Professional Documents
Culture Documents
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
Flexible Radio: A Framework for Optimized Multimodal Operation via Dynamic Signal Design,
Ioannis Dagres, Andreas Zalonis, Nikos Dimitriou, Konstantinos Nikitopoulos,
and Andreas Polydoros
Volume 2005 (2005), Issue 3, Pages 284-297
Flexible Transmission Scheme for 4G Wireless Systems with Multiple Antennas, François Horlin,
Frederik Petré, Eduardo Lopez-Estraviz, Frederik Naessens, and Liesbet Van der Perre
Volume 2005 (2005), Issue 3, Pages 308-322
Flexible Analog Front Ends of Reconfigurable Radios Based on Sampling and Reconstruction with
Internal Filtering, Yefim S. Poberezhskiy and Gennady Y. Poberezhskiy
Volume 2005 (2005), Issue 3, Pages 364-381
A Reconfigurable Spiral Antenna for Adaptive MIMO Systems, Bedri A. Cetiner, J. Y. Qian, G. P. Li,
and F. De Flaviis
Volume 2005 (2005), Issue 3, Pages 382-389
Multimode Communication Protocols Enabling Reconfigurable Radios, Lars Berlemann, Ralf Pabst,
and Bernhard Walke
Volume 2005 (2005), Issue 3, Pages 390-400
Towards a Fraud-Prevention Framework for Software Defined Radio Mobile Devices,
Alessandro Brawerman and John A. Copeland
Volume 2005 (2005), Issue 3, Pages 401-412
EURASIP Journal on Wireless Communications and Networking 2005:3, 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
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).
274 EURASIP Journal on Wireless Communications and Networking
Friedrich K. Jondral
Universität Karlsruhe (TH), Institut für Nachrichtentechnik, D-76128 Karlsruhe, Germany
Email: fj@int.uni-karlsruhe.de
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.
reproduction in any medium, provided the original work is properly cited. wideband receivers.
276 EURASIP Journal on Wireless Communications and Networking
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
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
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.
278 EURASIP Journal on Wireless Communications and Networking
I/Q balancing
∼
∼
π ···
−
2 + + ··· +
Quadrature
Figure 4: Polyphase filter for sampling rate adaptation.
× ∼ A/D component
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
SDR—Basics and Evolution to Cognitive Radio 281
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.
282 EURASIP Journal on Wireless Communications and Networking
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
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.
EURASIP Journal on Wireless Communications and Networking 2005:3, 284–297
c 2005 Ioannis Dagres et al.
Ioannis Dagres
Institute of Accelerating Systems & Applications (IASA), National Kapodistrian University of Athens (NKUA),
P.O. Box 17214, 10024 Athens, Greece
Email: jdagres@phys.uoa.gr
Andreas Zalonis
Institute of Accelerating Systems & Applications (IASA), National Kapodistrian University of Athens (NKUA),
P.O. Box 17214, 10024 Athens, Greece
Email: azalonis@phys.uoa.gr
Nikos Dimitriou
Institute of Accelerating Systems & Applications (IASA), National Kapodistrian University of Athens (NKUA),
P.O. Box 17214, 10024 Athens, Greece
Email: nikodim@phys.uoa.gr
Konstantinos Nikitopoulos
Institute of Accelerating Systems & Applications (IASA), National Kapodistrian University of Athens (NKUA),
P.O. Box 17214, 10024 Athens, Greece
Email: cnikit@cc.uoa.gr
Andreas Polydoros
Institute of Accelerating Systems & Applications (IASA), National Kapodistrian University of Athens (NKUA),
P.O. Box 17214, 10024 Athens, Greece
Email: polydoros@phys.uoa.gr
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.
286 EURASIP Journal on Wireless Communications and Networking
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
Flexible Radio Framework for Optimized Multimodal Operation 287
Tx1
Outer Inner Inner Outer
code code Rx1 . decoder decoder
Tx2 .
