MIMO-OFDM is anattractive air-interface solutionfor next-generationwireless local areanetworks (WLANs),wireless metropolitanarea networks(WMANs), andfourth-generationmobile cellular wireless systems.
AD VA N C E S I N SMART AN T E N N A S
INTRODUCTIONThe key challenge faced by future wireless com-munication systems is to provide high-data-ratewireless access at high quality of service (QoS).Combined with the facts that spectrum is ascarce resource and propagation conditions arehostile due to fading (caused by destructiveaddition of multipath components) and interfer-ence from other users, this requirement calls formeans to radically increase spectral efficiencyand to improve link reliability. Multiple-inputmultiple-output (MIMO) wireless technology [1]seems to meet these demands by offeringincreased spectral efficiency through spatial-multiplexing gain, and improved link reliabilitydue to antenna diversity gain. Even thoughthere is still a large number of open researchproblems in the area of MIMO wireless, bothfrom a theoretical perspective and a hardwareimplementation perspective, the technology hasreached a stage where it can be consideredready for use in practical systems. In fact, thefirst products based on MIMO technology havebecome available, for example, the pre-IEEE802.11n wireless local area network (WLAN)systems by Airgo Networks, Inc., Atheros Com-munications, Inc., Broadcom Corporation, Mar-vell Semiconductor, Inc., and MetalinkTechnologies, Inc. Current industry trends sug-gest that large-scale deployment of MIMO wire-less systems will initially be seen in WLANs andin wireless metropolitan area networks(WMANs). Corresponding standards currently
under definition include the IEEE 802.11nWLAN and IEEE 802.16 WMAN standards.Both standards define air interfaces that arebased on the combination of MIMO withorthogonal frequency division multiplexing(OFDM) modulation (MIMO-OFDM). Ongo-ing fourth-generation mobile cellular systemprestandardization efforts in Europe, which arecarried out in the context of various “IntegratedProjects,” funded by the European Commissionwithin its Sixth Framework Program (FP6), alsoshow strong support for a MIMO-OFDM airinterface.
The goal of this article is to provide a high-level review of the basics of MIMO-OFDMwireless systems with a focus on transceiverdesign, multiuser systems, and hardware imple-mentation aspects. The remainder of this arti-cle is organized as follows. The next sectioncontains a brief introduction into MIMO wire-less and OFDM. We then discuss space-fre-quency signaling and corresponding receiverdesign for MIMO-OFDM systems. An overviewof multi-user MIMO-OFDM systems is fol-lowed by a summary of recent results on theVLSI implementation of a four-stream spatial-multiplexing MIMO-OFDM transceiver. Final-ly, we provide a list of relevant open areas forfurther research.
MIMO SYSTEMS ANDOFDM MODULATION
PERFORMANCE GAINS INMIMO SYSTEMS
Traditionally, multiple antennas (at one side ofthe wireless link) have been used to performinterference cancellation and to realize diversityand array gain through coherent combining. Theuse of multiple antennas at both sides of the link(MIMO, Fig. 1a) offers an additional fundamen-tal gain — spatial multiplexing gain, whichresults in increased spectral efficiency. A briefreview of the gains available in a MIMO systemis given in the following.
Spatial multiplexing yields a linear (in theminimum of the number of transmit and receiveantennas) capacity increase, compared to sys-tems with a single antenna at one or both sides
HELMUT BÖLCSKEI, ETH ZURICH
ABSTRACTMultiple-input multiple-output (MIMO)
wireless technology in combination with orthog-onal frequency division multiplexing (MIMO-OFDM) is an attractive air-interface solution fornext-generation wireless local area networks(WLANs), wireless metropolitan area networks(WMANs), and fourth-generation mobile cellu-lar wireless systems. This article provides anoverview of the basics of MIMO-OFDM tech-nology and focuses on space-frequency signaling,receiver design, multiuser systems, and hardwareimplementation aspects. We conclude with a dis-cussion of relevant open areas for furtherresearch.
MIMO-OFDM WIRELESS SYSTEMS: BASICS, PERSPECTIVES, AND CHALLENGES
This work was supportedin part by the SwissNational Science Founda-tion (SNF) under grantno. 200020-109619.
BOLCSKEI LAYOUT 8/3/06 1:24 PM Page 31
Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY MALAYSIA. Downloaded on August 14, 2009 at 03:20 from IEEE Xplore. Restrictions apply.
