Future Internet

EURASIP Journal on Wireless Communications and Networking
Reconfigurable Radio for Future

Generation Wireless Systems
Guest Editors: Frederik Petré, Ahmet Kondoz,

Stefan Kaiser, and Ashish Pandharipande
EURASIP Journal on
Wireless Communications and Networking

EURASIP Journal on
Wireless Communications and Networking

Guest Editors: Frederik Petré, Ahmet Kondoz,
Stefan Kaiser, and Ashish Pandharipande
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 Petré, Ahmet Kondoz, Stefan Kaiser, and Ashish Pandharipande
Volume 2005 (2005), Issue 3, Pages 271-274
Software-Defined Radio—Basics and Evolution to Cognitive Radio, Friedrich K. Jondral

Flexible Radio: A Framework for Optimized Multimodal Operation via Dynamic Signal Design,
Ioannis Dagres, Andreas Zalonis, Nikos Dimitriou, Konstantinos Nikitopoulos,
and Andreas Polydoros
Adaptive Transmitter Optimization in Multiuser Multiantenna Systems: Theoretical Limits, Effect of

Delays, and Performance Enhancements, Dragan Samardzija, Narayan Mandayam,
and Dmitry Chizhik
Flexible Transmission Scheme for 4G Wireless Systems with Multiple Antennas, François Horlin,
Frederik Petré, Eduardo Lopez-Estraviz, Frederik Naessens, and Liesbet Van der Perre
Reconfigurable Signal Processing and Hardware Architecture for Broadband Wireless

Communications, Ying-Chang Liang, Sayed Naveen, Santosh K. Pilakkat, and Ashok K. Marath
Modular Software-Defined Radio, Arnd-Ragnar Rhiemeier

Adaptive Mobile Positioning in WCDMA Networks, B. Dong and Xiaodong Wang

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
Flexible Analog Front Ends of Reconfigurable Radios Based on Sampling and Reconstruction with
Internal Filtering, Yefim S. Poberezhskiy and Gennady Y. Poberezhskiy
A Reconfigurable Spiral Antenna for Adaptive MIMO Systems, Bedri A. Cetiner, J. Y. Qian, G. P. Li,
and F. De Flaviis
Multimode Communication Protocols Enabling Reconfigurable Radios, Lars Berlemann, Ralf Pabst,
and Bernhard Walke
Towards a Fraud-Prevention Framework for Software Defined Radio Mobile Devices,
Alessandro Brawerman and John A. Copeland
EURASIP Journal on Wireless Communications and Networking 2005:3, 271–274
c 2005 Hindawi Publishing Corporation
Editorial
Frederik Petré
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 band processor for every communication mode. However,

Future-generation wireless systems aim to support a spec- with the growing number of standards and communication
trum of services over a variety of networks in a way transpar- modes, this approach is becoming increasingly infeasible and
ent to the user. Flexibility and adaptivity are key ingredients economically unacceptable. A more efficient approach to-
of such future-generation wireless systems in order to deliver wards this design is to adopt a reconfigurable (as opposed to
optimal quality of service (QoS) for different applications fixed) radio concept, such that the terminal can adapt to the
over diverse communication environments. Rather than re- best-suited communication mode under the control of a QoS
lying on the traditional horizontal communication model, manager. A high degree of flexibility is required not only for
consisting of a single wireless access system, these future 4G the digital baseband processing but also for the analog radio
systems will employ a vertical communication model, which frequency (RF) front end, which should accept a large range
integrates different existing and new evolving wireless access of carrier frequencies, possess a flexible bandwidth, and deal
systems on a common IP-based platform, to complement with a wide variety of operational conditions. Likewise, the
each other for different service requirements and radio envi- same high degree of flexibility is called for not only at the
ronments. To enable seamless and transparent interworking physical layer but also at the medium access control (MAC)
between these different wireless access systems, or communi- (and possibly higher) layer(s), to be compatible with the pro-
cation modes, through horizontal (intrasystem) and vertical tocols of the different standards.
(intersystem) handovers, multimode functionality is needed
to support the different existing air interfaces and the newly 2. OVERVIEW OF THE SPECIAL ISSUE
emerging ones.
It is expected that multimode capabilities will be ulti- This special issue, which has been conceptualized within the
mately focussed on the terminal side to target a larger mar- framework of the IST-FP6 Network of Excellence in Wire-
ket base. New challenges then appear in terms of minimiz- less COMmunications (NEWCOM), and, more specifically,
ing the terminal cost, size, and power consumption, while within the context of NEWCOM Project D on “Flexible Ra-
at the same time maximizing its flexibility with respect to dio,” contains 3 invited papers and 9 regular papers.
communication standards as well as its adaptivity with re- The first (invited) paper “Software-defined radio—
spect to varying user requirements and changing communi- Basics and evolution to cognitive radio,” by F. K. Jondral,
cation conditions. The conventional approach to the design reviews the basic concepts and terminology of software-
of a multimode terminal is the provision of a custom base- defined radio (SDR) and discusses its future evolution
272 EURASIP Journal on Wireless Communications and Networking
towards cognitive radio. The author further emphasizes the on the utility of pipelining as a general design guideline for
importance of standardization and introduces the so-called modular software-defined radio.
software communications architecture (SCA) as an exam- The seventh paper “Adaptive mobile positioning in
ple framework that allows an object-oriented development WCDMA networks,” by B. Dong and X. Wang, introduces
of SDRs. a technique for mobile tracking in wideband code-division
multiple-access (WCDMA) systems employing multiple re-
2.1. Flexible baseband processing ceive antennas. To achieve a high estimation accuracy, the al-
The second (invited) paper “Flexible radio: A framework gorithm utilizes the time difference of arrival (TDOA) mea-
for optimized multimodal operation via dynamic signal de- surements in the forward link pilot channel, the angle of ar-
sign,” by I. Dagres et al., introduces a general framework rival (AOA) measurements in the reverse-link pilot channel,
for the study and design of flexible/reconfigurable radio sys- as well as the received signal strength. The proposed algo-
tems, with a special focus on the baseband portion of the rithm jointly tracks the unknown system parameters as well
physical layer and its interactions with procedures taking as the mobile position and velocity.
place in the higher layers. Furthermore, the authors describe
specific tools and fundamentals that underpin such flexible 2.2. Flexible analog RF front ends
transceiver architectures to provide multistandard capabili- The eighth paper “Flexible frequency discrimination sub-
ties, channel adaptivity, and user/service personalization. systems for reconfigurable radio front ends,” by B. Carey-
The third (invited) paper “Adaptive transmitter opti- Smith et al., surveys recent advances in flexible, frequency-
mization in multiuser multiantenna systems: Theoretical selective, circuit components (including bandpass and band-
limits, effect of delays, and performance enhancements,” by stop filters, and narrowband tunable antennas) applicable to
D. Samardzija et al., considers optimum linear precoders for software-defined radio front ends. In this perspective, the au-
multiantenna, multiuser systems. Optimality is considered thors discuss the filtering requirements in the SDR context
in terms of maximizing the sum rate capacity subject to an and advocate the use of intelligent, adaptive control to pro-
average transmitter power constraint. Performance limits of vide environment-aware frequency discrimination.
the proposed schemes under channel prediction and delayed The ninth paper “Flexible analog front ends of recon-
feedback are presented. figurable radios based on sampling and reconstruction with
The fourth paper “Flexible MIMO transmission scheme internal filtering,” by Y. Poberezhskiy and G. Poberezhskiy,
for 4G wireless systems with multiple antennas,” by F. Hor- pursues several ways to overcome the challenges of practi-
lin et al., presents a generic transmission scheme that allows cal realization and implementation of novel sampling and
to instantiate combinations of OFDM and cyclic-prefixed reconstruction techniques with internal filtering. In this per-
single-carrier modulation schemes with DS-CDMA. Addi- spective, the impact of these novel techniques on the analog
tionally, space-division multiplexing (SDM) and orthogonal front-end architectures and capabilities of software-defined
space-time block coding (STBC) have been integrated in the radios is discussed.
generic transmission scheme. For each resulting mode, the The tenth paper “A reconfigurable spiral antenna for
optimal linear MMSE multiuser receiver has been derived. A adaptive MIMO systems,” by B. Cetiner et al., studies the de-
mode selection strategy has also been proposed that trades sign of spiral antennas that are reconfigurable in the sense
off efficiently the communication performance in a typical that they can alter antenna characteristics through structural
suburban dynamic outdoor environment with the complex- change. In their work, the authors propose a reconfigurable
ity and PAPR at the mobile terminal. spiral antenna architecture based on RF-MEMS technology.
The fifth paper “Reconfigurable signal processing and The presented technology allows monolithic integration of
hardware architecture for broadband wireless communica- RF-MEMS with antenna structures on any microwave lami-
tions,” by Y.-C. Liang et al., proposes a flexible baseband nate substrate, with the capability to change the impedance
transceiver, which can be reconfigured to any type of cyclic- and radiation characteristics of the antenna. As a reference
prefix-based communication scheme. In addition, the au- model, the design, fabrication, and characterization of con-
thors introduce a corresponding reconfigurable hardware ventional single-arm rectangular spiral antennas radiating
architecture, and identify the common blocks that can be circularly polarized fields along their axes are presented in
reused across the different communication schemes. Finally, the paper.
they recognize that the major challenge is to have an efficient
system configuration and management function that will ini-
2.3. Flexible MAC and higher-layer protocols
tiate and control the reconfiguration based on user require-
ments and channel conditions. The eleventh paper “Multimode communication protocols
The sixth paper “Modular software-defined radio,” by A.- enabling reconfigurable radios,” by L. Berlemann et al., pro-
R. Rhiemeier, proposes a model of signal processing software poses a generic protocol stack, comprising common pro-
including irregular, connected, directed, acyclic graphs with tocol functionality for reconfigurable wireless communica-
random node weights and random edges. Several approaches tion systems. More specifically, the proposed generic proto-
for mapping such software to a given hardware are discussed. col stack contains parameterizable modules of basic proto-
Taking into account previous findings as well as new re- col functions that reside in the data link layer and the net-
sults from system simulations presented, the paper concludes work layer of the open systems interconnect (OSI) model.
Editorial 273
It is demonstrated that the presented parameterizable mod- Research Group of DESICS. He is investigating the baseband sig-
ules can be regarded as a toolbox for the timely and cost- nal processing algorithms and architectures for future wireless
efficient development of future communication protocols. communication systems, like third-generation (3G) and fourth-
The twelfth paper “Towards a fraud-prevention frame- generation (4G) cellular networks, and wireless local area net-
works (WLANs). His main research interests are in modulation
work for software defined radio devices,” by A. Brawerman
theory, multiple access schemes, channel estimation and equaliza-
and J. Copeland, considers a framework for security en- tion, smart antenna, and MIMO techniques. He is a Member of the
hancement in mobile SDR devices through the introduction ProRISC Technical Program Committee and the IEEE Benelux Sec-
of new hardware units and protocols. The presented frame- tion on Communications and Vehicular Technology (CVT). He is
work offers enhanced security by incorporating features like a Member of the Executive Board and Project Leader of the Flexible
monitoring against malicious attacks and viruses, authen- Radio project of the Network of Excellence in Wireless COMmuni-
tication, critical information-protection, and anticloning. cations (NEWCOM), established under the sixth framework of the
Proofs and experimental results are also given to validate the European Commission.
presented fraud-prevention framework.
Ahmet Kondoz after receiving his Ph.D. de-
gree in 1987 from the University of Surrey,
ACKNOWLEDGMENTS he became a lecturer in 1988, reader in 1995,
and in 1996 he was promoted to Professor
Many people deserve our gratitude for helping us to put to- in multimedia communication systems. He
gether this special issue. First of all, we wish to express our took part in the formation of the Centre for
gratitude to the Editor-in-Chief, Phil Regalia, for giving us Communication Systems Research (CCSR)
the opportunity and the support to realize this special is- and led the multimedia communication ac-
sue within the context of the IST FP6 Network of Excellence tivities within CCSR. He is the Founding
in Wireless COMmunications (NEWCOM). Naturally, we Director of I-Lab, the new multidisciplinary
would like to thank the authors of the regular papers for their media lab. Professor Kondoz’s current research interests are in
valuable and timely contributions. We are also grateful to the the areas of digital signal, image/video, speech/audio processing
authors of the three invited papers: Friedrich Jondral, An- and coding, wireless multimedia communications, error resilient
media transmission, immersive/virtual/augmented environments,
dreas Polydoros and his coauthors, and Narayan Mandayam
and the related human factors issues including human computer
and his coauthors. Fianlly, our appreciation goes to the many interaction/interface. He has published more than 250 journal and
obliging reviewers, without them our decision making would conference papers, two books, and 7 patents. Professor Kondoz has
have been impossible. been involved in ETSI, ITU, INMARSAT, and NATO standardiza-
tions of low-bit-rate speech communication activities. He is the
Frederik Petré Managing Director of MulSys Limited, a UniS spin-out company
Ahmet Kondoz marketing world’s first secure voice product over the GSM voice
Stefan Kaiser channel which has been pioneered by Professor Kondoz’s team in
Ashish Pandharipande the I-Lab. Professor Kondoz has been awarded several prizes, the
most significant of which are the Royal Television Societies’ Com-
munications Innovation Award and the IEE Benefactors Premium
Award.
Frederik Petré was born in Tienen, Bel-
gium, on December 12, 1974. He received Stefan Kaiser received the Dipl.-Ing. and
the E.E. degree and the Ph.D. degree in Ph.D. degrees in electrical engineering from
applied sciences from the Katholieke Uni- the University of Kaiserslautern, Germany,
versiteit Leuven (KULeuven), Leuven, Bel- in 1993 and 1998, respectively. From 1993
gium, in July 1997 and December 2003, to 2005, he was with the Institute of Com-
respectively. In September 1997, he joined munications and Navigation of the Ger-
the Design Technology for Integrated In- man Aerospace Center (DLR), Oberpfaf-
formation and Communication Systems fenhofen, Germany, where he was Head of
(DESICS) Division at the Interuniversity the Mobile Radio Transmission Group. He
Micro-Electronics Center (IMEC), Leuven, Belgium. Within the worked on multicarrier communications
Digital Broadband Terminals (DBATE) Group of DESICS, he (OFDM and MC-CDMA) for future wireless systems and digital
first performed predoctoral research on wireline transceiver de- terrestrial video broadcasting (DVB-T). In 1998, he was a Visiting
sign for twisted pair, coaxial cable, and power-line communi- Researcher at the Telecommunications Research Laboratories (TR-
cations. During the fall of 1998, he visited the Information Labs), Edmonton, Canada, working in the area of wireless com-
Systems Laboratory (ISL), Stanford University, California, USA, munications. Since 2005, he has been with DoCoMo Communica-
working on OFDM-based power-line communications. In Jan- tions Laboratories Europe GmbH, Munich, Germany, where he is
uary 1999, he joined the Wireless Systems (WISE) Group of Head of the Wireless Solution Laboratory. His research interests in-
DESICS as a Ph.D. researcher, funded by the Institute for Sci- clude among others multicarrier communications, multiple-access
entific and Technological Research in Flanders (IWT). Since Jan- schemes, and space-time processing for future mobile radio appli-
uary 2004, he has been a Senior Scientist within the Wireless cations (B3G, 4G).
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 staff at Samsung
Advanced Institute of Technology, Korea. His research interests are
in the areas of multicarrier (OFDM) and MIMO wireless commu-
nications, reconfigurable and cognitive wireless systems, multirate
signal processing, and signal processing techniques in communica-
tions.
c 2005 Friedrich K. Jondral
Software-Defined Radio—Basics and Evolution

to Cognitive Radio
Friedrich K. Jondral
Universität Karlsruhe (TH), Institut für 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-defined or reconfigurable radio systems. The need for software-
defined radios is underlined and the most important notions used for such reconfigurable transceivers are thoroughly defined.
The role of standards in radio development is emphasized and the usage of transmission mode parameters in the construction
of software-defined radios is described. The software communications architecture is introduced as an example for a framework
that allows an object-oriented development of software-defined radios. Cognitive radios are introduced as the next step in radio
systems’ evolution. The need for cognitive radios is exemplified by a comparison of present and advanced spectrum management
strategies.
Keywords and phrases: software-defined radio, reconfigurable transceiver, mobile communication standards, cognitive radio,
advanced spectrum management.
1. INTRODUCTION concerning the relation between SRs and SDRs is necessary at

Reconfigurability in radio development is not such a new this point: it is often argued that an SDR is a presently realiz-
technique as one might think. Already during the 1980s re- able version of an SR since state-of-the-art analog-to-digital
configurable receivers were developed for radio intelligence (A/D) converters that can be employed in SRs are not avail-
in the short wave range. These receivers included interesting able today. This argument, although it is correct, may lead to
features like automatic recognition of the modulation mode the completely wrong conclusion that an SR which directly
of a received signal or bit stream analysis. Reconfigurability digitizes the antenna output should be a major goal of future
became familiar to many radio developers with the publica- developments. Fact is that the digitization of an unnecessary
tion of the special issue on software radios of the IEEE Com- huge bandwidth filled with many different signals of which
munication Magazine in April 1995. only a small part is determined for reception is neither tech-
We refer to a transceiver as a software radio (SR) if its nologically nor commercially desirable.1 However, there is no
communication functions are realized as programs running reason for a receiver to extremely oversample the desired sig-
on a suitable processor. Based on the same hardware, differ- nals while respecting extraordinary dynamic range require-
ent transmitter/receiver algorithms, which usually describe ments for the undesired in-band signals at the same time.
transmission standards, are implemented in software. An SR Furthermore, the largest portion of the generated digital in-
transceiver comprises all the layers of a communication sys- formation, which stems from all undesired in-band signals,
tem. The discussion in this paper, however, mainly concerns is filtered out in the first digital signal processing step.
the physical layer (PHY). A cognitive radio (CR) is an SDR that additionally senses
The baseband signal processing of a digital radio (DR) is its environment, tracks changes, and reacts upon its findings.
invariably implemented on a digital processor. An ideal SR A CR is an autonomous unit in a communications environ-
directly samples the antenna output. A software-defined ra- ment that frequently exchanges information with the net-
dio (SDR) is a practical version of an SR: the received signals works it is able to access as well as with other CRs. From our
are sampled after a suitable band selection filter. One remark point of view, a CR is a refined SDR while this again repre-
sents a refined DR.
This is an open access article distributed under the Creative Commons

Attribution License, which permits unrestricted use, distribution, and 1 This is not an argument against the employment of multichannel or
reproduction in any medium, provided the original work is properly cited. wideband receivers.
Radio front end The rest of this paper is organized as follows. In Section 2
Receive
To user
we take a look at the most important wireless transmis-
sion standards currently used in Europe and specify their
Radio Analog-to-digital
frequency Baseband Data main parameters. Section 3 provides an overview of design
conversion processing processing
(RF) (A/D) approaches for mobile SDR terminals, especially over PaC-
From user
Transmit
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)
Control
(parameterization) is discussed in Section 5 and the need for a modified spec-
trum management in at least some major portions of the
Figure 1: SDR transceiver. electromagnetic spectrum is underlined in Section 6. Finally,
in Section 7 we propose the development of technology cen-
tric CRs as a first step towards terminals that may sense their
According to its operational area an SDR can be environment and react upon their findings. Conclusions are
(i) a multiband system which is supporting more than one drawn in Section 8.
frequency band used by a wireless standard (e.g., GSM
900, GSM 1800, GSM 1900), 2. MOBILE COMMUNICATION STANDARDS
(ii) a multistandard system that is supporting more than Standards are used to publicly establish transmission meth-
one standard. Multistandard systems can work within ods that serve specific applications employable for mass mar-
one standard family (e.g., UTRA-FDD, UTRA-TDD kets. The presently most important mobile communication
for UMTS) or across different networks (e.g., DECT, standards used in Europe are briefly described in the follow-
GSM, UMTS, WLAN), ing paragraphs.
(iii) a multiservice system which provides different services
(e.g., telephony, data, video streaming), Personal area networks
(iv) a multichannel system that supports two or more in- Bluetooth is a short distance network connecting portable
dependent transmission and reception channels at the devices, for example, it enables links between computers,
same time. mobile phones or connectivity to the internet.
Our present discussion is on multimode systems which are Cordless phone
combinations of multiband and multistandard systems. DECT (digital enhanced cordless telecommunications) pro-
The SDR approach allows different levels of reconfigura- vides a cordless connection of handsets to the fixed telephone
tion within a transceiver. system for in-house applications. Its channel access mode is
(i) Commissioning: the configuration of the system is FDMA/TDMA and it uses TDD. The modulation mode of
done once at the time of product shipping, when the DECT is Gaussian minimum shift keying (GMSK) with a
costumer has asked for a dedicated mode (standard or bandwidth (B) time (T) product of BT = 0.5. The transmis-
band). This is not a true reconfiguration. sion is protected only by a cyclic redundancy check (CRC).
(ii) Reconfiguration with downtime: reconfiguration is only
done a few times during product lifetime, for example, Wireless local area networks
when the network infrastructure changes. The recon- Today, IEEE 802.11b installations are the most widely used
figuration will take some time, where the transceiver is in Europe. Also, IEEE 802.11a systems are in operation. If
switched off. This may include the exchange of com- IEEE.11a is to be implemented into an SDR, it should be
ponents. recognized that its modulation mode is OFDM. It should
(iii) Reconfiguration on a per call basis: reconfiguration is be pointed out here that there are major efforts towards the
a highly dynamic process that works on a per call de- development of joint UMTS/WLAN systems which use the
cision. That means no downtime is acceptable. Only SDR approach.
parts of the whole system (e.g., front-end, digital base-
band processing) can be rebooted. Cellular systems
(iv) Reconfiguration per timeslot: reconfiguration can even GSM (global system for mobile communication) is presently
be done during a call. the most successful mobile communication standard world-
wide. Channel access is done via FDMA/TDMA and GSM
Figure 1 shows an SDR transceiver that differs from a uses FDD/TDD. The modulation mode of GSM is GMSK
conventional transceiver only by the fact that it can be recon- with a bandwidth time product of BT = 0.3. Error correction
figured via a control bus supplying the processing units with coding is done by applying CRC as well as a convolutional
the parameters which describe the desired standard. Such code. GSM was originally planned to be a voice communi-
a configuration, called a parameter-controlled (PaC) SDR, cation system, but with its enhancements HSCSD, GPRS, or
guarantees that the transmission can be changed instanta- EDGE, it served more and more as a data system, too. In Eu-
neously if necessary (e.g., for interstandard handover). rope, GSM systems are operating in the 900 MHz (GSM 900)
SDR—Basics and Evolution to Cognitive Radio 277
800 900 1000 1100 1200 1300 1400 1500 1600
890
915
935
960
1600 1700 1800 1900 2000 2100 2200 2300 2400 2500
1710
1785
1805
1880
1920
1980
2010
2025
2110
2170
2483.5
5100 5200 5300 5400 5500 5600 5700 5800 5900
··· f (MHz)
5150
5350
5470
5725
GSM MSS
DECT ISM
UTRA-TDD WLAN
UTRA-FDD
Figure 2: Mobile spectrum in Europe.
as well as in the 1800 MHz (GSM 1800) bands. The North Location and navigation
American equivalent of GSM is IS-136. Also, GSM 1900 as One important feature of mobile terminals is their ability to
well as IS-95, a second-generation CDMA system, are widely determine their own location as well as to track location in-
used in the US. UMTS (universal mobile telecommunication formation. Today many location-dependent services rely on
system) is the European version of the third-generation fam- the global positioning system (GPS). Currently the European
ily of standards within IMT-2000. One of the differences with satellite location and navigation system Galileo is under de-
respect to second-generation systems is that third-generation velopment.
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- Digital broadcast
nel access mode is CDMA. CRC, convolutional codes, as well There is a possibility that digital broadcast systems may be
as turbo codes [1] are employed for error protection. The used as downstreaming media within future mobile commu-
basic data modulation is QPSK. Furthermore, it should be nication infrastructures. The main developments in Europe
mentioned that one mobile user within an UTRA-FDD cell in this area are digital audio broadcast (DAB) and digital
can occupy up to seven channels (one control and six trans- video broadcast (DVB).
port channels) simultaneously. To have a sound basis for the description of a PaC-SDR
Figure 2 gives an overview over the present spectrum al- that can be switched between different standards, the most
location for mobile communications in Europe. Besides the important parameters of selected air interfaces are summa-
spectra of the standards mentioned above, also the spectra rized in Table 1.
allocated to mobile satellite system (MSS) as well as to indus-
trial, scientific, and medical (ISM) applications are specified.
The arrows within some of the bands indicate whether uplink 3. MOBILE SDR TERMINALS
(mobile to base station) or downlink (base station to mobile)
The general structure of a PaC-SDR terminal was already
traffic is supported.
given in Figure 1. Now we are going to look into the PaC-SDR
In connection with mobile communications, some addi-
transceiver structure in a more detailed way. The main pro-
tional groups of standards have to be discussed.
cessing modules of an SDR terminal are the radio front-end,
the baseband processing, and the data processing. Since a lot
Professional mobile radio of information about baseband processing can be found in
PMR standards are developed for police, firefighters, and the literature [2, 3] and since data processing is out of the
other administrative applications. The main difference to cel- scope of this paper, we are going to focus on the front-end
lular systems is that they allow direct handheld to hand- here.
held communication. The main PMR systems in Europe are The receiver branch transforms the analog RF antenna
TETRA (recommended by ETSI) and TETRAPOL. signal into its digital complex baseband representation.
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
Direct sequence (DS)
Access mode TDMA FDMA/TDMA FDMA/TDMA
CDMA
Duplex mode TDD TDD FDD FDD
Users per carrier frequency 8 maximum 12 8 —
FH sync. to master station,
Modulation GFSK with modulation GMSK GMSK QPSK
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
625 480 (DECT P32) 156.25 2560
(slot)
Frame duration — 10 ms 4.615 ms 10 ms
Number of bursts — 24 8 15
(slots)/frame
Burst (slot) duration 0.625 ms 0.417 ms 0.577 ms 0.667 ms
Maximum cell radius 5–10 m (1 mW Tx power) 300 m 36 km (10 km) Few km
— — — User specific OVSF codes,
Spreading sequences
call specific scrambling
— — — 2k (k = 2, 3, . . . , 8), 512
Spreading factor
for downlink only
Bit (chip) pulse shaping Root-raised cosine,
Gauss (BT = 0.5) Gauss (BT = 0.5) Gauss (BT = 0.3)
filter roll-off 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)
Direct sequence spread
Access mode TDMA FDMA/TDMA FDMA
spectrum
Duplex mode FDD/TDD Half duplex — —
Users/carrier — — —
4
frequency
OFDM with subcarrier OFDM with subcarrier
Modulation Π/4-DQPSK modulation BPSK, QPSK modulation
BPSK/QPSK/16QAM/64QAM QPSK/16QAM/64QAM
Error correction code CRC, Reed-Muller, RCPC Convolutional — Reed-Solomon, convolutional
9.143 Msamples/s for an
Bit (chip) rate 36 kbps 6/9/12/18/24/36/48/54 Mbps 50 bps
8 MHz channel
Number of bits (chips) 52 modulated symbols per — 2k mode: 2048 + guard int.
510 (255 symbols)
per burst (slot) OFDM symbol 8k mode: 8192 + guard int.
Frame duration 56.67 ms Packets of several 100 µs 15 s (7500 bit) 68 OFDM symbols
Number of bursts
4 Variable 5 subframes 68
(slots) per frame
1 OFDM symbol of 3.3 µs + 2k mode: 224 µs + guard time
Burst (slot) duration 14.167 ms 30 s
0.8 µs guard time 8k mode: 896 µs + 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 filtering
shaping filter roll-off factor 0.35 possible
Net data rate Up to 28.8 kbps Up to 25 Mbps — 49.8–131.67 Mbps
Evolutionary concepts — IEEE 802.11n Galileo —
Comparable systems TETRAPOL HiperLAN/2 GLONASS DAB
z−1 z−1 ··· z−1
Sampling rate adaptation

Inphase
× ∼ A/D
RF component
RF gj × gJ+ j × g2J+ j × · · · g(L−1)J+ j ×
I/Q balancing
∼
∼

π ···
−
2 + + ··· +
Quadrature
Figure 4: Polyphase filter for sampling rate adaptation.
× ∼ A/D component
The SDR transmitter branch consists of the procedures

Parameter control
inverse to that of the receiver branch. That is, the signal to be
transmitted is generated as a complex baseband signal, from
Figure 3: SDR/CR receiver front-end.
which, for example, the real part is taken to be shifted to the
(transmission) RF.
For SDRs, reconfigurability means that the radio is able
Figure 3 shows how it works: coming from the antenna, the to process signals of different standards or even signals that
RF signal is first bandpass filtered and then amplified. Fol- are not standardized but exist in specific applications. One
lowing a two-way signal splitter, the next step is an analog method to implement reconfigurability is parameterization
mixing with the locally generated RF frequency in the in- of standards. We look at a communication standard as a set
phase (I) path and with the same frequency phase shifted by of documents that comprehensively describe all functions of
−π/2 in the quadrature (Q) path. Afterwards, the I and Q a radio system in such a way that a manufacturer can de-
components of the signal are lowpass filtered and A/D con- velop terminals or infrastructure equipment on this basis.
verted. The sampling rate of the A/D converters should be Standardization is one necessary condition to make a com-
fixed for all signals and has to be chosen in such a way that the munication system successful on the market, as exemplified
conditions of Shannon’s sampling theorem are fulfilled for by GSM. Standardization pertains to all kinds of communi-
the broadest signal to be processed. Before the sampling rate cation systems, that is, especially to personal, local, cellular,
can be adapted to the signal’s standard, the impairments of or global wireless networks. Of course, a standard has to con-
the two-branch signal processing that come from the analog tain precise descriptions of all the functions of the system.
mixers and filters as well as from the A/D converters them- Especially for a mobile system, both the air interface and the
selves have to be corrected [4]. protocol stack have to be specified. Parameterization means
The reason for the Sampling rate adaptation is that the that every standard is looked upon as one member of a family
signal processor should work at the minimum possible rate. of standards [7]. The signal processing structure of the fam-
For a given standard, this minimum sampling rate depends ily is then developed in such a way that this structure may be
on fc = 1/Tc , the symbol or chip rate, respectively. Usually a switched by parameters to realize the different standards.
sampling rate of fs = 4 fc is sufficient for the subsequent sig- When developing an SDR, one has to pay attention to the
nal processing where, after the precise synchronization, the fact that there are substantial differences between the second-
sampling rate may be reduced once more by a factor of 4. generation FDMA/TDMA standards (GSM or IS-136), the
If the fraction of the sampling rates at the adaptor’s output third-generation CDMA standards (UMTS or cdma2000),
and input is rational (or may be sufficiently close approxi- and the OFDM-modulated WLAN standards (IEEE 802.11a
mated by a rational number), the sampling rate adaptation or HiperLAN/2) (cf. Table 1). Within UMTS, spreading at
can be implemented by an increasing of the sampling rate the transmitter and despreading at the receiver have to be
followed by an interpolation lowpass filter and a decreasing realized. IFFT and FFT operations are necessary for WLAN
of the sampling rate. If the interpolation lowpass is imple- transceivers. Aside from such fundamental differences, sim-
mented by an FIR filter, the impulse response usually be- ilarities among communication standards are predominant.
comes quite long. The solution is to take the up and down For example, when looking at the signal processing chains,
sampling into account within the filter process. Since the up- we remark that the error correction codes of all the second-
sampled signal is usually generated by the insertion of zeros, generation standards are very similar: a combination of a
the processing of these zeros can be omitted within the fil- block code for the most important bits and a convolutional
ter. This leads to the polyphase structure of Figure 4. Because code for the larger part of the voice bits is applied. Channel
different input/output ratios have to be realized for differ- coding for data transmission is done by a powerful convolu-
ent standards, the number of filter coefficients that must be tional code. UTRA, as a third-generation air interface, offers
stored may become large. If necessary, a direct computation net data rates of up to 2 Mbps and guarantees BERs, of up
of the filter coefficients can be more efficient than their ad- to 10−6 for specific applications. To reach these BERs turbo
vance storage [5]. After the sampling rate adaptation, the sig- codes are employed for data transmission. Of course, within
nal is processed within the complex baseband unit (demod- an SDR all these procedures have to be integrated into a gen-
ulation and decoding). The SDR data processing within the eral encoding/decoding structure. Also a common modula-
higher protocol layers [6] is not considered in the present pa- tor/demodulator structure has to be specified. Solutions to
per. these tasks are given, for example, in [2, 3, 7].
4. THE SOFTWARE COMMUNICATIONS The SCA is the description of an open architecture with
ARCHITECTURE distributed components. It strictly separates applications
(waveforms) from the processing platform (hardware, oper-
The Joint Tactical Radio System (JTRS) represents the fu- ating system, object request broker, core framework). It seg-
ture (mobile) communications infrastructure of the US joint ments the application functions and defines common inter-
forces. Introducing JTRS stands for an essential step towards faces for the management and the employment of software
the unification of radio communication systems, the trans- components. It defines common services and makes use of
parency of services, and the exchangeability of components. APIs to support the portability of hardware and software
The development of the JTRS is accompanied and supervised components and of applications.
by the US forces’ Joint Program Office (JPO). The connections between the applications and the core
Development, production, and delivery continue to be framework within the SCA are given by the APIs. Standard-
the tasks of competing industrial communications software ized APIs are essential in assuring the portability of applica-
and hardware suppliers. An important new aspect added by tions as well as for the exchangeability of devices. APIs guar-
the JTRS set-up is that the suppliers are guided to aim for antee that application and service programs may commu-
a most perfect interchangeability of components due to the nicate with one another, independent of the operating sys-
supervision function of the JPO. The tool used by the JPO is tem and the programming language used. APIs are waveform
the software communications architecture (SCA) [8], an open specific since uniform APIs for all waveforms would be inef-
framework that prescribes the developing engineers how the ficient for implementations with bounded resources. There-
hardware or software blocks have to act together within the fore, the goal is to have a standard set of APIs for each wave-
JTRS. The communication devices emerging from this phi- form. The single APIs are essentially given by the layers of the
losophy are clearly SDRs. ISO/OSI model.
A major group of suppliers and developers of communi- (i) A PHY API supports initialization and configuration
cation software and hardware founded the SDR Forum [9] to of the system in non-real-time. In real-time it takes care of
promote their interests. The importance of the SDR Forum, the transformation of symbols (or bits) to RF in the trans-
however, reaches well beyond the application of SDRs in the mitter branch. In the receiver branch it transforms RF signals
JTRS. This is underlined by the SDR Forum membership of to symbols (bits).
European and Asian industrial and research institutions that (ii) A MAC API supports all the MAC functions of the
usually are mainly interested in the evolution of commercial ISO/OSI layer model (e.g., timeslot control in TDMA or FEC
mobile communication networks. control).
The SCA describes how waveforms are to be implemented (iii) An LLC API makes available an interface for the
onto appropriate hardware devices. A waveform is defined waveform’s link layer performance (according to the ISO/OSI
by the determination of the lower three layers (network, layer model: data link services) on component level.
data link, physical) of the ISO/OSI model. Therefore, wave- (iv) A network API makes available an interface for the
form is a synonym of standard or air interface. Based on waveform’s network performance on component level.
the waveform definition, a transmission method is com- (v) A security API serves for the integration of data secu-
pletely determined. The definition of a waveform, there- rity procedures (INFOSEC, TRANSEC).
fore, lays down the modulation, coding, access, and duplex (vi) An input/output API supports the input and output
modes as well as the protocol structure of the transmission of audio, video, or other data.
method. The security relevant SCA aspects are written down in the
The SCA defines the software structure of an SDR that SCA security supplement [8]. The SCA security functions and
may be usable within the JTRS. The underlying hardware algorithms are of course defined with respect to the military
as well as the software is described in object-oriented terms. security requirements of JTRS.
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 5. USER CENTRIC AND TECHNOLOGY CENTRIC
form. COGNITIVE RADIO PROPERTIES
The JTRS operating environment (OE) defined in the The description of CR given by Mitola and Maguire in their
SCA consists of three main components: seminal paper [10] mainly focuses on the radio knowledge
(i) a real-time operating system, representation language (RKRL). CR is looked upon as a small
(ii) a real-time request broker, part of the physical world using and providing information
(iii) the SCA core framework. over very different time scales. Equipped with various sen-
sors, a CR acquires knowledge from its environment. Em-
When developing an SCA compliant radio device the ploying software agents, it accesses data bases and contacts
supplier gets the operating system and the CORBA middle- other sources of information. In this context, CR seems to
ware from the commercial market. The core framework as become the indispensable electronic aid of its owner. Read-
well as the waveform is developed by him or he also gets it ing [10] leads to the impression that a CR must be a complex
from the market or (in future) it may be contributed by the device that helps to overcome all problems of everyday life, all
JPO. the same whether they are recognized by the CR’s owner or
not. Of course, these visions as well as the recognition cycle for f0

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 different way.
The properties of CRs may be divided into two groups:
Frequency
(i) user centric properties that comprise support func-
tions like finding 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 fu
capabilities for the partitioning or the scheduling of 0 Time
processes, information gathering, and knowledge pro-
cessing. Figure 5: FDMA/TDMA signals over the time/frequency plane,
spectrum pool.
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 difficult de- unused [12, 13]. This unused area marks the pool from
sign choices need to be made to realize the wanted technol- which frequencies can be allocated to secondary users (SUs),
ogy centric properties of a CR. Therefore, we concentrate on for example, in a hotspot. In the following we denote the
the latter in the following sections. 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 fulfilled:
6. THE NEED FOR ADVANCED SPECTRUM
MANAGEMENT (i) the PUs’ system is not disturbed by the SUs’ system,
(ii) the PUs’ system remains unchanged (i.e., all signal
Today, spectrum is regulated by governmental agencies. Spec-
processing that has to be done to avoid disturbances
trum is assigned to users or licensed to them on a long-
of the PUs communications must be implemented in
term basis normally for huge regions like countries. Doing
the SUs’ system).
this, resources are wasted, because large-frequency regions
are used very sporadically. The vision is to assign appropriate Now we assume that the transmission method within the
resources to end users only as long as they are needed for a SUs’ system is OFDM. Figure 6 gives a brief overview over an
geographically bounded region, that is, a personal, local, re- OFDM transmitter: the sequential data stream is converted
gional, or global cell. The spectrum access is then organized to a parallel stream, the vectors of which are interpreted as
by the network, that is, by the users. First examples for self- signals in the frequency domain. By applying an inverse fast
regulation in mobile radio communications are to be found Fourier transform (IFFT), these data are transformed into
in the ISM (2400–2483.5 MHz) and in the WLAN (5150– the time domain and sent over the air on a set of orthogonal
5350 MHz and 5470–5725 MHz) bands. carriers with separation ∆ f on the frequency axis. If some
Future advanced spectrum management will comprise carriers should not be used, it is necessary to transmit zeros
[12] the following. on these carriers. This is the strategy to protect the PUs’ sys-
(i) Spectrum reallocation: the reallocation of bandwidth tem from disturbances originating from the SUs’ system. In
from government or other long-standing users to new order to make the SUs’ system work, the following problems
services such as mobile communications, broadband have to be solved.
internet access, and video distribution. (i) The reliable detection of upcoming PUs’ signals
(ii) Spectrum leases: the relaxation of the technical and within an extremely short time interval. (This means that
commercial limitations on existing licensees to use the detection has to be performed with a very high detection
their spectrum for new or hybrid (e.g., satellite and probability ensuring a moderate false alarm probability.)
terrestrial) services and granting most mobile radio li- (ii) The consideration of hidden stations.
censees the right to lease their spectrum to third par- (iii) The signaling of the present transmission situation
ties. in the PUs’ system to all stations of the SUs’ system such that
(iii) Spectrum sharing: the allocation of an unprecedented these do not use the frequencies occupied by the PUs.
amount of spectrum that could be used for unlicensed The solutions to these problems have recently been found
or shared services. [13]. The keywords for these solutions are distributed detec-
tion, boosting of the detection results and combining them
If we look upon the users’ behavior in an FDMA/TDMA in the hotspot’s access point to an occupancy vector, and dis-
system over the time/frequency plane (cf. Figure 5), we tributing the occupancy vector to all mobile stations in the
may find out that a considerable part of the area remains hotspot.
Bit
sequence
1
.
..
Serial-to-parallel conversion
Cod
Parallel-to-serial conversion
m
Spectrum
IDFT (IFFT)
Control SDR Localization
1 core monitoring
. RF-
.. Cod D/A
Mod.
m
1
.
.. Cod
m Information and knowledge processing
(a)
Figure 7: Technology centric cognitive radio.
∆f
(iii) in order to track the location’s or the spectral environ-

ment’s 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-
f−3 f−2 f−1 f0 f1 f2 f3 mands with the demands of other users, most probably in
f making decisions in a competitive environment using the re-
(b)
sults of game theory [15];
(vi) it has to keep its owner informed via a highly sophis-
Figure 6: OFDM. (a) Transmitter. (b) Spectrum. ticated man-machine interface.
A first block diagram of a technology centric CR is given
in Figure 7. One of the most important decisions that have to
The central point in our present discussion is that the SUs
be made in an open access environment is whether a control
system’s transceivers have in some sense to act like CRs. They
channel is to be implemented or not. The most challenging
have to sense their spectral neighborhood for PUs’ signals
development is that of the information and knowledge pro-
and to react upon their findings.
cessing.
7. TECHNOLOGY CENTRIC COGNITIVE RADIO

8. CONCLUSIONS
In a more advanced spectrum sharing system, CRs have to
apply more advanced algorithms. If a portion of the spec- Standardization of a transmission mode is necessary to en-
trum may be accessed by any access mode, the following sure its success on the market. From standards we can learn
procedure becomes imaginable: starting from the transmis- about the main parameters of a system and, by comparing
sion demand of its user, the CR decides about the data rate, different standards, we may conclude about similarities and
the transmission mode, and therefore about the bandwidth dissimilarities within their signal processing chains. Keep-
of the transmission. Afterwards it has to find an appropri- ing this knowledge in mind, we are able to construct PaC-
ate resource for its transmission. This presumes that the CR SDRs. A far more general setup is given by the SCA which is
knows where it is (self-location), what it is able to do (self- a framework for the reconfigurablity of transceivers and for
awareness), and where the reachable base stations are. To get the portability of waveforms from one hardware platform to
more information about possible interferences it should, for another. Starting from SDRs, the next step in the evolution
example, be able to detect signals active in adjacent frequency of intelligent transmission devices leads to CRs that may be
bands and to recognize their transmission standards [14]. looked upon as a small part of the physical world using and
Summing up, a CR should have implemented the follow- providing information over very different time scales. Since
ing technologies (possibly among others): this approach seems to be very futuristic, we take a look at
(i) location sensors (e.g., GPS or Galileo); the urgent problem of efficient spectrum usage. In order to
(ii) equipment to monitor its spectral environment in an introduce advanced spectrum management procedures (e.g.,
intelligent2 way; 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-
2 Intelligent means that searching for usable frequency bands is not done opment of technology centric CRs is proposed here as a first
by just scanning the whole spectrum. step towards general CRs.
ACKNOWLEDGMENT [17] J. Mitola III and Z. Zvonar, Eds., Software Radio Technologies:
Selected Readings, John Wiley & Sons, New York, NY, USA,
The author gratefully acknowledges the influence that the 6th 2000.
European Framework’s Integrated Project End-to-End Recon- [18] J. Mitola III and W. Tuttlebee, Eds., Software Defined Radio:
figurability (E2 R) as well as the Software-Defined Radio Fo- Origins, Drivers and International Perspectives, John Wiley &
rum have on his present work. Sons, Chichester, UK, 2002.
[19] W. Tuttlebee, Ed., Software Defined Radio: Enabling Technolo-
gies, John Wiley & Sons, Chichester, UK, 2002.
REFERENCES [20] M. Dillinger, K. Madani, and N. Alonistioti, Eds., Software De-
fined Radio: Architectures, Systems and Functions, John Wiley
[1] C. Heegard and S. B. Wicker, Turbo Coding, Kluwer Academic, & Sons, Chichester, UK, 2003.
Boston, Mass, USA, 1999. [21] J. Reed, Software Radio—a Modern Approach to Radio Engi-
[2] F. K. Jondral, “Parametrization—a technique for SDR imple- neering, Prentice-Hall, Upper Saddle River, NJ, USA, 2002.
mentation,” in Software Defined Radio—Enabling Technolo- [22] H. Harada and R. Prasad, Simulation and Software Radio for
gies, W. Tuttlebee, Ed., pp. 232–256, John Wiley & Sons, Lon- Mobile Communications, Artech House, Boston, Mass, USA,
don, UK, 2002. 2002.
[3] A. Wiesler and F. K. Jondral, “A software radio for second-and [23] S. Haykin, “Cognitive radio: brain-empowered wireless com-
third-generation mobile systems,” IEEE Trans. Veh. Technol., munications,” IEEE J. Select. Areas Commun., vol. 23, no. 2,
vol. 51, no. 4, pp. 738–748, 2002. pp. 201–220, 2005.
[4] P. Rykaczewski, D. Pienkowski, R. Circa, and B. Steinke, “Sig-
nal path optimization in software defined radio systems,”
IEEE Trans. Microwave Theory Tech., vol. 53, no. 3, pp. 1056– Friedrich K. Jondral received a Diploma
1064, 2005. and a Doctoral degree in mathematics from
[5] T. Hentschel and G. Fettweis, “Sample rate conversion for the Technische Universität Braunschweig,
software radio,” IEEE Commun. Mag., vol. 38, no. 8, pp. 142– Germany, in 1975 and 1979, respectively.
150, 2000. During the winter semester 1977/78, he was
[6] M. Sieber, “Design of a generic protocol stack for an adaptive a Visiting Researcher in the Department
terminal,” in Proc. 1st Karlsruhe Workshop on Software Radios, of Mathematics, Nagoya University, Japan.
pp. 31–34, Institut für Nachrichtentechnik, Universität Karl- From 1979 to 1992, Dr. Jondral was an em-
sruhe (TH), Karlsruhe, Germany, March 2000. ployee of AEG-Telefunken (now European
[7] F. K. Jondral, R. Machauer, and A. Wiesler, Software Radio— Aeronautic Defence and Space Company
Adaptivität durch Parametrisierung, J. Schlembach Fachverlag, (EADS)), Ulm, Germany, where he held various research and devel-
Weil der Stadt, Germany, 2002. opment, as well as management positions. Since 1993, Dr. Jondral
[8] “Software communications architecture specification, has been Full Professor and Head of the Institut für Nachrichten-
jtrs-5000sca v3.0,” Joint Tactical Radio System (JTRS) technik at the Universität Karlsruhe (TH), Germany. There, from
Joint Program Office, August 2004, available online on 2000 to 2002, he served as the Dean of the Department of Electri-
http://jtrs.army.mil. cal Engineering and Information Technology. During the summer
[9] “Software Defined Radio Forum,” available online on semester of 2004, Dr. Jondral was a Visiting Faculty in the Mobile
http://www.sdrforum.org. and Portable Radio Research Group of Virginia Tech, Blacksburg,
[10] J. Mitola III and G. Q. Maguire, “Cognitive radio: making Va. His current research interests are in the fields of ultra-wide-
software radios more personal,” IEEE Pers. Commun., vol. 6, band communications, software-defined and cognitive radio, sig-
no. 4, pp. 13–18, 1999.
nal analysis, pattern recognition, network capacity optimization,
[11] J. Mitola III, “Cognitive radio: an integrated agent architec- and dynamic channel allocation. Dr. Jondral is a Senior Member
ture for software defined radio,” Ph.D. dissertation, Computer
of the IEEE; he currently serves as an Associate Editor of the IEEE
Communication System Laboratory, Department of Telein-
Communications Letters and as a Member of the Software-Defined
formatics, Royal Institute of Technology (KTH), Stockholm,
Sweden, May 2000. Radio Forum’s Board of Directors.
[12] G. Staple and K. Werbach, “The end of spectrum scarcity,”
IEEE Spectr., vol. 41, no. 3, pp. 48–52, 2004.
[13] T. A. Weiss and F. K. Jondral, “Spectrum pooling: an inno-
vative strategy for the enhancement of spectrum efficiency,”
IEEE Commun. Mag., vol. 42, no. 3, pp. 8–14, 2004.
[14] M. Öner 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-
fined Radio Forum Technical Conference and Product Exhibi-
tion (SDR ’02), vol. 2, pp. NP–3–02, 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.
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
Email: azalonis@phys.uoa.gr
Nikos Dimitriou
Email: nikodim@phys.uoa.gr
Konstantinos Nikitopoulos
Email: cnikit@cc.uoa.gr
Andreas Polydoros
Email: polydoros@phys.uoa.gr
Received 16 March 2005; Revised 19 April 2005
The increasing need for multimodal terminals that adjust their configuration on the fly in order to meet the required quality of
service (QoS), under various channel/system scenarios, creates the need for flexible architectures that are capable of performing
such actions. The paper focuses on the concept of flexible/reconfigurable radio systems and especially on the elements of flexibility
residing in the PHYsical layer (PHY). It introduces the various ways in which a reconfigurable transceiver can be used to provide
multistandard capabilities, channel adaptivity, and user/service personalization. It describes specific tools developed within two
IST projects aiming at such flexible transceiver architectures. Finally, a specific example of a mode-selection algorithmic architec-
ture is presented which incorporates all the proposed tools and, therefore, illustrates a baseband flexibility mechanism.
Keywords and phrases: flexible radio, reconfigurable transceivers, adaptivity, MIMO, OFDM.
1. INTRODUCTION will be based on much higher bit rates than those pro-
vided by GSM, GPRS, and UMTS. The new services (video
The emergence of speech-based mobile communications
streaming, video broadcasting, high-speed Internet, etc.)
in the mid 80s and their exponential growth during the
will demand much higher bit rates/bandwidths and will
90s have paved the way for the rapid development of new
have strict QoS requirements, such as the received BER
wireless standards, capable of delivering much more ad-
and the end-to-end delay. The new and emerging stan-
vanced services to the customer. These services are and
dards (WiFi, WiMax, DVB-T, S-DMB, IEEE 802.20) will
have to compete with the ones based on wired commu-
This is an open access article distributed under the Creative Commons nications and overcome the barriers posed by the wireless
Attribution License, which permits unrestricted use, distribution, and medium to provide seamless coverage and uninterrupted
reproduction in any medium, provided the original work is properly cited. communication.
Flexible Radio Framework for Optimized Multimodal Operation 285
Another issue that is emerging pertains to the equipment change implies a circuit-design change, not just a numeric
that will be required to handle the plethora of the new stan- parameter change. Furthermore, the collection of adaptive
dards. It will be highly unlikely that the user will have avail- and reconfigurable transmitted-signal changes in response to
able a separate terminal for each of the introduced standards. some channel-state-information feedback may be termed dy-
There will be the case that the use of a specific standard will namic signal design (DSD). Clearly, certain potential changes
be dictated by factors such as the user location (inside build- may fall in a grey area between definitions.2
ings, in a busy district, or in a suburb), the user speed (pedes- A primitive example of flexibility is the multiband oper-
trian, driving, in a high-speed train), and the required quality ation of current mobile terminals, although this kind of flex-
(delay sensitivity, frame error rate, etc.). There might also be ibility driven by the operator is not of great research interest
cases in which it would be preferred that a service was de- from the physical-layer point of view. A more sophisticated
livered using a number of different standards (e.g., WiFi for version of such a flexible transceiver would be the one that
video, UMTS for voice), based on some criteria related to the has the intelligence to autonomously identify the incumbent
terminal capabilities (say, power consumption) and the net- system configuration and also has the further ability to ad-
work capacity constraints. Therefore, the user equipment has just its circumstances and select its appropriate mode of op-
to follow the rapid development of new wireless standards by eration accordingly. Software radio, for example, is meant to
providing enough flexibility and agility to be easily upgrade- exploit reconfigurability and modularity to achieve flexibil-
able (with perhaps the modification/addition of specific soft- ity. Other approaches may encompass other dimensions of
ware code but no other intervention in hardware). flexibility, such as adaptivity in radio resource management
We note that flexibility in the terminal concerns both the techniques.
analog/front-end (RF/IF) as well as digital (baseband) parts.
The paper will focus on the issues pertaining to the base-
3. FLEXIBILITY SCENARIOS
band flexibility and will discuss its interactions with the pro-
cedures taking place in the upper layers. In response to the demand for increasingly flexible radio
systems from industry (operators, service providers, equip-
ment manufacturers, chip manufacturers, system integra-
2. DEFINITIONS OF RADIO FLEXIBILITY tors, etc.), government (military communication and signal-
The notion of flexibility in a radio context may be defined intelligence systems), as well as various user demands, the
as an umbrella concept, encompassing a set of nonoverlap- field has grown rapidly over the last twenty years or so (per-
ping (in a conceptual sense) postulates or properties (each of haps more in certain quarters), and has intrigued and acti-
which must be defined individually and clearly for the overall vated R&D Departments, academia, research centers, as well
definition to be complete) such as adaptivity, reconfigurabil- as funding agencies. It is now a rapidly growing field of in-
ity, modularity, scalability, and so on. The presence of any quiry, development, prototyping, and even fielding. Because
subset of such features would suffice to attribute the quali- of the enormity of the subject matter, it is hard to draw solid
fying term flexible to any particular radio system [1]. These boundaries that exclusively envelop the scientific topic, but
features are termed “nonoverlapping” in the sense that the it is clear that such terms as SR, SDR, reconfigurable radio,
occurrence of any particular one does not predicate or force cognitive/intelligent/smart radio, and so on are at the cen-
the occurrence of any other. For example, an adaptive sys- ter of this activity. Similar arguments would include work
tem may or may not be reconfigurable, and so on. Additional on flexible air-interface waveforms and/or generalized (and
concepts can be also added, such as “ease of use” or “seam- properly parameterized) descriptions and receptions thereof.
lessly operating from the user’s standpoint,” as long as these Furthermore, an upward look (from the physical-layer “bot-
attributes can be quantified and identified in a straightfor- tom” of the communication-model pyramid) reveals an ever-
ward way, adding a new and independent dimension of flex- expanding role of research on networks that include recon-
ibility. Reconfigurability, for instance, which is a popular di- figurable topologies, flexible medium-access mechanisms,
mension of flexibility, can be defined as the ability to rear- interlayer optimization issues, agile spectrum allocation [4],
range various modules at a structural or architectural level and so on. In a sense, ad hoc radio networks fit the concept,
by means of a nonquantifiable1 change in its configuration. as they do not require any rigid or fixed infrastructure. Simi-
Adaptivity, on the other hand, can be defined as the radio sys- larly, looking “down” at the platform/circuit level [5], we see
tem response to changes by properly altering the numerical intense activity on flexible and malleable platforms and de-
value of a set of parameters [2, 3]. Thus, adaptive transmitted signs that are best suited for accommodating such flexibility.
(Tx) power or adaptive bit loading in OFDM naturally fall in In other words, every component of the telecommunication
the latter category, whereas dynamically switching between,
say, a turbo-coded and a convolutional-coded system in re-
2 This terminology is to a certain degree arbitrary and not universally
sponse to some stimulus (or information) seems to fit better
agreed upon; for instance, some authors call a radio system “reconfigurable”
the code-reconfigurability label, simply because that type of 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
1 “Nonquantifiable” here means that it cannot be represented by a nu- coefficient values change slowly as a function of the observation), and the
merical change in a parametric set. definition proposed here conforms to that understanding.
and radio universe can be seen as currently participating in scenario could involve the case of peer-to-peer communica-
the radio-flexibility R&D work, making the field exciting as tion whereby two flexible terminals could have the advan-
well as difficult to describe completely. tage of reconfiguring to a specific PHY (according to condi-
Among the many factors that seem to motivate the tions, optimization criteria) and establish a peer-to-peer ad
field, the most obvious seems to be the need for multistan- hoc connection.
dard, multimode operation, in view of the extreme pro- The aforementioned scenarios of flexibility point to the
liferation of different, mutually incompatible radio stan- fact that the elements of wireless communications equip-
dards around the globe (witness the “analog-to-digital-to- ment (on board both future terminals and base station sites)
wideband-to-multicarrier” evolution of air interfaces in the will have to fulfill much more complicated requirements than
various cellular-system generations). The obvious desire for the current ones, both in terms of multistandard capabilities
having a single-end device handling this multitude in a com- as well as in terms of intelligence features to control those
patible way is then at the root of the push for flexibility. This capabilities. For example, a flexible terminal on either of the
would incorporate the desire for “legacy-proof ” functional- aforementioned scenarios must be able to sense its environ-
ity, that is, the ability to handle existing systems in a single ment and location and then alter its transmission and recep-
unified terminal (or single infrastructure access point), re- tion parameters (frequency band, power, frequency, modula-
gardless of whether this radio system is equipped with all the tion, and other parameters) so as to dynamically adapt to the
related information prestored in memory or whether this is chosen standard/mode. This could in theory allow a multidi-
software-downloaded to a generically architected terminal; mensional reuse of spectrum in space, frequency, and time,
see [6] for details. In a similar manner, “future-proof ” sys- overcoming the various spectrum usage limitations that have
tems would employ flexibility in order to accommodate yet- slowed broadband wireless development and thus lead to one
unknown systems and standards with a relative ease (say, by a vision of cognitive radio [7], according to which radio nodes
mere resetting of the values of a known set of parameters), al- become radio-domain-aware intelligent agents that define
though this is obviously a harder goal to achieve that legacy- optimum ways to provide the required QoS to the user.
proofness. Similarly, economies of scale dictate that radio It is obvious that the advantageous operation of a truly
transceivers employ reusable modules to the degree possible flexible baseband/RF/IF platform will eventually include the
(hence the modularity feature). Of course, truly optimized use of sophisticated MAC and RRM functionalities. These
designs for specific needs and circumstances, lead to “point will have to regulate the admission of new users in the system,
solutions,” so that flexibility of the modular and/or generic the allocation of a mode/standard to each, the conditions of
waveform-design sort may imply some performance loss. In a vertical handover (from one standard to another), and the
other words, the benefit of flexibility may come at some cost, scheduling mechanisms for packet-based services. The cri-
but hopefully the tradeoff is still favorable to flexible designs. teria for assigning resources from a specific mode to a user
There are many possible ways to exploit the wide use of a will depend on various parameters related to the wireless
single flexible reconfigurable baseband transceiver, either on channel (path loss, shadowing, fast fading) and to the spe-
the user side or on the network side. One scenario could be cific requirements imposed by the terminal capabilities (min-
the idea of location-based reconfiguration for either multi- imization of power consumption and transmitted power),
service ability or seamless roaming. A flexible user terminal the generated interference, the user mobility, and the service
can be capable of reconfiguring itself to whichever standard requirements. That cross-layer interaction will lead to the ul-
prevails (if there are more than one that can be received) or timate goal of increasing the multiuser capacity and coverage
exists (if it is the only one) at each point in space and time, while the power requirements of all flexible terminals will be
either to be able to receive the ever-available (but possibly kept to a minimum required level.
different) service or to receive seamlessly the same service.
Additionally, the network side can make use of the future-
proof reconfiguration capabilities of its flexible base stations 4. FLEXIBLE TRANSCEIVER ARCHITECTURE
for “soft” infrastructure upgrading. Each base station can be AT THE PHY-DYNAMIC SIGNAL DESIGN
easily upgradeable to each current and future standard. An-
4.1. Transmission schemes and techniques
other interesting scenario involves the combined reception
of the same service via more than one standard in the same Research exploration of the next generation of wireless sys-
terminal. This can be envisaged either in terms of “standard tems involves the further development of technologies like
selection diversity,” according to which a flexible terminal OFDM, CDMA, MC-CDMA, and others, along with the use
will be able to download the same service via different air- of multiple antennas at the transmitter and the receiver. Each
interface standards and always sequentially (in time) select of these techniques has its special benefits in a specific envi-
the optimum signal (to be processed through the same flex- ronment: for example, OFDM is used successfully in WLAN
ible baseband chain) or, in terms of service segmentation systems (IEEE 802.11a), whereas CDMA is used successfully
and standard multiplexing, meaning that a flexible termi- in cellular 2G (IS-95) and 3G (UMTS) systems. The selection
nal will be able to collect frames belonging to the same ser- of a particular one relies on the operational environment of
vice via different standards, thus achieving throughput maxi- each particular system. In OFDM, the available signal band-
mization for that service, or receive different services (via dif- width is split into a large number of subcarriers, orthog-
ferent standards) simultaneously. Finally, another flexibility onal to each other, allowing spectral overlapping without
Tx1
Outer Inner Inner Outer
code code Rx1 . decoder decoder
Tx2 .
.
RxMr
TxMt
MIMO channel
SISO channel
CSI
Figure 1: MIMO code design procedure.
interference. The transmission is divided into parallel sub- The difference between adaptive modulation and cod-
channels whose bandwidth is narrow enough to make them ing (AMC) and dynamic signal design (DSD) is that AMC
effectively frequency flat. A cyclic prefix is used to combat ISI, is a design approach with a main focus on developing algo-
in order to avoid (or simplify) the equalizer [8]. rithms for numerical parameter changes (constellation size,
The combination of OFDM and CDMA, known as Tx power, coding parameters), based on appropriate feed-
MC-CDMA [9], has gained attention as a powerful trans- back information, in order to approach the capacity of the
mission technique. The two most frequently investigated underlying channel. The type of channel code in AMC is pre-
types are multicarrier CDMA (MC-CDMA) which employs determined for various reasons, such as known performance
frequency-domain spreading and multicarrier DS-CDMA of a given code in a given channel, compatibility with a given
(MC-DS-CDMA) which uses time-domain spreading of the protocol, fixed system complexity, and so on. Due to the va-
individual subcarrier signals [9, 10]. As discussed in [9], riety of channel models, system architectures, and standards,
MC-CDMA using DS spread subcarrier signals can be fur- there is a large number of AMC point solutions that will suc-
ther divided into multitone DS-CDMA, orthogonal MC-DS- ceed in the aforementioned capacity goal.
CDMA, and MC-DS-CDMA using no subcarrier overlap- In a typical communication system design, the algorith-
ping. In [11, 12], it is shown that the above three types of mic choice of most important functional blocks of the PHY
MC-DS-CDMA schemes with appropriate frequency spacing layer is made once at design time, based on a predetermined
between two adjacent subcarriers can be unified in the family and restricted set of channel/system scenarios. For example,
of generalized MC-DS-CDMA schemes. the channel waveform is selected based on the channel (fast
Multiple antennas with transmit and receive diversity fading, frequency selective) and the system characteristics
techniques have been introduced to improve communication (multi/single-user, MIMO). On the other hand, truly flexi-
reliability via the diversity gain [13]. Coding gain can also ble transceivers should not be restricted to one specific sce-
be achieved by appropriately designing the transmitted sig- nario of operation, so that the choice of channel waveform,
nals, resulting in the introduction of space-time codes (STC). for instance, must be broad enough to adapt either para-
Combined schemes have already been proposed in the lit- metrically or structurally to different channel/system condi-
erature. MIMO-OFDM has gained a lot of attention in re- tions. One good example of such a flexible waveform would
cent years and intensive research has already been performed. be fully parametric MC-CDMA, which can adjust its spread-
Generalized MC-DS-CDMA with both time- and frequency- ing factor, the number of subcarriers, the constellation size,
domain spreading is proposed in [11, 12] and efforts on and so on. Similarly, MIMO systems that are able to change
MIMO MC-CDMA can be found in [14, 15, 16, 17, 18]. the number of active antennas or the STC, on top of a flexi-
ble modulation method like MC-CDMA, can provide a large
4.2. Dynamic signal design number of degrees of freedom to code designers.
Flexible systems do not just incorporate all possible point so- With respect to the latter point, we note that STC de-
lutions for delivering high QoS under various scenarios, but sign has relied heavily on the pioneering work of Tarokh
possess the ability to make changes not only on the algorith- et al. in [19], where design principles were first established.
mic but also on the structural level in order to meet their Recent overall code design approaches divide coding into
goals. Thus, the DSD goal is to bring the classic design proce- inner and outer parts (see Figure 1), in order to produce
dure of the PHY layer into the intelligence of the transceiver easily implementable solutions [20, 21]. Inner codes are
and initiate new system architectural approaches, capable of the so-called ST codes, whereas outer codes are the clas-
creating the tools for on-the-fly reconfiguration. The mod- sic SISO channel codes. Each entity tries to exploit a dif-
ule responsible for all optimization actions is herein called ferent aspect of channel properties in order to improve the
supervisor, also known as controller and the like. overall system performance. Inner codes usually try to get
Table 1: Flexible design tools and inputs.
Physical-layer Modulation (a flexible

Space-time coding Channel coding
flexibility scheme like MC-CDMA)
Adjustable FFT size, Adjustable number of Tx/Rx Flexible FEC codes (e.g.,
spreading code length, antennas used, flexible ST turbo, convolutional, LDPC)
Tools constellation size (bit coding scheme as opposed to with adjustable coding rate,
loading), Tx power per (diversity/multiplexing/coding/SNR block size,
carrier (power loading) gain) code polynomial
Channel variation in time

