WO2001019013A1 - Turbo detection of space-time codes - Google Patents

Abstract

Communication systems which employ multiple transmit and receive antenna-element arrays. Data streams for transmission may be interleaved among the transmit antenna elements in order to reduce decision errors. Turbo processing of equalizer output from a number of layers in a layered space-time processing architecture may be employed to reduce decision errors. Additionally, space-time equalization may be performed to maximize signal to noise ratio such as via minimum mean square error processing, rather than zero forcing, in order to achieve the Shannon limit, reduce multi-path effects and/or reduce intersymbol interference. Moreover, the receiver can select number and/or identity of receive antenna elements from among a larger group in order to optimize performance of the system.

Description

1/19013 ^rL '

TURBO DETECTION OF SPACE-TIME CODES

RELATED APPLICATIONS

This document claims priority to and incorporates by reference copending provisional USSN 60 / 152,982 entitled "Turbo Space-Time Processing to Improve Wireless Channel Capacity" filed on September 9, 1999.

FIELD OF INVENTION

The present invention relates to systems and processes for radio communications using multiple-element antenna array technology.

BACKGROUND

Turbo processing and space-time equalization are terms that comprehend several conventional ways to increase wireless channel capacity. Generally, turbo coding and/or processing refers to techniques aimed at approaching the Shannon limit in a channel, while space-time processing refers to techniques for processing signals from multi-element antenna arrays to exploit the multi-path nature of fading wireless environments.

European patent application no. EP 817 401 A2 published July 1 , 1998 in the name of Foschini discloses the use of a number of processing layers for space time processing of signals from multiple-receiver antenna elements. There, the transmitter feeds a number of transmitter antenna elements by cyclically apportioning segments of the modulated encoded stream of data to transmitter antenna elements. At the receiver, a number of receiver antenna elements are coupled to a number of processing layers, in order to perform the space-time processing. Signal components received during respective periods of time over a plurality of the receive antenna elements are formed into respective space and time relationships in which space is associated with respective transmitter antenna elements. Preprocessing occurs so that a collection of signal components having the same space-time relationship

-l- 1/19013 r^ ^{i /}u

forms a signal vector such that particular decoded signal contributions can be subtracted from the signal vector while particular undecoded contributions can be nulled out of the signal vector. The resulting vector is then supplied to a decoder for decoding to reform the data stream. Such conventional systems and techniques are further described in documents referred to in the "Detailed Description" section of this document.

SUMMARY OF THE INVENTION

Systems and processes according to the present invention employ a number of transmitter antenna elements and a number of receiver antenna elements coupled to multiple space-time processing layers in the receiver. In the present invention, however, portions of the information stream being communicated can be interleaved among transmitter antenna elements such as on a random or pseudo random basis; among other things, such interleaving decreases decision errors in the space-time equalization process. Furthermore, each processing layer preferably includes turbo processing in order to feed soft decisions about information being processed back to the equalizers. Moreover, space-time equalization processes according to the present invention preferably seek to maximize signal to noise ratio rather than zero forcing, as well as reduce multi-path effects and intersymbol interference. A preferred process uses minimum mean square error processing which allows the Shannon limit actually to be achieved. Furthermore, systems and processes according to the present invention preferably allow selection of the number and identity of receiver antenna elements to which the receiver may be coupled in order to optimize performance.

According to one embodiment of the invention, an information source is coupled to provide a plurality of data streams to a plurality of transmit antennas, via, for each stream, an encoder, interleaver and symbol mapper. On the receiver side, a plurality of M receiver elements are coupled to a plurality of processing layers. The number of receiver antenna elements M is preferably greater than or equal to the number N of transmit antenna elements, since equalization according to the present invention does not require an extra degree of freedom. The M receiver antenna elements are coupled to the first processing layer by coupling to a space-time equalizer which preferably applies minimum mean square error processing in order to maximize signal to noise ratio. The output of the equalizer is applied to a deinterleaver, after which the deinterleaved stream is supplied to a decoder in the layer. Output of the decoder is provided for output common with the output from the other decoders in the other layers. Preferably, each layer also includes an interleaver which receives output from the decoder and deinterleaver and supplies its interleaved output back to the equalizer in the layer in order to provide soft decision making to the equalizer. In successive processing layers, output from the decoder of the preceding layer is combined with information from the interference canceler of the layer preceding the preceding layer (except the second layer, which receives signals from an interference canceller which is coupled to the decoder of the first layer and to the receive antenna elements).

According to an alternate embodiment, the deinterleaver, interleaver and decoder are shared among layers, so that the equalizer of each layer outputs to a deinterleaver common to all layers. The output of the deinterleaver may then be coupled to a decoder which again is common to all layers. An interleaver may be provided which receives output from the deinterleaver and the decoder and applies it to each equalizer for soft decisions to be applied to the equalizers.

Accordingly, components for deinterleaving, decoding and reinterleaving may be functionally located in each layer, or common to the layers. In the first case, each layer below the first layer processes signals from an interference canceller which receives signals from a decoder in the preceding layer and from the antenna elements (in the case of the second layer) or the interference canceller in the next-preceding layer (in the case of other layers). In the second case, each layer below the first processes signals from an interference canceller which receives signals from the equalizer in the preceding layer and from the antenna elements (in the case of the second layer) or the interference canceller in the next preceding layer (in the case of other layers). Such turbo processing architectures can be used in connection with layered space-time equalization which relies on zero forcing rather than minimum square error processing. They can also be used in multi array systems in which the data streams are periodically cycled rather than interleaved.

It is accordingly an object of the present invention to provide improved layer space-time processing for communication systems which employ turbo processing techniques in order, among other things, to reduce decision errors.

It is an additional object of the present invention to provide layered space-time processing for communication systems which seeks to maximize signal to noise ratio, thereby better addressing the Shannon limit, and which also addresses mulit-path effects and / or intersymbol interference.

It is an additional object of the present invention to provide processing for communication systems in which data streams may be interleaved rather than periodically cycled among transmit antenna elements, in order, among other things, to reduce decision errors.

It is an additional object of the present invention to provide layered space-time processing for communication systems in which a receiver can select a set of antenna elements, including their number and / or identity, from among a larger group of antenna elements in order to optimize performance of the system.

Other objects, features, and advantages of present invention will become apparent with respect to the remainder of this document. BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1(a) is a schematic diagram showing a first embodiment of communications systems according to the present invention.

Fig. 1(b) is a schematic diagram showing a second embodiment of communications systems according to the present invention.

Fig. 2 is schematic diagram showing one form of space-time processing according to the present invention. Figs. 3(a) and 3(b) are diagrams which compare performance between two coding schemes according to the present invention.

Fig. 4 is a diagram which shows different capacity bounds for processing according to the present invention over a flat Rayleigh fading channel.

Fig. 5 is a diagram which shows simulation results for a system according to the first embodiment of the present invention with two transmit and two receive antenna elements.

Fig. 6 is a diagram which shows simulation results for a system according to the first embodiment of the present invention with two transmit and four receive antenna elements.

Fig. 7 is a diagram which shows simulation results for a system according to the first embodiment of the present invention with two, four and eight receive antenna elements, using soft decisions and six turbo iterations.

Fig. 8 is a diagram which shows simulation results for a system according to the second embodiment of the present invention with two, four and eight receive antenna elements, using soft decisions and two turbo iterations.

Figs. 9(a) and 9(b) are diagrams which show simulated performance of the first embodiment of the present invention using soft decisions and four antenna elements for typical urban and hilly terrain profiles.

DETAILED DESCRIPTION

The documents and references cited in the following disclosure are incorporated herein by this reference.

