EURASIP Journal on Applied Signal Processing
Space-Time Coding and its Applications—Part II
Guest Editors: Dirk Slock, Vahid Tarokh, and Xiang-Gen Xia
EURASIP Journal on Applied Signal Processing Space-Time Coding and its Applications—Part II
EURASIP Journal on Applied Signal Processing Space-Time Coding and its Applications—Part II
Guest Editors: Dirk Slock, Vahid Tarokh, and Xiang-Gen Xia
Copyright © 2002 Hindawi Publishing Corporation. All rights reserved.
This is a special issue published in volume 2002 of “EURASIP Journal on Applied Signal Processing.” 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 K. J. Ray Liu, University of Maryland, College Park, USA
Associate Editors Kiyoharu Aizawa, Japan Jiri Jan, Czech Antonio Ortega, USA Gonzalo Arce, USA Shigeru Katagiri, Japan Mukund Padmanabhan, USA Jaakko Astola, Finland Mos Kaveh, USA Ioannis Pitas, Greece Mauro Barni, Italy Bastiaan Kleijn, Sweden Raja Rajasekaran, USA Sankar Basu, USA Ut Va Koc, USA Phillip Regalia, France Shih-Fu Chang, USA Aggelos Katsaggelos, USA Hideaki Sakai, Japan Jie Chen, USA C. C. Jay Kuo, USA William Sandham, UK Tsuhan Chen, USA S. Y. Kung, USA Wan-Chi Siu, Hong Kong M. Reha Civanlar, USA Chin-Hui Lee, USA Piet Sommen, The Netherlands Tony Constantinides, UK Kyoung Mu Lee, Korea John Sorensen, Denmark Luciano Costa, Brazil Y. Geoffrey Li, USA Michael G. Strintzis, Greece Irek Defee, Finland Heinrich Meyr, Germany Ming-Ting Sun, USA Ed Deprettere, The Netherlands Ferran Marques, Spain Tomohiko Taniguchi, Japan Zhi Ding, USA Jerry M. Mendel, USA Sergios Theodoridis, Greece Jean-Luc Dugelay, France Marc Moonen, Belgium Yuke Wang, USA Pierre Duhamel, France José M. F.Moura, USA Andy Wu, Taiwan Tariq Durrani, UK Ryohei Nakatsu, Japan Xiang-Gen Xia, USA Sadaoki Furui, Japan King N. Ngan, Singapore Zixiang Xiong, USA Ulrich Heute, Germany Takao Nishitani, Japan Kung Yao, USA Yu Hen Hu, USA Naohisa Ohta, Japan
Contents
Editorial, Dirk Slock, Vahid Tarokh, and Xiang-Gen Xia Volume 2002 (2002), Issue 5, Pages 445-446
On the Capacity of Certain Space-Time Coding Schemes, Constantinos B. Papadias and Gerard J. Foschini Volume 2002 (2002), Issue 5, Pages 447-458
Space-Time Turbo Trellis Coded Modulation for Wireless Data Communications, Welly Firmanto, Branka Vucetic, Jinhong Yuan, and Zhuo Chen Volume 2002 (2002), Issue 5, Pages 459-470
On Some Design Issues of Space-Time Coded Multi-Antenna Systems, Hsuan-Jung Su and Evaggelos Geraniotis Volume 2002 (2002), Issue 5, Pages 471-481
Space-Time Trellis Coded 8PSK Schemes for Rapid Rayleigh Fading Channels, Salam A. Zummo and Saud A. Al-Semari Volume 2002 (2002), Issue 5, Pages 482-486
Blind Identification of Convolutive MIMO Systems with 3 Sources and 2 Sensors, Binning Chen, Athina P. Petropulu, and Lieven De Lathauwer Volume 2002 (2002), Issue 5, Pages 487-496
Maximum Likelihood Blind Channel Estimation for Space-Time Coding Systems, Hakan A. Çırpan, Erdal Panayırcı, and Erdinc Çekli Volume 2002 (2002), Issue 5, Pages 497-506
Pilot-Symbol-Assisted Channel Estimation for Space-Time Coded OFDM Systems, King F. Lee and Douglas B. Williams Volume 2002 (2002), Issue 5, Pages 507-516
Low-Complexity Iterative Receiver for Space-Time Coded Signals over Frequency Selective Channels, Noura Sellami, Inbar Fijalkow, and Mohamed Siala Volume 2002 (2002), Issue 5, Pages 517-524
Maximum-Likelihood Sequence Detection of Multiple Antenna Systems over Dispersive Channels via Sphere Decoding, Haris Vikalo and Babak Hassibi Volume 2002 (2002), Issue 5, Pages 525-531
Linear Equalization Combined with Multiple Symbol Decision Feedback Detection for Differential Space-Time Modulation, Genyuan Wang, Aijun Song, and Xiang-Gen Xia Volume 2002 (2002), Issue 5, Pages 532-537 EURASIP Journal on Applied Signal Processing 2002:5, 445–446 c 2002 Hindawi Publishing Corporation
Editorial
Dirk Slock Mobile Communication Department, EURECOM Institute, 2229 route des Cretes, BP 193, 06904 Sophia Antipolis Cedex, France Email: [email protected]
Vahid Tarokh Department of EECS MIT, Cambridge, MA 02139, USA Email: [email protected]
Xiang-Gen Xia Department of ECE, University of Delaware, Newark, DE 19716, USA Email: [email protected]