Number of users
(Doppler), Rx antenna Effective channel
sharing the same BW,
Inputs correlation factor, feedback parameters (including
channel type
dealy, goodness of channel STC effects)
(indoor/outdoor)
estimation
diversity/multiplexing/SNR gain, while outer codes try to sense of Section 2) architectures for indoor, high-bit-rate
get diversity/coding gain. The best choice of an inner/outer wireless modems. OFDM was the signal modulation of
code pair relies on channel characteristics, complexity, and choice [22], along with a powerful turbo-coded scheme.
feedback-requirement (CSI) considerations. The Stingray Project targeted a Hiperman-compatible [23]
There are several forms of diversity that a system can of- MIMO-OFDM system for Fixed Wireless Access (FWA) ap-
fer, such as time, frequency, and space. The ability to change plications. It relied on a flexible architecture that exploited
the number of antennas, subcarriers, spreading factor and the channel state information (CSI) provided by a feedback
the ST code provides great control for the purpose of reach- channel from the receiver to the transmitter, driven by the
ing the diversity offered by the current working environment. needs of the supported service.
There are many STCs presented in the literature which ex- In the following sections, the key algorithmic choices
ploit one form of diversity in a given system/environment. of both projects are presented, which can be incorporated
All these point solutions must be taken into account in order in a single design able to operate in a variety of environ-
to design a system architecture that efficiently incorporates ments and system configurations. Since a flexible transceiver
most of them. must operate under starkly different channel scenarios, the
Outer channel codes must also be chosen so as to ob- transmission-mode-selection algorithm must rely solely on
tain the best possible overall system performance. In some instantaneous channel measurements and not on the aver-
cases, the diversity gain of the cascade coding can be analyti- age behavior of a specific channel model. This imposes the
cally derived, based on the properties of both coding options restriction of low channel dynamics in order to have the ben-
[20]. Even in these idealized scenarios, however, individually efit of feedback information. On both designs, a maximum
maximizing the diversity gain of both codes does not im- of one bit per carrier is allowed for feedback information,
prove performance. This means that, in order to maximize along with the mode selection number. The simplicity of this
the overall performance of the system, a careful tradeoff is feedback information makes both designs robust to channel
necessary between multiplexing gain, coding gain, and SNR estimation errors or feedback delay.
gain.
New channel estimation methods must also be developed 5.1. AMC in WIND-FLEX
in order to estimate not only the channel gain values but also The WIND-FLEX (WF) system was placed in the 17 GHz
other related inputs (see Table 1). For example, the types of band, and has been measured to experience high frequency
diversity that can be exploited by the receiver or the corre- selectivity within the 50 MHz channel widths. The result
lation factor between multiple antennas are important in- is strong performance degradation due to few subcarri-
puts for choosing the best coding option. Another input is ers experiencing deep spectral nulls. Even with a power-
the channel rate of change (Doppler), normalized to the sys- ful coding scheme such as turbo codes, performance degra-
tem bandwidth, in order to evaluate the feedback delay. In dation is unacceptable. The channel is fairly static for a
most current AMC techniques, this kind of input informa- large number of OFDM symbols, allowing for efficient de-
tion has not been employed, since the channel characteristics sign of adaptive modulation algorithms in order to deal
have not been considered as system design variables. with this performance degradation. In order to keep imple-
mentation complexity at a minimum, and also to minimize
the required channel feedback traffic, two design constraints
5. FLEXIBILITY TOOLS
have been adopted: same constellation size for all subcarri-
The paper is based on techniques developed in two IST ers, as well as same power for all within an OFDM sym-
projects, WIND-FLEX and Stingray. The main goal of bol, although both these parameters are adjustable (adap-
WIND-FLEX was the development of flexible (in the tive).
Target BER
10−1
Target throughput
(i.e., code (type,rate), Required uncoded
constellation) BER LUT
10−2
BER
Estimated channel Tx power
gains (frequency domain)
needed
Mode Tx power
Estimated noise PSD evaluation
10−3
2 4 6 8 10 12 14 16 18
Figure 2: Simplified block diagram of algorithm 1.
SNR/bit
4-QAM without bit loading Bit-loading rate-1 case

Bit-loading rate-2 case WSCE (20%) rate-2 case
Two algorithms have been proposed in order to optimize WSCE (10%) rate-1 case
the performance. The first algorithm (Figure 2) evaluates the
required Tx power for a specific code, constellation, and Figure 3: Uncoded performance for WIND-FLEX NLOS channel.
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- power-loading techniques is
quirements) takes place. This approach exhibits low com-
plexity and limited feedback information requirements. The  
relationship of the uncoded versus the coded BER perfor- 1 N

E
C = E log2 1 + SNRk  bits/carrier, (1)
mance in an OFDM system have been given in [24] for turbo N k=1
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 where the expectation operator is over the stochastic chan-
degrades system performance even when a strong outer code nel. For a system employing WSCE, the summation is over
is used. To counter, the technique of Weak Subcarrier ex- the used carriers along with appropriate transmit energy nor-
cision (WSCE) is introduced as a way to exclude a certain malization. These capacity results are based on the “qua-
number of subcarriers from transmission. The second pro- sistatic” assumption. For each burst, it is also assumed that
posed algorithm employs WSCE along with the appropriate a sufficiently large number of bits are transmitted, so that
selection of code/constellation size. This is called the “coded the standard infinite time horizon of information theory is
weak subcarrier excision” (CWSCE) method. meaningful. In Figure 4, the system average capacity (SAC)
In WIND-FLEX channel scenarios performance im- and the 1% system outage capacity (SOC) of the WF system
proved when using a fixed number of excised subcarriers. employing various WSCE scenarios are presented. Here, the
The bandwidth penalty introduced by this method was com- definitions are as follows.
pensated by the ability to use higher code rates. In Figure 3, (i) SAC (system average capacity). This is equivalent to
bit error rate (BER) simulation curves are shown for the un- the mean or ergodic capacity [27] applied to the ef-
coded performance of fixed WSCE and are compared with fective channel. It serves as an upper bound of systems
the bit loading algorithm presented in [26] for the NLOS with boundless complexity or latency that use a spe-
channel scenario. {Rate 1} and {Rate 2} are the system cific inner code.
throughputs when using 4-QAM with 10% and 20% WSCE, (ii) SOC (system outage capacity). This is the 1% outage
respectively. The BER performance without bit loading or capacity of the STC-effective channel.
WSCE is also plotted for a 4-QAM constellation. (iii) AC and OC. This is the average capacity and outage
There is a clear improvement by just using a fixed WSCE capacity of the actual sample-path channel.
scheme, and there is a marginal loss in comparison to the
nearly optimum bit-loading algorithm. Based on the average The capacity of an AWGN channel is also plotted as
SNR across the subcarriers, semianalytic computation of the an upper bound for a given SNR. At low SNR regions, the
average and outage capacity for the effective channel is possi- capacity of a system employing as high as 30% WSCE is
ble in order to evaluate a performance upper bound of a sys- higher than a system using all carriers without power load-
tem employing such WSCE plus uniform power loading. The ing. At high SNR, the capacity loss asymptotically approaches
use of an outer code helps to come close to this bound. We the bandwidth percentage loss of WSCE. The capacity using
note that the average capacity of an OFDM system without adaptive WSCE is also plotted. In some channel realizations,
7 The algorithm calculates the triplet that needs the min-

imum Tx power for a given target BER. If the mini-
6 mum required power is greater than the maximum avail-
able/allowable Tx power, it renegotiates the QoS. Transmit-
5
power adaptation is usually avoided, although it can be han-
Capacity (bits/s/Hz)
dled with the same algorithm. The triplet selection will still
4
be the one that needs the minimum Tx power. The extra
computation load is mainly due to the channel-tap sorting.
3
Proper exploitation of the channel correlation in frequency
2 (coherence bandwidth) can reduce this complexity overhead.
Instead of sorting all the channel taps, one can sort groups
1 of highly correlated taps. These groups can be restricted to
have an equal number of taps. There are many sorting algo-
0 rithms in the literature with different performance-versus-
0 2 4 6 8 10 12 14 16 18 20 complexity characteristics that can be employed, depending
Average channel SNR on implementation limitations.
0% WSCE SOC 30% WSCE SOC Simulation results using algorithm 1 for adaptive
10% WSCE SAC Optimum selection SOC transmission-power minimization are presented in Figure 6.
10% WSCE SOC AWGN The performance gain of the proposed algorithm is shown
20% WSCE SAC Optimum selection SAC for 4-QAM, the code rates 1/2 and 2/3. Performance is plot-
20% WSCE SOC 0%WSCE SAC ted for no adaptation, as well as for algorithm 1 in an NLOS
30% WSCE SAC
scenario. The performance over a flat (AWGN) channel is
also shown for comparison reasons, since it represents the
Figure 4: System average capacity and system 1% outage capacity
coded performance limit (given that these codes are designed
of different WSCE options.
to work for AWGN channels). The main simulation system
parameters are based on the WIND-FLEX platform. It uses a
in the low-to-medium SNR region, a 30% to 50% WSCE is parallel-concatenated turbo code with variable rate via three
needed. This result motivates the design of the second algo- puncture patterns (1/2, 2/3, 3/4) [28]. The recursive system-
rithm. The impact of CWSCE is the ability to choose between atic code polynomial used is (13, 15)oct . Perfect channel esti-
different code rates for the same target rate, a feature absent mation and zero phase noise are also assumed.
from the first algorithm. Assume an ordering of the different In addition to the transmission power gain, the adaptive
pairs {code rate-constellation size} based on the SNR neces- schemes practically guarantee the desired QoS for every chan-
sary to achieve a certain BER performance. It is obvious that nel realization. Note that in the absence of adaptation, users
this ordering also applies to the throughput of each pair (a experiencing “bad” channel conditions will never get the re-
system will not include pairs that need more power to pro- quested QoS, whereas users with a “good” channel would
vide lower throughput). For each of these pairs, the fixed per- correspondingly end up spending too much power versus
centage of excised carriers is computed so that they all pro- what would be needed for the requested QoS. By adopting
vide the same final (target) throughput. these algorithms, one computes (for every channel realiza-
The block diagram of CWSCE algorithm is given in tion) the exact needed power for the requested QoS, and thus
Figure 5. The respective definitions are as follows: can either transmit with minimum power or negotiate for a
lower QoS when channel conditions do not allow transmis-
(i) xi , i = 1, . . . ,l, is one of the system-supported constel- sion. An average 2 dB additional gain is achieved by using the
lations; second algorithm versus the first one.
(ii) yi , i = 1, . . . , M, is one of the supported outer chan-
nel codes. These can be totally different codes like 5.2. Adaptive STC in Stingray
turbo, convolutional, LDPC, or the codes resulting As mentioned, Stingray is a Hiperman-compatible 2 ×
from puncturing one mother code, or both; 2 MIMO-OFDM adaptive system. The adjustment rate,
(iii) zi , i = 1, . . . ,n, are the resulting WSCE percentages for namely, the rate at which the system is allowed to change the
the n competitive triplets; Tx parameters, is chosen to be once per frame (one frame =
(iv) Pos(zi ) are the positions of the zi % of weakest gains. 178 OFDM symbols) and the adjustable sets of the Tx pa-
(v) H is the vector of the estimated channel gains in the rameters are
frequency domain;
(1) the selected Tx antenna per subcarrier, called trans-
(vi) N
0 is the estimated power spectral density of the noise. mission selection diversity (TSD),
(vii) RUBi , i = 1, . . . ,n, is the required uncoded BER for (2) the {outer code rate, QAM size} set.
constellation xi and code yi ;
(viii) PTxi , i = 1, . . . ,n, is the required Tx power for the ith The antenna selection rule in TSD is to choose, for ev-
triplet. ery carrier k, to transmit from the Tx antenna T(k) with the
List of supported Target BER

 
channel codes (x1 , y1 , z1 )
 
 . 
 .. 
  [(x1 , y1 ), . . . , (xn , yn )]
(xn , yn , zn ) Required uncoded
Competitive .. .. BER LUT
Target . .
triplet evaluation WSCE
throuhput
[Pos(z1 ), . . . , Pos(zn )]  
(x1 , RUB1 )
 
 .. 
 
 . 
List of supported
H (xn , RUBn )
constellations

H
Channel/noise Mode Tx power  
N
0 evaluation PTx1
estimator  
 .. 

 . 

PTxn
Figure 5: Simplified block diagram of algorithm 2.
Each of the three STC schemes can be treated as an ordi-

nary OFDM SISO system producing (ideally) N independent
10−2 Gaussian channels [31]. This is the effective SISO-OFDM
channel. For the Stingray system (2 × 2), the corresponding
effective SNR (ESNR) per carrier is as follows:
10−3
T(k),0 2 T(k),1 2
H + H Es
k k
ESNRk =
BER
For TSD, , (2)

10−4
N0
for Alamouti, ESNRk
0,0 2 0,1 2 1,0 2 1,1 2
10−5
H + H + H + H Es
k k k k
= , (3)
2N0
λmax Es
10−6 for beamforming, ESNRk = k , (4)
2 4 6 8 10 12 14 N0
SNR/bit
where λmax
k is the square

of the maximum

eigenvalue of the
NLOS rate 2/3 NLOS alg. 1 rate 1/2
Hk00 Hk10 i, j
NLOS rate 1/2 AWGN rate 2/3 2 × 2 channel matrix , Hk is the frequency re-
NLOS alg. 1 rate 2/3 AWGN rate 1/2 Hk01 Hk11
sponse of the channel between the Tx antenna i and Rx an-
Figure 6: Simulation results using algorithm 1: max-log map, 4 it- tenna j at subcarrier k = 0, 1, . . . , N − 1, and N0 is the one-
erations, NLOS, 4-QAM, rate = 1/2 and 2/3. 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
best performance from a maximum-ratio combining (MRC) and estimated CSI scenarios are presented. The channel es-
perspective. For the second set of parameters, the optimiza- timation procedure uses the preamble structure described in
tion procedure is to choose the set that maximizes the system [32].
throughput (bit rate), given a QoS constraint (BER). For all simulations, path delays and the power of chan-
In order to identify performance bounds, TSD is com- nel taps have been selected according to the SUI-4 model
pared with two other rate-1 STC techniques, beamforming for intermediate environment conditions [33]. The average
and Alamouti. Beamforming is the optimal solution [29] for channel SNR is employed in order to compare adaptive sys-
energy allocation in an NT ×1 system with perfect channel tems that utilize CSI. Note that this average channel SNR is
knowledge at the transmitter side, whereby the same symbol independent of the employed STC. Having normalized each
is transmitted from both antennas multiplied by an appro- Tx-Rx path to unit average energy, the channel SNR is equal
priate weight factor in order to get the maximum achiev- to one over the power of the noise component of any one of
able gain for each subcarrier. Alamouti’s STBC is a blind the receivers. Alamouti is the most sensitive scheme to esti-
technique [30], where for each OFDM symbol period two mation errors. This is expected, since the errors in all four
OFDM signals are simultaneously transmitted from the two channel taps are involved in the decoding procedure. Based
antennas. on the ESNR, a semianalytic computation of the average and
7
10−1
6
Bits/carrier
4
10−2
3
BER
10−3 1
0
0 5 10 15
Average channel SNR
0 2 4 6 8 10
Average channel SNR/bit 2 × 2 TSD SAC 2 × 2 BF SOC
2 × 2 TSD SOC 1 × 1 AC
BF-PCSI BF-ECSI 2 × 2 ALA SAC 1 × 1 OC
TSD-PCSI TSD-ECSI 2 × 2 ALA SOC 2 × 2 AC
ALA-PCSI ALA-ECSI 2 × 2 BF SAC 2 × 2 OC
Figure 7: STCs BER performance for perfect/estimated CSI Figure 8: System average capacity and system 1% outage capacity
(PCSI/ECSI) and 4-QAM constellation. of different STC options.
outage capacity for the effective channel is possible in order

to evaluate a performance upper bound of these inner codes. 10−1
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 10−2
2 × 2 and 1 × 1 systems with no channel knowledge at the
transmitter and perfect knowledge at the receiver are also
BER
10−3
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 10−4
gain offered by the 2 × 2 channel may be expected to have a
slope similar to the capacity of the real channel (AC, OC). It
10−5
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- 2 4 6 8 10 12 14 16 18 20
curs the price of either enhanced feedback requirements or ESNR
higher complexity. Comparing the three candidate schemes,
4-QAM, 1/2 16-QAM, 3/4
we conclude that beamforming is a high-complexity solution 4-QAM, 2/3 64-QAM, 1/2
with considerable feedback requirements, whereas Alamouti 4-QAM, 3/4 64-QAM, 2/3
has low complexity with no feedback requirement. TSD has 16-QAM, 1/2 64-QAM, 3/4
lower complexity than Alamouti, whereas in comparison 16-QAM, 2/3
with beamforming, it has a minimal feedback requirement.
The gain over Alamouti is approximately 1.2 dB, while the Figure 9: TSD-turbo system performance results.
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 which consists of SNR regions and code-rate/constellation
one of the main reasons why bit loading and WSCE gave size sets for all the QoS operation modes (BER) that will be
marginal performance gain. The metric for selecting the sec- supported by the system. The selected inner code is TSD and
ond set of parameters was the effective average SNR at the the outer code is the same used in the WF system. Since per-
receiver (meaning the average SNR at the demodulator affect channel and noise-power knowledge are assumed, ESNR
ter the ST decoding). The system performance simulation is in fact the real prevailing SNR. This turns out to be a
curves based on the SNR at the demodulator (Figure 9) were good performance metric, since the outer (turbo) code per-
the basis for the construction of the Tx mode table (TMT), formance is very close to that achieved on an AWGN channel
Table 2: Transmission mode table in the case of perfect channel 7

SNR estimation.
6
Thr/put 4-QAM 4-QAM 4-QAM 16-QAM
BER 1/2 2/3 3/4 1/2 5
10−3
> 3.6 > 5.6 > 6.6 > 8.6
Bits/carrier
10−4 > 4.2 > 6.4 > 7.6 > 9.2 4
10−5 > 4.7 >7 > 8.4 > 9.8

3
10−6 >5 > 7.6 > 8.9 > 10.7
Thr/put 16-QAM 16-QAM 64-QAM 64-QAM 2
BER 2/3 3/4 2/3 3/4
10−3 > 11 > 12.2 > 15.9 > 17.3 1
10−4 > 11.7 > 12.9 > 16.5 > 17.9
0
10−5 > 12.3 > 13.6 > 16.9 > 18.6 0 5 10 15
10−6 > 13.1 > 14.5 > 17.5 > 19.8 Channel SNR
SAC ST, QOS = 10−4

SOC ST, QOS = 10−5
with equivalent SNR. Ideally, an estimation process should ST, QOS = 10−3 ST, QOS = 10−6
be included for assessing system performance as a function
of the actual measured channel, which would then be the in- Figure 10: TSD-turbo system throughput (perfect CSI-SNR esti-
put to the optimization. Using this procedure in Stingray, the mation).
related SNR fluctuation resulted in marginal performance
degradation.
Based on those curves, and assuming perfect channel- 1 × 1 (SISO) case. In [35, 36] it is shown that the quality
SNR estimation at the receiver, the derived TMT is presented of the CE estimate, which is typically characterized by the
in Table 2. Variance of the estimation error (VEE), affects drastically
By use of this table, the average system throughput (ST) the performance of the ST-OFDM schemes. In [34, 35, 36]
for various BER requirements is presented in Figure 10. The it is shown that the VEE is a function of the number and
system outage capacity (1%) is a good measure of through- the position of the subcarriers used for estimation purposes,
put evaluation of the system and is also plotted in the same of the corresponding channel taps and of the pilot modu-
figure. The average capacity is also plotted, in order to show lation method (when pilot-assisted modulation methods are
the difference from the performance upper bound. adopted). Figure 11 depicts the dependence of the symbol er-
The system throughput is very close to the 1% outage ca- ror rate of an Alamouti STC OFDM system with tentative de-
pacity, but it is 5 to 7 dB away from the performance limit, cisions on the number of subcarriers assigned for estimation
depending on the BER level. Since the system is adaptive, purposes. It is clear that this system is very sensitive to the
probably the 1% outage is not a suitable performance tar- estimation error, and therefore to the selection of the corre-
get for this system. The SNR gain achieved by going from sponding “pilot” number.
one BER level to the next is about 0.8 dB. This marginal gain Additionally, the working range of the decision-directed
is expected due to the performance behavior of turbo codes approaches is mainly dictated by the mean CE and the SNR,
(very steep performance curves at BER regions of interest). which should be such that most of the received symbols are
within the bounds of correct decisions (i.e., the resulting er-
5.3. Flexible algorithms for phase noise and residual ror from the tentative decisions should be really small). This
frequency offset estimation may be difficult to ensure, especially when transmitting high-
Omnipresent nuisances such as phase noise (PHN) and order QAM constellations. An improved supervisor has to
residual frequency offsets (RFO), which are the result of a take into account the effect of the residual CE error on the
nonideal synchronization process, compromise the orthogo- overall system performance for selecting the optimal triplet,
nality between the subcarriers of the OFDM systems (both by inserting its effect into the overall calculations.
SISO and MIMO). The resulting effect is a Common Er- Two approaches can be followed for the system optimiza-
ror (CE) for all the subcarriers of the same OFDM sym- tion. When the system protocol forces a fixed number of pilot
bol plus ICI. Typical systems adopt CE compensation algo- symbols loaded on fixed subcarriers (as in Hiperman), the
rithms, while the ICI is treated as an additive, Gaussian, un- corresponding performance loss is calculated and the possi-
correlated per subcarrier noise parameter [34]. The phase- ble triplets are decided. It is noted that an enhanced super-
impairment-correction schemes developed in Stingray and visor device could decide on the use of adaptive pilot modu-
WF can be implemented either by the use of pilot symbols or lation in order to minimize estimation errors by maximizing
by decision-directed methods. They are transparent to the se- the received energy, since the pilot modulation may signif-
lection of the Space-Time coding scheme, and they are easily icantly affect the system performance. Figure 12 depicts the
adaptable to any number of Tx/Rx antennas, down to the effect of the pilot modulation method for the 2 × 2 Alamouti
100 100
10−1 10−1
SER
SER
10−2 10−2
10−3 10−3
10−4 10−4
6 7 8 9 10 11 12 13 14 15 5 6 7 8 9 10 11 12 13 14
Eb /N0 (dB) Eb /N0 (dB)
No PHN, no RFO No PHN, no RFO OGP

L0 = 256 RGP FPP
L0 = 8
Figure 11: Effect of number of subcarriers used for estimation pur- Figure 12: Effect of pilot modulation on 2 × 2 ST-OFDM system,
poses on decision-directed, 2 × 2 ST-OFDM system, Alamouti en- Alamouti encoded, loaded with 16-QAM (L0 = 8; 2 estimators).
coded, loaded with 16-QAM.
ST-OFDM system including 8 pilots, 256 subcarriers, and as- The aforementioned CWSCE and TSD methods do be-
suming independent compensation per receiver antenna for long to this category of flexible (partial) solutions. The ca-
a realization of an SUI-4 channel. Three modulation meth- pacity penalty for their use (compared to the optimal solu-
ods are considered: randomly generated pilots (RGPs), or- tions) has been shown herein to be small. Both require com-
thogonal generated pilots (OGPs), and fixed pilot pattern mon feedback information (1 bit/carrier) and can be incor-
(FPP), where the same pilots are transmitted from any Tx porated appropriately in a system able to work under a va-
antenna. Thus, the selection of the pilot modulation scheme riety of antenna configurations, when such limited feedback
is another parameter to be decided, since its affects system information is available. When feedback information is not
performance in a significant way. available, CWSCE has the appropriate modules for mode se-
On the other hand, when the system protocol allows for lection (algorithm 1) for the SISO case, while Alamouti can
a variable number of pilot symbols, the optimization proce- be the choice for the MIMO case. Both STC schemes trans-
dure becomes more complex. After a training period of some form the MIMO channel into an inner SISO one, allowing
OFDM symbols, the mean CE can be roughly estimated. for the use of AMC (mode selection) techniques designed for
Using this estimate and taking into account that the whole SISO systems. In the Stingray system, as already explained,
OFDM symbol is loaded with the same QAM constellation, the average ESNR at the demodulator is a sufficient met-
it can be decided whether a specifically chosen constellation ric for choosing the Tx mode, whereas in WIND-FLEX the
is robust to the CE, so that the decision directed methods uncoded BER is, respectively, used. Employing TMT tables
(based on tentative decisions) are reliable. For the constella- with the required uncoded BER and code-rate/constellation-
tions where the pilot-symbol use is necessary, the supervisor size sets for all the QoS operation modes in MIMO sys-
has to select appropriately the position and the number of tems will increase the complexity, but it will permit seam-
pilot symbols. less incorporation of both systems into one single common
architecture. The uncoded performance of the effective chan-
nel is thus the only metric that need be used for choosing
6. TOWARDS A FLEXIBLE ARCHITECTURE
the Tx mode and can be computed for a variety of STC op-
As already mentioned, a flexible transceiver must be tions. Furthermore, the fully parametric PHN and RFO algo-
equipped with the appropriate robust solutions for all possi- rithms mentioned above are transparent to the selection of
ble widely ranging environments/system configurations. To the ST coding scheme and can provide the appropriate in-
target the universally best possible performance translates to formation about their performance under different environ-
high complexity. A first step towards a generic flexible ar- ments/modes.
chitecture should be one that efficiently incorporates simple The overall block diagram of a proposed architecture
tools in order to deliver not necessarily the best possible, but for the mode selection algorithm is given in Figure 13. It is
an acceptable performance under disparate system/channel meant to be able to work for all systems employing one or
environments. two antennas at the Tx/Rx.
List of supported Target BER

 
channel codes (x1 , y1 , z1 )
 
 .  [(x1 , y1 ), . . . , (xn , yn )]
 .. 
WSCE   Required uncoded
(on/off) Competitive (xn , yn , zn ) WSCE .. .. BER LUT
Target triplet evaluation & PCE . .
throughput [Pos(z1 ), . . . , Pos(zn )]
 
  (x1 , RUB1 )
PN(x1 )  
   . 
 .   . 
 .   . 
 .  HEF
List of supported PN(xl ) (xn , RUBn )
constellations
HEF HEF
PHN/RFO Effective channel/ Mode Tx power  
estimator N0EF noise estimator N0EF evaluation PTx1
Operation mode WSCE/TSD/ALA  
 . 
 . 
 . 
PTxn
[VEE(z1 ), . . . , VEE(zl )]
Figure 13: Block diagram of proposed algorithm for mode selection.
The related parameters are defined as follows: of dynamic signal design for instantiating some of these at-
(i) PN(xi ), i = 1, . . . , l, is the number of needed pilots for tributes in a specific application environment. Two design
a specific PHN/RFO performance, when the operation approaches are presented (based on the WF and Stingray
mode enables variable number of pilots; projects) and the key algorithmic choices of both are pre-
sented and incorporated into one flexible design capable of

EF is the vector of the estimated effective channel
(ii) H successfully operating in a variety of environments and sys-
gains in the frequency domain (STC dependent); tem configurations. It is evident that physical-layer flexi-
(iii) PCE : pilot carrier excision (an enhancement of the bility requires not only novel system architectures but also
WSCE module which provides the pilot positions for a new algorithms that efficiently utilize existing and/or new
given number of used pilots). modulation/coding techniques that can be adjusted to var-
Here, WSCE is active only when the system is 1 × 1. On ious environment and system scenarios, in order to offer
all other Tx-Rx antenna choices, all subcarriers are assumed QoS close to that delivered by corresponding point-optimal
“on.” When only a fixed number of pilot symbols are permit- solutions.
ted (e.g., when a specific protocol is used), the PHN/RFO es-
timator provides the VEE for each constellation choice to the ACKNOWLEDGMENT
Tx power evaluation module. In a peer-to-peer communica-
tion system, where two flexible terminals could have the pos- The work presented in this paper has been supported by
sibility of reconfiguring to a specific PHY, the number of pi- STINGRAY (IST-2000-30173) and WIND-FLEX (IST-1999-
lots can be allowed to change and the optimum solution de- 10025) projects that have been partly funded by the Euro-
pends on the constellation size. The competitive-triplet eval- pean Commission.
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- REFERENCES
nel/system characteristics but also on the selected optimiza- [1] A. Polydros, J. Rautio, G. Razzano, et al., “WIND-FLEX: de-
tion criteria such as maximizing the throughput, minimizing veloping a novel testbed for exploring flexible radio concepts
the Tx power, and so on. in an indoor environment,” IEEE Commun. Mag., vol. 41,
no. 7, pp. 116–122, 2003.
[2] A. J. Goldsmith and S.-G. Chua, “Variable-rate variable-
7. CONCLUSIONS power MQAM for fading channels,” IEEE Trans. Commun.,
The scientific field of radio flexibility is growing in impor- vol. 45, no. 10, pp. 1218–1230, 1997.
tance and appeal. Although still in fairly nascent form for [3] T. Keller and L. Hanzo, “Adaptive modulation techniques
for duplex OFDM transmission,” IEEE Trans. Veh. Technol.,
commercial use, flexible radio possesses attractive features vol. 49, no. 5, pp. 1893–1906, 2000.
and attributes that require further research. The present pa- [4] F. K. Jondral, “Flexible spectrum allocation and cognitive ra-
per presents the flexibility concept, definition, and related dio,” in Proc. 11th National Symposium of Radio Science (URSI
scenarios while, in parallel, explores in some depth the tool ’05), pp. 27–32, Poznan, Poland, April 2005.
[5] G. Masera, “Flexible hardware for wireless communications: Software Defined Radio Technical Conference and Product Ex-
a case of study,” in Proc. 11th National Symposium of Radio position (SDR ’03), Orlando, Fla, USA, November 2003.
Science (URSI ’05), Poznan, Poland, April 2005. [25] K. Babionitakis, I. Dagres, K. Nakos, and D. Reisis, “A VLSI
[6] E. Buracchini, “The software radio concept,” IEEE Commun. architecture for minimizing the transmission power in OFDM
Mag., vol. 38, no. 9, pp. 138–143, 2000. transceivers,” in Proc. 10th IEEE International Conference on
[7] J. Mitola and G. Maguire, “Cognitive radio: making software Electronics, Circuits and Systems (ICECS ’03), vol. 1, pp. 308–
radios more personal,” IEEE Personal Communications Maga- 311, Sharjah, United Arab Emirates, December 2003.
zine, vol. 6, no. 4, pp. 13–18, 1999. [26] P. Chow, J. M. Cioffi, and J. A. C. Bingham, “A practical dis-
[8] R. van Nee and R. Prasad, OFDM Wireless Multimedia Com- crete multitone transceiver loading algorithm fordata trans-
munications, Artech House, Boston, Mass, USA, 2000. mission over spectrally shaped channels,” IEEE Trans. Com-
[9] R. Prasad and S. Hara, “Overview of multicarrier CDMA,” mun., vol. 43, no. 234, pp. 773–775, 1995.
IEEE Commun. Mag., vol. 35, no. 12, pp. 126–133, 1997. [27] G. Foschini and M. Gans, “On the limits of wireless commu-
[10] L. Hanzo, M. Munster, B. J. Choi, and T. Keller, OFDM nications in a fading environment when using multiple anten-
and MC-CDMA for Broadband Multi-User Communications, nas,” Wireless Personal Communications, vol. 6, no. 3, pp. 311–
WLANs and Broadcasting, John Wiley & Sons, New York, NY, 335, 1998.
USA, 2003. [28] M. Benedix, J. Ertel, and A. Finger, “Turbo coding for an
[11] L.-L. Yang and L. Hanzo, “Performance of generalized mul- OFDM-based wireless LAN at 17 GHz,” in Proc. 7th Inter-
ticarrier DS-CDMA over Nakagami-mfading channels,” IEEE national OFDM-Workshop, pp. 137–141, Hamburg, Germany,
Trans. Commun., vol. 50, no. 6, pp. 956–966, 2002. September 2002.
[12] L.-L. Yang and L. Hanzo, “Multicarrier DS-CDMA: a multiple [29] R. Knopp and G. Caire, “Power control and beamforming
access scheme for ubiquitous broadband wireless communi- for systems with multiple transmit and receive antennas,”
cations,” IEEE Commun. Mag., vol. 41, no. 10, pp. 116–124, IEEE Transactions on Wireless Communications, vol. 1, no. 4,
2003. pp. 638–648, 2002.
[13] A. J. Paulraj, D. A. Gore, R. U. Nabar, and H. Bolcskei, “An [30] S. Alamouti, “A simple transmit diversity technique for wire-
overview of MIMO communications—a key to gigabit wireless communications,” IEEE J. Select. Areas Commun., vol. 16,
less,” Proc. IEEE, vol. 92, no. 2, pp. 198–218, 2004. no. 8, pp. 1451–1458, 1998.
[14] M. Juntti, M. Vehkapera, J. Leinonen, et al., “MIMO MC- [31] I. Dagres, A. Zalonis, and A. Polydoros, “An efficient adap-
CDMA communications for future cellular systems,” IEEE tive space time coding scheme for MIMO-OFDM systems,” in
Commun. Mag., vol. 43, no. 2, pp. 118–124, 2005. Proc. 14th IST Mobile and Wireless Telecommunications Sum-
[15] M. Vehkaperä, D. Tujkovic, Z. Li, and M. Juntti, “Layered mit, Dresden, Germany, June 2005, to appear.
space-frequency coding and receiver design for MIMO MC- [32] I. Harjula, S. Boumard, and A. M. Gallardo, “WP4-Training
CDMA,” in Proc. International Conference on Communications symbol design for Stingray demonstrator,” 2002, (document
(ICC ’04), vol. 5, pp. 3005–3009, Paris, France, June 2004. Stingray 4VTTN009d.doc in IST-2000-30173 STINGRAY).
[16] J. Tang and X. Zhang, “Link-adaptation-enhanced dynamic [33] V. Erceg, K. V. S. Hari, M. S. Smith, and D. S. Baum, “Channel
channel allocation for MIMO-OFDM wireless networks,” in Models for Fixed Wireless Applications,” 2001, IEEE 802.16.3
Proc. IEEE Wireless Communications and Networking Confer- Task Group Contributions, Doc. IEEE 802.16.3.c-01/29r4.
ence (WCNC ’05), New Orleans, La, USA, March 2005. [34] K. Nikitopoulos and A. Polydoros, “Decision-directed com-
[17] J.-H. Deng, J.-Y. Wu, and T.-S. Lee, “Space-path spreading for pensation of phase noise and residual frequency offset in a
high rate MIMO MC-CDMA systems with transmit diversity,” Space-Time OFDM receiver,” IEEE Commun. Lett., vol. 8,
in Proc. IEEE Global Telecommunications Conference (GLOBE- no. 9, pp. 573–575, 2004.
COM ’03), vol. 6, pp. 3397–3401, San Francisco, Calif, USA, [35] K. Nikitopoulos and A. Polydoros, “Phase-impairment ef-
December 2003. fects and compensation algorithms for OFDM systems,” IEEE
[18] Y. Kim, T. Kim, C. Kang, and D. Hong, “Union bound of max- Trans. Commun., vol. 53, no. 4, pp. 698–707, 2005.
imum likelihood detection in downlink MIMO MC-CDMA [36] K. Nikitopoulos and A. Polydoros, “Phase noise and resid-
systems,” in Proc. International Conference on Communica- ual frequency offset compensation in space-time OFDM sys-
tions (ICC ’04), vol. 6, pp. 3280–3283, Paris, France, June tems,” in Proc. IST Mobile and Wireless Telecommunications
2004. Summit, pp. 77–82, Aveiro, Portugal, June 2003.
[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. Ioannis Dagres holds a B.S. degree in com-
Theory, vol. 44, no. 2, pp. 744–765, 1998. puter engineering and an M.S. degree in sig-
[20] Y. Gong and K. B. Letaief, “Concatenated space-time block nal processing from the Technical Univer-
coding with trellis coded modulation in fading channels,” sity of Patras, Greece, in 1997 and 1999, re-
IEEE Transactions on Wireless Communications, vol. 1, no. 4,
spectively. He is currently working towards
pp. 580–590, 2002.
his Ph.D. degree at the same university.
[21] B. Hochwald and S. ten Brink, “Achieving near-capacity on
a multiple-antenna channel,” IEEE Trans. Commun., vol. 51, Since 1999, he has participated in several
no. 3, pp. 389–399, 2003. EU projects such as ADAMAS (ADAptive
[22] L. J. Cimini, “Analysis and simulation of a digital mobile chan- Multicarrier Access System), WIND-FLEX
nel using orthogonal frequency division multiplexing,” IEEE (Wireless INDoor FLEXible Modem Archi-
Trans. Commun., vol. 33, no. 7, pp. 665–675, 1985. tecture), and STINGRAY (Space Time CodING for Adaptive Re-
[23] ETSI, “Broadband Radio Access Networks (BRAN)— configurAble Systems) as a Research Associate. His research inter-
HIPERMAN, OFDM Physical (PHY) layer,” draft, 2002. ests include the topics of signal processing for communications
[24] I. Dagres and A. Polydoros, “Dynamic transceivers: adaptiv- and, in particular, adaptive signal design for modern communica-
ity and reconfigurability at the signal-design level,” in Proc. tion systems.
Andreas Zalonis received the B.S. degree Andreas Polydoros was born in Athens,
in physics and the M.S. degree in telecom- Greece, in 1954. He received the Diploma
munications and electronics, both from the in electrical engineering from the National
National Kapodistrian University of Athens Technical University of Athens in 1977, the
(NKUA), Greece, in 2000 and 2002, re- M.S.E.E. degree from the State University of
spectively. He is currently working towards New York at Buffalo in 1979, and the Ph.D.
his Ph.D. degree in fourth-generation dig- degree in electrical engineering from the
ital communications systems at the same University of Southern California in 1982.
university. Since 2001, he has participated He was a faculty member in the Department
in several national and European research of Electrical Engineering-Systems and the
projects in the area of wireless communications. Since 2003 he Communication Sciences Institute (CSI), the University of South-
works as a Research Associate in the Institute of Accelerating Sys- ern California, in 1982–1997, and has been a Professor since 1992.
tems and Applications (IASA), NKUA. He codirected CSI in 1991–1993. Since 1997, he has been a Pro-
fessor and Director of the Electronics Laboratory, Division of Ap-
Nikos Dimitriou holds a Diploma degree plied Physics, Department of Physics, University of Athens, Greece.
in electrical and computer engineering from His general areas of scientific interest are statistical communication
the National Technical University of Athens, theory and signal processing with applications to spread-spectrum
Greece (1996), an M.S. degree with distinc- and multicarrier systems, signal detection and classification in un-
tion in mobile and satellite communica- certain environments (he is the coinventor of per-survivor process-
tions from the University of Surrey, United ing, granted a US patent in 1995), and multiuser radio networks.
Kingdom (1997), and a Ph.D. degree from He cofounded and currently heads the Technology Advisory Board
the same University (2000). During that of Trellis Ware Technologies, and also serves on numerous scientific
period his work was focused on radio re- boards of academic nonprofit organizations in Greece and abroad.
source management issues and techniques Professor Polydoros is the recipient of a 1986 US National Science
for next-generation mobile multimedia communication systems Foundation Presidential Young Investigator Award. He became a
(UMTS, W-CDMA). He participated in projects related to traffic Fellow of the IEEE in 1995, cited “for contributions to spread-
modeling and network dimensioning for UMTS and GPRS (led spectrum systems and networks.”
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 specifically
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 Reconfig-
urAble Systems), and in the Network of Excellence in Wireless
Communications (NEWCOM). His research is focused on cross-
layer optimization techniques for next-generation flexible 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 EU re-
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).
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 reconfigurable radios have rendered feasible transmitter optimization schemes that can greatly
improve the performance of multiple-antenna multiuser systems. Reconfigurable 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 prefilter, (2) modified zero-forcing (MZF) spatial prefilter, and (3) triangularization spatial prefilter 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 that can provide high-capacity (i.e., spectral efficiency) wire-

less communications in rich scattering environments. It has
For a wide range of emerging wireless data services, the
been shown that the theoretical capacity (approximately) in-
application of multiple antennas appears to be one of the
creases linearly as the number of antennas is increased [1, 2].
most promising solutions leading to even higher data rates
With the advent of flexible and programmable radio
and/or the ability to support greater number of users.
technology, transmitter optimization techniques used in
Multiple-transmit multiple-receive antenna systems rep-
conjunction with MIMO processing can provide even greater
resent an implementation of the MIMO (multiple-input
gains in systems with multiple users. Reconfigurable ra-
multiple-output) concept in wireless communications [1]
dio platforms are particularly suitable for implementation
of transmitter optimization at the base station. Such opti-
This is an open access article distributed under the Creative Commons mization techniques have great potential to enhance perfor-
Attribution License, which permits unrestricted use, distribution, and mance on the downlink of multiuser wireless systems. From
reproduction in any medium, provided the original work is properly cited. 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 Mobile 1

Transmitter y1
have also focussed on multiple-antenna systems with mul-
tiple users include [4, 5, 6, 7, 8, 9, 10] and the references 1
therein. x1 Mobile 2
TX 2 MIMO
In this paper, we study multiple-antenna transmitter op- x2
channel y2
.. transform . ..
timization (i.e., spatial prefiltering) schemes that are based . S .
. H .
on linear preprocessing and transmit power optimization M
xN
(keeping the average transmit power conserved). Specifically, Mobile N
we consider the downlink of a wireless system with multiple yN
transmit antennas at the base station and a number of mo-
bile terminals (i.e., users) each with a single receive antenna. Figure 1: System model consisting of M transmit antennas and N
We consider the maximum achievable sum data rates in the mobile terminals.
case of (1) zero-forcing spatial prefilter, (2) modified zero-
forcing spatial prefilter, and (3) triangularization spatial pre-
filter coupled with dirty-paper coding transmission scheme
where x = [x1 , . . . , xN ]T is the transmitted vector (E[xxH ] =
[11]. We study the relationship between the above schemes
Pav IN ×N ), n is AWGN (E[nnH ] = N0 IN ×N ), H is the MIMO
as well as the impact of the number of antennas on perfor-
channel response matrix, and S is a transformation (spatial
mance.
prefiltering) performed at the transmitter. Note that the vec-
After characterizing the fundamental performance lim-
tors x and y have the same dimensionality. Further, hnm is
its, we then study the performance of the above transmitter
the nth row and mth column element of the matrix H corre-
optimization schemes with respect to delayed channel state
sponding to a channel between mobile terminal n and trans-
information (CSI). The delay in CSI may be attributed to the
mit antenna m. If not stated otherwise, we will assume that
delay in feeding back this information from the mobiles to
N ≤ M.
the base station or alternately to the delays in the ability to
Application of the spatial prefiltering results in the com-
reprogram/reconfigure the transmitter prefilter. Without ex-
posite MIMO channel G given as
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- G = HS, G ∈ C N ×N , (2)
der to alleviate this problem, we exploit correlations in the
channel by designing a linear MMSE predictor of the chan-
where gnm is the nth row and mth column element of the
nel state. We then show how the application of the MMSE
composite MIMO channel response matrix G. The signal re-
predictor can improve performance of transmitter optimiza-
ceived at the nth mobile terminal is
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-
N
ious transmitter optimization schemes including their fun- yn = gnn xn + gni xi +nn . (3)
i=1, i
=n
damental performance limits as well as the effect of delayed Desired signal for user n
CSI. In Section 4, a formal channel model capturing chan- Interference
nel correlations and a linear MMSE predictor of the channel
state which is used to overcome the effect of delayed CSI are In the above representation, the interference is the signal that
presented. is intended for mobile terminals other than terminal n. As
said earlier, the matrix S is a spatial prefilter at the transmit-
2. SYSTEM MODEL ter. It is determined based on optimization criteria that we
address in the next section and has to satisfy the constraint
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 trace SSH ≤ N (4)
terminals (each with a single receive antenna). In other words
each mobile terminal presents a MISO channel as seen from which keeps the average transmit power conserved. We rep-
the base station. resent the matrix S as
In Figure 1, xn is the information bearing signal intended
for mobile terminal n and yn is the received signal at the cor-
responding terminal (for n = 1, . . . , N). The received vector S = AP, A ∈ C M ×N , P ∈ C N ×N , (5)
y = [y1 , . . . , yN ]T is
where A is a linear transformation and P is a diagonal ma-
y = HSx + n, y ∈ CN , x ∈ CN , n ∈ CN ,
(1) trix. P is determined such that the transmit power remains
S ∈ C M ×N , H ∈ C N ×M , conserved.
3. TRANSMITTER OPTIMIZATION SCHEMES See Appendix A for a definition of the corresponding vir-
tual uplink and a proof of the above proposition.
Considering different forms of the matrix A, we study the (3) Triangularization spatial prefiltering with dirty-paper
following transmitter optimization schemes. coding (DPC) where the matrix A assumes the form
(1) Zero-forcing (ZF) spatial prefiltering scheme where A
is represented by
A = HH R−1 , (11)
−1
A = HH HHH . (6) where H = (QR)H and Q is unitary and R is upper triangular
(see [12] for details on QR factorization). In general, R−1 is a
As can be seen, for N ≤ M, the above linear transformation pseudoinverse of R. The composite MIMO channel G in (2)
is zeroing the interference between the signals dedicated to becomes G = L = HS, a lower triangular matrix. It permits
different mobile terminals, that is, HA = IN ×N . xn are as- application of dirty-paper coding designed for single-input
sumed to be circularly symmetric complex random variables single-output (SISO) systems. We refer the reader to [4, 5, 6,
having Gaussian distribution NC (0, Pav ). Consequently, the 13, 14, 15, 16] for further details on the DPC schemes.
maximum achievable data rate (capacity) for mobile termi- By applying the transformation in (11), the signal in-
nal n is tended for terminal 1 is received without interference. The
signal at terminal 2 suffers from the interference arising from
the signal dedicated to terminal 1. In general, the signal at
Pav | pnn |2
RZF
n = log2 1 + , (7) terminal n suffers from the interference arising from the sig-
N0
nals dedicated to terminals 1 to n − 1. In other words,
where pnn is the nth diagonal element of the matrix P defined y1 = g11 x1 + n1 ,
in (5). In (6) it is assumed that HHH is invertible, that is, the
y2 = g22 x2 + g21 x1 + n2 ,
rows of H are linearly independent.
(2) Modified zero-forcing (MZF) spatial prefiltering ..
.
scheme that assumes
−1
n
−1 yn = gnn xn + gni xi + nn , (12)
N0 i=1
A = HH HHH + I . (8)
Pav ..
.
In the case of the above transformation, in addition to the −1
N
knowledge of the channel H, the transmitter has to know the yN = gNN xN + gNi xi + nN .
noise variance N0 . xn are assumed to be circularly symmet- i=1
ric complex random variables having Gaussian distribution
NC (0, Pav ). The maximum achievable data rate (capacity) for Since the interference is known at the transmitter, DPC can
mobile terminal n now becomes 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
Pav |gnn |2
RMZF
n = log2 1 + N . (9)
Pav =n |gni | + N0
2 Pav |gnn |2 Pav |rnn pnn |2
i=1, i RDPC
n = log2 1 + = log2 1 + ,
N0 N0
While the transformation in (8) appears to be similar in form (13)
to an MMSE linear receiver, the important difference is that where rnn is the nth diagonal element of the matrix R defined
the transformation is performed at the transmitter. Using the in (11). Note that DPC is applied just in the case of the lin-
virtual uplink approach for transmitter beamforming (intro- ear transformation in (11), with corresponding rate given by
duced in [7, 8]), we present the following proposition. (13).
Note that trace(AAH ) = N, thereby satisfying the con-
Proposition 1. If the nth diagonal element of P is selected as straint in (4). Consequently, we can select P = IN ×N and
present the following proposition.
1
pnn = (n = 1, . . . , N), (10)
anH an Proposition 2. For high SNR (Pav N0 ) and P = IN ×N , the
achievable sum rate of the triangularization with DPC scheme
is equal to the rate of the equivalent (open loop) MIMO system.
where an is the nth column vector of the matrix A, the constraint
In other words, for Pav N0 ,
in (4) is satisfied with equality. Consequently, the achievable
downlink rate RMZFn for mobile n is identical to its correspond-

N
Pav
ing virtual uplink rate when an optimal uplink linear MMSE RDPC = log2 det IN ×N + HHH . (14)
n
receiver is applied. n=1
N0
Proof. Starting from the right-side term in (14) and with 9

HHH = RH R, for Pav N0 , 8

Average user rate (bits/symbol)

Pav H 7
log2 det IN ×N + R R
N0 6

Pav H
≈ log2 det R R 5
N0

Pav
r11 2 · · · Pav rNN 2 4
= log2
N0 N0 3

N N
Pav
rii 2 ≈ P 2
= log2 log2 1 + av rii 2
i=1
N0 i=1
N0
1

N
= RDPC
n 0
n=1 0 5 10 15 20 25
(15) SNR (dB)
which concludes the proof.
No prefiltering
TDMA-CBF
The ZF and MZF schemes should be viewed as trans- ZF
mitter beamforming techniques using conventional channel MZF
coding to approach the achievable rates [7, 8]. The trian- DPC
gularization with DPC scheme is necessarily coupled with a CL-MIMO
nonconventional coding, that is, the DPC scheme.
Figure 2: Average rate per user versus SNR (M = 3, N = 3, Rayleigh
Once the matrix A is selected, the elements of the diag- channel).
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 This solution is not practical because the terminals are nor-
mally individual entities in the network and they do not co-
trace APPH A H
≤ N. (16) operate when receiving signals on the downlink.
In Figure 2, we present average rates per user versus
The elements of the matrix P are selected such that SNR = 10 log (Pav /N0 ) for a system consisting of M = 3
T transmit antennas and N = 3 terminals. The channel is
diag(P) = p11 , . . . , pNN Rayleigh, that is, the elements of the matrix H are complex

N (17) independent and identically distributed Gaussian random
= arg max Rn . variables with distribution NC (0, 1). From the figure we ob-
trace(APP A )≤N i=1
H H
serve the following. The triangularization with DPC scheme
is approaching the closed-loop MIMO rates for higher SNR.
3.1. Fundamental limits The MZF solution is performing very well for lower SNRs
To evaluate the performance of the above schemes, we con- (approaching CL-MIMO and DPC rates), while for higher
sider the following baseline solutions. SNRs, the rates for the ZF scheme are converging to the MZF
(1) No prefiltering solution where each mobile terminal rates. The TDMA-CBF rates are increasing with SNR, but still
is served by one transmit antenna dedicated to that mobile. significantly lower than the rates of the proposed optimiza-
This is equivalent to S = I. A transmit antenna is assigned tion schemes. The solution where no prefiltering is applied
to a particular terminal corresponding to the best channel clearly exhibits properties of an interference limited system
(maximum channel magnitude) among all available transmit (i.e., after a certain SNR, the rates are not increasing). Corre-
antennas and that terminal. sponding cumulative distribution functions (cdf) of the sum
(2) Equal resource TDMA and coherent beamforming rates normalized by the number of users are given in Figure 3
(denoted as TDMA-CBF) is a solution where signals for dif- for SNR = 10 dB (see more on the “capacity-versus-outage”
ferent terminals are sent in different (isolated) time slots. In approach in [17]).
this case, there is no interference, and each terminal is us- In Figure 4, we present the behavior of the average rates
ing 1/N of the overall resources. When serving a particular per user versus number of transmit antennas. The average
mobile, ideal coherent beamforming is applied using all M rates are observed for SNR = 10 dB, N = 3, and variable
transmit antennas. number of transmit antennas (M = 3, 6, 12, 24). The rates in-
(3) Closed-loop MIMO (using the water-pouring opti- crease with the number of transmit antennas and the differ-
mization on eigenmodes) is a solution that is used as an up- ence between the rates for different schemes becomes smaller.
per bound on the achievable sum rates. In the following, it As the number of transmit antennas increases, while keeping
is denoted as CL-MIMO. This solution would require that the number of users N fixed, the spatial channels (i.e., rows of
multiple terminals act as a joint multiple-antenna receiver. the matrix H) are getting less cross-correlated (approaching
1 5.5
0.9
5

0.8
0.7 4.5
Probability
0.6 4
0.5
3.5
0.4
0.3 3
0.2
2.5
0.1
0 2
0 1 2 3 4 5 6 3 4 5 6 7 8 9 10 11 12
Rate (bits/symbol) Number of available terminals
ZF ZF
TDMA-CBF MZF
No prefiltering DPC
MZF
DPC Figure 5: Average rate per user versus number of available terminals
CL-MIMO (SNR = 10 dB, M = 3, Rayleigh channel).
Figure 3: CDF of rates, SNR = 10 dB, per user (M = 3, N = 3,
Rayleigh channel).
antennas and each terminal, there is (1 × M)-dimensional
8
spatial channel. For each set of the terminals there is a matrix
channel H j ∈ C M ×M where each row corresponds to a differ-
ent spatial channel of the corresponding terminal in the set.
7
The selected terminals are the ones corresponding to the set
−1

6 J = arg min HHj H j HHj , (18)
j
5
where · is the Frobenius norm. The above criterion will
favor the terminals whose spatial channels have low cross-
4 correlation. In Figure 5, we present the average rates per user
(the average sum rates divided by N = M) versus number of
3 available terminals. The increase in the rates with the number
of available terminals is a result of multiuser diversity (i.e.,
2 having more terminals allows the transmitter to select more
1 2 3 4 5 6 7 8 favorable channels).
M/N
3.2. Effect of CSI delay
ZF As a motivation for the analysis presented in the follow-
MZF
DPC ing sections, we now present the effects of imperfect chan-
nel state knowledge. In practical communication systems the
Figure 4: Average rate per user versus M/N (SNR = 10 dB, N = channel state H has to be estimated at the receivers, and
3, variable number of transmit antennas M = 3, 6, 12, 24, Rayleigh then fed to the transmitter. Specifically, mobile terminal n
channel). 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-
orthogonality for M → ∞). It can be shown that for orthog- matched) estimates being available to the transmitter to per-
onal channels, all three schemes perform identically. form the optimization. As said earlier, the MIMO channel
We now illustrate a case when the number of available is time varying. Let Hi−1 and Hi correspond to consecutive
terminals Nt (i.e., users) is equal to or greater than the num- block-faded channel responses. The temporal characteristic
ber of transmit antennas M. Out of Nt terminals, the trans- of the channel is described using the correlation
mitter will select N = M terminals and perform the above
transmitter optimization schemes for the selected set. There E h(i−1)nm h∗inm
k= , (19)
are Nt !/((Nt − M)!M!) possible sets. Between the transmit Γ
5.5 transmit antennas at the base station. The MISO channel

5 hn = [hn1 · · · hnM ] for mobile terminal n (n = 1, . . . , N) cor-
4.5
responds to the nth row of the channel matrix H, and we
assume that it is independent of other channels (i.e., rows of

4 the channel matrix). The temporal evolution of the MISO
3.5 channel hn may be represented as [19, 20]
3 hn (t) = [1 · · · 1]Dn Nn , Dn ∈ C N f ×N f , Nn ∈ C N f ×M ,
2.5 (20)
2
where Nn is an (N f × M)- dimensional matrix with elements
1.5 corresponding to complex i.i.d. random variables with dis-
1
tribution NC (0, 1/N f ). Dn is an N f × N f diagonal Doppler
shift matrix with diagonal elements
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Correlation
dii = e jωi t (21)
No prefiltering representing the Doppler shifts that affect N f plane waves

ZF and
MZF
2π
ωi = vn cos γi , for i = 1, . . . , N f , (22)
Figure 6: Average rate per user versus temporal channel correlation λ
k (SNR = 10 dB, M = 3 (solid lines), M = 6 (dashed lines), N = 3,
Rayleigh channel). where vn is the velocity of mobile terminal n and the angle of
arrival of the ith plane wave at the terminal is γi (generated
as U[0 2π]).
where Γ = E[hinm h∗inm ], and hinm is a stationary random pro- It can be shown that the model in (20) strictly conforms
cess (for m = 1, . . . , M and n = 1, . . . , N, denoting transmit- to the Jakes model for N f → ∞. This model assumes that
and receive-antenna indices, respectively). Low values of the at the mobile terminal the plane waves are coming from all
correlation k correspond to higher mismatch between Hi−1 directions with equal probability. Further, note that each di-
and Hi . Note that the above channel is modeled as a first- agonal element of Dn corresponds to one Doppler shift. Dn
order discrete Markov process. In the case of the Jakes model, and Nn are independently generated. With minor modifica-
k = J0 (2π fd τ), where fd is the maximum Doppler frequency tions, the above model can be modified to capture the spatial
and τ is the time difference between h(i−1)nm and hinm . In ad- correlations as well (see [21]).
dition, the above simplified model assumes that there is no We assume that the transmitter has a set of previous
spatial correlation. channel responses (for mobile terminal n) hn (t) where t =
We assume that the mobile terminals feed back Hi−1 kTch and k = 0, −1, . . . , −(L − 1). The time interval Tch
which is used at the base station to perform the transmitter may correspond to a period when a new CSI is sent from
optimization for the ith block. In other words, the downlink the mobile terminal to the base station. Knowing that the
transmitter is ignoring the fact that Hi = Hi−1 . In Figure 6, wireless channel has correlations, based on previous channel
we present the average rate per user versus the temporal responses the transmitter may perform a prediction of the
channel correlation k in (19). From these results we note the channel response hn (τ) at the time moment τ. In this paper
very high sensitivity of the schemes to the channel mismatch. we assume that the prediction is linear and that it minimizes
In this particular case, the performance of the ZF and MZF the mean square error between true and predicted channel
schemes becomes worse than when there is no prefiltering. state. The MMSE predictor Wn is
See also [18] for a related study of channel mismatch and 2
achievable data rates for single-user MIMO systems. Note Wn = min arg ETH hun − hn (τ)H , (23)
T
that the above example and the model in (19) is a simplifi-
cation that we only use to illustrate the schemes’ sensitivity where hun is a vector defined as
to imperfect knowledge of the channel state. In the following
section, we introduce a detailed channel model incorporat- T
hun = hn (0)hn − Tch · · · hn − (L − 1)Tch . (24)
ing correlations in the channel state information.
In other words, the vector is constructed by stacking up the
4. CHANNEL STATE PREDICTION FOR previous channel responses available to the transmitter. We
PERFORMANCE ENHANCEMENT define the following matrices:
In the following, we first address the temporal aspects of the
Un = E hun hH
un ,
channel H. For each mobile terminal, there is a (1 × M)- (25)
dimensional channel between its receive antenna and M Vn = E hun hn (τ) .
3.5 3.5
3 3

2.5 2.5
2 2
1.5 1.5
1 1
0.5 0.5
0 1 2 3 4 5 6 7 8 0 20 40 60 80 100 120
CSI delay (ms) Velocity (km/h)
No prefiltering No prefiltering
ZF ZF
MZF MZF
Figure 7: Average rate per user versus CSI delay, with MMSE pre- Figure 8: Average rate per user versus terminal velocity, with
diction (dashed lines) and without MMSE prediction (solid lines) MMSE prediction (dashed lines) and without MMSE prediction
(SNR = 10 dB, M = 3, N = 3, channel based on model in (20), (solid lines) (SNR = 10 dB, M = 3, N = 3, channel based on model
fc = 2 GHz, v = 30 kmph). in (20), fc = 2 GHz, τ = 2 milliseconds).
It can be shown that the linear MMSE predictor Wn is [22] optimization at the moment t + τ. Figure 7 presents average
rates for the ZF and MZF schemes, for SNR = 10 dB. Results
Wn = Un−1 Vn . (26) depicted by the solid lines correspond to the application of
the delayed CSI H(t) instead of the true channel state H(t+τ).
The above predictor exploits the correlations of the MISO The dashed lines depict results when the MMSE predicted
channel. Note that different linear predictors are needed for channel state HMMSE (t +τ) is used instead of the true channel
different mobile terminals. state H(t + τ). Without any particular effort to optimally se-
A practical implementation of the above prediction can lect the implementation parameters, in this particular exam-
use sample estimates of Un and Vn as ple, we use L = 10 previous channel responses to construct
−1
the vectors in (24). Further, the length of the integration win-
n =
1 H dow in (27) is selected to be Nw = 100. The results clearly
U hun iTch hun iTch ,
Nw i=−Nw point to improvements in the performance of the schemes
(27) when the MMSE channel state prediction is used. The re-
−1
n =
1 sults suggest that the temporal correlations in the channel
V hun iTch hn τ + iTch .
Nw i=−Nw alone are significant enough to support the application of
the MMSE prediction. The presence of spatial correlations in
Underlying assumption in using the above estimates is that the channel model will further improve the benefits of such
the channel is stationary over the integration window Nw Tch . channel state prediction schemes used in conjunction with
Further, if the update of the CSI is performed at discrete time transmitter optimization.
moments kTch (k = 0, −1, . . .), the update period Tch should For the above assumptions, in Figure 8 we present the av-
be such that erage rates per user versus the terminal velocity with the CSI
delay τ = 2 milliseconds. From the results, we can see that
1 the prediction scheme significantly extends the gains of the
Tch < . (28)
2 fdoppler transmitter optimization even for higher terminal velocities.
In Figure 7, we present the average rates per user versus

5. CONCLUSION
the delay τ of the CSI. The system consists of M = 3 trans-
mit antennas and N = 3 terminals. The channel is modeled The advances in programmable and reconfigurable radios
based on (20) (assuming that the carrier frequency is 2 GHz have rendered feasible transmitter optimization schemes that
and the velocity of each mobile terminal is 30 kmph and set- can greatly improve the performance of multiple-antenna
ting the number of plane waves N f = 100). Because the ideal multiuser systems. In this paper, we presented a study on
channel state H(t + τ) is not available at the transmitter, we multiple-antenna transmitter optimization schemes for mul-
assume that H(t) is used instead to perform the transmitter tiuser systems that are based on linear transformations and
transmit power optimization. We considered the maximum Proof of Proposition 1. Note that W = A in (8), for the MZF
achievable sum data rates in the case of the zero-forcing, transmitter spatial prefiltering. We normalize the column
the modified zero-forcing, and the triangularization spatial vectors of the matrix W in (A.3) as
prefiltering coupled with the dirty-paper coding transmis-
sion scheme. We showed that the triangularization with DPC Wnor = WP, (A.4)
approaches the closed-loop MIMO rates (upper bound) for
where P is defined in (10). In other words, the nth diagonal
higher SNR. Further, the MZF solution performed very well
element of P is selected as
for lower SNRs (approaching closed-loop MIMO and DPC
rates), while for higher SNRs, the rates for the ZF scheme 1
pnn = (n = 1, . . . , N), (A.5)
converged to that of the MZF rates. A key impediment to the wnH wn
successful deployment of transmitter optimization schemes
is the delay in the channel state information (CSI) that is where wn is the nth column vector of the matrix W (where
used to accomplish this. We characterized the degradation in wn = an , which is the column vector of A for n = 1, . . . , N).
the performance of transmitter optimization schemes with It is well known that any normalization of the columns of the
respect to the delayed CSI. A linear MMSE predictor of the MMSE receiver in (A.3) does not change the SINRs. In other
channel state was introduced which then improved the per- words, the SINR for the nth uplink user (n = 1, . . . , N) is
formance in all cases. The results have suggested that the tem-
2
poral correlations in the channel alone are significant enough Pav wnH h̄n
to support the application of the MMSE prediction. In the SINRUL
n = N 2
Pav i=1, i=n wnH h̄i + N0 wnH wn
presence of additional spatial correlations, the usefulness of (A.6)
2
such prediction schemes will be even greater. Pav wnH h̄n / wnH wn
= N 2 ,
Pav i=1, i=n wnH h̄i / wnH wn + N0
APPENDICES
where h̄n is the nth column vector of the matrix H̄. Note
A. DEFINITION OF THE VIRTUAL LINK AND n = hn which is the nth row vector of the downlink
that h̄H
PROOF OF PROPOSITION 1 MIMO channel H. The corresponding downlink SINR when
the MZF spatial perfiltering is used (with P defined in (10))
We now describe the corresponding virtual uplink for the is
system in Figure 1. Let x̄n be the uplink information-bearing 2
signal transmitted from mobile terminal n (n = 1, . . . , N) Pav hn an / anH an
SINRMZF = N 2 . (A.7)
Pav i=1, i=n hi ai / anH an + N0
and let ȳm be the received signal at the mth base sta- n
tion antenna (m = 1, . . . , M). x̄n are assumed to be cir-

cularly symmetric complex random variables having Gaus- As said earlier, wn = an and h̄Hn = hn . Thus, SINRn
MZF
=
sian distribution NC (0, Pav ). Further, the received vector ȳ = UL
SINRn for n = 1, . . . , N leading to identical rates, which
[ ȳ1 , . . . , ȳM ]T is concludes the proof.
ȳ = H̄x̄ + n̄ = HH x̄ + n̄, ȳ ∈ C M , x̄ ∈ C N ,
(A.1) B. SPATIAL PREFILTERING WITH DPC
n̄ ∈ C M , H̄ ∈ C M ×N ,
One practical, but suboptimal, single-dimensional DPC so-
lution is described in [14, 15]. Starting from that solution we
where x̄ = [x̄1 , . . . , x̄N ]T is the transmitted vector (E[x̄x̄H ] = introduce the DPC scheme.
Pav IN ×N ), n̄ is AWGN (E[n̄n̄H ] = N0 IM ×M ), and H̄ = HH is The transmitted signal in (1) intended for terminal n is
the uplink MIMO channel response matrix.
It is well known that the MMSE receiver is the opti- xn = fmod xn − In , (B.1)
mal linear receiver for the uplink (multiple-access channel)
[23, 24]. It maximizes the received SINR (and rate) for where xn is the information-bearing signal for terminal n and
each user. The decision statistic is obtained after the receiver fmod (·) is a modulo operation (i.e., a uniform scalar quan-
MMSE filtering as tizer). For a real variable x, fmod (x) is defined as

x̄dec = WH ȳ, (A.2) fmod (x) = (x + Z) mod(2Z) − Z (B.2)
and in the case of a complex variable a + jb, fmod (a + jb) =

where the MMSE receiver is fmod (a) + j fmod (b). The constant Z is selected such that
H E[xn xn∗ ] = Pav . Further, from (12), In is the normalized in-
N0 −1
W= HHH + I H terference at terminal n:
Pav
(A.3) −1
n
N0 −1 gni xi
= HH HHH + I . In = , (B.3)
Pav i=1
gnn
assuming that gnn = 0. Note that In is only known at Wireless Transmission, vol. 62, pp. 107–122, Rutgers Univer-
the transmitter. At terminal n, the following operation is sity, Piscataway, NJ, USA, October 2002.
performed: [7] E. Rashid-Farrohi, L. Tassiulas, and K. J. R. Liu, “Joint op-
timal power control and beamforming in wireless networks

yn using antenna arrays,” IEEE Trans. Commun., vol. 46, no. 10,
fmod = xn + n∗
n, (B.4) pp. 1313–1324, 1998.
gnn
[8] E. Visotsky and U. Madhow, “Optimum beamforming using
∗
where nn is a wrapped-around AWGN (due to the nonlin- transmit antenna arrays,” in Proc. 49th IEEE Vehicular Tech-
nology Conference (VTC ’99), vol. 1, pp. 851–856, Houston,
ear operation fmod (·)). For high SNR and with xn being uni-
Tex, USA, May 1999.
formly distributed over the single-dimensional region, the [9] E. Visotsky and U. Madhow, “Space-time transmit precoding
achievable rate is approximately 1.53 dB away from the rate with imperfect feedback,” IEEE Trans. Inform. Theory, vol. 47,
in (13) [14, 15]. no. 6, pp. 2632–2639, 2001.
To further approach the rate in (13), based on [14], the [10] S. Thoen, L. Van der Perre, M. Engels, and H. De Man,
following modifications of the suboptimal scheme in (B.1) “Adaptive loading for OFDM/SDMA-based wireless net-
are needed. The transmitted signal intended for terminal n is works,” IEEE Trans. Commun., vol. 50, no. 11, pp. 1798–1810,
2002.
now [11] D. Samardzija and N. Mandayam, “Multiple antenna trans-
mitter optimization schemes for multiuser systems,” in Proc.
xn = fk xn − ξn In + dn , (B.5) 58th IEEE Vehicular Technology Conference (VTC ’03), vol. 1,
pp. 399–403, Orlando, Fla, USA, October 2003.
where fk (·) is a modulo operation over a k-dimensional re- [12] G. Strang, Linear Algebra and Its Applications, Harcourt Brace
gion. ξn is a parameter to be optimized (0 < ξn ≤ 1) and Jovanovich, San Diego, Calif, USA, 3rd edition, 1988.
dn is a dither (uniformly distributed pseudonoise over the k- [13] M. H. M. Costa, “Writing on dirty paper (Corresp.),” IEEE
dimensional region). At terminal n, the following operation Trans. Inform. Theory, vol. 29, no. 3, pp. 439–441, 1983.
is performed: [14] G. J. Foschini and A. H. Diaz, “Dirty paper coding: perturb-
ing off the infinite dimensional lattice limit,” in Proc. DI-

yn MACS Workshop on Signal Processing for Wireless Transmis-
fk = xn + 1 − ξn un + ξn n∗
n, (B.6) sion, vol. 62, pp. 141–160, Rutgers University, Piscataway, NJ,
gnn
USA, October 2002.
where n∗n is a wrapped-around AWGN (due to the nonlin- [15] U. Erez, R. Zamir, and S. Shamai, “Additive noise channels
with side information at the transmitter,” in Proc. 21st IEEE
ear operation fk (·)) and un is uniformly distributed over the Convention of Electrical and ELectronic Engineers in Israel, pp.
k-dimensional region. For k → ∞ and xn being uniformly 373–376, Tel-Aviv, Israel, April 2000.
distributed over the k-dimensional region, the rate in (13) [16] A. S. Cohen and A. Lapidoth, “Generalized writing on dirty
can be achieved [14]. Further details on selecting ξn and dn paper,” in Proc. IEEE International Symposium on Information
are beyond the scope of this paper. We refer the reader to [14] Theory (ISIT ’02), pp. 227–227, Lausanne, Switzerland, June–
and references therein. July 2002.
[17] E. Biglieri, J. Proakis, and S. Shamai, “Fading channels:
information-theoretic and communications aspects,” IEEE
ACKNOWLEDGMENT Trans. Inform. Theory, vol. 44, no. 6, pp. 2619–2692, 1998.
This work is supported in part by the National Science Foun- [18] D. Samardzija and N. Mandayam, “Pilot-assisted estimation
dation under Grant no. FMF 0429724. of MIMO fading channel response and achievable data rates,”
IEEE Trans. Signal Processing, vol. 51, no. 11, pp. 2882–2890,
2003, Special Issue on MIMO.
REFERENCES [19] H. Xu, D. Chizhik, H. Huang, and R. Valenzuela, “A wave-
[1] G. J. Foschini, “Layered space-time architecture for wireless based wideband MIMO channel modeling technique,” in
communication in a fading environment when using multiple Proc. 13th IEEE International Symposium on Personal, Indoor
antennas,” Bell Labs Technical Journal, vol. 1, no. 2, pp. 41–59, and Mobile Radio Communications (PIMRC ’02), vol. 4, pp.
1996. 1626–1630, Lisbon, Portugal, September 2002.
[2] G. J. Foschini and M. J. Gans, “On limits of wireless commu- [20] D. Chizhik, “Slowing the time-fluctuating MIMO channel by
nications in a fading environment when using multiple anten- beam-forming,” IEEE Transactions on Wireless Communica-
nas,” Wireless Personal Communications, vol. 6, no. 3, pp. 311– tions, vol. 3, no. 5, pp. 1554–1565, 2004.
335, 1998. [21] D. Samardzija and N. Mandayam, “Downlink multiple an-
[3] T. M. Cover and J. A. Thomas, Elements of Information Theory, tenna transmitter optimization on spatially and temporally
John Wiley & Sons, New York, NY, USA, 1st edition, 1991. correlated channels with delayed channel state information,”
[4] G. Caire and S. Shamai, “On the achievable throughput of in Proc. Conference on Information Sciences and Systems (CISS
a multiantenna Gaussian broadcast channel,” IEEE Trans. In- ’04), Princeton University, Princeton, NJ, USA, March 2004.
form. Theory, vol. 49, no. 7, pp. 1691–1706, 2003. [22] S. Haykin, Adaptive Filter Theory, Prentice-Hall, Englewood
[5] D. N. C. Tse and P. Viswanath, “On the capacity region of Cliffs, NJ, USA, 2nd edition, 1991.
the vector Gaussian broadcast channel,” in Proc. IEEE Interna- [23] U. Madhow and M. L. Honig, “MMSE interference sup-
tional Symposium on Information Theory (ISIT ’03), pp. 342– pression for direct-sequence spread-spectrum CDMA,” IEEE
342, Yokohoma, Japan, June–July 2003. Trans. Commun., vol. 42, no. 12, pp. 3178–3188, 1994.
[6] S. Vishwanath, G. Kramer, S. Shamai , S. Jafar, and A. Gold- [24] S. Verdú, Multiuser Detection, Cambridge University Press,
smith, “Capacity bounds for Gaussian vector broadcast chan- Cambridge, UK, 1998.
nels,” in Proc. DIMACS Workshop on Signal Processing for
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 field 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 staff 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 floor, 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.
c 2005 François Horlin et al.
Flexible Transmission Scheme for 4G Wireless

