CA2727876C - Efficient computation of spatial filter matrices for steering transmit diversity in a mimo communication system - Google Patents

Abstract

Techniques for efficiently computing spatial filter matrices are described. The channel response matrices for a MIMO channel may be highly correlated if the channel is relatively static over a range of transmission spans. In this case, an initial spatial filter matrix may be derived based on one channel response matrix, and a spatial filter matrix for each transmission span may be computed based on the initial spatial filter matrix and a steering matrix used for that transmission span. The channel response matrices may be partially correlated if the MIMO channel is not static but does not change abruptly. In this case, a spatial filter matrix may be derived for one transmission span and used to derive an initial spatial filter matrix for another transmission span m. A spatial filter matrix for transmission span m may be computed based on the initial spatial filter matrix, e.g., using an iterative procedure.

Description

EFFICIENT COMPUTATION OF SPATIAL FILTER MATRICES FOR
STEERING TRANSMIT DIVERSITY IN A MIMO COMMUNICATION SYSTEM
This is a Divisional of Canadian Patent Application Serial No. 2,572,591 filed June 27, 2005.
BACKGROUND
I. Field [00011 The present invention relates generally to communication, and more specifically to spatial "processing for data transmission in a multiple-input multiple-output (MEMO) communication system.
H. Background 100021 A MEMO system employs multiple (NT) transmit antennas at a transmitting entity and multiple (NR) receive antennas at a receiving entity for data transmission. A
= MEMO channel formed by the NT transmit antennas and NR receive antennas may be decomposed into Ns spatial channels, where Ns 5_ min {N.T, NR}. The Ns spatial channels may be used to transmit data in parallel to achieve higher throughput and/or redundantly to achieve greater reliability.
[00031 Each spatial channel may experience various deleterious channel conditions such as, e.g., fading, multipath, and interference effects. The Ns spatial channels may also experience different channel conditions and may achieve different signal-to-noise-and-interference ratios (SNRs). The SNR of each spatial channel determines its transmission capacity, which is typically quantified by a particular data rate that may be reliably transmitted on the spatial channel. For a time variant wireless channel, the channel conditions change over time and the SNR of each spatial channel also changes over time.
[00041 To improve performance, the MEMO system may utilize some form of feedback whereby the receiving entity evaluates the spatial channels and provides feedback information indicating the channel condition or the transmission capacity of each spatial channel. The transmitting entity may then adjust the data transmission on each spatial channel based on the feedback information. However, this feedback information may not be available for various reasons. For example, the system may not support feedback transmission from the receiving entity, or the wireless channel may change more rapidly than the rate at which the receiving entity can estimate the wireless channel and/or send back the feedback information. In any case, if the transmitting entity does not know the channel condition, then it may need to transmit data at a low rate so that the data transmission can be reliably decoded by the receiving entity even with the worst-case channel condition. The performance of such a system would be dictated by the expected worst-case channel condition, which is highly undesirable.
[0005] To improve performance (e.g., when feedback information is not available), the transmitting entity may perform spatial processing such that the data transmission does not observe the worst-case channel condition for an extended period of time, as described below. A higher data rate may then be used for the data transmission.
However, this spatial processing represents additional complexity for both the transmitting and receiving entities.
100061 There is therefore a need in the art for techniques to efficiently perform spatial processing to improve performance in a MIMO system.
SUMMARY
[0007] Techniques for efficiently computing spatial filter matrices used for spatial processing by a receiving entity are described herein. A transmitting entity may transmit data via a MIMO channel using either full channel state information ("full-CSI") or "partial-CSI" transmission, as described below. The transmitting entity may also utilize steering transmit diversity (STD) for improved performance. With STD, the transmitting entity performs spatial processing with different steering matrices so that the data transmission observes an ensemble of effective channels and is not stuck on a "bad" channel realization for an extended period of time. The receiving entity performs the complementary receiver spatial processing for either full-CSI or partial-CSI
transmission and for steering transmit diversity. The spatial filter matrices used for receiver spatial processing may be efficiently computed if the MEMO channel is relatively static or does not change abruptly.
[0008] If the MIMO channel is relatively static over a range of transmission spans (e.g., a range of symbol periods or frequency subbands), then the channel response matrices for the MIMO channel over these transmission spans may be highly correlated.
In this case, an initial spatial filter matrix may be derived based on a channel response matrix and a selected receiver processing technique, as described below. A spatial filter matrix for each transmission span within the static range may then be computed based on the initial spatial filter matrix and the steering matrix used for that transmission span.
[0009] If the MIMO channel is not static but does not change abruptly, then the channel response matrices for different transmission spans may be partially correlated. In this case, a spatial filter matrix yl_x(t) may be derived for a given transmission span t and used to derive an initial spatial filter matrix for another transmission span m. A spatial filter matrix 111,(m) for transmission span m may then be computed based on the initial spatial filter matrix, e.g., using an iterative procedure. The same processing may be repeated over a range of transmission spans of interest, so that each newly derived spatial filter matrix may be used to compute another spatial filter matrix for another transmission span.
According to one aspect of the present invention, there is provided a method of deriving spatial filter matrices in a wireless multiple-input multiple-output (MIMO) communication system, comprising: deriving a first spatial filter matrix for a first transmission span; determining a first initial spatial filter matrix for a second transmission span based on the first spatial filter matrix, and deriving a second spatial filter matrix for the second transmission span based on the first initial spatial filter matrix.
According to another aspect of the present invention, there is provided an apparatus in a wireless multiple-input multiple-output (MIMO) communication system, comprising: a processor operative to derive a first spatial filter matrix for a first transmission span, determine a first initial spatial filter matrix for a second transmission span based on the first spatial filter matrix, and derive a second spatial filter matrix for the second transmission span based on the first initial spatial filter matrix.
According to still another aspect of the present invention, there is provided an apparatus in a wireless multiple-input multiple-output (MIMO) communication system, comprising: means for deriving a first spatial filter matrix for a first transmission span; means for determining a first initial spatial filter matrix for a second transmission span based on the first spatial filter matrix; and means 3a for deriving a second spatial filter matrix for the second transmission span based on the first initial spatial filter matrix.
According to yet another aspect of the present invention, there is provided a method of deriving spatial filter matrices in a wireless multiple-input multiple-output (MIMO) communication system, comprising: deriving a first spatial filter matrix for a first transmission span, the first transmission span comprising a transmission span in time, frequency, or time and frequency; determining a first initial spatial filter matrix for a second transmission span based on the first spatial filter matrix and on at least one steering matrix, the second transmission span comprising a transmission span in time, frequency, or time and frequency, and deriving a second spatial filter matrix for the second transmission span based on the first initial spatial filter matrix.
According to a further aspect of the present invention, there is provided an apparatus in a wireless multiple-input multiple-output (MIMO) communication system, comprising: a processor operative to derive a first spatial filter matrix for a first transmission span, the first transmission span comprising a transmission span in time, frequency, or time and frequency, determine a first initial spatial filter matrix for a second transmission span based on the first spatial filter matrix and on at least one steering matrix, the second transmission span comprising a transmission span in time, frequency, or time and frequency, and derive a second spatial filter matrix for the second transmission span based on the first initial spatial filter matrix.
According to still a further aspect of the present invention, there is provided an apparatus in a wireless multiple-input multiple-output (MIMO) communication system, comprising: means for deriving a first spatial filter matrix for a first transmission span, the first transmission span comprising a transmission span in time, frequency, or time and frequency; means for determining a first initial spatial filter matrix for a second transmission span based on the first spatial filter matrix and on at least one steering matrix, the second transmission span comprising a transmission span in time, frequency, or time and frequency;
and means for deriving a second spatial filter matrix for the second transmission span based on the first initial spatial filter matrix.