This is the second part of the special issue “Space-Time Cod- of blind identification of a convolutive MIMO system with ing and Its Applications.” In this part, there are ten papers more inputs than outputs. It considers the problem in the covering capacity of space-time coded systems, space-time frequency domain where, for each frequency, it constructs code designs, decoding methods for space-time coded trans- two tensors based on cross-polyspectra of the output. In- missions, and MIMO systems. novative solutions are proposed to resolve frequency depen- The first paper by C. B. Papadias and G. J. Foschini is dent scaling and permutation ambiguities. The paper by H. in the area of capacity issues of space-time coded MIMO A. C¸ ırpan, E. Panayırcı, and E. C¸ ekli considers the problem systems. This paper considers some capacity issues of some of blind estimation of space-time coded signals along with space-time coded systems. It proposes attainable capaci- the channel parameters. In this paper, both conditional and ties that mean the capacities achieved by different tech- unconditional maximum likelihood approaches are devel- niques with the use of progressively stronger known encod- oped and iterative solutions are proposed. The paper by K. ing/decoding techniques. F. Lee and D. B. Williams considers space-time coded or- The second three papers are in the area of space-time thogonal frequency division multiplexing (OFDM) systems code designs. The paper by W. Firmanto, B. Vucetic, J. Yuan, with multi-transmit antennas. In this paper, a low complex- and Z. Chen presents a design of space-time turbo trellis ity, bandwidth efficient, pilot-symbol-assisted channel esti- coded modulation by proposing a new recursive space-time mator for multi-transmit antenna OFDM systems is pro- trellis coded modulation. The proposed scheme is less than posed. 3 dB away from the theoretical capacity bound for MIMO The final three papers are in the area of decoding/ channels. The paper by H.-J. Su and E. Geraniotis consid- demodulation of space-time coded systems. The paper by ers some detailed design issues and tradeoffsofaspace-time N.Sellami,I.Fijalkow,andM.Sialapresentsalowcom- coded MIMO system. The paper by S. A. Zummo and S. A. plexity turbo-detector scheme for space-time coded fre- Al-Semari presents an 8PSK trellis space-time code design quency selective MIMO channels. The paper by H. Vikalo that is suitable for rapid fading channels. They propose two and B. Hassibi presents a sphere decoder for sequence de- approaches or their design: (i) to maximize the symbol-wise tection in multiple-antenna communication systems over Hamming distance between signals leaving from or remerg- dispersive channels. The sphere decoder provides the ML ing to the same encoder’s state; (ii) to partition a set based on sequence estimate with computational complexity compa- maximizing the sum of squared Euclidean distances and also rable to standard space-time decision-feedback equalizing the branch-wise Hamming distance. (DFE) algorithms. The paper by G. Wang, A. Song and X.- The next three papers focus on the topic of channel esti- G. Xia introduces linear equalization to reduce a convolutive mation for space-time coded systems. The paper by B. Chen, MIMO channel to a flat MIMO channel that may possibly A. P. Petropulu, and L. De Lathauwer addresses the problem be only partially known. Detection of differential space-time 446 EURASIP Journal on Applied Signal Processing modulation becomes hence feasible. Multiple symbol deci- Xiang-Gen Xia received his B.S. degree in sion feedback detection is considered for improved perfor- mathematics from Nanjing Normal Univer- mance. sity, Nanjing, China, his M.S. degree in mathematics from Nankai University, Tian- jin, China, and his Ph.D. degree in Electrical Dirk Slock Engineering from the University of Southern Vahid Tarokh California, Los Angeles, USA in 1983, 1986, Xiang-Gen Xia and 1992, respectively. He was a Lecturer at Nankai University, China during 1986–1988, a Teaching Assistant at University of Cincinnati, USA during 1988– 1990, a Research Assistant at the University of Southern California, Dirk Slock received the engineer’s degree USA during 1990–1992, and a Research Scientist at the Air Force In- from the University of Gent, Belgium in stitute of Technology during 1993–1994. He was a Senior/Research 1982. In 1984, he was awarded a Fulbright Staff Member at Hughes Research Laboratories, Malibu, California, scholarship for Stanford University, USA, during 1995–1996. In September 1996, he joined the Department where he received his M.S. in Electrical Engi- of Electrical and Computer Engineering, University of Delaware, neering, M.S. in Statistics, and Ph.D. in Elec- Newark, Delaware, USA, where he is currently an Associate Profes- trical Engineering in 1986, 1989, and 