install a new CFG, the SDR-MD must execute the
CUP. According to the CUP denition, before receiving a
new CFG the SDR-MD authenticates the WO by verifying
{R2}
WO
= [MD(CFG)]. If the comparison is positive, then
the SDR-MD authenticates the WO. Thus, masquerade and
replay attacks are eliminated.
After authentication, the SDR-MD receives a new CFG =
[Phone no.]
WO
. Since masquerade and replay attacks fail,
only the WO could have sent this message, and the nal step
to validate the CFG occurs. The SDR-MD veries the WOs
signature in the CFG. When the signature is successfully ver-
ied, the CFG is considered valid and the TPHP stores and
installs the new CFG.
Theorem 1. It is guaranteed that there is no possibility to clone
an SDR-MD over the air.
Proof. In order to clone an SDR-MDover the air, one attacker
must obtain the EK of the victim or a combination of valid
AK pair, valid AC, and valid CFG.
Since the EK and AK
private
are never disclosed by the
TPHP, the attacker has no possibility to obtain the EK nor
the AK pair of a victim. According to Lemma 2, the attacker
must prove its identity to obtain a valid CFG, thus if the at-
tacker uses an AC that is not his/hers, the WO will notice it
and deny a new valid CFG.
A Fraud-Prevention Framework for SDR Mobile Devices 411
Table 4: Possible attacks and how the secure protocol avoids them.
Attacks Description Protection
Access control
Clients using unauthorized services
Protocol employs client authentication
or trying to download data they should not
Masquerade
An entity pretends to be the manufacturer server
Protocol uses mutual authentication
or a client
Condentiality R-CFG might be condential
By establishing secure connections or
encrypting, the R-CFG proprietary information
are kept secret
Replay Messages are captured and retransmitted later Mutual authentication avoids replay attacks
Invalid R-CFGs
Installing R-CFGs that are not approved by the RA
Every R-CFG is digitally signed
by the RA and veried by the SDR-MD
R-CFG Integrity R-CFG modied after it has been approved
Protocol employs one-way hash functions
to guarantee data integrity
With no other way to clone an SDR-MD over the air, the
only way to bill someone elses account is to capture his/her
AC when transmitted over the air. However, the WOcloning-
aware procedure will detect that the captured AC does not
belong to that unit and it will deny any service.
Theorem 2. It is guaranteed that only a valid SDR-MD can
use the wireless operator services.
Proof. According to the WO cloning-aware procedure, in or-
der to use the network services the SDR-MD must present
a valid AC. By Lemma 1, only SDR-MDs with valid EKs are
able to obtain a valid AC. Therefore, unit with an invalid EK
does not have a valid AC and cannot use the WOs services.
According to Theorem 1, there is no way to clone an
SDR-MD over the air, and impersonation of other SDR-MDs
by capturing their AC is noticed by the WO cloning-aware
procedure. Thus, the only other way to clone an SDR-MD is
to have physical access to its TPHP.
However, if an attacker successfully disassembles the
TPHP without damaging it and is able to copy the TPHP
to another SDR-MDs TPHP, Lemma 3 and the SDR-MD
cloning-aware procedure guarantee that the SDR-MD that
received the cloned TPHP denies the use of the network ser-
vices. The value of R2 on the cloned TPHP and the value of
the current MD(CFG) in the device are dierent. Thus, the
SDR-MD blocks the use of any services.
Since the SDR-MD cloning-aware procedure blocks the
use of any service by cloned units and the WO cloning-aware
procedure notices masquerade attacks, it is guaranteed that
only a valid SDR-MD can use the wireless operator services.
In summary, the fraud-prevention framework elevates
the level of diculty to clone an SDR-MD. The only way to
clone one SDR-MD that employs the framework would be
disassembling the TPHP from the SDR-MD and reading its
contents. Since the TPHP is physically bound to other parts
of the SDR-MD, attempts to disassemble it would probably
damage the TPHP. Even if an attacker successfully disassem-
bles the TPHP without damaging it, the equipment to read
and copy the TPHP is so expensive that the attacker would
practically have no gain, if any, in doing so.
6. CONCLUSION
To greatly enhance the overall security of SDR-MDs, a fraud-
prevention framework is proposed. The fraud-prevention
framework is composed of new pieces of hardware, mod-
ules, and protocols. The framework oers security monitor-
ing against malicious attacks and viruses, protects sensitive
information, creates and protects an identity for the system,
employs a secure protocol for radio conguration download,
and nally, establishes an anticloning scheme which guaran-
tees that no units can be cloned over the air, and elevates the
level of diculty to clone units if the attacker has physical
access to the mobile device. Even if cloned units exist, the an-
ticloning scheme is able to identify and deny services to those
units.
Preliminary experiments show that the framework is
able to identify invalid R-CFGs with minimal delay. Proofs
that analyze correctness of the framework show that the
fraud-prevention framework provides an environment free
of cloned units.
Future work includes the execution of several ex-
periments that will measure performance of the fraud-
prevention framework, and comparisons with other state-of-
the-art related works.
REFERENCES
[1] B. Bing and N. Jayant, A cellphone for all standards, IEEE
Spectr., vol. 39, no. 5, pp. 3439, 2002.
[2] Software Dened Radio Forum website, http://www.
sdrforum.org.
[3] Java 2 Micro Edition Technology website, http://wireless.
java.sun.com/j2me.