IEEE Wireless Communications • August 200632
of the link, at no additional power or bandwidthexpenditure [2–4]. The corresponding gain isavailable if the propagation channel exhibits richscattering and can be realized by the simultane-ous transmission of independent data streams inthe same frequency band. The receiver exploitsdifferences in the spatial signatures induced bythe MIMO channel onto the multiplexed datastreams to separate the different signals, therebyrealizing a capacity gain.
Diversity leads to improved link reliabilityby rendering the channel “less fading” and byincreasing the robustness to co-channel inter-ference. Diversity gain is obtained by trans-mitting the data signal over multiple (ideally)independently fading dimensions in time, fre-quency, and space and by performing propercombining in the receiver. Spatial (i.e., anten-na) diversity is particularly attractive whencompared to time or frequency diversity, as itdoes not incur an expenditure in transmissiontime or bandwidth, respectively. Space-timecoding [5] realizes spatial diversity gain insystems with multiple transmit antennas with-out requiring channel knowledge at the trans-mitter.
Array gain can be realized both at the trans-mitter and the receiver. It requires channelknowledge for coherent combining and results inan increase in average receive signal-to-noiseratio (SNR) and hence improved coverage.
Multiple antennas at one or both sides of thewireless link can be used to cancel or reduce co-channel interference, and hence improve cellularsystem capacity.
OFDM MODULATION
MIMO technology will predominantly be used inbroadband systems that exhibit frequency-selec-tive fading and, therefore, intersymbol interfer-ence (ISI). OFDM modulation turns thefrequency-selective channel into a set of parallelflat fading channels and is, hence, an attractiveway of coping with ISI. Figure 1 depicts theschematic of a MIMO-OFDM system. The basicprinciple that underlies OFDM is the insertionof a guard interval, called cyclic prefix (CP),which is a copy of the last part of the OFDMsymbol (Fig. 1c), and has to be long enough toaccommodate the delay spread of the channel.The use of the CP turns the action of the chan-nel on the transmitted signal from a linear con-volution into a cyclic convolution, so that theresulting overall transfer function can be diago-nalized through the use of an IFFT at the trans-mitter and an FFT at the receiver (Fig. 1b).Consequently, the overall frequency-selectivechannel is converted into a set of parallel flatfading channels, which drastically simplifies theequalization task. However, as the CP carriesredundant information, it incurs a loss in spec-tral efficiency, which is usually kept at a maxi-mum of 25 percent.
In general, OFDM has tighter synchroniza-tion requirements than single-carrier (SC) mod-ulation and direct-sequence spread spectrum(DSSS), is more susceptible to phase noise, andsuffers from a larger peak-to-average powerratio. While general statements on overall imple-mentation point-of-view comparisons of OFDM,SC, and DSSS are difficult to make, recentindustry trends show a clear preference forOFDM-based solutions (e.g., IEEE 802.11nWLANs, IEEE 802.16 WMANs, Flarion Tech-nologies’ Flash-OFDM, and the system conceptdeveloped in the context of the European FP6Integrated Project WINNER).
SPACE-FREQUENCY SIGNALING INMIMO-OFDM SYSTEMS
The signaling schemes used in MIMO systemscan be roughly grouped into spatial multiplexing[1], which realizes capacity gain, and space-timecoding [5], which improves link reliability throughdiversity gain. Most multi-antenna signalingschemes, in fact, realize both spatial-multiplexingand diversity gain. A framework for characteriz-ing the trade-off between spatial-multiplexingand diversity gains in flat-fading MIMO chan-nels was proposed in [6]. In the following, wedescribe the basics of spatial multiplexing andspace-time coding with particular emphasis onthe aspects arising from frequency-selective fad-ing through multipath propagation and from theuse of OFDM.
SPATIAL MULTIPLEXING INMIMO-OFDM SYSTEMS
The basic idea of spatial multiplexing isdescribed above. It was shown in [3, 4] that thespatial-multiplexing gain or, equivalently, thenumber of spatial data pipes that can be
n Figure 1. (a) Schematic of a MIMO-OFDM system. OMOD and ODEMODdenote an OFDM-modulator and demodulator, respectively; (b) single-anten-na OFDM modulator and demodulator; (c) adding the cyclic prefix.
c0
IFFT FFTCP
OFDM modulator OFDM demodulator
OMOD
c1
CP
cN-1
r0r1
rN-1
OMOD
(a)
(b)
(c)
OMOD
ODEMODand
Separation
BOLCSKEI LAYOUT 8/3/06 1:24 PM Page 32
Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY MALAYSIA. Downloaded on August 14, 2009 at 03:20 from IEEE Xplore. Restrictions apply.