Systems with Multiple Antennas
François Horlin
Wireless Research, Interuniversity Micro-Electronics Center (IMEC), Kapeldreef 75, B-3001 Leuven, Belgium
Email: francois.horlin@imec.be
Frederik Petré
Email: petre@imec.be
Eduardo Lopez-Estraviz
Email: lopezest@imec.be
Frederik Naessens
Email: naessen@imec.be
Liesbet Van der Perre

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-prefixed 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 efficiently trades
off 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-prefix single carrier, space-division multiplexing, space-time
block coding, joint detection.
1. INTRODUCTION ments. Combinations of the multicarrier (MC) and spread-

Because of the limited frequency bandwidth, on the one spectrum (SS) modulations, named multicarrier spread-
hand, and the potential limited power of terminal stations, spectrum techniques, could be interesting candidates. They
on the other hand, spectral and power efficiency of fu- might benefit from the main advantages of both MC and SS
ture systems should be as high as possible. New air inter- schemes such as high spectral efficiency, multiple-access ca-
faces need to be developed to meet the new system require- pabilities, narrowband interference rejection, simple one-tap
equalization, and so forth.
Cellular systems of the third generation are based
This is an open access article distributed under the Creative Commons on the recently emerged direct-sequence code-division
Attribution License, which permits unrestricted use, distribution, and multiple-access (DS-CDMA) technique [1]. Intrinsically,
reproduction in any medium, provided the original work is properly cited. 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 link reliability relative to single-antenna systems [18, 19, 20].
in the uplink. Second, soft handover is supported between MIMO systems explicitly rely on the fact that the channels
two cells making use of different codes at the base stations. created by the additional spatial dimension are independent
However, the system suffers from intersymbol interference from each other. This is approximately true in rich scattering
(ISI) and multiuser interference (MUI) caused by multipath environments. However, when the channels are correlated,
propagation, leading to a high loss of performance. the gain obtained by the use of multiple antennas is lim-
The use of the orthogonal frequency-division multiplex- ited. Until very recently, the main focus of MIMO research
ing (OFDM) modulation is widely envisaged for wireless was on single-user communications over narrowband chan-
local area networks [2]. At the cost of the addition of a nels, thereby neglecting the multiple-access aspects and the
cyclic prefix, the time dispersive channel is seen in the fre- frequency-selective fading channel effects, respectively.
quency domain as a set of parallel independent flat sub- First, if the multiantenna propagation channels are suf-
channels and can be equalized at a low-complexity. An al- ficiently uncorrelated, MIMO systems can create Nmin par-
ternative approach to OFDM, that benefits from the same allel spatial pipes, which allows to realize an Nmin -fold ca-
low complexity equalization property, is single-carrier block pacity increase, where Nmin = min {NT , NR } (NT and NR
transmission (SCBT), also known as single-carrier (SC) with denote the number of transmit and receive antennas, resp.)
frequency-domain equalization. Since the SCBT technique is called the spatial multiplexing gain [18, 19, 20]. Specifi-
benefits from a lower peak-to-average power ratio (PAPR), cally, space-division multiplexing (SDM) techniques exploit
[3] encourages the use of SCBT in the uplink and OFDM in this spatial multiplexing gain, by simultaneously transmit-
the downlink in order to reduce the constraints on the analog ting Nmin independent information streams at the same fre-
front end and the processing complexity at the terminal. quency [21, 22]. In [23], SDM is combined with SC-CDMA
There are potential benefits in combining OFDM (or to increase the data rate of multiple users in a broadband cel-
SCBT) and DS-CDMA. Basically the frequency-selective lular network.
channel is first equalized in the frequency domain using the Second, MIMO systems can also create NT NR indepen-
OFDM modulation technique. DS-CDMA is applied on top dently fading channels between the transmitter and the re-
of the equalized channel, keeping the interesting orthogo- ceiver, which allows to realize an NT NR -fold diversity in-
nality properties of the codes. The DS-CDMA signals are crease, where NT NR is called the multiantenna diversity gain.
either spread across the OFDM carriers (referred to as in- Specifically, space-time coding (STC) techniques exploit di-
trablock spreading) leading to multicarrier CDMA (MC- versity and coding gains, by encoding the transmitted signals
CDMA) [4, 5, 6, 7], or across the OFDM blocks (referred to not only over the temporal domain but also over the spa-
as interblock spreading) leading to multicarrier block-spread tial domain [24, 25, 26]. Space-time block coding (STBC)
CDMA (MCBS-CDMA) [8, 9, 10, 11]. The SCBT counter- techniques, introduced in [25] for NT = 2 transmit anten-
parts named here single-carrier CDMA (SC-CDMA) and nas, and later generalized in [26] for any number of trans-
single-carrier block-spread CDMA (SCBS-CDMA) have also mit antennas, are particularly appealing because they facili-
been proposed in [12, 13, 14, 15], respectively. The differ- tate maximum-likelihood (ML) detection with simple linear
ent flavors to mix the MC and SS modulations complement processing. However, these STBC techniques have originally
each other and allow to make an efficient tradeoff between been designed for frequency-flat fading channels exploiting
the spectral and power efficiency according to the user re- only multiantenna diversity of order NT NR . Therefore, time-
quirements, channel propagation characteristics (time and reversal (TR) STBC techniques, originally proposed in [27]
frequency selectivity), and terminal resources. For example, for single-carrier serial transmission, have been combined
it has been recently proposed, in [16] for the downlink, and with SCBT in [28, 29] for signaling over frequency-selective
in [17] for the uplink, to perform a two-dimensional spread- fading channels. In [30, 31], the TR-STBC technique of [28]
ing (combination of MC-CDMA with MCBS-CDMA) in or- is combined with SC-CDMA to improve the performance
der to cope with the time and frequency selectivity of the of multiple users in a broadband cellular network. Although
channels. The chips of a given symbol are mapped on adja- this technique enables low-complexity chip equalization in
cent channels where the fading coefficients are almost con- the frequency domain, it does not preserve the orthogonality
stant so that the orthogonality properties of the codes are among users, and hence, still suffers from multiuser inter-
preserved and a low-complexity single-user detector can be ference. The space-time coded multiuser transceiver of [32],
used. which combines the TR-STBC technique of [29] with SCBS-
To meet the data rate and quality-of-service require- CDMA, preserves the orthogonality among users as well
ments of future broadband cellular systems, their spectral as transmit streams, regardless of the underlying multipath
efficiency and link reliability should be considerably im- channels. This allows for deterministic ML user separation
proved, which cannot be realized by using traditional single- through low-complexity code-matched filtering as well as de-
antenna communication techniques. To achieve these goals, terministic ML transmit stream separation through linear
multiple-input multiple-output (MIMO) systems, which de- processing. Another alternative to remove MUI determinis-
ploy multiple antennas at both ends of the wireless link, ex- tically in a space-time coded multiuser setup [33, 34] com-
ploit the extra spatial dimension, besides the time, frequency, bines generalized multicarrier (GMC) CDMA, originally de-
and code dimensions, which allows to significantly increase veloped in [35], with the STBC techniques of [26] but imple-
the spectral efficiency and/or to significantly improve the mented on a per-carrier basis.
s̃m [ j] x̃m [n] xm [n] transmission, the transmission scheme applies to both the
up- and downlink. In the uplink, the different user’s signals
sm [ j] θm Ninter cm [n] FQH T um [n]
are multiplexed at the receiver, after propagation through
Intrablock Interblock
their respective multipath channels. In the downlink, the dif-
IFFT redundancy ferent user signals are multiplexed at the transmitter, before
spreading spreading
the inverse fast Fourier transform (IFFT) operation.
Figure 1: Single-antenna transmitter model. The transmission scheme comprises four basic opera-
tions: intrablock spreading, interblock spreading, IFFT, and
We propose a transmission chain composed of generic adding transmit redundancy. The symbols sm [ j] are first
blocks and able to instantiate all the communication modes serial-to-parallel converted into blocks of B symbols, lead-
combining OFDM/SCBT and DS-CDMA as special cases. In ing to the sequence sm [ j] := [sm [ jB] · · · sm [( j + 1)B − 1]]T .
contrast with the transceiver proposed in [35, 36, 37] that The blocks sm [ j] are linearly precoded with a Q × B (Q ≥ B)
relies on a sharing of the set of carriers to retain the orthog- matrix, θ m , which possibly introduces some redundancy and
onality between the users, our transmission scheme relies on spreads the symbols in sm [ j] with length-Q codes as follows:
orthogonal CDMA, and thus inherits the nice advantages of
CDMA related to universal frequency reuse in a cellular net- s̃m [ j] := θ m · sm [ j]. (1)
work, like increased capacity and simplified network plan-
ning. Furthermore, the focus is especially put on the com- We refer to this first operation as intrablock spreading, since
munication modes emerging in the standards. We demon- the information symbols sm [ j] are spread within a single pre-
strate the rewarding synergy between existing and evolving coded block s̃m [ j]. The precoded block sequence s̃m [ j] is
MIMO communication techniques and our generic trans- then block spread with the elements cm [n] of a length-Ninter
mission technique, which allows to increase the spectral ef- code sequence, leading to Ninter successive chip blocks:
ficiency and to improve the link reliability of multiple users
in a broadband cellular network. Considering realistic prop-
x̃m [n] := s̃m [ j]cm n − jNinter , (2)
agation channels, we also advise a strategy for mode selec-
tion according to the communication conditions, making an
efficient tradeoff between the desired performance and the where j = n/Ninter . We refer to this second operation as
required computational complexity. interblock spreading, since the information symbols sm [ j] are
The paper is organized as follows. In Section 2, the spread along Ninter different chip blocks. The third operation
generic transmission scheme is presented that can capture involves the transformation of the frequency-domain chip
the standard emerging communication modes as special block sequence x̃m [n] into the time-domain chip block se-
cases. It is shown how the MC and SC modes can be instan- quence:
tiated. Section 3 is devoted to the extension of the transmis-
sion scheme introduced in Section 2 to multiple-antenna sys- xm [n] := F H
Q
· x̃m [n], (3)
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- where F HQ
is the Q×Q IFFT matrix. Finally, the K ×Q (K ≥ Q)
alized matrix model. Finally, a strategy for communication transmit matrix T possibly adds some transmit redundancy
mode selection is proposed in Section 5, based on the evalu- to the time-domain chip blocks:
ation of the cost and performance of each mode in a highly
mobile cellular environment. um [n] := T · xm [n]. (4)
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 0M ×N is a matrix of zeros With K = Q + L denoting the total block length T = T cp :=
T
of size M × N. The operators (·)∗ , (·)T , and (·)H denote, re- [I Tcp , I TQ ] , where I cp consists of the last L rows of I Q , T
spectively, the complex conjugate, transpose, and transpose adds redundancy in the form of a length-L cyclic prefix (cp).
conjugate of a vector or a matrix. The operator ⊗ is the Kro- The chip block sequence um [n] is parallel-to-serial converted
necker product between two vectors or matrices. We index into the scalar sequence [um [nK]· · ·um [(n + 1)K − 1]]T :=
the transmitted block sequence by i, the MIMO coded block um [n], and transmitted over the air at a rate 1/Tc (Tc stands
sequence and the intrablock chip block sequence by j, and for the chip duration).
the interblock chip block sequence by n. In the following, we will detail how our generic transmis-
sion scheme instantiates different communication modes,
and, thus, supports different emerging communication stan-
2. SINGLE-ANTENNA GENERIC
dards. We distinguish between the MC modes, on the one
TRANSMISSION SCHEME
hand, and the SC modes, on the other hand. In Figure 2, the
The transmission scheme for the mth user (m = 1, . . . , M) principle of CDMA spreading in two possible dimensions is
is depicted in Figure 1. Since we focus on a single-user illustrated.
Spreading
Code factor
Code
Spreading
factor
Frame block
Frame block
···
Subchannel
Subchannel
(MC carrier frequency or SC time instant)
(MC carrier frequency or SC time instant)
(a) (b)
Figure 2: (a) Intrablock spreading pattern (MC/SC-CDMA). (b) Interblock spreading pattern (MCBS/SCBS-CDMA).
2.1. Instantiation of the multicarrier modes CDMA retains the orthogonality among users in both up-
The MC modes always comprise the IFFT operation, and add and downlink [11]. Hence, it has the potential to convert a
transmit redundancy in the form of a cyclic prefix (T = T cp ). difficult multiuser detection problem into an equivalent set
of simpler and independent single-user equalization prob-
The MC modes comprise MC-CDMA and MCBS-CDMA as
lems. However, it will be shown in Section 5 that the perfor-
particular instantiations of the generic transmission scheme.
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-
2.1.1. MC-CDMA
formation (CSI) is available at the transmitter, it performs
As we have indicated in the introduction, MC-CDMA first linear precoding to robustify the transmitted signal against
performs classical DS-CDMA symbol spreading, followed by frequency-selective fading. In case CSI is available at the
OFDM modulation, such that the information symbols are transmitter, it allows to optimize the transmit spectrum of
spread across the different subcarriers located at different fre- each user separately through adaptive power and bit load-
quencies and characterized by a different fading coefficient ing. Note that classical MC-DS-CDMA can be seen as a spe-
[4, 5, 6]. With Q = BNintra and Nintra the intrablock spreading cial case of MCBS-CDMA, because it does not include linear
code length, the Q × B intrablock spreading matrix θ m = βm precoding, but, instead, only relies on bandwidth-consuming
spreads the chips across the subcarriers, where the mth user’s forward error correction (FEC) coding to enable frequency
Q × B spreading matrix βm is defined as diversity [8, 10, 38].
2.2. Instantiation of the single-carrier modes

βm := I B ⊗ am , (5)
The SC modes employ a fast Fourier transform (FFT) as part
of the intrablock spreading operation to annihilate the IFFT
with am := [am [0] · · · am [Nintra − 1]]T the mth user’s Nintra × operation. For implementation purposes, however, the IFFT
1 code vector. The interblock spreading operation is dis- is simply removed (and not compensated by an FFT), in
carded by setting Ninter = 1. Since intrablock spread- order to minimize the implementation complexity. The SC
ing does not preserve the orthogonality among users in a modes rely on cyclic prefixing (T = T cp ) to make the chan-
frequency-selective channel, MC-CDMA requires advanced
multiuser detection for uplink reception in the base station, nel appear circulant. The SC modes comprise SC-CDMA
and frequency-domain chip equalization for downlink re- and SCBS-CDMA as particular instantiations of the generic
ception in the mobile station. MC-CDMA has been proposed transmission scheme.
as a candidate air interface for future broadband cellular sys-
2.2.1. SC-CDMA
tems [7].
The SC-CDMA transmission scheme, which combines SCBT
2.1.2. MCBS-CDMA with DS-CDMA, can be interpreted as the SC counterpart
The MCBS-CDMA transmission scheme is the only MC of MC-CDMA [12, 13]. This mode is captured through our
mode that comprises the interblock spreading operation general transmission scheme, by setting Q = BNintra . The
Ninter > 1. Since the CDMA spreading is applied on each intrablock spreading matrix θ m = F Q · βm , with βm de-
carrier independently, which can be seen as a constant fad- fined in (5), performs symbol spreading on the B informa-
ing channel if the propagation environment is static, MCBS- tion symbols, followed by an FFT operation to compensate
m
sm
1 [ j] s̃1 [ j] x̃1m [n] x1m [n] streams on each transmit antenna nT , as expressed in
d1m [i] θm Ninter cm [n] FQH T um
1 [n]
m
sm
nT [ j] = d nT [i], (6)
. MIMO .
.. ..
coding sm m
NT [ j] s̃NT [ j]
m
x̃N [n] xm [n]
T NT where j = i.
dNmT [i] θm Ninter cm [n] FQH T um
NT [n]
3.2. Space-time block coding
Figure 3: MIMO transmitter model.

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
for the subsequent IFFT operation. Like MC-CDMA, SC- conciseness, we limit ourselves to the case of NT = 2 transmit
CDMA does not preserve the orthogonality among the users antennas (Alamouti scheme [25, 27]). The STBC coding is
in a frequency-selective channel. It requires advanced mul- implemented by coding the two antenna streams across two
tiuser detection for uplink reception at the base station and time instants, as expressed in
chip equalization for downlink reception at the mobile ter-
minal. On the contrary to MC-CDMA, each user symbol is sm
1 [ j] dm
1 [i]
= ,
spread amongst multiple subchannels of the same power at-
m
s2 [ j] dm
2 [i]
tenuation. The interblock spreading operation is left out by (7)
sm
1 [ j + 1] dm1 [i]
∗
setting Ninter = 1. =χ· m ,
m
s2 [ j + 1] d 2 [i]∗
2.2.2. SCBS-CDMA
where i = j/2 and
The SCBS-CDMA transceiver can be considered as the SC
counterpart of MCBS-CDMA. It is the only SC mode that 0 −1
entails the interblock spreading operation (Ninter > 1). The χ := χ ⊗χ with χ := . (8)
NT B NT 1 0
intrablock spreading matrix θ m = F Q only performs an FFT
operation to compensate for the subsequent IFFT operation. In the case of the MC modes, the STBC coding is applied in
If the propagation channels are static, SCBS-CDMA retains the frequency domain on a per-carrier basis so that
the orthogonality among users in both the up- and downlink,
even after propagation through a frequency-selective chan- χ := I B . (9)
nel (like MCBS-CDMA). It also converts a difficult multiuser B
detection problem into an equivalent set of simpler and in-

In the case of the SC modes, the STBC coding is applied in
dependent single-user equalization problems. The orthogo-
the time domain by further permuting the vector elements
nality property is however lost in time-selective channels, as
so that
studied numerically in Section 5.
χ := F TB · F B (10)
B
3. EXTENSION TO MULTIPLE ANTENNAS
The generic transmission model is extended in Figure 3 to is a B × B permutation matrix implementing a time reversal.
include two types of MIMO techniques. We assume NT an- It is easily checked that the transmitted block at time in-
tennas at the transmit side and NR antennas at the receive stant j + 1 from one antenna is the time-reversed conjugate
side. The information symbols dnmT [i], which are assumed of the transmitted symbol at time instant j from the other
independent and of variance equal to σd2 , are first serial-to- antenna (with possible permutation and sign change). As we
parallel converted into blocks of B symbols, leading to the will show later, this property allows the deterministic trans-
mit stream separation at the receiver, regardless of the under-
block sequence d m T
nT [i] := [dnT [iB] · · · dnT [(i + 1)B − 1]] for
m m
lying frequency-selective channels.
nT = 1, . . . , NT . A MIMO coding operation is performed
across the different transmit antenna streams, that results
into the NT antenna sequences snmT [ j] input to the generic 4. RECEIVER DESIGN
transmission scheme. The overall rate increase obtained by
4.1. Cyclostationarization of the channels
the use of multiple antennas is either 1 or NT depending on
the MIMO technique that is selected. Adopting a discrete-time baseband equivalent model, the
chip-sampled received signal at antenna nR (nR = 1, . . . , NR ),
3.1. Space-division multiplexing vnR [n], is the superposition of a channel-distorted version of
the MNT transmitted user signals, which can be written as
In this section, we combine our generic transmission scheme
with SDM, which allows to instantiate all combinations of m

M
NT
L
SDM with OFDM/SCBT and CDMA as special cases. The vnR [n] = hm m
nR ,nT [l]unT [n − l] + wnR [n], (11)
SDM technique is implemented by sending independent m=1 nT =1 l=0
where hm nR ,nT [l] is the chip-sampled finite impulse response where ΛmnR ,nT
is a diagonal matrix composed of the frequency-
(FIR) channel of order Lm that models the frequency- domain channel response between the mth user’s antenna nT
selective multipath propagation between the mth user’s an- and the base station antenna nR .
tenna nT and the base station antenna nR , including the ef-
fect of transmit/receive filters and the remaining asynchro- 4.2. Matrix model
nism of the quasisynchronous users, and wnR [n] is additive
white Gaussian noise (AWGN) at the base station antenna nR Based on (13), a generalized matrix model is developed that
with variance σw2 . Furthermore, the maximum channel or- relates the vector of transmitted user’s symbols to the vector
der L, that is, L = maxm {Lm }, can be well approximated by of received samples. It encompasses all the combinations of
L ≈ (τmax,a + τmax,s )/Tc + 1, where τmax,a is the maximum OFDM/SCBT with CDMA considered in this paper as special
asynchronism between the nearest and the farthest user of cases. Based on this model, a multiuser joint detector opti-
the cell, and τmax,s is the maximum excess delay within the mized according to the MMSE criterion will be derived and
given propagation environment [35]. its complexity will be reduced by exploiting the cyclostation-
Assuming perfect time and frequency synchronization, arity properties of the matrices.
the sequence vnR [n] is serial-to-parallel converted into the The generalized input/output matrix model that relates
sequence vnR [n] := [vnR [nK] · · · vnR [(n + 1)K − 1]]T . From the MIMO-coded symbol vector defined as
the scalar input/output relationship in (11), we can derive T
the corresponding block input/output relationship: s̄[ j] := s̄1 [ j]T · · · s̄M [ j]T (14)
NT
M
vnR [n] = Hm
nR ,nT
[0] · um
nT [n] with
m=1 nT =1
T
+ Hm
nR ,nT
[1] · um
nT [n − 1] + w nR [n], s̄m [ j] := sm
1 [ j]
T
· · · sm
NT [ j]
T
(15)
(12)
for m = 1, . . . , M to the received and noise vectors defined as
where wnR [n] := [wnR [nK] · · · wnR [(n + 1)K − 1]]T is the
corresponding noise block sequence, H m [0] is a K × K T T
T
nR ,nT ȳ[ j] := ȳ 1 [ j] · · · ȳ NR [ j] ,
lower triangular Toeplitz matrix with entries [H mnR ,nT
[0]] = T (16)
p,q
hm m m
nR ,nT [p − q] for p − q ∈ [0, L ] and [H nR ,nT [0]] = 0 z̄[ j] := z̄1 [ j]T · · · z̄NR [ j]T
p,q
else, and H m
nR ,nT
[1] is a K × K upper triangular Toeplitz
matrix with entries [H m [1]] = hm with
nR ,nT nR ,nT [K + p − q] for
p,q
K + p − q ∈ [0, Lm ] and [H m [1]] = 0 else (see, e.g.,
T
T T
nR ,nT p,q ȳ n [ j] := y n jNinter ··· y n ( j + 1)Ninter − 1 ,
[35] for a detailed derivation of the single-user case). The R

R R
T
T T
delay-dispersive nature of multipath propagation gives rise z̄nR [ j] := znR jNinter ··· znR ( j + 1)Ninter − 1 ,
to so-called interblock interference (IBI) between successive (17)
blocks, which is modeled by the second term in (12).
The Q × K receive matrix R removes the redundancy for nR = 1, . . . , NR , is given by
from the chip blocks, that is, y n [n] := R · vnR [n]. With
R
R = Rcp = [0Q×L , I Q ], R again discards the length-L cyclic ȳ[ j] = C · F H · Λ · θ · s̄[ j] + z̄[ j], (18)
prefix. The purpose of the transmit/receive pair is twofold.
First, it allows for simple block-by-block processing by re- where the channel matrix is defined as
moving the IBI, that is, R · H m [1] · T = 0Q×Q , provided the
nR ,nT  
CP length to be at least the maximum channel order L. Sec- Λ11 · · · 0Q×Q
 .. 
ond, it enables low-complexity frequency-domain processing  ..
.. 
 . . . 
by making the linear channel convolution to appear circu-  
0 M 
lant to the received block. This results in a simplified block  Q×Q · · · Λ1 
 
 .. .. .. 
input/output relationship in the time domain: Λ := 
 . . . 
(19)
 1 
 ΛN · · · 0Q×Q 

M
NT  R 
y n [n] =
m
Ḣ nR ,nT · xm  . .. .. 
nT [n] + z nR [n], (13)  .
R
m=1 nT =1  . . . 

0Q×Q · · · ΛM N R
m
where Ḣ nR ,nT = R · H m
nR ,nT
[0] · T is a circulant channel ma-
trix, and znR [n] = R · wnR [n] is the corresponding noise with
block sequence. Note that circulant matrices can be diago-
m
nalized by FFT operations, that is, Ḣ nR ,nT = F HQ
· Λm
nR ,nT
· FQ, Λm
nR
:= Λm
nR ,1
· · · Λm
nR ,NT
(20)
for m = 1, . . . , M and nR = 1, . . . , NR , and the intrablock 4.2.2. Space-time block coding

spreading, Fourier, and interblock spreading matrices are de-
Taking (7) into account, the model (18) is instantiated to the
fined, respectively, as
STBC input/output matrix model
 
I ⊗ θ 1 · · · 0NT Q×NT B
H
 NT  ȳ stbc [i] = C stbc · F stbc · Λstbc · θ stbc · χ · d̄[i]
 .. .. .. 
θ :=  . . . , stbc (28)
 
M + z̄stbc [i],
0NT Q×NT B · · · I NT ⊗ θ (21)
F := I NR M ⊗ F Q ,
where the vector of transmitted symbols is given in (25)
C := I NR ⊗ C 1 · · · C M , assuming that NT = 2, the received and noise vectors are
defined as
in which
ȳ[ j]
ȳ stbc [i] := ,
C m := c m ⊗ I Q , (22) ȳ[ j + 1]∗
(29)
z̄[ j]
T
with cm := [cm [0] · · · cm [Ninter − 1]] . Note that the model z̄stbc [i] := .
z̄[ j + 1]∗
(18) only holds for static channels. In case of time-selective
m
channels, the channel matrices Ḣ nR ,nT cannot be diagonalized The MIMO encoding, intrablock spreading, channel,
anymore and are different from one chip block to the next Fourier, and interblock spreading matrices are, respectively,
one. given by
4.2.1. Space-division multiplexing  
I MNT B
Taking (6) into account, the model (18) is instantiated to the χ :=  ,
stbc IM ⊗ χ
SDM input/output matrix model
 
H θ 0MNT Q×MNT B
ȳ sdm [i] = C sdm · F sdm · Λsdm · θ sdm · χ · d̄[i] θ stbc :=  ,
sdm (23) 0MNT Q×MNT B θ∗
+ z̄sdm [i],  
Λ 0NR MQ×MNT Q
Λstbc :=  , (30)
where the vector of transmitted symbols is defined as 0NR MQ×MNT Q Λ∗
 
F 0NR MQ×NR MQ
d̄[i] := d̄1 [i]
T M T T
, (24) F stbc :=  ,
· · · d̄ [i] 0NR MQ×NR MQ F∗
 
with C 0NR Ninter Q×NR MQ
C stbc :=  .
T 0NR Ninter Q×NR MQ C∗
m T T
d̄ [i] := d m
1 [i]
m
· · · d NT [i] (25)
4.3. Multiuser joint detector
for m = 1, . . . , M, and the received and noise vectors are de- In order to detect the transmitted symbol block of the pth
fined as p
user d̄ [i] based on the received sequence of blocks within
ȳ sdm [i] := y[ j], the received vector ȳ mode [i] (“mode” stands for “sdm” or
(26) “stbc”), a first solution consists of using a single-user receiver
z̄sdm [i] := z[ j]. that inverts successively the channel and all the operations
performed at the transmitter. The single-user receiver relies
The MIMO encoding, intrablock spreading, channel, implicitly on the fact that CDMA spreading has been applied
Fourier, and interblock spreading matrices are, respectively, on top of a channel equalized in the frequency domain. After
given by CDMA despreading, each user stream is handled indepen-
dently. However the single-user receiver fails in the uplink
χ := I MNT B , where multiple channels have to be inverted at the same time.
sdm
θ sdm := θ, The optimal solution is to jointly detect the transmitted
symbol blocks of the different users within the transmitted
Λsdm := Λ, (27) vector d̄[i] based on the received sequence of blocks within
F sdm := F, the received vector ȳ mode [i]. The optimum linear joint detec-
tor according to the MMSE criterion is computed in [39]. At
C sdm := C.
the output of the MMSE multiuser detector, the estimate of
the transmitted vector is 5.1. Complexity

−1 To evaluate the complexity of the different receivers, we dis-
σw2 tinguish between the initialization phase, when the receiver is
d̄ˆ [i] = I + GH · Gmode · GH · ȳ [i], calculated, and the data processing phase, when the received
σd2 MNT B mode mode mode
(31) data is actually processed. The rate of the former is related

to the channel’s fading rate, whereas the latter is executed
where continuously at the symbol block rate. The complexity will
be described in terms of complex multiply/accumulate cycle
H (MAC) operations. It is assumed that 2N 3 complex MAC op-
Gmode := C mode · F mode · Λmode · θ mode · χ . (32) erations are required to invert a matrix of size N, and that
mode
N log2 (N) complex MAC operations are required to com-
pute an FFT of size N.
The MMSE linear joint detector consists of two main opera-
tions [39, 40]. 5.1.1. Initialization complexity
(i) First, a filter matched to the composite impulse re- The complexity required to compute the MMSE receiver is
sponses multiplies the received vector in order to minimize given in Table 1 for the base station and for the terminal,
the impact of the white noise. The matched filter consists respectively. It has been assumed that the inversion of the
of the CDMA interblock despreading, the FFT operator to MMSE equalizer block diagonal inner matrix is dominant.
move to the frequency domain, the maximum ratio combin- Since the size of each block is equal to the number of inter-
ing (MRC) of the different received antenna channels, the fering users times the number of interfering antennas, the
CDMA intrablock despreading, the IFFT to go back to the computation of the equalizer is more complex in case of
time domain in case of the SC modes, and the STBC decod- intrablock spreading (number of interfering users equal to
ing. the number of users) than in case of interblock spreading
(ii) Second, the output of the matched filter is still multi- (number of interfering users equal to one), and in case of
plied with the inverse of the composite impulse response au- SDM (number of interfering antennas equal to the number
tocorrelation matrix of size MNT B that mitigates the remain- of transmit antennas) than in case of STBC (number of in-
ing intersymbol, interuser, and interantenna interference. terfering antennas equal to one).
In the case of interblock spreading (MCBS and SCBS-
CDMA), the spreading matrix C has the property that 5.1.2. Data processing complexity
C H C = I NR QM if the CDMA codes are chosen orthogo- The complexity needed during the data processing phase to
nal. When the propagation channels are static, the different transmit and receive each user’s transmitted complex symbol
user streams are perfectly separated at the output of the in- is further given in the Table 2. The computational effort is al-
terblock despreading operation and further treated indepen- most equally shared between the transmitter and the receiver
dently. The MMSE multiuser joint detector exactly reduces in the case of the MC-based modes. In the case of the SC-
to independent single-user detectors. In case of time-varying based modes, all the computational effort has been moved
propagation channels, model (18) is not valid anymore and to the receiver. From a terminal complexity point of view, it
the multiuser MMSE joint detector cannot be simplified to is clearly advantageous to use the SC-based modes in uplink
single-user detectors. and the MC-based modes in downlink. Both the transmis-
In the case of intrablock spreading (MC- and SC- sion and the reception are less complex in case of interblock
CDMA), however, the linear MMSE receiver is different from spreading than in case of intrablock spreading, because the
the single-user receiver, and suffers from a higher computa- (I) FFT operators can be executed at a lower rate in the for-
tional complexity. Fortunately, both the initialization commer case (namely, before the spreading at the transmitter and
plexity, which is required to compute the MMSE receiver, and after the despreading at the receiver). Since STBC combines
the data processing complexity can be significantly reduced NT successive symbol blocks, it is on the overall more com-
for MC- and SC-CDMA, by exploiting the initial cyclosta- plex than the SDM scheme.
tionarity property of the channels. Based on a few permuta-
tions and on the properties of the block circulant matrices 5.2. Peak-to-average power ratio
given in [12], it can be shown that the initial inversion of the
The PAPR is illustrated in Figure 4 for each mode as a func-
square autocorrelation matrix of size MNT B can be replaced
tion of the spreading factor. The user signals are spread by pe-
by the inversion of B square autocorrelation matrices of size
riodic Walsh-Hadamard codes for spreading, which are over-
MNT .
laid with an aperiodic Gold code for scrambling. QPSK mod-
ulation is assumed with Q = 128 subchannels, and a cyclic
5. MODE SELECTION STRATEGY prefix length of L = 32. The PAPR is defined as the maxi-
mum instantaneous peak power over all combinations of the
In this section, a cost and performance comparison between transmitted symbols divided by the average transmit power.
the different communication modes is made, which can serve While the MC-based communication modes feature a
as an input for an efficient mode selection strategy. very large PAPR, the SC-based communication modes have
Table 1: Initialization complexity, M users, NT transmit antennas, Q carriers/subchannels, and B complex symbols per block.
Base station Mobile terminal

SDM STBC SDM STBC
MC/SCBS-CDMA 2QM(NT )3 2MQ 2Q(NT )3 2Q
MC/SC-CDMA 2B(MNT )3 2B(M)3 2B(MNT )3 2B(M)3
Table 2: Data processing complexity, M users (M = 1 at the terminal), NT transmit antennas, NR receive antennas, Q carriers/subchannels,
B complex symbols per block.
Transmitter Receiver
SDM STBC SDM STBC
NR
MCBS-CDMA log2 Q NT log2 Q log2 Q + NR NR log2 Q + NT NR
NT

NR

SCBS-CDMA 0 0 + 1 log2 Q + NR NR + 1 log2 Q + NT NR
NT
Q Q NR Q Q Q Q
MC-CDMA log2 Q NT log2 Q log2 Q + NR NRlog2 Q + NT NR
MB MB NT MB B MB B

NR Q Q Q Q
SC-CDMA 0 0 + 1 log2 B + NR NR + 1 log2 B + NT NR
NT MB B MB B
25 5.3. Goodput performance

We consider a mobile cellular system which operates in an
20 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
15 cell is equal to 3 km. The mobile terminals are moving at a
PAPR (dB)
speed ranging from 0 (static) to 250 km/h (highly mobile).

The system operates at a carrier frequency of 2 GHz, with a
10
system bandwidth of 5 MHz. Linear antenna arrays are as-
sumed at the mobile terminals and at the base station. An
5 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-
0 lengths is selected at the base station, which results in a cor-
0 1 2 3 4 5 6 7
relation smaller than 0.1 between the antennas. We assume
log2 (spreading factor)
a six-sector antenna at the base station and omnidirectional
SC-CDMA antennas at the mobile terminals. The propagation environ-
SC-BS-CDMA ment is characterized by a specular multipath composed of
MC-CDMA
MC-BS-CDMA
6 paths (each path represents the reflection on a scatterer)
and 10 subpaths per path (each subpath represents the re-
Figure 4: PAPR as a function of the spreading factor, QPSK, 128 flection on a specific part of the scatterer). For each path, the
carriers. excess delay standard deviation is equal to 238 nanoseconds,
the angle spread standard deviation at the base station and
a PAPR close to 0 dB. MC-CDMA has a PAPR improving for that at the mobile terminals are equal to 6 and 68 degrees, re-
an increasing spreading factor due to the fact that the total spectively. For each subpath, the angle spread standard vari-
number of possible chip combinations at the output of the ation at the base station and that at the mobile terminals are
spreading operation is decreasing. The use of the SC-based equal to 2 and 5 degrees, respectively. The computation of the
communication modes is encouraged in the uplink to re- path loss is based on the COST231 Hata urban propagation
duce the power amplifier backoff at the mobile terminal and, model.
thus, to increase its power efficiency (even if the power am- Monte Carlo simulations have been performed to aver-
plifier backoff is not directly settled based on the low prob- age the bit error rate (BER) over 500 stochastic channel re-
ability power peak values in the MC-based systems, it is ex- alizations, and to compute the corresponding goodput, de-
pected that the variation in instantaneous transmitted power fined as the throughput offered to the user assuming a re-
is much lower in the SC-based systems). transmission of the erroneous packets until they are correctly
received. The information bandwidth is spread by the

spreading factor equal to 8. The user signals are spread by pe- 2
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- 1.5
Goodput (Mbps)
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 1
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 0.5
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 0
the uplink, transmit power control is applied such that the 0 5 10 15 20 25 30
received symbol energy is constant for all users. The power Eb /N0 (dB)
transmitted by each terminal depends on the actual channel QPSK, 1/2 16-QAM, 3/4
experienced by it. The BER (or the goodput) is determined QPSK, 2/3 64-QAM, 1/2
as a function of the received bit energy, or, equivalently, as QPSK, 3/4 64-QAM, 2/3
16-QAM, 1/2 64-QAM, 3/4
a function of the transmit power averaged over the differ-
16-QAM, 2/3
ent channel realizations. In the downlink, no transmit power
control is applied. For a constant transmit power at the base Figure 5: Tradeoff between channel coding rate and constellation,
station, the received symbol energy at each terminal depends static downlink, MCBS-CDMA, using MRC.
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- robustness due to the use of a higher constellation. As it will
ergy, the required transmit powers for the different commu- be shown later, the conclusion may be reversed in a mobile
nication modes appear to be very similar. environment due to better robustness of lower-order map-
Rather than comparing all possible modes for the two pings to changing channel conditions.
link directions, we only consider the relevant modes for each Figures 6 and 7 illustrate the gain obtained by the use of
direction. For the uplink, the SC modes demonstrate two different multiple-antenna techniques in the downlink of an
pronounced advantages compared to MC modes [3]. First, MCBS-CDMA-based static system and in the uplink of an
the SC modes exhibit a smaller PAPR than the MC modes, SCBS-CDMA-based static system, respectively. Four differ-
which leads to increased terminal power efficiency. Second, ent system configurations are considered (1×1, 1×2, 2×1, 2×
the SC modes allow to move the IFFT at the transmitter to 2). A typical user’s load is assumed (number of users equal to
the receiver, which results in reduced terminal complexity. 5). Two optimal combinations of the constellation and chan-
For the downlink, the MC modes are the preferred modula- nel coding rate are considered (16-QAM with coding rate
tion schemes, since they only incur a single FFT operation at 1/2, 64-QAM with coding rate 3/4). Exploiting the spatial
the receiver side which also leads to reduced terminal com- diversity at the mobile terminal or at the base station by the
plexity. use of two antennas (1 × 2 or 2 × 1 configurations), MRC
Figure 5 illustrates the user’s goodput in the downlink reception, or STBC transmission enables a significant gain in
of a MCBS-CDMA-based static communication system for transmit power (up to 12 dB gain is achieved in the down-
different combinations of the constellation size and channel link, up to 7 dB gain is achieved in the uplink). However, the
coding rate. The signal is received at the terminal through 2 additional gain obtained by combining STBC transmission
antennas. A typical user’s load of 5 users has been assumed. with MRC reception (2 × 2 configuration) is relatively small
Looking to those combinations that give the same asymptotic (less than 2 dB). SDM suffers from a high 8 dB loss in the
goodput (16-QAM and CR 3/4 compared to 64-QAM and goodput regions that can also be reached by a single-antenna
CR 1/2), it is always preferable to combine a high constel- system (this loss can be reduced if more complex nonlinear
lation size with a low code rate. The same conclusion holds receivers are considered). One can only achieve an increase of
in order to select the combination constellation-coding rate capacity by the use of multiple antennas at very high signal-
that maximizes the goodput for each SNR value. The enve- to-noise ratio (SNR) values. As a result, we advise to use one
lope is obtained by progressively employing QPSK, 16-QAM, directive antenna at the base station to exploit the very small
64-QAM constellations and a code rate equal to 1/2, and then angle spread and increase the cell capacity, and two diversity
by increasing the code rate progressively to the values 2/3, antennas at the mobile terminal to improve the link reliabil-
3/4 while keeping the constellation size fixed to 64-QAM. ity. The receive MRC technique is performed in the down-
The gain in communication robustness provided by the use link, while the STBC coding scheme is applied in the up-
of more channel coding exceeds the loss in communication link.
3
2
1.8
2.5
1.6
Goodput (Mbps)
2 1.4
Goodput (Mbps)
1.2
1.5 1
0.8
1
0.6
0.5 0.4
0.2
0 0
0 5 10 15 20 25 30 35 0 5 10 15 20
SISO 1 × 1 STBC 2 × 2 IA 1 user IA 8 users
MRC 1 × 2 SDM 2 × 2 IA 5 users IE x users
STBC 2 × 1
Figure 6: Multiple-antenna gain in the downlink, static environ- Figure 8: Impact of the user load on the downlink, static environ-
ment, MCBS-CDMA; dashed curves: 16-QAM and coding rate 1/2, ment, MCBS-CDMA (IE stands for interblock spreading) and MC-
solid curves: 64-QAM and coding rate 3/4. 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.
3
2
1.8
2.5
1.6
2 1.4
Goodput (Mbps)
Goodput (Mbps)
1.2
1.5 1
0.8
1
0.6
0.4
0.5
0.2
0 0
0 5 10 15 20 25 30 35 0 5 10 15 20
SISO 1 × 1 STBC 2 × 2 IA 1 user IA 8 users

MRC 1 × 2 SDM 2 × 2 IA 5 users IE x users
STBC 2 × 1
Figure 9: Impact of the user load on the uplink, static environment,
Figure 7: Multiple-antenna gain in the uplink, static environment, SCBS-CDMA (IE stands for interblock spreading) and SC-CDMA
SCBS-CDMA; dashed curves: 16-QAM and coding rate 1/2, solid (IA stands for intrablock spreading), using STBC; dashed curves:
curves: 64-QAM and coding rate 3/4. 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- single-user receiver, which performs channel-independent
tivity to the user’s load assuming static channels. Interblock block despreading followed by MMSE single-user equaliza-
spreading (MCBS-CDMA and SCBS-CDMA) is compared tion. MCBS-CDMA and SCBS-CDMA are MUI-free trans-
to intrablock spreading (MC-CDMA and SC-CDMA). The mission schemes, such that their user’s goodput remains un-
number of users ranges from 1 to 8. Again the same two com- affected by the user’s load. In case of intrablock spreading,
binations of the constellation and channel coding rate have the MMSE multiuser receiver outperforms the single-user
been selected (16-QAM with coding rate 1/2, 64-QAM with detector. In the downlink, the performance of the MMSE
coding rate 3/4). In case of interblock spreading, the MMSE multiuser joint detector is slightly decreasing for an in-
multiuser receiver reduces to an equivalent but simpler creasing number of users and converges to the one of the
2 2
1.8 1.8
1.6 1.6
1.4 1.4
Goodput (Mbps)
Goodput (Mbps)
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 50 100 150 200 250 0 50 100 150 200 250
Speed (km/h) Speed (km/h)
16-QAM, 1/2, IA, SNR = 8 dB 16-QAM, 1/2, IA, SNR = 8 dB

16-QAM, 1/2, IE, SNR = 8 dB 16-QAM, 1/2, IE, SNR = 8 dB
64-QAM, 3/4, IA, SNR = 16 dB 64-QAM, 3/4, IA, SNR = 16 dB
64-QAM, 3/4, IE, SNR = 16 dB 64-QAM, 3/4, IE, SNR = 16 dB
Figure 10: Impact of the terminal speed on the downlink, MCBS- Figure 11: Impact of the terminal speed on the uplink, SCBS-
CDMA (IE stands for interblock spreading) and MC-CDMA (IA CDMA (IE stands for interblock spreading) and SC-CDMA (IA
stands for intrablock spreading), using MRC. stands for intrablock spreading), using STBC.
single-detector at full user load. MMSE multiuser reception be used for higher speeds. This conclusion is in line with
is especially needed in the uplink, since a single-user receiver the two-dimensional spreading strategy proposed in [16, 17]
cannot get rid of the MUI, and features a BER curve flat- for a MC-based system, that prioritizes the spreading in the
tening already at low SNRs. The impact of the user’s load is time domain rather than in the frequency domain, for the
much more pronounced in the SC-CDMA uplink than in the sake of complexity and performance. It is also interesting to
MC-CDMA downlink since the signals propagate through note that the goodput achieved with high constellations at
different channels, which is more difficult to compensate for. high speed is smaller than the goodput achieved with low
In the downlink, MC-CDMA always outperforms MCBS- constellations.
CDMA since it benefits from the frequency diversity offered
by the CDMA spreading. In the uplink, the performance of
SCBS-CDMA is equivalent to the one of SC-CDMA for a typ- 6. CONCLUSIONS
ical user’s load, worse for a small user’s load and better for a A generic transmission scheme has been designed that al-
high user’s load. lows to instantiate all the combinations of OFDM and
Figures 10 and 11 compare the effect of the terminal cyclic-prefixed SC modulations with DS-CDMA. The SDM
speed on the user’s goodput in case of intrablock and in- and STBC multiple-antenna techniques have been inte-
terblock spreading. A typical user’s load of 5 users has grated in the generic transmission scheme. For each resulting
been assumed, which makes MC-CDMA (SC-CDMA) and mode, the optimal linear MMSE multiuser receiver has been
MCBS-CDMA (SCBS-CDMA) perform equally well in static derived.
conditions. For each combination of the constellation and A mode selection strategy has also been proposed that
coding rate, the SNR value corresponding to the maximum trades off efficiently the communication performance in a
slope in the goodput curves has been chosen (realistic work- typical suburban dynamic outdoor environment with the
ing point in the curves illustrated in Figures 8 and 9). Since complexity and PAPR at the mobile terminal.
the symbol block duration is higher in case of interblock (i) A hybrid modulation scheme (MC in the downlink,
spreading than in case of intrablock spreading, the perfor- SC in the uplink) should be used in order to minimize the
mance of MCBS-CDMA (SCBS-CDMA) is significantly re- mobile terminal PAPR and data processing power.
duced, while the performance of MC-CDMA (SC-CDMA) (ii) Under low-to-medium mobility conditions, it is bet-
remains acceptable when the speed of the mobile terminals ter to use a high constellation and a low channel coding rate
increases. Since the orthogonality between the users is lost to achieve the maximum goodput for a given SNR value.
when the speed increases, the impact is more severe in the However, the lowest constellation order should be selected
uplink than in the downlink. If a 10-percent performance under high mobility conditions.
loss is acceptable, interblock spreading should be used up to (iii) One directive antenna should be used at the base
60 km/h in the downlink or 10 km/h in the uplink for detec- station to increase the cell capacity and multiple antennas
tion complexity reasons, while intrablock spreading should should be used at the mobile terminal in combination with
diversity techniques like MRC reception and STBC transmis- [14] S. Zhou, G. B. Giannakis, and C. Le Martret, “Chip-
sion to improve the link reliability. interleaved block-spread code division multiple access,” IEEE
(iv) Since interblock and intrablock spreading perform Trans. Commun., vol. 50, no. 2, pp. 235–248, 2002.
equally well in typical user loads, interblock spreading should [15] F. Petré, G. Leus, L. Deneire, and M. Moonen, “Downlink
frequency-domain chip equalization for single-carrier block
be used at low terminal speeds to minimize the data pro- transmission DS-CDMA with known symbol padding,” in
cessing complexity while intrablock spreading should only be Proc. IEEE Global Telecommunications Conference (GLOBE-
used at high terminal speeds. COM ’02), vol. 1, pp. 453–457, Taipei, Taiwan, November
It has been demonstrated that an adaptive transceiver is 2002.
interesting to support different communication modes and [16] H. Atarashi, N. Maeda, S. Abeta, and M. Sawahashi, “Broad-
to efficiently track the changing communication conditions. 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
REFERENCES (PIMRC ’02), vol. 3, pp. 992–997, Lisbon, Portugal, Septem-
[1] T. Ojanperä and R. Prasad, Wideband CDMA for Third Gen- ber 2002.
eration Mobile Communications, Artech House, Norwood, [17] N. Chapalain, D. Mottier, and D. Castelain, “An OFCDM up-
Mass, USA, 1998. link transmission system with channel estimation based on
[2] R. van Nee, G. Awater, M. Morikura, H. Takanashi, M. Web- spread pilots,” in Proc. IST Mobile and Wireless Communica-
ster, and K. W. Halford, “New high-rate wireless LAN stan- tions Summit, Lyon, France, June 2004.
dards,” IEEE Commun. Mag., vol. 37, no. 12, pp. 82–88, 1999. [18] G. J. Foschini and M. J. Gans, “On limits of wireless commu-
[3] D. Falconer, S. L. Ariyavisitakul, A. Benyamin-Seeyar, and nications in a fading environment when using multiple an-
B. Eidson, “Frequency domain equalization for single-carrier tennas,” Kluwer Journal of Wireless Personal Communications,
broadband wireless systems,” IEEE Commun. Mag., vol. 40, vol. 6, no. 3, pp. 311–335, 1998.
no. 4, pp. 58–66, 2002. [19] G. G. Raleigh and J. M. Cioffi, “Spatio-temporal coding
[4] N. Yee, J.-P. Linnartz, and G. Fettweis, “Multicarrier CDMA for wireless communication,” IEEE Trans. Commun., vol. 46,
in indoor wireless radio networks,” in Proc. IEEE International no. 3, pp. 357–366, 1998.
Symposium on Personal, Indoor and Mobile Radio Communi- [20] D. Gesbert, H. Bolcskei, D. A. Gore, and A. J. Paulraj, “Out-
cations (PIMRC ’93), vol. 1, pp. 109–113, Yokohama, Japan, door MIMO wireless channels: models and performance pre-
September 1993. diction,” IEEE Trans. Commun., vol. 50, no. 12, pp. 1926–
[5] K. Fazel, “Performance of CDMA/OFDM for mobile commu- 1934, 2002.
nication system,” in Proc. 2nd IEEE International Conference [21] A. J. Paulraj and T. Kailath, “Increasing capacity in wireless
on Universal Personal Communications (ICUPC ’93), vol. 2, broadcast systems using distributed transmission/directional
pp. 975–979, Ottawa, Ontario, Canada, October 1993. reception (DTDR),” U.S. Patent 5345599, Stanford University,
[6] K. Fazel, S. Kaiser, and M. Schnell, “A flexible and high Stanford, Calif, USA, 1994.
performance cellular mobile communications system based [22] G. J. Foschini, “Layered space-time architecture for wireless
on multi-carrier SSMA,” Kluwer Journal of Wireless Personal communication in a fading environment when using multiple
Communications, vol. 2, no. 1/2, pp. 121–144, 1995. antennas,” Bell Labs Technical Journal, vol. 1, no. 2, pp. 41–59,
[7] S. Kaiser, “OFDM code-division multiplexing in fading chan- 1996.
nels,” IEEE Trans. Commun., vol. 50, no. 8, pp. 1266–1273, [23] F. W. Vook, T. A. Thomas, and K. L. Baum, “Transmit array
2002. processing for cyclic-prefix CDMA,” in Proc. 56th IEEE Ve-
[8] V. M. DaSilva and E. S. Sousa, “Multicarrier orthogonal hicular Technology Conference (VTC ’02), vol. 1, pp. 270–274,
CDMA signals for quasi-synchronous communication sys- Vancouver, British Columbia, Canada, September 2002.
tems,” IEEE J. Select. Areas Commun., vol. 12, no. 5, pp. 842– [24] V. Tarokh, N. Seshadri, and A. R. Calderbank, “Space-time
852, 1994. codes for high data rate wireless communication: perfor-
[9] S. Kondo and L. B. Milstein, “Performance of multicarrier mance criterion and code construction,” IEEE Trans. Inform.
DS-CDMA systems,” IEEE Trans. Commun., vol. 44, no. 2, Theory, vol. 44, no. 2, pp. 744–765, 1998.
pp. 238–246, 1996. [25] S. M. Alamouti, “A simple transmit diversity technique for
[10] L.-L. Yang and L. Hanzo, “Multicarrier DS-CDMA: a multiple wireless communications,” IEEE J. Select. Areas Commun.,
access scheme for ubiquitous broadband wireless communi- vol. 16, no. 8, pp. 1451–1458, 1998.
cations,” IEEE Commun. Mag., vol. 41, no. 10, pp. 116–124, [26] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, “Space-time
2003. block codes from orthogonal designs,” IEEE Trans. Inform.
[11] F. Petré, G. Leus, M. Moonen, and H. De Man, “Multi- Theory, vol. 45, no. 5, pp. 1456–1467, 1999.
carrier block-spread CDMA for broadband cellular down- [27] E. Lindskog and A. J. Paulraj, “A transmit diversity scheme for
link,” EURASIP Journal on Applied Signal Processing, vol. 2004, channels with intersymbol interference,” in Proc. IEEE Inter-
no. 10, pp. 1568–1584, 2004, Special Issue on Multi-Carrier national Conference on Communications (ICC ’00), vol. 1, pp.
Communications and Signal Processing . 307–311, New Orleans, La, USA, June 2000.
[12] M. Vollmer, M. Haardt, and J. Gotze, “Comparative study of [28] N. Al-Dhahir, “Single-carrier frequency-domain equalization
joint-detection techniques for TD-CDMA based mobile ra- for space-time block-coded transmissions over frequency-
dio systems,” IEEE J. Select. Areas Commun., vol. 19, no. 8, selective fading channels,” IEEE Commun. Lett., vol. 5, no. 7,
pp. 1461–1475, 2001. pp. 304–306, 2001.
[13] K. L. Baum, T. A. Thomas, F. W. Vook, and V. Nangia, [29] S. Zhou and G. B. Giannakis, “Single-carrier space-time
“Cyclic-prefix CDMA: an improved transmission method for block-coded transmissions over frequency-selective fading
broadband DS-CDMA cellular systems,” in Proc. IEEE Wire- channels,” IEEE Trans. Inform. Theory, vol. 49, no. 1, pp. 164–
less Communications and Networking Conference (WCNC ’02), 179, 2003.
vol. 1, pp. 183–188, Orlando, Fla, USA, March 2002. [30] S. Barbarossa and F. Cerquetti, “Simple space-time coded
SS-CDMA systems capable of perfect MUI/ISI elimination,” Frederik Petré was born in Tienen, Bel-
IEEE Commun. Lett., vol. 5, no. 12, pp. 471–473, 2001. gium, on December 12, 1974. He received
[31] F. W. Vook, T. A. Thomas, and K. L. Baum, “Cyclic-prefix the Electrical Engineering degree and the
CDMA with antenna diversity,” in Proc. 55th IEEE Vehicu- Ph.D. degree in applied sciences from the
lar Technology Conference (VTC ’02), vol. 2, pp. 1002–1006, Katholieke Universiteit Leuven (KULeu-
Birmingham, Ala, USA, May 2002. ven), Leuven, Belgium, in July 1997 and
[32] F. Petré, G. Leus, L. Deneire, M. Engels, M. Moonen, and December 2003, respectively. In September
H. De Man, “Space-time block coding for single-carrier 1997, he joined the Design Technology for
block transmission DS-CDMA downlink,” IEEE J. Select. Ar- Integrated Information and Communica-
eas Commun., vol. 21, no. 3, pp. 350–361, 2003, Special Issue tion Systems (DESICS) Division at the In-
on MIMO Systems and Applications . teruniversity Micro-Electronics Center (IMEC) in Leuven, Bel-
[33] Z. Liu and G. B. Giannakis, “Space-time block-coded multi- gium. Within the Digital Broadband Terminals (DBATE) Group
ple access through frequency-selective fading channels,” IEEE of DESICS, he first performed predoctoral research on wireline
Trans. Commun., vol. 49, no. 6, pp. 1033–1044, 2001.
transceiver design for twisted pair, coaxial cable, and powerline
[34] A. Stamoulis, Z. liu, and G. B. Giannakis, “Space-time block-
communications. During the fall of 1998, he visited the Informa-
coded OFDMA with linear precoding for multirate services,”
IEEE Trans. Signal Processing, vol. 50, no. 1, pp. 119–129, 2002. tion Systems Laboratory (ISL) at Stanford University, Calif, USA,
[35] Z. Wang and G. B. Giannakis, “Wireless multicarrier commu- working on OFDM-based powerline communications. In January
nications,” IEEE Signal Processing Mag., vol. 17, no. 3, pp. 29– 1999, he joined the Wireless Systems (WISE) Group of DESICS
48, 2000. as a Ph.D. researcher, funded by the Institute for Scientific and
[36] G. B. Giannakis, P. A. Anghel, and Z. Wang, “Wideband gen- Technological Research in Flanders (IWT). Since January 2004, he
eralized multicarrier CDMA over frequency-selective wire- has been a Senior Scientist within the Wireless Research Group of
less channels,” in Proc. IEEE Int. Conf. Acoustics, Speech, Sig- DESICS. He is investigating the baseband signal processing algo-
nal Processing (ICASSP ’00), vol. 5, pp. 2501–2504, Istanbul, rithms and architectures for future wireless communication sys-
Turkey, June 2000. tems, like third-generation (3G) and fourth-generation (4G) cel-
[37] G. B. Giannakis, Z. Wang, A. Scaglione, and S. Bar- lular networks, and wireless local area networks (WLANs). His
barossa, “AMOUR-generalized multicarrier transceivers for main research interests are in modulation theory, multiple-access
blind CDMA regardless of multipath,” IEEE Trans. Commun., schemes, channel estimation and equalization, smart antenna, and
vol. 48, no. 12, pp. 2064–2076, 2000. MIMO techniques. He is a Member of the ProRISC Technical Pro-
[38] Q. Chen, E. S. Sousa, and S. Pasupathy, “Performance of a gram Committee and the IEEE Benelux Section on Communica-
coded multi-carrier DS-CDMA system in multipath fading tions and Vehicular Technology (CVT). He is a Member of the Ex-
channels,” Kluwer Journal of Wireless Personal Communica- ecutive Board and Project Leader of the Flexible Radio Project of
tions, vol. 2, no. 1-2, pp. 167–183, 1995. the Network of Excellence in Wireless Communications (NEW-
[39] A. Klein, G. K. Kaleh, and P. W. Baier, “Zero forcing and mini- COM), established under the sixth framework of the European
mum mean-square-error equalization for multiuser detection
Commission.
in code-division multiple-access channels,” IEEE Trans. Veh.
Technol., vol. 45, no. 2, pp. 276–287, 1996. Eduardo Lopez-Estraviz was born in Fer-
[40] L. Vandendorpe, “Performance analysis of IIR and FIR lin- rol, Spain, in 1977. He received the
ear and decision-feedback MIMO equalizers for transmulti- Telecommunications Engineering degree
plexers,” in Proc. IEEE International Conference on Commu-
from the Universidad de Vigo (UVI), Vigo,
nications (ICC ’97), vol. 2, pp. 657–661, Montreal, Quebec,
Spain, in 1996 and 2002, respectively. In
Canada, June 1997.
[41] 3GPP Technical specification group, “TR25.996 v6.1.0, Spatial June 2002, he joined the Interuniversity
channel model for Multiple Input Multiple Output (MIMO) Micro-Electronics Center (IMEC), Leuven,
simulations,” September 2003. Belgium, as a researcher. His fields of inter-
est are in digital signal processing for broad-
band communications. He is now working
François Horlin was born in Bruxelles, Bel- on the integration of a multistandard mobile terminal. In this
gium, in 1975. He received the Electri- context, he is working on developing functionality for a fourth-
cal Engineering degree and the Ph.D. de- generation (4G) cellular system, targeting high capacity in very
gree from the Université catholique de Lou- high mobility conditions.
vain (UCL), Louvain-la-Neuve, Belgium, in
1998 and 2002, respectively. In September Frederik Naessens was born in Roeselare,
2002, he joined the Interuniversity Micro- Belgium, in 1979. He received the Engi-
Electronics Center (IMEC), Leuven, Bel- neering degree from the Catholic School for
gium, as a Senior Researcher. He is currently Higher Education Bruges-Ostend (KHBO)
the Head of the research activities on digital Academy in Ostend, Belgium, in 2001. Af-
signal processing for broadband communications. His fields of inter this studies, he joined the Interuniversity
terest are in digital signal processing, mainly for high-bit-rate mul- Micro-Electronics Center (IMEC), Leuven,
tiuser communications. During his Ph.D. thesis, he proposed a new Belgium, as a development engineer. Cur-
solution for jointly optimizing the transmitter and the receiver in a rently he is also a part-time student follow-
multiuser multi-input multi-output (MIMO) type of communica- ing an M.S. degree in ICT at the University
tion systems. He is now working on the integration of a multistan- of Kent, United Kingdom. He has been involved in a demonstra-
dard mobile terminal. In this context, he is responsible for the de- tor setup for GPS and GLONASS satellite navigation systems. He
velopment of the functionality of a fourth-generation (4G) cellular is now working on developing functionality for future cellular sys-
system, targeting high capacity in very high mobility conditions. tems (4G).
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
IMEC’s OFDM ASICs development and a
Project Leader for the turbo codec. Cur-
rently, she is leading the Software-Defined
Radio Baseband Project, and she is the Scientific Director of Wire-
less Research in IMEC.
c 2005 Ying-Chang Liang et al.
Reconfigurable 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
Email: naveen@i2r.a-star.edu.sg
Santosh K. Pilakkat
Email: pilakkat@i2r.a-star.edu.sg
Ashok K. Marath
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 reconfigured to any type of cyclic-prefix (CP) -based com-
munication systems, including orthogonal frequency-division multiplexing (OFDM), single-carrier cyclic-prefix (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 identified. 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 reconfigurable transceiver onto the proposed hardware platform. The different functional entities which will
be required to perform the reconfiguration and realize the transceiver are explained.
Keywords and phrases: reconfigurable signal processing, broadband communications, software-defined radio, cyclic prefix,
frequency-domain equalization.
1. INTRODUCTION cognitive radios (CRs) [2] are the well-suited implementa-

tions of such network independent radios.
A number of wireless standards govern personal wireless
Different frame format and modulation schemes have
communications, to name a few, GSM and CDMA for 2G
been proposed for different networks. For example, orthog-
cellular networks; WCDMA, CDMA-2000, and TDS-CDMA
onal frequency-division multiplexing (OFDM) [3] has been
for 3G cellular networks; IEEE 802.11a/b/g for wireless lo-
adopted in IEEE 802.11a and IEEE 802.11g, and IEEE 802.16.
cal area networks; IEEE 802.16 for wireless wide area net-
Single-carrier cyclic-prefix (SCCP) system is selected for
works (WiMAX); and IEEE 802.15 for personal area net-
IEEE 802.16 as well. Cyclic-prefix (CP) -based CDMA sys-
works (PAN). In order to satisfy the need of customers’ mo-
tems are considered for beyond 3G systems. For example,
bility, the designed radio transceivers should not be tied to
CP-CDMA and CP-based direct-sequence CDMA (CP-DS-
any specific network. Software-defined radios (SDRs) [1] or
CDMA) are candidate schemes for enhanced versions of 3G
DS-CDMA systems [4, 5, 6, 7]; multicarrier CDMA (MC-
This is an open access article distributed under the Creative Commons CDMA) and multicarrier direct-sequence CDMA (MC-DS-
Attribution License, which permits unrestricted use, distribution, and CDMA) [8] are being extensively studied for potential adop-
reproduction in any medium, provided the original work is properly cited. tion in beyond 3G (B3G) or fourth-generation (4G) cellular
systems [10, 21]. The common feature of these systems is data sub-block of N symbols. Thanks to the use of CP, the
that they are block-based transmission systems, and a CP channel matrix is a circular matrix, which can be represented
portion is inserted to each data block in order to suppress as H = WH ΛW, where W is the N-point DFT matrix, and
the interblock interference (IBI) and to simplify the receiver Λ = diag{λ1 , . . . , λN } is a diagonal matrix, the elements of
design. SDR structures and implementations have been ex- which are the channel’s frequency-domain responses in each
plored in [11, 12, 13, 14, 15, 16, 17, 18, 19] in different ways. subcarrier.
In this paper, we propose a universal mobile transceiver For multicarrier systems, the data symbol sub-block is
which is configurable to any type of CP-based systems. We pretransformed using the N-point IDFT matrix WH ; while
present a hardware platform for the proposed transceiver for single-carrier systems, the data symbol sub-block is sent
and identify the reusable common blocks. In particular, a out directly. At the receiver side, both single-carrier and mul-
study of the equalizer design at the mobile terminals is car- ticarrier systems first pretransform the received data block
ried out for each system. Single-user environments are con- with the DFT matrix W, the output of which is then utilized
sidered for OFDM and SCCP systems. For CDMA mul- to recover the transmitted signals via either linear or nonlin-
tiuser systems, we target a receiver architecture for the down- ear receivers.
link only. It is found that after block despreading operation, For the multiuser CDMA case, spreading and despread-
MC-DS-CDMA and CP-DS-CDMA have the same equal- ing are added in the transmitter and receiver, respectively.
ization blocks as OFDM and SCCP systems, respectively, Since the transceiver design for the mobile terminal is of in-
showing that hardware and software sharing is possible for terest, for the CDMA multiuser case, we concentrate on the
these systems. More specifically, denoting N and G as the receiver architecture for the downlink only. For OFDM and
block size and processing gain, respectively, these two sys- SCCP systems, single-user environments are considered.
tems perform signal detection of N symbols over G consec-
utive time blocks. However, MC-CDMA and CP-CDMA re- 2.1. OFDM
quire only one time block to perform signal detection. The For OFDM systems, the input-output relation at the nth time
receiver complexity of MC-CDMA and CP-CDMA can be block after FFT can be expressed as [3]
much higher if near ML detection is required [5] due to the
requirement of N-dimensional signal detection within each y(n) = Λs(n) + u(n), (1)
time block.
In [11], a reconfigurable architecture for multicarrier where s(n) is the N × 1 input signal vector, y(n) is the N × 1
based CDMA systems is proposed. Specifically, the systems FFT output of the received signal vector, and u(n) is the FFT
discussed there include MC-CDMA, MC-DS-CDMA, and output of the AWGN vector, which is of dimension N × 1.
MT-CDMA of [8] only. The architecture proposed in our pa-
per is more generic in the sense that it can be reconfigured 2.2. SCCP systems
to support both multicarrier systems and single-carrier sys- For SCCP systems, the input-output relation at the nth block
tems, including SCCP, CP-CDMA, and CP-DS-CDMA. Fur- after FFT can be expressed as [20]
thermore, we propose a hardware platform to support the re-
configurable architecture. An attempt has been made to map y(n) = ΛWs(n) + u(n), (2)
the functional reconfigurable transceiver onto the proposed
hardware platform. The different functional entities which where s(n), y(n), and u(n) are defined as in OFDM systems.
will be required to perform the reconfiguration and realize Next, we look at four CDMA-based systems: MC-DS-
the transceiver are explained. CDMA, MC-CDMA, CP-CDMA, and CP-DS-CDMA. Sup-
This paper is organized as follows. In Section 2, the pose all users have the same processing gain G. We introduce
system models are described for various CP-based sys- the following common notations for all systems: T for to-
tems. In Section 3, the linear MMSE receivers are specified tal number of users; D(n) for long scrambling codes at the
for different systems. Section 4 proposes the reconfigurable nth block, where D(n) = diag{d(n; 0), . . . , d(n; N − 1)}; ci
transceiver and details the configurations for each system. for short codes of user i, ci = [ci (0), . . . , ci (G − 1)]T , with
We also analyze the complexity of the equalizer for each sys- cH H
i c j = 1 for i = j, and ci c j = 0 for i = j.
tem. The hardware architecture for the proposed transceiver
is proposed in Section 5. Finally, conclusions are drawn in 2.3. MC-DS-CDMA
Section 6.
MC-DS-CDMA performs time-domain spreading [8] as fol-
lows.
2. SYSTEM DESCRIPTION (1) N symbols from each user-specific spreading codes
are taken.
In this section, we review the input-output relations for the (2) The chip signals corresponding to the same symbol
CP-based systems, which serve as the basis for the design of index from all users are summed and transmitted over the
reconfigurable signal processing and hardware architecture. same subcarrier but through different times blocks. Thus it
The transmission is on a block-by-block basis, with each takes G consecutive blocks to transmit out the whole chip
block consisting of the CP sub-block of P symbols and the sequences of the N symbols.
Reconfigurable Signal Processing for Wireless Communications 325
Mathematically, the nth received block after FFT where

operation can be written as T
y(n) = y(n; 0), . . . , y(n; N − 1) ,
−1
T
D(n) = diag d(n; 0), . . . , d(n; N − 1) ,
y(n) = ΛD(n) (n),
ci (n)si + u (3) (10)
i=0 C = diag C̄, . . . , C̄ ,
T
s(n) = s̄T1 (n), . . . , s̄TM (n) ,
where
T and u(n) is the FFT output of the AWGN vector. Further,
y(n) = y(n; 0), . . . , y(n; N − 1) ,

Λ = diag λ0 , . . . , λN −1 , C̄ = c0 , . . . , cT −1 ,
T (11)
D(n) = diag d(n; 0), . . . , d(n; N − 1) , (4) s̄i (n) = s0 (n; i), . . . , sT −1 (n; i) .
T
si = si (0), . . . , si (T − 1) , Dividing y(n) into M nonoverlapping short-column vectors,
T each with G elements, (9) can then be decoupled as
(n) = u(n; 0), . . . , u(n; N − 1)
u .
Note that u (n) is the FFT output of the AWGN vector. On yi (n) = Λi Di (n)C̄s̄i (n) + ui (n) (12)
the other hand, gathering the received signals of the kth sub-
carrier from 0th block to (G − 1)th block yields for i = 1, . . . , M, where Λi and Di (n) are the ith sub-blocks
of Λ and D(n). From (12), MC-CDMA experiences MAI, but

M not ISI.
y(k) = λk
si (k)D(k)ci + ũ(k), (5)
i=0 2.5. CP-CDMA
CP-CDMA is a single-carrier dual of MC-CDMA. The M =
where N/G symbols of each user are first spread out with user-
T specific spreading codes, then the chip sequence for all users
y(k) = y(0; k), . . . , y(G − 1; k) , are summed up; the total chip signal of size N is then passed
to CP inserter, which adds a CP. Using the duality between

D(k) = diag d(0; k), . . . , d(G − 1; k) , (6)
CP-CDMA and MC-CDMA, from (12), the nth received
T block of CP-CDMA after FFT can be written as [5]
ũ(k) = u(0; k), . . . , u(G − 1; k) .