1989, sor. His current research interests include communication systems respectively. While at Stanford, he developed including equalization and coding; SAR and ISAR imaging of mov- new fast recursive least-squares algorithms ing targets, wavelet transform and multirate filterbank theory and for adaptive filtering. In 1989–1991, he was a member of the re- ff applications; time-frequency analysis and synthesis; and numeri- search sta at the Philips Research Laboratory, Belgium. In 1991, he cal analysis and inverse problems in signal/image processing. Dr. joined the Eurecom Institute where he is now Associate Professor. At Xia has over 80 refereed journal articles published, and four U.S. Eurecom, he teaches statistical signal processing and speech coding patents awarded. He is the author of the book “Modulated Coding for mobile communications. His research interests include DSP for for Intersymbol Interference Channels” (New York, Marcel Dekker, mobile communications: antenna arrays for (semi-blind) equaliza- 2000). Dr. Xia received the National Science Foundation (NSF) Fac- tion/interference cancellation and spatial division multiple access, ulty Early Career Development (CAREER) Program Award in 1997, space-time processing and audio coding. More recently, he has been the Office of Naval Research (ONR) Young Investigator Award in focusing on receiver design, downlink antenna array processing, and 1998, and the Outstanding Overseas Young Investigator Award from speech coding for third generation systems, and introducing spa- the National Nature Science Foundation of China in 2001. He also tial multiplexing in existing wireless systems. He received one best received the Outstanding Junior Faculty Award of the Engineering journal paper award from the IEEE-SP and one from EURASIP in School of the University of Delaware in 2001. He is currently an As- 1992. He is the coauthor of two IEEE-Globecom98 best student pa- sociate Editor of the IEEE Transactions on Mobile Computing, the per awards. He has been an Associate Editor for the IEEE-SP Trans- IEEE Transactions on Signal Processing and the EURASIP Journal actions. on Applied Signal Processing. He is also a Member of the Signal Vahid Tarokh received his Ph.D. degree Processing for Communications Technical Committee in the IEEE in Electrical Engineering from the Uni- Signal Processing Society. versity of Waterloo, Ontario, Canada in 1995. From August 1995 to May 1996, he was employed by the Coordinated Science Laboratory of the University of Illinois Urbana-Champaign, as a visit- ing Professor. He then joined the AT&T Labs-Research, where he was employed as a Senior Member of Technical Staff, Principal Member of Technical Staff, and the Head of the Department of Wireless Communications and Signal Processing until August 2000. In the fall of 2000, Dr. Tarokh joined the Department of Electri- cal Engineering and Computer Sciences of MIT as an Associate Professor, where he is currently employed. Dr. Tarokh received numerous awards including the 1987 Gold Tablet of the Iranian Math Society, the 1995 Governor General of Canada’s Academic Gold Medal, the 1999 IEEE Information Theory Society Prize Paper Award (jointly with A. R. Calderbank and N. Seshadri), and more recently the 2001 Alan T. Waterman Award. EURASIP Journal on Applied Signal Processing 2002:5, 447–458 c 2002 Hindawi Publishing Corporation
On the Capacity of Certain Space-Time Coding Schemes
Constantinos B. Papadias Global Wireless Systems Research Department, Bell Laboratories, Lucent Technologies, 791 Holmdel-Keyport Road, Holmdel, NJ 07733, USA Email: [email protected] Gerard J. Foschini Wireless Communications Research Department, Bell Laboratories, Lucent Technologies, 791 Holmdel-Keyport Road, Holmdel, NJ 07733, USA Email: [email protected]
Received 30 May 2001 and in revised form 22 February 2002
We take a capacity view of a number of different space-time coding (STC) schemes. While the Shannon capacity of multiple- input multiple-output (MIMO) channels has been known for a number of years now, the attainment of these capacities remains a challenging issue in many cases. The introduction of space-time coding schemes in the last 2–3 years has, however, begun paving the way towards the attainment of the promised capacities. In this work we attempt to describe what are the attainable information rates of certain STC schemes, by quantifying their inherent capacity penalties. The obtained results, which are validated for a number of typical cases, cast some interesting light on the merits and tradeoffsofdifferent techniques. Further, they point to future work needed in bridging the gap between the theoretically expected capacities and the performance of practical systems. Keywords and phrases: MIMO systems, space-time coding, Bell labs Layered Space Time (BLAST), space-time spreading (STS), channel capacity, space-time processing (STP), transmit diversity.