Abstract — By deriving a generalized Shannon capacity formula The basic information theory results reported by Foschini and for multiple-input, multiple-output Rayleigh fading channels, and Gans [ 1 ] have promised extremely high spectral efficiencies by suggesting a layered space-time architecture concept that attains a tight lower bound on the capacity achievable, Foschini has possible through multiple-element antenna array technology. shown a potential enormous increase in the information capacity In high scatteπng wireless environments (e g., troposcatter, of a wireless system employing multiple-element antenna arrays cellular, and indoor radio), the use of multiple spatially sepaat both the transmitter and receiver. The layered space-time rated and/or differently polaπzed antennas at the receiver has architecture allows signal processing complexity to grow linearly, been very effective in providing diversity agamst fading [2], rather than exponentially, with the promised capacity increase. [3]. Receiver diversity techniques also create signal processing This paper includes two important contributions: First, we show that Foschini 's lower bound is, in fact, the Shannon bound when the opportunities for mterference suppression and equalization output signal-to-noise ratio (SNR) of the space-time processing in (e g., [4]-[6]) However, using multiple antennas at either the each layer is represented by the corresponding "matched filter" transmitter or the receiver does not enable a significant gain in bound. This proves the optimality of the layered space-time the possible channel capacity According to [ 1 ], the Shannon concept. Second, we present an embodiment of this concept capacity for a system with 1 transmit and N receive antennas for a coded system operating at a low average SNR and in the presence of possible intersymbol interference. This embodiment scales only logarithmically with N. as .V — <• oc. For a system utilizes the already advanced space-time filtering, coding and using .V transmit and 1 receive antennas, asymptotically turbo processing techniques to provide yet a practical solution there is no additional capacity to be gained, assuming that the to the processing needed. Performance results are provided for transmit power is divided equally among the N antennas. quasi-static Rayleigh fading channels with no channel estimation Foschini and Gans [1 ] have shown that the asymptotic errors. We see for the first time that the Shannon capacity for wireless communications can be both increased by N times (where capacity of multiple-input, multiple-output (MIMO) Rayleigh N is the number of the antenna elements at the transmitter fading channels grows, instead, linearly with .V when N and receiver) and achieved within about 3 dB in average SNR, antennas are used at both the transmitter and the receiver. about 2 dB of which is a loss due to the practical coding scheme Furthermore, in [7], Foschini suggested a layered space-time we assume — the layered space-time processing itself is nearly architecture concept that can attain a tight lower bound on the information-lossless! capacity achievable. In this layered space-time architecture,

Index Terms — Equalization, interference suppression, space- .V information bit streams are transmitted simultaneously time processing, turbo processing. (in the same frequency band) using .V diversity antennas. The receiver uses another N diversity antennas to decouple

I INTRODUCTION and detect the .V transmitted signals, one signal at a time. The decoupling process in each of the ,V processing "layers"

T URBO" and "space-time" are two of the most explored involves a combination of nulling out the mterference from concepts in modem-day communication theory and yet undetected signals ( ^r diversity antennas can null up to wireless research From a communication theoπst's viewpoint, iV — 1 mterferers, regardless of the angles-of-arπval [5]) and "'turbo" coding/processing is a way to approach the Shannon canceling out the interference from already detected signals. limit on channel capacity, while "space-time" processing is One very significant aspect of this architecture is that it a way to increase the possible capacity by exploiting the πch allows an N -dimensional signal processing problem — which multipath nature of fading wireless environments We will see would otherwise be solvable only through multiuser detection through a specific embodiment in this paper that combining the methods [8] with complexity (m is the signal constellation two concepts provides even a practical way to both increase size) — to be solved with only N similar 1-D processing steps. and approach the possible wireless channel capacity. Namely, the processing complexity grows only linearly with

With growing bit rate demand in wireless communications, the promised capacity it is especially important to use the spectral resource efficiently

This paper includes two important contributions. First, we show that Foschini's lower bound is. in fact, the Shannon bound when the output SNR of the space-time processing in

Paper approved bv K B Letaief. the Editor for Wireless Svstems ot the IEEE Communications Society Manuscπpt received September 15 1999 revised Deeach layer is represented by the corresponding "matched filter" cember i 1999 This paper was presented at the IEEE International Coπterence bound [6], i e . the maximum SNR achievable in a hypothetical on Communications New Orleans. LA June 2000

The author is with the Home Wireless Networks Norcross GA 30071 USA situation where the array processing weights to suppress the (e-mail lek@homewιretess com) remaining interference in each layer are chosen to maximize the

Publisher Item Identifier S 0090-6778(00)071 1 1 7 output signal-to-interference-plus-noise ratio and any possible intersymbol interference ( ISI) is assumed to be completely able using 1-D processing and coding techniques that are aleliminated by some means ot equalization The 'matched filter" ready practical and "legacy-compatible" with the EDGE stanbound has been shown to be approachable using minimum dard, e g , the use of bit-interleaved 8-ary phase-shift keying mean-square error (MMSE) space-time filtering techniques (8-PS ) with rate- 1/3 convolutional coding and an equalizer [6] ' By showing the equivalence ot the generalized Foschim's w ith a similar length and structure bound and the Shannon bound, we essentially prove the A slightly different lavered space-time approach based on optimahtv of the layered space-time concept space-time coding [23]. [24j has been studied in [25]. Although

Second, we present an embodiment of Foschini 's lavered it is difficult to make a general compaπson. we will see later that space-time concept for a coded system operating at a low avour coded layered space-time approach does by far outperform erage SNR and in the presence ot unavoidable ISI Previously, the results reported in [25] for .V = 4 and N = 8. On the a different embodiment has been provided in [9] for an uncoded other hand, for N = 2, space-time coded quaternary phase-shift system with vaπable signal constellation sizes, operatmg at keying (QPSK) without layered processing appears to be the a high average SNR without ISI. Adding coding redundancy best known technique tor achieving a spectral efficiency of 2 might, at first, seem conflicting with the desire to increase the bps Hz channel bit rate Our justification is as follows- First, we seek This paper is organized as follows Section II provides a to enhance the channel capacity from a system perspective. bπef review of Foschim's layered space-time concept. Section We use "noise^" in SNR to represent all system impairments, III descπbes the two coded layered space-time architectures including thermal noise and multiuser interference. The ability and presents a capacity analysis which reveals the equivalence to operate at low SNR's means that more users per unit area of a generalized Foschim's lower bound formula and the true can occupy the same bandwidth simultaneously Second, we capacity bound Section IV provides details on the array proanticipate the use of adaptive-rate coding schemes to permit cessing, equalization, and iterative MAP techniques. Section V different degrees ot error protection according to the channel presents performance results. A summary and conclusions are SNR's. Incremental redundancy transmission [ 10], currently given in Section VI being considered for the Enhance Data Services for GSM Evolution (EDGE, GSM stands for Global System for Mobile II. BACKGROUND THEORY Communications) standard, is an efficient way to implement We bπefly review the theory behind Foschim's layered adaptive code rates without requiπng channel SNR monitoring. space-time concept. The generalized Shannon capacity for a With such adaptive-rate coding, the system does not "waste" MIMO Rayleigh fading system with N transmit and M receive spectral resources under good channel conditions. antennas is given in [ 1 ] as

Meanwhile, the iterative processing pπnciple used in turbo and seπal concatenated coding [ 1 1 ]-[ 15] has been successfully C = log₂ [det (/ + ##')] (1) applied to a wide vaπety of joint detection and decoding problems. One such application is the so-called "turbo equalization" where H is an M x N matπx. the (i. j)ιh element of which [ 16]— [ 19], where successive maximum a posteriori (MAP) is the normalized channel transfer function of the transmission processing is performed by the equalizer and channel decoder link between the ₇th transmit antenna and the nh receive anto provide a priori information about the transmit sequence tenna. / is the M x M identity matπx. p is the average SNR to one another. Similar to the layered space-time concept, per receive antenna, and det( )and superscript t denote deterturbo processing allows a multi-dimensional (fu'o-dimensional minant and con_jugate transpose. It is assumed that the transmit in this case) problem to be optimally solved with successive power is equally divided among the N transmit antennas. The 1 -D processing steps without much performance penalty. In normalization of the channel transfer function is done such that this paper, we apply the turbo pπnciple to layered space-time the average (over Rayleigh fading) of its squared magnitude is processing in order to prevent decision errors produced in each equal to unity layer from catastrophically affecting the signal detection in The lower bound on capacity is provided in [ 1] as subsequent layers.