Despreading y(k) using D(k)ci , we obtain y(n) = ΛWD(n)Cs(n) + u(n). (13)
zi (k) = cH H (k)
i D (k)y From the above, it is seen that CP-CDMA experiences both
(7) MAI and ISI.
= λk si (k) + u(k).
Thus we have 2.6. CP-DS-CDMA

CP-DS-CDMA is the single-carrier dual of MC-DS-CDMA.
zi = Λsi + u
i, (8) It performs block spreading as follows: the N symbols of each
user are first spread out with its own spreading codes, then
the chip sequence for all users are summed up; the total chip
where zi = [zi (0), . . . , zi (N − 1)]T , and ui is defined accord- signal corresponding to different chip indices is transmitted
ingly. Therefore, after block despreading, an MC-DS-CDMA over different time block, thus it takes G blocks to send out
is equivalent to an OFDM, and it does not experience any the whole chip sequence of the N symbols.
MAI and ISI. The nth received block after FFT can be written as [5]
2.4. MC-CDMA −1
T
y(n) = ΛWD(n) ci (n)si + u(n), (14)
MC-CDMA performs spreading over frequency domain. De- i=0
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
where n = 0, . . . , G − 1.
as
Simplification is available if the same set of long scram-
bling codes are chosen for blocks from 0 to (G − 1), or if long
y(n) = ΛD(n)Cs(n) + u(n), (9) scrambling codes are not used (D(n) = I). We consider the
case when long scrambling codes are not used, and collect 3.1. MC-DS-CDMA/OFDM
the kth subcarrier’s received signals from the 0th to (G − 1)th
An MC-DS-CDMA after block despreading is equivalent to
blocks, we obtain
an OFDM. Thus both can employ the same one-tap equalizer
to retrieve the transmitted signals: (σs2 [σs2 |Λ|2 + σn2 I]−1 Λ)H .
−1
T
Note that MC-DS-CDMA performs symbol-level equaliza-
y(k) = λk W(k, :)si ci + ũ(k), (15)
i=0
tion.
where W(k, :) denotes the kth row of matrix W, 3.2. MC-CDMA
T For MC-CDMA, the signal separation is done for each sub-

y(k) = y(0; k), . . . , y(G − 1; k) , block. The channel matrix for the ith sub-block of the nth
T block is
ci = ci (0), . . . , ci (G − 1) , (16)
T
ũ(k) = u(0; k), . . . , u(G − 1; k) . Hi = Λi Di (n)C̄. (22)
Using ci to perform despreading, we obtain Thus the MMSE equalizer is given by
zi (k) = cH (k)
i y 2
−1
= λk W(k, :)si + u
(k).
(17) Pi = σs2 σs2 Λi + σn2 I Λi Di (n)C̄ (23)
Thus we have for i = 1, . . . , Q, which can realized through a one-tap equal-

izer, (σs2 [σs2 |Λ|2 +σn2 I]−1 Λ)H , followed by multiple despread-
zi = ΛWsi + ui , (18) ers, [Di (n)C̄]H , i = 1, . . . , Q. Note that MC-CDMA performs
chip-level equalization.
which is the single-carrier duality of the MC-DS-CDMA
model (8). From (18), it is seen that CP-DS-CDMA does not 3.3. CP-CDMA
experience MAI, but it does have ISI. For CP-CDMA, the channel matrix for the nth block is given
by
3. LINEAR EQUALIZERS
H = ΛWD(n)C. (24)
We first look at the generic multiple-input multiple-output
(MIMO) channel model. Suppose the MIMO channel is de- Thus the MMSE equalizer is given by
scribed by the following input-output relation:
−1
P = σs2 σs2 |Λ|2 + σn2 I ΛWD(n)C (25)
x = Hs + n, (19)
which is realized through a one-tap equalizer, (σs2 [σs2 |Λ|2 +
where s, x, and n denote the transmitted signal vector, re-
σn2 I]−1 Λ)H , followed by an IDFT operator WH , followed by a
ceived signal vector, and received noise vector, respectively;
despreader [D(n)C]H . Note that CP-CDMA performs chip-
H is the channel matrix, representing the responses from the
level equalization.
transmit antennas to the receive antennas. Without loss of
generality, we assume that s, x, and n are all N × 1 vectors,
3.4. CP-DS-CDMA/SCCP
and H is an N × N matrix. Assume that E [ssH ] = σs2 I,
E [nnH ] = σn2 I, and E [snH ] = 0. Denote P as the linear A CP-DS-CDMA performs block despreading using G con-
equalizer for the channel (19), which generates the output secutive blocks, and after block despreading, it is equivalent
to a SCCP system both have the following channel matrix:
z = PH x. (20) H = ΛW. (26)
Then, the minimum mean square error (MMSE) equalizer is Thus the MMSE equalizer is given by
given by
−1
−1
P = σs2 σs2 |Λ|2 + σn2 I ΛW (27)
P = σs σs2 HHH + σn2 I
2
H. (21)
which is realized through a one-tap equalizer, (σs2 [σs2 |Λ|2 +
Now, we are ready to specify the linear receivers for each of σn2 I]−1 Λ)H , followed by a IDFT operator WH . Note that
the CP-based systems. chip-level equalization is performed for CP-DS-CDMA.
Channel
Source coding Constellation
Scrambling Burst formation demapping
Modem Receiver FE Modem
Channel Tx FE Descrambling
Tx processing Rx
coding processing Viterbi/turbo
Interleaving decoding
Pilot insertion Deinterleaving
Figure 1: Block diagram of the proposed transceiver configurable for CP-based systems.
Serial Parallel
to Code Block interleaving IFFT CP insertion to
parallel spreading mapping serial
Figure 2: Functional block of the reconfigurable Modem Tx.
Serial Block One-tap

to CP removal FFT IFFT Despreading
despread equalizer
parallel
Synchronization
Figure 3: Functional block of the reconfigurable Modem Rx.
4. PROPOSED TRANSCEIVER 4.2. System configurations

Based on the input-output relations and the detailed equal- The Modem Tx and Modem Rx are shown in Figures 2 and
izers presented in the previous section for each CP-based sys- 3, respectively, and are realized as parameterized functions.
tem, in this section, a universal transceiver is proposed which We define a flag parameter i(A) for block A; if i(A) = 1, the
is configurable to any type of these systems. block will be functioning; and if i(A) = 0, it will be unity.
We also use “CS” to denote code spreading block; “BI” for
4.1. Transceiver overview block interleaving block; “IFFT” for IFFT block; “FFT” for
The proposed transceiver is illustrated in Figure 1, which FFT block; “BDI” for block despreading and deinterleaving
contains source coding, scrambling, channel coding, interblock; “OTE” for one-tap equalizer; and “DS” for despread-
leaving, and pilot insertion blocks, as well as Modem Tx ing block.
and burst formation block at the transmitter side; and re- Denote G and N as the processing gain and the size of
ceiver front end (FE) and Modem Rx as well as constellation each block. Note that G = 1 for non-CDMA systems. The
demapper, descrambling, decoding, and deinterleaving at the Modem Tx takes in a set of M data symbols, where M = N
receiver side. The data symbols are passed to the reconfig- for OFDM, SCCP, MC-DS-CDMA, and CP-DS-CDMA, and
urable Modem Tx block where they are modulated accord- M = N/G for MC-CDMA and CP-CDMA.
ing to the required specification. The burst formation and The flag parameters and selection of M for Modem Tx
Tx front-end format the processed symbols to proper fram- are shown in Table 1 for different systems. For non-CDMA
ing structure with suitable preambles and they are modulated systems, the CS block is a unity function. The BI block works
onto a carrier and transmitted. After undergoing the channel for MC-DS-CDMA and CP-DS-CDMA, which translates the
and additive noise distortion, the received signal is down- spread output of size NG into G serial blocks, each with size
converted in the RX FE processing and fed to the config- N. The BI block may also work for MC-CDMA to achieve
urable Modem Rx, which does the synchronization (frame, almost equal diversity for each transmitted symbol [9]. The
code, and frequency), CP removal, FFT, channel estimation, IFFT processing block in Modem Tx is switched off in the
channel equalization, and phase compensation. The detected case of single-carrier systems (SCCP, CP-CDMA, and CP-
symbols are further processed in the subsequent block to DS-CDMA). A CP insertion block is added to remove IBI
generate the information bits. and to translate the channel matrix into a circular matrix.
In the reconfigurable transceiver, though each block has Finally, parallel-to-serial conversion is done for signal trans-
to be reconfigured to meet the particular standard’s require- mission.
ment, here we restrict our discussion to the Modem Tx and At Modem Rx, a common synchronization (SYN) block
Modem Rx blocks. is applied to each system. For-single-user case, this SYN
Table 1: Parameter configurations for Modem Tx. Table 3: Equalizer complexity per block for CDMA-based systems.
Schemes M i(CS) i(BI) i(IFFT) Schemes FFT BDI OTE IFFT DS

OFDM N 0 0 1 MC-DS-CDMA 1 N/G 1/G 0 0
SCCP N 0 0 0 MC-CDMA 1 0 1 0 N/G
MC-DS-CDMA N 1 1 1 CP-CDMA 1 0 1 1 N/G
MC-CDMA N/G 1 0/1 1 CP-DS-CDMA 1 N/G 1/G 1/G 0
CP-CDMA N/G 1 0 0
CP-DS-CDMA N 1 1 1
as follows:
(i) MC-DS-CDMA: N log2 (N) + 3(N/G);
Table 2: Parameter configurations for Modem Rx.
(ii) MC-CDMA: N log2 (N) + 3N;
Schemes i(BDI) i(OTE) i(IFFT) i(DS) (iii) CP-CDMA: 2N log2 (N) + 3N;
OFDM 0 1 0 0 (iv) CP-DS-CDMA: N log2 (N) + 3(N/G) + (N/G) log2 (N).
SCCP 0 1 1 0 It is seen that MC-CDMA and CP-CDMA have relatively
MC-DS-CDMA 1 1 0 0 higher complexity than the other two. However, we point out
MC-CDMA 0 1 0 1 here that the lower complexity of MC-DS-CDMA and CP-
CP-CDMA 0 1 1 1 DS-CDMA is achieved with the assumption that the wireless
CP-DS-CDMA 1 1 1 0 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-
block implements time and frequency synchronization as tunately, the complexity increment for these two systems is
well as phase noise estimation. Code acquisition is needed around 2 times only.
for CDMA case. Channel estimation block is also common As we mentioned earlier, we target the reconfigurable re-
for each system, which can be implemented using preamble ceiver architecture for the downlink only. Now, we quan-
blocks or common pilot channels or dedicated pilot chan- tify the gain of doing so with respect to the multimode re-
nels, based on the systems to be configured to. The FFT ceiver which gathers the implementation for each system.
is performed in Modem Rx for all systems. The other pa- Adding the NCM for OFDM, which is N log2 (N) + 3N, and
rameters are system dependent, which are summarized in the NCM for SCCP, which is 2N log2 (N) + 3N, with those
Table 2. for CDMA systems, the required NCM per data block for
the simple architecture supporting the six modes is given by
(i) For OFDM, only OTE is needed; for MC-DS-CDMA, (8N + (N/G)) log2 (N) + 12N + 6(N/G). With the reconfig-
both BDI and OTE are required. urable receiver, the required blocks are FFT, BDI, OTE, IFFT,
(ii) For SCCP, OTE and IFFE blocks function; for CP-DS- and DS. Again, ignoring the complexity for BDI and DS, then
CDMA, BDI, OTE, and IFFT blocks are required. the total NCM required by the reconfigurable receiver per
(iii) For MC-CDMA, both OTE and DS blocks are needed; data block is 2N log2 (N) + 3N. For large processing gain, the
for CP-CDMA, OTE, IFFT, and DS blocks are all re- required NCM for the reconfigurable receiver is about 25%
quired. of that for the multimode receiver.
Finally, we list out the typical values of block size N for
different standards. In 802.11a/g, N = 64, and in the propos-
4.3. Complexity issues
als for 802.11n, N = 64 or N = 128. N = 256 for 802.16a/e
The equalizer complexities of different systems are com- OFDM mode, and N is as high as 1024 [21] for B3G systems.
pared. For non-CDMA systems, such as OFDM and SCCP, For CDMA-based systems, when the chip rate is fixed, the
the block size N is usually small, say 64 in IEEE 802.11a. For processing gain is usually variable depending on the user’s
CDMA systems, the block size N can be much larger, say at data rate [21].
least 256. Thus we only compare the receiver complexities of
CDMA-based systems.
5. HARDWARE ARCHITECTURE
Table 3 illustrates the required operations per block for
each CDMA-based system. Note that for both BDI and DS The proposed transceiver architecture is mapped onto the
blocks, the number of despreaders of size G is considered. We hardware architecture shown in Figure 4. In this figure,
use the number of complex multiplications (NCM) per block generic high-level functional subsystems and interfaces of the
as the complexity metric. As the despreading operation may software-defined radio (SDR) transceiver system are shown.
only involve addition operations since the codes are usually The main entities of the architecture are the general-purpose
BPSK or QPSK modulated, this complexity is ignored. Treat processor (GPP) and SDR modem subsystem.
the required NCM for an N-dimensional OTE as 3N, and The GPP is an x86-based processor with a PCI-based in-
the NCM for an N-point FFT (IFFT) as N log2 (N). Then the terface for the SDR modem subsystem. The processing re-
required NCMs per block for each CDMA system are given sources available on the SDR modem subsystem are Ti DSP
User
RF front end RF front end System & interface
configuration
RTR Engine manager
Baseband Baseband
engine1 engine < n > Config. database
Upper-layer protocols (TCP/IP)

L1 / L2 L1 / L2 and application software
engine1 engine < n >
Firmware engine Waveform Waveform

SDR modem driver1 Waveform
driver2
driver (802.11) driver < n >
(WCDMA)
Platform Host interface
manager manager
RTOS
SDR executive
DSP Kernel/OS
SDR modem subsystem General-purpose processor
Figure 4: Transceiver architecture blocks and interfaces.
TMS320C6416 running at a clock speed of 600 MHz, a Xil- The platform has a run-time reconfigurable (RTR)
inx FPGA (XCV2V6000), and the RF front end. Along with programmable logic hardware engine. This is used in
the analog RF circuitry, the RF front end also has the ADC computation-intensive front-end signal processing of Mo-
(analog devices AD6544), DAC (AD9777), and some recon- dem Tx shown in Figure 2 and Modem Rx of Figure 3. This
figurable logic for digital front-end processing. subsystem consists of reconfigurable hardware and a man-
The logical functions performed in these resources will agement firmware module that runs on the embedded Ti
be dependent on the type of system (OFDM, MC-DS- DSP processor, managing the resources on the hardware.
CDMA, SCCP, CP-DS-CDMA, MC-CDMA, CP-CDMA) The major functions implemented in the RTR engine are
that is being realized. channelization, synchronization and timing recovery, chan-
A set of logical entities should be realized on the hard- nel estimation and equalization, demodulation, despreading,
ware platform comprised of GPP and SDR modem subsys- and de-interleaving. These are typically computation inten-
tem. The specific functions performed in these resources will sive processing suitable for parallel hardware implementa-
be dependent on the type of system being realized in the tion. Corresponding reverse processing is done in the uplink.
platform. The transient functions in the architecture which The rest of the functions such as source/channel cod-
are configuration dependent are shown in dotted outlines in ing, scrambling, pilot insertion, and so forth and their cor-
Figure 4. The functions that are always present irrespective of responding reverse in the receiver path are performed using
the configured system are shown in solid outlines. the processing resources of DSP. Again the exact functions
SDR modem subsystem as shown in Figure 4 covers the performed by these engines depend on the system applica-
RF and digital signal processing functions in the system. This tions that are configured.
subsystem interfaces to the GPP as a device (specifically as a The firmware engine, which is implemented in the DSP,
PCI device). represents the run-time reconfigurable firmware subsystem.
Burst formation, transmitter front-end (FE) processing This typically implements some of the Layer-1 control pro-
and receiver FE processing functions are realized in the RF cedures like scheduling and Layer-1 management and MAC
front end. In the receive path, the RF signals down-converts processing (for example in IEEE 802.11a). These functions
(5 GHz band signals for 802.11a WLAN) to a 70 MHz IF and also depend on the waveform application being reconfigured
quadrature is digitized into I and Q for further processing by to as the processing requirements for different CP-based sys-
baseband run-time reconfigurable (RTR) engine in the re- tems are different.
ceive section. In the transmit section, the RF subsystem up- An SDR executive provides the platform services needed
converts the baseband I and Q signals to the required fre- to reconfigure and activate waveform applications, such as
quency (5 GHz band for 802.11a WLAN) for transmission. 802.11a for OFDM system. It is independent of any other ap-
The RF front end is configured to the required configura- plication running on the platform. The SDR executive con-
tion during the initial configuration process. The parameters sists of the RTOS, a basic set of resource and configuration
for configuration are provided by the respective waveform management processes and communication and data path
driver. management functions. The executive interfaces with the
system and configuration manager (SCM) on GPP through some of them may be modified versions of standard drivers.
the SDR Modem driver and waveform driver executed on the For example in the case of OFDM, this will be a modi-
GPP. The SDR executive incorporates a command interpreter fied, dynamically loadable version of the standard wireless
to receive and process the commands from the drivers. It also LAN 802.11a drivers. As for CDMA application, this may
manages the communication channels between the GPP and be a specific driver developed for the purpose, incorporat-
SDR Modem. ing the Layer 2 and higher protocols interfacing with Layer-
To enable instantiation of different waveform applica- 1 functions inside the SDR Modem. These driver modules
tions on the same hardware, the SDR executive supports dy- are loaded and unloaded by the SCM, whenever the cor-
namic loading of COFF modules and reconfiguration of FP- responding waveform application is created and destroyed.
GAs. The SDR executive will be stored on a nonvolatile stor- Upper-layer protocols and application software will make use
age and will be loaded at boot time. This module receives of the communication capabilities provided by the waveform
firmware and bit map information from the system and con- drivers and interface with them.
figuration manager and configures on the platform subsys- A configuration database holds all the configuration and
tems. status information including all the system images for each
System and configuration manager (SCM) is the central of the waveform applications that are supported. At least part
management function of the architecture. The SCM uses the of it will be nonvolatile (e.g., disk based). The SCM manages
GPP resources to perform high-level functions like system the configuration database, it maintains the current status of
management, configuration/reconfiguration management of the system, and will estimate, along with the SDR executive,
the platform, configuration storage, user interface, and so the resource availability for the selected application. If suf-
forth. This module performs its functions in conjunction ficient resources are available, the waveform application will
with other platform functions, specifically the SDR executive, be installed and feedback will be given to the user. If resources
kernel/OS, user interface, and configuration database. are not available, it will abort the loading and free up all the
The major functions performed by the SCM are resources reserved for this waveform application. The load-
ing process includes, in addition to loading of SDR Modem
(a) SDR Modem control (reset, platform loading, initial- firmware modules and FPGA bit maps, any host driver mod-
ization, etc.); ules. A loaded waveform application can be activated only
(b) waveform application database management (add/ upon explicit activation by the SCM. This involves activa-
delete waveform applications); tion of RF subsystem, hardware and firmware components
(c) waveform application management (loading, activa- (loaded and configured during the loading process), as well
tion, deactivation, unloading); as loading of the driver on the host side. The SCM imple-
(d) diagnostic support (diagnostic commands, error and ments error logging and diagnostic command support for
message logging, etc.). the development and testing support. This includes ability to
log module-wise diagnostic messages into nonvolatile stor-
The SCM uses the SDR Modem driver to communicate age.
with the SDR executive to download platform components Finally, the user interface exposes the user features pro-
as needed. vided by the SCM to the user.
SDR Modem driver, a kernel driver, interfaces the SDR
modem subsystem platform to the system and configuration
6. CONCLUSIONS
manager process and acts as the root device driver for the
SDR Modem hardware. The driver will be loaded when the In this paper, we have proposed a broadband mobile
SDR Modem device is detected and basically interfaces with transceiver and a hardware architecture which can be con-
the SDR executive to pass command and response between figured to any cyclic-prefix (CP) -based system. We have
the SCM and SDR executive. This also provides facility to identified the reusable common blocks and studied the re-
transfer the COFF files and FPGA bit maps to configure the ceiver complexity of the equalization block for different sys-
DSP and FPGA, respectively. tems. It is found that after the despreading operation, MC-
Kernel/OS block in Figure 4 is the operating-system ker- DS-CDMA and CP-DS-CDMA have the same equalization
nel on the GPP. The operating system used is a general- blocks as OFDM and SCCP systems, respectively, showing
purpose OS, LINUX, with real-time extensions, or a dedi- that both hardware and software sharing is possible for those
cated real-time OS like LynxOS could also be used. The real- systems. We also noticed that although MC-CDMA and CP-
time extensions are necessary as part of the waveform appli- CDMA require only one block to perform signal detection,
cations, depending on the application, may be implemented the receiver complexity is higher than the other two CDMA
inside the kernel. systems. However, MC-DS-CDMA and CP-DS-CDMA rely
Waveform drivern, are function-specific drivers in the on the assumption that the wireless channel is static within
host OS. These are shown in dotted outline to indicate that the G blocks. For fast-fading environments, MC-CDMA and
they are not always present. A driver will be loaded into CP-CDMA may still be the better choice to compromise the
the host memory only when corresponding waveform ap- fast fading and complexity issues. Further, functionality of
plication is being instantiated; and it is unloaded when the the different functions in the proposed hardware architecture
application is unloaded. These are technology specific and is elaborated. It is seen that though different functions can be
realized on a reconfigurable hardware, the major challenge is cations (ISSSTA ’04), pp. 854–858, Sydney, Australia, August–
to have an efficient system configuration and management September 2004.
function which will initiate and control the reconfiguration [18] A. Gupta, A. Forenza, and R. W. Heath Jr., “Rapid MIMO-
OFDM software defined radio system prototyping,” in Proc.
as per waveform application requirements.
IEEE Workshop on Signal Processing Systems (SIPS ’04), pp.
182–187, Austin, Tex, USA, October 2004.
REFERENCES [19] S. Chuprun, J. Kleider, and C. Bergstrom, “Emerging software
defined radio architectures supporting wireless high data rate
[1] J. Mitola III, “The software radio architecture,” IEEE Com-
OFDM,” in Proc. IEEE Radio and Wireless Conference (RAW-
mun. Mag., vol. 33, no. 5, pp. 26–38, 1995.
CON ’99), pp. 117–120, Denver, Colo, USA, August 1999.
[2] J. Mitola III and G. Q. Maguire Jr., “Cognitive radio: making
[20] D. Falconer, S. L. Ariyavisitakul, A. Benyamin-Seeyar, and
software radios more personal,” IEEE Pers. Commun., vol. 6,
B. Eidson, “Frequency domain equalization for single-carrier
no. 4, pp. 13–18, 1999.
broadband wireless systems,” IEEE Commun. Mag., vol. 40,
[3] L. J. Cimini Jr., “Analysis and simulation of a digital mobile
no. 4, pp. 58–66, 2002.
channel using orthogonal frequency division multiplexing,”
[21] N. Maeda, Y. Kishiyama, H. Atarashi, and M. Sawahashi,
IEEE Trans. Commun., vol. 33, no. 7, pp. 665–675, 1985.
“Variable spreading factor-OFCDM with two dimensional
[4] Y.-C. Liang, A. S. Madhukumar, and F. Chin, “Single carrier
spreading that prioritizes time domain spreading for forward
cyclic prefix based CDMA systems,” Tech. Rep., Centre for
link broadband wireless access,” in Proc. 57th IEEE Semian-
Wireless Communications (Now Institute for InfoComm Re-
nual Vehicular Technology Conference (VTC ’03), vol. 1, pp.
search), Singapore, Singapore, 2001.
127–132, Jeju, Korea, April 2003.
[5] Y.-C. Liang, “Block-Iterative GDFE (BI-GDFE) for CP-
CDMA,” in Proc. 61st IEEE Semiannual Vehicular Technology
Conference (VTC ’05), Stockholm, Sweden, May–June 2005. Ying-Chang Liang received the Ph.D. de-
[6] K. L. Baum, T. A. Thomas, F. W. Vook, and V. Nangia, gree in electrical engineering from Jilin
“Cyclic-prefix CDMA: an improved transmission method for University, China, in 1993. He is now a
broadband DS-CDMA cellular systems,” in Proc. IEEE Wire-
Lead Scientist in the Institute for Info-
less Communications and Networking Conference (WCNC ’02),
comm Research (I2R), Singapore. He also
vol. 1, pp. 183–188, Orlando, Fla, USA, March 2002.
[7] S. Zhou, P. Xia, G. Leus, and G. B Giannakis, “Chip- holds Adjunct Associate Professor positions
interleaved block-spread CDMA versus DS-CDMA for cel- in Nanyang Technological University and
lular downlink: a comparative study,” IEEE Transactions on the National University of Singapore, Sin-
Wireless Communications, vol. 3, no. 1, pp. 176–190, 2004. gapore. From December 2002 to December
[8] S. Hara and R. Prasad, “Overview of multicarrier CDMA,” 2003, he was a Visiting Scholar in the De-
IEEE Commun. Mag., vol. 35, no. 12, pp. 126–133, 1997. partment of Electrical Engineering, Stanford University. From Au-
[9] X. Cai, S. Zhou, and G. B. Giannakis, “Group-orthogonal gust 1997 to November 2001, he was a Senior Member of techni-
multicarrier CDMA,” IEEE Trans. Commun., vol. 52, no. 1, cal staff in the Centre for Wireless Communications, Singapore.
pp. 90–99, 2004. His research interests include space-time wireless communications,
[10] A. Jamalipour, T. Wada, and T. Yamazato, “A tutorial on multi- reconfigurable signal processing and cognitive radio, smart an-
ple access technologies for beyond 3G mobile networks,” IEEE tennas, ultra-wideband communications, and capacity-achieving
Commun. Mag., vol. 43, no. 2, pp. 110–117, 2005. schemes for MIMO fading channels. He received the Best Paper
[11] K.-C. Chen and S.-T. Wu, “A programmable architecture for Award from the 50th IEEE Vehicular Technology Conference. He
OFDM-CDMA,” IEEE Commun. Mag., vol. 37, no. 11, pp. 76– was also the corecipient of the 1997 National Natural Science Award
82, 1999. from China. He has been an Associate Editor for IEEE Transactions
[12] S.-T. Wu and K.-C. Chen, “Programmable multiuser synchro- on Wireless Communications since 2002. He was the Publication
nization for OFDM-CDMA,” in Proc. 53rd IEEE Vehicular Chair for 2001 IEEE Workshop on Statistical Signal Processing and
Technology Conference (VTC ’01), vol. 2, pp. 830–834, Rhodes, is a Senior Member of IEEE.
Greece, May 2001.
[13] J. Helmschmidt, E. Schuler, S. Rao, S. Rossi, S. di Matteo, and Sayed Naveen is an Associate Scientist in the
R. Bonitz, “Reconfigurable signal processing in wireless termi- Communications & Devices Division, In-
nals [mobile applications],” in Proc. Design, Automation and stitute for Infocomm Research (I2R), Sin-
Test in Europe Conference and Exhibition (DATE ’03), pp. 244– gapore. He received his B.E. degree in
249, Munich, Germany, March 2003. electronics and communication engineer-
[14] J. R. Moorman, “Implementation of a 3G W-CDMA software ing from the University of Madras, India,
radio,” in Proc. IEEE International Conference on Communi- in 1992, and the M.Eng. degree in commu-
cations (ICC ’03), vol. 4, pp. 2494–2499, Anchorage, Alaska,
nication systems from the Regional Engi-
USA, May 2003.
neering College, Trichy, India, in 1995. After
[15] I. P. Seskar and N. B. Mandayam, “A software radio architec-
ture for linear multiuser detection,” IEEE J. Select. Areas Com- the M.S. degree, he joined the Central Re-
mun., vol. 17, no. 5, pp. 814–823, 1999. search Laboratory, Bharat Electronics Limited, Bangalore, India, as
[16] J. Palicot and C. Roland, “FFT: a basic function for a recon- a member of research staff, where he was working on power ampli-
figurable receiver,” in Proc. 10th International Conference on fier linearization and wireless transceivers for defense communi-
Telecommunications (ICT ’03), vol. 1, pp. 898–902, Papeete, cations. In 1999, he joined the WCDMA base station development
Tahiti, February–March 2003. team, Sanyo LSI Technology India, Bangalore, India, as a Senior En-
[17] P. Burns and M. C. Reed, “A complexity cost function for the gineer of signal processing. In 2000, he joined I2R (formerly Cen-
signal processing in a WCDMA base station for dimension- tre for Wireless Communications). His research interests include
ing of a software defined radio,” in Proc. 8th IEEE Interna- reconfigurable signal processing and architectures for broadband
tional Symposium on Spread Spectrum Techniques and Appli- wireless communications and software-defined radio.
Santosh K. 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-defined radios, and
reconfigurable 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.
c 2005 Arnd-Ragnar Rhiemeier
Modular Software-Defined Radio
Arnd-Ragnar Rhiemeier
Institut für Nachrichtentechnik, Universität 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-defined 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-defined radio. Modularity and flexibility
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 findings as well as new results from system simulations presented here, the paper finally concludes with the
utility of pipelining as a general design guideline for modular software-defined radio.
Keywords and phrases: flexible digital baseband signal processing, firmware support for reconfiguration, computing resource
allocation, multiprocessing, modeling of SDR software.
1. INTRODUCTION The present paper puts an emphasis on the physical-

layer signal processing because its capabilities represent the
Software-defined and hardware reconfigurable radio systems fundamental limits for higher layers in delivering their ser-
have attracted more and more attention recently because vices. Therefore, mastering all aspects of physical-layer sig-
they are expected to be among the key techniques to serv- nal processing in software-defined and hardware reconfig-
ing future wireless communication market needs. In con- urable radios is vital for delivering best end-to-end service
trast to the strong convergence tendency in wired networks, to the end user, by means of a single communication de-
a growing number of standards and communication modes vice. Modular software-defined radio (Mod-SDR) strives for
can be observed in wireless access networks. Presumably, this casting light on one important aspect which has been largely
trend will prevail, eventually due to the natural diversity in neglected hitherto: the design guidelines which govern the
service requirements and radio environments. The better the coordinated interplay of signal processing software modules
match between the physical channel (the properties of which embedded in logical structures of some arbitrary wireless
are determined in part by the user mobility) and the signal communication standards.
processing in the transceiver, the easier to achieve the opti-
mal quality of service (QoS) on the physical layer. Further- 1.1. Related work
more, the ongoing introduction of UMTS in Europe shows
A great number of important contributions on software-
that diversity in standards is not only a technical challenge; if
defined radio [1, 2, 3] and reconfigurability [4, 5, 6, 7, 8] can
market response falls short of business expectations (based
be found in the literature. However, many authors narrow
on a particular communication standard), or if user’s de-
down their research interest to one particular aspect of the
mand shifts to a different wireless access technology (and
signal processing chain: sample rate adaptation [9, 10, 11],
thus to a different sort of underlying signal processing), it
RF front-end design [12, 13, 14, 15, 16], A/D conversion [17,
would be beneficial for any manufacturing company to be
18], or channel coding [19, 20, 21], just to name a few exam-
able to respond quickly to such situations. Software-defined
ples. Notably, work related to signal processing in the digital
and reconfigurable radio systems have the potential to allow
baseband is centered around algorithms [22, 23]. However,
short time-to-market product designs under these commer-
structural properties of signal processing software (including
cial conditions.
an abstract way for representation) and the principles of or-
ganizing the execution of multiple algorithms in a distributed
This is an open access article distributed under the Creative Commons multiprocessing hardware system have not been studied in-
Attribution License, which permits unrestricted use, distribution, and tensively in the context of software radio. One major contri-
reproduction in any medium, provided the original work is properly cited. bution of Mitola [24] attempts to reexamine software radio
from a truly unorthodox point of view, but his findings still hardware architecture, providing sufficient computing power
pertain to algorithms and eventually do not reach beyond the at a much lower electric power consumption than that of a
Turing’s theory of computing. Nevertheless, his contribution single comparable general-purpose processor. Furthermore,
hints at the fact that SDR requires an understanding which all hardware resources of a Mod-SDR device such as pro-
is radically different from classical communications and its cessors, buses, memories, and interfaces are administered by
BER curves. a nonpreemptive operating system. Supplied by the termi-
nal manufacturer, this piece of firmware is protected from
1.2. Motivation for modular software-defined radio any direct manipulation on the part of the end user. How-
The motivation for introducing a novel view of flexible radio ever, by means of a physical layer API, both user applications
system design is twofold. First, the fact that a radio terminal and the network may address transmission mode requests to
accomplishes its signal processing by software or reconfig- the QoS manager which belongs to the core framework ser-
urable hardware rather than by dedicated hardware does not vices/applications running immediately on top of the oper-
render the rules of real-time computing obsolete, but so far ating system (see SCA, [28, Figure 2-1]). A request will cer-
this aspect has not been considered systematically in the con- tainly include an abstract representation of the signal pro-
text of software-defined radio. Second, any BER produced by cessing software which needs to be executed to realize the
some software implementation has to be at least as good as transmission mode. modular software-defined radio makes
the BER of its equivalent ASIC implementation. Hence, it ap- use of directed acyclic graphs to represent signal processing
pears to be questionable to present BER curves as a measure software.
of quality for the design of a software-defined radio. The main task of the QoS manager consists of the map-
The goal of modular software-defined radio is to estab- ping of signal processing software to the available hardware
lish general guidelines for designing and operating flexible while respecting the real-time requirements which are de-
signal processing systems. In order to make these guidelines fined by the air interface of the requested communication
general, they need to be independent of particular commu- standard [29]. In return for requests, the QoS manager ac-
nication standards, independent of algorithmic implementa- cepts or rejects transmission modes, based on the availability
tion details, and independent of technological advances in of resources and on a decision process which includes parti-
microelectronics. The model for SDR software will reflect tioning of the graph and scheduling of all software modules.
these important aspects. Partitioning is the process of uniquely assigning each module
to a processor, while scheduling means determining individ-
1.3. Compatibility with the real world ual trigger instants for each module.
In the same way as the ideal software radio concept [1, 25] Throughout this work, a software module is defined to be
has evolved into some compromise approaches usually sum- the smallest entity of executable machine code, which can-
marized under the term of software-defined radio, the advent not be preempted by the operating system. Software mod-
of reconfigurable, distributed signal processing hardware in ules are self-contained, independent entities, and there is no
radio devices [26, 27] can be seen as another step in this evo- data exchange across module boundaries other than input
lution towards implementations which are both technologi- data from predecessor modules and output data to successor
cally feasible and economically attractive. modules. In principle, any module may be connected to any
Critics generally claim that SDRs will always be noto- other module, the only requirement being data type compat-
riously power-inefficient and inherently overpriced, hence ibility in between all modules.
never prove competitive against carefully designed ASICs. 2.1. Details of the software model
This may be true indeed if the flexibility of SDR is uncon-
ditionally passed on to the end user in the form of “future For the Mod-SDR software model to be independent of pro-
upgradability.” Therefore, it is more reasonable to predict cessor types (DSP, FPGA, ASSP, specific coprocessor, and
that the flexibility of SDR is likely to stay under the imme- ASIC) and of technological advances in microelectronics, the
diate control of manufacturers, all the more so to support a processing runtime of software modules is taken into ac-
sustainable business model. Actually, time-to-market is the count as the main behavioral attribute of signal processing
master argument in favor of software-defined radio tech- software. At the same time, algorithmic details are abstracted
niques. The present paper shares this view suggesting to per- away in this manner. Given the framework of directed acyclic
ceive software-defined and reconfigurable radios as wireless graphs, the nodes of a graph represent software modules car-
communications embedded real-time systems which are tun- rying some signal processing runtime as the node weight.
able to end-user needs or network operator needs, but not Due to the vast variety of unpredictable influences on the
more. Modular software-defined radio provides those em- processing runtime of software modules in an SDR envi-
bedded systems with design guidelines and the core of a QoS ronment [29], node weights are subject to random variation
manager for the physical layer. 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
2. MODELING OF HARDWARE AND SOFTWARE be produced at the output of a software module:
One fundamental assumption is that SDR will become real-
ity soonest in the form of some distributed multiprocessing pm = α · c · rm , ∀ nodes m : 1 ≤ m ≤ M, (1)
Modular Software-Defined Radio 335
where pm is the processing runtime of some node m and rm 5 36 27

is m’s output memory resource demand. Formally, α ∈ R+ 23 4
6 7
is the constant of proportionality translating a memory de- 22 13 38 21 11
mand into a runtime, hence its unit [α] = (bit/s)−1 . It can 31 18 39
9 29
be interpreted as the absolute speed of a processor when 8
28
executing the signal processing machine code behind node 2
1 33 17
m. The constant factor c is unitless, and its value is drawn
16 15
from a random experiment, for each m anew. The charac-
37
teristics of the SDR environment as well as all shortcomings
26 35
of a strictly linear resource runtime model are modeled by 19 12
the real-valued random variable C. A complete description 3 32 34 30
40
of this variable is given by its probability density function 10 25 20 24 14
(pdf) fC (c). Throughout this paper, all realizations c stem
from independent identically distributed random variables Figure 1: Irregular, connected, directed, acyclic graph.
C for all nodes m. The pdf employed in this work has a
rectangularly windowed Gaussian shape with the parameters M1 M2
µC = 1.0, σC,eff = E{(C − µC )2 } = 0.25, and windowed in- ···
terval [0.5; 1.5]. The choice of a Gaussian is certainly arbi- P1 P2
trary, but there are strong hints that the actual shape of the B B
pdf has much less influence on the performance of the QoS i/o Shared
manager than the effective relative spread σP,eff /µP,eff of pro- memory memory
B
cessing runtimes [31].
The structural properties of Mod-SDR software are cap- Figure 2: Symmetric multiprocessor architecture.
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 generally includes L ∈ N identical processors Pl with associ-
in their node runtimes, but also in their directed edges. ated (distributed) memory Ml , a shared memory and an in-
The random graph generator which is used for computer put/output (i/o) memory for interfacing of the physical layer
simulations produces irregular, connected, directed, acyclic signal processing subsystem to the outside world, that is, to
graphs with a fixed number of nodes M = 40. The gener- other processing subsystems or to the analog front end of the
ation process starts out from a chain of two nodes indexed Mod-SDR transceiver.
m = 1 (referred to as the source node) and m = 2 (referred All of these hardware resources are connected by B ∈ N
to as the target node). Subsequently, nodes are added in an separate data buses. It is assumed that processors are actively
iterative process by placing them somewhere relative to the involved in bus transfers, that is, no useful signal processing
existing graph. For every node to be placed, one predeces- code of a module can be executed by a processor while this
sor and one successor are selected at random and with equal processor is exchanging data with the shared memory or with
probability from the existing nodes of the graph. Should a the i/o memory via some bus. This is certainly a conservative
new node shortcut the edge between adjacent existing nodes, assumption, which can also be interpreted as a worst case on
then that edge is removed with probability of 0.5. Connect- one hand. On the other hand, however, it is very unlikely
edness of the graph is enforced in a simple way by exempting that simultaneous signal processing and bus communica-
m = 1 from the node selection procedure, whereas the prop- tions can be achieved in the general case of Mod-SDR graphs.
erty of noncyclicity needs to be verified throughout the en- Even if a processor architecture supports communication la-
tire graph generation process. Finally, some additional graph tency hiding behind useful core computations, those mech-
properties, which are related to the data input and output anisms would require coordination on the code program-
behavior of nodes, need to be determined. First, the output- ing level across module boundaries. However, this intermod-
to-input data ratio of all nodes be 1.0. Second, all edges out- ule control flow contradicts the Mod-SDR self-containment
going from any demultiplexing node convey the full output paradigm, where any module may indeed be linked to any
data volume. Third, all edges incoming to any multiplexing other module (logically, by the directed edge of a graph), in
node convey only one Kth of the predecessors’ output data order to accomplish a useful signal processing task, but with-
volumes, where K is the number of incoming edges. These out mutual knowledge of their respective machine code in-
rules make sure that edge weights remain in the same order ternals.
of magnitude throughout the entire graph. Figure 1 shows Naturally, bus access is exclusive. Conflicting access tim-
one realization of a random graph as an example. ings on the part of processors need to be arbitrated by the
scheduler, which is built into the QoS manager. The bus
2.2. Details of the hardware model speed is given relative to the signal processing speed α of the
A symmetric multiprocessor architecture (see Figure 2) processors, in the form of a relative bus speed β ∈ R+ . The
serves as the model of a modular, easy-to-extend multi- unitless factor β describes how much faster a processor can
processing hardware system for Mod-SDR. The architecture transfer an amount of data over the bus rather than produce
the same amount of data as the output of a module (by reg- over their entire utilization period. Potentially, QoS parame-
ular signal processing). Large values of the relative bus speed ters may be negotiated during connection setup, but once a
(β 1) represent fast buses, whereas β < 1 represents slow transmission mode has been accepted, the (static) scheduling
buses. Partitioning always entails a cut affecting a certain sub- situation is deterministic, and so is the demand for comput-
set of edges in the graph. The logical data flow along cut edges ing power [29].
then translates into an asynchronous, physical data flow be- All SDR design techniques related to mapping structured
tween processors via the shared memory. How this is orga- signal processing software to hardware may be roughly classi-
nized using pairs of bus transfer nodes is described in detail fied by the number of graph copies involved in the partition-
in [32]. The basic idea is that bus transfers require runtime in ing and scheduling process; some approaches use but a sin-
the same way as regular signal processing nodes, but paired gle copy, while others imply multiple identical copies of the
for intermediate results to be written to and read from the graph. In principle, techniques of both classes are applicable
shared memory. Therefore, the resulting runtime model for to signal processing for circuit-switched services as well as
bus transfer nodes is similar to (1), but includes β in the place for packet-switched services. However, ordinary protocol re-
of c: quirements in packet-oriented networks (e.g., stop-and-wait
ARQ) may oftentimes force the QoS manager to operate on
pm,n = α · β−1 · rm , ∀ edges m, n affected by the cut. a single copy of the given graph. As a matter of principle,
(2) single-copy variants are less demanding in terms of program
memory and data memory.
These runtimes do not depend on the SDR environment, The following partitioning approaches are employed in
but on the deterministic capabilities of the Mod-SDR device. Mod-SDR system simulations to be discussed in a later sec-
Their values pm,n reappear later as edge weights that repre- tion of the this paper.
sent potential link cost for all edges m, n. Actual costs are (i) Implicit partitioning by the Hu algorithm [29]—
only incurred by those edges affected by the partitioning cut. eventually, a scheduling algorithm.
The focus of the present work is on fundamental design (ii) Kernighan-Lin (KL) algorithm [30]—a method for
and operating principles in Mod-SDR systems. Before tack- local search of the design space.
ling the general case of L ∈ N processors, the case L = 2 (iii) Spectral partitioning [33]—an application of alge-
should be well understood first. With L = 2 processors, only braic graph theory.
B ≤ 2 is reasonable. In this paper, the case of B = 1 bus is These approaches primarily operate on a single copy of
studied. the graph which is given by some transmission mode request.
The originally implied scheduling idea is that radio signals
2.3. Proposed measure of quality are processed on a frame-by-frame basis, only one frame per
It is the declared goal of Mod-SDR to go beyond designing real-time period, and all computations relate to a single radio
for some particular set of standards or transmission modes. frame only. An accurate pseudocode of the static scheduler is
Therefore, concrete real-time requirements such as deadline given in [34].
periods (which are clearly determined by the air interfaces of However, pipelined scheduling [33, 35] in combination
any standard) are missing. Instead, the more general require- with these partitioning approaches revokes the memory ad-
ment for maximum speedup of a multiprocessing hardware vantage of a single graph copy, because pipelining involves
system is considered. The speedup s ∈ R+ is defined as the the processing of radio signals related to several different
factor by which the multiprocessor Mod-SDR implementa- radio frames within the time of one real-time period. The
tion terminates faster than the same software implementa- number of different radio frames involved in these computa-
tion on a single processor running at the same speed of α. tions is called the depth of the pipeline. Indeed, the program
The advantage of speedup is evident; it is a relative measure memory remains unaffected by pipelining, but not the data
of quality for the design of (and the way of operating) a mul- memory; multiple intermediate computation results from
tiprocessor computing system. The actual value of α is not of several different frames must be kept in memory, which re-
importance because it cancels out in the speedup s. sults in a much higher demand than before.
A radical alternative to the above partitioning approaches
is graph duplication; one complete copy of the graph is as-
3. MAPPING APPROACHES signed to the first processor, another copy is assigned to the
Although the Mod-SDR software model includes random second processor. No partitioning algorithm is needed, the
graphs, statistics in this approach should not be confused workload on both processors is perfectly balanced by con-
with any “statistical structure of computational demand” struction, and bus access is reduced to the necessary i/o data
[24]. The way of building and using a population of SDR ter- transfers. The associated scheduling scheme has been named
minals involves a random process, but transmission mode re- graph duplication pipelining (GDP) [35]. Despite its obvi-
quests are realizations of this random process. Therefore, the ous advantages, GDP suffers from both increased data mem-
QoS manager has to deal with realizations of random graphs. ory and program’s memory demand due to the duplication
Per request, the total number of nodes and all associated process. An alternative pipelining approach that returns to
node weights are arbitrary, but fixed. Likewise, the logical operating on a single copy of the graph is half-frame pipelin-
structures captured in the set of directed edges remain fixed ing (HFP) [36, 37]. HFP is primarily based on a scheduling
1 in principle. At a sample size of 2000 realizations per β, the

observed measurement range is often so densely populated
0.9
by dots that the latter amalgamate into vertical lines. There-
Fraction of max speedup
fore, in addition to the individual speedup measurements,

the contour lines of the 5%, the 50% (median), and the 95%
0.8 quantile are estimated and overlayed to the figures. A con-
tour line represents the maximum speedup achieved by the
0.7 given quantile of Mod-SDR realizations (95%: topmost, 5%:
bottommost, median: in between) as a function of β.
0.6 4.1. Circuit-switched services

Figure 3 shows the speedup results for implicit Hu partition-
ing and nonpipelined scheduling. Obviously, the faster the
10−1 100 101 102 bus, the better the speedup. Since Hu’s algorithm is eventu-
Relative bus speed β ally a pure scheduling algorithm, it does not take into ac-
count any link cost while partitioning graphs implicitly. A
Figure 3: Implicit Hu partitioning, no pipelining. 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 β → ∞. Figure 3 dis-
idea where the QoS manager supports the following kind proves this assumption. The behavior observed over the en-
of time-interleaved frame processing; while one processor is tire β range can be explained in part by the occurrence of
still busy processing the second half of some frame indexed PHYSICAL WAIT IDLE and LOGICAL WAIT IDLE condi-
i, the other processor already starts processing its successor tions [30]. The former arises after the arbitration of concur-
frame indexed (i + 1). In contrast to GDP, this approach re- rent bus access requests, whereas the latter originates from
quires load-balanced partitioning of the graph under one ad- logical interdependencies of nodes in the graph; although the
ditional side condition, namely maintaining a vertical cut rel- bus is fully accessible, one processor is forced to remain idle
ative to the source node and the target node. Both the KL al- waiting for some intermediate results to be produced by the
gorithm and a modified variant of spectral partitioning have other processor.
been tested for this purpose. It can be shown [38] that the Both conditions cause idle times in the processors, and
KL algorithm is a simple and efficient partitioning support thus reduce speedup. While concurrent bus access requests
for HFP. become more and more unlikely as bus speed increases, the
In principle, all of these mapping approaches are eligible LOGICAL WAIT IDLE condition continues to prevail in all
for application by the QoS manager of a Mod-SDR, inde- schedules independent of β. Another reason for speedup loss
pendent of the service type. It remains to be shown which against the limit can be identified in a special property of Hu’s
approach achieves a high probability of good speedup in the algorithm; the approach strictly aims at maximum paral-
highly unpredictable SDR environment. 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
4. SYSTEM SIMULATIONS OF MOD-SDR
due to its given structural properties (small d˜ value below
The three partitioning approaches have been discussed ex- d˜ = 0.5), the Hu algorithm systematically fails to generate
tensively elsewhere, but the comparison of performance was high speedup. Load imbalance between the two processors
merely based on one particular sample graph. The node (equal to the difference of aggregate runtimes between the
weights did stem from random experiments, but the struc- two partitions) is the resulting effect of this failure.
ture of the graph remained fixed throughout all simula- Pipelining eliminates LOGICAL WAIT IDLE conditions
tion runs. Here, in contrast, the random graph model of by deliberately constructing a dense schedule in a first step.
Section 2.1 (also including random edges) is applied to im- All resulting anticausal data dependencies between the parti-
prove the expressiveness of Mod-SDR system simulations. tions are resolved in a second step by rescheduling bus trans-
Those new results are discussed in the following. fers of intermediate results across the boundaries of real-
All figures show speedup measurements (dots) as a func- time processing periods. In this way, a radio frame pipeline
tion of the relative bus speed β. These results are given as of some depth is created (cf. Section 3). Figure 4 shows the
a fraction of the maximum speedup s which is theoreti- speedup results for implicit Hu partitioning under pipelin-
cally achievable [33] by perfect parallelization. Hence, on ing. Indeed, the overall speedup behavior has improved; all
one hand, a fractional value of 1.0 represents the upper per- contour lines indicate higher speedup for the same quantile
formance limit for any Mod-SDR realization. On the other of realizations, and the speedup spread for fast buses is re-
hand, the fractional value of 0.5 represents a reasonable lower duced. Nevertheless, implicit partitioning pursuant to Hu’s
limit, because below s = 0.5, the two-processor system would algorithm continues to suffer from its systematic drawbacks
effectively work slower than a single-processor system, thus mentioned above: strong dependency on graph structure and
rendering any distributed processing approach meaningless complete insensitivity to link cost.
1 1
0.9 0.9

0.8 0.8
0.7 0.7
0.6 0.6
10−1 100 101 102 10−1 100 101 102

Relative bus speed β Relative bus speed β
Figure 4: Implicit Hu partitioning, pipelining. Figure 6: KL partitioning, pipelining.
1 by pipelining. Figure 6 shows the resulting performance

of the KL algorithm. The contour lines reveal the highest
0.9
speedup and the lowest speedup spread observed so far. Nat-
urally, the speedup increases as the relative bus speed β in-
creases, because link cost tends to be reduced and bus con-

0.8 flicts become less and less likely. Nevertheless, the results of
this figure prove that there are better partitions than Hu’s for
0.7
all β. It can be concluded that the approach of Kernighan and
Lin successfully effects a good compromise between maxi-
mum load balance and minimum link cost.
0.6 Since the KL algorithm is based on a local search of
the design space (taking Hu’s solution as a starting config-
uration), it may terminate in local optimum points. Global
10−1 100 101 102 search methods, in contrast, should be able to avoid local
Relative bus speed β optima and finally produce a better overall speedup. Spec-
tral partitioning is a global search method, because it assesses
Figure 5: KL partitioning, no pipelining. 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
The KL algorithm has been introduced to Mod-SDR [30] of the partitioning cut [39]. What’s more, W’s diagonal el-
as a remedy for this situation. Figure 5 shows its performance ements wm,m = pm are the nodes’ processing runtimes ac-
without pipelining. Astonishingly enough at a first glance, cording to (1) and its off-diagonal elements wm,n = 2 · pm,n
although the approach explicitly considers link cost (and it amount to the potential runtimes of bus transfer node pairs,
is even capable of trading load balance for link cost), the where pm,n is from (2). The weight matrix W is real-valued
KL algorithm shows no systematic superiority compared to and symmetric, and spectral partitioning deliberately ex-
Hu’s algorithm under these operating conditions. Speedup ploits this property [40, 41].
degradation in the low β range is a bit more graceful than Figure 7 shows the performance results for spectral par-
in Figure 3, but at high bus speeds, the KL algorithm is eas- titioning and nonpipelined scheduling. Unfortunately, these
ily outperformed by an approach as simple as Hu’s. This can results are much worse than those of Kernighan and Lin; the
be explained by the fact that the partitioning approach by 95% contour line just achieves 0.8 at high bus speeds, and
Kernighan and Lin indeed considers link cost, but tacitly as- the speedup spread remains large in the [0.5; 0.8] interval.
sumes that bus transfers are nonconflicting at all times. Fur- Clearly, such a behavior is completely unacceptable in prac-
thermore, its partitioning cut has an arbitrary orientation tice; a fractional value of 0.5 means that the SDR implemen-
relative to the source node and the target node. As a con- tation on the two-processor system meets real-time deadlines
sequence, a large number of LOGICAL WAIT IDLE condi- in the exact same way as on a single-processor system. Con-
tions still occurs causing a large spread over the [0.7; 0.9] sequently, because the investment into the second processor
range of speedup values. does not pay off at all in the form of speedup, it must be con-
As before, processor idle times associated with LOGI- cluded that the two-processor system is either ill-designed in
CAL WAIT IDLE conditions can be completely eliminated its hardware or ill-conditioned in its operations.
1 1
0.9 0.9

0.8 0.8
0.7 0.7
0.6 0.6
10−1 100 101 102 10−1 100 101 102

Relative bus speed β Relative bus speed β
Figure 7: Spectral partitioning, no pipelining. Figure 8: Spectral partitioning, pipelining.
Previous figures have shown that, as a matter of fact, bet- (or 3, if continuous RF transmission is to be automatically
ter contour lines and narrower spread are possible using the supported by the physical layer processing subsystem [38]).
same hardware and nonpipelined scheduling. Therefore, the Second, GDP is optimal regarding speedup; it reaches the
reason for the inferior speedup behavior of spectral parti- speedup limit s, or the fractional value of 1.0 in Figures 4, 6,
tioning needs to be identified. Figure 8 shows its speedup re- and 8, independent of β. In view of the previously discussed
sults under partitioning. Obviously, only a small part of all difficulties in approaching the limit, GDP is certainly the best
realizations experience an improvement in speedup due to design choice for circuit-switched services. If GDP’s mem-
the elimination of LOGICAL WAIT IDLE conditions. No- ory demand is still an issue, the QoS manager could easily
tably, the 5% contour line remains at the same speedup level resort to HFP. Its speedup performance for circuit-switched
for high bus speeds. These results back previous findings on services (see [38, Figure 2]) is suboptimal, but totally com-
spectral partitioning; the approach in its current form [33] parable to that of the KL algorithm under pipelining (see
does not generate well-balanced partitions. Load imbalance, Figure 6), however, at a constant pipeline depth of 2 and at
as mentioned before in the context of Hu’s algorithm, is a less program and data memory demand than GDP’s.
genuine feature of partitioning, not of scheduling. Failure
to generate load-balanced partitions cannot be compensated 4.2. Packet-switched services
for by any scheduling technique. As mentioned in the beginning of Section 3, packet-oriented
To sum up, the KL algorithm is able to outperform Hu’s networks may require the QoS manager to operate on a sin-
algorithm (but not systematically) in the low-to-midrange β gle copy of the graph. Then the graph contains the complete
region (β < 5), if frame-by-frame signal processing is the de- set of computational tasks necessary for processing a single
sired way of operating the software-defined physical layer of packet, but subsets of these tasks may be repetitive in nature.
a Mod-SDR. True systematic superiority of the KL algorithm Taking the IEEE 802.11a wireless LAN standard as an exam-
in speedup can only be observed under pipelining. A big dis- ple, it is easy to identify such tasks, even when looking at a
advantage, however, is the depth of the pipeline; it depends single packet only: intercarrier/intraconstellation interleav-
on the graph structure, its actual value is not predictable, and ing, constellation mapping, and IFFT [42]. All of these need
it is quite a large integer number. (Histograms not shown to be repeated for every single OFDM symbol alike, just op-
graphically here: mean value around 20 at M = 40 nodes erating on different data within the packet. In contrast, non-
per graph, spread over a window of [5; 35], independent of repetitive computations (per single packet) include scram-
β for implicit Hu and spectral partitioning, depending on β bling and channel coding of IEEE 802.11a.
for KL.) Memory demand increases linearly with the depth The speedup which could be expected under the con-
of the pipeline, and therefore pipelined operation in combi- ditions of the random graph model and a completely
nation with the above approaches does not lead to workable nonrepetitive task graph would be identical to that of Figures
Mod-SDR implementations. 3, 5, and 7. However, if a dominant subset of tasks were in fact
In retrospect, frame-by-frame signal processing must be repetitive, then pipelining approaches such as HFP and GDP
considered inadequate for circuit-switched services. There is could result in better speedup, when applied to the subset.
simply no need to restrict partitioning algorithms to operat- The following results of HFP and GDP for packet pro-
ing on a single graph copy anyway under these conditions. cessing are conditional on the assumption that there are ex-
If the restriction is dropped, the QoS manager can approach actly NF frames per packet which need to be processed iden-
partitioning and scheduling in a different, much simpler way. tically. Furthermore, both processors are considered to be
First of all, it turns out [35] that GDP (cf. Section 3) is op- exclusively reserved for physical layer signal processing as
timal regarding delay; the depth of the pipeline is exactly 2 soon as a packet has arrived. That is to say, one processor
1 1
NF = 7
NF = 6
0.9 0.9
NF = 7