1. INTRODUCTION exist many cases of interest, where more research is needed in order to approach the capacities of MIMO systems in The combined use of antenna arrays and sophisticated practice. multiple-input multiple-output (MIMO) transceiver tech- In this paper, we will attempt to quantify the perfor- niques has boosted the anticipated spectral efficiencies of mance of certain STC techniques, in terms of their “at- wireless links in the last five years or so. The MIMO channel tainable” capacities. By “attainable capacities” we mean the capacity expressions derived in [1] indicate that the spectral capacities achieved by different techniques with the use efficiencies of MIMO channels can grow approximately lin- of progressively stronger known (typically SISO) encod- early with the (minimum of the) number of antennas avail- ing/decoding techniques. In other words, we will quantify the able on each side of the link. irreducible capacity penalties inherent in certain STCs, due to Similar to the case of single-input single-output (SISO) the way they process signals at the transmitter, as well as at channels, the attainment of the theoretically promised ca- the receiver. Our general framework targets STC techniques pacities in practice, has to rely on strong encoding/decoding which, when viewed end-to-end, can be broken down to a techniques. In the SISO case, it took about fifty years to number of SISO problems. This will allow the evaluation of approach closely (with the advent of Turbo codes [2]) the attainable capacities by quantifying the spectral efficiency of channel capacities predicted by Shannon [3]. In the MIMO each component SISO channel. We then present a number of case, an initial (the so-called D-BLAST) architectural super- STC techniques that fit well within our framework and de- structure was proposed in [1] that is theoretically capable rive their capacities. The evaluation of these capacities helps of achieving the channel capacity. The quest for practical not only compare some of the existing techniques, but also capacity-approaching STC techniques is however ongoing. to identify cases where further research is needed. Interestingly, it seems that some existing STC techniques The remainder of the paper is organized as follows. In [4, 5], already allow to approach closely the channel capac- Section 2, we provide our working assumptions, as well as ities in a number of cases [6, 7]. These quite rapid advance- some background on the topic. In Section 3, we define the ments are of course not unrelated to the mature state of SISO notion of decomposable STCs, as well as some other rele- encoding/decoding techniques. At the same time, there still vant features of such codes. In Section 4, we focus on a num- 448 EURASIP Journal on Applied Signal Processing
s1(k) x1(k) RF RF . . b˜(t) DEMUX . . . . STP bˆ(i) ...... ENCODE . . DECODE STM sM (k) xN (k) REMUX RF RF
Figure 1: A generic (M, N) multiple antenna system. ber of recently developed STC techniques for multiple-input The way in which we will view MIMO systems through- single-output (MISO) channels, and show what are their out the paper is the following. The specific way in which attainable capacities. In Section 5, we present similar results the operations at the transmitter and the receiver mentioned for MIMO systems. In Section 6, we show some numerical above take place, imposes a number of constraints to the results for outage capacities of MIMO Rayleigh-faded chan- problem of achieving the MIMO capacity. We refer to the nels in a number of cases of interest. Finally, in Section 7 we system that results after the imposition of these constraints present our conclusions, as well as some directions for future as an architectural “STC super-structure.” Each STC super- work. structure then admits a whole class of specific STCs, by ap- plying different types of temporal error-correction codes. Our goal will be to identify what are the capacity penalties 2. BACKGROUND AND ASSUMPTIONS inherent to these STC super-structures, due to the imposi- Figure 1 shows a generic architecture