3b [0010] The steering matrices may be defined such that the computation of the spatial filter matrices can be simplified. Various aspects and embodiments of the invention are described in further detail below.
BRIEF DESCRIPTION OF THE DRAWINGS
[0011] FIG. 1 shows a transmitting entity and a receiving entity in a MIMO
system;
[0012] FIG. 2 shows a model for data transmission with steering transmit diversity;
[0013] FIGS. 3A and 3B show data transmission in a single-carrier MIMO system and a multi-carrier MIMO system, respectively;
[0014] FIGS. 4 and 5 show processes to compute spatial filter matrices for fully and partially correlated channel response matrices, respectively;
[0015] FIG. 6 shows a block diagram of an access point and a user terminal; and [0016] FIG. 7 shows a block diagram of a processor for spatial filter matrix computation.
DETAILED DESCRIPTION
[0017] The word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.
[0018] FIG. 1 shows a simple block diagram of a transmitting entity 110 and a receiving entity 150 in a MIMO system 100. At transmitting entity 110, a transmit (TX) spatial processor 120 performs spatial processing on data symbols (denoted by a vector s(rn)) to generate transmit symbols (denoted by a vector x(m)). As used herein, a WO 2006/00-1706 PC1'4152005/0228-40 "data symbol" is a modulation symbol for data, a "pilot symbol" is a modulation symbol for pilot (which is data that is blown a priori by both the transmitting and receiving entities), a "transmit symbol" is a symbol to be sent from a transmit antenna, a "received symbol" is a symbol obtained from a receive antenna, and a modulation symbol is a complex value for a point in a signal constellation used for a modulation scheme (e.g., M-PSK, M-QAM, and so on). The spatial processing is performed based on steering matrices V(m) and possibly other matrices. The transmit symbols are further conditioned by a transmitter unit (TMTR) 122 to generate NT modulated signals, which are transmitted from NT transmit antennas 124 and via a MIMO channel.
100191 At receiving entity 150, the transmitted modulated signals are received by NR
receive antennas 152, and the NR received signals are conditioned by a receiver unit (RCVR) 154 to obtain received symbols (denoted by a vector r(nz) ). A receive (RX) spatial processor 160 then performs receiver spatial processing (or spatial matched filtering) on the received symbols with spatial filter matrices (denoted by 1µ41(nz)) to obtain "detected" data symbols (denoted by a vector (m)).g The detected data symbols are estimates of the data symbols sent by transmitting entity 110. The spatial processing at the transmitting and receiving entities are described below.
[00201 . The spatial filter matrix computation techniques described herein may be used for a single-carrier MIMO system as well as a multi-carrier MLMO system.
Multiple carriers may be obtained with orthogonal frequency division multiplexing (OFDM), discrete multi tone (DMT), some other multi-carrier modulation techniques, or some other construct. OFDM effectively partitions the overall system bandwidth into multiple (NF) orthogonal subbands, which are also referred to as tones, subcarriers, bins, and frequency channels. With OFDM, each subband is associated with a respective subcarrier that may be modulated with data.
(0021] In MIMO system 100, the M11\40 channel formed by the NT transmit antennas at transmitting entity 110 and the NR receive antennas at receiving entity 150 may be characterized by an NR X NT channel response matrix 11(m) , which may be given as:
111,2(i11) ' = h(111) 122,1(m) '2,2 (in) '" h2,NT(in) 11(m) = Eq (1) "'No ("0 hNR,2(177) /1NR.NT ("7)_.