NF = 5
0.8 0.8
NF = 3
0.7 0.7
NF = 5
0.6 NF = 4 0.6
NF = 3
NF = 2
0.5 0.5
10−1 100 101 102 10−1 100 101 102
β β
Upper bounds NF = 6
NF = 4
NF = 2
Figure 9: Half-frame pipelining, packet processing. Sample size Lower bounds
2000 per (β, NF ). Figure 10: Graph duplication pipelining, packet processing. Sam-
ple size 2000 per (β, NF ).
cannot finish the remainder of some higher-layer computa- to real-time multiprocessing and modularity in flexible sig-
tional task, while the other processor already starts process- nal processing software, have been neglected at large in the
ing the radio signals of a physical layer packet. existing SDR literature. Modular software-defined radio ad-
Figures 9 and 10 (adopted from [38]) show the speedup dresses these issues by looking into the organizational princi-
performance of HFP and GDP, respectively. For reasons of ples of signal processing rather than into the signal process-
legibility, only the contour line triplets are drawn, parameter- ing itself. Therefore, a novel way of modeling SDR software
ized by the number of frames per packet NF : 2 ≤ NF ≤ 7. In had to be introduced. Several techniques for mapping such
comparison to circuit-switched processing, GDP is no longer software to hardware have been briefly reviewed. Mod-SDR
optimal, since the filling and the emptying of the pipeline system simulations presented in Section 4 allow to draw the
cause idle times on the processors. A detailed discussion of following conclusions.
HFP and GDP can be found in [38]. However, the crucial (i) With respect to circuit-switched services, frame-by-
point in the above figures lies in the dashed lines representing frame signal processing has proven inadequate. Pipelining
upper bounds on HFP speedup, but lower bounds on GDP methods such as GDP and HFP are to be preferred a priori.
speedup. Therefore, it can be concluded that GDP systemat- GDP is optimal regarding speedup and delay. HFP is subop-
ically outperforms HFP in packet processing. timal, but requires less memory than GDP. Therefore, HFP
Even for reasonable values of β and low numbers of can only prove competitive against GDP if memory is a se-
frames per packet (or task repetitions in parts of a graph), rious design issue. HFP can merely establish a compromise
GDP closely approaches the speedup limit s. So far, only between the achievable speedup and dynamic power dissipa-
transmissions with a constant NF have been examined. How- tion of the bus.
ever, IP traffic in real WLANs consists of a mix of packet sizes, (ii) With respect to packet-switched services, frame-by-
and hence physical layer packets contain different numbers frame signal processing generally retains its right to exist.
of radio frames. To gain more insight into this matter, packet Pipelining is a viable alternative only if repetitive signal pro-
size statistics of some tangible system have to be known. For cessing tasks can be identified. If so, GDP should be used.
the example of IEEE 802.11a, it has been shown [37] that As for the repetitive task in isolation, HFP is systematically
small NF (values of 10 and below) occur in the great majority outperformed by GDP. Even frame-by-frame signal process-
of packets, so smart signal processing of small-sized packets ing (which is independent of the number of repetitions) may
is indeed an important issue in established WLAN standards. show higher speedup than HFP.
Given its superior speedup performance, GDP should be the Here, pipelining has been employed as a technique for
first choice for packet-oriented signal processing. software execution. However, additional work on Mod-SDR
[34] provides strong hints that hardware subsystem pipelin-
5. SUMMARY AND CONCLUSION ing also helps reducing dynamic power dissipation in CMOS
hardware, at the same time keeping speedup high. Therefore,
As a starting point, some technical and commercial bound- whenever signal processing in wireless communications is
ary conditions of SDR have been briefly reviewed. It follows repetitive in nature, the insertion of pipelining is the prefer-
from this account that certain design issues, which are related able design guideline for Mod-SDR systems.
Future research directions include the improvement of MTT-S International Microwave Symposium Digest (IMS ’01),
spectral partitioning for direct comparison with the (non- vol. 2, pp. 1085–1088, IEEE, Phoenix, Ariz, USA, May 2001.
pipelined) KL approach and a more comprehensive study of [15] A. Wiesler, Parametergesteuertes Software Radio für Mobil-
funksysteme, Ph.D. dissertation, Forschungsberichte aus dem
terminal behavior in packet-oriented networks. Further on, a
Institut für Nachrichtentechnik, Universität Karlsruhe (TH),
suitable extension of the current hardware model to hetero- Karlsruhe, Germany, May 2001.
geneous multiprocessor systems and interconnect topologies [16] M. Beach, J. MacLeod, and P. Warr, “Radio frequency trans-
other than a bus would advance the design theory of modular lation for software defined radios,” in Software Defined Radio:
software-defined radio. Enabling Technologies, W. Tuttlebee, Ed., pp. 25–78, John Wi-
ley & Sons, London, UK, 2002.
[17] P. B. Kennington and L. Astier, “Power consumption of A/D
REFERENCES
converters for software radio applications,” IEEE Trans. Veh.
[1] J. Mitola, “The software radio architecture,” IEEE Commun. Technol., vol. 49, no. 2, pp. 643–650, 2000.
Mag., vol. 33, no. 5, pp. 26–38, 1995. [18] J. Singh, “High speed analog-to-digital converter for software
[2] IEEE J. Select. Areas Commun., vol. 17, no. 4, 1999, Special radio applications,” in Proc. IEEE 11th International Sympo-
Issue on Software Radio. sium on Personal, Indoor, and Mobile Radio Communications
[3] IEEE Commun. Mag., vol. 37, no. 2, 1999, Special Issue on (PIMRC ’00), vol. 1, pp. 39–42, IEEE, London, UK, Septem-
Software Radio. ber 2000.
[4] A. Ivers and D. Smith, “A practical approach to the imple- [19] G. Ahlquist, M. Rice, and B. Nelson, “Error control coding in
mentation of multiple radio configurations utilizing reconfig- software radios: an FPGA approach,” IEEE Personal Commu-
urable hardware and software building blocks,” in Proc. IEEE nications, vol. 6, no. 4, pp. 35–39, 1999.
Military Communications Conference (MILCOM ’97), vol. 3, [20] M. Valenti, “An efficient software radio implementation of the
pp. 1327–1332, IEEE, Monterey, Calif, USA, November 1997. UMTS turbo codec,” in Proc. IEEE 12th International Sympo-
[5] A. Kountouris, C. Moy, L. Rambaud, and P. Le Corre, “A sium on Personal, Indoor, and Mobile Radio Communications
reconfigurable radio case study: a software based multi- (PIMRC ’01), vol. 2, pp. G108–G113, IEEE, San Diego, Calif,
standard transceiver for UMTS, GSM, EDGE and Bluetooth,” USA, September 2001.
in Proc. IEEE Vehicular Technology Conference (VTC ’01), [21] V. Thara and M. Siddiqi, “Power efficiency of software radio
vol. 2, pp. 1196–1200, Atlantic City, NJ, USA, October 2001. based turbo codec,” in Proc. IEEE Region 10 Conference on
[6] O. Faust, B. Sputh, D. Nathan, S. Rezgui, A. Weisensee, and Computers, Communications, Control, and Power Engineering
A. Allen, “A single-chip supervised partial self-reconfigurable (TENCON ’02), vol. 2, pp. 1060–1063, IEEE, Beijing, China,
architecture for software defined radio,” in Proc. 17th In- October 2002.
ternational Symposium on Parallel and Distributed Processing [22] A. Wiesler and F. K. Jondral, “A software radio for 2nd and 3rd
(IPDPS ’03), pp. 191–191, IEEE, Nice, France, April 2003. generation systems,” IEEE Trans. Veh. Technol., vol. 51, no. 4,
[7] H. Miranda, P. Pinto, and S. Silva, “A self-reconfigurable re- pp. 738–748, 2002.
ceiver architecture for software radio systems,” in Proc. IEEE [23] F. K. Jondral, “Parametrization—a technique for SDR imple-
Radio and Wireless Conference (RAWCON ’03), pp. 241–244, mentation,” in Software Defined Radio: Enabling Technologies,
IEEE, Boston, Mass, USA, August 2003. W. Tuttlebee, Ed., pp. 232–256, John Wiley & Sons, London,
[8] A. Pacifici, C. Vendetti, F. Frescura, and S. Cacopardi, “A re- UK, 2002.
configurable channel codec coprocessor for software radio [24] J. Mitola, “Software radio architecture: a mathematical per-
multimedia applications,” in Proc. International Symposium spective,” IEEE J. Select. Areas Commun., vol. 17, no. 4,
on Circuits and Systems (ISCAS ’03), vol. 2, pp. II-41–II-44, pp. 514–538, 1999.
IEEE, Bangkok, Thailand, May 2003. [25] J. Mitola, “Software radios—survey, critical evaluation and
[9] T. Hentschel and G. Fettweis, “Sample rate conversion for future directions,” in Proc. National Telesystems Conference
software radio,” IEEE Commun. Mag., vol. 38, no. 8, pp. 142– (NTC ’92), pp. 13/15–13/23, IEEE, Washington, DC, USA,
150, 2000. May 1992.
[10] W. Abu-Al-Saud and G. Stuber, “Efficient sample rate conver- [26] C. Dick, “Reinventing the signal processor,” Xcell Journal,
sion for software radio systems,” in Proc. IEEE Global Telecom- vol. 45, pp. 72–75, Spring 2003.
munications Conference (GLOBECOM ’02), vol. 1, pp. 559– [27] P. Galicki, “FPGAs have the multiprocessing I/O infrastruc-
563, IEEE, Taipeh, Taiwan, Republic of China, November ture to meet 3G base station design goals,” Xcell Journal,
2002. vol. 45, pp. 80–84, Spring 2003.
[11] W. Abu-Al-Saud and G. Stuber, “Modified CIC filter for sam- [28] “Software communications architecture specification,
ple rate conversion in software radio systems,” IEEE Signal jtrs-5000sca v2.2.1,” Joint Tactical Radio System (JTRS)
Processing Lett., vol. 10, no. 5, pp. 152–154, 2003. Joint Program Office, April 2004, [Online] available:
[12] J. Ming, H. Y. Weng, and S. Bai, “An efficient IF architecture http://jtrs.army.mil.
for dual-mode GSM/W-CDMA receiver of a software radio,” [29] A.-R. Rhiemeier and F. K. Jondral, “Mathematical modeling
in Proc. IEEE International Workshop on Mobile Multimedia of the software radio design problem,” IEICE Transactions on
Communications (MoMuC ’99), pp. 21–24, IEEE, San Diego, Communications, vol. E86-B, no. 12, pp. 3456–3467, 2003,
Calif, USA, November 1999. Special Issue on Software Defined Radio Technology and Its
[13] J. Dodley, R. Erving, and C. Rice, “In-building software radio Applications.
architecture, design and analysis,” in Proc. IEEE 11th Interna- [30] A.-R. Rhiemeier and F. K. Jondral, “A software partitioning al-
tional Symposium on Personal, Indoor, and Mobile Radio Com- gorithm for modular software defined radio,” in Proc. 6th In-
munications (PIMRC ’00), vol. 1, pp. 479–483, IEEE, London, ternational Symposium on Wireless Personal Multimedia Com-
UK, September 2000. munications (WPMC ’03), pp. 42–46, Yokosuka, Japan, Octo-
[14] W. Schacherbauer, A. Springer, T. Ostertag, C. Ruppel, and R. ber 2003.
Weigel, “A flexible multiband frontend for software radios us- [31] A.-R. Rhiemeier and F. K. Jondral, “On the design of modu-
ing high IF and active interference cancellation,” in Proc. IEEE lar software defined radio systems,” in Proc. IEE Colloquium
on DSP Enabled Radio, Institute for System Level Integration Signal Processing Research (CCSPR), New Jersey Institute of Tech-
(ISLI), Alba Campus, Livingston, Scotland, UK, September nology, Newark, NJ, USA, and received a Dipl.-Ing. degree in elec-
2003. trical engineering from the Universität Karlsruhe (TH), Germany.
[32] A.-R. Rhiemeier and F. K. Jondral, “Enhanced resource uti- Subsequently, he committed himself to a teaching and research as-
lization in software defined radio terminals,” in Interna- sistantship position at the Institut für Nachrichtentechnik. In 2004,
tionales Wissenschaftliches Kolloquium (IWK ’03), Technische he received a Ph.D. degree summa cum laude in telecommuni-
Universität, Ilmenau, Germany, September 2003. cations for his work on software-defined radio architectures and
[33] U. Berthold, A.-R. Rhiemeier, and F. K. Jondral, “Spectral par- algorithms. His current professional interests include design flow
titioning for modular software defined radio,” in IEEE 59th methodologies and the productization of advanced concepts in
Vehicular Technology Conference (VTC ’04), vol. 2, pp. 1218– communications such as MIMO and software-defined radio. He
1222, Milano, Italy, May 2004. has been an IEEE Member for 11 years and served as the Chairman
[34] A.-R. Rhiemeier and F. K. Jondral, “Software partitioning and
of the IEEE Student Branch Karlsruhe for two years.
hardware architecture for modular SDR systems,” in Proc.
Software Defined Radio Technical Conference and Product Ex-
position (SDR ’03), vol. 2, pp. 9–15, 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 defined radio,” in
Proc. IEEE/Sarnoff Symposium on Advances in Wired and Wire-
less Communication (SARNOFF ’04), pp. 201–204, Princeton,
NJ, USA, April 2004.
[36] A.-R. Rhiemeier, “A comparison of scheduling approaches in
modular software defined radio,” in Proc. 3rd Karlsruhe Work-
shop on Software Radios (WSR ’04), pp. 33–38, Karlsruhe, Ger-
many, March 2004, also appeared as reprint in: Frequenz: Jour-
nal of Telecommunications, vol. 58, no. 5/6, pp. 115–120, 2004.
[37] A.-R. Rhiemeier, T. Weiss, and F. K. Jondral, “Half-frame
pipelining for modular software defined radio,” in Proc. IEEE
15th International Symposium on Personal, Indoor, and Mobile
Radio Communications (PIMRC ’04), vol. 3, pp. 1664–1668,
IEEE, Barcelona, Spain, September 2004.
[38] A.-R. Rhiemeier, T. Weiss, and F. K. Jondral, “A simple and ef-
ficient solution to half-frame pipelining for modular software
defined radio,” in Proc. Software Defined Radio Technical Con-
ference and Product Exposition (SDR ’04), vol. A, pp. 119–125,
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. 107–114, San Jose, Calif, USA, November
2001.
[40] M. Fiedler, “Algebraic connectivity of graphs,” Czechoslovak
Mathematical Journal, vol. 23, no. 98, pp. 298–305, 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-
fications,” 1999 Edition, and IEEE 802.11a-1999, High-speed
Physical Layer in the 5 GHz Band.
Arnd-Ragnar Rhiemeier received a first

degree in electrical engineering from the
Ruhr-Universität Bochum, Germany, in
1995, then continued his studies at the
Universität 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 field of pattern recog-
nition. In 1998, he resumed his graduate studies at the Institut
für Nachrichtentechnik, Universität Karlsruhe (TH). In 1999, he
completed his final project at the Center for Communications and
c 2005 B. Dong and X. Wang
Adaptive Mobile Positioning in WCDMA Networks
B. Dong
Department of Electrical and Computer Engineering, Queen’s 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 difference 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 first-order autoregressive (AR) vector process with an additional
discrete state variable as the motion offset, 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 efficient auxiliary particle filtering 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 filter.
1. INTRODUCTION a gap between its performance and the optimal performance

since it is based on a linear approximation of the highly non-
Mobile positioning [1, 2, 3, 4], that is, estimating the location linear mobility model. Moreover, that work deals with the
of a mobile user in wireless networks, has recently received static scenario only and does not address mobility tracking in
significant attention due to its various potential applications a dynamic environment. In [2, 9, 10], the extended Kalman
in location-based services, such as location-based billing, in- filter (EKF) is used to track the user mobility. It is well known
telligent transportation systems [5], and the enhanced-911 that the EKF is based on linearization of the underlying non-
(E-911) wireless emergence services [6]. In addition to fa- linear dynamic system and often diverges when the system
cilitating these location-based services, the mobility infor- exhibits strong nonlinearity.
mation can also be used by a number of control and man- On the other hand, the recently emerged sequential
agement functionalities in a cellular system, such as mobile Monte-Carlo (SMC) methods [11, 12] are powerful tools for
location indication, handoff assistance [3], transmit power online Bayesian inference of nonlinear dynamic systems. The
control, and admission control. SMC can be loosely defined as a class of methods for solv-
ing online estimation problems in dynamic systems, by re-
Various mobile positioning schemes have been proposed
cursively generating Monte-Carlo samples of the state vari-
in the literature. Typically, they are based on the measure-
ables or some other latent variables. In [3], an SMC algo-
ments of received signal strength [7], time of arrival (TOA)
rithm for mobility tracking and handoff in wireless cellular
or time difference of arrival (TDOA) [8], and angle of arrival
networks is developed. In [8], several SMC algorithms for
(AOA) [4]. In [4], a hybrid TDOA/AOA method is proposed
positioning, navigation, and tracking are developed, where
and the mobile user location is calculated using a two-step
the mobility model is simpler than the one used in [3]. Note
least-square estimator. Although this scheme offers a higher
that in both works, the trial sampling density is based only
location accuracy than the pure TDOA scheme, there is still
on the prior distribution and does not make use of the mea-
surement information, which renders the algorithms less ef-
This is an open access article distributed under the Creative Commons
ficient. Moreover, the model parameters are assumed to be
Attribution License, which permits unrestricted use, distribution, and perfectly known, which is not realistic for practical mobile
reproduction in any medium, provided the original work is properly cited. positioning systems.
In this paper, we propose to employ a more efficient SMC as a first-order discrete-time Markov chain with finite state
method, the auxiliary particle filter, to jointly estimate both S = {S1 , S2 , . . . , SN } and the transition probability matrix
the mobility information (location, velocity, acceleration, A [ai, j ], ai, j P(sk = S j | sk−1 = Si ). It is assumed that
and the state sequence of commands) and the unknown sys- ai, j = p for i = j and ai, j = (1 − p)/(N − 1) for i = j, where
tem parameters. We assume the mobility estimation is based N is the total number of states. The correlated random ac-
on TDOA measurements at the mobile station (MS) and celerator rk is modelled as the first-order autoregressive (AR)
AOA measurements as well as the received signal strength model, that is, rk = αrk−1 + wk , where α is the AR coefficient,
measurements in the neighbor base stations (BSs). All these 0 < α < 1, and wk is a Gaussian noise vector with covariance
measurements are available in WCDMA networks. The re- matrix σw2 I.
mainder of this paper is organized as follows. In Section 2, Based on the above discussion, the motion model can be
we describe the nonlinear dynamic system model under con- expressed as
sideration, and present the mathematical formulation for the
 
problem of mobility tracking in a WCDMA wireless network. ∆t 2
In Section 3, we briefly introduce some background materi-   1 ∆t 0 0 0  
xk  2  xk−1
als on sequential Monte-Carlo techniques. The new mobil- vx,k   1 ∆t 0 0 
0  vx,k−1 
   0
 
r   r 
ity tracking algorithms are developed in Section 4. Section 5  x,k   0 0 α 0 0 0   x,k−1 

 = 
provides the simulation results; and Section 6 contains the  yk   
∆t  
2 yk−1 
  0 1 ∆t  
conclusions. v y,k  
0 0
2  v y,k−1 
 
r y,k 0 0 0 0 1 ∆t  r y,k−1

2. SYSTEM DESCRIPTIONS xk 0 0 0 0 0 α xk−1

2.1. Mobility model B
    (2)
Assume that a mobile of interest moves on a two- ∆t 2 ∆t 2
 2 0   0 
dimensional plane, and the motion state xk [xk , vx,k ,    2 
 ∆t 0   ∆t 0 
rx,k , yk , v y,k , r y,k ]T corresponds to the observation measure-    
   
 0 0   1 0 
ments at tk = t0 + ∆t · k, where ∆t is the sampling time in- +   sx,k +   wx,k .

∆t 2  s y,k  ∆t 2 
terval; xk and yk are, respectively, the horizontal and vertical    w y,k
 0 
 0 

Cartesian coordinates of the mobile position at time instance  2   2 
  sk   wk
k; vx,k and v y,k are the corresponding velocities; rx,k and r y,k  0 ∆t   0 ∆t 
are the corresponding accelerators. The discrete-time mov- 0 0 0 1

ing equation can be expressed as [2, 3] Cs Cw
 
∆t 2 In short,
      0 
xk 1 ∆t 0 0 xk−1  2



v  0   
0  vx,k−1   ∆t 0 
 x,k   1 0  ax,k−1 , xk = Bxk−1 + Cs sk + Cw wk . (3)
 =  +
 yk  0 0 1 ∆t   yk−1    ∆t 2 
 a y,k−1
v y,k v y,k−1  0 
0 0 0 1  2  2.2. Measurement model
0 ∆t
Some new features in WCDMA systems (e.g., cdma2000)
(1) such as network synchrony among the BSs, dedicated
reverse-link for each MS, adaptive antenna array for AOA es-
where ak [ax,k , a y,k ]T is the driving acceleration vector at timation, and forward link common broadcasting channel,
time k. Note that in mobility tracking applications, the time make several measurements available in practice for mobile
interval ∆t between two consecutive update intervals is typi- tracking.
cally on the order of several hundred symbol intervals to al- First of all, methods for determining the time differ-
low for the measurements of TDOA, AOA, and RSS. Such ence of arrival (TDOA) from the spread-spectrum signal,
a relatively large time scale also makes it possible to employ including the coarse timing acquisition with a sliding cor-
more sophisticated signal processing methods for more ac- relator or matched filter, and fine timing acquisition with a
curate mobility tracking. delay-locked loop (DLL) or tau-dither loop (TDL) [13, 14],
In practical cellular systems, a mobile user may have sud- can be applied in WCDMA systems. Coarse timing acqui-
den and unexpected changes in acceleration caused by traffic sition can achieve the accuracy within one chip duration
lights and/or road turn; on the other hand, the acceleration whereas fine synchronization by the DLL can achieve the
of the mobile may be highly correlated in time. In order to accuracy within fractional portion of chip duration. More-
incorporate the unexpected as well as the highly correlated over, in WCDMA systems, much higher chip rate is used
changes in acceleration, we model the motion of a user as a than that in IS-95 systems with shorter chip period, thereby
dynamic system driven by a command sk [sx,k , s y,k ]T and improving the precision of timing. Furthermore, with mul-
a correlated random acceleration rk [rx,k , r y,k ]T , that is, tiple antennas to collect the radio signal at the base sta-
ak = sk + rk . Following [2, 3], the command sk is modelled tion (in particular, phaseinformation), we could apply array
Adaptive Mobile Positioning in WCDMA Networks 345
signal processing algorithms (e.g., MUSIC or ESPRIT) [15] following measurement equation of the underlying dynamic
to estimate the angle of arrival (AOA). In addition, the re- model:
ceived signal strength indicator (RSSI) signal in WCDMA
systems contains the distance information between a mobile zk = h xk + vk , (5)
and a given base station, which is quantified by the large-
scale path-loss model with lognormal shadowing [16]. Note p p p β β β
where vk [nk,1 , nk,2 , nk,3 nτk,1 , nτk,2 , nk,1 , nk,2 , nk,3 ] with co-
that by averaging the received pilot signal, the rapid fluctu-
ation of multipath fading (i.e., small-scale fading effect) is variance matrix Q = diag(ηd I, ηt I, ηβ I); and h(xk )
mitigated. Based on the above discussion, we consider the [h1 (xk ), . . . , h8 (xk )] where the form of each hi (·), is given by
measurements for mobility tracking to include RSS pk,i , AOA (4). Note that the availability of TDOA and AOA will enhance
βk,i at the BSs, and TDOA τk,i fed back from MS. Denote the mobility tracking accuracy. In practice, if in some served
mobiles such information is not available, mobility tracking
Dk,i = [(xk − ai )2 + (yk − bi )2 ]1/2 , where (ai , bi ) is the po-
can still be performed based only on the RSS measurement,
sition of the ith BS. We have
with less accuracy.
p
pk,i = p0,i − 10η log Dk,i + nk,i , i = 1, 2, 3,
2.3. Problem formulation
1
τk,i = Dk,i − Dk,1 + nτk,i , i = 2, 3, (4)
Based on the discussions above, the nonlinear dynamic sys-
c tem under consideration can be represented by a jump-

y k − bi β Markov model as follows:
βk,i = tan(−1) + nk,i , i = 1, 2, 3,
xk − ai
sk ∼ MC(π, A), xk = Bxk−1 + Cs sk + Cw wk , zk = h(xk )+vk ,
where i is the BS index; p0,i is a constant determined by (6)
p
the wavelength and the antenna gain of the ith BS; nk,i ∼
N (0, ηd ) is the logarithm of the shadowing component, where MC(π, A) denotes a first-order Markov chain with
which is modelled as Gaussian distribution; c is the speed of initial probability vector π and transition matrix A. De-
light and η is the path-loss factor; nτk,i ∼ N (0, ητ ) is the mea- note the observation sequence up to time k as Zk
surement noise of TDOA between the ith BS and the serving [z1 , z2 , . . . , zk ], the corresponding discrete state sequence
β
BS; and nk,i ∼ N (0, ηβ ) is the estimation error of AOA at the Sk [s1 , s2 , . . . , sk ], and the continuous state sequence
ith BS. The noise terms in (4) are assumed to be white both Xk [x1 , x2 , . . . , xk ]. Let the model parameters be θ =
in space and in time. {π, A, ηw , ηd , ηt , ηβ }. Given the observations Zk up to time
Denote the measurements at time instance k as zk k, our problem is to infer the current position and velocity.
[pk,1 , pk,2 , pk,3 , τk,2 , τk,3 , βk,1 , βk,2 , βk,3 ]T . Then, we have the This amounts to making inference with respect to

p xk | Zk = ··· p s1 , . . . , sk , x1 , . . . , xk−1 | Zk dx1 , . . . , dxk−1
Sk ∈S k

k

(7)
∝ ··· p zi | xi p xi | xi−1 , si p si | si−1 dx1 , . . . , dxk−1 .
Sk ∈S k i=1
i.i.d. i.i.d.
The above exact expression of p(xk | Zk ) involves very high- where vk ∼ Nc (0, ηv I), εk ∼ Nc (0, ηε I), and sk is the
dimensional integrals and the dimensionality grows linearly discrete hidden state evolving according to a discrete-time
with time, which is prohibitive to compute in practice. In Markov chain with initial probability vector π and transi-
what follows, we resort to the sequential Monte-Carlo tech- tion probability matrix A. Denote yk {xk , sk } and the
niques to solve the above inference problem. system parameters θ = {π, A, ηv , ηε }. Suppose we want to
make an online inference about the unobserved states Yk =
3. BACKGROUND ON SEQUENTIAL MONTE CARLO (y1 , y2 , . . . , yk ) from a set of available observations Zk =
(z1 , z2 , . . . , zk ). Monte-Carlo methods approximate such in-
Consider the following jump-Markov model: ( j)
ference by drawing m random samples {Yk }mj=1 from the
posterior distribution p(Yk | Zk ). Since sampling directly
xk = A sk xk−1 + B sk vk ,
(8) from p(Yk | Zk ) is often not feasible or computationally
zk = C sk xk + D sk εk , too expensive, we can instead draw samples from some trial
sampling density q(Yk | Zk ), and calculate the target infer- weighted with respect to p(Yk | Zk ). In [18], it is shown that
ence E p {ϕ(Yk ) | Zk } using samples drawn from q(·) as ( j)
the optimal trial distribution is p(yk | Yk−1 , Zk ), which min-
imizes the conditional variance of the importance weights.
1 ( j) ( j)
m The SMC recursion at time k is as follows [17, 19].

E p ϕ Yk | Zk ∼= w ϕ Yk , (9) For j = 1, . . . , m,
Wk j =1 k
( j)
(i) draw a sample yk from the trial distribution
( j) ( j) ( j) m ( j)
where wk = p(Yk | Zk )/q(Yk | Zk ), Wk = j =1 wk , ( j)
( j)

( j) ( j)
and the pair {Yk , wk }mj=1 is called a set of properly weighted p yk | Yk−1 , Zk ∝ p zk | yk p yk | yk−1
samples with respect to the distribution p(Yk | Zk ) [17]. (10)
( j) ( j) ( j)
Suppose a set of properly weighted samples {Yk−1 , = p zk | xk , sk p xk | xk−1 , sk p sk | sk−1 ,
( j) m
wk−1 } j =1 with respect to p(Yk−1 | Zk−1 ) has been drawn
( j) ( j) ( j)
at time (k − 1), the sequential Monte-Carlo (SMC) proce- and let Yk = (Yk−1 , yk ),
( j) ( j)
dure generates a new set of samples {Yk , wk }mj=1 properly (ii) update the importance weight
( j) ( j)
( j)

wk ∝ wk−1 p zk | Yk−1 , Zk−1
( j)
N ( j)
( j)
(11)
= wk−1 p sk | sk−1 p zk | xk , sk , Zk−1 p xk | xk−1 , sk dxk .
sk =1
( j) ( j) ( j) ( j) ( j) ( j)
Apparently, it is difficult to use such an optima trial sampling ρk ∝ wk−1 ψ(xk−1 , sk−1 , zk ). Since the term ψ(xk−1 , sk−1 , zk )
density because the importance weight update equation does ( j) ( j)
is independent of sk and xk , we use it as the p.d.f. for
not admit a closed-form and involves a high-dimension inte- generating the auxiliary index κk before we sample the state
gral for each sample stream [19]. To approximate the integral variables (sk , xk ). Such a scheme is termed as the auxiliary
in (11), we use particle filter [20], where some auxiliary variable is intro-
duced in the sampling space such that the trial distribu-
( j)

p zk | xk , sk , Zk−1 p xk | xk−1 , sk dxk tion for the auxiliary variable can make use of the cur-
rent measurement zk . In order to utilize the observation in
( j)
( j)

the trial sampling density of sk , we sample sk according to
≈ p zk | xk , sk , Zk−1 δ xk = µk xk−1 , sk dxk ( j) ( j)
(κ ) (κ )
( j)
( j)
p(zk | µk (xk−k1 , sk ))p(sk | sk−k1 ) and sample xk according to
= p zk | Zk−1 , µk xk−1 , sk , ( j)
(κ ) ( j)
p(xk | Xk−k1 , sk ). The importance weights are then updated
(12) according to
( j)
where µk (xk−1 , sk ) is the mean of p(xk | xk−1 , sk ). Using (12),
( j) ( j)
( j)
(κ ) ( j)
( j)
( j)
(κ )

the importance weight update is approximated by ( j)
p zk | xk p xk | xk−k1 , sk p sk | sk−k1
wk ∝ (κ( j) ) ( j) ( j) (κ( j) ) ( j)
(κ ) ( j)
p zk | µk xk−k1 , sk p sk | sk−k1 p xk | xk−k1 , sk
( j) ( j)
N ( j)
( j)
( j)

wk ≈ wk−1 p zk | µ k xk−1 , sk , Zk−1 p sk | sk−1 . p zk | xk
( j)
sk =1

= ( j) .
( j) ( j)
p zk | µk xk−k1 , sk
(κ ) ( j)
ψ xk−1 ,sk−1 ,zk
(13) (14)
To make the SMC procedure efficient in practice, it is Considering the jump-Markov model (8), we have
( j) ( j) ( j)
( j) (κ ) (κ )
necessary to use a resampling procedure as suggested in µk (xk(κ−1 ) , sk ) = E{xk | sk , Xk−k1 } = A(sk )xk−k1 . The
[17, 18]. Roughly speaking, the aim of resampling is to du- auxiliary particle filter algorithm at the kth recursion is
plicate the sample streams with large importance weights summarized in Algorithm 1.
while eliminating the streams with small ones. In [19], it is If the system parameter θ is unknown, we need to aug-
( j)
suggested that we resample {Yk−1 } according to the weights ment the unknown parameter θ to the state variable yk as
(i) For j = 1, . . . , m and sk = 1, . . . , N, calculate the trial (i) For j = 1, . . . , m,

( j) ( j) ( j) ( j)
sampling density ρk ∝ wk−1 ψ(xk−1 , sk−1 , zk ). (a) draw samples of the unknown parameter {θ ( j) }mj=1
(ii) For j = 1, . . . , m, from p(θ | Yk−1 , Zk−1 ),
( j)
( j) ( j)
(a) draw the auxiliary index κk with probability ρk , (b) calculate the auxiliary variable sampling density
(b) draw a sample sk from the trial distribution ( j) ( j) ( j)
( j) ( j)
ρk ∝ wk−1 Nsk =1 p(zk | µk (xk−1 , sk ), θ ( j) )p(sk |
(κ ) (κ )
p(zk | µk (xk−k1 , sk ))p(sk | sk−k1 ) and let ( j)
sk−1 , θ ( j) ).
( j)
(κk )
( j)
Sk = (Sk−1 , sk ),
( j)
(ii) For j = 1, . . . , m,
( j) ( j)
(c) draw a sample xk from the trial distribution (a) draw the auxiliary index κk with probability ρk ,
( j) ( j) ( j)
(κ ) ( j) ( j) (κ ) ( j) (b) draw a sample sk from the trial distribution
p(xk | Xk−k1 , sk ) and let Xk = (Xk−k1 , xk ), ( j) ( j)
(d) update the importance weight (κ ) (κ )
( j)
p(zk | µk (xk−k1 , sk ), θ ( j) )p(sk | sk−k1 , θ ( j) ),
( j) ( j) (κ ) ( j) ( j)
wk ∝ p(zk | xk )/ p(zk | µk (xk−k1 , sk )). (c) draw a sample xk from the trial distribution
( j)
(κ ) ( j) ( j) ( j) ( j)
p(xk | Xk−k1 , sk , θ ( j) ) and let yk = (sk , xk ) and
Algorithm 1: The auxiliary particle filter algorithm at the kth re- ( j)
( j) (κk ) ( j)
cursion. let Yk = (Yk−1 , yk ),
(d) update the importance weight
( j)
the new state variable. Therefore, we have to sample from the ( j) ( j)
wk ∝ p(zk | xk , θ ( j) )/ p(zk | µk (xk−k1 , sk ), θ ( j) ),
(κ ) ( j)
joint density ( j)
(e) update the sampling density p(θ | Yk , Zk ) based on
( j) ( j)
( j)
( j)
( j)
p(θ | Yk−1 , Zk−1 ), yk and zk .
p yk , θ | Yk−1 , Zk = p yk | Yk−1 , zk , θ p θ | Yk−1 , Zk−1
( j)
Algorithm 2: The auxiliary particle filter algorithm of the kth re-
∝ p zk | yk , θ p yk | yk−1 , θ cursion for the case of unknown parameters.
( j)

× p θ | Yk−1 , Zk−1 .
( j)
(15) where Wk = mj=1 wk . Following the auxiliary particle filter
framework discussed in Section 3, we choose the sampling
And the importance weights are updated according to density for generating the auxiliary index κk as

( j) ( j)
wk ∝ wk−1 p zk | Yk−1 , Zk−1 , θ ( j)
( j) q κk = j
( j) ( j)
( j)

N ∝ wk−1 p zk | µk xk−1 , s p s | sk−1 , j = 1, . . . , m.
( j) ( j) ( j)
≈ wk−1 p zk | µk xk−1 , sk , θ ( j) p sk | sk−1 , θ ( j) . s∈S
sk =1 (18)
(16)
Considering the motion equation (3) and the measurement
( j) ( j)
For each sample stream j, the trial sampling density for the equation (5), we have µk (xk−1 , s) = Bxk−1 +Cs s. Next we draw
state variable (sk , xk ) and the importance weight update are a sample of state sk from the trial distribution
both based on the sampled unknown parameter θ ( j) . At the

end of the kth iteration, we update the trial sampling density ( j)
(κ ) (κ )
( j)
q sk = s ∝ p zk | µk xk−k1 , s · p s | sk−k1
( j) ( j) ( j)
p(θ | Yk , Zk ) based on p(θ | Yk−1 , Zk−1 ), yk and zk . The ( j) (19)
auxiliary particle filter algorithm at the kth recursion for the (κ )
= φ h µk xk−k1 , s , Q · a (κ( j) ) ,
case of unknown parameters is summarized in Algorithm 2. sk−k1 ,s
4. NEW MOBILITY TRACKING ALGORITHM where φ(µ, Σ) denotes the p.d.f. of a multivariate Gaussian
distribution with mean µ and covariance Σ. The trial sam-
4.1. Online estimator with known parameters pling density for xk is given by
We next outline the SMC algorithm for solving the prob-
lem of mobility tracking based on the jump-Markov model ( j)
( j)

(κ ) ( j) (κ ) ( j)
given by (6). Let yk = (xk , sk ), Xk = (x1 , x2 , . . . , xk ), Sk = p xk | xk−k1 , sk = φ Bxk−k1 + Cs sk , ηw Cw CTw . (20)
(s1 , . . . , sk ), Yk = (y1 , . . . , yk ), and Zk = (z1 , . . . , zk ). The
aim of mobility tracking is to estimate the posterior distri- And the importance weight is updated according to
bution of p(Yk | Zk ). Using SMC, we can obtain a set of
( j) ( j)
Monte-Carlo samples of the unknown states {Yk , wk }mj=1 p zk | xk
( j)
( j)
that are properly weighted with respect to the distribution wk ∝ ( j) , (21)
(κ )
p(Yk | Zk ). The MMSE estimator of the location and veloc- p zk | µk xk−k1 , sk
ity at time k can then be approximated by
where p(zk | xk ) = φ(h(xk ), Q). Finally, we summarize the
1 ( j)
m

E xk | Zk ∼
( j) adaptive mobile positioning algorithm with known parame-
= x · wk , k = 1, 2, . . . , (17)
Wk j =1 k ters in Algorithm 3.
( j) variable and the unknown parameter from

(I) Initialization: for j = 1, . . . , m, draw the state vector x0
from the multivariate Gaussian distribution N (x0 , 10I) ( j)

( j)
and draw s0 uniformly from S; all importance weights p yk , θ | Yk−1 , Zk
( j)
are initialized as w0 = 1. ( j)
( j)

∝ p zk | yk , θ p yk | yk−1 , zk , θ p θ | Yk−1 , Zk−1 ,
(II) For k = 1, 2, . . .,
(a) for j = 1, . . . , m, calculate the trial sampling (24)
density for the auxiliary index according to (18),
(b) for j = 1, . . . , m, ( j)
( j) where p(θ | Yk−1 , Zk−1 ) is the trial sampling density for the
(i) draw an auxiliary index κk with the unknown parameter at time (k − 1) and can be decomposed
probability q(κk = j),
( j) as
(ii) draw a sample sk according to (19),
( j)
(iii) draw a sample xk according to (20), ( j)
( j)

( j) p θ | Yk−1 , Zk−1 = p π, A, ηw , ηv , ηt , ηβ | Yk−1 , Zk−1
(iv) update the importance weight wk according N
to (21), ( j)
( j)
( j)

( j) ( j) ( j)
(v) append yk = {xk , sk } to Y(κ
( j) )
=p π| s0 p ai | π, Sk−1 p ηw | Xk−1 , Zk
k−1 to form
( j) i=1
( j)
Yk = {Y(κ ) (
k−1 , yk }.
j)
( j)
( j)
( j)

× p ηd | Xk−1 , Zk−1 p ηt | Xk−1 , Zk−1 p ηβ | Xk−1 , Zk−1 .
Algorithm 3: Adaptive mobile positioning algorithm with known (25)
system parameters.
Suppose we have updated the trial sampling density for θ at
the end of time (k − 1). Based on the sampled parameters
( j) ( j) ( j) ( j) ( j) ( j)
Complexity θ k = {π ( j) , A( j) , ηw , ηd , ηt , ηβ } ∼ p(θ | Yk−1 , Zk−1 ) at
The major computation involved in Algorithm 3 includes time k, we draw samples of the auxiliary index κk , the dis-
evaluations of Gaussian densities (i.e., mN evaluations in crete state sk , and the continuous state xk according to (18),
(18), mN evaluations in (19), and m evaluations in (21))), (19), and (20) and update the importance weight using (21).
and simple multiplications (i.e., mN multiplications in (18) In (18), (19), (20), and (21), the known system parameter θ
( j)
(κ )
and mN multiplications in (19)). Note that Algorithm 3 is is replaced by θ k k and the noise covariance matrix Q is sub-
well suited for parallel implementations. ( j) ( j)
(κ ) (κ )
( j) ( j)
(κ )
stituted by Q(κk ) = diag(ηd k I, ηt k I, ηβ k I). The location
4.2. Online estimator with unknown parameters and velocity are estimated through (17) and the minimum
mean-squared error (MMSE) estimate of the unknown pa-
We next treat the problem of jointly tracking the state Yk and ( j) ( j)
the unknown parameters θ = {π, A, ηw , ηd , ηt , ηβ }. We first rameter θ at time k is given by θ̂ k = (1/Wk ) mj=1 θ k wk ,
( j)
specify the priors for the unknown parameters. For the initial where Wk = mj=1 wk . At the end of time k, we update the
probability vector π and the ith row of the transition prob- trial sampling density for θ as follows.
ability matrix A, we choose a Dirichlet distribution as their At the end of time k, we update the trial sampling density
priors: for the initial state probability vector π as
( j)

π ∼ D α1 , α2 , . . . , αN , p π | s0 ∼ D α1 + δs( j) −1 , α2 + δs( j) −2 , . . . , αN + δs( j) −N .
(22) 0 0 0
ai ∼ D α1 , α2 , . . . , αN , i = 1, . . . , N. (26)
Given the prior distribution of the ith row ai of the tran-
For the noise variances, ηw , ηd , ηt , and ηβ , we use the inverse
sition probability matrix A at the end of time (k − 1), that is,
chi-square priors: ( j) (k−1, j) (k−1, j) (k−1, j)
p(ai | π, Sk−1 ) ∼ D(αi,1 , αi,2 , . . . , αi,N ), at time k,
the trial sampling density for ai is updated according to
ηw ∼ χ −2 ν0,w , λ0,w , ηd ∼ χ −2 ν0,d , λ0,d ,
(23) ( j)
ηt ∼ χ −2 ν0,t , λ0,t , ηβ ∼ χ −2 ν0,β , λ0,β . ( j) ( j) ( j)
p ai | π, Sk ∝ p sk | π, Sk−1 , ai p ai | π, Sk−1

Suppose that at time (k − 1), we have m sample streams of
 (k−1, j) (k−1, j)
( j) ( j)
state Yk−1 and parameter θ, {Yk−1 , θ k−1 }mj=1 , and the asso- ∼ D  αi,1 + δs( j) −i δs( j) −1 , αi,2 + δs( j) −i δs( j) −2 , . . . ,
( j)
k −1 k

k−1 k

ciated importance weights {wk−1 }mj=1 , representing an im- (k, j)
αi,1
(k, j)
αi,2
portant sample approximation to the posterior distribution 
p(Yk−1 , θ | Zk−1 ) at time (k − 1). Note that here the index k (k−1, j) 
on the parameter samples indicates that they are drawn from αi,N + δs( j) δ ( j)
−i sk −N
.
the posterior distribution at time k rather than implying that
k−1
(k, j)
θ is time-varying. By applying Bayes’ theorem and consider- αi,N
ing the system equations (6), at time k, we sample the state (27)
And given the noise variance sampling density at time (k − 1), (I) Initialization: for j = 1, . . . , m, draw the samples of the
( j) ( j)
p(ηw | Xk−1 , Zk−1 ) ∼ χ −2 (νk−1,w , λk−1,w ), at time k, the trial initial probability vector π, the ith row ai of the
sampling density for ηw is updated according to transition probability matrix, the noise variance ηw , ηd ,
ηt , and ηβ according to their prior distributions in (22)
( j)
( j) ( j)
( j)
( j)
p ηw | Yk , Zk ∝ p xk | xk−1 , sk , ηw p ηw | Xk−1 , Zk−1 and (23), respectively. Draw the state vector x0 from
the multivariate Gaussian distribution N (x0 , 10I), and
( j) ( j)
∼ χ −2 νk−1 + 1, λk,w , draw s0 uniformly from S, all importance weights are
( j)
(28) initialized as w0 = 1.
(II) For k = 1, 2, . . .,
( j) ( j) 2 (a) for j = 1, 2, . . . , m, calculate the trial sampling
where λk,w = (ν0,w + k − 1)/(ν0,w + k)λk−1,w + i=1 (xk,3i − density for the auxiliary index according to (18),
αxk−1,3i )2 /2(ν0,w + k). Similarly, we have where the actually unknown parameter θ is
( j)
replaced by θ k−1 ,
( j) ( j) (b) for j = 1, 2, . . . , m,
p ηd | Yk , Zk ∼ χ −2 νk−1,d + 1, λk,d , (29) ( j)
(i) draw an auxiliary index κk with the
( j) ( j)
p ηt | Yk , Zk ∼ χ −2 νk−1,t + 1, λk,t , (30) probability q(κk = j),
( j)
( j)
( j)
(ii) draw a sample sk according to (19),
p ηβ | Yk , Zk ∼ χ −2 νk−1,β + 1, λk,β , (31) ( j)
(iii) draw a sample xk according to (20),
( j)
(iv) update the importance weights wk
( j) ( j) 3 according to (21),
where λk,d = (ν0,d + k − 1)/(ν0,d + k)λk−1,d + i=1 (pi,k − (κ
( j)
)
( j) ( j) ( j)
( j) 2 ( j) ( j) (v) append yk = {xk , sk } and Yk−k1 to form
hi (xk )) /3(ν0,d + k), λk,t = (ν0,t + k − 1)/(ν0,t + k)λk−1,t + ( j)
(κk )
2 ( j) 2 ( j)
( j)
Yk = {Yk−1 , yk },
( j)
i=1 (τi,k − hi+3 (xk )) /2(ν0,t + k) and λk,β = (ν0,β + k − (vi) update the trial sampling density for θ
( j) ( j)
1)/(ν0,β + k)λk−1,β + 3i=1 (βk,i − hi+5 (xk ))2 /3(ν0,β + k). Fi- according to (26), (28), (29), (30), and (31),
nally, we summarize the adaptive mobile positioning algo- (vii) sample the unknown system parameters
( j) ( j) ( j) ( j) ( j)
rithm with unknown system parameters Algorithm 4. θ k = (π ( j) , A( j) , ηw , ηd , ηt , ηβ ) according
to (26), (28), (29), (30), and (31),
respectively.
Complexity
Compared with the known parameter case, that is, Algorithm 4: Adaptive mobile positioning algorithm with un-
Algorithm 3, the additional computation in Algorithm 4 is known system parameters.
introduced by the updates of the trial densities of the un-
√
knowns and the draws of these parameters, which at it- standard deviation of TDOA ηt = 100/c; whereas in sce-
√
eration, involve 4m simple multiplications, as well as the nario 2, the standard deviation of AOA ηβ = 2/360, the
√
m(N + 1) samplings from the Dirichlet distribution and 4m standard deviation of TDOA ηt = 50/c; where c = 3 · 108
samplings from the inverse chi-square distribution. As noted m/s is the speed of light. In both scenarios, the base station
previously, since in mobility tracking applications the up- transmission power p0,i = 90 mW, the path-loss index η = 3,
date is performed at a time scale of several hundred symbols, and the number of samples m = 250. All simulation results
the above SMC-based tracking algorithm is feasible to imple- are obtained based on M = 50 random realizations.
ment in practice.
5.1. Performance comparison with existing techniques
We first compare the performance of the extended Kalman
5. SIMULATION
filter (EKF) mobility tracker [2], the standard particle fil-
Computer simulations are performed on a WCDMA ter mobility tracker [3], and the proposed auxiliary parti-
hexagon cellular network to assess the performance of the cle filter (APF) mobility tracker (Algorithm 3) in terms of
proposed adaptive mobile positioning algorithms. The net- the normalized mean-squared error (NMSE) assuming that
work under investigation contains 64 BSs with cell radius the system parameters are known. The NMSE is defined as

2 km. The mobile trajectories within the network are gener- NMSE = (1/L) Lk=1 ((x̂k − xk )2 +( ŷk − yk )2 )/(xk2 +yk2 ), where
ated randomly according to the mobility model described in L is the observation window size. The NMSE results based
Section 2.1 and fixed for all simulations. On the other hand, on the different observations (i.e., RSS only, RSS/AOA and
the pilot signals are generated randomly according to the ob- RSS/AOA/TDOA) for scenarios 1 and 2 are reported in Tables
servation model (5) for each simulation realization. Some 1 and 2, respectively. It is seen that both the standard PF and
parameters used in the simulations are the sampling interval the APF significantly outperform the EKF in the above two
∆t = 0.5 seconds; the correlation coefficient of the random scenarios under the same observations. In fact, the perfor-
accelerator in (3) is α = 0.6; the variance of each random mance gain varies from 5–10 dB for different scenarios and
variable in wk is ηw = 1; the standard deviation of lognor- observations. Moreover, by utilizing the current observations
√
mal shadowing ηd = 5 dB. We consider two scenarios. In in the trial sampling density, the APF demonstrates further
√
scenario 1, the standard deviation of AOA ηβ = 4/360, the improvement over the standard PF (roughly 3 dB).
Table 1: Performance comparisons between EKF, standard PF, and APF in terms of NMSE based on different 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 different 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
6600 400
6400 350
6200 300
6000 250
RSE (m)
Y 5800 200
5600 150
5400 100
5200 50
5000 0
3000 3500 4000 4500 5000 5500 6000 6500 7000 7500 0 50 100 150 200 250 300
X t
True Estimated RSS/AOA RSS

Estimated RSS Estimated RSS/TDOA/AOA RSS/AOA
RSS/TDOA/AOA
Figure 1: Estimated trajectories based on different observations for
scenario 1. Figure 2: The root-squared error as a function of time for different
mobile positioning schemes for scenario 1.
We also compare the APF mobility tracker (Algorithm 4) mance and the one based on RSS only performs the
with the standard PF mobility tracker assuming that the sys- worst. We report the corresponding root-squared error
tem parameters are unknown. The NMSE results for scenar- (RSE) as a function of time for scenarios 1 and 2 in
ios 1 and 2 are reported in Tables 1 and 2, respectively. It is Figures 2 and 3, respectively. RSE is defined as RSE =
seen that performance penalty due to unknown system pa- !
rameters is less than 3 dB whereas the APF is still 3-4 dB bet- ((x̂k − xk )2 + ( ŷk − yk )2 ). It is observed that by incorporat-
ter than the standard PF. ing the AOA measurements into the observation function,
the RSE is significantly reduced. Further RSE reduction is
5.2. Tracking performance of the proposed algorithm achieved by using additional TDOA measurements. Figures
In Figure 1, we compare the trajectories estimated by the 4 and 5 show the empirical cumulative distribution function
APF algorithm (Algorithm 4) based on RSS only, RSS/AOA, (CDF) of root-squared error (RSE) based on different ob-
and RSS/AOA/TDOA, respectively for scenario 1. It is seen servations (i.e., RSS only, RSS/AOA, and RSS/TDOA/AOA)
that the online estimation algorithm based on the combined measurements. It is seen that the estimated location based
observations RSS/AOA/TDOA achieves the best perfor- on RSS only is most likely to have large deviation from the
350 1
0.9
300
0.8
250 0.7
200 0.6
RSE (m)
CDF
0.5
150
0.4
100 0.3
0.2
50
0.1
0 0
0 50 100 150 200 250 300 0 200 400 600 800 1000 1200 1400
t Root-squared error (m)
RSS RSS
RSS/AOA RSS/AOA
RSS/TDOA/AOA RSS/TDOA/AOA
Figure 3: The root-squared error as a function of time for different Figure 5: CDF of root-squared error based on different mobile po-
mobile positioning schemes for scenario 2. sitioning schemes for scenario 2.
1 50
0.9 45
0.8 40
0.7
35
RMSE (m)
0.6
30
CDF
0.5
25
0.4
20
0.3
15
0.2
0.1 10
0 5
0 200 400 600 800 1000 1200 10 20 30 40 50 60 70 80 90 100
Root-squared error (m) σt
RSS RSS/AOA, scenario 1 RSS/AOA, scenario 2

RSS/AOA RSS/TDOA/AOA, scenario 1 RSS/AOA, scenario 2
RSS/TDOA/AOA
√
Figure 4: CDF of root-squared error based on different mobile po- Figure 6: RMSE as a function of σt ηt in Algorithm 4 using
sitioning schemes for scenario 1. different observations.
actual location whereas that based on RSS/TDOA/AOA has in both scenarios and the performance gain over that
the smallest outage probability. By comparing the estimation of RSS/AOA diminishes as the variance of TDOA mea-
performance in scenarios 1 and 2, it is seen that the algorithm surements increases. When the TDOA measurement noise
achieves better performance for scenario 2 due to the smaller variance is small, a large performance improvement by the
measurement noise. TDOA/AOA is achieved. However, when the AOA measure-
We also report the effect of the variance of TDOA ment error increases above a certain level, the performance
measurement on the estimation performance in Figure 6 improvements become negligible. The RMSE in scenario 2
in terms !of root mean-squared error (RMSE) defined as is smaller than that in scenario 1 in both RSS/AOA and

RMSE = (1/L) Lk=1 ((x̂k − xk )2 + ( ŷk − yk )2 ). It is seen that RSS/TDOA/AOA location because of a better accuracy in
the RMSE with RSS/TDOA/AOA monotonically increases AOA and TDOA measurements.
0.8 6. CONCLUSIONS
0.6 We have considered the problem of mobile user position-
ing under the sequential Monte-Carlo Bayesian framework.
a1,1
0.4
We have developed a new adaptive mobile positioning algo-
0.2
rithm based on the auxiliary particle filter algorithm. The al-
0 gorithm makes use of the measurements of time difference of
0 50 100 150 200 250 300 350 400 450 500
arrival, angle of arrival as well as received signal strength, all
Iteration no.
of which are available in practical WCDMA networks. The
proposed algorithm jointly tracks the unknown system pa-
0.7
0.6 rameters as well as the mobile position and velocity. Simu-
0.5 lation results show that the proposed algorithm has an ex-
0.4 cellent mobility tracking and parameter estimation perfor-
a2,3
0.3 mance and it significantly outperforms the existing mobility

0.2
0.1
estimation schemes.
0
0 50 100 150 200 250 300 350 400 450 500
Iteration no. ACKNOWLEDGMENTS
This work was supported in part by the US National Science
Foundation (NSF) under Grants DMS-0225692 and CCR-
Figure 7: Parameter tracking performance of the transition proba-
bility matrix A as a function of the iteration number for scenario 1. 0225826, and by the US Office of Naval Research (ONR) un-
der Grant N00014-03-1-0039.
REFERENCES
2
[1] J. Caffery and G. L. Stuber, “Subscriber location in CDMA
1.5 cellular networks,” IEEE Trans. Veh. Technol., vol. 47, no. 2,
pp. 406–416, 1998.
ηw,1
1
[2] T. Liu, P. Bahl, and I. Chlamtac, “Mobility modeling, loca-
0.5 tion tracking, and trajectory prediction in wireless ATM net-
works,” IEEE J. Select. Areas Commun., vol. 16, no. 6, pp. 922–
0 936, 1998.
0 50 100 150 200 250 300 350 400 450 500
[3] Z. Yang and X. Wang, “Joint mobility tracking and handoff in
Iteration no.
cellular networks via sequential Monte Carlo filtering,” IEEE
Trans. Signal Processing, vol. 51, no. 1, pp. 269–281, 2003.
1.4 [4] L. Cong and W. Zhuang, “Hybrid TDOA/AOA mobile user lo-
1.2 cation for wideband CDMA cellular systems,” IEEE Transac-
1 tions on Wireless Communications, vol. 1, no. 3, pp. 439–447,
0.8
ηw,2
2002.
0.6 [5] C.-D. Wann and Y.-M. Chen, “Mobile location tracking with
0.4 velocity estimation,” in Proc. 5th International Conference on
0.2
Intelligent Transportation Systems (ITS ’02), pp. 566–571, Sin-
0
0 50 100 150 200 250 300 350 400 450 500 gapore, Singapore, September 2002.
Iteration no. [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.,
Figure 8: Parameter tracking performance of the motion variance vol. 36, no. 4, pp. 30–37, 1998.
ηw as a function of the iteration number for scenario 1. [7] M. Hellebrandt and R. Mathar, “Location tracking of mobiles
in cellular radio networks,” IEEE Trans. Veh. Technol., vol. 48,
no. 5, pp. 1558–1562, 1999.
[8] F. Gustafsson, F. Gunnarsson, N. Bergman, et al., “Particle
We next illustrate the parameter tracking behavior of the filters for positioning, navigation, and tracking,” IEEE Trans.
Signal Processing, vol. 50, no. 2, pp. 425–437, 2002.
proposed adaptive mobile positioning algorithm with un- [9] M. McGuire and K. N. Plataniotis, “Dynamic model-based fil-
known parameters in scenario 1. The estimates of the pa- tering for mobile terminal location estimation,” IEEE Trans.
rameters a1,1 and a2,3 as a function of time index k for one Veh. Technol., vol. 52, no. 4, pp. 1012–1031, 2003.
vehicle trajectory are plotted in Figure 7. We also plot the es- [10] R. Togneri and L. Deng, “Joint state and parameter estimation
timates of the noise variances ηw,1 and ηw,2 in Figure 8. It is for a target-directed nonlinear dynamic system model,” IEEE
observed that although the initial estimates of the unknown Trans. Signal Processing, vol. 51, no. 12, pp. 3061–3070, 2003.
[11] P. M. Djuric, J. H. Kotecha, J. Zhang, et al., “Particle filtering,”
parameters are far from the actual value, after a short period
IEEE Signal Processing Mag., vol. 20, no. 5, pp. 19–38, 2003.
of time, the estimates of these unknown parameters converge [12] A. Doucet, N. de Freitas, and N. Gordon, Sequential Monte
to the true values, demonstrating the excellent tracking per- Carlo Method in Practice, Springer-Verlag, New York, NY,
formance of the proposed algorithm. USA, 2000.
[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. 67–94, 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. 1–3, pp. 89–105, 2002.
[18] A. Doucet, S. J. Godsill, and C. Andrieu, “On sequential
Monte Carlo sampling methods for Bayesian filtering,” Statis-
tics and Computing, vol. 10, no. 3, pp. 197–208, 2001.
[19] M. Davy, C. Andrieu, and A. Doucet, “Efficient particle fil-
tering for jump Markov systems. Application to time-varying
autoregressions,” IEEE Trans. Signal Processing, vol. 51, no. 7,
pp. 1762–1770, 2003.
[20] M. Pitt and N. Shepard, “Filtering via simulation: auxiliary
particle filters,” Journal of the American Statistical Association,
vol. 94, no. 446, pp. 590–599, 1999.
B. Dong received his M.S. degree in electrical engineering from