IEEE Wireless Communications • August 2006 33
opened up within a given frequency band, isgiven by the minimum of the number of trans-mit and receive antennas, provided the receiverknows the channel perfectly. The transmitterdoes not need to have channel state informa-tion (CSI). While the analysis in [3, 4] was car-ried out for flat fading MIMO channels, it wasshown in [7, 8] that the corresponding resultsare robust with respect to multipath-inducedfrequency-selective fading. Moreover, in [8] itwas demonstrated that under real-world propa-gation conditions such as spatial fading corre-lation (caused, e.g., by insufficient antennaspacing), multipath propagation (leading tofrequency-selective fading) can be highly bene-ficial in terms of spatial-multiplexing gain.Multipath propagation tends to increase theangle spread perceived by the transmitter andthe receiver, which, in turn, increases the rankof the channel matrix and hence the spatial-multiplexing gain. This comes, however, at thecost of increased receiver complexity due tothe need to separate the multipath componentsor, equivalently, to equalize the (ISI) MIMOchannel.
Spatial Multiplexing in MIMO-OFDM Systems — In anOFDM-based MIMO system, spatial multiplex-ing is performed by transmitting independentdata streams on a tone-by-tone basis with thetotal transmit power split uniformly acrossantennas and tones. Although the use ofOFDM eliminates ISI, the computational com-plexity of MIMO-OFDM spatial-multiplexingreceivers can still be high. This is because thenumber of data-carrying tones typically rangesbetween 48 (as in the IEEE 802.11a/g stan-dard) and 1728 (as in the IEEE 802.16e stan-dard) and spatial separation has to beperformed for each tone.
Recently, a new class of algorithms that alle-viate this problem was proposed in [9]. The basicidea underlying these algorithms is to exploit thefact that the matrix-valued transfer function in aMIMO-OFDM system is “smooth” across tonesbecause the delay spread in the channel is limit-ed. Computational complexity reductions areobtained by performing channel inversion in thecase of a minimum mean-squared error (MMSE)receiver, or QR decomposition in a spheredecoder (or a successive cancellation receiver)on a subset of tones only and computing theremaining inverses or QR factors, respectively,through interpolation. The resultant savings,compared to brute-force tone-by-tone channelinversion or QR decomposition, are proportion-al to the number of tones divided by the productof the number of transmit antennas and thechannel order (upper-bounded by the length ofthe CP). In practice, a reduction in computation-al complexity of up to 50 percent can beobtained. The performance-complexity trade-off,numerical properties, and memory requirementsof this new class of algorithms remain to beinvestigated in detail.
Noncoherent MIMO-OFDM Systems — With perfectCSI at the receiver and no CSI at the transmit-ter, and fixed transmit power, capacity increaseswith bandwidth until it saturates and is given by
the receive SNR. In the noncoherent case, whereneither the transmitter nor the receiver haveCSI, the capacity behavior as a function of band-width is markedly different: for full-band OFDMsystems (i.e., the transmit signal occupies alltime-frequency slots), beyond a certain criticalbandwidth, “overspreading” occurs, and thecapacity goes to zero. The “overspreading” phe-nomenon was first described in [10] in the con-text of SISO systems and can be explained asfollows. Increasing the bandwidth results in aproportional increase in the number of indepen-dent frequency-diversity branches (provided thechannel satisfies the uncorrelated scatteringassumption). Since the receiver is not assumedto have CSI, these diversity branches contributeto “channel uncertainty” which leads to a capaci-ty penalty. For large bandwidths (and hencesmall SNR per degree of freedom) this penaltyeventually drives the capacity to zero. In theMIMO case, increasing the number of transmitand receive antennas, on the one hand, increasesthe total number of degrees of freedom for com-munication and, on the other hand, results in anincrease in channel uncertainty. Since the totalavailable transmit power is split uniformly acrosstransmit antennas, increasing the number oftransmit antennas results in a smaller SNR perdegree of freedom which leads to the existenceof a finite optimum (in the sense of capacitymaximizing) number of transmit antennas.Increasing the number of receive antennas, onthe other hand, yields an increase in the receiveSNR and is hence always beneficial. In summary,for MIMO-OFDM systems operating at band-widths of several GHz, such as MIMO-basedultra-wideband systems, it is generally not advis-able to use a large number of transmit antennas.Figure 2 provides a numerical result illustratingthis phenomenon.
n Figure 2. Top: capacity lower bound for 4 receive antennas and for varyingnumber of transmit antennas MT as a function of bandwidth W. Bottom: cor-responding optimum (w.r.t. capacity) number of transmit antennas MT (with a maximum of MT = 8). Figure taken from [20].