=

where entry 121 j(m), for i = 1 ... NR and j = 1 ... NT, denotes the coupling or complex channel gain between transmit antenna] and receive antenna i for transmission span in.
A transmission span may cover time and/or frequency dimensions. For example, in a single-carrier MIMO system, a transmission span may correspond to one symbol period, which is the time interval to transmit one data symbol. In a multi-carrier MIMO
system, a transmission span may correspond to one subband in one symbol period. A
transmission span may also cover multiple symbol periods and/or multiple subbands.
For simplicity, the MIMO channel is assumed to be full rank with Ns = NT NR .
100221 The MIMO system may support data transmission using one or more operating modes such as, for example, a "calibrated" mode and an "uncalibrated" mode.
The calibrated mode may employ full-CSI transmission whereby data is transmitted on orthogonal spatial channels (or "eigenmodes") of the MIMO channel. The uncalibrated mode may employ partial-CSI transmission whereby data is transmitted on spatial channels of the MIIVIO channel, e.g., from individual transmit antennas.
100231 The MIMO system may also employ steering transmit diversity (STD) to improve performance. With STD, the transmitting entity performs spatial processing with steering matrices so that a data transmission observes an ensemble of effective channels and is not stuck on a single bad channel realization for an extended period of time. Consequently, performance is not dictated by the worst-case channel condition.
1. Calibrated Mode ¨ Full-CSI Transmission 100241 For full-CSI transmission, eigenvalue decomposition may be performed on a correlation matrix of 11(m) to obtain Ns eigenmodes of H(m), as follows:
R(m) 11H (m) - II(m) = E(m) A(m) =E" (iii) , Eq (2) where R(m) is an NT X NT correlation matrix of 11(m);
E(m) is an NT X NT unitary matrix whose columns are eigenvectors of R(m);
A(m) is an NT X NT diagonal matrix of eigenvalues of ROO; and denotes a conjugate transpose.
A unitary matrix U is characterized by the property Un - U = I, where I is the identity matrix. The columns of a unitary matrix are orthogonal to one another, and each column has unit power. The matrix E(m) may be used for spatial processing by the transmitting entity to transmit data on the Ns eigenmodes of 11(m) . The eigenrnodes may be viewed as orthogonal spatial channels obtained through decomposition.
The diagonal entries of A(m) are eigenvalues of R(m), which represent the power gains for the Ns eigenmodes. Singular value decomposition may also be performed to obtain matrices of left and right eigenvectors, which may be used for full-CSI
transmission.
[0025] The transmitting entity performs spatial processing for full-CSI
transmission with steering transmit diversity, as follows:
xf (m) = E(m) = V(m) - s(m) , Eq (3) where s(m) is an NT X 1 vector with up to Ns data symbols to be sent in transmission span in;
V(m) is an NT X NT steering matrix for transmission span in;
E(in) is the matrix of eigenvectors for transmission span in; and x( in) is an NT X 1 vector with NT transmit symbols to be sent from the NT
transmit antennas in transmission span in.
As shown in equation (3), each data symbol in s(m) is effectively spatially spread with a respective column of V(m) . If Ns <N1, then Ns data symbols in s(m) are spatially spread with an Ns x Ns matrix V(m) to obtain Ns "spread" symbols. Each spread symbol includes a component of each of the Ns data symbols. The Ns spread symbols from the spatial spreading are then sent on the Ns eigenmodes of 11(m). Each steering matrix V(m) is a unitary matrix and may be generated as described below.
[0026] The receiving entity obtains received symbols from the NR receive antennas, which may be expressed as:
r j(m) = H(m) = x f (m) + n(m) =H(m) = E(m) = Y(n) s(m) +
Eq (4) Hf_eff (iii) = s(m) + n(m) where r f (in) is an NR x 1 vector with NR received symbols obtained via the NR
receive antennas in transmission span in;