of a wireless system tion of constraints on both the transmitter and the receiver. with M transmitter and N receiver antennas. Such a system Said differently, we will attempt to quantify the Shannon ca- will be denoted in the remainder of the paper as (M, N). The pacities that are attainable in each case. continuous-time input stream b˜(t) is assumed to be carrying Now we define the notation and assumptions that will the original primitive data stream {b˜(i)} that is to be com- be used throughout the rest of the paper. After error correc- municated to the receiver. The input stream is then processed tion coding, interleaving, and demultiplexing (irrespective of by the shown DEMUX/ENCODE/STM unit, whose output the order into which these operations occur), the original bit is an ensemble of M parallel data streams, each one of which stream {b˜(i)} is converted to a number, say Q,ofencoded is separately upconverted and transmitted over the MIMO sub-streams, denoted as {b1(k)},...,{bQ(k)}. Note that the channel. number of encoded sub-streams Q will most often equal the The DEMUX/ENCODE/STM unit includes the follow- number of transmitter antennas M, however this may not ing operations: always be the case. Finally, the Q sub-streams are mapped through a spatial multiplexing operation, as shown in (1) demultiplexing; Figure 1, to the M sub-streams that are transmitted from the (2) encoding; M antennas. We denote the sub-stream transmitted from the (3) spatial multiplexing. mth antenna by {sm(k)}. We assume that the physical chan- These operations may be ordered differently and can be done nel between the mth transmitter and the nth receiver antenna in a more or less joint fashion. For example, the original is flat-faded in frequency, it can be hence represented, at bit stream may be encoded first as a whole, and then de- baseband, through the complex scalar hnm. The baseband re- multiplexed onto the M antennas. Alternatively, {b˜(i)} may ceived signal at the receiver antenna array is then represented be first demultiplexed onto a number of sub-streams, each by the following familiar (narrow-band) mixing model: one of which is afterwards separately encoded independently. x(k) = Hs(k)+n(k), (1) Either way, the encoded/demultiplexed sub-streams are then mapped through the so-called spatial multiplexer onto the where the involved quantities are defined as follows: M antennas for transmission. This mapping may be a sim- T ple 1-1 streaming of each encoded sub-stream on each an- • s(k): = [s1(k) ··· sM(k)] is the M ×1 vector snapshot tenna (such as in the original so-called V-BLAST transmis- of transmitted sub-streams, each assumed of equal 2 sion mentioned in [8]), or a more complex spatial map- variance σs ; ping. At the receiver, after the signals are received with an • H is the N × M channel matrix; antenna array, they are first converted to baseband. Then, • x(k) is the N × 1 vector of received signal snapshots; they are processed in space and time (STP), decoded, and re- • n(k) is the N × 1 vector of additive noise samples, as- multiplexed in the STP/DECODE/REMUX unit (again the sumed i.i.d. and mutually independent, each of vari- order of these operations may be arbitrary). These opera- 2 ance σn . tions attempt to recover as reliably as possible a replica of the original primitive bit stream {b˜(i)}. For the purposes of We also denote by the superscripts ∗, T, † the complex con- this paper, temporal interleaving is not explicitly accounted jugate, transpose, and Hermitian transpose, respectively, of for, but it can be easily accomodated. a scalar or matrix. The open-loop Shannon capacity of the On the Capacity of Certain Space-Time Coding Schemes 449
(M, N)flat-fadedchannelisgiven(see[1]) by the now fa- v(k) is “spatially” white, that is, E(v(k)v †(k)) = I.Weare miliar (so-called “log-det”) formula: now ready to define a number of useful properties of end-to- end linear STC super-structures. ρ † C = log det IN + HH [bps/Hz], (2) 2 M Decomposability = 2 2 We call an end-to-end linear STC super-structure fully de- where ρ Mσs /σn . This formula assumes the transmitter is constrained to communicate using i.i.d. random processes composable, when the matrix F in (4) is diagonal (possibly of equal power from each of the antennas. Later we will after a rearrangement of its entries). In this case, the orig- M × refine the context to fully accommodate an open loop chan- inal M N problem has reduced into Q spatially single- nel outage mode. As mentioned above, the