.
RxMr
TxMt
MIMO channel
SISO channel
CSI
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
288 EURASIP Journal on Wireless Communications and Networking
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).
Flexible Radio Framework for Optimized Multimodal Operation 289
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
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
Flexible Radio Framework for Optimized Multimodal Operation 291
H (xn , RUBn )
constellations
H
Channel/noise Mode Tx power
N
0 evaluation PTx1
estimator
..
.
PTxn
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.
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 af- fect 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
Flexible Radio Framework for Optimized Multimodal Operation 293
Bits/carrier
10−4 > 4.2 > 6.4 > 7.6 > 9.2 4
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)
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.
Flexible Radio Framework for Optimized Multimodal Operation 295
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.
296 EURASIP Journal on Wireless Communications and Networking
[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 wire- less 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.
Flexible Radio Framework for Optimized Multimodal Operation 297
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.
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
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.
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.
300 EURASIP Journal on Wireless Communications and Networking
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
Transmitter Optimization in Multiuser Multiantenna Systems 301
1 5.5
0.9
5
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
Average user rate (bits/symbol)
−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
Transmitter Optimization in Multiuser Multiantenna Systems 303
3.5 3.5
3 3
Average user rate (bits/symbol)
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.
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
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)
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
Transmitter Optimization in Multiuser Multiantenna Systems 307
François Horlin
Wireless Research, Interuniversity Micro-Electronics Center (IMEC), Kapeldreef 75, B-3001 Leuven, Belgium
Email: francois.horlin@imec.be
Frederik Petré
Wireless Research, Interuniversity Micro-Electronics Center (IMEC), Kapeldreef 75, B-3001 Leuven, Belgium
Email: petre@imec.be
Eduardo Lopez-Estraviz
Wireless Research, Interuniversity Micro-Electronics Center (IMEC), Kapeldreef 75, B-3001 Leuven, Belgium
Email: lopezest@imec.be
Frederik Naessens
Wireless Research, Interuniversity Micro-Electronics Center (IMEC), Kapeldreef 75, B-3001 Leuven, Belgium
Email: naessen@imec.be
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.
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.
310 EURASIP Journal on Wireless Communications and Networking
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.
Flexible Transmission Scheme for 4G Wireless Systems 311
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].
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
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)
314 EURASIP Journal on Wireless Communications and Networking
Table 1: Initialization complexity, M users, NT transmit antennas, Q carriers/subchannels, and B complex symbols per block.
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
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.
318 EURASIP Journal on Wireless Communications and Networking
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
Eb /N0 (dB) Eb /N0 (dB)
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
Eb /N0 (dB) Eb /N0 (dB)
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
Flexible Transmission Scheme for 4G Wireless Systems 319
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)
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
320 EURASIP Journal on Wireless Communications and Networking
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
Flexible Transmission Scheme for 4G Wireless Systems 321
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 in- ter 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).
322 EURASIP Journal on Wireless Communications and Networking
Ying-Chang Liang
Institute for InfoComm Research (I2R), 21 Heng Mui Keng Terrace, Singapore 119613
Email: ycliang@i2r.a-star.edu.sg
Sayed Naveen
Institute for InfoComm Research (I2R), 21 Heng Mui Keng Terrace, Singapore 119613
Email: naveen@i2r.a-star.edu.sg
Santosh K. Pilakkat
Institute for InfoComm Research (I2R), 21 Heng Mui Keng Terrace, Singapore 119613
Email: pilakkat@i2r.a-star.edu.sg
Ashok K. Marath
Institute for InfoComm Research (I2R), 21 Heng Mui Keng Terrace, Singapore 119613
Email: ashok@i2r.a-star.edu.sg
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.
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
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).
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
326 EURASIP Journal on Wireless Communications and Networking
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.
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)
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.