W (Hz)108107
x 108
02
b/s
4
6
8
10
12
109 1010 1011 1012
W (Hz)108107
0
2
MT
4
6
8
109 1010 1011 1012
MT
BOLCSKEI LAYOUT 8/3/06 1:24 PM Page 33
Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY MALAYSIA. Downloaded on August 14, 2009 at 03:20 from IEEE Xplore. Restrictions apply.
IEEE Wireless Communications • August 200634
SPACE-FREQUENCY CODING INMIMO-OFDM SYSTEMS
While spatial multiplexing aims at increasingspectral efficiency by transmitting independentdata streams, the basic idea of space-time coding[5] is to introduce redundancy across space andtime to realize spatial diversity gain without theneed for CSI at the transmitter.
In single-antenna OFDM systems, frequencydiversity is obtained by coding and interleavingacross tones (and employing appropriate decod-ing algorithms). In frequency-selective fadingMIMO channels, two sources of diversity areavailable: frequency diversity and spatial diversi-ty. It is therefore sensible to ask how these twosources of diversity can be exploited concurrent-ly. Simply using a space-time code to code acrossspace and frequency (rather than time) wasshown in [11], in general, to yield spatial diversi-ty gain only. A straightforward way to realizespace-frequency diversity is to combine thisapproach with forward-error-correction codingand interleaving across tones; most practical sys-tems employ bit-interleaved coded modulation[12]. The problem can, however, be approachedin a more systematic fashion through space-fre-quency codes [11], which essentially spread thedata symbols across space (antennas) and fre-quency (tones), that is, coding is performedwithin one OFDM symbol and not across OFDMsymbols. The resultant code design rules [11],taking the presence of ISI explicitly into account,differ significantly from those for the flat fadingcase [5]: In the coherent case, where the receiverhas perfect CSI, because of ISI, low correlationbetween shifted versions of the transmitted sig-nal is required in addition to the propertiesrequired in the flat-fading case. A framework fordesigning codes that achieve full rate and fulldiversity in frequency-selective fading multi-antenna channels was proposed in [13]. In thenoncoherent case, a good code will allow thereceiver to implicitly “learn” the channel. Code
design for noncoherent MIMO-OFDM systemswas addressed recently, in a systematic fashion,in [14]. In particular, [14] presents space-fre-quency code design criteria, taking the presenceof ISI into account, and provides explicit con-structions of codes that achieve full diversity inspace and frequency. Again, the resulting designcriteria differ significantly from those for the fre-quency-flat fading case. Unlike in the coherentcase, noncoherent space-frequency codesdesigned to achieve full spatial diversity in fre-quency-flat fading channels can fail completelyto exploit not only frequency diversity, but alsospatial diversity, when used in frequency-selec-tive fading environments [14].
MULTIUSER MIMO-OFDM SYSTEMSTo date, research in the MIMO area has focusedpredominantly on point-to-point links. The wire-less industry has just started to integrate MIMOtechnology into WLAN, WMAN, and mobilecellular standards. However, little is knownabout how to optimally leverage the new degreesof freedom resulting from multi-antenna termi-nals and multi-antenna access points or base sta-tions in a network context. A notable exceptionis the multi-antenna broadcast channel with per-fect transmit CSI [15], where the full-capacityregion is known and current research focus is onthe design of low-complexity precoding schemes.In the remainder of this section, we brieflyreview recent results on space-frequency codingand multiple-access in multiuser MIMO-OFDMsystems.
SPACE-FREQUENCY CODING FOR THEMULTIUSER CASE
The main difference between space-frequencycoding in point-to-point channels and in multipleaccess channels (representative of the uplink in amultiuser system) is that in the point-to-pointcase joint encoding across all transmit antennasis possible, while in the multiple-access case indi-vidual users cannot coordinate their transmis-sion. This observation suggests that thespace-frequency code-design problem in themultiple-access case is fundamentally differentfrom the point-to-point case, and joint (acrossusers) code designs that take the multiuseraspect explicitly into account will be required ingeneral. We emphasize, however, that eventhough a joint code book is employed the userswill, of course, not cooperate in selecting theircodeword. It was recently demonstrated in [16]that, depending on the individual users’ trans-mission rates, joint code designs may or may notbe necessary. As a general guideline, the resultsin [16] allow us to conclude that joint codedesigns are necessary whenever multiple userstransmit concurrently at high rates; in this case,the joint code design has to extend over the cor-responding group of users. Otherwise, employingindependently chosen codes designed for point-to-point channels for each of the users is optimal(in terms of error probability). The number ofreceive (base station) antennas plays an impor-tant role in delineating the regions where jointcode designs are necessary from those regions
n Figure 3. Multiple access based on variable amount of user collision in frequen-cy (signal space). Figure taken from [17].