capacity in (2) dimensional problems. can be only achieved with the use of strong encoding (STC) techniques. In the remainder of the paper, we will attempt to Partial decomposability quantify how much of this capacity is allowed to be attained We call, similarly, an end-to-end linear STC super-structure within certain STC super-structures. partially decomposable, when the matrix F in (4) is block- diagonal (again after a possible rearrangement of entries). In × 3. FEATURES OF STC ARCHITECTURAL other words, instead of a coupled Q Q problem, we are faced SUPER-STRUCTURES with a number of decoupled lower-dimension problems. When viewing (1), the only visible imposed constraints are Balance the equality between the powers of each sub-stream, the in- We call an end-to-end linear STC super-structure fully bal- dependent equal-power noise, and the flat channel charac- anced, when each of the Q sub-streams in (4) experiences teristic. The absence of additional constraints would allow, the same amount of interference from the other Q − 1sub- in theory, the attainable capacity of the (M, N)systemtobe streams as any other sub-stream. given by (2). The imposition of further constraints though, reflecting operations at both the transmitter and the receiver, Partial balance may reduce the capacity in (2). We call this (potentially) re- duced capacity the constrained capacity of a given architec- We call an end-to-end linear STC super-structure partially tural super-structure. balanced, when, the Q sub-streams can be arranged in groups At the receiver, the received encoded vector signal x(k) of sub-streams of dimension lower than Q, such that each is processed in order to produce attempted replicas of group experiences the same amount of interference from the the Q encoded sub-streams. These replicas, denoted as other groups as any other group of sub-streams. {d1(k)},...,{dQ(k)}, are then driven to the (joint or dis- The features defined above, as well as some other that will joint) decoder/de-interleaver, which will attempt to recover be discussed later, will help us classify different STC super- the original uncoded sub-streams, and eventually, the origi- structures, regarding their ability to attain their respective nal primitive bit stream. Leaving out the encoding/decoding capacities. More precisely, they affect the way in which er- stages, we can take an end-to-end view that relates the en- ror correction coding can be embedded in them, so as to coded sub-streams at the transmitter {b1(k)},...,{bQ(k)} approach these capacities. We will now see how some of to their processed attempted (soft) replicas at the receiver these properties and features are reflected into some particu- {d1(k)},...,{dQ(k)}. Quite often, these relate in a linear lar STC super-structures. fashion, that is, according to the following model: 4. (M, 1) SYSTEMS d(k) = Fb(k)+v(k), (3) Inthissection,weconsidersomerepresentativeSTCsuper- where F is a square (Q × Q) matrix and all vectors in (3)are structures that were developed for cases of multiple-input of dimension Q × 1. We will refer to STC super-structures single-output (M, 1) systems. Due to the fact that MISO that admit the end-to-end representation in (3)asend-to- antenna systems provide diversity-type gains, which are, at end linear. Further, depending on the specific structure and most, logarithmic in M, (as opposed to the linear capacity in- attributes of the mixing matrix F and the noise impairment crease of true MIMO systems), they are usually called “trans- v(k), we can define some extra attributes. Before describ- mit diversity systems.” The few techniques that will be shown ing these attributes, we define, for convenience, a noise-pre- are examples that fit well within the framework defined in whitened version of model (3). Denoting by Rv the covari- Section 3, and as such, allow for an analytical evaluation of † ance matrix of v(k)(i.e.,Rv = E(v(k)v (k))), assumed full their theoretical constrained capacities. Before proceeding, rank, an equivalent representation of (3)is we mention the open-loop capacity of flat-faded (M, 1) sys- tem, which is given by = d (k) F b(k)+v (k), (4) ρ M = Φ−1 = Φ−1 = Φ−1 = max = 2 where d (k) v d(k), F v F, v (k) v v(k), and Rv CM,1 log2 1+ hm (5) † † M m=1 E(v(k)v (k)) = ΦvΦv. Note that the new noise impairment 450 EURASIP Journal on Applied Signal Processing which is obtained by substituting, in (2), N = 1andH = (1) fully decomposable to two (1, 1) systems; [h1 ··· hM]. (2) fully balanced.