Reconfigurable Signal Processing for Wireless Communications 327
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
Synchronization
Table 1: Parameter configurations for Modem Tx. Table 3: Equalizer complexity per block for CDMA-based systems.
User
RF front end RF front end System & interface
configuration
RTR Engine manager
Baseband Baseband
engine1 engine < n > Config. database
RTOS
SDR executive
DSP Kernel/OS
SDR modem subsystem General-purpose processor
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
330 EURASIP Journal on Wireless Communications and Networking
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
Reconfigurable Signal Processing for Wireless Communications 331
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.
332 EURASIP Journal on Wireless Communications and Networking
Arnd-Ragnar Rhiemeier
Institut für Nachrichtentechnik, Universität Karlsruhe (TH), 76128 Karlsruhe, Germany
Email: rhiemeier@int.uni-karlsruhe.de
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.
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
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
Modular Software-Defined Radio 337
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
1 1
0.9 0.9
Fraction of max speedup
0.7 0.7
0.6 0.6
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
340 EURASIP Journal on Wireless Communications and Networking
1 1
NF = 7
NF = 6
0.9 0.9
NF = 7
Fraction of max speedup
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.
Modular Software-Defined Radio 341
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
342 EURASIP Journal on Wireless Communications and Networking
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.
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
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.
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
346 EURASIP Journal on Wireless Communications and Networking
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
Adaptive Mobile Positioning in WCDMA Networks 347
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.
348 EURASIP Journal on Wireless Communications and Networking
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)
Adaptive Mobile Positioning in WCDMA Networks 349
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).
350 EURASIP Journal on Wireless Communications and Networking
Table 1: Performance comparisons between EKF, standard PF, and APF in terms of NMSE based on different observations for scenario 1.
Table 2: Performance comparisons between EKF, standard PF, and APF in terms of NMSE based on different observations for scenario 2.
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
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
Adaptive Mobile Positioning in WCDMA Networks 351
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
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.
352 EURASIP Journal on Wireless Communications and Networking
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
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.
Adaptive Mobile Positioning in WCDMA Networks 353
Bruce E. Carey-Smith
Centre for Communications Research, University of Bristol, Bristol BS8 1UB, UK
Email: b.carey-smith@bristol.ac.uk
Paul A. Warr
Centre for Communications Research, University of Bristol, Bristol BS8 1UB, UK
Email: paul.a.warr@bristol.ac.uk
Phill R. Rogers
Centre for Communications Research, University of Bristol, Bristol BS8 1UB, UK
Email: phill.rogers@bristol.ac.uk
Mark A. Beach
Centre for Communications Research, University of Bristol, Bristol BS8 1UB, UK
Email: m.a.beach@bristol.ac.uk
Geoffrey S. Hilton
Centre for Communications Research, University of Bristol, Bristol BS8 1UB, UK
Email: geoff.hilton@bristol.ac.uk
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.
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
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
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
Flexible Subsystems for Radio Front Ends 357
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
358 EURASIP Journal on Wireless Communications and Networking
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
Flexible Subsystems for Radio Front Ends 359
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)
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
−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.
Flexible Subsystems for Radio Front Ends 361
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.
362 EURASIP Journal on Wireless Communications and Networking
[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.
Flexible Subsystems for Radio Front Ends 363
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
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
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
Flexible Analog Front Ends of Reconfigurable Radios 367
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.
Flexible Analog Front Ends of Reconfigurable Radios 369
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).
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)
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).
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
372 EURASIP Journal on Wireless Communications and Networking
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
374 EURASIP Journal on Wireless Communications and Networking
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.