We consider two possible coded layered space-time strucC > ∑ log₂ [l + - χ² _2k] = C_F (2) tures: one applying coding across the multiple signal processing k- - \l + \ layers, and the other assuming independent coding within each layer. Similar to [ 1 ]. we assume a quasi-static random Rayleigh where

is a chi-squared random vaπable with 2k degrees of channel model, where the channel characteπstics are stationary freedom For M = within each data block, but statistically independent between different data blocks, different antennas, and. in the case of disC_F = ∑ log, [l t ],] (3) persive multipath channels, different paths The system is assumed to have similar ISI situations as in EDGE and GSM,

Since χ _k represents a fading channel with a diversity order where multipath dispersions may last up to several symbol peot k. the lower-bound capacity in ( 3) can be viewed as the sum riods [20] We show that near-capacity performance is achiev- ot the capacities of V independent channels with increasing di¬

¹ In a flat fading case MMSE arrav processing achieves exactlv the matched versity orders trom 1 to V This suggests a layered space-time tiller bound pertorniαncc approach [7] for detecting the V transmitted signals as follows In the first layer, the receiver detects a first transmitted signal H,_j(f) is the channel transfer function (not normalized) of the by nulling out mterference from ,V - 1 other transmitted signals transmission link between the zth transmit antenna and the through array processing Assuming a "zero forcing ' (ZF) con₇th receive antenna, and superscπpts * and T denote complex straint, one receive antenna is needed to completely correlate conjugate and transpose Note in (7) and (8) that we consider and subtract each interference [5] Thus, the overall process of the folded spectra H_t](f - {m/T)) and «_,(/ - (m/T)) of nulling Λ — 1 interferences leaves the receiver with .V — (Λ^* — the channel transfer function and noise power density, where 1 ) = 1 degree of freedom to provide diversity for detecting the m — —J. . J (J is finite because the signal sources are first signal, I e , a diversity order of 1 (or simply no diversity) assumed to be band-limited) This is to take into account the Once detected, the first signal is subtracted out from the received effect of excess bandwidth and symbol-rate sampling when signals on all N antennas the frequency selectivity of the channel is not symmetπcal

In the second layer, the receiver performs similar mterference around the Nyquist band edges. Even though we assume white nulling to detect a second transmitted signal. This time, since Gaussian noise, the noise power density near and outside the there are only .V - 2 remaining mterferences. the receiver afNyquist band edges actually attenuates with the receive filter fords a diversity order of 2. The detected signal is again subtransfer function. From our expeπment (assuming a square-root tracted out from the received signals provided by the first layer Nyquist filter with a 50% rolloff factor), the computed capacity

Repeating the above mterference nulling/canceling step can be underestimated by as much as 0.5 dB if this attenuation through .V layers, we see that the receiver affords an increasing is not taken into account. order of diversity from 1 to N. If the capacities achieved in individual layers can be combined in some manner, then the in. CODED LAYERED SPACE-TIME ARCHITECTURES layered space-time approach just mentioned will achieve the A Basic Concepts capacity lower bound expressed in (3) We will explore two capacity combining possibilities in the next section. We consider two coded layered space-time approaches as

Note that the capacity and capacity low bound given in shown in Fig. 1(a) and (b). In the first approach, named "LST-I" ( 1H3) are actually frequency-dependent. We here provide an (LST stands for "layered space-time"), the coded mformation explicit capacity formula for band-limited, frequency-selective bits are interleaved across the N parallel data streams xι, channels (some vanables are redefined to be consistent with x₂, , x,v, where x, denotes a sequence of complex-valued, later analytical development). transmit data symbols (e.g., 8-PS symbols). The receiver first decouples the N data streams through interference nulling/can¬

C = (log₂[det(»«-¹ )]) (4) cellation, as descnbed in Section II, then demterleaves and decodes all the JV data streams as one information block. In where, as shown in equations (5)-(8) at the bottom of the page. the second approach, "LST-II," the information is first divided Si is the frequency-domain correlation matπx of the signals into N uncoded bit sequences m, u₂, . . " T, each of which on M receive antennas, N_; (/) is the noise power density at is independently encoded, interleaved, and symbol-mapped to frequency / on the th receive antenna. T is the symbol peπod. generate one of the N parallel data streams At the receiver, the

/^■( 1/2T)

< > = T / [ } df (5)

Λ/ Transmit

Fig I Coded layered space-time architecture ( a) LST-I and (b) LST-II

.V data streams are decoupled and independently deinterleaved cient condition for nulling Λ^' - 1 interference), we only consider and decoded. The output of LST-II produces Λ^r information Λ/ = N in this study blocks at a rate of l/.V times the output rate of LST-I. Similar to [9], the underlying assumption of our layered

In Fig. 1(a) and (b), "space-time equalizer" refers to a comspace-time architecture is that the receiver can order the detecbined array processing (for interference nulling) and equalizations of iV data streams such that an undetected layer always tion function. Instead of the ZF cπteπon, we assume that the ophas the strongest received SNR In LST-I. the space-time timization of the antenna equalizer weights is based on a MMSE equalizer in each layer must provide data decisions x _\{l) (Λ decπteπon. which in general provides better performance than a notes the permutation due to layer ordeπng) to the interference ZF approach. Foschini [7] has also indicated a potential percanceller, since decoding cannot be done until all the layers formance benefit of using MMSE (or "maximum SNR") rather are processed In LST-II. the interference cancellation in each than ZF in a layered space-time architecture. Although we show laver can use more reliable data decisions u _{\(l )} provided by \l receive antennas in Fig. 1 (a) and ( b) (Λ/ > V is the suffi- the decoder Thus. LST-I ts more prone to decision errors than LST-II In order to minimize the effects of decision errors, and also to improve the joint detection/decoding performance in general, we assume the use of turbo processing m our layered space-time architecture As shown in Fig 1(a) and (b), the space-time equalizers and the decoders provide extrinsic soft information to one another by subtracting the received soft information from the newly computed soft information. Details on MMSE space-time equalization and turbo processing will be provided m the Section IV

B Capacity Analysis Fig 2 Space-time DDFSE with MAP processing

Without getting into the detail of all the processing functions, we first discuss the general differences between the two coded Note that ( 1 1) is an explicit formula similar to (4), it shows the layered space-time approaches. In particular, we are most interfrequency dependence of the output SNR and the mtegration of ested in the capaciry combining aspects of the two approaches capacity over the signal bandwidth. Also, we assume that the

Let SNR_fc denote the output SNR of the array processing in kt layer has k — 1 interferences. the fcth layer. First, we note that, in LST-II, the capacity of each In the process of analyzing the meaning of ( 1 1 ), we discovprocessing layer is bounded by the spectral efficiency R of the ered an identical relationship between ( 11 ) and (4) regardless of modulation and coding in each layer, e g , R = 1 for 8-PS how the layers are ordered We show the proof in the Appendix with rate 1/3 coding Thus, the total capacity of LST-II is given (this proof is valid even when M ≠ N). Thus, Foschim's lower by (similar to ( l)-(3), we wπte capacity without showing the bound (3) is actuallv the true Shannon capacity bound when frequency dependence) the output SNR of the space-time processing in each layer is represented by the corresponding ' matched filter" bound. This proves the optima ty of the layered space-time concept.