4.1. (2, 1) systems: the Alamouti scheme The total constrained capacity of the system equals the sum of the capacities of the two SISO systems (recall that each An ingenious transmit diversity scheme for the (2, 1) case was introduced a few years ago by Alamouti [4], and remains to SISO system operates at half the original information rate): date the most popular scheme for (2, 1) systems. We denote A ρ 2 2 by S the 2 × 2 matrix whose ( ) element is the encoded sig- C = log 1+ h1 + h2 . (11) i, j 2,1 2 2 nal going out of the jth antenna at odd (i = 1) or even (i = 2) time periods (the length of each time period equals the du- Note that, by contrasting (11)to(5), we see that ration of one encoded symbol). In other words, one could A = max think of the vertical dimension of S as representing “time” C2,1 C2,1 . (12) and of its horizontal dimension as representing “space.” The Alamouti scheme transmits the following signal every two This result is summarized in the following theorem. encoded symbol periods: Theorem 1. The (2, 1) Alamouti transmit diversity scheme has b1(k) b2(k) a constrained capacity equal to the (2, 1) open-loop channel ca- S( ) = s ( ) s ( ) = (6) 2 k 1 k 2 k ∗ − ∗ . pacity. b2(k) b1(k) Notice that in this case, Q = M = 2. Notice also that the Moreover, since each component of v (k) is a stationary spatial multiplexing is done according to a block scheme, the noise process, the capacity of each (1, 1) system is attain- block length being equal to L = 2 time periods. Having as- able through conventional (i.e., spatially single-dimensional) sumed, as noted earlier, the channel to be flat in frequency, state-of-the-art encoding techniques. For example, each of the two sub-streams can be encoded independently with a the (2, 1) channel is characterized through H = [h1 h2]. The baseband signal arriving at the single receiver antenna at two Turbo code, which is suitable for the classical additive white consecutive time instants can be expressed as Gaussian noise channel (with stationary noise). ∗ 4.2. A (4, 1) scheme r(k) = h1 b1(k)c1 + b (k)c2 2 (7) − ∗ The nice property of the full log-det capacity being attain- + h2 b2(k)c1 b1(k)c2 + n(k), able in the (2, 1) case does not unfortunately hold in general for (M, 1) systems with M>2 (see, e.g., [10, 11]). However, where cT = [1 | 0], cT = [0 | 1].1 After sub-sampling at 1 2 some schemes have been developed recently for special cases. the receiver and complex-conjugating the second output, we In the following, we describe a scheme that we recently de- obtain rived for the (4, 1) case (see [12]) and evaluate its capacity = T = constrained on different receiver processing options. d1(k) c1 r(k) h1b1(k)+h2b2(k) + v1(k), (8) The original information sequence b˜(i)isfirstdemul- ∗ = T ∗ = − ∗ ∗ d2(k) c2 r (k) h2b1(k) h1b2(k) + v2(k), tiplexed into four sub-streams bm(k)(m = 1,...,4). The 4-dimensional transmitted signal is now organized in blocks = T = where vm(k) cmn(k), m 1, 2. Equation (8)canbeequiva- of L = 4 (encoded) symbol periods, it is hence represented by lently written as a4× 4matrixS, which is arranged as follows: = h1 h2 b1(k) = b1 b2 b3 b4 d(k) ∗ ∗ + v(k) Hb(k)+v(k), (9) ∗ ∗ ∗ ∗ −h h b2(k) b −b b −b 2 1 S = 2 1 4 3 , (13) b3 −b4 −b1 b2 where vT ( ) = ( ) ( ) and H is a unitary matrix. After ∗ ∗ ∗ ∗ k v1 k v2 k b b −b b match-filtering to H,weobtain 4 3 2 1 where we have dropped the time index k for convenience. d(k) = H†d(k) The channel matrix is again assumed flat-faded, it can be 2 2 (10) hence represented by = . The received sig- = h1 + h2 0 H h1 h2 h2 h4 2 2 b(k)+v (k), nal will then be given by 0 h1 + h2 h1 where v (k) remains spatially white. Comparing to (3), it is h2 clear that this (2, 1) system is r = S + n = Sh + n, (14) h3 h4 1 By suitably redefining c1 and c2, the scheme can be modified for use with direct-sequence CDMA systems, where it is referred to as space-time spreading (STS) [9]. 2This result has been previously reported in [6, 7]. On the Capacity of Certain Space-Time Coding Schemes 451