n(m) is a noise vector for transmission span in; and f _aff (m) is an NR X NT "effective" MIMO channel response matrix observed by the data vector s (m) for full-CSI transmission with steering transmit diversity, which is:
eff Hf (in) 11(m) - E(m)- Y(in) Eq (5) For simplicity, the noise is assumed to be additive white Gaussian noise (AWGN) with a zero mean vector and a covariance matrix of co = 2I, where o-2 is the variance of ¨nn the noise and 1 is the identity matrix.
100271 The receiving entity can recover the data symbols in s(m) using various receiver processing techniques. The techniques applicable for full-CSI transmission include a full-CSI technique and a minimum mean square error (MMSE) technique_ [0028] For the full-CSI technique, the receiving entity may derive a spatial filter matrix fcsi (in) as follows:
Mfcsi(m) = (110 - A-1 (in) = EH (in) 1111 (in) Eq (6) The receiving entity may perform receiver spatial processing using M (in) , as follows:
f,( Mfrs; (in) = r f (in) , = V H (in) = A-I (in) = Eli (m) = le (in) - [Wm). E(m). V(m) - s(m)+ 14(m)1 , Eq (7) s(m) lir (in) , where gfc,i(m) is an NT X I vector with Ns detected data symbols; and n1 (in) is the post-detection noise after the receiver spatial processing.
[0029] For the MMSE technique, the receiving entity may derive a spatial filter matrix _õ,õ,õ (m) as follows:
[11Leff(m)- HI _eff (in) + cr2 = 1]-1 = H iff_eff (m) . Eq (8) The spatial filter matrix Mf _,õõ,õ(m) minimizes the mean square error between the symbol estimates from the spatial filter and the data symbols in s(nz) .
[00301 The receiving entity may perform MMSE spatial processing, as follows:
n-i f _rnmse(in) f _muse (in) - M f _mmse(in) = r f (in) , Dfimmõ (in) M f _minõ(M) f_eff (in) s(m) n(m)] Eq (9) D-ft_mmse (iii) .M f _ntnise(in) _eff (in) s(m) f _nunse (in) , mmse s(in is where Di_ a diagonal matrix containing the diagonal elements of M1 (in) I-If_eff (311) , or Di _.õ,õ(m) = diag [M1_,õõõe(m)= Hf_ep- (In)] ; and n f õõõõ(m) is the MMSE filtered noise.
The symbol estimates from the spatial filter Mfin nue (in,_ 1 are unnonnalized estimates of -%
the data symbols. The multiplication with the sealing matrix Dintmse _ ( ) provides normalized estimates of the data symbols.
[00311 Full-CSI transmission attempts to send data on the eigenmodes of 11(m) .
However, a full-CSI data transmission may not be completely orthogonal due to, for example, an imperfect estimate of 11(m), error in the eigenvalue decomposition, finite arithmetic precision, and so on. The MilvISE technique can account for (or "clean up") any loss of orthogonality in the full-CSI data transmission.
100321 Table 1 summarizes the spatial processing at the transmitting and receiving entities for full-CSI transmission with steering transmit diversity.
Table 1 Entity Calibrated Mode - Full-CSI Transmission Transmitter If (m) = E(ni) = V(m) = s(m) Spatial Processing 111_,ff (m)=- 11(m) = E(nz) = V(771) Effective Channel Receiver M fcsi (n) = EH (in). HH (in) Spatial full-CSI Filter Matrix 110 700(.111114706 PCPUS20057022841) Spatial (m) M r -(m) Processing (m ) = [-1-1feff (m)- Hr_ 400 + t - Spatial Receiver D _ ,,we (u1) diag [Mj-iNe(m). f _off (1/1).1 Filter Matrix mms E _________________________________________________________________ Sf Spatial miw,!(.1") " .inmse(in) = M (n) r j-(iii) Processing 2. Uncalikated Mode ¨ Partial-CSI Transmission 100331 For partial-CSI transmission with steering transmit diversity, the transmitting entity performs spatial processing as follows:
x, (in)(m) - s(/n) , Eq (10) where xi(m) is the transmit data vector for transmission span in. As shown in equation (10), each data symbol in s(m) is spatially spread with a respective column of V(in).
The NT spread symbols resulting from the multiplication with V(m) are then transmitted from the NT transmit antennas.
100341 The receiving entity obtains received symbols, which may be expressed as:
rp(m) , H(m). x 1,00+ n(in) = 1-1(in) = V(m)- s(m) + n(in) , Eq (11) = p_ ,yr(m)- s(m) + n(m) , where r(m) is the received symbol vector for transmission span in; and ¨P
IlL
(n) is an NR x NT effective MEMO channel response matrix observed by s(m) tor partial-CSI transmission with steerine, transmit diversity, which is:
ti(mn)- Y(m) . Eq (12) 100351 The receiving entity can recover the data symbols in s(m) using various receiver processing techniques. ftc techniques applicable for partial-CSI transmission include a channel correlation matrix inversion (CCM1) technique (which is also commonly called WO 200(/004706 PCT/US2005/022840 a zero-forcing technique), the MMSE technique, and a successive interference cancellation (SIC) technique.
100361 For the CCMI technique, the receiving entity may derive a spatial filter matrix Meõõ-(m), as follows:

[Hp 1T('1 p (101- - p (in) = (13) The receiving entity may perfomi CCMI spatial processing, as follows:
= rp(m) , 0-(m)-{11, .eff (in) = s(in)+ r)(in)] , Eq (14) =
s(in) n (in) , where cc,õi(m) is the CCMI filtered noise. Due to the structure of (in) , the CCMI technique may amplify the noise.
[00371 For the MMSE technique, the receiving_ entity may derive a spatial filter matrix NI,õ,õ(m), as follows:
Mp_mmõ(211) = [1-1Th_cif (in) 00+ a2 tip _ell (in) . Eq (15) Equation (15) for the partial-CSI transmission has the same form as equation (8) for the full-CSI transmission. However, Ikp_eff (m) (instead of [Lett-00 ) is used in equation (15) for partial-CSI transmission.
[0038] The receiving entity may perform MMSE spatial processing, as follows:
p _01111Se ( Pr', nanSe(in) M p M11715e(M) r p (110 Eq (16) = (m) = M (in) = (in) - (111) -4-(in) , p _ p mrivli:
where D p (hag [M,, ("1) = I p õff and nr, (m) is the MMSE
filtered noise for partial-CSI transmission.
100391 For the SIC technique, the receiving entity recovers the data symbols in s(m) in successive stages. For clarity, the following description assumes that each element of I
S(m) and each element of r(m) corresponds to one data symbol stream. The receiving entity processes the NR received symbol streams in r r(m) in Ns successive stages to recover the Ns data symbol streams in s(m) . Typically, the SIC
processing is such that one packet is recovered for one stream, and then another packet is recovered for another stream, and so on. For simplicity, the following description assumes Ns = NT.
100401 For each stage e where C =I ... Ns , the receiving entity performs receiver spatial processing on NR input symbol streams rp(m) for that stage. The input symbol streams for the first stage (1! = ) are the received symbol streams, or r'p(m)--=
The input symbol streams for each subsequent stage (P = 2 ... Ns ) arc modified symbol streams from a preceding stage. The receiver spatial processing for stage C is based on a spatial filter matrix M,(m), which may be derived based on a reduced effective channel response matrix I-Cp' (in) and further in accordance with the CCMI, MMSE, or some other technique. 1-14,_.01(m) contains N, ¨ e +1 columns in ft, J.ff(m) corresponding to Ns + I
data symbol streams not yet recovered in stage C. The receiving entity obtains one detected data symbol stream f.,} for stage t! and further processes (e.g., demodulates, deinterleaves, and decodes) this stream to obtain a corresponding decoded data stream Id( 100411 The receiving entity next estnnates the interference that data symbol stream causes to the other data symbol streams not yet recovered. To estimate the interference, the receiving entity processes (e.g., re-encodes, interleaves, and symbol maps) the decoded data stream {e} in the same manner performed by thc transmitting entity for this stream and obtains a stream of "remodulated" symbols {ie} , which is an estimate of the data symbol stream (se} just recovered. The receiving entity then performs spatial processing on the remodulated symbol stream with steering matrices V(m) and further multiplies the result with channel response matrices 11(m) to obtain NR
interference .e components (m) caused by stream ts, . The receiving entity then subtracts the NR