CLST-II = ∑ mιn{Λ, log₂[l + SNR*] } (9)

The capaciry analysis presented above is based on the as¬

*=ι sumption of perfect layer detection, I e., no decision errors af¬

Without layer ordeπng, it is most likely that the overall perforfecting the detection in subsequent layers In reality, LST-I is mance of LST-II will be largely influenced by the error probamore prone to decision errors than LST-II and layer ordeπng bility of the first processing layer with a diversity order of only becomes important for both schemes Our simulation results in 1 In contrast, our simulation results in Section V will show that Section V will demonstrate how decision errors affect the actual LST-II with layer ordeπng can actually achieve a diversity order performance of the two coded layered space-time approaches. of approximately N

Since coding is performed across all the processing layers in IV. SIGNAL PROCESSING FUNCTIONS LST-I, the achieved SNR in each layer will contribute to the overall layer processing performance. As Foschini [7] indicated, A Space-Time Equalization such a coding scheme should be able to achieve the capacity We consider combined array processing and equalization m lower bound in (3). Here, we provide a generalized formula of order to cope with dispersive channels A space-time equalizer, Foschi 's lower bound by removing the ZF constraint and inconsisting of a spatial/temporal whitening filter, followed by stead using SNR*. as the generalized output SNR. a decision-feedback equalizer (DFE) or maximum-likelihood sequence estimator, can suppress both ISI and dispersive

C_F = ∑ log₂[l + SNR_fc] ( 10) interference [6]. The space-time equalizer used in this study is shown in Fig 2 It consists of a linear feedforward filter W_j(f), j = 1, M, on each diversity branch, a combiner,

Reference [6] provides output SNR formulas for different symbol-rate sampler, soft-input, soft-output (SISO) MAP types of optimum space-time processors. Here, it is of great sequence estimator, and synchronous iinear feedback filter interest to express the capacity lower bound using the best perE ' (f). The feedforward filters { W_j (f) \ are shown as conformance achievable In the following equation, we represent tinuous-time filters, but they can be implemented in practice SNR*. in ( 10) by the "matched filter" bound-the maximum using fractionally-spaced tapped delay lines The combmed achievable SNR by any space-time processing receiver: use ot a sequence estimator and feedback filter after diversity combining is similar to the structure ot a delayed decision-feedc_F stF = ( ∑ iog₃[ι + r_k f)\ (I D back sequence estimator (DDFSE) [27] Thus, we refer to the =ι space-time equalizer in Fig 2 as a space-time DDFSE.^" A "space-time DFE" is a structure where the sequence estimator where I (/) is the "matched filter" bound² given by equation is replaced by a memoryless hard sheer ( 15) in Section IV-A (simply a rewπting of the result in [6])

It has been shown in [20) that a space-time DDFSE with a

^:Note thai the matched tilter bound usually refers to the integrated SNR sequence estimator memory of μ and a feedback filter of length (l\ ( / )) over the signal bandwidth (e g , [61) However in the capacity context La — μ can be optimized in a MMSE manner as if it was a we assume the beM possible way lo exploit the SNR s in all frequency components space-time DFE with a feedback filter ot length Lo- In fact. numencai results in [6] showed that an optimum space-time TABLE I DFE (with unconstrained filter lengths and no feedback decision GSM TYPICAL LRBAN (TU) CHANNEL MODEL errors) can perform within only 1-2 dB of the ideal "matched filter" bound performance Thus, in order to have a practical Path Delay ( μs) 0 0 0 2 0 5 1 6 2 3 5 0 receiver structure for layered space-time processing, we consider a space-time DDFSE with a minimum sequence estimator Path Power (dB) -3 0 00 -2 0 -6 0 -8 0 -10 0 memory, I e , μ = 1 The sequence estimator is used only to provide a trellis structure needed for turbo processing, and presumably more reliable feedback decisions than the sheer used TABLE II GSM HILLY TERRAIN (HT) CHANNEL MODEL in a space-time DFE. Details on MAP processing will be given in Section IV-B

We first provide a bπef review of the space-time filtering Path Delay (μs) 0 0 0 2 0 4 0 6 15 0 17 2 theory Based on the space-time DFE equivalent model de- scπbed above, the MMSE solution tor the feedforward filters Path Power (dB) 0 0 -2 0 -1 0 -7 0 -6 0 -12 0 {W (/) } with unconstrained length can be given using the results of [6] (see also [28]) forward filter on each branch should have the following causal and anticausal lengths to achieve near-optimum performance

W = X^HKl + B(f)) = X^Hl } +?!^{P_t. (12) i + 1 / L_c K(N - l)(p_dB/W) L » K + KN(_PdB/ W) ( 16) where where K is the channel memory, .V is used here to indicate the total number of signals, including the desired and mterfer¬

( 13) ence signals, and p _B is the average SNR in decibels. In our case, K = 5, and assume for example that the system has four w έ w_v f transmit and four receive antennas (N = 4) and the operating

- f) range of average SNR is around 5 dB ( as = 5) The required filter length, including the center tap, will be Lc + £ -ι + l ~

Wy. I f + W^'vz / + ( 14) 23.5. This is a highly impractical number, considering that four such filters are required, one per each receive antenna. Fur(15) thermore, as mentioned earlier, the optimum feedforward filters should be implemented using fractionally-spaced tapped delay

In the above equations, we assume that there are a total of k lines. If a T/2-spaced filters are used, the total number of taps signal sources and we use H_k to indicate the channel vector will be doubled. Such a space-time system with about 200 coefficients would be nearly impossible to compute in any radio link design.

Faced with such impracticality of an ideal signal processing arrangement, we proceed to consider a suboptimum option. First, we will use svmbol-spaced instead of fractionally-spaced feedforward filters In order to avoid significant performance penalties, a channel estimation-based timing recovery algo- πthm descπbed in [29] will be used to optimize the symbol timing and the decision delay of the center tap relative to the measured channel impulse response In pπnciple, such timing optimization also allows the DFE to use a feedforward filter with a shorter span than the channel memory while achieving a reasonable performance [29], [30] After expeπmenting with a number of significantly reduced filter length options, we decided on the following suboptimum space-time equalizer structure The feedforward filter on each branch has a total ot nine symbol-spaced taps, which are positioned such that L_c = L = 4 The feedback filter has a length of 8. i e .

L ₀ - 0 with the MAP processor memory μ = 1 included bles I and II). and assuming the same symbol rate of 270 833 ( in order to completely cancel postcursor ISI. B must be at kbaud (T = i 692 μs) with Nyquist filteπng (partial response least as large as the channel memory plus the number of causal signaling is used in EDGE and GSM) the ISI lasts up to five taps in the feedforward filter) The method m [29] is used to symbol periods for the hilly terrain (HT) profile in Table II Acoptimize the symbol timing and the decision delay of the center cording to the empirical filter length formulas in [6], the feed- tap as descπbed above Direct matπx inversion is used to set all the filter coefficients in a standard MMSE linear processing effect in bit-mterleaved coded modulation [36]-[38]. However, fashion [4], [26], [31 ], assuming perfect channel estimation this effect can be overcome by iterative decoding [38], which is implicit in our turbo space-time processing approach.

B Turbo Processing As noted earlier, in LST-I. the space-time equalizer in each layer must provide immediate data decisions to be used for mter¬

The turbo processing technique used- in this study is also ference cancellation. Since these decisions are not "protected" based on a standard approach — the reader is referred to the by coding, they are prone to errors. In this study, we explore πch literature [ 1 1 ]-[ 19], [21]-[22], [32]-[35] for a thorough a soft decision technique to minimize the effect of decision ertreatment of this subject. The space-time equalizer and the rors The optimum soft decision can be computed by averaging decoder both performs SISO sequence estimation to compute all the possible transmit symbols weighted by their APP's [39] the a posteriori probability ( PP) of the transmit data symbols. This sequence estimation is done using the Bahl-Cocke-Je- lmek-Raviv (BCJR) forward backward algorithm. In the Xk = ∑ sP[x_k = x\y] (18) following, we descπbe the basic principle of the iterative i€.Y detection/decoding process. where X includes all the complex-valued 8-PSK constellation

Using the BCJR algomhm. the MAP processor in the points. Since P{x_k = x\y] can be obtained along with the comspace-time equalizer with τn^μ states (m is the signal constelputation of the APP P[c \y], this soft decision approach can be lation size, e.g., = 8 for 8-PSK, and μ = 1 in our case) implemented with nearly no additional cost in complexity Simcomputes the APP P[c,t|j/] of the kt coded bit c based on ilarly, we apply the same technique to compute soft decision the observation y, where y is the equalizer output sequence outputs in LST-II. corresponding to all the data symbols in a received block (see Fig. 2), and the a priori information provided by the decoder (this is not available in the first "turbo" iteration). The logarithm V PERFORMANCE RESULTS \(c ) = log_c(P[c_k\y}) of this APP can be regarded as the sum A Performance Criteria and System Assumptions of two terms