Queen’s 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. Wang’s 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.
c 2005 Bruce E. Carey-Smith et al.
Flexible Frequency Discrimination Subsystems

for Reconfigurable 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
Email: paul.a.warr@bristol.ac.uk
Phill R. Rogers
Email: phill.rogers@bristol.ac.uk
Mark A. Beach
Email: m.a.beach@bristol.ac.uk
Geoffrey S. Hilton
Email: geoff.hilton@bristol.ac.uk
Received 8 October 2004; Revised 14 March 2005
The required flexibility of the software-defined 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 filtering to reduce spurious spectral content in the transmitter and limit out-of-band interference in the receiver.
In this paper, recent advances in flexible, frequency-selective, circuit components applicable to reconfigurable SDR front ends
are reviewed. The paper contains discussion regarding the filtering 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 filters and narrowband tunable antennas. The suitability of these elements to the
mobile wireless SDR environment is discussed.
Keywords and phrases: software-defined radio, tunable antenna, reconfigurable front end, MEMS, tunable bandpass filter, tunable
bandstop filter.
1. INTRODUCTION mobile telephones, which are predefined to operate on a fixed

number of standards, SDR must be capable of adapting to
The intention of software-defined radio (SDR) is to pro- both current and future mobile telecommunications stan-
vide a flexible radio platform capable of operating over a dards [1]. The SDR concept imposes demanding require-
continuously evolving set of communications standards and ments on the transceiver front end.1 Current spectrum al-
modes. In contrast to the majority of currently available locations and recent regulatory reforms (e.g., [2]) suggest
This is an open access article distributed under the Creative Commons 1 The front end is that part of the radio which performs channelisation
Attribution License, which permits unrestricted use, distribution, and and up- and down-conversion. These functions may be performed in either
reproduction in any medium, provided the original work is properly cited. the analogue or digital domain.
Flexible Subsystems for Radio Front Ends 355
that transceiver operation from 600 MHz to 6 GHz will be major advantage of the wideband approach is the wide band-
required to cover existing and emerging telecommunication width at the analogue-to-digital converter interface, allowing
services.2 The front end must not only be capable of oper- concurrent operation at multiple frequencies. Both the re-
ating across this entire band but must do so whilst attaining ceiver and transmitter are thus able to operate on multiple
performance comparable with fixed frequency solutions. In standards simultaneously. However, design for wideband op-
order to achieve the wide frequency coverage required from eration leads to compromise in the transceiver performance,
the SDR front end, two approaches are possible: wideband caused by limitations in the front-end components. The non-
operation,3 with all filtering being carried out in the dig- linearity of the active analogue front-end components lim-
ital domain; or flexible narrowband operation, using tun- its the distortion-free dynamic range of both the transmitter
able narrowband analogue circuits. In this paper, it is argued and receiver. Wideband linearisation techniques for ampli-
that, when compared to wideband solutions, flexible narrow- fiers and mixers have demonstrated some linearity improve-
band operation holds significant advantages, and with cur- ment [6, 7] and are necessary to limit in-band intermodu-
rent technology would offer improved overall performance lation distortion. However, linearity improvements come at
in the SDR transceiver. A key component of narrowband a significant cost to overall radio efficiency. Alternatively, by
transceivers is appropriate filtering to reduce spurious spec- band-limiting the signal of interest, unwanted spectral prod-
tral content in the transmitter and limit out-of-band interfer- ucts can be suppressed, leading to a significant reduction in
ence in the receiver. The main focus of this paper is to review the overall linearity requirements of the transceiver front end
techniques for providing tunable frequency discrimination [8].
in the analogue front end of a narrowband reconfigurable Band limiting within the SDR environment must be flex-
SDR receiver. ible since both the signals of interest and the unwanted dis-
tortion products and interference will vary as a function of
1.1. Practical SDR front ends the user’s environment and choice of standard. In this pa-
Practical implementations of the SDR front end must cur- per, techniques of introducing flexible frequency discrimina-
rently include analogue circuitry to perform frequency up- tion into the transceiver front end are considered. Section 2
and down-conversion, amplification, and filtering. In the examines the different requirements for filtering within the
ideal implementation of SDR, maximum front-end flexibil- SDR transmitter and receiver. Tunable bandpass and band-
ity is achieved by the conversion of the analogue signals to stop filters are discussed in Sections 3.1 and 3.2, respectively.
and from the digital domain as close to the antenna as pos- Bandpass filters offer general frequency discrimination to re-
sible [1] employing direct-digital conversion. However, per- duce the wideband interference level while bandstop filters
forming direct-digital conversion at RF and microwave fre- are particularly suited to removing high-level interferers and
quencies places currently unachievable demands on the con- providing selective isolation between the transmit and re-
version hardware and digital processing circuitry. Commer- ceive bands. Tunable, narrowband antennas are not normally
cially available analogue-to-digital converters (ADCs) can associated with frequency discrimination; however, they are
achieve analogue bandwidths in the order of 70 MHz with tuned transducers between guided and unguided electro-
greater than 90 dB spurious-free dynamic range (SFDR)4 but magnetic waves and exhibit filtering behaviour. Narrowband
at the present rate of advance it will be some time before tunable antennas offer significant advantages over wideband
direct-digital conversion is possible for the wireless commu- antennas, one of which is the additional frequency discrim-
nications environment [3]. The alternative to directly sam- ination which they offer. These advantages are discussed in
pling the RF signal is to retain the analogue-digital divide Section 3.3.
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 2. FRONT-END REQUIREMENTS OF SDR
are being explored at both the component and system level The design of the air interface of narrowband transceivers
[4, 5]. concentrates on achieving compliance with a particular
standard’s receiver blocking mask and transmitter emission
1.2. Limitations of wideband front ends mask. The latter must be met in order to comply with the
Although having attractive features, a wideband front end requirements of the current operating standard while the
may not necessarily be the optimum solution for SDR. The former provides a guide as to the maximum signal levels
that will be encountered by the receiver. In conventional
(fixed standard) radio, fixed filters provide the necessary fre-
2 A more conservative goal, encompassing the majority of current stan- quency discrimination. An example of a typical transceiver
dards, still requires a two-octave frequency coverage from GSM at 800 MHz is shown in Figure 1. Some compromise in the performance
to the lower ISM band at 2.4 GHz. of individual filtering elements must be made to introduce
3 For the purposes of this paper, the term wideband is used to describe
components whose operational bandwidth (frequency range over which

the flexibility required for SDR. It may be possible, how-
they are impedance matched) covers all the communication standards of ever, to recover some of this through combining their ca-
interest; multioctave for SDR. pabilities in an intelligent way. This idea is explored in
4 For example, analog devices, AD6645.
Section 2.2.
Mixer
LNA
Band-select Image-reject
filter filter
Duplex Receiver
filter Frequency Local
down-conversion oscillator
Frequency Local
Isolator Transmitter up-conversion oscillator
Harmonic Band-select
rejection filter filter Preamplifier
PA
Mixer
Figure 1: Typical transceiver front end.
2.1. Frequency discrimination requirements in the the receiver will be desensitised. Secondly, leakage of the high
transmitter and receiver power transmit signal into the receiver can cause receiver
The high power amplification of transmitted signals leads overload. In both cases, the SDR front end must provide
to harmonic (HD) and intermodulation distortion (IMD) some form of flexible frequency discrimination in the trans-
products which contaminate the spectrum of the emitted mit and receive paths. The use of separate, independently
signal [9]. Some IMD products appear in-band and can- tuned, transmit and receive narrowband antennas will lead
not be filtered out, placing fundamental linearity require- to some improvement in the isolation. Tunable notch filters
ments on the power amplifier (PA). Unnecessary IMD can could be used to provide additional suppression.
be avoided by harmonic filtering before the PA. This prevents
harmonic energy generated in the up-conversion and pream- 2.2. Environment-aware frequency discrimination
plification stages from producing additional IMD in the PA. For any given operating standard, the SDR transceiver must
It also reduces the wideband noise requirements of these select an appropriate filtering profile. These profiles could
stages. Harmonic filtering after the PA is usually manda- be predefined for different standards, however, flexibility
tory in order to meet the emission mask for any given stan- in the front-end circuitry raises the possibility of intelli-
dard. Being well removed in frequency from the wanted sig- gent, real-time adaptation of the transceiver’s frequency dis-
nal, harmonic attenuation is usually straightforward using a crimination profile to the current operating environment.
lowpass or bandpass filter. However, the multioctave air in- By constantly assessing the user’s signal environment, the
terface of SDR demands that this filtering be tunable. The transceiver can respond with the appropriate level of sup-
greatest challenge for tunable filters in the transmitter, par- pression on a frequency-by-frequency basis. The potential
ticularly post-PA, is in tolerating high power levels. There are advantage of this scheme is that the necessary filtering can be
two factors to consider: the linearity of the tuning elements provided using lower specification circuit blocks. Through
and their absolute power handling capability. Both of these the intelligent and flexible combination of lower perfor-
factors must be considered when selecting the tuning mech- mance elements, filtering can be supplied where it is needed
anism for transmit filters. rather than by unnecessary blanket coverage.
In the receiver, the requirement for filtering is usually This concept has greater applicability to the receiver,
more stringent. Signals in adjacent channels are controlled by where the nature of the unwanted signals is changing in an
the current standard and can therefore be tolerated, assum- unpredictable way. The customary means of dealing with this
ing a degree of linearity in the active front-end components. unpredictability is to provide a maximum amount of filter-
However, out-of-band signals are not controlled; the only ing to match the blocking mask for a particular standard. The
guarantee is that they will fall below some specified maxi- blocking mask, however, assumes the worst case, where all
mum level, defined by the blocking mask. The required re- out-of-band signals are at the maximum interference level
jection is usually achieved using a bandpass filter. (usually 0 dBm). In reality, troublesome interference will be
In frequency division duplex (FDD) systems, the trans- limited to specific frequency bands which will change as a
mitter can interfere detrimentally with the receiver since both function of the environment. Wideband spectrum measure-
operate simultaneously. This can occur in two ways. Firstly, ments at a variety of high-communication traffic locations
the transmitter noise floor is generally much higher than in a typical European city indicate that seldom do multiple
the sensitivity of the receiver and, given inadequate isolation, signals approach the typical wideband blocking specification
0 1
Received signal power (dBm)
−10
−20
−30
−40 0.1
Probability
−50
−60
−70
−80 0.01
−90
−100
4 8 12 16 20 24 28 32 36 40 44 48 52 56 60
Frequency (MHz) ×102
0.001
−50 −45 −40 −35 −30 −25 −20 −15
Signal level L
Figure 2: Worst-case power profile.
N =1
N =2
N =3
simultaneously [10]. The worst-case power profile, com-
bined from measurements taken in 44 different locations, is Figure 3: Results of spectrum measurements taken in a typical Eu-
shown in Figure 2 where a zero dB gain, omnidirectional an- ropean city. The graph shows the probability that there will be more
tenna is assumed in order to calculate the absolute power lev- than N signals whose power is greater than the signal level L.
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- material placed on a nonferrous substrate, creating a two-
ceiving a number of high-level interfering signals reduces layered structure. The dielectric permittivity of the struc-
quickly with their average signal strengths (Figure 3). For ex- ture can be varied by applied external electric field. Currently
ample, the probability of receiving more than two interfering these materials exhibit high loss and require large bias volt-
signals above −25 dBm is less than 0.4%. The data used in ages, however, greater than 30% tuning range has been doc-
this analysis is the long-term cumulative maximum, so even umented and resonator Q’s greater than 800 can be obtained
lower instantaneous probabilities can be expected. These re- [11, 12].
sults indicate that, by providing a high level of narrowband The introduction and continuing improvement of mi-
suppression at a small number of frequencies, the wideband croelectromechanical systems (MEMS) has led to the possi-
interference level can be reduced significantly from the typ- bility of MEMS-tuned filters for applications where size and
ical receiver blocking specification. This, in turn, eases the tuning speed are critical. Recently, considerable attention has
general filtering requirements, reducing the absolute stop- been given to their use at high frequencies. MEMS for RF
band attenuation required of a front-end bandpass filter. This (RF MEMS) and microwave applications have been fabri-
may enable the use of a lower order filter or a more compact cated and characterised for frequencies up to 40 GHz [13].
implementation. The remaining sections of this paper look MEMS-tuned microwave filter design, although being a rel-
at the performance of various narrowband tunable front-end atively new research area is attracting considerable interest.
components, able to provide flexible frequency discrimina- The majority of lumped-element MEMS-tuned filter designs
tion. use fixed air-core inductors and achieve tuning via adjustable
MEMS capacitors and are generally limited to frequencies be-
3. FLEXIBLE FREQUENCY FRONT-END SUBSYSTEMS low 3 GHz. A group at Raytheon have developed a number
of MEMS-tuned lumped element filters using MEMS digi-
3.1. Tunable bandpass filters tal capacitor arrays to give continuous tuning in frequency
All practical transceivers employ bandpass filtering to pro- bands from 70 MHz to 2.8 GHz [14]. An octave tuning range
vide wideband suppression of unwanted interference and lumped element MEMS filter with concurrent bandwidth
distortion products. In an SDR, this element must be flexi- tuning from 7 to 42% is presented in [15]. This is impres-
ble. There are a number of filter tuning technologies appli- sive performance although the 4-bit MEMS capacitor arrays
cable for use in mobile telecommunication. Varactor-tuned lead to uneven coverage across the tuning band.
filters have been widely used due to their fast tuning speeds Numerous planar distributed designs have also been
and octave-frequency tuning range. However, their consider- published, for example, [16]. The majority of these designs
able loss and poor linearity have limited their use to IF sec- use some form of tunable capacitive loading to alter the res-
tions of the transceiver. The most promising emerging tech- onant frequency of transmission-line resonators. The tun-
nologies for RF are tunable ferroelectric films and RF micro- ing ranges tend to be more modest due to the reduction in
electromechanical systems (MEMS). Both technologies re- resonator Q as the capacitance is increased at lower tuning
sult in filters which have fast tuning speeds, are small, intro- frequencies. An alternative is to directly adjust the length of
duce little distortion, and consume minimal power. Ferro- the resonators. The use of MEMS switches to connect addi-
electric films such as BST consist of a thin film of ferroelectric tional lengths of transmission-line to a hairpin-line filter has
Hairpin resonator RF MEMS switch MM ls ls ls

10
Cc Cc Cc
Y0e , Y0o , v pe , v po
(a) 1 2 n
(a)
0
−20
Insertion loss (dB)
LDCL pairs
−40
Lumped
−60
capacitors
−80 (b)
0 1 2 3 4
Frequency (GHz)
Figure 5: (a) Schematic of a pair of lumped-distributed coupled
MEMS switches off lines (LDCLs). (b) Partially cropped image of 3-element filter with
MEMS switches on
four pairs of LDCLs.
(b)
Figure 4: (a) Filter layout and (b) measured performance of a 3.2. Tunable bandstop filters
MEMS-tuned coupled hair-pin filter.
Bandstop filters 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-
been proposed [17]. Measured results show a tuning range able to bandpass filters, particularly in the transmitter, due
from 2.05 GHz to 2.25 GHz with a constant percentage band- to their low passband insertion loss and high attenuation in
width of 4.5%. The filter layout and performance are shown the stopband. Bandstop filters may find application in the re-
in Figure 4. ceiver path where a high level of suppression is needed over a
A limitation of distributed filter designs is that it is dif- limited bandwidth. Their low passband insertion loss means
ficult to alter the interresonator coupling and, because this they can be employed without greatly affecting the receiver
coupling dictates the overall filter Q, this leads to difficulty sensitivity.
in bandwidth tuning. This limitation is overcome by us- The majority of tunable bandstop filters use some form
ing capacitively loaded dual-behaviour resonators which al- of variable capacitance to tune the electrical length in a ca-
low for independent control of the attenuation poles of the pacitively coupled shunt stub design [20]. With some mod-
filter [18]. The reported frequency tuning range is some- ification, these designs can yield tuning ranges of almost
what limited however, due to the frequency invariance of the an octave [21], although the use of lumped capacitive ele-
impedance inverters. ments produces an asymmetric stopband response and in-
A wide-range, multioctave centre-frequency tunable fil- troduces spurious parasitic behaviour at frequencies above
ter with concurrent bandwidth tuning capability has been the stopband. Discrete tuning of bandstop filters has re-
reported by the authors [19]. This design, suitable for cently been demonstrated using MEMS switches to alter the
use with MEMS bistable contact and capacitive switches, resonator properties also reaching almost an octave tuning
utilises switches and switched capacitors distributed at in- range [22]. The disadvantage of discrete tuning is that, to af-
tervals along pairs of coupled transmission lines (Figure 5) fect a high tuning resolution, large numbers of tuning ele-
to achieve both bandwidth and centre-frequency tuning. Be- ments are needed.
ing discrete in nature, there are a finite number of tuning Wide-range, continuous tuning can be attained by
points. However, the distributed topology means that the employing a composite tuning mechanism, consisting of
tuning range and resolution are limited only by the place- varactor-loading and discrete transmission-line length ad-
ment density and electrical size of the resonators. Illustrative justment using PIN diode switches [23] as shown in Figure 7.
performance is given in Figure 6. This composite approach not only extends the relative
0 Port 1 Port 2
50Ω
−10
λ/4 at 1.75 GHz
Insertion loss (dB)
−20
−30
−40 G1
. .
−50 . .
G2 . . G2
−60
−70 . .
.. ..
Return loss (dB)
90Ω
90Ω
−10
−20 G9
−30
200 300 400 500 700 1000 2000
Figure 7: Tunable notch filter schematic.
Frequency (MHz)
(a)
3.3. Narrowband tunable antennas
The performance of an SDR terminal hinges on its interface
0 with the radio channel: the antenna. Acceptable radiation ef-
−10 ficiency must be attained across the complete operational fre-
quency range. In addition, current market expectations place
−20
Insertion loss (dB)
constraints on the size of the SDR terminal. These require-

−30 ments, wide operational bandwidth and small size, are in-
−40 compatible. Reducing the size of an antenna results in either
its efficiency decreasing (which is unacceptable for a termi-
−50
nal antenna) or its bandwidth narrowing, for example, [24].
−60 Conversely, the (instantaneous) operational bandwidth of an
−70 antenna may be widened by increasing its size, for example,
300 350 400 450 500 550 600 [25].
Frequency (MHz) The current practice is to design a passive antenna with
a wide operational bandwidth, and then conform its shape
(b) to fit 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
Figure 6: Measured and modelled performance of LDCL filter at a important characteristic of an antenna: its ability to radi-
selection of tuning points. (a) Measured (solid) and modelled (bro- ate. When radiation patterns are considered, they are usu-
ken line) centre-frequency tuning. (b) Measured bandwidth tuning ally simple 2D cuts,5 for example, [26]; though most show
at a nominal centre frequency of 450 MHz. the trend of pattern degradation as a function of frequency
[27]. Comparison of channel capacity data based on anten-
nas whose only significant difference is polarisation and pat-
centre-frequency tuning range beyond that achievable using tern purity suggests that antennas with high polarisation and
solely reactive loading, but also permits the independent tun- pattern purities can yield higher channel capacities relative to
ing of the pass- and stopbands. Independent placement of those that have poor polarisations and varying patterns [28].
both the pass- and stopbands is advantageous in that it re- An alternative to a large passive wideband structure is a
laxes the passband constraints by reducing the range of fre- narrowband tunable antenna. Here the antenna can be made
quencies over which minimum loss must be maintained. The arbitrarily small provided its instantaneous input response
region of minimum loss can be tuned to the required fre- (bandwidth) covers at least one channel of the standard. By
quency and the loss of the filter at other frequencies becomes loading the antenna with a variable reactance, be that capac-
less important. Measured results of the filter, showing a two- itive (e.g., using a varactor diode) or inductive, it is possible
octave tuning range and variable, relative passband position,
are given in Figure 8. The use of active semiconductor com- 5 Although this may seem a succinct way of characterising an antenna’s
ponents leads to moderate linearity performance. This may radiation, it is insufficient as it only accounts for one or two planes about
be alleviated by employing MEMS counterparts throughout the structure; for an accurate characterisation, full 3D (co- and cross-polar)
the filter. patterns should be measured.
0 0
2
4
6
Insertion loss (dB)
−10 8
10
12
14
|S11 | / dB
16
−20 18
20
22
24
−30 26
0 0.5 1 1.5 2 2.5 28
30
Frequency (GHz) 32
6 pF, G9 0.6 pF, G4 34
36
1.6 pF, G7 0.5 pF, G3 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8
0.9 pF, G5
Frequency/GHz ×109
(a) 0v 6v
1v 7v
2v 8v
3v 9v
0 10 v
4v
5v 15 v
−5
Figure 9: Measured S11 response of the tunable narrowband an-

Insertion loss (dB)
−10
tenna with variation of varactor bias voltage.
−15
−20
has shown the efficiency of a tunable antenna to be com-
−25 parable to that of a passive wideband conformal antenna
[30], and the nonlinearities to be within thresholds of cur-
−30 rent standards [31]. It is envisaged that with the development
0 0.5 1 1.5 2 2.5 3 of MEMS technologies, tunable devices with lower loss will
Frequency (GHz) emerge.
2.7 pF, G1 Effectively, a tunable narrowband antenna is little more
1.2 pF, G3 than a tunable filter. Previous work has shown that the use
0.7 pF, G6 of such an antenna, whose instantaneous input bandwidth is
optimised for operation on a single channel, offers significant
(b) interference rejection relative to a passive wideband antenna
[32]. Given the congested nature of the spectrum this filter-
ing prior to the RF front-end is highly desirable.
Figure 8: Measured performance of a compositely tuned bandstop
filter showing (a) centre-frequency tuning and (b) decrease in filter
Q and the associated shift in passband frequency. (Approximate var-
3.4. Discussion
actor capacitance is measured in pF and G denotes ground position There remain significant challenges in the design and pro-
with reference to Figure 7.) duction of flexible components for frequency discrimina-
tion in SDR transceivers. New technologies capable of mul-
tioctave tuning ranges have been demonstrated and emerg-
to vary, in a controlled fashion, the resonant frequency of ing technologies promise lower loss and higher linearity per-
the antenna. Figure 9 shows the tuning range of an electri- formance. However, for mobile applications, where size is a
cally small antenna that is capacitively tuned using a varac- primary consideration, the fabrication of such devices must
tor diode. An example of the measured copolarisation pat- be addressed. Suitable filters and antennas must be mono-
tern for this antenna is shown in Figure 10. Because only lithic in order to simplify production and achieve appropri-
the resonant frequency of a single mode is varied, the radia- ate footprints while at the same time being environmentally
tion characteristics remain relatively constant over the tuning robust. For transmitter applications, power handling capa-
range, when compared with those of a passive wideband an- bilities must be addressed. High power RFMEMS-tuned fil-
tenna [29]. However, the use of a reactive element will intro- ters capable of 25 W have been investigated at VHF frequen-
duce losses (which are a function of bias voltage) and nonlin- cies [33], however, the tuning capabilities of these filters are
earities (due to the semiconductor junction). Prior research limited.
tuning space. The latter introduces the cost and complexity

of an additional receiver. The requirements of this receiver
are reduced significantly 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
filter.
4. SUMMARY
z
A wideband radio front end is attractive for SDR, however,
y the resulting compromise in performance suggests that a nar-
rowband tunable approach is worthy of consideration. The
x
2000 MHz essential advantage of narrowband systems is the frequency
discrimination they offer, thereby reducing the linearity and
Figure 10: Typical measured copolarisation pattern of the tunable dynamic range requirements of the transceiver. This paper
narrowband antenna. has surveyed recent techniques to extend the tuning range
and overall flexibility of three key elements of the narrow-
band radio front end; bandpass and bandstop filters and an-
From a system implementation perspective, the SDR air tennas.
interface requirements will play a key role in determining the Multioctave centre-frequency tuning ranges have been
optimum combination of flexible filtering subsystems. The demonstrated in bandpass and bandstop filters with concur-
majority of the frequency discrimination requirements for rent bandwidth tuning. However, miniaturisation and inte-
the transceiver can be determined from known signal infor- gration issues must be addressed to yield solutions suitable
mation. For example, the suppression of transmit noise in for wireless terminals. Tunable narrowband antennas have
the receive band can be determined from a knowledge of been shown not only to contribute helpful frequency dis-
the transmitter noise profile and the receiver cochannel in- crimination but also to display far superior performance for
terference rejection ratio for a given set of up- and down- a given size when compared to wide operational bandwidth
link channel parameters. For those requirements which deal designs.
with signal information that is not known beforehand (i.e., With a selection of tunable filtering components in the
receiver blocking), some relaxation in front-end filtering re- front end, the potential exists for intelligent real-time adap-
quirements may be gained by defining them on a statisti- tive control of the transceivers frequency discrimination pro-
cal basis (i.e., number of simultaneous interferers above a file. Analysis of radio spectrum measurements suggests that,
given power level which must be tolerated by the receiver). for the receiver, a high-level of discrimination is needed only
In this case, the wideband suppression level of the receiver at a few select frequencies at any given time. An adaptive fil-
may be able to be reduced by employing notch filters to tering profile may allow the specifications of the individual
deal with high-level interferers. Once the complete specifica- filtering components to be relaxed.
tions have been defined, the optimum combination of sub-
systems can be designed, based on their individual perfor- REFERENCES
mance.
[1] J. Mitola, “The software radio architecture,” IEEE Commun.
A method of intelligent control is a critical prerequisite
Mag., vol. 33, no. 5, pp. 26–38, 1995.
for the use of flexible subsystems in an SDR front end. In all [2] M. Cave, “Review of Radio Spectrum Management,” In-
cases, it is necessary to have knowledge of the tuning rela- dependent review prepared for the department of trade
tionship between the frequency response of the subsystems and industry and HM treasury, March 2002, available on
and the signals used to control them. However, the major www.ofcom.org.uk .
difficulty lies in determining the appropriate frequency pro- [3] R. H. Walden, “Performance trends for analog to digital con-
file for each subsystem. Within the wanted channel, centre verters,” IEEE Commun. Mag., vol. 37, no. 2, pp. 96–101, 1999.
[4] H. Tsurumi and Y. Suzuki, “Broadband RF stage architec-
frequency and bandwidth information can be used to tune ture for software-defined radio in handheld terminal appli-
bandpass filters and antennas for optimum passband loss and cations,” IEEE Commun. Mag., vol. 37, no. 2, pp. 90–95,
radiation efficiency. The appropriate receiver stopband pro- 1999.
files are not known a priori, and the most efficient profile [5] Z. D. Liu and P. S. Hall, “Dual-band antenna for hand
will assign bandstop filter suppression at the frequency of the held portable telephones,” Electronics Letters, vol. 32, no. 7,
most problematic interference. Two control approaches ex- pp. 609–610, 1996.
ist: blind adaptation, where the filters are tuned to minimise [6] P. A. Warr, M. A. Beach, and J. P. McGeehan, “Gain-element
transfer response control for octave-band feedforward ampli-
broadband detected power; and spectrum monitoring, where fiers,” Electronics Letters, vol. 37, no. 3, pp. 146–147, 2001.
a parallel receiver determines the offending signal charac- [7] T. Nesimoglu, M. A. Beach, P. A. Warr, and J. R. MacLeod,
teristics. The former suffers from search algorithm latency; “Linearised mixer using frequency retranslation,” Electronics
complex algorithms are needed due to local minima in the Letters, vol. 37, no. 25, pp. 1493–1494, 2001.
[8] J. R. MacLeod, M. A. Beach, P. A. Warr, and T. Nesimoglu, [27] K.-L. Wong and W.-H. Hsu, “Broadband triangular mi-
“Filter considerations in the design of a software defined ra- crostrip antenna with U-shaped slot,” Electronics Letters,
dio,” in Proc. IST Mobile Communications Summit (IST ’01), vol. 33, no. 25, pp. 2085–2087, 1997.
pp. 363–368, Barcelona, Spain, September 2001. [28] Radiocommunications Agency, “Ref: AY4476B—Antenna Ar-
[9] N. Pothecary, Feedforward Linear Power Amplifiers, Artech ray Technology and MIMO Systems,” March 2004, available
House, Norwood, Mass, USA, 1999, pp. 49–52. on http://www.ofcom.org.uk.
[10] G. T. Watkins, Flexible linearity profile feedforward low noise [29] P. R. Urwin-Wright, G. S. Hilton, I. J. Craddock, and P. N.
amplifiers for Software Defined Radio, Ph.D. thesis, University Fletcher, “On the pattern control of an annular slot operating
of Bristol, Bristol, UK, 2003. in its ‘DC’ mode,” in Proc. 3rd Management Meeting of Cost
[11] I. Vendik, O. Vendik, V. Pleskachev, and M. Nikol’ski, “Tun- 284, Budapest, Hungary, April 2003.
able microwave filters using ferroelectric materials,” IEEE [30] P. R. Urwin-Wright, G. S. Hilton, I. J. Craddock, and P. N.
Trans. Appl. Superconduct., vol. 13, no. 2, pp. 716–719, 2003. Fletcher, “A tuneable electrically-small antenna operating in
[12] B. Noren, “Thin film barium strontium titanate (BST) for the ‘DC’ mode,” in Proc. 5th European Personal Mobile Com-
a new class of tunable RF components,” Microwave Journal, municatons Conference, vol. 5, pp. 524–528, Glasgow, UK,
vol. 47, no. 5, pp. 210–220, 2004. April 2003.
[13] G. M. Rebeiz and J. B. Muldavin, “RF MEMS switches and [31] P. R. Urwin-Wright, G. S. Hilton, I. J. Craddock, and P. N.
switch circuits,” IEEE Microwave, vol. 2, no. 4, pp. 59–71, Fletcher, “A reconfigurable electrically-small antenna operat-
2001. ing in the ‘DC’ mode,” in Proc. 57th IEEE Semiannual Vehicu-
[14] J. Brank, J. Yao, M. Eberly, A. Malczewski, K. Varian, and C. lar Technology Conference (VTC ’03), vol. 2, pp. 857–861, Jeju,
Goldsmith, “RF MEMS-based tunable filters,” International Korea, April 2003.
Journal of RF and Microwave Computer-Aided Engineering, [32] P. R. Rogers, An electrically-small tuneable annular slot an-
vol. 11, no. 5, pp. 276–284, 2001. tenna for future mobile terminals, Ph.D. thesis, University of
[15] R. M. Young, J. D. Adam, C. R. Vale, et al., “Low-loss bandpass Bristol, Bristol, UK, 2004.
RF filter using MEMS capacitance switches to achieve a one- [33] C. A. Hall, R. C. Luetzelschwab, R. D. Streeter, and J. H. Van-
octave tuning range and independently variable bandwidth,” patten, “A 25 watt RF MEM-tuned VHF bandpass filter,” in
in Proc. IEEE MTT-S International Microwave Symposium Di- Proc. IEEE MTT-S International Microwave Symposium Digest,
gest, vol. 3, pp. 1781–1784, Philadelphia, Pa, USA, June 2003. vol. 1, pp. 503–506, Philadelphia, Pa, USA, June 2003.
[16] A. Abbaspour-Tamijani, L. Dussopt, and G. M. Rebeiz,
“Miniature and tunable filters using MEMS capacitors,” IEEE
Bruce E. Carey-Smith received a B.E. de-
Trans. Microwave Theory Tech., vol. 51, no. 7, pp. 1878–1885,
gree in electrical and electronic engineer-
2003.
ing from the University of Canterbury,
[17] J. R. MacLeod, T. Nesimoglu, M. A. Beach, and P. A. Warr,
“Miniature distributed filters for software re-configurable ra- Christchurch, NZ, in 1995, and is currently
dio applications,” in Proc. IST Mobile Wireless Telecommunica- pursuing postgraduate study at the Uni-
tions Summit, pp. 159–163, Thessaloniki, Greece, June 2002. versity of Bristol, Bristol, UK. From 1995
[18] E. Fourn, C. Quendo, E. Rius, et al., “Bandwidth and cen- to 2002, he was with Tait Electronics Ltd.,
tral frequency control on tunable bandpass filter by using Christchurch, NZ, involved in the design of
MEMS cantilevers,” in Proc. IEEE MTT-S International Mi- RF circuits and systems for mobile radio ap-
crowave Symposium Digest, vol. 1, pp. 523–526, Philadelphia, plications. Subsequently he joined the Uni-
Pa, USA, June 2003. versity of Bristol, Bristol, UK, as a Research Associate in the Centre
[19] B. Carey-Smith, P. A. Warr, M. A. Beach, and T. Nesimoglu, for Communications Research where his current research interests
“Wide tuning-range planar filters using lumped-distributed are in the areas of tunable microwave circuits and amplifier lineari-
coupled resonators,” IEEE Trans. Microwave Theory Tech., sation for software reconfigurable radio.
vol. 53, no. 2, pp. 777–785, 2005.
[20] I. C. Hunter and J. D. Rhodes, “Electronically tunable mi- Paul A. Warr received his Ph.D. degree in
crowave bandstop filters,” IEEE Trans. Microwave Theory 2001 from the University of Bristol, Bris-
Tech., vol. 82, no. 9, pp. 1354–1360, 1982. tol, UK, for his work on octave-band lin-
[21] S. Toyoda, “Quarter-wavelength coupled variable bandstop ear receiver amplifiers, his M.S. degree in
and bandpass filters using varactor modes,” IEEE Trans. Mi- communications systems and signal pro-
crowave Theory Tech., vol. 82, no. 9, pp. 1387–1389, 1982. cessing also from Bristol in 1996, and his
[22] G. Zheng and J. Papapolymerou, “Monolithic reconfigurable B.Eng. degree in electronics and commu-
bandstop filter using RF MEMS switches,” International Jour- nications from the University of Bath, UK,
nal of RF and Microwave Computer-Aided Engineering, vol. 14, in 1994. He is currently a Lecturer in ra-
no. 4, pp. 373–382, 2004.
dio frequency engineering at the University
[23] B. Carey-Smith and P. A. Warr, “Broadband configurable
of Bristol where his research covers the front-end aspects of soft-
bandstop filter with composite tuning mechanism,” Electron-
ics Letters, vol. 40, no. 25, pp. 1587–1589, 2004. ware (reconfigurable) radio and diversity-exploiting communica-
[24] S. D. Rogers and C. M. Butler, “Wide-band sleeve-cage tion systems; responsive linear amplifiers; flexible filters; and linear
and sleeve-helical antennas,” IEEE Trans. Antennas Propagat., frequency translation. Funding sources for this research have in-
vol. 50, no. 10, pp. 1409–1414, 2002. cluded UK DTI/EPSRC alongside CEC ACTS and IST programmes
[25] R. B. Waterhouse, “Broadband stacked shorted patch,” Elec- and industrial collaborators. Dr. Warr is a Member of the Execu-
tronics Letters, vol. 35, no. 2, pp. 98–100, 1999. tive Committee of the IEE Professional Network on Communica-
[26] T. Huynh and K.-F. Lee, “Single-layer single-patch wide- tion Networks & Services. Prior appointments have included the
band microstrip antenna,” Electronics Letters, vol. 31, no. 16, Marconi Company where he worked on secure, high-redundancy,
pp. 1310–1312, 1995. cross-platform communications.
Phill R. Rogers obtained the M.Eng. degree

(first 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 staff 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 definable 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.
Geoffrey 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. From 1984, to 1986, he worked
as a design engineer at GEC-Marconi, be-
fore commencing research in the area of
microwave antennas, first as a postgraduate
student and later as a member of research
staff at Bristol University. This work included design and analysis of
printed antenna elements and arrays, and also involved the devel-
opment of finite-difference time-domain (FDTD) models for these
structures. In 1993, he joined the academic staff 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.
c 2005 Y. S. Poberezhskiy and G. Y. Poberezhskiy
Flexible Analog Front Ends of Reconfigurable

Radios Based on Sampling and Reconstruction
with Internal Filtering
Yefim S. Poberezhskiy
Rockwell Scientific 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 filtering in analog front ends potentially provide the best performance of
software defined radios. However, conventional techniques used for these procedures limit reconfigurability and adaptivity of
the radios, complicate integrated circuit implementation, and preclude achieving potential performance. Novel sampling and
reconstruction techniques with internal filtering 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 filtering on the analog front end architectures and capabilities of software defined
radios is discussed.
Keywords and phrases: software defined radios, reconfigurable and adaptive transceivers, sampling, analog signal reconstruction,
antialiasing filtering, A/D.
1. INTRODUCTION It is shown in this paper that the changes in the AMP de-
sign have to be related first of all to the methods of sampling,
Next generation of software defined radios (SDRs) should
reconstruction, and antialiasing filtering. It is also shown
be reconfigurable to support future wireless systems operat-
that implementation of novel sampling and reconstruction
ing with different existing and evolving communication stan-
techniques with internal filtering [17, 18, 19, 20, 21, 22, 23]
dards while providing a wide variety of services over vari-
will make the AMPs of SDRs almost as flexible as their DPs
ous networks. These SDRs should also be extremely adap-
and significantly improve performance of SDRs. To this end,
tive to achieve high performance in dynamic communica-
conventional architectures of the radio AMPs are briefly ex-
tion environment and to accommodate varying user needs.
amined in Section 2. It is shown that the architectures that
Modern radios, virtually all of which are digital, do not
potentially can provide the best performance have the low-
meet these requirements. They contain large analog front
est flexibility and scale of integration. The main causes of
ends, that is, their analog and mixed-signal portions (AMPs)
the conventional architectures’ drawbacks are determined.
[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16]. The AMPs are
In Section 3, novel sampling and reconstruction techniques
much less flexible and have much lower scale of integration
with internal filtering are described. The sampling technique
than the radios’ digital portions (DPs). The AMPs are also
was obtained as a logical step in the development of inte-
sources of many types of interference and signal distortion. It
grating sample-and-hold amplifiers (SHAs) in [17, 18]. In
can be stated that reconfigurability, adaptivity, performance,
[19, 20], it was derived from the sampling theorem. The re-
and scale of integration of modern SDRs are limited by their
construction technique with internal filtering was derived
AMPs. Therefore, only radical changes in the design of the
from the sampling theorem in [21]. Initial analysis of both
AMPs allow development of really reconfigurable SDRs.
techniques was performed in [22, 23]. Section 3 contains ex-
amination of their features and capabilities, which is more
This is an open access article distributed under the Creative Commons detailed than that in [22, 23]. Challenges of these techniques’
Attribution License, which permits unrestricted use, distribution, and implementation and two methods of modification of sam-
reproduction in any medium, provided the original work is properly cited. pling circuits (SCs) with internal antialiasing filtering are
Flexible Analog Front Ends of Reconfigurable Radios 365
analyzed in Section 4. Since SCs and reconstruction circuits

(RCs) with internal filtering are inherently multichannel, RF
mitigation of the channel mismatch impact on the perfor- section
mance of the SDRs is discussed in Section 5. Architectures
of the AMPs modified to accommodate sampling and recon- LPF SHA A/D
struction with internal filtering are considered in Section 6. cos 2π fr t I channel
To DP
2. CONVENTIONAL ARCHITECTURES OF LPF SHA A/D

THE RADIO AMPS sin 2π fr t Q channel
2.1. AMPs of receivers
(a)
In digital receivers, the main purpose of AMPs is to create
conditions for signal digitization. Indeed, AMPs, regardless
IF strip
of their architectures, carry out the following main func-
tions: antialiasing filtering, amplification of received sig- RF LPF IF
section filter
nals to the level required for the analog-to-digital converter
(A/D), and conversion of the signals to the frequency most LO
convenient for sampling and quantization. Besides, they of-
LPF SHA A/D
ten provide band selection, image rejection, and some other
cos 2π f0 t I channel
types of frequency selection to lower requirements for the To DP
dynamic range of subsequent circuits. Most AMPs of mod- LPF SHA A/D
ern receivers belong to one of three basic architectures: direct
sin 2π f0 t Q channel
conversion architecture, superheterodyne architecture with
baseband sampling, and superheterodyne architecture with
(b)
bandpass sampling. The examples of these architectures are
shown in Figure 1.
In a direct conversion (homodyne) architecture (see 1st IF strip
Figure 1a), a radio frequency (RF) section performs prelimi- RF LPF1
1st IF
nary filtering and amplification of the sum of a desired signal, section filter
noise, and interference. Then, this sum is converted to the 1st LO
baseband, forming its in-phase (I) and quadrature (Q) com-
ponents. A local oscillator (LO), which generates sine and 2nd IF strip
cosine components at radio frequency fr , is tunable within LPF2 2nd IF
the receiver frequency range. Lowpass filters (LPFs) provide filter
antialiasing filtering of the I and Q components while SHAs 2nd LO
and A/Ds carry out their sampling and quantization. Chan- To DP
nel filtering is performed digitally in the receiver DP. For sim- SHA A/D
plicity, circuits providing frequency tuning, gain control, and
other auxiliary functions are not shown in Figure 1 and sub- (c)
sequent figures. Although integrated circuit (IC) implemen-
tation of this architecture encounters many difficulties, it is Figure 1: Receiver AMP architectures: (a) direct conversion archi-
simpler than that of the architectures shown in Figures 1b tecture, (b) superheterodyne architecture with baseband sampling,
and 1c. and (c) superheterodyne architecture with bandpass sampling.
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) f0 af- typically chosen to simplify antialiasing filtering and digitiza-
ter image rejection and preliminary amplification in the RF tion. Double frequency conversion also allows division of the
section. Antialiasing filtering is performed at a fixed IF. This AMP gain between the 1st and 2nd IF strips. This architec-
enables the use of bandpass filters with high selectivity, for ture performs real-valued bandpass sampling, representing
example, surface acoustic wave (SAW), crystal, mechanical, signals by the samples of their instantaneous values. In the
and ceramic. Then, the sum is converted to the baseband and DP, these samples are converted to the samples of I and Q
its I and Q components are formed. components (complex-valued representation), to make digi-
An example of a superheterodyne architecture with tal signal processing more efficient.
bandpass sampling is shown in Figure 1c. In most cases, The results of comparative analysis of the described ar-
such receivers have two frequency conversions. The 1st IF chitectures are reflected in Table 1. This analysis is not de-
is usually selected high enough to simplify image rejection tailed because each basic architecture has many modifica-
and reduce the number of spurious responses. The 2nd IF is tions. For example, superheterodyne architectures may have
Table 1: Comparison of various AMP architectures of modern receivers.
Architecture Advantages Drawbacks

Absence of spectral images Significant phase and amplitude
caused by frequency conversion imbalances between I and Q channels
High nonlinear distortions due to the
Better adaptivity compared to fall of substantial part of IMPs within
other modern architectures the signal spectrum
Better compatibility of AMP technology LO leakage that creates interference to
Direct conversion
with IC technology compared to other other receivers and contributes to the
receiver architecture
architectures DC offset
Relatively low requirements for Relatively low selectivity of
SHA and A/D antialiasing filtering
Direct current (DC) offset caused
Minimum cost, size, and weight
by many factors
Flicker noise
High nonlinear distortions due to the
Radical reduction of LO leakage due
fall of substantial part of IMPs within
to the offset frequency conversion
signal spectrum
High selectivity of antialiasing filtering Low adaptivity and reconfigurability of the
provided by SAW, crystal, mechanical, or receiver AMP due to the use of SAW,
ceramic IF filters crystal, mechanical, or ceramic IF filters
Superheterodyne receiver Slight reduction of phase and amplitude imbalances
Incompatibility of AMP technology with
architecture with between I and Q channels compared to the direct
IC technology due to the use of SAW,
baseband sampling conversion architecture (due to conversion from a
crystal, mechanical, or ceramic IF filters
constant IF to zero frequency)
Reduction of flicker noise due to Still significant phase and amplitude
lesser gain at zero frequency imbalances between I and Q channels
Relatively low requirements for Spurious responses due to
SHA and A/D frequency conversions
Still significant flicker noise
Low adaptivity and reconfigurability of
Radical reduction of LO leakage
the receiver AMP due to the use of SAW,
due to offset frequency conversion
crystal, mechanical, or ceramic IF filters
High selectivity of antialiasing filtering Incompatibility of AMP technology with
provided by SAW, crystal, mechanical, IC technology due to the use of SAW,
Superheterodyne receiver crystal, mechanical, or ceramic IF filters
or ceramic IF filters
architecture with
bandpass sampling Exclusion of phase and amplitude Still high nonlinear distortions due to
imbalances between I and Q channels large input current of SHA
Exclusion of DC offset and flicker Spurious responses due to

noise frequency conversions
Minimum IMPs falling within the Highest requirements for SHA and
signal spectrum A/D
different number of frequency conversions, and even the ar- number of intermodulation products (IMPs) falling within
chitectures with a single conversion have different properties the signal spectrum. It also excludes phase and amplitude
depending on the parameters of their IF strips. For instance, imbalances between I and Q channels, DC offset, and flicker
selection of a low IF in a single-conversion architecture en- noise. The drawbacks of this architecture have the following
ables replacement of high-selectivity off-chip IF filters with causes. Low adaptivity, reconfigurability, and scale of inte-
active filters. This increases flexibility and scale of integra- gration of the AMPs are caused by inflexibility of the best IF
tion of an AMP at the expense of more complicated image filters (e.g., SAW, crystal, mechanical, and ceramic) and in-
rejection. compatibility of their technology with IC technology. Inflex-
Despite the absence of some details, Table 1 conclusively ibility of these filters also does not allow avoiding spurious
shows that the superheterodyne architecture with bandpass responses. Two times higher sampling frequency required for
sampling has advantages that cannot be provided by other bandpass sampling raises requirements for SHA and A/D. At
architectures. Indeed, only bandpass sampling minimizes the present, track-and-hold amplifiers (THAs) are usually used
as SHAs for bandpass sampling. A THA does not suppress

out-of-band noise and IMPs of all the stages between the an- D/A LPF
tialiasing filter and the THA capacitor. As a result of sam- From I channel cos 2π fr t
pling, these noise and IMPs fall within the signal spectrum. DP BPF PA
The impact of this phenomenon is especially significant in D/A LPF
receivers with bandpass sampling. THAs need large input
current because they utilize only a small fraction of signal Q channel sin 2π fr t
energy for sampling. The large input current requires a sig-
nificant AMP gain. This makes sampling close to the antenna (a)
impossible. The large input current also increases nonlinear
distortions. Higher frequency of bandpass signals compared IF strip
to baseband ones further increases the required THA input IF BPF PA
current and, consequently, nonlinear distortions. THAs are filter
very susceptible to jitter. LO
It is important to add that conventional sampling pro-
cedures have no theoretical basis. In contrast, sampling with D/A LPF
internal antialiasing filtering has been derived from the sam- From I channel cos 2π f0 t
pling theorem. As shown in Section 3, it eliminates the draw- DP
backs of conventional sampling. D/A LPF
Q channel sin 2π f0 t
2.2. AMPs of transmitters
An AMP of a digital transmitter, regardless of its architecture, (b)
has to perform reconstruction filtering, conversion of recon-
structed signals to the RF, and their amplification. Similar 2nd IF strip
to the receivers, modern transmitters have three basic AMP
architectures: direct up-conversion architecture, offset up- 2nd IF BPF PA
filter
conversion architecture with baseband reconstruction, and
offset up-conversion architecture with bandpass reconstruc- 2nd LO
tion. Simplified block diagrams of these architectures are 1st IF strip
shown in Figure 2.
In a direct up-conversion architecture (see Figure 2a), 1st IF
filter
modulation and channel filtering are carried out in the trans-
1st LO
mitter DP using complex-valued representation. The I and
From DP
Q outputs of the DP are converted to analog samples by D/A SHA
D/As. After baseband filtering and amplification of I and
Q components, an analog bandpass signal is formed di- (c)
rectly at the transmitter RF. An LO, which generates cos 2π fr t
and sin 2π fr t, is tunable within the transmitter frequency Figure 2: Transmitter AMP architectures: (a) direct up-conversion
range. The formed RF signal passes through a bandpass filter architecture, (b) offset up-conversion architecture with baseband
(BPF) that filters out unwanted products of frequency up- reconstruction, (c) offset up-conversion architecture with bandpass
conversion, and enters a power amplifier (PA). This archi- reconstruction.
tecture is the most flexible and suitable for IC implementa-
tion among modern architectures. However, it cannot pro-
vide high performance. The baseband reconstruction causes of the transmitter RF band, the BPF in the AMP can be re-
significant amplitude and phase imbalances between the I placed by an LPF. This architecture still has all the drawbacks
and Q channels, DC offset, and nonlinear distortions that re- related to baseband reconstruction.
duce the accuracy of modulation. The DC offset also causes These drawbacks are eliminated in an offset up-
the LO leakage through the antenna. Additional problem of conversion architecture with bandpass reconstruction shown
this architecture is that a voltage-controlled oscillator (VCO), in Figure 2c. Here, a bandpass IF signal is formed digitally
used as an LO, is sensitive to pulling from the PA output. in the DP. This reduces nonlinear distortions and excludes
An AMP architecture with offset up-conversion and DC offset and amplitude and phase imbalances between I
baseband reconstruction (see Figure 2b) is not susceptible to and Q channels. As a result, modulation becomes more ac-
VCO pulling. It provides better reconstruction filtering than curate, and a spurious carrier is not present. However, like
the previous architecture due to the use of SAW, crystal, me- in the case of receivers, these advantages are achieved at the
chanical, or ceramic IF filters and allows slightly more accu- expense of reduced adaptivity of the AMP and incompatibil-
rate formation of bandpass signals since it is performed at a ity of its technology with IC technology caused by the most
constant IF. If the IF is selected higher than the upper bound effective IF reconstruction filters. Besides, the sample mode
Table 2: Comparison of various AMP architectures of modern transmitters.
Architecture Advantages Drawbacks
Low accuracy of modulation due to

Better compatibility of AMP technology
significant phase and amplitude imbalances
Direct with IC technology compared to other
between I and Q channels, DC offset, and
up-conversion modern architectures
nonlinear distortions
transmitter
architecture Better adaptivity compared to LO leakage through the antenna caused
other modern architectures by DC offset and other factors
Pulling voltage-controlled LO from PA output
Low accuracy of modulation due to
Insusceptibility to pulling the
significant phase and amplitude
voltage-controlled LO from the PA
imbalances between I and Q channels, DC
output
Offset offset, and nonlinear distortion
up-conversion High selectivity of reconstruction Low adaptivity and reconfigurability of AMP
transmitter filtering due to the use of SAW, crystal, due to the use of SAW, crystal, mechanical, or
architecture with mechanical, or ceramic IF filters ceramic IF filters
baseband
Slightly better accuracy of modulation Incompatibility of AMP technology with
reconstruction
due to forming a bandpass signal at a IC technology due to the use of SAW,
constant IF crystal, mechanical, or ceramic IF filters
Reduction of LO leakage
The highest accuracy of modulation due to
Low adaptivity and reconfigurability
radical reduction of phase and amplitude
of AMP due to the use of SAW,
imbalances between I and Q channels, DC
crystal, mechanical, or ceramic filters
Offset offset, and nonlinear distortion
up-conversion Insusceptibility to pulling Incompatibility of AMP technology with
transmitter voltage-controlled LO from PA IC technology due to the use of SAW,
architecture output crystal, mechanical, or ceramic filters
with bandpass High selectivity of reconstruction
reconstruction Incomplete utilization of D/A output
filtering due to the use of SAW, crystal,
power
mechanical, or ceramic filters
Radical reduction of LO leakage High requirements for D/A
length ∆ts in a conventional SHA at the D/A output should 3. SAMPLING AND RECONSTRUCTION WITH
meet the condition INTERNAL FILTERING
1 3.1. General
∆ts ≤ , (1)
2 f0
As shown in Section 2, AMPs with bandpass sampling, re-
where f0 is a center frequency of the reconstructed signal, construction, and filtering provide the best performance of
which coincides with the 1st IF. Condition (1) can be by- both receivers and transmitters (see Figures 1c and 2c). At
passed by using SHA with weighted integration. However, the same time, conventional methods of sampling, recon-
they are not used. Condition (1) does not allow efficient uti- struction, and filtering limit flexibility of the AMPs, compli-
lization of the D/A output current and, consequently, signal cate their IC implementation, and prevent achieving poten-
reconstruction close to the antenna. tial performance. Novel sampling and reconstruction tech-
The results of the comparative analysis of the described niques with internal filtering [17, 18, 19, 20, 21, 22, 23] al-
transmitter AMP architectures are reflected in Table 2. Since low elimination of these drawbacks and provide additional
each basic architecture has many modifications, this anal- benefits. These techniques have a strong theoretical founda-
ysis is not detailed. However, it shows that the offset up- tion because they are derived from the sampling theorem.
conversion architecture with bandpass reconstruction pro- They can be used for both bandpass and baseband sam-
vides the highest accuracy of modulation. As to the draw- pling and reconstruction. However, this paper is mainly fo-
backs of this architecture, they can be eliminated by imple- cused on bandpass applications of the proposed techniques
mentation of the proposed reconstruction technique with in- since the techniques are more beneficial for these applica-
ternal filtering (see Section 3). tions.

Su ( f ) Su ( f )
B B
−2 fs − f0 − fs 0 fs f0 2 fs f
−2 fs − fs 0 fs 2 fs f
(a)
(a)

Ga ( f ) Su1 ( f )

Ga ( f ) Su1 ( f )
−2 fs − f0 − fs 0 fs f0 2 fs f
−2 fs − fs 0 fs 2 fs f
(b)
(b)
Figure 3: Amplitude spectra and the desired AFR: (a) |Su ( f )|, Figure 4: Amplitude spectra and the desired AFR: (a) |Su ( f )|,
(b) |Su1 ( f )| and |Ga ( f )| (dashed line). (b) |Su1 ( f )| and |Ga ( f )| (dashed line).
3.2. Antialiasing and reconstruction filtering where f0 is a center frequency of Su ( f ). A plot of Su ( f ) is

To better describe operation of sampling and reconstruction shown in Figure 4a. For bandpass sampling and reconstruc-
circuits (SCs and RCs) with internal filtering, we first spec- tion, fs usually meets the condition
ify requirements for antialiasing and reconstruction filter-
f
ing. An antialiasing filter should minimally distort analog fs = 0 . (5)
signal u(t) intended for sampling and maximally suppress floor f0 / fs + 0.5 ± 0.25
noise and interference that can fall within the signal spec-
trum Su ( f ) as a result of sampling. Selection of fs according to (5) minimizes aliasing and sim-
When baseband sampling takes place, spectrum Su ( f ) of plifies both digital forming of I and Q components at the
a complex-valued u(t), represented by its Iand Q compo- output of the receiver A/D and digital forming of a band-
nents, occupies the interval (see Figure 3a) pass signal at the input of the transmitter D/A. Therefore, fs
that meets (5) is considered optimal. When fs is optimal, an
[−0.5B, +0.5B], (2) antialiasing filter should cause minimum distortion within
intervals (4) and suppress noise and interference within the
where B is a bandwidth of u(t). Sampling with frequency intervals
fs causes replication of Su ( f ) with period fs (see Figure 3b)
and mapping the whole f -axis for u(t) into the region − f0 + 0.5B + 0.5r fs , − f0 − 0.5B + 0.5r fs
[−0.5 fs , 0.5 fs [ for the sampled signal u(nT s ), where Ts = 1/ fs (6)
∪ f0 − 0.5B + 0.5r f s , f0 + 0.5B + 0.5r f s ,
is a sampling period. Thus, antialiasing filter should cause
minimum distortion within interval (2) and suppress noise
where r is an integer, r ∈ [(0.5 − 2 f0 / fs ), ∞[, r
= 0. Figure 4b
and interference within the intervals
shows amplitude spectrum |Su1 ( f )| of u(nT s ), and the de-
sired AFR |Ga ( f )| of an antialiasing filter for bandpass sam-
k f s − 0.5B, k f s + 0.5B , (3)
pling. Thus, a bandpass antialiasing filter has to suppress
where replicas of Su ( f ) are located in the spectrum Su1 ( f ) noise and interference within intervals (6) and minimally
of u(nT s ). In (3), k is any nonzero integer. In principle, noise distort u(t) within intervals (4). Suppression of noise and in-
and interference within the gaps between all intervals (3) and terference within the gaps between intervals (4) and (6) is not
(2) can be suppressed in the DP. However, if these noise and mandatory, but it can be used to lower requirements for the
interference are comparable with or greater than u(t), weak- resolution of an A/D and DP.
ening them by an SC lowers requirements for the resolution Bandpass reconstruction requires only suppression of
of an A/D and DP. A desired amplitude-frequency response u(nT s ) images within intervals (6) and minimum distortion
(AFR) |Ga ( f )| of an antialiasing filter is shown in Figure 3b within intervals (4).
by the dashed line.
In the case of reconstruction, it is necessary to suppress 3.3. Canonical sampling circuits
all the images of u(nT s ) within intervals (3) and minimally The block diagrams of two canonical SCs with internal an-
distort the image within interval (2). No suppression within tialiasing filtering are shown in Figure 5. In Figure 5a, an in-
the gaps between intervals (3) and (2) is usually required. put signal ui (t) is fed into L parallel channels, whose out-
When bandpass sampling takes place, spectrum Su ( f ) of puts are in turn connected to an A/D by a multiplexer (Mx).
real-valued bandpass u(t) occupies the intervals The spectrum of ui (t) is not limited by an antialiasing filter
and includes the spectrum of the signal u(t) that should be
− f0 − 0.5B, − f0 + 0.5B ∪ f0 − 0.5B, f0 + 0.5B , (4) sampled. The lth channel (l ∈ [1, L]) forms all samples with

5
ui (t)
0
1 u(nTs )
ui (t) . −5
. . 0 1 2 3 4 5 6 7 8 9 10
. . Mx A/D
.
t/Ts
(a)
L
1
wn (t)
...
... 0
Control −1
WFG unit 0 1 2 3 4 5 6 7 8 9 10
t/Ts
(a) (b)
1
wn (t)
A/D
0
1
ui (t) −1
.. .. 0 1 2 3 4 5 6 7 8 9 10
u(nTs ) . Mx
. t/Ts
(c)
A/D
L
1
wn (t) 0
...
...
−1
Control 0 1 2 3 4 5 6 7 8 9 10
WFG unit t/Ts
(d)
(b)
1
wn (t)
Figure 5: Canonical SCs with internal antialiasing filtering: (a) 0

single-A/D version and (b) multiple-A/D version.
−1
0 1 2 3 4 5 6 7 8 9 10
t/Ts
numbers l + kL, where k is any integer. The operational cy-
(e)
cle of each channel is equal to LTs , consists of three modes
(sample, hold, and clear), and is shifted by Ts relative to the 1
operational cycle of the previous channel. The length of the
wn (t)
0
sample mode is equal to Tw , where Tw is the length of weight
function w0 (t). −1
0 1 2 3 4 5 6 7 8 9 10
During the sample mode, ui (t) is multiplied by wn (t) = t/Ts
w0 (t − nT s ), and the product is integrated. As a result,
(f)
nT s +0.5Tw

u nT s = ui (τ) · wn (τ) · dτ. (7)
nT s −0.5Tw Figure 6: Positions of samples and corresponding wn (t).
Equation (7) reflects sampling, accumulation of the signal

energy with weight w0 (t), and antialiasing filtering with im- by Ts . Positions of samples and corresponding wn (t) are illus-
pulse response h(t) = w0 (nT s + 0.5Tw − t). Throughout the trated by Figure 6 for L = 5. As follows from (7), w0 (t) deter-
hold mode with length Th , a channel is connected to the A/D mines filtering properties of SCs. Examples of baseband and
that quantizes the channel output. In the clear mode with bandpass weight functions w0 (t) and the AFRs |Ga ( f )| of the
length Tc , the channel is disconnected from the A/D, and the SCs with these w0 (t) are shown in Figures 7 and 8, respec-
capacitor of its integrator is discharged. It is reasonable to tively. Since an SC performs finite impulse response (FIR)
allocate Ts for both hold and clear modes: Th + Tc = Ts . A filtering with AFR |Ga ( f )|, which is the amplitude spectrum
weight function generator (WFG) simultaneously generates of w0 (t), it suppresses interference using zeros of its AFR.
L − 1 copies wn (t) of w0 (t) because, at any instant, L − 1 chan- When baseband sampling takes place, the distances between
nels are in the sample mode, and one channel is in the hold or the centers of adjacent intervals (2) and (3) are equal to fs
clear mode. Each wn (t) is shifted relative to the previous one (see Figure 3). To suppress all intervals (3), |Ga ( f )| should
1 1
0.8
0.5
0.6
w0 (t)
w0 (t)
0.4 0
0.2 −0.5
0
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2 −1
−4 −3 −2 −1 0 1 2 3 4
t/Ts
t/Ts
(a)
(a)
0
0
AFR |Ga ( f )|
−20
−20
AFR |Ga ( f )|
−40
−40
−60
−60
−80
0 0.5 1 1.5 2 2.5 3 3.5 4
−80
f / fs 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
f / fs
(b)
(b)
Figure 7: Baseband SC (a) w0 (t) and (b) AFR |Ga ( f )|, in dB. Figure 8: Bandpass SC (a) w0 (t) and (b) AFR |Ga ( f )|, in dB.
have at least one zero within each of them. To achieve this, length of the sample and clear modes is equal to Ts . In the
condition Tw ≥ 1/ fs = Ts is necessary. For bandpass sam- subsequent multiply mode with duration Tw , the signal from
pling, the distances between the centers of adjacent intervals the SHA is multiplied by the appropriate copy of w0 (t) gener-
(4) and (6) are equal to 0.5 fs (see Figure 4). Consequently, ated by the WFG, and the product enters an adder that sums
Tw ≥ 1/(0.5 fs ) = 2Ts is required. When Th + Tc = Ts , the the output signals of all the channels. Since at any instant,
length of the channel operational cycle is L − 1 channels are in the multiply mode and one channel is
in the sample or clear mode, the WFG simultaneously gen-
LTs = Tw + Th + Tc ≥ 3Ts for bandpass u(t), erates L − 1 copies of w0 (t), each delayed by Ts relative to
(8) the previous one. The RC reconstructs an analog signal u(t)
LTs = Tw + Th + Tc ≥ 2Ts for baseband u(t).
according to the equation
It follows from (8) that L ≥ 3 is required for bandpass sam- ∞ ∞

pling and L ≥ 2 is necessary for baseband sampling. Only u(t) ≈ u nT s · wn (t) = u nT s · w0 t − nT s .
bandpass sampling is considered in the rest of the paper. n=−∞ n=−∞
In the SC shown in Figure 5b, the required speed of A/Ds (9)
is lower and an analog Mx is replaced with a digital one. This
version is preferable when the maximum speed of the A/Ds It follows from (9) that the RC performs reconstruction fil-
is lower than fs , or when L slower A/Ds cost less and/or con- tering with transfer function determined by w0 (t).
sume less power than faster one. In the version of a canonical RC shown in Figure 9b, digi-
tal words are distributed by a digital DMx among L channels.
3.4. Canonical reconstruction circuits Presence of a D/A in each channel allows removal of SHAs.
The block diagrams of canonical RCs with internal filtering Here, the channel operational cycle consists of two modes:
are shown in Figure 9. In Figure 9a, a demultiplexer (DMx) convert and multiply. In the first mode, the D/A converts
periodically (with period LTs ) connects the output of a D/A digital words into samples u(nT s ), which are multiplied by
to each of L channels. The lth channel (l ∈ [1, L]) processes wn (t) during the multiply mode. This version has the follow-
samples with numbers l + kL, where k is any integer. Opera- ing advantages: lower requirements for the speed of D/As,
tional cycle duration of each channel is equal to LTs and de- replacement of an analog DMx by a digital one, and removal
layed by Ts relative to that of the previous channel. The cycle of SHAs.
consists of three modes: clear, sample, and multiply. In the
clear mode, the SHA capacitor is discharged. Then, during 3.5. Advantages of the SCs and RCs and challenges
the sample mode, this capacitor is connected to the D/A by of their realization
the DMx and charged. Throughout these modes, there is no Both SCs and RCs with internal filtering make AMPs highly
signal at the channel output because zero level enters the sec- adaptive and easily reconfigurable because w0 (t), which
ond input of a multiplier from a WFG. The reasonable total determines their filtering properties, can be dynamically
At the same time, practical realization of the SCs and

Digital SHA
1 RCs with internal filtering and their implementation in SDRs
words u(nTs )
. u(t) present many technical challenges. Canonical structures of
D/A DMx . .
.. . the SCs and RCs (see Figures 5 and 9) are rather complex.
Therefore, their simplification is highly desirable. This sim-
SHA plification is intended, first of all, to reduce complexity and
L number of multiplications.
... ...
Control
WFG
unit 4. SIMPLIFICATION OF THE SCs AND RCs
(a) 4.1. Approaches to the problem
Approaches to simplification of the SCs and RCs depend
D/A on the ways of wn (t) generation and multiplications. Ana-
Digital 1 log generation of wn (t) implies that multiplications of ui (t)
words
. u(t) in the SCs and u(nT s ) in the RCs by wn (t) are performed
DMx . by analog multipliers. Since only simple wn (t) can be gen-
. u(nTs ) .
.
. erated by analog circuits, and this generation is not flexible
D/A
enough, digital generation is preferable. When wn (t) are gen-
erated digitally, they can be converted to the analog domain
L in the WFG (see Figures 5 and 9) or sent to the multipliers in
... ...
digital form. In the first case, multiplications in the SCs and
Control
unit
WFG RCs are analog. In the second case, these multiplications can
be carried out by multiplying D/As.
(b) Since digital wn (t) have unwanted spectral images, spec-
tral components of an input signal ui (t) in the SCs and a re-
Figure 9: Canonical RCs with internal reconstruction filtering: constructed signal u(t) in the RCs corresponding to the un-
(a) single-D/A version and (b) multiple-D/A version. wanted images should be suppressed. The suppression can
be performed by a wideband filter with fairly low selectiv-
ity that allows IC implementation. Such a filter is sufficient
changed. Internal filtering performed by these structures al- because a required sampling rate of w0 (t) representation is
lows removal of conventional antialiasing and reconstruc- much higher than that of the A/D used in a receiver and the
tion filters or their replacement by wideband low-selectivity D/A used in a transmitter when bandpass sampling and re-
filters realizable on a chip. This makes the AMP technol- construction take place. In practice, some kind of prefilter-
ogy uniform and compatible with the IC technology. The ing is performed in all types of receivers, and some kind of
RCs with internal filtering utilize the D/A output current postfiltering is performed in transmitters. Usually, these pre-
more efficiently than conventional devices, then bandpass filtering and postfiltering can provide the required suppres-
reconstruction takes place. The SCs with internal antialiasing sion. Since prefiltering and postfiltering automatically sup-
filtering accumulate signal energy in their storage capacitors press stopbands (6) remote from passband (4), internal fil-
during the sample mode. This accumulation filters out jitter tering performed by SCs and RCs should first of all sup-
and reduces the charging current of the storage capacitors press stopbands (6) closest to the passband. Complexity of
by 20–40 dB in most cases. Reduced jitter enables the devel- the SCs and RCs, caused by high sampling rate of w0 (t) rep-
opment of faster A/Ds. The decrease in the charging current resentation, can be compensated by its low resolution. The
lowers both the required gain of an AMP and its nonlinear goal is to lower the required resolution of w0 (t) represen-
distortions. The reduced AMP gain allows sampling close to tation or to find other means that can reduce multiplying
the antenna. Smaller charging current also lowers input volt- D/As (or analog multipliers) to a relatively small number of
age of the SCs. Indeed, although the same output voltage has switches.
to be provided by an SC with internal antialiasing filtering Simplification of the SCs and RCs can be achieved by
and a conventional SHA, the SC input voltage can be signifi- proper selection of w0 (t) and optimization of their architec-
cantly lower when the integrator operational amplifier has an tures for a given w0 (t). Below, attention is mostly focused on
adequate gain. As mentioned in Section 2.1, a conventional the SCs because their practical realization is more difficult
SHA does not suppress out-of-band noise and IMPs of all than that of RCs due to higher requirements for their dy-
the stages between the antialiasing filter and its capacitor. As namic range. Achieving a high dynamic range of multiplica-
a result of sampling, these noise and IMPs fall within the sig- tions in the SCs is still a challenging task, although low input
nal spectrum. The SCs with internal antialiasing filtering op- current (compared to conventional SHAs) makes it easier.
erate directly at the A/D input and reject out-of-band noise Brief information on w0 (t) selection is provided in
and IMPs of all preceding stages. Thus, they perform more Section 4.2, and two examples of the SC simplification are
effective antialiasing filtering than conventional structures. described and analyzed in Sections 4.3, 4.4, and 4.5. It is
important to emphasize that possible simplifications of the where wm0 (t) are partial weight functions, whose envelopes
SCs are not limited to these examples. are equal to cm W0 (t) and whose carriers are harmonics of f0 .
The spectra of wm0 (t) are partial transfer functions Gm ( f ). It
4.2. Selection of weight functions follows from (12) that when f0 / fs is high enough ( f0 / fs > 3
is usually sufficient), the distances between adjacent harmon-
Selection of w0 (t) is application specific and requires multi- ics of f0 are relatively large, and overlapping of Gm ( f ) does
ple tradeoffs. For example, w0 (t) that maximizes the dynamic not notably affect the suppression within those stopbands (6)
range of an AMP and w0 (t) that provides the best internal fil- that are close to the passband. Since remote stopbands (6)
tering are different. Indeed, w0 (t) with rectangular envelope are suppressed by prefiltering or postfiltering, the simplest
maximizes the dynamic range due to its minimum peak fac- c(t), which is a squarewave, can be used when f0 / fs is suf-
tor and the most efficient accumulation of the signal energy, ficient. Combining a squarewave c(t) with an appropriately
but it provides relatively poor internal filtering. At the same selected K-level W0 (t) allows reducing the multiplying D/As
time, w0 (t) that provides the best internal filtering for given L to a small number of switches. Note that, besides w0 (t) with
and fs /B has high peak factor and relatively poor accumula- K-level W0 (t), there are other classes of w0 (t) that allow us
tion of signal energy. When both features are desirable, w0 (t) to do this. If discontinuities in W0 (t) and c(t) are properly
has to be selected as a result of a certain tradeoff, and this re- aligned and f0 / fs > 3, overlapping of Gm ( f ) can be insignifi-
sult can be different depending on specific requirements for cant even if condition Tw = kTs is not met.
the radio. To provide the best antialiasing filtering for given The lower f0 / fs is, the more significantly Gm ( f ) are over-
L and fs /B, w0 (t) should be optimized using the least mean lapped. As a result, both W0 (t) and c(t) influence the filtering
square (LMS) or Chebyshev criterion [23]. Any w0 (t), op- properties of the SCs and RCs. When f0 / fs = 0.25, c(t) has
timal according to one of these criteria, requires high accu- the greatest impact on their transfer functions. To reduce the
racy of its representation and multiplications. This compli- multiplying D/As to a small number of switches in this case,
cates realization of the SCs. Suboptimal w0 (t) that provide c(t) should also be a several-level function.
effective antialiasing filtering with low accuracy of represen-
tation and multiplications are longer than optimal w0 (t) and,
4.3. Separate multiplying by W0 (t) and c(t)
consequently, require larger L. An increase of fs /B simplifies
antialiasing filtering and allows reduction of L or accuracy The following method of the SC realization can be obtained
of multiplications for a given quality of filtering [20]. Tech- using separate multiplication of ui (t) by the envelope W0 (t)
nology of the SCs and technical decisions regarding these and carrier c(t) of w0 (t). The nth sample at the output of the
and other units of the SDRs also influence selection of w0 (t). SC is as follows:
Due to the complexity of these multiple tradeoffs, there is
0.5Tw +nTs
no mathematical algorithm for w0 (t) selection, and heuristic
procedures combined with analysis and simulation are used u nT s = ui (t)w0 t − nTs dt. (13)
−0.5Tw +nTs
for this purpose.
In general, a bandpass w0 (t) can be represented as Taking into account (10), we can write

w0 (t) = W0 (t)c(t) for t ∈ − 0.5Tw , 0.5Tw ,
(10) w0 t − nTs = W0 t − nTs · c t − nTs . (14)
w0 (t) = 0 for t ∈
/ − 0.5Tw , 0.5Tw ,
When condition (5) is met, (14) can be rewritten as
where W0 (t) is a baseband envelope, and c(t) is a periodic
carrier (with period T0 = 1/ f0 ) that can be sinusoidal or
T0
nonsinusoidal. To provide linear phase-frequency response, w0 t − nTs = W0 t − nTs · c t − (n mod 4) . (15)
4
W0 (t) should be an even function, and c(t) should be an even
or odd function. Assuming that Tw = kTs where k is a nat- Since c(t ± T0 /2) = −c(t),
ural number, we can expand c(t) into Fourier series over the
time interval [−0.5Tw , 0.5Tw ]: 

c(t)(−1)n/2 if n is even,

c t − nTs = T
(16)
∞


c t − 0 (−1)(n±1)/2 if n is odd.
c(t) = cm e jm2π f0 t , (11) 4
m=−∞
Substituting (16) into (14), and (14) into (13), we obtain
where m is any integer and cm are coefficients of the Fourier 0.5Tw +nTs
series. Taking into account (10) and (11), we can write that
u(nT s ) = ui (t)W0 t − nTs
within the interval [−0.5Tw , 0.5Tw ], 
−0.5Tw +nTs


 if n is even
 (17)
∞ ∞
c(t)(−1)n/2 

× dt.
w0 (t) = W0 (t) cm e jm2π f0 t =  T0
wm0 (t), (12) 
c t − (−1)(n±1)/2 if n is odd 


m=−∞ m=−∞ 4
C Channel 1 Channel 1

+ C
+
− From From
. From WFEG CU .. −
From
ui (t) . . From CU .
cos 2π f0 t . CU . Mx A/D ui (t) .
. CU .
cos 2π f0 t . Mx A/D
+
+
− C Channel L
− Channel L
sin 2π f0 t
... sin 2π f0 t
... C
WFEG CU ...
... CU
...
Figure 10: Modified SC.
Figure 11: Modified SC with rectangular W0 (t).