NVO 2006/004706 1)CT/US2005/022840 interference components i (in) from the NR input symbol streams r p(in) for the current stage C to obtain NR input symbol streams rff:" (m) for the next stage, or f f . 4-1 (Ill) = (111) . The input symbol streams rt+1(/il) represent the streams that the receiving entity would have received if the data symbol stream {sr} had not been transmitted, assuming that the interference cancellation was effectively performed. The -receiving entity then repeats the same processing on the NR input symbol streams rf+1 (In) to recover another data stream. However, the effective channel response matrix ¨ P
Elep+1 eff (711) for the subsequent stage C + 1 is reduced by one column corresponding to the data symbol stream {se} recovered in stage e .
100421 For the SIC technique, the SNR of each data symbol stream is dependent on (1) the receiver processing technique (e.g., CCMI or MMSE) used for each stage, (2) the specific stage in which the data symbol stream is recovered, and (3) the amount of interference due to the data symbol streams recovered in later stages. In general, the SNR progressively improves for data symbol streams recovered in later stages because the interference from data symbol streams recovered in prior stages is canceled. This may then allow higher rates to be used for data symbol streams recovered in later stages.
[0043] Table 2 summarizes the spatial processing at the transmitting and receiving entities for partial-CSI transmission with steering transmit diversity. For simplicity, the SIC technique is not shown in Table 2.
Table 2 Entity Uncalibrated Mode - Partial-CSI Transmission Transmitter xp (in) = (ni)- s(m) Spatial Processing Effective nr_eir (m) = 11(m) V(m) Channel Mõõ,;(m) =- [HH (117) - H (iii)]] (in) Spatial ¨ P = - P ¨P _eff Receiver Filter Matrix =
CCM I
= M=(/17) = r(in)Spatial Processing WO 2011(01047W, PCTI1S2005/022S40 p_xt1(in)-11 a- -II -li.(ni) Spatial Receiver diag [Nip FL (in)] Filter Matrix MMSE
Spatial (in) =I) r (in) ¨p in:n ¨p MOV., ¨"
Processing [00441 FIG. 2 shows a model for data transmission with steering transmit diversity.
Transmitting entity 110 performs spatial processing (or spatial spreading) for steering transmit diversity (block 220) and spatial processing for either full-CSI or partial-CSI
transmission (block 230). Receiving entity 150 pert onusreceiver spatial processing for full-CSI or partial-CSI transmission (block 2601 and receiver spatial processing (or spatial despreading) for steering transmit diversity (block 270). As shown in FIG. 2, the transmitting entity performs spatial spreading for steering transmit diversity prior to the spatial processing (if any) for full-CSI and partial-CS! transmission. The receiving entity may perform the complementary receiver spatial processing for full-CSI
or partial-CSI transmission t011owed by spatial despreading for steering transmit diversity.
3. Spatial Filter Matrix Computation [00451 With steering transmit diversity, different steering matrices V(m) may be used for different transmission spans to randomize the effective MEMO channel observed by a data transmission. This may then improve performance since the data transmission does not observe a "bad" MEMO channel realization for an extended period of time.
The transmission spans may correspond to symbol periods for a single-carrier MIMO
system or subbands for a multi-carrier MINIO system.
(0046) FIG. 3A shows a partial-CSI transmission with steering transmit diversity for a single-carrier MEMO system. For this system, the transmission span index in may be equal to a symbol period index n (or in = n). One vector s(n) of data symbols may be transmitted in each symbol period n and spatially spread with a steering matrix V(n) selected for that symbol period. Each data symbol vector s(n) observes an effective MINIO channel response o 11 ff(n) = 11(n)- V (n) and is recovered using a spatial filter matrix )1,(n).
100471 FIG. 313 shows a 1artial-CS1 transmission µvith steering transmit diversity in a multi-carrier M1M0 system. For this system. the transmission span index in may be WO 200(004706 PLIAIS2005/022840 equal to a subband index k (or in = k ). For each symbol period, one vector s(k) of data symbols may be transmitted in each subband k and spatially spread with a steering matrix V(k) selected for that subband. Each data symbol vector s(k) observes an effective MIMO channel response of tip 1(k) = V(k) and is recovered using, a spatial filter matrix M, (k) . The vector s(k) and the matrices V(k), 11(k), and ______ (k) are also a function of symbol period n, but this is not shown for simplicity.
100481 As shown in FIGS. 3A and 3B, if different steering matrices are used for different transmission spans, then the spatial filter matrices used by the receiving entity are a function of the transmission span index in. This is true even if the channel response matrix 11(m) is fixed or constant over a range of transmission spans.
For example, in a multi-carrier MIMO system, 11(k) may be fixed across a set of subbands for a flat fading MIMO channel with a flat frequency response. As another example, in a single-carrier MIMO system, II(n) may be fixed over a given time interval for a MIMO channel with no temporal fading. This time interval may correspond to all or a portion of the time duration used to transmit a block of data symbols that is coded and decoded as a block.
[00491 A degree of correlation typically exists between the channel response matrices for adjacent transmission spans, e.g., between 11(m) and 11(tii 1). This correlation may be exploited to simplify the computation for the spatial filter matrices at the receiving entity. The computation is described below for two cases -- full-correlation and partial-correlation.
A. Full Correlation [0050] With full-correlation, the channel response matrix for the MEMO
channel is fixed over a range of transmission span indices of interest, e.g., for in =1 ... M , where M may be any integer value greater than one. Thus, 11(1) 1-1(2) 11(M) .
[0051] For the full-CSI technique, the spatial filter matrix Mii(//i) with fully correlated channel response matrices may be expressed as:
NI Fs/00 = V"(///) A = Ku . Fq (1 7 ) The spatial filter matrix Mr,õ(m) may then be computed as:

wo /006/0047116 PCT/US20051022840 N't fõi (In) H (in) = M , for ni = M , Eq (IS) where M = A =
EH = II" is a- base spatial filter matrix, which is the spatial filter jcsi _base.
matrix for the full-CSI teclmique without steering transmit diversity. The base spatial filter matrix M is not a function of transmission span in because the channel response matrix H is fixed. Equation (IS) indicates that the spatial filter matrix Mft.õ(m) for each transmission span in may be obtained by pre-multiplying the base spatial filter matrix M. with the steering matrix VII (m) used for that transmission span.
100521 Alternatively, the spatial filter matrix may be computed as:
IVI = (n) M (1) for in = 2 M , Eq (19) where M(1) = V " (1)- A I- Ea = Hil and W., (in) = (in)- V(1).
Equation (19) indicates that the spatial filter matrix Mfi,si(m) for each transmission span in may be obtained by pre-multiplying the spatial filter matrix M.,(1) for transmission span 1 with the matrix W 1(in). The matrices W1(111) , for in = 2 ... M , are unitary matrices, each of which is obtained by multiplying two unitary steering matrices V(m) and V(1). The matrices W1(in) may be pre-computed and stored in a memory.
10053] For the MMSE technique for full-CSI transmission, the spatial filter matrix fõõ(m) with fully correlated channel response matrices may be expressed as:
f õ,õõ, ( ) [HI; (in ) -L or(1i1) o-2 -,{v11(,7)= EH IC E- V(in) +a -IL =V" (m)- En , .Eq (20) VH(m)- [EH E = 11 = Ell -11" .
Equation (20) is derived using the properties: (A = B)-' 13-1 -A and V = VI' 1. The _ term within bracket in the second equality M equation (20) may he expressed as:

[VH = EH = 11 =I-4 = E = V + o-2 = -= {\"(F1' = H" = H = E+ cy2 V" )- V]
, = [V" (E" - = H = E+ (72 Vi , where "(m)" has been omitted for clarity. The inverse of the term in the second equality above may then be expressed as:
[V"(E" = H" = H = E+ - 1)- =[V"(E" -H" flEa2 -0.1 = V] , where V" = .
[00541 The spatial filter matrix Mf ( ) may be computed as:
µ,11 MM f aans,,'(in ¨ kin f moist. base for in =1 ... M , Eq (21) where = [EH - - 11- -1]-1 - EH - .
Similar to the full-CSI
M f 'anise base technique, the spatial filter matrix Mv,õõse(m) for transmission span in may be obtained by pre-multiplying the base spatial filter matrix Mf base with the steering muse _ matrix VH(m). The spatial filter matrix M(m) may also be computed as:
Mf _nan5C(111) W 1(m)-M _____ f rriniSe(1) for in = 2 M , Eq (22) where M (1) = NTH (1)- [Ell H -E+ I] -E" -H11.
¨ ¨ ¨
[0055] For the CCIvIl technique, the spatial filter matrix IN/1,1(m) with fully correlated channel response matrices may be expressed as:
= [1 1 õif (In) IL co, 00]-1 H ,./1- (in) , =[V" (in) = I-1" -II = V(m)].1 -V" (m)-1114 [V (m)= R V(in)] VH(m)- , Eq (23) (m) - R IV/1(m)] V H (111)-HH
, y (m)-R , where [V" (m)1-1 V(m) because V(m) is a unitary matrix.
100561 The spatial filter matrix M.õõ,;(m) may thus be computed as:
M1 (in) V' 00 M emi _base for m =I M , Eq (24) where M = R- =
H The spatial filter matrix M,,1(in) may also be computed as:
_ _ (m) M1(1) for in = 2 M , Eq (25) where Mce,õ;(1) =Y11 (1)- R -1 = H"
100571 For the MMSE technique for partial-CSI transmission, the spatial filter matrix 1V1(m) with fully correlated channel response matrices may be expressed as:
-I H
r--[H (in) = ; (m) + a2 = I] = H p (in) , = [V H (m)- H" V(m) +o-2 -11-1 -VH (w)-H" , Eq (26) = V"(m)-[H" -H + a2 II -1 = 11" .
Equation (26) may be derived in similar manner as equation (20) above.
100581 The spatial filter matrix Mp_õ(m) may be computed as:
Ni e (in) = ¨ (in) "I. p_ntruse_base for in = I M , Eq (27) where Mp [He 1-I a2 - I] -1 = H" . The spatial filter matrix Allp2,õnse (n) inny also he computed as:
Wi(m)* M p niõ:õ(1) for in 2 M , Eq (28) where M. nuns,. (1) = V" (1) =[H" = H + o-2 = H" .
¨
100591 Table 3 summarizes the computation for the spatial filter matrices for full-CSI
and partial-CSI transmissions with fully correlated channel response matrices over transmission spans in = I ... M
Table 3 Spatial Filter Matrices with Full Correlation WO 2006/00-1706 pc-rluS70(151IFY,84(1 Mode Spatial Filter Matrix Technique ,r/
M - = A = E -11 and Full-CSI
M ksi H (1n). M fesi _base Full-CS I _______________________________________________ = [E" ut - H -E+ (3-2 -11'EH - , and MMSE
V" rtir M f inmse Julre(n1) f _mrnse _base H
M ceõ,; = R - H , and CCMI
_______________________ (in) VI' (iii) = Mccmi base Partial-CSI _______ M , = [1-1"-H +o-2 = H" , and MMSE
õa, (in) = V (iii) -M p nanse_base =
100601 In general, the spatial filter matrix for transmission span in may be computed as (m)= V" (m)- M r _base , where the subscript "x" denotes the receiver processing technique and may be "fesi", 'muse", "ccmi", or "p_ininse". The base spatial filter matrix Ms basõ may be computed as if steering transmit diversity was not used.
[0061] FIG. 4 shows a flow diagram of a process 400 to compute spatial filter matrices with fully correlated channel response matrices over transmission spans in =1 M .
An initial spatial filter matrix M., is first computed (block 412). This initial spatial filter matrix may be the base spatial filter matrix Mx_basõ that is derived based on (1) the channel response matrix H and (2) the receiver processing technique selected for use (e.g., full-CSI, MMSE for full-CSI, CCIV1I, or MMSE for partial-CSI).
Alternatively, the initial spatial filter matrix may be the spatial filter matrix 1\1,(1) for transmission span in =1 , which may be derived based on H and V(1) .
[0062] The transmission span index in is then set to 1 if Mr ,õ = M, base (as shown in FIG. 4) or set to 2 if =
M,(1) (block 414). The spatial filter matrix M ,(m) for transmission span in is then computed based on the Initial spatial filter matrix Msiõõ
and the steering matrix V(m) used for transmission span HI (block 416). In particular, M, (iii) may be computed based on either 114,0,aõ and V(m) or M., (1) and W, 11) , as described above. A determination is then made whether in < M (block 420). If the WO /006/004706 1CTIUS20051022:140 answer is 'yes, then the index in is incremented (block 422), and the process returns to block 416 to compute the spatial filter matrix for another transmission span.
Otherwise, if in = M in block 420, then the spatial filter matrices ,(1) through M(M) are used for receiver spatial processing of received symbol vectors r,(1) through rõ(M) , respectively (block 424). Although not shown in FIG. 4 for simplicity, each spatial filter matrix may be used for receiver spatial processing as soon as both the spatial filter matrix M(in) is generated and the received symbol vector r,(m) are obtained.
100631 For full-CSI transmission, the spatial processing at the transmitting entity may also be simplified as: xf (m) = E = V(m) s(m) . A matrix E = V(m) may be computed for each transmission span in based on the steering matrix V(m) for that transmission span and the matrix E. which is not a function of transmission span for the full correlation case.
B. Partial Correlation 100641 With partial-correlation, the channel response matrices for the MIMO channel are less than fully correlated across a range of transmission span indices of interest. In this case, a spatial filter matrix computed for a transmission span C may be used to facilitate the computation of a spatial filter matrix for another transmission span in.
[00651 In an embodiment, a base spatial filter matrix M(f) for transmission span is obtained from a spatial filter matrix M,(e) computed for transmission span e by removing the steering matrix V(f) used for transmission span I!, as follows:
= V(C)- M, (0 . Eq (29) The base spatial filter matrix Mx_bõ,(r) is then used to derive a base spatial filter matrix Mx_bõ,(m) for transmission span in (e.g., in = C 1). Alx_bõ,(m) may be computed, e.g., using an iterative procedure or algorithm that iteratively performs a set of computations on Mx baõ(t) to obtain a final solution for b,('fl). Iterative procedures for computing an MMSE solution (e.g., adaptive MMSE algorithms, gradient algorithm, lattice algorithms, and so on) are known in the art and not described herein. The spatial filter matrix M, (in) for transmission span in may be computed as:

M, (In) = V" (11) - ba,(m) . Eq (30) The processing order for this embodiment may thus be given as:
M M (Juise(M) M i(m) , where " -->" denotes a direct computation and "" denotes possible iterative computation. The base spatial filter matrices m(e) and M(m) do not contain steering matrices, whereas the spatial filter matrices M() and M ,(m) contain steering matrices v(e) and V(m) used for transmission spans e and m, respectively.
[0066] hi another embodiment, the spatial filter matrix M(m) for transmission span in is computed using an iterative procedure that iteratively performs a set of computations on an initial guess M(m). The initial guess may be derived from the spatial filter matrix Mx (f) derived for transmission span , as follows:
Mx(m), We(m)-M,Y) , Fq (31) where We(m) = V (m) V() . The processing order for this embodiment may be given as: M --> (m) Mx (/7) . The spatial filter matrices M, (11I) and Mx (m) both contain the steering matrix V(m) used for transmission span M.
[0067] For the above embodiments, mx_base(e) and Mx(m) may be viewed as the initial spatial filter matrices used to derive the spatial filter matrix 114,00 for a new transmission span in. In general, the amount of correlation between ivt,(e) and Mx(m) is dependent on the amount of correlation between M
_baseM and Mx_bõ,(m), which is dependent on the amount of correlation between H() and 11(m) for transmission spans c and in. A higher degree of correlation may result in faster convergence to the final solution for M().
[00681 FIG. 5 shows a flow diagram of a process 500 to compute spatial filter matrices with partially correlated channel response matrices for transmission spans in =1 M .
The indices for the current and next transmission spans are initialized as e =
I and = 2 (block 512). A spatial filter matrix M1(t) is computed for transmission span (2.
in accordance with the receiver processing technique selected for use (block 514). An WO 2006100170(. 1)("FIUSNO5/0228441 initial spatial filter matrix M., for transmission span in is then computed based on the spatial filter matrix .N."I JO and the proper steering matrix/matrices V(0 and V(nn) , e.g., as shown in equation (29) or (31) (block 516). The spatial filter matrix (m) for transmission span in is then computed based on the initial spatial filter matrix M. õ e.g., using an iterative procedure (block 518).

determination is then made whether in < M (block 520). If the answer is 'yes', then the indices .e and in are updated, e.g., as P = in and in =-in + I
(block 522).
The process then returns to block 516 to compute a spatial filter matrix for another transmission span. Otherwise, if all spatial filter matrices have been computed, as determined in block 520, then the spatial filter matrices M,(1) through M,(M) are used for receiver spatial processing of received symbol vectors r, (l) through r,.(1\4), respectively (block 524).
10070! For simplicity, FIG. 5 shows the computation of M spatial filter matrices for M
consecutive transmission spans in =LI M .
The transmission spans do not need to be contiguous. In general, a spatial filter matrix derived for one transmission span P is used to obtain an initial guess of a spatial filter matrix for another transmission span in, where P and in may be any index values.