We now present performance results of the layered space-time concepts descπbed so far. The performance mea¬

Mc_k) = λ"(c_k) + λ'(c_k) (17) sure is the block-error rate (BLER) over Rayleigh fading. The results are obtained through Monte Carlo simulation. The where λ^p{c ) = log_e(F[c*]) is the logarithm of the a prior inBLER is averaged over up to 40000 channel realizations. Each formation provided by the decoder, and A^e ( c_k ) is called the "ex- block contains 400 information bits (before coding). tπnsic" information. In each "turbo" iteration, the space-time

In comparing the performance results to channel capacity, equalizer subtracts λ^p(c ) from the newly computed value of we follow the convention of a number of previous works (e.g , λ(c_fc ) to obtain the extrinsic information λ'{c_k) [see Fig. 1(a) [9], [23]) to compare the computed BLER with the "outage caand (b)] The entire sequence {X'(c_k ) } is demterleaved and forpacity" [1 L i.e.. the probability that a specified bit rate is not warded to the decoder. supported by the channel capacity. This is a vague companson,

Similarly. the decoder computes the log-APP since the Shannon limit refers to the highest enor-free bit rate v(c_k) = \og_c {P[c_k\{X'{c_k) }]) based on the demterleaved possible for long encoded blocks but it does not specify how extnnsic information provided by the space-time equalizer, long the blocks should be. Nevertheless, such a companson and subtract λ'(c ) from it to obtain extnnsic information should still be meaningful as long as the block length and BLER υ'(c ). The extnnsic mformation is then interleaved and are specified. This is similar to the way a bit-error rate of 10^-5 forwarded to the equalizer as the new a priori information is commonly used as the "error free" reference for an additive X^p{c ) for the next "turbo" iteration. white Gaussian noise (AWGN) channel.

The interleaver considered in this study is a pseudo-random In order to assess the best performance achievable, we asinterleaver, i.e., we generate a pseudo-random permutation of sume that the channel characteπstics can be perfectly estimated numbers from 1 to /. where / is the block length, and then use at the receiver Similarly, the choices of 1-D processing and this permutation as a fixed interleaver. coding techniques are important to deliver the best possible

In combining the branch metπc obtained from the equalizer performance. We try to optimize these choices while keeping output with the branch metnc obtained from the soft input them as practical as possible. Except for the use of array provided by the decoder, the MAP processor in the space-time processing and iterative MAP algonthms, all the radio link equalizer must compute the α priori information for each 8-PS techniques assumed in this study are "legacy-compatible" with symbol x_k from the three soft inputs (X^p(c_3k ), X^p(r_{ik+ 1} ), the EDGE standard (note also that vast research interest in turbo λ^p(c. ₊ ) ) ^We assume that this is done by way of summing coding has made simplified MAP algoπthms available [21 ], the three soft inputs as if the three coded bits were transmuted [221 that are not much more complex than the conventional from independent sources (these soft inputs are actually not Viterbi algoπthm) None of these techniques are claimed independent when conditioned on the observed waveform ot the to be optimum Yet. our results indicate that near-capacity entire data burst) This is a suboptimai method, which is known performance is achievable when combining them through the to cause a ' random modulation^' performance degradation coded lavered space-time architectures which appeared to perform slightly better than other generators we tested) Both schemes assume the use of bit-interleaved 8-PSK with Gray mapping The turbo coding scheme has an additional interleaver within the encoder which uses another pseudo-randomly generated permutation The receiver structure is consistent with what we have descπbed so far Note, however that we assume a minimum number of filter taps (only one feedforward tap per branch and no feedback filter) whenever there is no delay spread assumed, although the MAP processor in the DDFSE is always used for iterative detection decoding as de- scnbed in Section IV B For the turbo coding scheme, "one iteration ' means a full cvcle of three processes 1 ) MAP processing in DDFSE 2) turbo decoding by the first decoder, and 3) turbo decoding by the second decoder

Fig 3(a) shows the performance of the two coding schemes in an AWGN channel First we note that the performance of convolutional cod g also benefits from iterative processing This is due to the suboptimal nature of the decoding scheme, l e the 'random modulation effect descπbed earlier, which can be improved through iterative decoding Fig 3(a) shows that most of the improvement is achieved within two decoding cycles For turbo coding, the performance still improves even after five iterations, but saturates quickly after ten iterations At 10^-3 BLER (approximately equivalent to 10^"° bit-enor rate), turbo coding outperforms convolutional coding by about 2 2 dB, and the required SNR is within only 2 4 dB of the 0-dB Shannon limit for a spectral efficiency of 1 bps/Hz (8-PSK with rate- 1/3 codmg)

However, when we look at the average BLER performance

Average SNR (dB) over quasi-static flat fading channels in Fig 3(b), the benefit

(b) of turbo coding (with ten iterations) over convolutional coding (with two iterations) is reduced to only about 05 dB at any value

Fig 3 Performance of bn-uiterleaved 8 PSK with rate- 1/3 convolutional and of the average SNR and for all the assumed numbers of receive turbo coding over (a) AWGN channel and (b) quasi-static flat Rayleigh fading channels with V receive diversity antennas diversity antennas This is not surpπsing for two reasons First it is well known that the average BLER is determined mostly by the probability of fading events that results in high BLER s

B Choosing the Coding Scheme If we look at the relative performance at a BLER of, say above

We consider a bit-interleaved coded modulation scheme 10% in Fig 3(a) the difference between the two coding schemes using 8-PSK with Gray mapping and rate- 1/3 coding is indeed less than 1 dB Second the performance of convoluSquare-root Nyquist filtering with 30% rolloff is assumed at the tional coding over fading channels is already within about 2 dB transmitter and receiver Bit-interleaved coded modulation has of the capacity bound — the capacity bound in this case is debeen shown [36], [37] to outperform traditional trellis-coded fined as the probability that the combined output SNR of all modulation in fast fading channels (where time diversity can diversity branches is below the 0-dB Shannon lirmt Thus, there be exploited through sufficient interleaving) and it can be is not much room for fuπher improvement improved upon by considering a better mapping technique Based on the fact that the performances of the two coding that permits a large Euclidean distance without sacπficmg the schemes are quite similar in quasi static fading channels we maximum Hamming distance of the baseline coding scheme will only consider convolutional coding in the remainder of this [38] In this paper, though, since quasi-static fading is our paper basic assumption, the code by itself must be able to withstand deep fades In pπnciple, any code that performs well in an C Lavered Space-Time Performance AWGN channel is considered a good candidate — turbo codes We first look at the performance of the two coded lavered are among the strongest candidates that come to mind space-time approaches over a flat Rayleigh fading channel

Fig 3(a) and (b) provide a performance companson between Fig 4 shows the different capacity bounds for this channel two rate 1/3 coding schemes one using a 64-state convolutional assuming V = 2 1 and 8 where V is the number ot transmit code with (octal) generators ( d G₂ G₃ ) = ( 155 117 123) and receive antennas Again although we plot the results as (the same code as proposed for EDGE [20)) and the other using block eπor rate the capacity bound is defined as the probaa turbo code with two identical 16-state recursive encoders simbility that the specified spectral efficiency R (R = V in this ilar to the scheme oπgmally proposed by Beπou and Glavieux case) is not supported bv each of the differentlv defined channel [ 12] (the results here assume generators ( Ci G_>) = (23 31 ) capacities C denotes the Shannon capacity bound given bv (4)

Average SNR (dB) Average SNR (dB)

Ftg. . Capacity bounds for quasi-static flat Rayleigh fading channels with .V _Fig. 5. Layered space-time performance of LST-I with two transmit and two transmit and Λ^" receive antennas. C: Shannon capacity, C_F : Foschini (original) receive antennas (.V = 2 ). Quasi-static flat Rayleigh fading channel, bound, and - ₁1 : capacity bound for LST-II.