In (16) and (17), “±” corresponds to “±” in (5). In particular,
when c(t) = cos 2π f0 t, (17) can be rewritten as follows:
be lowered compared to the canonical structures when some
0.5Tw +nTs other W0 (t) are used. Note that single-ended circuits are used

u nT s = ui (t)W0 t − nTs in Figures 10 and 11 only for simplicity of illustration. In
−0.5Tw +nTs
  practical applications, differential circuits are preferable.
cos 2π f t (−1)n/2 if n is even (18)
0
×
sin 2π f t (−1)(n±1)/2 if n is odd 
dt. 4.4. Analysis of the modified SC
0
Many features of the canonical and modified SCs are the
same. Indeed, when the same w0 (t) are used, their filtering
The algorithm described by (18) can be carried out by the
properties are identical and they accumulate equal amounts
modified SC shown in Figure 10. Here, ui (t) enters two mul-
of signal energy. Consequently, they provide the same reduc-
tipliers where it is multiplied by cos 2π f0 t and sin 2π f0 t.
tion of the input current compared to conventional sampling
These multiplications are similar to the beginning of the
structures. They are equally adaptive and equally suitable for
procedure used for baseband sampling of bandpass signals
IC implementation. However, there is still substantial differ-
(see Figures 1a and 1b). However, further processing is dif-
ence between them. A canonical SC is not sensitive to DC off-
ferent. Instead of baseband filtering of the lowest spectral
set, while the outputs of the modified SCs are influenced by
image after each multiplier, the differential outputs of both
DC offsets in the first two multipliers. Besides, the number
multipliers enter L channels through 4-contact switches. The
and values of IMPs that fall within the signal spectrum are
switches are necessary because each sample in any channel is
higher in the modified SCs than in the canonical ones. In-
shifted by ±πL/2 relative to the previous one in this channel
deed, multiplication of ui (t) by wn (t) in each channel of the
when (5) is true. A control unit (CU) provides proper opera-
canonical SC creates a spectral image at the frequency fs /4
tion of the switches. This switching shifts the multiplier out-
because wn (t) are centered around corresponding sampling
put spectral image from zero frequency to fs /4. After passing
instants and fs meets (5), whereas the first two multiplica-
decoupling capacitor C, it is processed in the channel. Sim-
tions in the modified SCs create baseband spectral images.
ilar to the canonical structure in Figure 5a, the operational
Below, this is proven analytically.
cycle of each channel is equal to LTs , consists of three modes
Assuming that the DC offset in the multiplier of the lth
(sample, hold, and clear), and is shifted by Ts relative to the
channel in a canonical SC is Ul , where l = [(n − 1) mod L]+1,
operational cycle of the previous channel. The difference is
we can rewrite (13) as
that the channel input signal is multiplied by the appropri-
ate copy Wn (t) of W0 (t) instead of wn (t) during the sample 0.5Tw +nTs

mode. A weight function envelope generator (WFEG) forms u nT s = ui (t)w0 t − nTs + Ul dt. (19)
−0.5Tw +nTs
Wn (t). Each Wn (t) is shifted relative to the previous one by Ts
to be in phase with the operational cycle of the correspond-
It follows from (16) and (19) that
ing channel.
At first glance, the structure in Figure 10 looks even more 0.5Tw +nTs

complex than the canonical one shown in Figure 5a. How- u nT s = ui (t)W0 t − nTs
−0.5Tw +nTs
ever, appropriate selection of W0 (t) can significantly simplify  
it. For example, when W0 (t) is a rectangular function, the 
c(t)(−1)n/2 if n is even


WFEG and the multipliers in the channels are unnecessary. × T dt + Ul Tw .

c t −
0
(−1) (n±1)/2 if n is odd 

As shown in Figure 11, the modified SC contains only 2 mul- 4
tipliers for any L in this case. Complexity of the SC can also (20)
Equation (20) can be rewritten as 80

0.5Tw +nTs

u nT s = (−1)floor [(n+0.5∓0.5)/2] ui (t)W0 t − nTs
−0.5T +nT
  s
w

c(t) 

if n is even
60

× T dt + Ul Tw ,
 0
if n is odd 
SER (dB)
c t + 
4
(21)
40
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 offset,
is located at zero frequency, while its desired component
is located at the frequency fs /4, as indicated by coefficient 20
0 0.02 0.04 0.06 0.08 0.1
(−1)floor [(n+0.5∓0.5)/2] . Thus, the DC offset can be easily fil-
γ
tered out in the receiver DP.
For the modified SC, we can write f0 / fs = 0.25
f0 / fs = 0.75
u nT s = f0 / fs = 1.25
0.5Tw +nTs

W0 t − nTs Figure 12: SER(γ) for various f0 / fs .
−0.5Tw +nTs
 

 ui (t)c(t) + U1 (−1)n/2 if n is even


× T dt, sin 2π f0 t in the first two multipliers of the modified SC can

 ui (t)c t −
0
+ U2 (−1)(n±1)/2 if n is odd 

4 be replaced by squarewaves with frequency f0 and time shift
(22) of 0.25T0 = 0.25/ f0 relative to each other when f0 / fs > 3,
and sufficient prefiltering is provided. This replacement fur-
where U1 and U2 are DC offsets in the first two multipliers. ther simplifies the modified SCs. Thus, the described modi-
Similar to (20), this equation can be rewritten as fication of the SCs substantially simplifies their realization at
the expense of slightly lower performance.
0.5Tw +nTs
floor [(n+0.5∓0.5)/2]

u nT s = (−1) W0 t − nTs 4.5. Use of orthogonality of WFG outputs
−0.5Tw +nTs
 
 As mentioned in Section 4.2, increase of fs /B makes inter-
 ui (t)c(t) + U1
if n is even

× T dt. nal filtering easier and may allow reduction of L. In addition

ui (t)c t +
0
+ U2 if n is odd 
 to reducing L, high-ratio fs /B makes possible reducing the
4
(23) number of multipliers N for given L if f0 /B is also high. This
possibility is discussed below.
It follows from (23) that both signal and DC offset compo- When (5) is true, the carrier of wn (t) generated for nth
nents after sampling are located at the frequency fs /4, as indi- sample is rotated by ±π/2 relative to the carrier of wn+2m+1 (t)
cated by coefficient (−1)floor [(n+0.5∓0.5)/2] . Therefore, the DC generated for (n + 2m + 1)th sample, where m is any inte-
component cannot be filtered out. ger. Thus, if the envelope of w0 (t) is rectangular, in some
Thus, DC offset and increased number and values of cases, wn (t) and wn+2m+1 (t) can be sent to the same multi-
IMPs lower the performance of the modified SC compared plier of the SC or RC with internal filtering even when these
to the canonical one. However, their performance is still sig- weight functions overlap. This property can be used to re-
nificantly better than that of the conventional baseband sam- duce N for a given L. For example, if Tw /Ts = 2 (L = 3) and
pling. Indeed, the entire signal processing is performed at u(t) = U0 cos(2π f0 t + ϕ0 ), one multiplier can be used for all
zero frequency in the conventional procedure. Consequently, 3 channels and perform, ideally accurate sampling. However,
besides multipliers, all subsequent analog stages contribute a pure sinewave cannot carry information. In the case of a
to the increase in the DC offset and nonlinear distortion. In bandpass signal u(t) = U(t) cos[2π f0 t + ϕ(t)], sampling er-
addition, baseband antialiasing filters create significant phase ror is unavoidable, and signal-to-error power ratio (SER) for
imbalance between I and Q channels. In the modified SCs, this error is
signal processing after 4-contact switches is performed at 2 2
fs /4, and subsequent analog stages do not increase nonlin- 16π 2 π f0 Tw − 1
SER = 3 (24)
ear distortions and DC offset. The phase mismatch among γ2 π f0 Tw ∓ 1
channels of the modified SC is negligible because all clock
impulses are generated in the control unit using the same when B 1/Tw . Here, “∓” corresponds to “±” in (5), γ =
reference oscillator, and proper design allows us to mini- BRMS / f0 , and BRMS is root mean square bandwidth of u(t).
mize time skew. As follows from Section 4.2, cos 2π f0 t and Figure 12 illustrates the dependence SER(γ) for several values
of f0 / fs . Since the spectrum of the error determined above is |Cm |

|C1 | |C1 | |C1 |
generally wider than Su ( f ), a part of this error can be filtered |C2 | |C2 | |C2 |
out in the receiver DP. Therefore, (24) is a lower bound of the 0.5 fs
actual SER. −0.5 fs −0.4 fs −0.2 fs 0 0.2 fs 0.4 fs 0.6 fs 0.8 fs f
This method of reducing the number of multipliers N
can also be used for L > 3 if the corresponding SER is suffi- (a)
ciently small. In this case, the minimally required N is

Ga ( f ), Su ( f )
 0.5Tw 0.5Tw B Bt

 +1 if is even,
Ts Ts
N = (25)

 0.5T w 0.5Tw
 if is odd.
Ts Ts
−0.5 fs −0.25 fs 0 0.25 fs 0.5 fs 0.75 fs f
For N > 1, this method can complicate the channel mis-
match compensation in the receiver DP described in the next (b)
section. It is important to mention that (24) can be used for
any N.
Gd ( f ), Su1 ( f ), Se ( f )
It follows from (24) and Figure 12 that the described
method can be used only with very high-ratios fs /B that cor-
respond exclusively to sigma-delta A/Ds.
−0.5 fs −0.25 fs 0 0.25 fs 0.5 fs 0.75 fs f
5. CHANNEL MISMATCH MITIGATION
(c)
5.1. Approaches to the problem
The SCs and RCs with internal filtering are inherently mul- Figure 13: Amplitude spectra and AFRs: (a) spectral components
tichannel. Therefore, the influence of channel mismatch on of d(t); (b) |Su ( f )|—solid line, |Ga ( f )|—dashed line; (c) |Su1 ( f )|
the performance of SDRs must be minimized. This is espe- and |Se ( f )|—solid line, |Gd ( f )|—dashed line.
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 dl = gl − g0 in the lth channel. Since samples of u(t) are gen-
power of n(t) is larger than that of s(t), the average power of erated in turn by all channels, the deflections d1 , d2 , . . . , dL ,
the error e(t) caused by the channel mismatch can be com- d1 , d2 , . . . , dL , d1 , d2 , . . . appear at sampling instants t = nT s
parable with or even exceed the power of s(t). as a periodic function d(t) with period LTs :
There are three approaches to this problem. The first of
them includes technical and technological measures that re- ∞

L

duce this mismatch: placing all the channels on the same die, d(t) = dl δ t − (kL + l)Ts , (26)
simplifying w0 (t), and correcting circuit design. The second k=−∞ l=1
approach is based on preventing an overlap of the signal and

mismatch error spectra. In this case, the error spectrum can where δ(t) is the delta function. The spectrum of d(t) is
be filtered out in the DP. The third approach is adaptive com- ∞

m
pensation of the channel mismatch in the DP. The first ap- Sd ( f ) = Cm δ f − f s , (27)
proach alone is sufficient for many types of transmitters and m=−∞ L
for receivers with limited dynamic range. In high-quality re-
ceivers, the measures related to this approach are necessary where coefficients
but usually not sufficient. Therefore, the second and third
L
approaches are considered below. 1
− j2πml
Cm = dl exp . (28)
LTs l=1 L
5.2. Separation of signal and error spectra
To determine the conditions that exclude any overlap be- As reflected by (27) and (28), Sd ( f ) is a periodic function
tween spectra Su1 ( f ) of u(nT s ) and Se ( f ) of e(t), we first find with the period fs because d(t) is discrete with sampling
Se ( f ). The phase mismatch among channels can be made period Ts . Therefore, it is sufficient to consider Sd ( f ) only
negligible because all clock impulses are generated in the within the interval [−0.5 fs , 0.5 fs [. Since d(t) is real-valued,
control unit using the same reference oscillator, and proper Sd ( f ) is even. Since d(t) is periodic with period LTs , Sd ( f ) is
design minimizes time skew. Therefore, it is sufficient to take discrete with the harmonics located at frequencies ±m fs /L,
into account only the amplitude mismatch caused by the dif- m = 1, 2, . . . , floor (L/2) within the interval [−0.5 fs , 0.5 fs [.
ferences among the channel gains g1 , g2 , . . . , gL . The average The spectral components of d(t) are shown in Figure 13a for
gain is g0 = (g1 +g2 + · · · +gL )/L, and the deflection from g0 is L = 5. When (5) is true, the images of the spectrum Su1 ( f )
of u(nT s ) occupy the following bands within the interval Table 3: Minimum values of fs /B.
[−0.5 fs , 0.5 fs [:
M↓ L→ 3 5 7 9 11
100 fs /B 6.1 10.4 14.9 19.6 24.5
− 0.25 fs − 0.5B, −0.25 fs + 0.5B
(29) 1000 fs /B 6.01 10.04 14.08 18.15 22.22
∪ 0.25 fs − 0.5B, 0.25 fs + 0.5B ,
where B is a bandwidth of u(t). Figure 13b shows |Su ( f )| and that it is relatively easy to avoid an overlap of Se ( f ) and Su ( f )
the AFR |Ga ( f )| of antialiasing filtering performed by the SC and exclude an impact of the SC channel mismatch on the re-
for f0 = 0.25 fs . Spectrum Se ( f ) is a convolution of Su ( f ) and ceiver performance when L = 3. For odd L > 3, significant
Sd ( f ). Taking (5) into account, we get increase of fs is required. Consequently, combining the SCs
and sigma-delta A/Ds almost automatically excludes this im-
∞

m pact if L is odd.
Se ( f ) = Cm Su f − fs − 0.25
m=−∞ L When L is odd, but (32) is not met, Se ( f ) and Su ( f ) over-
(30) lap. However, the overlap can be lowered by increasing fs /B
m
+ Su f − fs + 0.25 . and, when L ≥ 5, by reducing the Sd ( f ) harmonics adjacent
L
to ±0.5 fs since they create the closest-to-the-signal images of
Since e(t) is a real-valued discrete function with sampling pe- Se ( f ). Changing the succession of channel switching can re-
riod Ts , |Se ( f )| is an even periodic function with the period duce the harmonics. The succession that makes d(t) close to
fs that is unique within the interval [−0.5 fs , 0.5 fs [. a sampled sinewave minimizes the overlap.
It follows from (30) that if L is even, the error image cor- Figure 13c shows |Su1 ( f )| and |Se ( f )| for the situation
responding to m = ±L/2 falls to the frequencies ±0.25 fs , that when L is odd and condition (32) is met. Here, the error im-
is, within the signal spectrum. Therefore, Se ( f ) and Su ( f ) ages adjacent to the signal are created by C2 , and the more
cannot be separated. When L is odd, the situation is differ- distant images by C1 . The AFR |Gd ( f )| of the DP channel
ent. The centers of the images caused by the channel mis- filter is shown by the dashed line.
match are located at frequencies ±(r + 0.5) fs /(2L), where
5.3. Compensation of channel mismatch in DP
r = 0, 1, . . ., 0.5(L − 1) − 1, 0.5(L − 1) + 1, . . ., L − 1 within
the interval [−0.5 fs , 0.5 fs [. The bandwidth of each image is If, despite all the measures, the residual error caused by the
B1 = B + 2Bt1 , where Bt1 is the image one-sided transition mismatch still degrades the receiver performance, it can be
band. The images of Se ( f ) are created by coefficients Cm . adaptively compensated in the DP. This compensation can
Since these coefficients are different, the images have differ- be performed either during the operation mode simultane-
ent transition bands. However, we assume for simplicity that ously with signal processing or during a separate calibration
transition bands of all images are equal to those of the most mode. In all cases, channel gains gl are estimated first, and
powerful image. Mean values of u(t) and e(t) are equal to then deflections dl are compensated.
zero. Denoting the standard deviation of u(t) and e(t) as σu There are many methods of fast channel gain estima-
and σe , respectively, we can state that σu
σe . The standard tion in calibration mode. For example, when all the copies
deviation σe1 of the most powerful spectral image of e(t) al- wn (t) of w0 (t) are simultaneously applied to the SC multipli-
ways meets condition σe1 ≤ σe . It is reasonable to assume ers and a test signal is sent to the SC input, estimation time is
Bt /Bt1 = σu /σe1 = M where Bt is the antialiasing-filter one- Te = Tw + LTs = (2L − 1)Ts , assuming that Ts is required for
sided transition band. Thus, Bt1 = Bt /M and M > 1. Taking the hold and clear modes in each channel. A sinewave with
into account that Bt ≤ 0.5 fs − B, we obtain frequency f0 is the simplest test signal. The estimation can
also be done when wn (t) are delayed relative to each other by

Ts , like in the operational mode. If (5) is true and the test sig-
0.5 fs − B
B1 ≤ B + 2 . (31) nal is a sinewave with frequency f0 and arbitrary initial phase,
M
Te = 2Tw + (L + 1)Ts = (3L − 1)Ts because two consecutive
Since channel filtering in the receiver DP removes all the samples are required for each channel to estimate its gain.
spectral components of e(t) outside bands (29), only the part When the phase shift between the sinewave and the carrier of
of Se ( f ) which falls within these bands degrades the receiver w0 (t) is equal to ±45◦ , the estimation time can be reduced to
performance. It follows from (29) and (30) that Se ( f ) and Te = Tw + LTs = (2L − 1)Ts .
Su ( f ) do not overlap if (B + B1 ) ≤ fs /L. Inequality (31) al- Channel mismatch compensation performed during the
lows us to rewrite this condition as follows: operation mode requires much longer estimation because
u(t) is a stochastic process. The block diagram of a simplified
fs (M − 1) version of such a compensator is shown in Figure 14. Here, a
≥ 2L and M ≥ L. (32)
B (M − L) demultiplexer (DMx) distributes digital words resulting from
the SC samples among L digital channels. Each digital chan-
According to (32), fs /B → 2L when M → ∞. In practice, nel corresponds to the SC channel with the same number.
M ≥ 100. Table 3 shows the minimum values of fs /B re- Averaging units (AU) calculate the mean magnitudes of sam-
quired to filter out e(t) when L is odd. It follows from Table 3 ples in each channel. The mean magnitudes are processed
out antialiasing filtering and sampling, and an A/D quantizes

From 1
A/D AU 1 1 the output of the SC. All further processing is performed in
DMx .. K1 . Mx DF a DP.
.
. .
When multiplication of ui (t) by w0 (t) is performed in the
L AU L L analog domain, the carrier c(t) of w0 (t) is a sinewave, the en-
. . . KL velope W0 (t) of w0 (t) is a smooth function, and the AMP has
. sufficient dynamic range, prefiltering in the RF strip is used
GS ..
only to limit the receiver frequency range R. The same type
of prefiltering can be used when c(t) is nonsinusoidal and/or
Figure 14: Digital channel mismatch compensator. W0 (t) is not a smooth function, but R is narrower than half
an octave. Such prefilters do not require any adjustment dur-
ing frequency tuning.
If the conditions above are not met, the prefilter band-
To DP width should be narrower than R. Nonsinusoidal c(t) and
RF strip SC A/D
nonsmooth W0 (t) require the prefilter bandwidth that does
not exceed half an octave. In practice, the prefilter bandwidth
Figure 15: Modified receiver AMP architecture with digitization at is determined as a result of a tradeoff. Indeed, on the one
the RF. hand, reduction of the prefilter bandwidth allows increasing
its transition band. This simplifies IC implementation of the
prefilter. On the other hand, increase in the prefilter band-
in a gain scaler (GS), which generates coefficients Kl that
width simplifies its frequency tuning.
compensate the channel mismatch. A multiplexer (Mx) com-
In any case, signal u(t) intended for digitization is only
bines the outputs of all the channels. A subsequent digital fil-
a part of ui (t), and u(t) usually has wider spectrum than a
ter (DF) provides the main frequency selection. In practice,
desired signal s(t) since channel filtering is performed in the
channel mismatch compensation during the operation mode
DP. Therefore, a reasonable algorithm of the automatic gain
requires the most statistically efficient methods of gl estima-
control (AGC) is as follows. The RF strip gain should be reg-
tion, and the compensator should be designed together with
ulated by a control signal generated at the output of a digital
automatic gain control (AGC) of the receiver.
channel filter and constraints generated at the input of the
SC and at the output of the D/A. These constraints prevent
6. AMPS’ ARCHITECTURES BASED ON SAMPLING clipping of u(t) caused by powerful interference, which is fil-
AND RECONSTRUCTION WITH INTERNAL tered out by the digital channel filter, and clipping of ui (t)
FILTERING caused by powerful interference, which is filtered out by the
SC. To compensate level variations due to the constraints,
6.1. General
feed-forward automatic scaling is usually required in the DP
It is shown in Section 2 that the SDR front ends with band- with fixed-point calculations.
pass sampling, reconstruction, and antialiasing filtering po- Reconfiguration or adaptation of the receiver at the same
tentially provide the best performance. At the same time, f0 usually can be achieved by varying only W0 (t). Frequency
conventional methods of sampling, reconstruction, and fil- tuning requires shifting the AFR of the SC along the fre-
tering limit flexibility of the front ends, complicate their IC quency axis and, sometimes, adjusting the prefilter AFR. The
implementation, and do not allow achieving their potential AMP has to carry out only coarse frequency tuning. Fine
parameters. It follows from Section 3 that implementation of tuning with the required accuracy can be performed in the
the novel sampling and reconstruction techniques with inter- receiver DP. The reasonable increment ∆ f of coarse tun-
nal filtering can eliminate these drawbacks and provide some ing is about 0.1B, where B is the bandwidth of u(t). Thus,
additional benefits. Sections 4 and 5 demonstrate that chal- the number of different center frequencies f0 within the fre-
lenges of the proposed techniques realization can be over- quency range R is about 10 R/B. In most cases, coarse tun-
come. The impact of these techniques on the architectures of ing requires changing both c(t) and W0 (t). Indeed, when
the radios’ AMPs is discussed below. f0 is changed, usually fs should also be changed to preserve
condition (5). This in turn necessitates changing W0 (t) be-
6.2. Modified receiver AMPs cause certain relations between fs and Tw are necessary to
Implementation of sampling with internal antialiasing filter- suppress noise and interference within intervals (6). During
ing in digital receivers requires modification of their front coarse tuning, W0 (t) can remain unchanged only when pre-
ends. Since accumulation of the signal energy in the storage vious and subsequent frequencies f0 have the same optimal
capacitors of the SCs significantly reduces the required gain fs and keeping unchanged W0 (t) does not cause additional
of AMPs, and antialiasing filtering performed by the SCs is discontinuities in w0 (t). However, this happens rarely, and
flexible, it is reasonable to consider the possibility of signal frequency tuning in the AMP shown in Figure 15 is rela-
digitization at the receiver RF. This leads to the simplest AMP tively complex. The SCs described in Section 4.3 cannot be
architecture shown in Figure 15. Here, an RF strip performs used in this architecture due to possible leakage of the c(t)
prefiltering and all the required amplification, an SC carries generator.
IF strip From DP
D/A RC RF strip PA
RF Low-Q
LPF
section filter
LO Figure 17: Modified transmitter AMP architecture with recon-
struction at the RF.
To DP
SC A/D
IF strip
Figure 16: Modified superheterodyne receiver AMP architecture
with sampling at the IF. Low-Q
BPF PA
filter
LO
A superheterodyne architecture of the receiver AMP
modified to accommodate sampling with internal antialias- From DP
ing filtering at the IF is shown in Figure 16. Compared to the D/A RC
previous architecture, this one simplifies both frequency tun-
ing and prefiltering. Here, an RF section performs image re- Figure 18: Modified offset up-conversion transmitter AMP archi-
jection and preliminary amplification of the sum of a desired tecture with reconstruction at the IF.
signal, noise, and interference. Prefiltering and further signal
amplification are carried out in an IF strip. This prefiltering
is performed by a filter with low quality factor (Q) that can 6.3. Modified transmitter AMPs
be implemented on a chip. Similar to the case of receivers, implementation of recon-
In principle, prefiltering is necessary only when c(t) is struction with internal filtering in transmitters requires
nonsinusoidal and/or W0 (t) is not a smooth function. Oth- modification of their AMPs. This modification affects only
erwise, it can be excluded. However, as shown in Section 4.2, the transmitter drive (exciter) and does not influence the
use of a K-level W0 (t) and a squarewave carrier c(t) radi- transmitter PA. Signal reconstruction at the transmitter RF
cally simplifies the SC due to reducing multiplying D/As to leads to the simplest AMP architecture shown in Figure 17.
a relatively small number of switches. Besides, it allows in- Here, digital words corresponding to the samples of a band-
creasing the receiver IF, which, in turn, simplifies image re- pass signal are formed in a DP at the transmitter RF. Then,
jection in the RF section. When the receiver frequency range they are converted to the analog samples by a D/A. An RC
R is wide, a variable IF allows avoiding spurious responses. In reconstructs the bandpass analog signal and carries out main
practice, two or three different f0 ’s are sufficient, and they can analog filtering. A subsequent RF strip amplifies the signal
be selected so that transitions from one f0 to another require to the level required at the input of a PA and performs post-
minimum adjustment. For example, these transitions may filtering. This postfiltering is absolutely necessary when c(t)
require changing only c(t). If these frequencies are within the is nonsinusoidal and/or W0 (t) is not a smooth function. Al-
bandwidth of the low-Q filter, the latter does not require any though the AMP in Figure 17 looks simple, its implementa-
adjustment when the IF is changed. In both AMP architec- tion causes problems related to frequency tuning of the trans-
tures shown in Figures 15 and 16, all measures that reduce mitter and digital-to-analog conversion of bandpass signals
the influence of the SC channel mismatch on the receiver per- at the varying RF.
formance (see Section 5) should be taken. Therefore, when These problems are solved in the offset up-conversion
condition (32) is not met, digital channel mismatch compen- AMP architecture modified to accommodate bandpass re-
sator has to be implemented in the receiver DP. construction with internal filtering at the IF shown in
Despite the differences, the AMP architectures in Figures Figure 18. The fact that reconstruction, preliminary ampli-
15 and 16 utilize the advantages of sampling with internal fication, and postfiltering of bandpass analog signals are car-
antialiasing filtering (see Section 3.5). First of all, removal ried out at the transmitter IF significantly simplifies realiza-
of high-quality conventional antialiasing filters (e.g., SAW, tion of this procedures. An RC performs main reconstruction
crystal, mechanical, ceramic) and implementation of the SCs filtering, while postfiltering is carried out by a low-Q IF filter
with variable w0 (t) make these architectures realizable on a that can be placed on a chip.
chip, reconfigurable, and adaptive. Then, the proposed sam- Implementation of reconstruction with internal flexible
pling significantly improves performance by adding to the filtering makes the transmitter AMPs easily reconfigurable
benefits of bandpass sampling described in Section 2.1 the and highly adaptive and increases their scale of integration.
following advantages. A variable IF allows avoiding spuri- This reconstruction also reduces the required AMP gain due
ous responses in a superheterodyne AMP. Nonlinear distor- to more efficient utilization of the D/A output current. As a
tions are radically reduced due to rejection of out-of-band result, reconstruction can be performed closer to the antenna
IMPs of all preceding stages and lower input current of the than in conventional architectures.
SC caused by accumulation of the signal energy. This accu-
mulation also filters out jitter, improving performance and
7. CONCLUSIONS
speed of the A/D. Finally, the accumulation of signal energy
lowers the required AMP gain and allows sampling close to In modern SDRs, analog front end architectures with band-
the antenna. pass sampling, reconstruction, and antialiasing filtering can
potentially provide the best performance of both receivers lowered by increasing fs /B and, when L ≥ 5, by reducing the
and transmitters. However, conventional methods of per- Sd ( f ) harmonics adjacent to ±0.5 fs . The third approach is
forming these procedures limit flexibility, complicate IC im- based on adaptive compensation of the channel mismatch in
plementation, and do not allow achieving the potential per- the DP.
formance of the radios. In principle, sampling and reconstruction with internal
Novel sampling and reconstruction techniques with in- filtering can be carried out at the radios’ RFs. However, fre-
ternal filtering derived from the sampling theorem elimi- quency conversion to an IF significantly simplifies practical
nate these problems. The techniques provide high flexibility realization of the modified SDRs.
because their filtering and other properties are determined Implementation of the SCs and RCs with internal filter-
by weight functions w0 (t) that can be dynamically changed. ing in SDRs radically increases reconfigurability, adaptivity
Since technology of the SCs and RCs with internal filtering is and scale of integration of their front ends. Simultaneously, it
compatible with IC technology, they radically increase scale improves performance of the radios due to significant reduc-
of integration in the AMPs. The RCs provide more efficient tion of nonlinear distortions, rejection of out-of-band noise
utilization of the D/A output current than conventional tech- and IMPs of all stages preceding sampling, avoiding spurious
niques. The SCs accumulate the input signal energy. This responses, and filtering out jitter. This implementation also
accumulation filters out jitter, improving performance and substantially reduces front ends of SDRs, enabling sampling
speed of A/Ds, and reduces the input current. The reduc- and reconstruction close to the antenna.
tion of the input current lowers nonlinear distortions and
required gain of AMPs. REFERENCES
Technical challenges of the SCs and RCs practical real-
[1] T. Anderson and J. W. Whikohart, “A digital signal processing
ization can be overcome by proper selection of w0 (t) and
HF receiver,” in Proc. 3rd International Conference on Com-
optimization of their architectures for a given w0 (t). Selec- munication Systems & Techniques, pp. 89–93, London, UK,
tion of w0 (t) requires multiple tradeoffs. Simplification of February 1985.
the SCs and RCs is usually intended to reduce complexity [2] C. M. Rader, “A simple method for sampling in-phase and
and/or number of multiplications. Minimum complexity of quadrature components,” IEEE Trans. Aerosp. Electron. Syst.,
multiplications is achieved when multiplying D/As or ana- vol. 20, no. 6, pp. 821–824, 1984.
log multipliers can be replaced by a relatively small number [3] M. V. Zarubinskiy and Y. S. Poberezhskiy, “Formation of read-
of switches. This can be accomplished, for instance, by us- outs of quadrature components in digital receivers,” Telecom-
munications and Radio Engineering, vol. 40/41, no. 2, pp. 115–
ing w0 (t) with K-level envelope W0 (t) and squarewave car-
118, 1986.
rier c(t) when W0 (t) and c(t) are properly synchronized and [4] Y. S. Poberezhskiy, Digital Radio Receivers, Radio & Commu-
f0 / fs is adequately high ( f0 / fs > 3 is usually sufficient). When nications, Moscow, Russia, 1987 (Russian).
f0 / fs is low, c(t) should also be a several-level function. There [5] J. B.-Y. Tsui, Digital Microwave Receivers: Theory and Concepts,
are other classes of w0 (t) that allow replacing multipliers by Artech House, Norwood, Mass, USA, 1989.
a small number of switches. [6] M. E. Frerking, Digital Signal Processing in Communication
Separate multiplications of the input signal ui (t) by Systems, Van Nostrand Reinhold, New York, NY, USA, 1994.
W0 (t) and c(t) and use of only two multipliers for multiply- [7] W. E. Sabin and E. O. Schoenike, Eds., Single-Sideband Sys-
ing by c(t) lead to a method that significantly simplifies the tems and Circuits, McGraw-Hill, New York, NY, USA, 2nd edi-
SCs. Although this is achieved at the expense of slightly re- tion, 1995.
[8] J. Mitola III, “The software radio architecture,” IEEE Com-
duced performance compared to the canonical SCs, the sim-
mun. Mag., vol. 33, no. 5, pp. 26–38, 1995.
plified SCs still provide significantly better performance of [9] R. I. Lackey and D. W. Upmal, “Speakeasy: the military soft-
the radios than conventional sampling. ware radio,” IEEE Commun. Mag., vol. 33, no. 5, pp. 56–61,
Increase of fs /B simplifies antialiasing and reconstruc- 1995.
tion filtering and allows reduction of L in some cases. When [10] B. Razavi, “Recent advances in RF integrated circuits,” IEEE
both fs /B and f0 /B are sufficiently high, use of WFG outputs’ Commun. Mag., vol. 35, no. 12, pp. 36–43, 1997.
orthogonality allows reduction of N for a given L. However, [11] H. Meyr, M. Moeneclaey, and S. A. Fechtel, Digital Commu-
this method is practical only for very high fs /B. nications Receivers, John Willey & Sons, New York, NY, USA,
Since SCs and RCs with internal filtering are inherently 1998.
[12] A. A. Abidi, “CMOS wireless transceivers: the new wave,” IEEE
multichannel, the impact of channel mismatch on the per- Commun. Mag., vol. 37, no. 8, pp. 119–124, 1999.
formance of SDRs should be minimized. There are three ap- [13] J. Mitola III, Software Radio Architecture, John Willey & Sons,
proaches to the problem. The first of them includes all tech- New York, NY, USA, 2000.
nical and technological measures that reduce the mismatch. [14] C. Chien, Digital Radio Systems on a Chip: a System Approach,
The second one is based on preventing an overlap of signal Kluwer Academic, Boston, Mass, USA, 2000.
and mismatch error spectra. This can be achieved only when [15] M. Helfenstein and G. S. Moschytz, Eds., Circuits and Sys-
L is odd, and condition (32) is met. In this case, the error tems for Wireless Communications, Kluwer Academic, Boston,
Mass, USA, 2000.
spectrum can be filtered out in the DP. Combination of the [16] Y. Sun, Ed., Wireless Communication Circuits and Systems, IEE,
SCs with odd L and sigma-delta A/Ds almost automatically London, UK, 2004.
excludes the overlap. When L is odd, but condition (32) is [17] Y. S. Poberezhskiy and G. Y. Poberezhskiy, “Sampling with
not met, the overlap cannot be avoided. However, it can be weighted integration for digital receivers,” in Proc. Digest of
IEEE MTT-S Symposium on Technologies for Wireless Appli-

cations, pp. 163–168, Vancouver, British Columbia, Canada,
February 1999.
[18] Y. S. Poberezhskiy and G. Y. Poberezhskiy, “Sampling tech-
nique allowing exclusion of antialiasing filter,” Electronics Let-
ters, vol. 36, no. 4, pp. 297–298, 2000.
[19] Y. S. Poberezhskiy and G. Y. Poberezhskiy, “Sample-and-hold
amplifiers performing internal antialiasing filtering and their
applications in digital receivers,” in Proc. IEEE Int. Symp. Cir-
cuits and Systems (ISCAS ’00), vol. 3, pp. 439–442, 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. 90–92, 2001.
[21] Y. S. Poberezhskiy and G. Y. Poberezhskiy, “Signal reconstruc-
tion technique allowing exclusion of antialiasing filter,” Elec-
tronics Letters, vol. 37, no. 3, pp. 199–200, 2001.
[22] Y. S. Poberezhskiy and G. Y. Poberezhskiy, “Sampling and sig-
nal reconstruction structures performing internal antialias-
ing filtering,” in Proc. 9th International Conference on Elec-
tronics, Circuits and Systems (ICECS ’02), vol. 1, pp. 21–24,
Dubrovnik, Croatia, September 2002.
[23] Y. S. Poberezhskiy and G. Y. Poberezhskiy, “Sampling and sig-
nal reconstruction circuits performing internal antialiasing
filtering and their influence on the design of digital receivers
and transmitters,” IEEE Trans. Circuits Syst I: Regular Papers,
vol. 51, no. 1, pp. 118–129, 2004, erratum ibid. vol. 51, no. 6,
p. 1234, 2004.
Yefim 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 Scientific 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.
c 2005 Bedri A. Cetiner et al.
A Reconfigurable Spiral Antenna

for Adaptive MIMO Systems
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
Email: gpli@uci.edu
F. De Flaviis
Email: franco@uci.edu
Received 17 October 2004; Revised 15 March 2005
We present a reconfigurable spiral antenna for use in adaptive MIMO systems. The antenna is capable of changing the sense of
polarization of the radiated field. It is fabricated by using an RF-MEMS technology compatible with microwave laminate substrates
developed within the author’s 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
configuring 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 MIMO systems, reconfigurable spiral antenna, radio-frequency microelectromechanical systems.
1. INTRODUCTION smart antennas and adaptive coding and modulation tech-

niques, which have been proven to dramatically increase
Reconfigurable wireless communication systems, which can
the wireless channel capacity and improve the diversity. Al-
dynamically adapt themselves to constantly changing envi-
though in these studies considerable attention has been given
ronmental propagation characteristics, will be the key for
to the performance analysis of these systems in the context
the next-generation communication scenarios. A communi-
of coding and signal processing architectures, the investiga-
cation system, capable of changing its output dynamically
tion of the antenna aspect is mainly limited to the impact
through reconfigurability features, allows optimal system-
of the number of antenna elements with little consideration
level performance at all times, regardless of changing char-
on their radiation and polarization characteristics as well as
acteristics of the communication environment.
array geometry. Multiple antenna elements of these systems
There has recently been enormous research performed
are fixed by the initial design and cannot change their prop-
on MIMO systems [1] with associated technologies such as
erties, that is, radiation pattern, polarization, and operating
frequency.
This is an open access article distributed under the Creative Commons We have recently developed a microwave-laminate-
Attribution License, which permits unrestricted use, distribution, and compatible RF-MEMS technology [2, 3, 4] that allows
reproduction in any medium, provided the original work is properly cited. fabricating multifunction reconfigurable antennas (MRAs)
A Reconfigurable Spiral Antenna for Adaptive MIMO Systems 383
on microwave laminate substrates that best meet the an- The proposed reconfigurable spiral antenna architec-
tenna performance characteristics. An MRA can dynami- ture is built on a number of printed rectangular-shaped
cally configure its structural architecture and thus alter its metal strips interconnected by RF-MEMS actuators on a
performance properties, that is, polarization, radiation pat- microwave-laminate printed circuit board (PCB) substrate,
tern, and operating frequency. Therefore, an adaptive MIMO RO4003-FR4 (εr = 3.38, tan δ = 0.002). Shown in Figure 1
system equipped with MRAs will not be constrained to em- are two adjacent strips interconnected by an RF-MEMS ac-
ploy a fixed antenna design over varying channel condi- tuator, which is made of a metallic movable membrane, sus-
tions. This is an additional degree of freedom in adapt- pended over a metal stub protruding from an adjacent strip,
able parameters of an adaptive MIMO system and permits fixed to both ends of the strip through metallic posts. The op-
the selection of the best antenna properties and configura- timized height of these posts was found to be 8 µm for a good
tion in conjunction with the adapted transmission scheme tradeoff between up-position switch coupling and actuation
with respect to the channel condition [5]. Thus the gap be- voltage. Metal stubs are covered by silicon-nitride (SiNx ) film
tween theoretical MIMO performance and practice is mini- to prevent metallic membrane from sticking onto the stub
mized. upon contact. This film also provides a capacitive contact
Motivated by the features of next-generation wireless for actuator down state isolating RF signal from DC. A DC
MIMO communication systems, as stated above, we have bias voltage of approximately 50 V applied between the mem-
aimed at developing innovative antenna architectures, which brane and the stub causes an electrostatic force that pulls
combine multiple functions in one single antenna. One such the suspended membrane on top of the stub (actuator down
application example, a spiral antenna capable of changing its state or actuator on) and the actuator connects the strips
polarization state through microwave-laminate-compatible (see Figure 1c); otherwise strips are disconnected (actuator
RF-MEMS technology, is presented in this paper. In this up state or actuator off) (see Figure 1b). Judicious activation
application example, we only reconfigure the polarization of interconnecting actuators, that is, by keeping some of the
property, but there would be no process limitation to accom- actuators in the up position (zero bias) while activating the
modate reconfigurability features in operation frequency and rest of them by applying DC bias voltages, allows the recon-
radiation characteristics of a more complex reconfigurable figurable spiral to configure its architecture into single-arm
antenna design. 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
2. ANTENNA STRUCTURE, OPERATIONAL achieved. In Figure 2, for the clarity of illustration, each con-
MECHANISM FABRICATION, AND RESULTS figured geometry is depicted separately and actuators in the
up state are shown without metallic membrane. The antenna
Spiral antennas are attractive for communication appli-
is fed by a single coaxial probe as shown in Figure 2c. The
cations where broadband characteristics with respect to
supply voltage is connected to the proper locations on the
both input impedance and radiation pattern are required.
antenna segments through resistive bias lines so as to prevent
There have been extensive investigations regarding radiation
RF signal from being shorted by the DC power supply.
characteristics of spiral antennas with different geometri-
The proposed prototype antenna is aimed to radiate an
cal shapes such as circular, rectangular, and eccentric [6, 7].
axial beam of RHCP and LHCP fields. It is known that a
These antennas are mainly used to radiate circularly po-
single-arm rectangular spiral antenna with outermost arm
larized wave forming either an axial beam—normal to the
peripheral length (circumference) of C,
plane of the spiral—or tilted beam pattern—off-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 1λeff < C < 2λeff , (1)
circuit between the spiral and the feed line, which is needed
for multiarm spiral antennas. excites only the first radiation mode giving rise to an ax-
ial beam of circular polarization [8], where λeff = λ0 /[(εr +
1)/2]1/2 is the effective wavelength of the current traveling on
2.1. Antenna structure and operational mechanism
the spiral. The strip number and size are optimized so that
While the microwave-laminate-substrate-compatible RF- circumference of the antenna, C = 42 mm = 1.04λeff , satis-
MEMS technology has been used in [2, 4] for monolithic in- fies (1), and minimum number of actuators with associated
tegration of antenna elements with RF-MEMS switches, the bias circuitries are needed.
role of RF-MEMS switches, which are located on antenna
feed lines, has been limited to routing the RF signal feed- 2.2. Fabrication and results
ing the antennas. In this work, we integrate a number of RF- For reference, as a first step, conventional single-arm rect-
MEMS actuators within the geometrical structure of the an- angular spiral antennas radiating circularly polarized field
tenna to construct a reconfigurable spiral antenna. In other along their axes have been designed, fabricated, and charac-
words, RF-MEMS actuators are used as part of the physical terized. We chose to use RO4003-FR4 (εr = 3.38, tan δ =
structure of the antenna, owing to the monolithic integra- 0.002) microwave laminated substrate [9] to realize the an-
tion capability of the processing technique, providing a large tennas due to its low cost and widespread use in wireless sys-
degree of structural reconfigurability. tems. The substrate is conductor-backed to ensure that the
t2
A g A
t1
PCB
w2
w3
Aluminum Substrate
A A Electroplated copper SiNx
PCB copper
(b)
w2
h1
h2
w1
Aluminum Substrate
Aluminum Substrate Electroplated copper SiNx
PCB copper SiNx PCB copper
(a) (c)
Figure 1: RF-MEMS actuator interconnecting two adjacent metallic strips. (a) Top view; width of metal strip w1 = 800 µm; width of stub
w2 = 100 µm; width of membrane w3 = 150 µm. (b) Side view (up position); thickness of nitride t1 = 0.2 µm; thickness of membrane
t2 = 0.5 µm; air gap g = 7.8 µm. (c) Side view (down position); thickness of electroplated copper h1 = 8 µm; thickness of PCB copper
h2 = 16 µm.
z (coax feed)
50 V 0V y
x
Actuator
Coax feeding in the up 50 V
state
point 0V
Actuator
in the down
50 V 0V 50 V 0V state
50 V 0V
0V 50 V
(a) (b)
εr = 3.38 H
Coaxial cable
(c)
Figure 2: Schematics of the single-arm rectangular spiral antennas, which are reconfigured from the proposed reconfigurable 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.
Spiral
segments
DC bias path Coax inner

conductor
(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
reconfigurable spiral antenna. (a) Antenna pattern, DC bias path, and via formation. (b) Dielectric layer deposition and sacrificial layer
planarization. (c) Aluminum (Al) membrane deposition. (d) Final release.
−5
Return loss (dB)
−10
−15
−20
−25
−30
3 3.5 4 4.5 5 5.5 6
Figure 4: Photograph of the microfabricated reconfigurable spiral
Frequency (GHz)
antenna.
Calculated
antenna radiates broadside to the printed spiral surface. Sub- Measured
strate thickness is chosen to be 7.6 mm, which is one of the
standard thicknesses for PCB family substrates, the closest Figure 5: Return loss of the antenna for RHCP radiation.
one to the quarter wavelength at a center design frequency
of 5 GHz. Theoretical characterization of the antenna struc- both sides. We first form the segments of the antenna and
ture is conducted by using Ansoft HFSS 8.5 full-wave anal- the planar part of the bias circuitry by wet-etching the cop-
ysis tool [10] based on finite element method which takes per layer. Vertical vias for bias circuitry and coax feed are
into account the edge effects due to finite-size dielectric and created by standard PCB processes. After this step, a thin
conducting plane of the antenna. layer of HDICP CVD SiNx is deposited and etched by re-
A brief fabrication sequence for monolithic integration active ion etching such that the SiNx covers only the tips of
of RF-MEMS actuators with rectangular-shaped strip seg- the metal stubs protruding from the antenna segments (see
ments of the spiral antenna is given in Figure 3. Details of Figure 1a). We continue fabricating RF-MEMS actuators fol-
the fabrication process can be found in [2, 3, 4], so here lowing the process flow shown in Figures 3b and 3d without
we only briefly explain the process. The fabrication begins affecting the antenna structure. A photograph of the micro-
with RO4003 laminate with copper layers of 16 micron on fabricated antenna is shown in Figure 4, and Figure 5 shows
90 90
135 45 135 45
−10 −10
−20 −20
−30 −20 −10 −30 −20 −10

180 0 180 0
225 315 225 315
270 270
Calculated ER Calculated EL Calculated ER Calculated EL

Measured ER Measured EL Measured ER Measured EL
(a) (b)
90 90
135 45 135 45
−10 −10
−20 −20
−30 −20 −10 −30 −20 −10

180 0 180 0
225 315 225 315
270 270

(c) (d)
Figure 6: Radiation patterns for the RHCP spiral antenna at 5 GHz in (a) φ = 0 plane, (b) φ = 45 plane, (c) φ = 90 plane, and (d) φ = 135
plane.
the return loss of the spiral antenna with counterclockwise forming an almost axial beam pattern. This slight tilt from
sense of winding corresponding to the RHCP radiation. The the z-axis is due to the asymmetry of the antenna struc-
simulated result is also validated by comparison with experi- ture with respect to the z-axis. The measured average half-
mental data in this figure. Due to the symmetry between two power beamwidth (HPBW) is approximately 105◦ . The an-
antenna configurations, the RHCP and LHCP spirals exhib- tenna radiates almost entirely circularly polarized wave in the
ited almost identical return loss with a VSWR of less than z-direction with axial ratio value of 0.9 dB. The gain at this
two covering the frequency band of 4.3–5.4 GHz. The size direction is 5.3 dB. Variations of axial ratio and gain in the
of the antenna geometry can be scaled to change the oper- z-direction with respect to frequency are shown in Figure 8.
ational bandwidth to make it compatible with popular IEEE The circular polarization bandwidth over which the axial ra-
networking standards such as IEEE 802.11a. tio is less than 3 dB is approximately 11%. Gain of the an-
Measured and calculated radiation patterns at 5 GHz in tenna with average value of 4.9 dB shows small variation over
four different planes of φ = 0◦ , 45◦ , 90◦ , 135◦ are shown in this bandwidth. The difference in performance characteris-
Figures 6 and 7 for RHCP and LHCP spirals, respectively. tics between the RF-MEMS integrated spiral antenna and
As seen from these figures, the antennas radiate circularly conventional single-arm rectangular spiral antenna was ob-
polarized wave slightly off-broadside to the plane of spiral, served to be negligible.
90 90
135 45 135 45
−10 −10
−20 −20
−30 −20 −10 −30 −20 −10

180 0 180 0
225 315 225 315
270 270

(a) (b)
90 90
135 45 135 45
−10 −10
−20 −20
−30 −20 −10 −30 −20 −10

180 0 180 0
225 315 225 315
270 270

(c) (d)
Figure 7: Radiation patterns for the LHCP spiral antenna at 5 GHz in (a) φ = 0 plane, (b) φ = 45 plane, (c) φ = 90 plane, and (d) φ = 135
plane.
3. CONCLUSION reconfigurable antenna systems with MEMS integration at

low cost with high system-level performances. The defining
Motivated by the fact that the antenna properties (polariza- feature of this technology is its capability of allowing mono-
tion, operating frequency, and radiation behavior) can be lithic integration of RF-MEMS, with antenna structures on
used as additional degrees of freedom in adaptive MIMO any microwave laminate substrate that best meets the an-
system parameters, we presented a reconfigurable antenna tenna performance characteristics. Experimental impedance
architecture employing RF-MEMS as a vehicle to achieve po- and radiation characteristics of the proposed architecture are
larization reconfigurability. The antenna builds on a num- in excellent agreement with theoretical results. The results
ber of rectangular-shaped metallic strips monolithically in- showed that the antennas radiate right-hand and left-hand
terconnected with RF-MEMS actuators. Its architecture is circularly polarized axial beam waves with good axial ratio
dynamically reconfigured into RHCP and LHCP single-arm and gain values covering the design frequency bandwidth of
rectangular spirals with opposite sense of windings by acti- 4.3–5.4 GHz. If desired, this bandwidth can be changed by
vating some of the actuators while keeping the rest in the scaling the size of the antenna to make it compatible with
off-state. RF-MEMS technology compatible with microwave IEEE networking standards such as 802.11a. The presented
laminate substrates is the key enabling multifunctional application example has been intended to demonstrate an
12 [6] J. A. Kaiser, “The archimedean two-wire spiral antenna,” IRE

Transactions on Antennas and Propagation, vol. 8, pp. 312–323,
1960.
Axial ratio and gain (dB)
9 [7] R. T. Gloutak Jr. and N. G. Alexopoulous, “Two-arm eccentric

spiral antenna,” IEEE Trans. Antennas Propagat., vol. 45, no. 4,
pp. 723–730, 1997.
[8] H. Nakano, J. Eto, Y. Okabe, and J. Yamauchi, “Tilted-
6
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–
3 24, 2002.
[9] Microwave Products Technical Information, Rogers, Chandler,
Ariz, USA, 2001.
0 [10] Ansoft HFSS 8.5, Ansoft Corporation, Pittsburgh, Pa, USA.
4.6 4.8 5 5.2 5.4 5.6
Frequency (GHz)
Bedri A. Cetiner was born in Gazi Magosa,
Calculated axial ratio Calculated gain Cyprus, in 1969. He received the Ph.D. de-
Measured axial ratio Measured gain gree in electronics and communications en-
gineering from the Yildiz Technical Uni-
Figure 8: Frequency responses of axial ratio and gain in the z- versity, Istanbul, in 1999. From November
direction for RHCP spiral antenna. 1999 to June 2000 he was with the Univer-
sity of California, Los Angeles, as a North-
ern Atlantic Treaty Organization (NATO)
initial design and fabrication that will pave the way for novel Science Fellow. He then joined the Depart-
ment of Electrical Engineering and Com-
antennas into which multifunctional features are dynami-
puter Science, the University of California (UCI), Irvine, where he
cally combined by making use of large number of actua- worked as a postdoctoral researcher and Research Specialist from
tors. Multifunction reconfigurable antenna is very promis- June 2000 to June 2004. He is currently an Assistant Professor at
ing in the establishment of the next-generation multifunc- the Space Science Center, Morehead State University. His research
tion highly integrated reconfigurable communication archi- interest is focused on the analysis and design of microwave circuits
tectures. 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 reconfigurable antennas for use in adaptive
ACKNOWLEDGMENTS multi-input multi-output (MIMO) systems. He is the principal in-
This work was supported in part by the US Air Force under ventor of microwave-laminate-compatible RF-MEMS technology.
He is a Member of the IEEE Antennas and Propagation, Microwave
Grant F04611-03-C-004. The authors would like to acknowl-
Theory and Techniques, and Communications Societies.
edge the staffs of the Integrated Nanosystems Research Facil-
ity clean room, University of California, Irvine, for their help J. Y. Qian received a B.S. degree in opto-
in this work. electronics and an M.S. degree in optics
from the University of Science and Technol-
REFERENCES ogy of China, in 1994 and 1997, respectively,
and a Ph.D. degree in electrical engineer-
[1] Special Issue on MIMO Systems and Applications, IEEE J. Se- ing and computer science from the Univer-
lect. Areas Commun., vol. 21, no. 3, April 2003. sity of California (UCI), Irvine, in 2004. He
[2] B. A. Cetiner, J. Y. Qian, H.-P. Chang, M. Bachman, G. P. is currently working as a postgraduate re-
Li, and F. De Flaviis, “Monolithic integration of RF MEMS searcher at UCI. His research interests in-
switches with a diversity antenna on PCB substrate,” IEEE clude application of RF-MEMS switches for
Trans. Microwave Theory Tech., vol. 51, no. 1, pp. 332–335,
development of microwave devices, and small-size reconfigurable
2003.
antennas for smart wireless communication systems.
[3] H.-P. Chang, J. Y. Qian, B. A. Cetiner, F. De Flaviis, M.
Bachman, and G. P. Li, “RF MEMS switches fabricated on
G. P. Li developed a silicon silicide molec-
microwave-laminate printed circuit boards,” IEEE Electron
ular beam epitaxy (MBE) system while be-
Device Lett., vol. 24, no. 4, pp. 227–229, 2003.
[4] B. A. Cetiner, J. Y. Qian, H. P. Chang, M. Bachman, G. P. ing with the University of California at Los
Li, and F. De Flaviis, “Microwave laminate PCB compatible Angeles (UCLA), and then in the area of sil-
RF MEMS technology for wireless communication systems,” icon bipolar developed very large-scale in-
in Proc. IEEE Antennas and Propagation Society International tegration (VLSI) technology and process-
Symposium, vol. 1, pp. 387–390, Columbus, Ohio, USA, June related device physics while being with the
2003. IBM T. J. Watson Research Center. During
[5] B. A. Cetiner, H. Jafarkhani, J. Y. Qian, H. J. Yoo, A. Grau, his tenure as a Staff Member and Manager
and F. De Flaviis, “Multifunctional reconfigurable MEMS in- of the Technology Group, IBM, he coordi-
tegrated antennas for adaptive MIMO systems,” IEEE Com- nated and conducted research efforts in technology development
mun. Mag., vol. 42, no. 12, pp. 62–70, 2004. of high-performance and scaled-dimension (0.5 and 0.25 µm)
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 defining 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 filters, 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.
c 2005 Lars Berlemann et al.
Multimode Communication Protocols Enabling

Reconfigurable Radios
Lars Berlemann
Chair of Communication Networks, RWTH Aachen University, Kopernikusstraße 16, D-52074 Aachen, Germany
Email: ber@comnets.rwth-aachen.de
Ralf Pabst
Email: pab@comnets.rwth-aachen.de
Bernhard Walke
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 reconfigurable wireless communication sys-
tems. This focus extends the field of software-defined radios which usually concentrates on the physical layer. The generic protocol
stack comprises common protocol functionality and behavior which are extended through specific 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-specific functionality of the protocol software is realized through adequate parameterization and com-
position of the generic modules. The generic protocol stack allows an efficient realization of reconfigurable protocol software and
enables a completely reconfigurable wireless communication system. It is a first step from side-by-side realized, preinstalled modes
in a terminal towards a dynamic reconfigurable anymode terminal. The presented modules of the generic protocol stack can also
be regarded as a toolbox for the accelerated and cost-efficient development of future communication protocols.
Keywords and phrases: generic protocol stack, link layer functions, modular layer composition, reconfigurability, software-
defined radio.
1. INTRODUCTION Software-defined radios (SDRs) [4, 5] are a promising ap-

proach towards these multimode devices. The recent tech-
The radio access of future ubiquitous communication net-
nological progress allows an extension of the key issues in
works will be released from the constrains of cellular wireless
research of SDRs from the signal processing of the physical
networks, as for instance universal mobile telecommunication
layer on the complete communication chain used for wire-
system (UMTS), or wireless local area networks (WLANs).
less communication. The current research efforts are target-
Wireless mobile broadband systems, providing a patchy cov-
ing at the realization of cognitive radios [4, 6, 7]: self-aware,
erage in densely populated urban areas, will play an im-
frequency-agile radio systems that are able to identify un-
portant role. For details on such a fixed and planned relay-
used radio spectrum. These cognitive radios require proto-
based radio network, see [1, 2]. The addressed future wireless
col reconfigurability to unfold their advantage of dynamic
network will have to combine several radio access technolo-
spectrum usage. Therefore, this paper extends the focus of
gies (RATs). Consequently, multimode capable terminals and
SDRs on the protocol software used for reliable commu-
base stations are required to enable the seamless interwork-
nication over the air. Especially, the data link layer (DLL)
ing between these RATs. Multimode architectures can already
and network layer corresponding to the ISO/OSI reference
be found in existing systems, like IEEE 802.16 [3] with differ-
model [8] are considered. This work supplements the re-
ent modes of the physical layer (PHY).
search of the integrated project E2R [9] dealing with end-to-
end reconfigurability. The approach taken in this work aims
This is an open access article distributed under the Creative Commons at maximizing flexibility by providing a framework that is
Attribution License, which permits unrestricted use, distribution, and both general enough to accommodate a wide range of proto-
reproduction in any medium, provided the original work is properly cited. cols, yet efficient enough to ensure competitive performance.
Multimode Protocols for Reconfigurable Radios 391
(Classic) single Reconfigurable

protocol stack protocol stack
Same
proprety System-specific Reconfiguration
protocol stack management/functions
Specific functionality Common functionality
Specific parts Generic
protocol stack
RAT 1 RAT x
Different RAT 2 Protocol Protocol
functions management
Protocol Protocol
architecture framework
Data
structures
Figure 1: UML diagram of the generic protocol stack in the context of protocol reconfigurability.
Similar goals have been formulated and followed in the soft- ity and maintainability, code/resource sharing, protocol de-
ware engineering domain. The x-Kernel [10] architecture velopment acceleration through reusability, and continuous
composes a protocol graph of protocol components together performance evaluation in the context of quality-of-service
into a system, but the approach does not permit dynamic re- (QoS) dimensioning.
configuration of the protocol graph and does not specifically The initial step towards a generic stack is a detailed,
target wireless communications. The DARPA active network- layer-by-layer analysis of communication protocols to iden-
ing program [11] tries to answer the key question of loca- tify their similarities. The realization of generic parts is cru-
tion (and nature) of programmability with the perspective cial for the success of the proposed concept in the face of a
to build a flexible distributed computing system, again with tradeoff between genericity, that is, general usability, and im-
a focus on fixed networks. plementation effort. As depicted in Figure 1, the generic pro-
This paper consolidates previous publications [12, 13] tocol stack comprises
and deduces summarizing conclusions. After introducing the
(i) fundamental protocol functions;
idea of a generic protocol stack in Section 2, its application
(ii) a protocol architecture;
for protocol reconfigurability in the context of a multimode
capable protocol architecture is outlined thereafter. The re- (iii) data structures;
alization of a generic protocol stack, based on fundamental (iv) a protocol framework;
protocol functions that can be parameterized, is summarized (v) protocol management.
in Section 3. The composition of specific protocol layers of
They form, together with RAT specific parts, a system-
adequately parameterized modules is shown in Section 4.
specific protocol stack. An efficient multimode capa-
Section 5 introduces the composition of system-specific lay-
ble reconfigurable stack is realized in adding cross-stack
ers at the example of the UMTS radio link control (RLC) layer,
management-related functions. The cross-stack manage-
the transmission control protocol (TCP), and the IEEE 802.11
ment of the generic protocol stack for enabling protocol re-
medium access control (MAC) layer, which differ in their
configurability is introduced in detail in Section 3.
development history as well as in their layer classification
corresponding to the ISO/OSI reference model. A segmen- 2.1. Two complementing approaches for realization
tation/reassembly module and an automatic repeat request
There are in general two possibilities from the software en-
(ARQ) protocol module are validated and evaluated analyt-
gineering perspective for approaching the generic protocol
ically and through simulations in Section 4. These modules
stack: (1) parameterizable functional modules and/or (2) in-
are used with different parameterization for composing the
heritance, depending on the abstraction level of the identified
above-mentioned three specific protocol layers.
protocol commonalities. As introduced above, this paper fo-
cuses more on the modular approach while the inheritance-
2. THE GENERIC PROTOCOL STACK based approach is considered in [14, 15]. Additionally, [16]
The rationale for approaching a generic protocol stack is that takes up the idea of a generic protocol stack in focusing on a
all communication protocols have common functions. These generic link layer for the cooperation of different access net-
commonalities can be exploited to build an efficient mul- works at the level of the DLL. However, the link layer proto-
timode capable wireless system. The aim is to gather these cols are not the only protocols that have to be considered in a
common parts in a single generic stack and specialize this multimode capable network but the complete protocol stack.
generic part. Thereby, the particular requirements of the tar- This implies, for instance, higher layer functions as the con-
geted RAT are considered, as depicted in Figure 1. The tar- trol and management of the radio resources as well as mobil-
geted advantages of this concept are runtime reconfigurability.
Applications 3.1. Protocol reconfiguration manager

Data The protocol reconfiguration manager has thereby the fol-
lowing tasks:
TCP/IP
Protocol Based on (i) management of the (permanently or temporally) par-
Full reconfi- RLC, LLC generic allel existing protocols and protocol stacks;
reconfigurability guration protocol
manager MAC (ii) creation, destruction and/or reconfiguration of a sin-
stack
PHY
gle protocol or complete protocol stack;
(iii) administration of the user data flow during the recon-
Data figuration process, as for instance the redirection of
Channel (modem) the user data from the old to the reconfigured protocol
stack;
Figure 2: A reconfigurable protocol stack based on generic func- (iv) cross-layer optimization through protocol conver-
tional modules in the context of a completely reconfigurable termi- gence, as introduced in the next section. This implies
nal. for example the transfer of protocol or user data from
the old stack to the new one;
(v) support and enabling of reconfiguration functions of
2.2. Identifying commonalities the network, as for instance the support of a network-
The evolution of the digital cellular mobile radio networks initiated reconfiguration or an update of the network
originated in the global system for mobile communication information about the status of the terminal.
(GSM) toward systems of the third generation, as for exam- The reconfiguration of the protocol stack, administrated
ple UMTS, has shown that in their standardization, devel- through the protocol reconfiguration manager, has two char-
opers have fallen back on well-proven functions and mecha- acteristics: (1) the creation of a new stack/layer consisting of
nisms which are adopted to the specific requirements of their adequate parameterized modules of the generic stack and de-
application. The approach towards an efficient multimode struction of the existing one and (2) the reconfiguration of
protocol stack, introduced in this paper, is based on these the existing protocol stack in exchanging the parameteriza-
protocol commonalities. tion of the corresponding modules.
As the architecture of modern communication protocols
cannot be forced into the classical layered architecture of the 3.2. Protocol convergence in future wireless networks
ISO/OSI reference model, it is rather difficult to identify sim-
The generic protocol stack enables the protocol convergence
ilarities and attribute these to specific layers. Therefore, this
in future wireless networks. The convergence of such multi-
paper deepens the level of examination and considers fun-
mode protocol stacks has two dimensions: on the one hand
damental protocol functions as introduced above as one ba-
the convergence between two adjacent layers, in the following
sis for a generic protocol stack, contrary to [14, 15] where
referred to as vertical convergence, and on the other hand, the
complete protocols are analyzed for genericity. The identified
convergence between layers located in the different modes of
protocol functions correspond mainly to the DLL as speci-
the protocol stack which have the same functions, in the fol-
fied in the ISO/OSI reference model. Nevertheless, they can
lowing referred to as horizontal convergence. The generic pro-
be found in multiple layers of today’s protocol stacks. The
tocol stack, as introduced above, enables both the horizontal
functions of segmentation/reassembly or an ARQ protocol
as well as the vertical protocol convergence.
used for error correction are an example for this. They are
Figure 3 depicts the transition between two protocol
located in the RLC as well as in the transport layer, namely in
stacks of different-air interface modes of a wireless network.
the TCP.
The different PHY options of IEEE 802.16 could serve as an
example, see [3]. Presently, these PHY options are not en-
3. ENABLING RECONFIGURABILITY visaged to coexist in terminal or access equipment, although
they share a common MAC protocol with very little option-
The generic protocol stack, with its pool of generic functions specific extensions. The protocol stack is separated into the
as introduced above, enables an efficient as well as flexible user and control plane (u- and c-plane) on the one hand
realization of reconfigurable protocol stack. Full, end-to-end and the management plane (m-plane) on the other hand.
reconfigurability from the modem part up to the applications The split between common and specific parts of the pro-
requires a layer overlapping management and additional re- tocol is here exemplary depicted for the MAC layer, as ex-
configuration functions. Therefore, as this paper considers plained above. This split may be also necessary in the PHY,
communication protocols implemented in software, a proto- RLC, or higher layers, depending on the targeted protocol
col reconfiguration manager is introduced in Figure 2, which architecture and functional flexibility of the common part.
accomplishes all reconfigurability-related tasks of the PHY, A cross-stack management logically connects the protocol
MAC, RLC/DLL, and transport layer. These system-specific stacks of the different modes on the m-plane.
layers are based on the pool of generic protocol functions and A seamless interworking and optimized transition be-
mechanisms. tween mode 1 and mode 2 have certain requirements to the
Protocol status
information
m-plane Transition Cross-stack management including
u-plane + c-plane modes convergence protocol
RLC RLC
Common Common
part MAC MAC part
MAC 1 Specific MAC 2 Specific

part part
PHY 1 PHY 2
Protocol stack Protocol stack
mode 1 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 MAC layer. This reflects, for instance, the case of IEEE 802.16 [3].
Service access point SAP