Variable amount of collision
Full collision (CDMA)
User 1
User 1
User 2
User 2
No collision (FDMA)
User 1 User 2
BOLCSKEI LAYOUT 8/3/06 1:24 PM Page 34
Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY MALAYSIA. Downloaded on August 14, 2009 at 03:20 from IEEE Xplore. Restrictions apply.
IEEE Wireless Communications • August 2006 35
where independent single-user codes are opti-mal. Generally, increasing the number of receiveantennas for fixed SNR results in an increase ofthe relative (compared to the capacity region)size of the latter region. This is due to the factthat for a large number of receive antennasthere are more spatial degrees of freedom avail-able to separate the individual users’ signals sothat imposing “separation” through appropriatejoint code design is required for a smaller set of(high) rates. We finally note that the discussionin this paragraph pertains to space-time codes aswell.
Space-frequency code design for broadcastchannels (representative of the downlink in amultiuser system) is a largely unexplored area.
MULTIPLE ACCESS IN MIMO-OFDM SYSTEMS
Multiple access and broadcasting is fundamen-tally different in systems with multi-antenna ter-minals and base stations compared to systemswith single-antenna terminals, base stations, orboth. The underlying reason is that realizing spa-tial-multiplexing gain requires the users to col-lide (interfere) in signal space. This favorscollision-based (nonorthogonal) multiple-accessschemes such as code division multiple-access(CDMA) over orthogonal multiple-accessschemes such as frequency division multiple-access (FDMA) or time division multiple-access(TDMA).
In OFDM-based systems it is particularly sim-ple to realize variable amounts of collision insignal space by assigning different subsets of theavailable OFDM tones to different users. Thecorresponding multiple-access or broadcastschemes, commonly referred to as OFDMA,range from FDMA (each OFDM tone isassigned to at most one user) to CDMA (eachOFDM tone is assigned to all users). The situa-tion is depicted schematically in Fig. 3. Note thathere the terminology CDMA is used solely toindicate that all users collide on all tones.Spreading, which introduces redundancy, yieldsan inferior capacity performance compared to aCDMA scheme according to our definition. Theimpact of variable amount of collision in OFDM-based multiple-access schemes was analyzed indetail in [17]. The main findings, assuming thatjoint decoding is employed, can be summarizedas follows: the capacity region obtained for anyamount of collision is outer-bounded by thecapacity region obtained for a fully collision-based multiple-access scheme (i.e., CDMA).This result holds, irrespective of the number ofantennas at the terminals and the base station.One may now be tempted to conclude that thereis no case for multiple-access schemes other thanone with full collision. In practice, however, min-imizing the amount of collision in frequency isdesirable, as this minimizes the receiver com-plexity incurred by having to separate the collid-ing (interfering) signals. In summary, there is atrade-off between capacity and receiver complex-ity. The following (rough) rules of thumb, appli-cable in the high-SNR case, may serve aspractically relevant guidelines for system design:• When the users are spatially well separated, as
measured by their spatial signatures induced
at the base station, and the number of basestation antennas is high, collision in frequencyis crucial to maximize the system (i.e., sum)capacity.
• For poor spatial separation or a small numberof base station antennas, or both, the impactof collision on system capacity is small.More detailed design guidelines can be
found in [17], which furthermore reveals that,when considering system capacity, the numberof base station antennas is typically the limitingfactor. Based on this result, one may be tempt-ed to conclude that there is no point for multi-antenna terminals. This is, however, not thecase, as using multi-antenna terminals willresult in higher individual data rates andimproved per-user link quality. An analysis ofthe impact of variable amount of collision inbroadcast channels does not seem to be avail-able at this point.