[which is equivalent to the generalized Foschini bound CF. Λ/F in ( 1 1 )], C_f denotes the original Foschini bound (with the ZF constraint) in (3), and CLST-Π is the capacity bound for LST-II in (9) (however, the results for C.sτ-11 are obtained simply by averaging the probability that R = 1 is not suppoπed by each processing layer). Note that CLST -₁₁ can indeed provide approximately a diversity order of N; this is attributed largely to the use of layer ordering as discussed earlier. Note also that all the bounds show an improvement with increasing JV. This means that the capacity actually increases more than linearly with the number of transmit and receive antennas. However, there is a diminishing improvement as N increases to a much larger number.

Fig. 5 shows the simulation results for LST-I with 2 transmit Average SNR (dB) and 2 receive antennas (i.e., N — 2). Three sets of results are provided, assuming: 1) soft decisions; 2) hard decisions; and 3) Fig. 6. Layered space-time performance of LST-I with four transmit and four coπect decisions in each layer (note that the DDFSE always uses receive antennas (.V = 4). Quasi-static flat Rayleigh fading channel. tentative decisions and provides soft outputs to the decoder). We see that, although soft decisions offer some improvernent over hard decisions, the impact of decision enors is still quite noticeable. Fig. 6 shows similar results for N = 4. Here, the impact of decision errors is not as significant as the previous results, and turbo processing and soft decisions help to reduce much of this impact. With three iterations, the effect of decision enors almost completely disappears when using soft decisions. Decision enors have a lesser effect for a larger ;V because of the greater diversity order available through anay processing and layer ordering.

In Fig. 7, we compare the results using soft decisions and six "turbo" iterations with the Shannon capacity bound. For N = 4 and 8, the performance of LST-I is within 2.5-3 dB of the capacity bound at 10% BLER (and about 3-3.5 dB at 1% BLER).

Since the BLER may vary as a function of the block size,³ it Average SNR (dB) is also important to consider the processing loss by discounting the loss due to the inefficiency of modulation and coding. As Fig. 7. Layered space-time performance of LST-I for .V = 2. 4. and 8. Soft decisions. 6 iterations. Quasi-static flat Rayleigh fading channel.

'As an example, when we double the block size, the required average SNR is 0.2-0.4 dB greater than the results shown here. However, this difference in average SNR applies uniformly to all results, with or without layered space-time shown in Fig. 3(b). there is already a gap of about 2 dB between processing the performance of our coding scheme and the Shannon limit. U over 4. Son

From the above results, we conclude that, for N = 4 and 8, LST-I outperforms LST-II by a margin of 0.5 dB (at 10% the suboptimum space-time equalizer structure we assume, as BLER) to 3 dB (at 1% BLER). For N = 2, the performance of LST-I is greatly affected by decision errors (note that, even in already discussed in Section IV-A. this case. LST-I still performs as well as LST-II at 10% BLER), whereas LST-II can reach a lower BLER at high average SNR. VI. CONCLUSION Based on these results, the layered space-time approach is not highly recommended for N = 2. As mentioned earlier, By deriving the generalized Shannon capacity formula and space-time coding is a better alternative to achieve a spectral suggesting a layered space-time architecture that attains a tight efficiency of 2 bps/Hz. For instance, a 64-state space-time lower bound on the capacity achievable. Foschini has laid a coded QPSK can perform to within 2 dB of the Shannon significant theoretical foundation for improving the wireless capacity bound [23]. channel capacity through multipie-eiement array technology. We have shown that Foschini's lower bound is actually the true Shannon bound when the output SNR of the space-time

D. Frequency-Selective Channels processing in each layer is represented by the corresponding

Finally, we present an example of performance results for fre"matched filter" bound. We then provided two coded layered quency-selective fading channels. This example assumes N = 4 space-time approaches as an embodiment of this concept. For and the use of soft decisions for both LST-I and LST-II. Fig.9(a) a large number of transmit and receive antennas, coding across and (b) show the results for the TU and HT profiles, defined in the layers provides a better performance than independent Tables I and II. Again, we only show results with two "turbo" itcoding within each layer. However, with two transmit and two erations for LST-H because little improvement can be achieved receive antennas, the former is heavily affected by decision with more iterations in this case. For both deiay profiles, the pererrors and. therefore, provides a poorer performance than the formance at 10% BLER is within 3 dB of the Shannon bound latter. for LST-I with six iterations, and within 4 dB for LST-II with The underlying coding and signal processing techniques used two iterations. At a lower BLER. the loss relative to the bound in this study are based on practical but suboptimal approaches. is greater for HT than for TU. This is due to the limitation of Yet. such subopti ality can be greatly compensated for by it- erative processing. Overall, our coded layered space-time apFurthermore, using the matrix identity [40] proaches can achieve a performance within about 3 dB of the Shannon bound at 10% BLER. about 2 dB of which is a loss (27) due to the practical coding scheme we assume. Thus, not only det(A) is the layered space-time architecture exactly what the Shannon where -4_adj is called the adjugate matrix of matrix A, we can limit has prescribed in a theoretical sense, but it also provides rewrite (26) as an attractive general methodology for improving and achieving the wireless channel capacity. dj

H •' _ = ^«d) HI

(28) det(A) ^{_ n} det(i3) l + r„(/) ^'

APPENDIX By replacing det( ) in the above equation using (21 ), we obtain

PROVING THE EQUIVALENCE OF FOSCHINI BOUND AND SHANNON CAPACITY B««i5r; det(A) n (29)

Using the mathematical induction method, we will prove that (4) and ( 11) are identical. In order to do so, we must show that fc I-Il^{(1 + r}*⁽/⁾⁾

We then show in the following that, given (21), (19) is also true diversity order, including the effects of both multiple antennas

For convenience, let = det[b_l, . ir;. ...

+ det a _r.:, _nl(f) y- s,₁K-^l and -3 = S„-ιK-¹. (25) H' , (34)

" Kt(/)

- tiot[6ι. ^{• ■ •} • *;. ^{■ ■} We can rewrite (24) as

_L ff«ι(/) det[H'_n. H'

Λ-'H = ι + r.(/) ^' (26) = <lct[6ι. • ■ Similarly, we can expand the result of (34) with respect to the [ 16] C Douillard. C. B. M. Jezequel. A. Picart. P. Didier. and A. Glavieux second column, the third column, and so on (except for the ₇th "Iterative correction of intersymbol interference: Turbo equalization. Eur Trans Telecommun . vol. 6. pp. 507-51 1. Sept. 1995 column). Eventually, we obtain [ 17] A Picart. P. Didier, and A. Glavieux. "Turbo-detection: A new approach to combat channel frequency selectivity." in Proc IEEE ICC^'97. Mon rreal. Quebec. Canada. June 1997, pp. 1498-1502. column [ 18] D Raphaeli and Y. Zarai. "Combined turbo equalization and turbo de det(j4_j ) = det [6ι . b₂, , H . b_M\ = det(B_}) coding." IEEE Commun Lett . vol. 2. pp. 107-109. Apr. 1998.

[ 19] J. Garcia-Fπas and J. D Villasenor. "Combined blind equalization anc (35) turbo decoding." in Proc IEEE ICC'99. Communication Theory Mini Conference. Vancouver. BC. Canada. June 1999. pp. 52-57. which proves (31 ). The proof of ( 19) is therefore complete. [20] S. L. Anyavisitakul. J. H. Winters, and N. R. Sollenberger. "Joint equalization and interference suppression for high data rate wireless systems,' in Proc. IEEE VTC 99. Houston. TX. May 1999. pp. 700-706.

ACKNOWLEDGMENT [21 ] P. Jung. "Novel low complexity decoder for turbo codes." Electron. Lett vol. 31. no. 2. pp. 86-87. Jan. 1995.