Toolbox of functional
Interface
modules Multiplexer PDU factory
Functional Multiplexer Segmentation
Interface module unit
Manager Interface
ARQ unit
Control Functional
Manager
module SAP
Functional
Figure 5: Toolbox of functional modules as part of the generic pro-
module
Layer or Data tocol stack.
sublayer Interface
Service access point
can be repeatedly used with a different set of parameters cor-
responding to the specific communication system. The mod-
ules of generic protocol functions form together with system-
System specific
specific modules a complete protocol layer, as depicted in
Generic Figure 4. The communication inside the layer is performed
with the help of generic data structures, that is, generic ser-
Figure 4: Composition of a protocol specific layer or sublayer on vice primitives and generic protocol data units (PDUs), which
the basis of generic and system-specific functional modules. 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.
cross-stack management. The protocol data, as for instance
A unique manager as well as interfaces for the service ac-
the protocol status information of the existing connections,
cess points (SAPs) to the adjacent layers complete the fully
has to be transferred between the two modes. This horizon-
functional protocol layer as depicted in Figure 4. In detail,
tal convergence is performed by a mode convergence proto-
the mentioned components have the following tasks.
col. Therefore, protocol functions of the c-plane, common
to the different modes, are necessary to access u-plane status (i) Functional module (generic or RAT specific): realizes a
information. These common functions rely on a well-defined certain fundamental functionality as black box. In case
interface towards the mode-specific part of the layer. For fur- of a generic module, a list of parameters for charac-
ther details on the proposed protocol architecture and corre- terizing the functionality is given and the underlying
sponding infrastructure, see [17]. functionality is hidden. The comprehensiveness of the
fulfilled function is limited to fit straightforward into
a single module.
4. MODULAR APPROACH—THE GENERIC PROTOCOL
(ii) Manager: composes and administrates the layer dur-
STACK AS TOOLBOX OF PROTOCOL FUNCTIONS
ing runtime. This implies the composition, rearrange-
Again, the generic protocol stack is the realization of the ment, parameterization, and data questioning of the
common parts, as illustrated in Figure 1, and implements functional modules. Additionally, the manager admin-
its common functions on the basis of modules. These com- istrates the layer’s internal communication, as for in-
mon protocol functions get their system-specific behavior stance the connection of the layer’s modules through
based on parameterization. Once specified, these modules generic service primitives. It is the layer’s counterpart
To RRC/ PDCP/ BMC (radio bearers) SAP
Interface
SAP
Acknow-
Interface ledged
Multiplexer
mode
To RRC Transpa- Unacknow-
rent mode ledged Segmentation unit
mode
Manager
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 reconfiguration manager as introduced (iv) discarding of several-times received segments∗ ;
above in Figure 2 and it realizes the reconfigurability (v) reordering of PDUs∗ ;
of the layer. (vi) multiplexing/demultiplexing of the data flow, as for in-
(iii) Interface: translates the generic service primitives with stance the mapping of different channels∗ ;
specific protocol information as payload to system- (vii) dynamic scheduling;
specific ones and enables thus the vertical as well as (viii) ciphering;
horizontal integration of the system-specific parts of (ix) header compression.
the layer.
(iv) Service access point (SAP): here, services of the layer are The asterisk ∗ marked functions are considered in
performed for the adjacent layers. The layer may com- this paper while the other functions are target of the au-
municate via generic primitives without a translation thors’ongoing work.
interface to an adjacent layer if the said layer has the
same modular composition. The interface is needed if 4.2. Parameterization of functional modules
it is demanded that the layer appears as a classic layer
fitting into an ordinary protocol stack. Parameterization implies not only specific values, as for in-
stance the datagram size of a segmentation module, but also
(v) PDU factory (as functional module, later depicted in
a configuration of behavior and characteristics of a module,
Figures 6, 7, and 8): composes layer-specific protocol
as for example the concretion of an ARQ module as a Go-
frames and places them as payload in generic PDUs.
back-N ARQ protocol with specified window sizes for trans-
This approach enables the simulation and performance mission and reception. This implies as well a configuration of
evaluation on several levels. A single (sub-) layer as well as a the modules’ interface to the outside. Thus, the parameteri-
complete protocol stack can be composed out of the intro- zation of functional modules may mean (1) a specification of
duced modules. To facilitate understanding, the parameteri- certain variables, (2) the switching on/off of certain function-
zation itself is introduced later. ality/behavior, and (3) an extension of the module’s interface
to the outside.
4.1. Generic protocol functions of the data link layer Taking the example of the segmentation/reassembly
Taking Section 2.2 into account, the following functions of module, the parameterization may imply among other
the DLL are considered as being part of the generic protocol things:
stack: (i) use of concatenation;
(i) error handling with the help of forward error correc- (ii) use of padding, that is, filling up of the PDU to reach a
tion (FEC) or ARQ protocols as for instance Send- certain size;
and-Wait ARQ, Go-back-N ARQ∗ , or Selective-Reject (iii) transmitter or/and receiver role;
ARQ; (iv) buffer size for SDUs concatenated in a single PDU;
(ii) flow control∗ ; (v) size of PDU after handling;
(iii) segmentation, concatenation, and padding of protocol (vi) behavior in case of error, that is, interworking with
data units (PDUs)∗ ; ARQ module.
To higher layer
Interface
Multiplexer
ARQ ARQ ARQ ARQ
Manager 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 by a Go-Back-N ARQ with a window length of 1. The per-
means the specification of formance of these three layers is evaluated in the subsequent
section.
(i) ARQ protocol characteristic, for instance Go-Back-N
ARQ or Selective-Reject ARQ;
(ii) transmitter or/and receiver role; 6. SIMULATIVE EVALUATION AND VALIDATION OF
(iii) receive and transmission window sizes; THE FUNCTIONAL MODULES
(iv) fixed, variable (TCP) window length or open/shut The parameterizable modules are implemented in the specifi-
mechanism (LLC); cation and description language (SDL), and evaluated with the
(v) timer value, after a packet is assumed to be lost; help of a modular object-oriented software and environment
(vi) connection service: inexistent (UMTS RLC), sepa- for protocol simulation (MOSEPS) that provides basic traf-
rated for each direction (802.11—CSMA/CA with fic generators, a model of an erroneous transmission chan-
RTS/CTS), 2-way handshake (GSM LLC), or 3-way nel and statistical evaluation methods. This section intro-
handshake (TCP); duces the modular approach to protocol functions in focus-
(vii) use of negative acknowledgments (NACKs). 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
5. COMPOSITION OF SYSTEM-SPECIFIC LAYERS protocol stack is shown. The modules are parameterized cor-
As introduced above, the link layer functions are not limited responding to a specific protocol layer and their simulative
in their appearance to the DLL. To illustrate the applicability behavior is compared to analytical models known from the
of the modular approach based on the toolbox of protocol literature.
functions of Figure 5, a composition of three exemplary pro-
tocol layers, all differently localized in a protocol stack cor- 6.1. UMTS radio link control layer
responding to the ISO/OSI reference model, is introduced in In general, segmentation is needed in all cases where higher
the following: (1) a UMTS RLC layer in Figure 6, (2) a TCP, layer PDUs, referred to as service data units (SDUs), need to
IP, and UDP layers in Figure 7, and (3) an IEEE 802.11 MAC be separated into multiple PDUs to be further handled by
layer in Figure 8. The consideration of Figure 7 is limited lower layers. This restriction concerning the size of a PDU
in the following to the TCP layer, marked through the gray may result for instance from reasonable limitations of a PDU
dash-dotted rectangle. The medium access of the distributed transmitted with error correction in an ARQ protocol. It may
coordination function (DCF) of 802.11 may be regarded as also be motivated through the capacity of a transport channel
a Send-and-Wait ARQ, simply realized in the ARQ module offered by the physical layer, using the notation of UMTS.
To LLC SAP
SAP Interface Mobility unit
Interface
Multiplexer
Mobility/data
To STA
MGMT
Segmentation unit
layer
Manager
ARQ unit
PDU factory
Interface
SAP
To PHY
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

1
segmentation can increase the channels efficiency in hav-
ing a multiplexing gain. Concatenation increases thereby the 0.9
channel utilization as it is outlined in Figure 9. The channel 0.8
Payload /channel capacity
utilization as quotient of user payload and available channel 0.7 4th channel
capacity is given through 3rd channel
0.6
0.5
payload lpayload 2nd channel
= (1) 0.4
channel capacity lpayload /packet size · packet size
0.3
0.2
in the case of no concatenation. The packet size of the
0.1
physical channel is assumed to be fixed and the packet 1st channel
length of the user data lpayload determines if the segmented 0
0 100 200 300 400 500
higher layer SDU(s) fit(s) into a single PDU. We assume
Payload packet size lpayload (byte)
that an additional physical channel is established if the user
data requires it. Thus, additional channel capacity is pro- Sim.: with concatenation
vided and the available capacity is increased. In case of no Sim.: without concatenation
Analyt.: with concatenation
concatenation, the fixed-sized packet is transmitted partly Analyt.: without concatenation
empty over the physical channel. Consequently, the chan-
nels’ overall utilization, that is, the effectively used capac-
Figure 9: Utilization of channels with a fixed packet segmentation
ity compared to the amount of provided capacity, is de- size of 128 bytes.
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- concatenation. The same stands for the usage of concatena-
mentation aspects of the UMTS RLC in the unacknowledged tion, as the observed channel utilization does not match the
mode (UM). The UM of the RLC has the responsibility to ideal one. The number of SDUs prepared for transmission in
concatenate SDUs to a PDU of a predefined length, here a single PDU is here limited so that for small lpayload values, a
128 bytes. The simulative results with and without concate- PDU is not completely filled. In summary, the parameteriz-
nation are given by the markers. The payload packet size, that able segmentation/reassembly module adequately reflects the
is, the user data, is increased up to 500 bytes and the channel expected behavior and can be validly used in a multimode
utilization as introduced above is evaluated. capable protocol stack.
In applying the segmentation in a communication proto-
col, the protocol overhead comes into play. Due to this over- 6.2. Transmission control protocol layer
head, more capacity than transmitted user data is required as As introduced above in Figure 7, a TCP layer can be com-
observable in Figure 9 for lpayload = 384 bytes in the case of no posed out of the functional modules of the DLL as being
1
0.9
1.2
0.8
1 0.7
Overhead to payload
Overhead to payload
0.8 0.6
0.5
0.6
0.4
0.4 0.3
0.2
0.2
0.1
0 0
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000
Payload data frame length (byte) Size of data per packet (byte)
Sim.: BER = 10−5 Sim.: w = 8

Sim.: BER = 10−6 Sim.: w = 64
Analyt.: BER = 10−5 Analyt.: w = 8
Analyt.: BER = 10−6 Analyt.: w = 64
(a) (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 the frame length of the payload data for (a) a BER of 10−5
mechanisms of the TCP layer’s ARQ module, we measure the and 10−6 on the one hand and (b) window length of 8 and
protocol overhead in dependency on the payload packet size 64 on the other hand. Figure 10a shows the expected per-
in the case of erroneous transmissions. The focus is thereby formance corresponding to the Go-Back-N ARQ. The over-
on the influence of two effects: the bit error ratio (BER) of head to payload ratio increases with increasing bit error ra-
the radio channel, that is, the wireless medium, and the size tio and an optimal frame length for the payload data to
of the send and receive window w. minimize the said ratio can be determined. From the cross-
With a packet length of lpacket = lheader + lpayload , where protocol optimization perspective, this frame length may be
TCP has fixed header length of lheader = 40 bytes, the packet used as a dimensioning rule for segmentation. The same
error ratio (PER) can be calculated to 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
PER = 1 − (1 − BER)lpacket ·8 . (2) corresponding to the Go-Back-N ARQ, increases. The differ-
ence between analysis and simulation for large payload data
Based hereon the overhead to payload quotient for the frames in Figures 10a and 10b is reasoned by the send/receive
Go-Back-N ARQ can be derived and approximated [12] to buffers of the implemented TCP layers. Data packets in these
buffers have to be discarded when an error is detected and are
overhead neglected in the approximation (3). In summary, the ARQ
payload module of the generic protocol stack fulfils adequately its in-
tended purpose.
lheader
=
lpayload 6.3. IEEE 802.11 medium access control layer
w/2
i=1 (w/2) − i + 1 · (1 − PER)i−1 · PER lpacket In this section, the modular composition of an IEEE 802.11
+ · ,
w/2 lpayload MAC layer as illustrated in Figure 8 is validated and evalu-
(3) ated. Therefore, the average throughput of the carrier sensing
multiple access with collision avoidance (CSMA/CA) -based
where w is the length of the transmission window leading decentralized medium access by the DCF with and without
to the analytical results as depicted in Figure 10. This figure request to send/clear to send (RTS/CTS) is analyzed and sim-
illustrates the overhead to payload ratio in dependency on ulated.
1 1
0.9 0.9
0.8 0.8
0.7 0.7
Throughput (Mbps)
Throughput (Mbps)
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100
Number of stations Number of stations
Sim.: without RTS/CTS Sim.: without RTS/CTS

Sim.: with RTS/CTS Sim.: with RTS/CTS
Analyt.: without RTS/CTS Analyt.: without RTS/CTS
Analyt.: with RTS/CTS Analyt.: with RTS/CTS
(a) (b)
Figure 11: 802.11 MAC layer evaluation of the DCF-based medium access under utilization of the ARQ module. The total system throughput
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 calculated to

the medium is empty (e), successfully (s) allocated, or a collision (c)
occurs [18, 19]. The fixed values result from the time length of an Ps Lpayload
tsaturation = , Taverage = Pe Te + Ps Ts + Pc Tc . (5)
RTS/CTS sequence. Taverage
Duration with Duration without
Probability We assume a channel rate of 1 Mbps. The slot length,
RTS/CTS RTS/CTS
short interframe space (SIFS) and distributed coordination
pe = (1 − τ)n Te = 1 Te = 1 function IFS (DIFS) are 1, 6, and 10 microseconds resulting
ps = nτ(1 − τ)n−1 Ts = 636 + 8lpayload Te = 636 + 8lpayload into the analytical as well as simulative results of Figure 11.
pc = 1 − pe − ps Te = 170 Ts = 234 + 8lpayload There, the overall system throughput, with and without
RTS/CTS, in dependency on the number of stations is de-
picted. For small packets, Lpayload = 128 bytes, Figure 11a,
the headers are the main cause for an inefficient use of the
The channel capacity is mainly wasted by two effects: medium. For larger frames, Lpayload = 4096 bytes, Figure 11b,
MAC header sending and collisions. One way to an analytical a collision wastes more time, as a transmitting station is only
approach for determination of the throughput is to calculate able to notice an interfered frame after its ending. Therefore,
the collision probability p and the access probability τ with the RTS/CTS mechanism is introduced, to have just a small
the help of a two-dimensional Markov chain for the mod- RTS frame lost in case of a collision. The simulation agrees
eling of the backoff window of the DCF [18, 19] resulting mainly with the analytic determination of the throughput
into of (5) and illustrates the superiority of the RTS/CTS-based
solution. As the ARQ module of the generic protocol stack
−1 reflects the expected behavior of RTS/CTS mechanism [19],
1 − (2p)8
p = 1 − (1 − τ)n−1 , τ = 2 · 1 + W0 + pW0 , this module can be legitimately used in an 802.11 MAC layer.
1 − 2p
(4)
7. CONCLUSION
where n is the number of stations and W0 the minimum The introduced concept of a generic protocol stack enables
backoff window size, here we chose W0 = 8. With the protocol software for future multimode capable systems
help of the average time slot length Taverage on the basis of under the consideration of protocol reconfiguration. The
Table 1, the average total system throughput tsaturation can be generic protocol stack, as a collection of modular protocol
functions, takes up the usual advance of software engineering [11] J. M. Smith and S. M. Nettles, “Active networking: one view of
in the field of protocol development and evaluation: it has the past, present, and future,” IEEE Trans. Syst., Man, Cybern.
fallen back on well-proven and known protocol functions C, vol. 34, no. 1, pp. 4–18, 2004.
and behavior from the portfolio of the engineers’ experience. [12] L. Berlemann, A. Cassaigne, and B. Walke, “Generic protocol
functions for design and simulative performance evaluation
A generic realization of these functions in the form of inde- of the link-layer for reconfigurable wireless systems,” in Proc.
pendent modules results in a toolbox of protocol functions as 7th International Symposium on Wireless Personal Multimedia
a construction kit for protocol development. Dimensioning Communications (WPMC ’04), pp. 5–5, Abano Terme, Italy,
rules for an adequate support of QoS in wireless communica- September 2004.
tion from the perspective of the protocols can be derived. In [13] L. Berlemann, A. Cassaigne, R. Pabst, and B. Walke, “Modu-
taking the tradeoff of genericity into account, these thought- lar link layer functions of a generic protocol stack for future
ful realized modules stimulate efficiency through reusabil- wireless networks,” in Proc. Software Defined Radio Technical
Conference (SDR ’04), pp. 6–6, Phoenix, Ariz, USA, November
ity and maintainability as well as accelerate the development
2004.
process itself. However, the consideration of common pro- [14] M. Siebert and B. Walke, “Design of generic and adaptive pro-
tocol functions and protocol convergence during the devel- tocol software (DGAPS),” in Proc. International Conference
opment of future protocols will increase by itself the grade on 3rd Generation Wireless and Beyond (3Gwireless ’01), San
of genericity and advantage of this approach. The efficiency Francisco, Calif, USA, June 2001.
of protocol reconfigurability benefits from the introduced [15] L. Berlemann, M. Siebert, and B. Walke, “Software defined
generic approach and implies a clearly identified effort of protocols based on generic protocol functions for wired and
protocol management. Thus, the introduced approach is a wireless networks,” in Proc. Software Defined Radio Technical
Conference (SDR ’03), Orlando, Fla, USA, November 2003.
first step to an end-to-end reconfigurable wireless system.
[16] J. Sachs, “A generic link layer for future generation wireless
networking,” in Proc. IEEE International Conference on Com-
ACKNOWLEDGMENTS munications (ICC ’03), Anchorage, Alaska, USA, May 2003.
[17] L. Berlemann, R. Pabst, M. Schinnenburg, and B. Walke, “A
The authors would like to thank the German Research Foun- reconfigurable multi-mode protocol reference model facilitat-
dation (DFG) for funding the work contributed to this paper ing modes convergence,” in Proc. 11th European Wireless Con-
in the form of a Research College (Graduiertenkolleg). This ference (EW ’05), Nicosia, Cyprus, April 2005.
work has been continously funded from third parties’ grants. [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-
REFERENCES many, October 1998.
[1] B. Walke, R. Pabst, and D. Schultz, “A mobile broadband sys- [19] A. Hettich, “Leistungsbewertung der Standards HIPERLAN/2
tem based on fixed wireless routers,” in Proc. International und IEEE 802.11 für drahtlose lokale Netze,” Ph.D. disser-
Conference on Communication Technology (ICCT ’03), vol. 2, tation, RWTH Aachen University, Aachen, Germany, 2001,
pp. 1310–1317, Beijing, China, April 2003. ABMT 23, http://www.comnets.rwth-aachen.de.
[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. 80–89, 2004. Lars Berlemann was born in Aachen, Ger-
[3] IEEE, “Standard for Local and metropolitan area networks, many, in 1977. After having absolved a stu-
Part 16: Air Interface for fixed Broadband Wireless Access dent internship at Infineon Technologies,
Systems—Amendment 2: Medium Access Control Modifica- San Jose, he received his Diploma degrees in
tions and Additional Physical Layer Specifications for 2-11 electrical engineering and economics from
GHz,” April 2003. Aachen University in 2002 and 2003, re-
[4] J. Mitola, “The software radio architecture,” IEEE Commun. spectively. Since 2002, he served as a Re-
Mag., vol. 33, no. 5, pp. 26–38, 1995. search Assistant at the Chair of Communi-
[5] W. Tuttlebee, Software Defined Radio: Enabling Technologies, cation Networks (with Professor B. Walke)
Wiley Series in Software Radio, John Wiley & Sons, New York, of RWTH Aachen University, where he is
NY, USA, 2002. working towards his Ph.D. degree. He is a scholarship holder of the
[6] J. Mitola, Cognitive Radio, Ph.D. thesis, KTH, Stockholm, Research College (Graduiertenkolleg) “Software for Mobile Com-
Sweden, 2000, pp. 45–90. munication Systems” of the German Research Foundation (DFG)
[7] P. Mähönen, “Cognitive trends in making: future of net- and received the “Siemens Mobile Academic Award” in 2003. He
works,” in Proc. IEEE 15th International Symposium on Per-
is responsible for the organization of the scientific program of Eu-
sonal, Indoor and Mobile Radio Communications (PIMRC ’04),
ropean Wireless Conference 2005 and PIMRC 2005. His research
Barcelona, Spain, September 2004, invited paper.
[8] ITU-T Recommendation X.200, “Information Technology— work has so far been taken up by various national and European
Open System Interconnection—Basic Reference Model: Union funded research projects, such as the Academic Network for
The Basic Model,” International Telecommunication Union Wireless Internet Research in Europe (ANWIRE), Wireless World
(ITU), Geneva, Switzerland, 1994. Initiative New Radio (WINNER), and CoCoNet. His research in-
[9] End-to-End Reconfigurability (E2R), IST-2003-507995 E2R, terests include cognitive radios, distributed QoS support in spec-
http://www.e2r.motlabs.com. trum sharing scenarios, spectrum etiquette, and flexible protocol
[10] N. C. Hutchinson and L. L. Peterson, “The X-Kernel: An ar- stacks. He is the author/coauthor of 18 reviewed scientific pub-
chitecture for implementing network protocols,” IEEE Trans. lications and he is a Student Member of the IEEE ComSoc and
Software Eng., vol. 17, no. 1, pp. 64–76, 1991. ITG/VDE.
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 scientific 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 specification of protocols,
fixed 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 fixed and mo-
bile communication networks. He has published more than 120 re-
viewed conference papers, 25 journal papers, and seven textbooks
on architecture, traffic 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.
c 2005 A. Brawerman and J. A. Copeland
Towards a Fraud-Prevention Framework for

Software Defined 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 reconfigurability of software defined 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 configuration data download, storage and installation, user’s privacy, and cloning. The objective of this paper
is to present a fraud-prevention framework for software defined radio mobile devices that enhances overall security through
the use of new pieces of hardware, modules, and protocols. The framework offers security monitoring against malicious attacks
and viruses, protects sensitive information, creates and protects an identity for the system, employs a secure protocol for radio
configuration download, and finally, establishes an anticloning scheme, which besides guaranteeing that no units can be cloned
over the air, also elevates the level of difficulty 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 defined radio mobile
devices.
1. INTRODUCTION the concerns are the R-CFG download, storage, and instal-
lation; user’s privacy, that is, protection of the user’s iden-
Software defined radio [1] allows multiple radio standards tity, location, and communication with other devices; and fi-
to operate on common radio frequency hardware, thereby nally, SDR-MD cloning, that is, illegally using services that
ensuring compatibility among legacy, current, and evolving are billed to someone else’s device.
wireless communication technologies. To address the SDR Forum concerns and greatly en-
A software defined radio mobile device (SDR-MD) is ca- hance the overall security of SDR-MDs, a fraud-prevention
pable of having its operation changed by dynamically load- framework is proposed. The proposed framework offers se-
ing radio reconfiguration data (R-CFG files) over the air. curity monitoring against malicious attacks and viruses that
With different R-CFGs, the device can operate using different may affect the configuration data, protects sensitive informa-
wireless communication technologies while having a single tion through the use of protected storage, creates and pro-
transceiver. A typical SDR-MD can manage communication tects an identity for the system, employs a secure protocol
via satellite, over different cellular technologies, VoIP (voice for R-CFG download, and finally, establishes an anticloning
over internet protocol), and operations over the internet. scheme which guarantees that no units can be cloned over
the air, and elevates the level of difficulty to clone units if the
One of the key issues in SDR wireless communication in-
attacker has physical access to the SDR-MD. Even if cloned
volves security. According to the SDR Forum [2], some of
units exist, the anticloning scheme is able to identify and
deny services to those units.
This is an open access article distributed under the Creative Commons Preliminary practical experiments using java 2 micro-
Attribution License, which permits unrestricted use, distribution, and edition (J2ME) [3] and proofs that analyze the correctness
reproduction in any medium, provided the original work is properly cited. of the fraud-prevention framework are also presented.
2. BACKGROUND to another card, thereby cloning the authentication informa-

tion of the user.
Research work has been done for each of the SDR concerns
Cloning the SIM card is a relevant flaw, however a much
previously described; however, no published work has devel-
more serious flaw was discovered. In [9] it is shown that the
oped a solution that encompasses more than one of the con-
cryptographic codes used for authentication are not strong
cerns at once. This section is divided according to the SDR
enough to resist attacks. To exploit this vulnerability, an in-
Forum concerns. For each subsection, some of the relevant
dividual would interact with the SIM card repeatedly to learn
related research is presented.
the secret key and would then be able to clone the phone
2.1. R-CFG download, storage, and installation without having to clone the SIM card. Although it was con-
sidered that the attacker had physical access to the SIM card,
In [4], the authors discuss a model for securing the R-CFG it was mentioned that over-the-air attacks are possible, mak-
download and installation that involves the use of secret de- ing cloning on GSM cellphones a more serious threat.
vice keys and signatures. All security operations take place The Universal Mobile Telecommunications System
within tamper-proof hardware that also contains the pro- (UMTS) [10] is an open air-interface standard for third-
grammable components of the transceiver. This approach generation wireless telecommunications. It provides higher
provides good security for the radio software that lies within data rates and fixes several security flaws encountered in the
the tamper-proof hardware, but leads to some drawbacks GSM standard. Despite several advantages that the UMTS
such as the use of nonstandard security methods, lack of a standard provides, it also stores vital information in the SIM
means for third-party vendors to provide R-CFGs, and, most card. Thus, like the GSM, someone might be able to copy the
important, lack of a means for securing radio software that authentication information from one SIM card to another.
resides outside the tamper-proof hardware. Another drawback concerns the KASUMI block cipher,
which is at the core of the integrity and confidentiality mech-
2.2. User’s privacy
anisms in the UMTS network. Hardware implementations
Some efforts, called privacy extension to Mobile IPv6, deal are required to use at most 10 000 gates and must achieve en-
with user’s privacy. The basic idea of these efforts is to re- cryption rates in the order of 2 Mbps (maximum data rate).
place the MAC address of a mobile device with a random Thus, a considerable effort must be performed in order to
one, called a temporal mobile identifier (TMI) [5] or pseu- implement a high-performance hardware component that
dorandom interface identifier (PII) [6]. carries out the operations of the KASUMI block cipher.
In those schemes, personal mobile location privacy con- As a final remark, UMTS devices are not capable of re-
trol relies on either the home administration, the foreign ad- configuring their radio parameters via software. Thus, dual
ministration, or both. Moreover, the home administration is mode or tri-mode expensive cell phones are necessary to
required to share some secrets with the foreign administra- guarantee backward compatibility with other standards.
tion to prevent eavesdroppers from having any knowledge Simpler schemes that only detect cloned units and do not
about the binding users temporal identifiers and real iden- try to prevent cloning have also been proposed. They can be
tifiers. These efforts cannot completely control mobile loca- found in [11, 12].
tion privacy by a mobile user since the administration can
associate any identifier (PII or TMI) with the corresponding
2.4. Trusted computing group
real ID of the mobile device.
The trusted computing group (TCG) [13] is an industry
2.3. SDR-MD cloning standards body comprising computer and device manufac-
The advanced mobile phone system (AMPS) [7] is the analog turers, software vendors, and others with an interest in en-
mobile phone system standard introduced in the Americas hancing the security of the computing environment across
during the early 1980s. Despite the fact that it was a great ad- multiple platforms and devices.
vance in its time, the AMPS presented several security flaws, The TCG claims that it will develop and promote open
and multiple copies of cloned mobile stations were created industry standard specifications for trusted computing hard-
with little difficulty. ware building blocks and software interfaces across multiple
The global system for mobile communication (GSM) [8] platforms, including personal computers (PCs), servers, per-
is a globally accepted standard for digital cellular communi- sonal digital assistants (PDAs), and digital phones.
cation. The GSM authentication framework relies on special So far the TCG has only presented specification for the
cryptographic codes to authenticate customers and bill them PC environment [14]. Some of the benefits include more se-
appropriately. A personalized smart card, called a SIM card, cure local data storage, a lower risk of identity theft, and the
stores a secret key that is used to authenticate the customer; deployment of more secure systems and solutions based on
knowledge of the key is sufficient to make calls billed to that open industry standards.
customer. Despite the fact that the TCG specification for the PC
The SIM card is easily removable so that the user can does point out and solve several security flaws, this specifi-
use other cell phones. The drawback is that someone who cation would not achieve a satisfactory performance if em-
has physical access to the SIM card can copy the information ployed by constrained SDR-MDs.
A Fraud-Prevention Framework for SDR Mobile Devices 403
To the outside world

Application layer
Environment Software core

SDR device manager discoverer
SDR framework
Secure/unsecured internet R-CFG/CFG
connection module security module
Unsecured LDDA
module
Other
applications
OS layer
Secure LDDA
module
R-CFG manager CFG Application
manager manager OS
R-CFG Storage
RAM
ROM CPU Embedded
Config. file Unencrypted
devices
Protected
Encrypted RAM
Hardware layer
Signalizes cloning
SDR
Tamper-protected hardware package
FPGA
Security hardware layer
Figure 1: The preliminary design of the fraud-prevention framework.
3. THE FRAUD-PREVENTION currently installed. It also informs the SDR-DM when a dif-
FRAMEWORK SPECIFICATION ferent R-CFG is needed. The CFG manager is responsible for
managing the configuration (CFG) file. The CFG file is pro-
The fraud-prevention framework is composed of new pieces vided by the wireless operator (WO) and is used to set the
of hardware, new modules, and new protocols. Figure 1 de- device’s phone number. Note that both the R-CFG and CFG
picts the preliminary design of the framework. The dashed files are stored in an encrypted storage. Standard encryption
squares are the main contributions of this work. algorithms such as RC5 [15] and RSA [16] can be used to
Note that the SDR device manager (SDR-DM) is respon- provide the encryption storage. Other modules as well as ba-
sible for managing all the communication with the outside sic definitions are discussed in separate subsections below.
world and for requesting the services of each module when
needed. Also, the environment discoverer module is respon- 3.1. Basic definitions
sible for detecting which wireless communication technolo- This section presents definitions, components, and entities
gies are available in the current SDR-MD’s environment. This that participate in the fraud-prevention framework. The
module is assumed to be present in the software core SDR nomenclature used to specify the framework is presented in
framework and is outside the scope of this work. Table 1.
The R-CFG manager is responsible for managing the The entities that participate in the framework as well as
R-CFG files currently stored in the device and the R-CFG their responsibilities are defined in Table 2.
Table 1: Basic definitions.

C A 48-bit random number (nonce)
KY {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
It is used to check integrity status of a certain component. It is defined as the function Att(X),
Attestation
which results in the hash of component X
It is used to obtain the attestation credential. Composed by the 2048-bit attestation private key
Attestation key pair (AK)
(AKpriv ) and public key (AKpub )
It is used to identify the SDR-MD. It is signed by the privacy credential authority (Privacy CA)
Attestation credential (AC)
and it is presented whenever the user tries to use the network services. AC = [AKpub ]Privacy CA
When the SDR-MD discovers it is a cloned unit, it sets its AC to null.
Null AC
Every bit in the AC is equal to 0
It is used to uniquely identify the SDR-MD. It is never disclosed by the device.
Endorsement key (EK)
Its size is also 2048 bits
R-CFG It is used to configure the radio of the SDR-MD
Valid R-CFG An R-CFG that has been approved by the regulatory agency
An R-CFG that has not been approved by the regulatory agency or it has been modified
Invalid R-CFG
after been approved by the regulatory agency
It is used to set up the phone number of the SDR-MD. It is signed by the WO.
CFG
CFG = [Phone no.]WO
Table 2: Entities and responsibilities.
Produces the SDR-MD. Generates the R-CFGs. Generates the SDR-MD’s EK and informs
Manufacturer (manuf.) 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
Tests, approves, and licenses the R-CFG. Basically, the RA tests the R-CFG in the specific hardware to ensure
Regulatory agency (RA)
that the device does not cause interference or function out of its defined spectrum, as defined in [17]
Sells the SDR-MD. Provides communication services. Generates the CFG
WO
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 files. Detects if it is a cloned or valid unit
3.2. The tamper-protected hardware package responsible for storing integrity values and protected non-
The TPHP must be physically protected from tampering. volatile memory to store the necessary keys. Notice that the
This includes physically binding it to the other physical parts TPHP comes from the manufacturer with the RA’s public key
of the SDR-MD such that it cannot be easily disassembled already stored.
and transferred to other devices. These mechanisms are in- The attestation engines are divided into the attesta-
tended to resist tampering. Tamper evidence measures are to tion measurement engine (AMEng), attestation store engine
be employed. Such measures enable detection of tampering (AS Eng), attestation report engine (AR Eng), and attestation
upon physical inspection. The package must limit pin prob- comparison engine (AC Eng). Table 3 presents the functions
ing and EMR scanning. Similar tamper-protected hardware of each attestation engine. Figure 2 depicts the components
is the trusted platform module of [13] and the Intel wireless of the TPHP as it comes from the manufacturer.
trusted platform processor [18].
The TPHP is composed of two tamper resistant chips 3.3. The secure SDR R-CFG download protocol
(TRCs): TRC1, which is read only, and TRC2, which is To install only valid R-CFGs, a secure SDR R-CFG download
read/write. The TRC1 contains the EK, the attestation en- protocol is defined as part of the fraud-prevention frame-
gines responsible for measuring, reporting, and comparing work. The secure protocol employs the mutual authentica-
integrity values, and a specialized hardware to generate 48-bit tion and R-CFG validation and verification steps described
random numbers. The TRC2 contains the attestation engine by the R-CFG/CFG security module.
Table 3: Attestation engines and functions. The steps of the secure SDR R-CFG download protocol
when using an unsecured connection, such as HTTP, are de-
Measures Att(EK), Att(R-CFG), and Att(CFG),
AM Eng picted in Figure 5. Dashed arrows indicate communication
and writes the results into R0, R1 and R2 inside the SDR-MD.
AS Eng
Stores the Att(EK) in register 0(R0), the Att(R-CFG) Although the protocol is specified using an unsecured
in register 1(R1), and the Att(CFG) in register 2(R2) connection, the R-CFG is still protected since it is encrypted
AR Eng Reads and reports the values of the registers with the EK, thus only that specific device which has initi-
Compares the values of R0, R1, and R2, reported by ated the connection can correctly decrypt and install the R-
AC Eng CFG. Details on how to obtain a lightweight secure connec-
the AR Eng, with the values measured by the AM Eng
tion using the Light SSL (LSSL) protocol, specified in the se-
cure/unsecured internet connection module, can be found in
Whenever a manufacturer generates a new R-CFG, it has [19].
to send the R-CFG to be approved and licensed by the RA. The SDR R-CFG download protocol initiates with the
This is called R-CFG validation. SDR-MD contacting the manufacturer’s server and estab-
To perform R-CFG validation, the protocol employs a lishing an unsecured connection. Next, the SDR-MD sends
public-private key mechanism. The manufacturer sends to MD(EK) and a nonce C encrypted by the EK. The manufac-
the RA a combination of a header, which contains manufacturer maintains a database of all available EKs (M EKDB),
turer, model, serial number range, and possibly some other indexed by MD(EK). The database has all information that
information; the new R-CFG; and the hardware in which the the manufacturer needs about each SDR-MD it has pro-
R-CFG is to be tested and used. duced.
The RA installs the R-CFG in the specified device and When the manufacturer receives the MD(EK), it searches
tests the device’s behavior. If no malfunction is observed, in its M EKDB for that value. If it does not find the MD(EK),
the RA approves the R-CFG and assigns it a license num- the manufacturer ends the connection. On the other hand, if
ber. During the test, the RA computes h = MD(headerR- MD(EK) is in M EKDB, then the manufacturer obtains the
CFG). The value h is then signed with the RA’s private key, EK of that device and generates a new nonce C . The C is
[h]RA . Figure 3 depicts the signing step. The signed hash then encrypted by the EK and sent, along with C, to the SDR-
value, [h]RA , is sent back to the manufacturer along with the MD.
assigned license number. Upon receiving C and C , the SDR-MD authenticates the
Once the R-CFG has been licensed, signed, and placed on manufacturer if the received C is equal to the one that the
a server, the SDR-MDs can contact the server at any time to SDR-MD has previously generated. If authentication fails,
download the combination of header, R-CFG, and [h]RA . the SDR-MD terminates the connection; otherwise, it ob-
After an SDR-MD has connected to the manufacturer’s tains C , sends it back to the manufacturer, and requests the
server, mutual authentication is performed. The mutual necessary R-CFG.
authentication step avoids masquerade and replay attacks. The manufacturer then authenticates the device. If au-
When using an unsecured connection, this is done by ex- thentication fails, the manufacturer terminates the connec-
changing random challenges (nonces) or by certificates, tion; otherwise, it sends the requested R-CFG encrypted by
while when using a secure connection, the protocol that pro- the EK. The SDR-MD receives the R-CFG, verifies it, and
vides the secure connection is assumed to take care of the checks the R-CFG data integrity. If the R-CFG tests show
mutual authentication. no negative results, the SDR-MD installs the R-CFG and ac-
After the mutual authentication step has been success- knowledges the manufacturer. The connection is then re-
fully completed, the SDR-MD requests and downloads the leased.
new R-CFG. Upon download completion, R-CFG verifica- After releasing the connection, the SDR-MD installs the
tion is necessary to guarantee that the R-CFG has been ap- R-CFG and stores the Att(R-CFG) value in R1. Whenever the
proved by the RA and properly signed. The verification step SDR-MD is booting up, the AM Eng calculates a new Att(R-
also tests whether the R-CFG is appropriate for the device CFG) value, which is then passed to the AC Eng to be com-
(Figure 4). pared with R1. If Att(R-CFG) = R1, the current radio config-
However, to guarantee that the R-CFG has not been mod- uration is trusted. On the other hand, if Att(R-CFG) = R1,
ified after being approved and signed by the RA, the follow- the SDR-DM blocks the use of any service.
ing steps are performed:
3.4. The anticloning scheme
(1) a new hash value h = MD(header R-CFG) is calcu-
lated; One of the more dangerous threats in SDR wireless commu-
(2) the received [h]RA is decrypted to obtain h; nication is cloning. SDR-MD cloning is considered a federal
(3) h and h are compared: if h = h , the received R-CFG crime. According to [20], telecommunication fraud losses
is accepted. However, if h
= h , the R-CFG is rejected. are estimated at more than a billion dollars yearly. A large
amount of this loss is due to cloning. Besides illegal billing,
Figure 4 also shows the data integrity check. If the new R- cloned units increase the competition of shared resources,
CFG has passed all the tests, it is then installed and the value which increases network congestion and degrades network
of Att(R-CFG) is stored in R1. services. Furthermore, the impact of overload traffic from
AS Eng
EK AM Eng Protected
0-cloned R0 storage
Att(EK)
1-valid
AC Eng R1 Att(R-CFG)
Random R2
no. AR Eng
generator RA pubkey
TRC1-read only TRC2-read/write
Figure 2: The tamper-protected hardware package in an invalid state.
Header 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
Header
the necessary steps to block the use of the network services.
Another advantage is that the anticloning framework is in-
R-CFG dependent of technology, working well for different wireless
H h technologies.
R-CFG
3.4.1. Entering a valid state

[h]RA 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
RA’s private key
SDR-MD enters a temporary state, that is, it is able to prove
Figure 3: R-CFG validation. its identity, however, it does not have a phone number yet, it
does not have the CFG file installed. After obtaining the CFG,
Header is checked the SDR-MD finally reaches a valid state. It is able to identify
Header 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
H h
R-CFG anytime after the SDR-MD has reached the valid state, it may
need a new R-CFG file or a new CFG file. While obtaining any
of those files, the SDR-MD goes to a temporary state. With
the new data locally stored, the security checks are executed
[h]RA h = 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
RA’s public key
the Privacy CA and it is executed only one time per each EK.
R-CFG verification Data integrity check Whenever the manufacturer generates a new EK, it in-
forms the Privacy CA, in a safe way, about that EK. The
Figure 4: R-CGF verification and data integrity check. 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-
cloned units is unpredictable. Thus, the estimation of traf- MD to its AC.
fic patterns is imprecise for network planning. The ACP steps are defined as follows. First, the SDR-
The anticloning scheme, which is part of the proposed MD contacts the Privacy CA and sends the value R0 =
fraud-prevention framework, is designed to provide a core Att(EK). The Privacy CA looks for a matching MD(EK) in
set of hardware and software technologies that provide the the CA EKDB. If it finds a match, the Privacy CA obtains the
basis for a wireless network environment free of cloned units. EK of that unit and acknowledges the unit. If no equivalent
Unlike other cloning detection schemes, the proposed MD(EK) is found, either the manufacturer failed to inform
anticloning scheme not only detects cloned units, but also the Privacy CA about this unit or this is an invalid EK. Thus,
elevates the level of difficulty to clone a valid unit. Also, as the Privacy CA does not provide an AC to the unit.
Manufacturer’s server SDR device
Verifies C
Generates C
MD(EK) in M EKDB ?
Gets EK
Obtains C
Establishes unsecure
MD(EK), KEK {C}
C , req. R-CFG type X
KEK {R-CFG}
connection
KEK {C C }
End
ACK
R-CFG data int.
MD(EK), KEK {C}
Obtains R-CFG
SDR-DM
Installs R-CFG
KEK {R-CFG}
R-CFG verif.
Generates C
KEK {C C }
SDR-DM
Obtains C
Obtains C
Verifies C
ACK
C
SDR-TPHP
Figure 5: The secure SDR R-CFG download protocol.
Gets CFG Upon receiving the AC, the WO verifies if the AC is null.
Invalid Gets AC Temp If the comparison is positive, the unit is a clone and the WO
Valid
state state state terminates the connection. Otherwise, the CUP continues its
Needs CFG normal flow.
Gets Needs
The WO uses the Privacy CA’s public key and decrypts
R-CFG R-CFC the AC, obtaining the AKpub . The WO has a database (DB),
indexed by the AKpub , that contains information about each
Temp SDR-MD in a valid state, such as phone number and user
state name. Next, the WO looks for a matching AKpub in the DB.
If it finds a match, it verifies MD(CFG) = R2. If the compar-
ison is negative, this is an invalid unit; either this is a cloned
Figure 6: Transition states of an SDR-MD. unit or a masquerade attack is occurring, and countermea-
sures are taken.
On the other hand, if the comparison is positive, this is a
Second, the Privacy CA generates an AK pair and the unit valid unit. The WO then obtains C and generates a nonce C
authenticates the Privacy CA. The unit generates a nonce C to authenticate the unit. C is concatenated with C and sent
and sends it to the Privacy CA encrypted by the EK. The Pri- encrypted by the AKpub to the SDR-MD. If the AKpub is not
vacy CA obtains C and sends it back along with an encrypted in the DB, this is a unit in the temporary state.
message containing the AK pair. Upon receiving the message, Upon receiving KAKpub {C C } from the WO, the unit au-
the unit verifies C, authenticating the Privacy CA. thenticates the WO if the received C is equal to the one pre-
Third, after authenticating the Privacy CA, the unit obviously generated. If authentication fails, the SDR-MD ter-
tains the AK pair and acknowledges the Privacy CA. The Pri- minates the connection. Otherwise, it sends C back to the
vacy CA then generates the AC = [AKpub ]Privacy CA and sends WO.
it, encrypted by the AKpub to the unit. The unit receives the Next, the WO authenticates the unit by verifying C .
AC, decrypts it, and stores it in its TPHP. After that, the con- If authentication fails, the WO terminates the connection.
nection is finally released. Otherwise, the WO generates a new CFG and stores the
After obtaining the AC, the final step to enter the valid MD(CFG) value in the DB. The unit receives the CFG en-
state is to have the SDR-MD executing the CFG update pro- crypted by its AKpub and decrypts it. The unit then stores the
tocol (CUP) to obtain a valid CFG. This protocol is executed CFG in the protected storage of TRC2 and installs the new
whenever the unit needs a new phone number. Figure 8 de- phone number.
picts the CUP step by step. Next, the AM Eng measures Att(CFG) and writes the
After connecting to the WO’s server, the unit sends its value in R2. The unit then sends this value encrypted by the
AC and the value of R2 = Att(CFG) along with a nonce C WO’s public key to the WO. The WO verifies the value and
encrypted by the WO’s public key. The WO’s public key is acknowledges the unit if the comparison is positive. Other-
obtained a priori through a secure protocol. If this is a new wise, it informs the unit that an error occurred during the
unit, the value of R2 is null. CFG installation step. This step is repeated in the case of
Privacy CA
Obtains EK
Generates AC
Generates AK pair
Att(EK) in EKDB ?
Obtains C
KEK {C (AK pair)}
Connection
Connection
established
released
Att(EK)
KEK {C}
ACK
ACK
ACK
AC
KEK {C (AK pair)}
SDR-DM
Authenticates CA
Obtains AK pair
R = Att(EK)
Request R0
KEK {C}
Generates C
SDR-MD
Obtains C
ACK
ACK
Stores AC
ACK
AC
SDR-TPHP
Figure 7: Attestation credential protocol.
WO
Stores MD(CFG)
Generates CFG
Verifies C
Obtains AKpub from AC
Verifies Att(CFG)
Att(CFG) = MD(CFG)
Generates C
Obtains Att(CFG)
Obtains C
KAKpub {C C }
KAKpub {CFG}
AC, {R2C }WO
Req. new CFG
Connection
Connection
established
{R2}WO
released
Verifies
ACK
C
KAKpub {CFG}
SDR-DM
{C C }
AC, {R2C }WO
R2 = Att{CFG}
Generates C
Req. AC, R2
SDR-MD
Obtains CFG
ACK
Installs CFG
C
Obtains C
Stores CFG
KAKpub
Obtains C
Verifies C
SDR-TPHP
Figure 8: The CFG update protocol.
errors. After receiving an acknowledgment, the unit releases by the AM Eng and sent to the AC Eng, which also receives
the connection. the current value of R0 and R2 from the AR Eng. The AC
After obtaining the AC from the Privacy CA and the CFG Eng compares the values and signalizes 1 for a valid unit, if
file from the WO, the SDR-MD finally reaches a valid state. Att(EK) = R0 and Att(CFG) = R2, or 0 for a cloned unit, if
Therefore, the unit is ready to use all the services offered Att(EK) = R0 or Att(CFG) = R2. In this fashion the SDR-
by the WO. Figure 9 depicts the tamper-protected hardware MD is aware of cloning. Figure 10 illustrates the procedure.
package when the SDR-MD is in the valid state. If the SDR-MD is a valid unit, the AC is sent and the WO
Note that the clone signal, sent by the AC Eng, propagates cloning-aware procedure begins.
outside the TPHP to the CPU and inside the TPHP to the In the WO side, the procedure works basically as an au-
TRC2, where it sets the AC to null if the SDR-MD is a clone thentication module. The WO obtains the AC and verifies if
unit. 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-
3.4.2. Cloning-aware procedure tains the AKpub 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 cloning-aware procedure is implemented in both sides, the WO prepares to authenticate the unit. If the unit is cor-
the WO and the SDR-MD, and is responsible for detecting rectly authenticated, the WO allows the use of the service.
whether the SDR-MD is a valid unit or a cloned unit. On the other hand, if the unit is not authenticated, the WO
After the unit has connected to the WO and requested concludes that this unit is trying to use other unit’s AC (mas-
a service, the cloning-aware procedure starts in the SDR- querade attack) and denies the service. Figure 11 illustrates
MD side. New Att(EK) and Att(CFG) values are measured the procedure.
AS Eng
EK AM Eng AKpriv AC
0-cloned R0 Att(EK)
1-valid AKpub
AC Eng R1 Att(R-CFG)
And
AK pair
Random R2 Att(CFG)
no. AR Eng
generator WOpub
AC CFG
TRC1-read only RA pubkey
TRC2-read/write
Figure 9: The tamper-protected hardware package in a valid state.
WO SDR-DM AM Eng AC Eng AR Eng Wireless link

PDA (11 Mbps) Manufacturer
Connection client server with SSL
established End-to-end
security
Starts proc. Figure 12: The experiment set-up.

Req. service Att(EK) Reads R0
Att(CFG) Reads R2
Figure 12. An SDR-MD, in this case a Sharp Zaurus PDA SL-
Att(EK), R0, R2 5600 with CPU speed of 400 MHz, 32 MB SDRAM, Linux
Att(CFG) OS, and J2ME support, connects through an 11 Mbps wire-
Compares less link to a Pentium 4 2.6 GHz server with 256 MB RAM.
Sends signal
4.1. The secure SDR R-CFG download protocol
Ends proc.
Two preliminary experiments involving the secure SDR R-
AC CFG download protocol and the secure R-CFG/CFG mod-
ule are described. In the first experiment, the time the R-
WO proc.
CFG/CFG security module takes to identify invalid R-CFGs
and delete them is measured. The second experiment com-
Figure 10: Cloning-aware procedure: SDR-MD side. pares the secure protocol execution when using an unsecured
connection : HTTP, a lightweight secure connection, LSSL
WO [19], and the SSL protocol [21].
Obtains AKpub from AC
Checks AKpub in DB
The graph in Figure 13 shows the results of the first ex-

Verifies C
Generates C
periment. The MD5 algorithm is used to calculate the finger-

SDR-MD is able to use services
KAKpriv {C}
Connection
established
print and to perform the data integrity check. As expected,

C
the larger the R-CFG is, the longer it takes to perform the
AC
security checks.
Figure 14 depicts the results of the second experiment.
Note that the secure protocol with unsecured connection
SDR-DM presents best performance, since it does not need to spend
KAKpriv {C}
SDR-MD
Obtains C
time with the cipher suite handshake and other extra steps
Req. AC
AC
needed by secure connections. In case secure connections are

necessary, the use of the LSSL is suggested since it presents
SDR-TPHP better performance than the SSL, as can be noticed in this
experiment.
Figure 11: Cloning-aware procedure: WO side.
4.2. Anticloning scheme
It is expected that the anticlone scheme will not add any
4. PRELIMINARY EXPERIMENTS
further delay on the obtainment of network services when
The experiments were executed using J2ME, which is a comparing with the GSM and UMTS techniques. Although
lightweight java version, specifically designed to be used with SDR mobile devices are constrained by nature, encryption
constrained devices. The experiments set-up is depicted in and decryption operations are only executed for small pieces
1100 The proofs continue with two final theorems. The first
theorem proves that there is no possibility to clone an SDR-
1000
MD over the air. The second theorem guarantees that only a
valid SDR-MD can use the network services.
Time (ms)
850
800 Lemma 1. The Privacy CA only attests the identity of SDR-
750 MDs that have valid EKs.
700
650
600 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
128 256 512 1M
is, hash of an EK that is not generated by the manufacturer,
R-CFG (KB) has the AC denied.
Figure 13: Time to identify invalid R-CFGs. A replay attack is not possible since the ACP is executed
only once per each EK. Impersonation of the SDR-MD, that
×102 is, masquerade attack, is noticed by the authentication step.
190
175
160
145
Lemma 2. No SDR-MD obtains a CFG file unless its identity
130 is successfully proved.
Time (ms)
115
100 Proof. According to the CUP definition, only after being au-
85 thenticated by the WO, the SDR-MD is given a new CFG.
70 This eliminates the possibility of masquerade attacks and re-
55
40
play attacks.
25 Only after responding correctly to the challenge gener-
10 ated by the WO, the SDR-MD is given a new CFG. Therefore,
128 256 512 1M no SDR-MD obtains a new CFG file unless it has proved its
R-CFG (KB) identity.
Secure protocol with HTTP
Secure protocol with LSSL Lemma 3. Only valid CFG files are installed in each SDR-MD.
Secure protocol with SSL
Figure 14: Comparing the secure protocol varying the connection Proof. To install a new CFG, the SDR-MD must execute the
type. CUP. According to the CUP definition, before receiving a
new CFG the SDR-MD authenticates the WO by verifying
{R2}WO = [MD(CFG)]. If the comparison is positive, then
of information such as the 2048-bit EK and AK pair, and the SDR-MD authenticates the WO. Thus, masquerade and
the 48-bit nonce C. Furthermore, the attestation engines and replay attacks are eliminated.
the random number generator in the TPHP are specialized After authentication, the SDR-MD receives a new CFG =
pieces of hardware that can quickly execute data integrity [Phone no.]WO . Since masquerade and replay attacks fail,
measurements and generate a 48-bit random number. only the WO could have sent this message, and the final step
to validate the CFG occurs. The SDR-MD verifies the WO’s
5. CORRECTNESS PROOFS signature in the CFG. When the signature is successfully ver-
ified, the CFG is considered valid and the TPHP stores and
This section presents a list of possible attacks involving the installs the new CFG.
R-CFG files and how the secure SDR R-CFG protocol avoids
those attacks. It then continues with correctness proofs that Theorem 1. It is guaranteed that there is no possibility to clone
show that the fraud-prevention framework provides an envi- an SDR-MD over the air.
ronment free of cloned units.
Table 4 illustrates common methods of attacks that fail Proof. In order to clone an SDR-MD over the air, one attacker
against the proposed protocol. must obtain the EK of the victim or a combination of valid
Next, the correctness proofs are presented. It begins with AK pair, valid AC, and valid CFG.
three lemmas. The first lemma shows that only an SDR-MD Since the EK and AKprivate are never disclosed by the
with a valid EK is provided an AC. The second lemma shows TPHP, the attacker has no possibility to obtain the EK nor
that an SDR-MD only obtains a new CFG when its identity the AK pair of a victim. According to Lemma 2, the attacker
is successfully proved. Finally, the third lemma shows that must prove its identity to obtain a valid CFG, thus if the at-
only valid CFGs, that is, CFGs that have been generated and tacker uses an AC that is not his/hers, the WO will notice it
signed by the WO, can be installed by an SDR-MD. and deny a new valid CFG.
Table 4: Possible attacks and how the secure protocol avoids them.
Attacks Description Protection

Clients using unauthorized services
Access control Protocol employs client authentication
or trying to download data they should not
An entity pretends to be the manufacturer server
Masquerade Protocol uses mutual authentication
or a client
By establishing secure connections or
Confidentiality R-CFG might be confidential encrypting, the R-CFG proprietary information
are kept secret
Replay Messages are captured and retransmitted later Mutual authentication avoids replay attacks
Every R-CFG is digitally signed
Invalid R-CFGs Installing R-CFGs that are not approved by the RA
by the RA and verified by the SDR-MD
Protocol employs one-way hash functions
R-CFG Integrity R-CFG modified after it has been approved
to guarantee data integrity
With no other way to clone an SDR-MD over the air, the damage the TPHP. Even if an attacker successfully disassem-
only way to bill someone else’s account is to capture his/her bles the TPHP without damaging it, the equipment to read
AC when transmitted over the air. However, the WO cloning- and copy the TPHP is so expensive that the attacker would
aware procedure will detect that the captured AC does not practically have no gain, if any, in doing so.
belong to that unit and it will deny any service.
Theorem 2. It is guaranteed that only a valid SDR-MD can 6. CONCLUSION
use the wireless operator services.
To greatly enhance the overall security of SDR-MDs, a fraud-
Proof. According to the WO cloning-aware procedure, in or- prevention framework is proposed. The fraud-prevention
der to use the network services the SDR-MD must present framework is composed of new pieces of hardware, mod-
a valid AC. By Lemma 1, only SDR-MDs with valid EKs are ules, and protocols. The framework offers security monitor-
able to obtain a valid AC. Therefore, unit with an invalid EK ing against malicious attacks and viruses, protects sensitive
does not have a valid AC and cannot use the WO’s services. information, creates and protects an identity for the system,
According to Theorem 1, there is no way to clone an employs a secure protocol for radio configuration download,
SDR-MD over the air, and impersonation of other SDR-MDs and finally, establishes an anticloning scheme which guaran-
by capturing their AC is noticed by the WO cloning-aware tees that no units can be cloned over the air, and elevates the
procedure. Thus, the only other way to clone an SDR-MD is level of difficulty to clone units if the attacker has physical
to have physical access to its TPHP. access to the mobile device. Even if cloned units exist, the an-
However, if an attacker successfully disassembles the ticloning scheme is able to identify and deny services to those
TPHP without damaging it and is able to copy the TPHP units.
to another SDR-MD’s TPHP, Lemma 3 and the SDR-MD Preliminary experiments show that the framework is
cloning-aware procedure guarantee that the SDR-MD that able to identify invalid R-CFGs with minimal delay. Proofs
received the cloned TPHP denies the use of the network ser- that analyze correctness of the framework show that the
vices. The value of R2 on the cloned TPHP and the value of fraud-prevention framework provides an environment free
the current MD(CFG) in the device are different. Thus, the of cloned units.
SDR-MD blocks the use of any services. Future work includes the execution of several ex-
Since the SDR-MD cloning-aware procedure blocks the periments that will measure performance of the fraud-
use of any service by cloned units and the WO cloning-aware prevention framework, and comparisons with other state-of-
procedure notices masquerade attacks, it is guaranteed that the-art related works.
only a valid SDR-MD can use the wireless operator services.
REFERENCES
In summary, the fraud-prevention framework elevates
the level of difficulty to clone an SDR-MD. The only way to [1] B. Bing and N. Jayant, “A cellphone for all standards,” IEEE
Spectr., vol. 39, no. 5, pp. 34–39, 2002.
clone one SDR-MD that employs the framework would be [2] Software Defined Radio Forum website, http://www.
disassembling the TPHP from the SDR-MD and reading its sdrforum.org.
contents. Since the TPHP is physically bound to other parts [3] Java 2 Micro Edition Technology website, http://wireless.
of the SDR-MD, attempts to disassemble it would probably java.sun.com/j2me.
[4] L. B. Michael, M. J. Mihaljevic, S. Haruyama, and R. Kohno, Alessandro Brawerman is a graduate stu-
“A framework for secure download for software-defined ra- dent pursuing a Doctoral degree in the
dio,” IEEE Commun. Mag., vol. 40, no. 7, pp. 88–96, 2002. School of Electrical and Computer Engi-
[5] C. Castelluccia and F. Dupont, A Simple Privacy Extension neering, Georgia Institute of Technology.
for Mobile IPv6, The Internet Engineering Task Force, Internet He received his B.S. and M.S. degrees in
Draft: Draft-Castellucia- MobileIP-Privacy, February 2001. computer science from the Federal Uni-
[6] A. Escudero, “Location privacy in IPv6—tracking binding versity of Parana, Brazil, during December
updates,” in Proc. International Workshop on Interactive Dis- 1997 and May 2000, respectively. His re-
tributed Multimedia Systems and Telecommunication (IDMS search interests are in security for software
’01), Lancaster, UK, September 2001. defined radio devices, such as cellphones
[7] CMS 88 Cellular Mobile TelephoneSystem, EN/LZT 101908, and PDAs, and security and management of wireless networks. He
Ericsson. currently holds a scholarship from Brazil and his research is funded
[8] Global System for Mobile Communication, “The GSM by the Brazilian National Council of Research (CNPq). He has pub-
security technical whitepaper for 2002,” http://www. lished and presented over 10 technical papers. He is an IEEE Mem-
hackcanada.com/blackcrawl/. ber and was a Fellow of the Panasonic Information & Network-
[9] UC Berkeley. Internet Security, Applications, Authen- ing Technologies Laboratory (2003–2004). He was also the Instruc-
tication and Cryptography Group, “GSM cloning,” tor of several computer science courses at the Federal University of
http://www.isaac.cs.berkeley.edu/isaac/gsm.html.
Parana (1998–2000).
[10] Overview of the Universal Mobile Telecommunication
System. DRAFT, July 2002, http://www.umtsworld.com/ John A. Copeland has been the John H.
technology/overview.htm. Weitnauer, Jr. Chaired Professor at the
[11] M. B. Frederick, “Cellular telephone fraud anit-fraud system,” School of Electrical and Computer Engi-
US Patent 5,448,760, September 1995. neering, Georgia Institute of Technology,
[12] M. S. M. Annoni Notare, F. A. da Silva Cruz, B. Goncalves from 1983 up to date. He teaches grad-
Riso, and C. B. Westphall, “Wireless communications: secu- uate and undergraduate courses on com-
rity management against clonedcellular phones,” in Proc. IEEE
munications networks and network se-
Wireless Communications and Networking Conference (WCNC
curity, and does research in those areas
’99), vol. 3, pp. 1412–1416 , New Orleans, La, USA, September
1999. (http://www.csc.gatech.edu). He was Direc-
tor of the Georgia Center for Advanced
[13] The Trusted Computing Group, http://www.trusted-
computinggroup.org. Telecommunications Technology from 1993 to 1996, Vice Pres-
ident of Technology at Hayes Microcomputer Products from
[14] The TCG PC Specific Implementation Specification, http://
www.trustedcomputinggroup.org/downloads/. 1985 to 1993, and Vice President of Engineering Technology at
Sangamo Weston, Inc. from 1982 to 1985, and did research at
[15] The RC5 encryption algorithm, http://www.secinf.net/
cryptography/The RC5 Encryption Algorithm.html. Bell Labs on semiconductor circuits and optical fiber networks
from 1965 to 1982. He received the B.S., M.S., and Ph.D. de-
[16] RSA encryption, http://mathworld.wolfram.com/RSA-
Encryption.html. grees in physics from the Georgia Institute of Technology. He
[17] Federal Communications Commission. Authorization and has been awarded 38 patents and has published over 60 tech-
use of software defined radio: first report and order, nical papers. In 1970, he received the IEEE’s Morris N. Lieb-
September 2001, http://www.fcc.gov/Bureaus/Engineering mann Award. He is a Fellow of the IEEE and has served as
Technology/Notices/2000/fcc00430.txt. the Editor of the IEEE Transactions on Electron Devices. He
[18] Intel, “Intel wireless trusted platform: security for mo- served on the Board of Trustees for the Georgia Tech Research
bile devices,” http://www.intel.com/design/pca/application- Corporation from 1983 to 1993. He is a Member of Infra-
sprocessors/whitepapers/300868.htm. gard, the ISSA, and the ACM SIGSAC. In 2000, he invented the
[19] A. Brawerman, D. Blough, and B. Bing, “Securing the down- StealthWatch network behavior anomaly detection system, and
load of radio configuration files for software defined radio founded Lancope (http://www.lancope.com) which has deployed
devices,” in Proc. ACM International Workshop on Mobility the StealthWatch system in over 100 corporate and government
Management and Wireless Access (MobiWac ’04), pp. 98–105, networks.
Philadelphia, Pa, USA, September–October 2004.
[20] US Secret Service Financial Crimes Division, http://www.
secretservice.gov/financial 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.

Future Internet

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Future Internet

Uploaded by

Copyright:

Available Formats

EURASIP Journal on Wireless Communications and Networking

Reconfigurable Radio for Future

Guest Editors: Frederik Petré, Ahmet Kondoz,

Reconfigurable Radio for Future

Reconfigurable Radio for Future

Software-Defined Radio—Basics and Evolution to Cognitive Radio, Friedrich K. Jondral

Adaptive Transmitter Optimization in Multiuser Multiantenna Systems: Theoretical Limits, Effect of

Reconfigurable Signal Processing and Hardware Architecture for Broadband Wireless

Modular Software-Defined Radio, Arnd-Ragnar Rhiemeier

Adaptive Mobile Positioning in WCDMA Networks, B. Dong and Xiaodong Wang

Flexible Frequency Discrimination Subsystems for Reconfigurable Radio Front Ends,

1. BACKGROUND band processor for every communication mode. However,

Ashish Pandharipande was born in India

Software-Defined Radio—Basics and Evolution

Received 24 February 2005; Revised 4 April 2005

1. INTRODUCTION concerning the relation between SRs and SDRs is necessary at

This is an open access article distributed under the Creative Commons

SDRs. In Section 4 the software communications architec-

800 900 1000 1100 1200 1300 1400 1500 1600

Figure 2: Mobile spectrum in Europe.

Table 1: Parameters of selected air interfaces.

TETRA IEEE 802.11a GPS DVB-T

z−1 z−1 ··· z−1

Sampling rate adaptation

The SDR transmitter branch consists of the procedures

not. Of course, these visions as well as the recognition cycle for f0

(iii) in order to track the location’s or the spectral environ-

7. TECHNOLOGY CENTRIC COGNITIVE RADIO

Flexible Radio: A Framework for Optimized Multimodal

Received 16 March 2005; Revised 19 April 2005

Figure 1: MIMO code design procedure.

Table 1: Flexible design tools and inputs.

Physical-layer Modulation (a flexible

Channel variation in time

4-QAM without bit loading Bit-loading rate-1 case

7 The algorithm calculates the triplet that needs the min-

List of supported Target BER

Figure 5: Simplified block diagram of algorithm 2.

Each of the three STC schemes can be treated as an ordi-

For TSD, , (2)

outage capacity for the eﬀective channel is possible in order

Table 2: Transmission mode table in the case of perfect channel 7

10−5 > 4.7 >7 > 8.4 > 9.8

SAC ST, QOS = 10−4

No PHN, no RFO No PHN, no RFO OGP

List of supported Target BER

Figure 13: Block diagram of proposed algorithm for mode selection.

Konstantinos Nikitopoulos was born in

Adaptive Transmitter Optimization in Multiuser

Received 6 March 2005; Revised 21 April 2005

1. INTRODUCTION that can provide high-capacity (i.e., spectral eﬃciency) wire-

to the case of a broadcast channel [3]. Recent studies that Mobile 1

Proof. Starting from the right-side term in (14) and with 9

Average user rate (bits/symbol)

Average user rate (bits/symbol)

5.5 transmit antennas at the base station. The MISO channel

assume that it is independent of other channels (i.e., rows of

No prefiltering representing the Doppler shifts that aﬀect N f plane waves

Average user rate (bits/symbol)

In Figure 7, we present the average rates per user versus

tion antenna (m = 1, . . . , M). x̄n are assumed to be cir-

and in the case of a complex variable a + jb, fmod (a + jb) =

Dragan Samardzija was born in Kikinda,

Narayan Mandayam received the B.Tech. (with honors) degree in