HARDWARE IMPLEMENTATION ASPECTSThe gains achievable in MIMO(-OFDM) sys-tems come at an (often significant) increase inhardware complexity. Little is known about suit-able VLSI architectures for MIMO(-OFDM)systems and the corresponding silicon complexi-ty. The first commercial MIMO(-OFDMA) chipset was developed by IospanWireless, Inc. in2002 for a proprietary fixed wireless system. Thischip set supported two-stream spatial multiplex-ing and space-time coding. Several companieshave announced MIMO-OFDM chip sets for the
n Figure 4. Layout and chip micrograph (upper left corner) of the MIMO-OFDMbaseband signal processing ASIC [18] manufactured in 0.25 µm 1P/5M 2.5 VCMOS technology.
BOLCSKEI LAYOUT 8/3/06 1:24 PM Page 35
Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY MALAYSIA. Downloaded on August 14, 2009 at 03:20 from IEEE Xplore. Restrictions apply.
IEEE Wireless Communications • August 200636
upcoming IEEE 802.11n WiFi standard. AirgoNetworks, Inc. offered a prestandard chip setearlier in 2005. In the mobile WiMAX area(IEEE 802.16e), Beceem Communications, Inc.has developed MIMO-OFDMA chip sets sup-porting two-stream spatial multiplexing, space-time coding, and beamforming.
A four-stream (four transmit and fourreceive antennas) MIMO-OFDM WLAN physi-cal layer testbed has recently been developed ina collaboration between the Integrated SystemsLaboratory (IIS) and the Communication Tech-nology Laboratory (CTL) at ETH Zurich. Nextwe briefly summarize the main features of thistestbed. The basic system architecture of thetestbed is based on the SISO IEEE 802.11 a/gOFDM physical layer (FFT length 64, CPlength 16, FFT bandwidth 20 MHz, symbolduration 4 µs, support of BPSK, QPSK, 16-QAM, and 64-QAM, and rate 1/2 convolutionalcoding) and is, therefore, most relevant to theupcoming IEEE 802.11n standard. Furtherspecifics of the testbed are as follows: with anintermediate frequency (IF) of 20 MHz, the(direct IF) sampling rate of the A/D and D/Aconverters is 80 Msamples/s, which is digitallydownconverted to a baseband sampling rate of20 Msamples/s. Each receive RF chain containsa gain control element.
The ASIC described in [18] and shown in Fig.4 contains the baseband digital-signal-processingfunctional blocks of the PHY layer describedabove, including an MMSE ordered-successive-interference-cancellation (OSIC) MIMO receiv-er. It operates at 80 MHz clock frequency andachieves uncoded data rates of up to 192 Mb/s ina 20 MHz channel, which corresponds to a spec-tral efficiency of 9.6 b/s/Hz. The die area break-
down of the ASIC according to functional blocksalong with die area figures for a correspondingSISO system, is summarized in Table 1. Com-pared to a SISO transceiver, the 4 × 4 MIMOtransceiver requires the four-fold replication ofmost functional blocks and, in addition, a chan-nel-matrix preprocessor for MIMO detectionand the MIMO detector itself; as a result, theoverall chip area increases by a factor of 6.5.The main bottleneck in implementing the 4 × 4MIMO system was found to be the latencyincurred by preprocessing the channel matricesfor MIMO-OFDM detection. We therefore con-clude that algorithms for computationally effi-cient MIMO-OFDM channel matrixpreprocessing, such as those described in [9], areof utmost importance for practical implementa-tions.
AREAS FOR FUTURE RESEARCH
We conclude this survey article with a brief dis-cussion of open problems in the area of MIMO-OFDM that need to be addressed so that thegains promised by the technology can be fullyleveraged in practical systems.
As mentioned above, multiuser MIMO sys-tems are largely unexplored. Making progress inthe area of multiuser MIMO systems is of keyimportance to the development of practical sys-tems that exploit MIMO gains on the systemlevel also. The recently launched EU FP6STREP project MASCOT (Multiple-AccessSpace-Time Coding Testbed) is aimed at devel-oping, analyzing, and implementing (in hard-ware) concepts and techniques for multiuserMIMO communications. Specific areas of rele-vance in the context of multiuser MIMO sys-tems include multiple-access schemes,transceiver design (including precoding), andspace-frequency code design. In particular, thevariable amount of collision-based frameworkfor multiple access, introduced in [17], needs tobe further developed to account for the pres-ence of out-of-cell interference and to allow forvariable amounts of collision in space, time, andfrequency. Flarion Technologies’ Flash-OFDMsystem can be seen as a special case of such ageneral system.