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
376 EURASIP Journal on Wireless Communications and Networking
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
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
378 EURASIP Journal on Wireless Communications and Networking
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
380 EURASIP Journal on Wireless Communications and Networking
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
Flexible Analog Front Ends of Reconfigurable Radios 381
Bedri A. Cetiner
Space Science Center, Morehead State University, Morehead, KY 40351, USA
Email: ba.cetiner@moreheadstate.edu
J. Y. Qian
Department of Electrical Engineering and Computer Science, University of California at Irvine, Irvine, CA 92697-2625, USA
Email: qianj@uci.edu
G. P. Li
Department of Electrical Engineering and Computer Science, University of California at Irvine, Irvine, CA 92697-2625, USA
Email: gpli@uci.edu
F. De Flaviis
Department of Electrical Engineering and Computer Science, University of California at Irvine, Irvine, CA 92697-2625, USA
Email: franco@uci.edu
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.
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
384 EURASIP Journal on Wireless Communications and Networking
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.
A Reconfigurable Spiral Antenna for Adaptive MIMO Systems 385
Spiral
segments
(a)
(b)
(c)
RF-MEMS
actuator
(d)
Figure 3: Fabrication sequence for monolithic integration of RF-MEMS actuators with rectangular-shaped strip segments in constructing
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
386 EURASIP Journal on Wireless Communications and Networking
90 90
135 45 135 45
−10 −10
−20 −20
270 270
(a) (b)
90 90
135 45 135 45
−10 −10
−20 −20
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.
A Reconfigurable Spiral Antenna for Adaptive MIMO Systems 387
90 90
135 45 135 45
−10 −10
−20 −20
270 270
(a) (b)
90 90
135 45 135 45
−10 −10
−20 −20
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.
Lars Berlemann
Chair of Communication Networks, RWTH Aachen University, Kopernikusstraße 16, D-52074 Aachen, Germany
Email: ber@comnets.rwth-aachen.de
Ralf Pabst
Chair of Communication Networks, RWTH Aachen University, Kopernikusstraße 16, D-52074 Aachen, Germany
Email: pab@comnets.rwth-aachen.de
Bernhard Walke
Chair of Communication Networks, RWTH Aachen University, Kopernikusstraße 16, D-52074 Aachen, Germany
Email: walke@comnets.rwth-aachen.de
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.
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 reconfigurabil- ity.
392 EURASIP Journal on Wireless Communications and Networking
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
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].
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
394 EURASIP Journal on Wireless Communications and Networking
Interface
SAP
Acknow-
Interface ledged
Multiplexer
mode
To RRC Transpa- Unacknow-
rent mode ledged Segmentation unit
mode
Manager
Segmentation unit ARQ unit
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.
Multimode Protocols for Reconfigurable Radios 395
To higher layer
Interface
Multiplexer
Manager IP address
Multiplexer Multiplexer
TCP port UDP port
TCP PDU factory UDP PDU factory
Multiplexer
Segmentation unit
IP PDU factory
Interface
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.
396 EURASIP Journal on Wireless Communications and Networking
To LLC SAP
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.
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
Multimode Protocols for Reconfigurable Radios 397
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)
(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.
398 EURASIP Journal on Wireless Communications and Networking
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
(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.
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.
400 EURASIP Journal on Wireless Communications and Networking
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
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.
402 EURASIP Journal on Wireless Communications and Networking
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
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.
404 EURASIP Journal on Wireless Communications and Networking
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.
A Fraud-Prevention Framework for SDR Mobile Devices 405
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 manufac- turer 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
406 EURASIP Journal on Wireless Communications and Networking
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
Verifies C
Generates C
MD(EK) in M EKDB ?
Gets EK
Obtains C
Establishes unsecure
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
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 ob- viously 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
408 EURASIP Journal on Wireless Communications and Networking
Privacy CA
Obtains EK
Generates AC
Generates AK pair
Att(EK) in EKDB ?
Obtains C
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
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
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.
A Fraud-Prevention Framework for SDR Mobile Devices 409
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
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
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.
A Fraud-Prevention Framework for SDR Mobile Devices 411
Table 4: Possible attacks and how the secure protocol avoids them.
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.
412 EURASIP Journal on Wireless Communications and Networking
[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.