T where r = [x(1) x(2) x(3) x(4)] contains 4-symbol snap- models above share the same 2×2 channel matrix ∆2 and have shots at the received signal. By complex-conjugating the sec- identically distributed, but statistically independent, 2×1ad- ond and the fourth entry of r in (14), we obtain ditive noise vectors. In order to facilitate the capacity eval- uation of this scheme, we present at this point the noise- r(1) h1 h2 h3 h4 b1 prewhitened version of (20): ∗ ∗ ∗ ∗ ∗ r (2) −h h −h h b2 r = = 2 1 4 3 + n , (15) − − r b1 n r(3) h3 h4 h1 h2 b3 1 = Λ + 1 , (22) ∗ ∗ ∗ ∗ ∗ − − r b3 n r (4) h4 h3 h2 h1 b4 3 3 where n is similarly obtained from n by complex- where conjugating its second and fourth entry. The received signal λκ is hence written as Λ = (23) −κλ = r Hb + n , (16) ∆ = Λ†Λ with 2 and where n1 and n3 are i.i.d. and mutually 2 where now H is defined as independent Gaussian variables of variance σn each. Again, { } an identical signal model to (22) holds for the pair b4,b2 . h h h h Similar to γ and α in (21), λ and κ in (23) are real and imagi- 1 2 3 4 −h∗ h∗ −h∗ h∗ nary, respectively. Similarly, the nonzero value of κ represents H = 2 1 4 3 . (17) = −h h h −h mutual interference√ between the two sub-streams (if α 0, 3 4 1 2 = = − ∗ − ∗ ∗ ∗ then λ γ and κ 0). h4 h3 h2 h1 Maximum allowable Shannon capacity We now perform matched filtering with respect to H We first compute the Shannon capacity constrained only γ 0 α 0 upon transmitter processing. By considering the two 2 × 2 † 0 γ 0 −α † models that describe the post-matched-filtering signals ac- r = H r = b + H n = ∆ b + n , (18) mf −α 0 γ 0 4 mf cording to (22), we deduce that the maximum achievable 0 α 0 γ capacity of a (4, 1) system within the space-time spreading scheme (13)isgivenby where 1 P † C constr,max = log det I + T ΛΛ , (24) 4 4,1 2 2 2 2 = † = 2 = ∗ ∗ 4σn γ h h hm ,α2jIm h1h3 + h4h2 . (19) m=1 = 2 where PT 4σb is the total average transmitted power from 2 ΛΛ† = The parameter α expresses some residual interference inher- the antenna array (σ is the variance of each bi). Since ∆ b ent in this (4, 1) technique, and is in general nonzero. Note 2, this gives both the particular sparse structure of the matrix ∆ ,aswell 4 1 ρ as the fact that γ is real and α is imaginary. These result in ∆ constr,max = ∆ 4 C4 1 log2 det I2 + 2 , (25) 2 2 2 , 2 4 being in general full rank (det(∆4) = (γ + α ) ). Comparing (18)to(3), we observe that this (4, 1) scheme is: = 2 where ρ PT /σn (all bandwidth-related normalizations (1) partially decomposable to two uncoupled (2, 2) sys- have been taken into account, so that (25) represents the tems; total capacity of the system). If the interference caused (2) fully balanced. by the quantity α vanished, the expression in (25)would reduce to Namely, by grouping the entries of rmf in two pairs, we obtain ργ Copt = log 1+ (26) 4,1 2 4 rmf,1 b1 nmf,1 = ∆2 + , rmf,3 b3 nmf,3 which is the open-loop capacity of the (4, 1) flat-faded (20) = constr,max system. However, for α 0, C4,1 falls short of r b n opt mf,4 = ∆ 4 + mf,4 , C . r 2 b n 4,1 mf,2 2 mf,2 As mentioned in Section 3, the constrained capaci- where ties that correspond to each scheme depend not only on the constraints imposed at the transmitter, but also on γα those caused by receiver processing. In the following, we ∆ = (21) 2 −αγ will describe a number of options for receiver process- ing, which will each correspond to a different constrained ∆† = ∆ capacity. (we note in passing that 2 2). The two (2, 2) signal 452 EURASIP Journal on Applied Signal Processing