The author wishes to thank G. J. Foschini for reviewing the [22] A. J. Viterbi. "An intuitive justification and a simplified implementatior information theory part of the oπginal manuscπpt of this paper of the MAP decoder for convolutional codes." IEEE J. Select. Area: and suggesting several improvements, and for the enormous inCommun.. vol. 16, pp. 260-264. Feb. 1998.

(231 V. Tarokh. N. Seshadn. and A. R. Calderbank. "Space-time codes foi spiration he provided by inventing the layered space-time conhigh data rate wireless communications: Performance analysis and code cept. The author also benefited from discussions with K. R. construction," IEEE Trans Inform Theory, vol. 44, pp. 744-765. Mar Narayanan. I. Lee. and X. Li. Finally, the author thanks the 1998.

[24] A. F. Naguib. V. Tarokh. N. Seshadn. and A. R. Calderbank. "A anonymous reviewers for valuable comments and suggestions. space-time coding modem for high-data-rate wireless communications." IEEE Select Areas Commun.. vol. 16. pp. 1459—1478. Oct

REFERENCES 1998.

[251 ^v- Tarokh. A. F. Naguib. N. Seshadn. and A. R. Calderbank. "Combined

[ 11 G. J. Foschini and M. J. Gans. "On limits of wireless communication array processing and space-u e coding," IEEE Trans. Inform. Theory in a fading environment when using multiple antennas." Wireless Pers. vol. 45. pp. 1121-1 128. May 1999 Commun . vol. 6. no. 3. pp. 311-335. Mar. 1998. [26] J. G. Proakis. Digital Communications. 2nd ed. New York: McGraw-

[2] W. C. Jakes Jr.. Ed.. Microwave Mobile Communications. New York: Hill. 1989. Wiley. 1974. [27] A. Duel-Hallen and C. Heegard. "Delayed decision-feedback sequence

[31 D. C. Cox. "Universal digital portable radio communications." Proc. estimation." IEEE Trans. Commun , vol. 37. pp. 428-436. May 1989 IEEE. vol. 75. pp. 436-477. Apr. 1987. [28] S. L. Anyavisitakul and I. Lee, "The equivalence of two unified solu¬

[4] R. A. Monziπgo and T. W Miller. Introduction to Adaptive Artions for optimum space-time processing," IEEE Trans. Commun . to rays. New York: Wiley. 1980. be published.

[51 J. H. Winters. J. Salz, and R. D. Gitlm. "The impact of antenna diver[29] S. Anyavisitakul and L. J. Greenstein. "Reduced-complexity equalizasity on the capaciry of wireless communications systems." IEEE Trans. tion techniques for broadband wireless channels." IEEEJ. Select Areas Commun.. vol. 42. pp. 1740-1751. Apr. 1994. Commun.. vol. 15, pp. 5-15. Jan. 1997.

[6] S. L. Anyavisitakul. J. H. Winters, and 1. Lee, "Optimum space-time [30] P. A. Voois. I. Lee. and J. M. Cioffi. "The effect of decision delay in processors with dispersive mterference: Unified analysis and required finite-length decision feedback equalization." IEEE Trans. Inform filter span. ' IEEE Trans. Commun.. vol. 47, pp. 1073-1083. July 1999. Theory, vol. 42. pp. 618-621. Mar. 1996

[7] G. J. Foschini. "Layered space-time architecture for wireless commu[31] S. Haykin. Adaptive Filter Theory Englewood Cliffs. NJ' Prentice- nication in a fading environment when using multiple antennas." Bell Hall. 1991. Labs Tech. J.. vol. 1. no. 2. pp. 41-59. Autumn 1996. [32] L. R. Bahl. J. Cocke. F. Jelinek. and J. Raviv. "Optimal decoding of linear

[81 S. Verdu. Multiuser Detection. Cambridge. U.K.: Cambridge U iv. codes for minimizing symbol error rate." IEEE Trans. Inform. Theon, Press. 1998. vol. IT-20. pp. 284-287. Mar. 1974.

[9| G. J. Foschini. G. D. Golden. R. A. Valenzuela. and P. W Wolaniansky, [33] J. Hagenauer. E. Offer, and L. Papke. "Iterative decoding of bina . "Simplified processing for high spectral efficiency wireless communicablock and convolutional codes." IEEE Trans Inform. Theory, vol. 42 tion employing multi-element arrays." IEEE J. Select. Areas Commun.. pp. 429-445. Mar. 1996. vol. 17. pp. 1841-1852. Nov. 1999. [34] S. Benedetto. D. Divsalar. G. Montorsi. and F. Pollara. "Soft-output de[ 10) R. van Nobelen. N. Seshadn. J. Whitehead. and S. Timiπ. "An adaptive coding algoπthms for continuous decoding of parallel concatenated conradio link protocol with enhanced data rates for GSM evolution." IEEE volutional codes." in Proc IEEE ICC 96. Dallas. TX, June 1996. pp Pers Commun . vol. 6. pp. 54-63. Feb. 1999. 1 12-1 17 [ I I ] C. Berrou. A. Glavieux. and P. Thitimajshima. "Near Shannon limit [35] R. J. McEliece. D. J. C. MacKay. and J.-F. Cheng. Turbo decoding as error-correction coding and decoding: Turbo codes." in Proc. IEEE an instance of Pearl's "belief propagation^" algonthm." IEEE J Select ICC93. Geneva. Switzerland. May 1993. pp. 1064-1070. Areas Commun . vol. 16. pp 140-152. Feb. 1998 [12] C. Benou and A. Glavieux. "Near optimum error correcting coding and [36] E. Zehavi. '8-PSK trellis codes for a Rayleigh channel." IEEE Trans decoding: Turbo codes." IEEE Trans Commun . vol.44. pp. 1261-1271. Commun.. vol. 40. pp. 873-883. May 1992. Oct. 1996 [37] G. Caire. G. Taπcco. and E. Biglien. "Bit-inierleaved coded modula[ 13] S. Benedetto and G. Montorsi. "Unveiling turbo codes: Some results tion." IEEE Trans. Inform Theory, vol 44. pp. 927-946. May 1998 on parallel concatenated coding schemes." IEEE Trans Inform. Theory. [38) X. Li and J Ritcey. "Trellis-coded modulation with bn interleaving and vol. 42. pp. 409-428. Mar. 1996. iterative decoding," IEEE Trans Commun . vol. 17. pp. 715-724. Apr [ I4| S Benedetto. D Divsalar. G. Montorsi. and F Pollara. "Serial concate1999. nation of interleaved codes: Design and performance analysis." IEEE [39] S. Anyavisitakul and Y Li. Joint coding and decision feedback Trans. Inform Theory, vol. 44. pp 409-429. Apr. 1998 equalization tor broadband wireless channels." IEEE J Select Areas [ I 5| K. R. Narayanan and G. L. Stuber. "A senal concatenation approach to Commun . vol 16. pp. 1670-1678. Dec 1998 iterative demodulation and decodine." IEEE Trans Commun . vol 47. [40| G Siring. Linear Λt ehra and Its Applnaiions Orlando. FL. Harcourt pp. 956-961. July 1999 Brace Jovanovich. 1980 1/19013 ^r

According to the present invention, the receiver can select a set of antenna elements, including their number and / or identity, from among a larger group of antenna elements in order, among other things, to improve performance of the system without increasing the extent of radiofrequency circuitry. One process for selecting antenna elements is to utilize equation 4 above as a measure of quality for the particular set of antenna elements being evaluated. That evaluation can occur for each permutation or combination of antenna elements in order to select the subset with optimum performance (as determined, for instance, by selecting the subset with greatest value calculated according to equation 4). This can occur at whatever desired points in time, including periodically.

Claims

CLAIMSWhat is claimed is:

1. A radio receiver coupled to a plurality of receiver antenna elements, comprising a plurality of layers for processing signals received by the receiver elements, each layer comprising: a space-time equalizer, the space-time equalizer in a first layer of the plurality of layers coupled to each of the receiver elements, the space-time equalizer in each of the other layers of the plurality of layers coupled to an interference canceller which receives output from the layer preceding each said other layer.