412 EURASIP Journal on Wireless Communications and Networking
[4] L. B. Michael, M. J. Mihaljevic, S. Haruyama, and R. Kohno,
A framework for secure download for software-dened ra-
dio, IEEE Commun. Mag., vol. 40, no. 7, pp. 8896, 2002.
[5] C. Castelluccia and F. Dupont, A Simple Privacy Extension
for Mobile IPv6, The Internet Engineering Task Force, Internet
Draft: Draft-Castellucia- MobileIP-Privacy, February 2001.
[6] A. Escudero, Location privacy in IPv6tracking binding
updates, in Proc. International Workshop on Interactive Dis-
tributed Multimedia Systems and Telecommunication (IDMS
01), Lancaster, UK, September 2001.
[7] CMS 88 Cellular Mobile TelephoneSystem, EN/LZT 101908,
Ericsson.
[8] Global System for Mobile Communication, The GSM
security technical whitepaper for 2002, http://www.
hackcanada.com/blackcrawl/.
[9] UC Berkeley. Internet Security, Applications, Authen-
tication and Cryptography Group, GSM cloning,
http://www.isaac.cs.berkeley.edu/isaac/gsm.html.
[10] Overview of the Universal Mobile Telecommunication
System. DRAFT, July 2002, http://www.umtsworld.com/
technology/overview.htm.
[11] M. B. Frederick, Cellular telephone fraud anit-fraud system,
US Patent 5,448,760, September 1995.
[12] M. S. M. Annoni Notare, F. A. da Silva Cruz, B. Goncalves
Riso, and C. B. Westphall, Wireless communications: secu-
rity management against clonedcellular phones, in Proc. IEEE
Wireless Communications and Networking Conference (WCNC
99), vol. 3, pp. 14121416 , NewOrleans, La, USA, September
1999.
[13] The Trusted Computing Group, http://www.trusted-
computinggroup.org.
[14] The TCG PC Specic Implementation Specication, http://
www.trustedcomputinggroup.org/downloads/.
[15] The RC5 encryption algorithm, http://www.secinf.net/
cryptography/The RC5 Encryption Algorithm.html.
[16] RSA encryption, http://mathworld.wolfram.com/RSA-
Encryption.html.
[17] Federal Communications Commission. Authorization and
use of software dened radio: rst report and order,
September 2001, http://www.fcc.gov/Bureaus/Engineering
Technology/Notices/2000/fcc00430.txt.
[18] Intel, Intel wireless trusted platform: security for mo-
bile devices, http://www.intel.com/design/pca/application-
sprocessors/whitepapers/300868.htm.
[19] A. Brawerman, D. Blough, and B. Bing, Securing the down-
load of radio conguration les for software dened radio
devices, in Proc. ACM International Workshop on Mobility
Management and Wireless Access (MobiWac 04), pp. 98105,
Philadelphia, Pa, USA, SeptemberOctober 2004.
[20] US Secret Service Financial Crimes Division, http://www.
secretservice.gov/nancial crimes.shtml#Telecommunications.
[21] A. O. Freier, P. Karlton, and P. C. Kocher, The SSL protocol
Version 3.0, http://home.netscape.com/eng/ssl3.
Alessandro Brawerman is a graduate stu-
dent pursuing a Doctoral degree in the
School of Electrical and Computer Engi-
neering, Georgia Institute of Technology.
He received his B.S. and M.S. degrees in
computer science from the Federal Uni-
versity of Parana, Brazil, during December
1997 and May 2000, respectively. His re-
search interests are in security for software
dened radio devices, such as cellphones
and PDAs, and security and management of wireless networks. He
currently holds a scholarship from Brazil and his research is funded
by the Brazilian National Council of Research (CNPq). He has pub-
lished and presented over 10 technical papers. He is an IEEE Mem-
ber and was a Fellow of the Panasonic Information & Network-
ing Technologies Laboratory (20032004). He was also the Instruc-
tor of several computer science courses at the Federal University of
Parana (19982000).
John A. Copeland has been the John H.
Weitnauer, Jr. Chaired Professor at the
School of Electrical and Computer Engi-
neering, Georgia Institute of Technology,
from 1983 up to date. He teaches grad-
uate and undergraduate courses on com-
munications networks and network se-
curity, and does research in those areas
(http://www.csc.gatech.edu). He was Direc-
tor of the Georgia Center for Advanced
Telecommunications Technology from 1993 to 1996, Vice Pres-
ident of Technology at Hayes Microcomputer Products from
1985 to 1993, and Vice President of Engineering Technology at
Sangamo Weston, Inc. from 1982 to 1985, and did research at
Bell Labs on semiconductor circuits and optical ber networks
from 1965 to 1982. He received the B.S., M.S., and Ph.D. de-
grees in physics from the Georgia Institute of Technology. He
has been awarded 38 patents and has published over 60 tech-
nical papers. In 1970, he received the IEEEs Morris N. Lieb-
mann Award. He is a Fellow of the IEEE and has served as
the Editor of the IEEE Transactions on Electron Devices. He
served on the Board of Trustees for the Georgia Tech Research
Corporation from 1983 to 1993. He is a Member of Infra-
gard, the ISSA, and the ACM SIGSAC. In 2000, he invented the
StealthWatch network behavior anomaly detection system, and
founded Lancope (http://www.lancope.com) which has deployed
the StealthWatch system in over 100 corporate and government
networks.