Even though it probably constitutes one ofthe most important areas in MIMO wirelessthat remain to be addressed, the MIMO com-munity has seen relatively little work on hard-ware implementation aspects arising in MIMOtransceiver design. An exception is the recentPh.D. thesis [19], which reports, among otherresults, the ASIC implementation of a spheredecoder. Hardware implementation problemsof significant current interest include efficientalgorithms for (soft-)sphere decoding and forchannel preprocessing in MIMO-OFDM sys-tems.
To date most of the work on (multiuser)MIMO has focused on physical layer aspects.Understanding the impact of MIMO technologyon the higher layers and, in particular, the devel-opment of link adaptation, scheduling, andretransmission algorithms that make explicit useof the MIMO nature of the system are of signifi-cant interest.
n Table 1. Chip area of baseband functionalblocks in 0.25 µm CMOS technology (FOE,FOC, and FSD stand for frequency offset estima-tion, frequency offset compensation, and framestart detection, respectively). Taken from [18].
ComponentArea (mm2)
SISO 4 × 4 MIMO
DDC, DUC 0.5 1.9
AGC 0.1 0.4
FOE, FOC, FSD 0.3 1.3
Modulator, I/FFT 0.9 1.4
Frame buffers — 3.3
Ch. est. and ch. mem. < 0.1 1.12
QR decomposition — 1.29
QR memory — 1.23
MIMO detector — 0.9
Total 1.9 12.8
To date most of thework on (multiuser)MIMO has focusedon physical layer aspects. Understanding theimpact of MIMOtechnology on thehigher layers is ofsignificant interest.
BOLCSKEI LAYOUT 8/3/06 1:24 PM Page 36
Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY MALAYSIA. Downloaded on August 14, 2009 at 03:20 from IEEE Xplore. Restrictions apply.
IEEE Wireless Communications • August 2006 37
ACKNOWLEDGMENTS
The author would like to thank his collabora-tors D. Baum, M. Borgmann, A. Burg, D.Cescato, M. Gärtner, S. Häne, D. Perels, U.Schuster, and S. Visuri, whose work was sur-veyed in this article. Fruitful and stimulatingdiscussions and feedback on this article fromthe above mentioned collaborators are acknowl-edged as well.
REFERENCES[1] A. J. Paulraj, R. U. Nabar, and D. A. Gore, Introduction
to Space-Time Wireless Communications, Cambridge,UK: Cambridge Univ. Press, 2003.
[2] A. J. Paulraj and T. Kailath, “Increasing Capacity inWireless Broadcast Systems Using Distributed Transmis-sion/Directional Reception,” U.S. Patent no. 5,345,599,1994.
[3] G. J. Foschini, “Layered Space-Time Architecture forWireless Communication in a Fading Environment whenUsing Multielement Antennas,” Bell Labs Tech. J.,Autumn 1996, pp. 41–59.
[4] I. E. Telatar, “Capacity of Multi-Antenna Gaussian Chan-nels,” Euro. Trans. Telecommun., vol. 10, no. 6,Nov./Dec. 1999, pp. 585–95.
[5] V. Tarokh, N. Seshadri, and A. R. Calderbank, “Space-Time Codes for High Data Rate Wireless Communica-tion: Performance Criterion and Code Construction,”IEEE Trans. Info. Theory, vol. 44, no. 2, Mar. 1998, pp.744–65.
[6] L. Zheng and D. N. C. Tse, “Diversity and Multiplexing:A Fundamental Trade-off in Multiple Antenna Chan-nels,” IEEE Trans. Info. Theory, vol. 49, no. 5, May2003, pp. 1073–96.
[7] G. G. Raleigh and J. M. Cioffi, “Spatio-Temporal Codingfor Wireless Communication,” IEEE Trans. Commun.,vol. 46, no. 3, 1998, pp. 357–66.
[8] H. Bölcskei, D. Gesbert, and A. J. Paulraj, “On theCapacity of OFDM-Based Spatial Multiplexing Systems,”IEEE Trans. Commun., vol. 50, no. 2, Feb. 2002, pp.225–34.
[9] M. Borgmann and H. Bölcskei, “Interpolation-Based Effi-cient Matrix Inversion for MIMO-OFDM receivers,” Proc.38th Asilomar Conf. Signals, Syst., and Computers,Pacific Grove, CA, Nov. 2004, pp. 1941–47.
[10] I. E. Telatar and D. N. C. Tse, “Capacity and MutualInformation of Wideband Multipath Fading Channels,”IEEE Trans. Info. Theory, vol. 46, no. 4, July 2000, pp.1384–1400.