2. A receiver according to claim 1 further comprising a deinterleaver coupled to each space time equalizer and to a decoder for output.

3. A radio receiver according to claim 1 in which each layer comprises its own deinterleaver and decoder, and further comprises an interleaver adapted to receive output from the decoder in said layer and from the deinterleaver in said layer, the interleaver feeding output to the equalizer in said layer, thereby allowing soft decisions about information being processed iteratively by said layer to be fed back and forth between said equalizer and said decoder in said layer.

4. A radio receiver according to claim 3 in which the interference canceller which receives output from the layer preceding each said other layer receives information from the decoder and the interference canceller in the preceding layer.

5. A radio receiver according to claim 1 in which the equalizer in each layer is connected to a common deinterleaver, which feeds a common decoder, and in which a common interleaver receives signals from the decoder, and is coupled to each equalizer in order to provide interleaved signals to the equalizer.

6. A radio receiver according to claim 5 in which the interference canceller which receives output from the layer preceding each said other layer receives signals from the equalizer and from the interference canceller in said layer preceding each said other layer.

7. A radio receiver according to claim 1 in which each space-time equalizer performs equalization using a minimum mean-square error criterion.

8. A radio receiver according to claim 1 in which the receiver is adapted to receive and process signals from a transmitter that is coupled a plurality of transmit antenna elements, the number of transmit antenna elements to which the transmitter is coupled being less than the number of receiver antenna elements to which the receiver is coupled.

9. A radio receiver according to claim 8 in which the receiver is adapted to receive and process signals from a transmitter that is coupled to N transmit antenna elements, the number of receiver antenna elements to which the receiver is coupled is M, and M is greater than N.

10. A radio receiver according to claim 9 adapted to be coupled to at least one set of M receiver antenna elements out of K available receiver antenna elements.

11. A radio receiver according to claim 1 in which the receiver is adapted to select the sequence in which information from the receiver antenna elements is to be processed. 1/19013

12. A communications system, comprising: a radio transmitter coupled to a stream of information and to a plurality of transmit antenna elements, said transmitter adapted to apportion a portion of the stream of information to each transmit antenna element by interleaving said portions of the information stream among the transmit antenna elements; and a radio receiver coupled to a plurality of receiver antenna elements, comprising a plurality of layers for processing signals received by the receiver elements, each layer comprising: a space-time equalizer, the space-time equalizer in a first layer of the plurality of layers coupled to each of the receiver elements, the space-time equalizer in each of the other layers of the plurality of layers coupled to an interference canceller which receives output from the layer preceding each said other layer.

13. A system according to claim 12 further comprising a deinterleaver coupled to each space time equalizer and to a decoder for output.

14. A system according to claim 12 in which each layer comprises its own deinterleaver and decoder, and further comprises an interleaver adapted to receive output from the decoder in said layer and from the deinterleaver in said layer, the interleaver feeding output to the equalizer in said layer, thereby allowing soft decisions about information being processed iteratively by said layer to be fed back and forth between said equalizer and said decoder in said layer.

15. A system according to claim 14 in which the interference canceller which receives output from the layer preceding each said other layer receives information from the decoder and the interference canceller in the preceding layer.

16. A system according to claim 12 in which the equalizer in each layer is connected to a common deinterleaver, which feeds a common decoder, and in which a common interleaver receives signals from the deinterleaver and decoder, and is coupled to each equalizer in order to provide interleaved signals to the equalizer.

17. A system according to claim 16 in which the interference canceller which receives output from the layer preceding each said other layer receives signals from the equalizer and from the interference canceller in said layer preceding each said other layer.

18. A system according to claim 12 in which each space-time equalizer performs equalization using a minimum mean-square error criterion.

19. A system according to claim 12 in which the number of transmit antenna elements to which the transmitter is coupled is less than the number of receiver antenna elements to which the receiver is coupled.

20. A system according to claim 12 in which the transmitter is coupled to N transmit antenna elements, the receiver is coupled to M antenna elements, and M is greater than N.

21. A system according to claim 20 in which the receiver is adapted to be coupled to at least one set of M receiver antenna elements out of K available receiver antenna elements.

22. A system according to claim 12 in which the receiver is adapted to select the sequence in which information from the receiver antenna elements is to be processed.

23. A communications system, comprising: a radio transmitter coupled to a stream of information and to a plurality of transmit antenna elements, said transmitter adapted to apportion a portion of the stream of information to each transmit antenna element by interleaving said portions of the information stream among the transmit antenna elements; and a radio receiver coupled to a plurality of receiver antenna elements, comprising a plurality of layers for processing signals received by the receiver elements, each layer comprising: a space-time equalizer, a deinterleaver and a decoder, the space-time equalizer in a first layer of the plurality of layers coupled to each of the receiver elements, the space-time equalizer in each of the other layers of the plurality of layers coupled to each of the receiver elements and to an interference canceller which receives output from the decoder in the layer preceding each said other layer, each space-time equalizer coupled to the deinterleaver in its layer, each deinterleaver coupled to the decoder in its layer; the equalizer in each layer adapted to perform minimum mean-square error equalization to signals being processed; and an output for said stream of information coupled to the decoders in each of said layers.

24. A system according to claim 23 in which the transmitter is adapted to interleave portions of the information stream among the transmit antenna elements randomly.

25. A system according to claim 23 in which the transmitter is adapted to interleave portions of the information stream among the transmit antenna elements pseudo-randomly.

26. A system according to claim 23 in which the number of transmit antenna elements to which the transmitter is coupled is less than the number of receiver antenna elements to which the receiver is coupled.

27. A system according to claim 26 in which the transmitter is coupled to N transmit antenna elements, the receiver is coupled to M antenna elements, and M is greater than N.

28. A system according to claim 27 in which the receiver is adapted to be coupled to at least one set of M receiver antenna elements out of K available receiver antenna elements.

29. A system according to claim 23 in which the receiver is adapted to select the sequence in which information from the receiver antenna elements is to be processed.

30. A process for communicating an information stream using a radio transmitter and a radio receiver, including: a. coupling a radio transmitter to the information stream and to a plurality of transmit antenna elements, and interieaving portions of the information stream among the transmit antenna elements; b. transmitting said portions of the information stream; c. coupling a radio receiver to a plurality of receiver antenna elements to receive the transmitted information stream; and d. processing the information stream in a plurality of processing layers, comprising:

(i) in a first layer, coupling the receiver antenna elements to an equalizer and space-time processing the signals from the receiver antenna elements in the equalizer; deinterleaving the output from the equalizer; decoding the deinterleaved output from the equalizer; feeding the decoded information to a common output for the information stream; and

(ii) in each successive layer, coupling to an equalizer the output from an interference canceller that is fed by output from the preceding layer and space-time equalizing said output from said interference canceller; deinterleaving the output from the equalizer; decoding the deinterleaved output from the equalizer; and feeding the decoded information to a common output for the information stream.

31. A process according to claim 30 in which steps of deinterleaving the output from the equalizer, decoding the deinterleaved output from the equalizer and feeding the decoded information to a common output for the information stream are performed inside each layer and further comprising, in each layer, reinterleaving deinterleaved and decoded output from said layer, and feeding the reinterleaved signal to the equalizer in said layer thereby allowing soft decisions about information being processed iteratively by said layer to be fed back and forth between said equalizer and said decoder in said layer.

32. A process according to claim 30 in which the equalizer in each layer performs minimum mean-square error equalization.

33. A process according to claim 30 in which said interleaving is performed randomly.

34. A process according to claim 30 in which said interleaving is performed pseudo-randomly.

35. A process according to claim 30 in which coupling of said receiver to said receiver antenna elements includes selecting a set of receiver antenna elements from a larger group of receiver antenna elements.

36. A process according to claim 30 in which coupling of said receiver to said receiver antenna elements includes coupling to a set of receiver antenna elements selected from a larger group of receiver antenna elements.

37. A process according to claim 36 in which said coupling further includes selecting the sequence in which signals from receiver antenna elements are to be processed.