[11] H. Bölcskei and A. J. Paulraj, “Space-Frequency CodedBroadband OFDM Systems,” Proc. IEEE Wireless Com-mun. and Networking Conf., Chicago, IL, Sept. 2000,pp. 1–6.
[12] G. Caire, G. Taricco, and E. Biglieri, “Bit-InterleavedCoded Modulation,” IEEE Trans. Info. Theory, vol. 44,no. 3, May 1998, pp. 927–46.
[13] X. Ma and G. B. Giannakis, “Full-Diversity Full-RateComplex-Field Space-Time Coding,” IEEE Trans. Sig. Pro-cessing, vol. 51, no. 11, Nov. 2003, pp. 2917–30.
[14] M. Borgmann and H. Bölcskei, “Noncoherent Space-Frequency Coded MIMO-OFDM,” IEEE JSAC, vol. 23, no.9, Sept. 2005, pp. 1799–1810.
[15] G. Caire and S. Shamai (Shitz), “On the AchievableThroughput of a Multiantenna Gaussian BroadcastChannel,” IEEE Trans. Info. Theory, vol. 49, no. 7, July2003, pp. 1691–1706.
[16] M. E. Gärtner and H. Bölcskei, “Multi-User Space-Time/Frequency Code Design,” Proc. IEEE ISIT, Seattle,WA, July 2006.
[17] S. Visuri and H. Bölcskei, “Multiple Access Strategiesfor Frequency-Selective MIMO Channels,” IEEE Trans.Info. Theory, Sept. 2006 (to appear).
[18] D. Perels et al., “ASIC Implementation of a MIMO-OFDM Transceiver for 192 Mb/s WLANs,” Euro. Solid-State Circuits Conf., Sept. 2005, pp. 215–18.
[19] A. P. Burg, “VLSI Circuits for MIMO CommunicationSystems,” Ph.D. dissertation, ETH Zurich, Switzerland,2006.
[20] M. Borgmann and H. Bölcskei, “On the Capacity ofNoncoherent Wideband MIMO-OFDM Systems,” IEEEInt’l. Symp. Info. Theory, Adelaide, Australia, Sept.2005, pp. 651–55.
ADDITIONAL READING[1] H. Bölcskei et al., Eds., Space-Time Wireless Systems:
From Array Processing to MIMO Communications, Cam-bridge, UK: Cambridge Univ. Press, 2006.
[2] D. Tse and P. Viswanath, Fundamentals of WirelessCommunication, Cambridge, UK: Cambridge Univ.Press, 2005.
BIOGRAPHYHELMUT BÖLCSKEI [M’98, SM’02] ([email protected])received his Dr.Techn. degree in electrical engineering fromVienna University of Technology, Austria, in 1997. In 1998he was with Vienna University of Technology. From 1999to 2001 he was a postdoctoral researcher with the Infor-mation Systems Laboratory, Department of Electrical Engi-neering, Stanford University, Palo Alto, California. He wason the founding team of Iospan Wireless Inc., a Silicon Val-ley based startup company (acquired by Intel in 2002) thatspecialized in MIMO wireless systems for high-speed Inter-net access. From 2001 to 2002 he was an assistant profes-sor of electrical engineering at the University of Illinois atUrbana-Champaign. Since 2002 he has been an assistantprofessor of communication theory at ETH Zurich, Switzer-land. He was a visiting researcher at Philips Research Labo-ratories, Eindhoven, The Netherlands; ENST, Paris, France,and the Heinrich Hertz Institute, Berlin, Germany. Hisresearch interests include communication and informationtheory with special emphasis on wireless communicationsand signal processing. He received the 2001 IEEE SignalProcessing Society Young Author Best Paper Award, the2006 IEEE Communications Society Leonard G. AbrahamBest Paper Award, and the ETH “Golden Owl” TeachingAward, and was an Erwin Schrödinger Fellow (1999–2001)of the Austrian National Science Foundation (FWF). He hasserved as an Associate Editor of IEEE Transactions on SignalProcesssing, IEEE Transactions on Wireless Communica-tions, and EURASIP Journal on Applied Signal Processing.He is currently on the editorial board of Foundations andTrends in Networking.
BOLCSKEI LAYOUT 8/3/06 1:24 PM Page 37
Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY MALAYSIA. Downloaded on August 14, 2009 at 03:20 from IEEE Xplore. Restrictions apply.