4. Steering Matrices [0071i A set of steering matrices (or transmit matrices) may be generated and used for steering transmit diversity. These steering matrices may be denoted as ty}, or V(i) for i =1 L , where L may be any integer greater than one. Each steering matrix V(i) should be a unitary matrix. This condition ensures that the NT data symbols transmitted simultaneously using V(i) have the same power and are orthogonal to one another after the spatial spreading with V(i [00721 The set of 1,, steering matrices may be generated in various manners. For example, the L steering matrices may be generated based on a unitary base matrix and a set of scalars. The base matrix may be used as one of the L steering matrices.
The other L-1 steering matrices may bc generated by multiplying the rows of the base matrix with different combinations of scalars. Each scalar may be any real or complex value. The scalars are selected to have unit magnitude so that steering matrices generated with these scalars are unitary matrices.
100731 The base matrix may be a Walsh matrix. A 2 x 2 Walsh matrix W,, and a larger size Walsh matrix W2NQN may be expressed as:

=
and W 1 N,.1N Eq (32) -1 Lw -w N.N _NxN_ Walsh matrices have dimensions that are powers of two (e.g., 2, 4, 8, and so on).
[0074] The base matrix may also be a Fourier matrix. For an N x N Fourier matrix DN,N , the element dõ,,õ in the n-th row and m-th column of D_N,N may be expressed as:
-j2 ___________________ d = e N , for ii = ... NI and III ... . Eq (33) Fourier matrices of any square dimension (e.g., 2, 3, 4, 5, and so on) may be formed.
Other matrices may also be used as the base matrix.
[0075] For an N x N base matrix, each of rows 2 through N of the base matrix may be independently multiplied with one of K different possible scalars. KN-1 different steering matrices may be obtained from KN-' different permutations of the K
scalars for N -1 rows. For example, each of rows 2 through N may be independently multiplied with a scalar of +1, -1, +j, or -/. For N =4 and K = 4 , 64 different steering matrices may be generated from a 4 x 4 base matrix with four different scalars. In general, each row of the base matrix may be multiplied with any scalar haying the form e' , where 8may be any phase value. Each element of a scalar-multiplied N x N
base matrix is fw-ther scaled by 1/ AiN to obtain an N x N steering matrix having unit power for each column.
[00761 Steering matrices derived based on a Walsh matrix (or a 4 x 4 Fourier matrix) have certain desirable properties. If the rows of the Walsh matrix are multiplied with scalars of 1 and j, then each element of a resultant steering matrix is +1, --I, j or - j. In this case, the multiplication of an element (or "weight-) of a spatial filter matrix with an element of the steering matrix may be performed with just hit manipulation. lithe elements of the I. steering matrices belong in a set composed of /006/0(1.4706 PCPUS/"M122340 f+l, ¨ 1. + j, ¨ A then the computation to derive the spatial filter matrices for the full correlation case can be greatly simplified.

5. MIMO System 100771 FIG. 6 shows a block diagram of an access point 610 and a user terminal 650 in a MEMO system 600. Access point 610 is equipped with Nõp antennas that may be used for data transmission and reception, and user terminal 650 is equipped with Nut antennas, where Nap > 1 and Nu, >1.
100781 On the downlink, at access point 610, a TX data processor 620 receives and processes (encodes, interleaves, and symbol maps) traffic/packet data and control/
overhead data and provides data symbols. A TX spatial processor 630 performs spatial processing on the data symbols with steering matrices V(m) and possibly eigenvcctor matrices E(m) for the downlink, e.g., as shown in Tables 1 and 2. TX spatial processor 630 also multiplexes in pilot symbols, as appropriate, and provides Nap streams of transmit symbols to Nap transmitter units 632a through 632ap. Each transmitter unit 632 receives and processes a respective transmit symbol stream and provides a corresponding downlink modulated signal. Nap downlink modulated signals from transmitter units 632a through 632ap arc transmitted from Nap antennas 634a through 634ap, respectively.
[00791 At user terminal 650, Nut antennas 652a through 652ut receive the transmitted downlink modulated signals, and each antenna provides a received signal to a respective receiver unit 654. Each receiver unit 654 performs processing complementary to that performed by receiver unit 632 and provides received symbols. An RX spatial processor 660 performs receiver spatial processing on the received symbols from all receiver units 654a through 654ut, e.g., as shown in Tables 1 ud 2, and provides detected data symbols. An RX data processor 670 processes (e.g., symbol demaps, deinterleaves, and decodes) the detected data symbols and provides decoded data for the downlink.
[00801 The processing for the uplink- may be the same or different from the processing for the downlink. Traffic and control data is processed (e.g., encoded, interleaved, and symbol mapped) by a TX data processor 688, spatially processed by a TX spatial processor 690 with steering matrices V(///) and possibly cigenvector matrices E(m) for WO 2006/110-17116 PCTTS2005/02284o the uplink, and multiplexed with pilot symbols to generate Nu, transmit symbol streams.
Nu, transmitter units 654a through 654ut condition the Nõ, transmit symbol streams to generate Nõ, uplink modulated signals, which arc transmitted via Nu, antennas 652a through 652ut.
[0081] At access point 610, the uplink modulated signals are received by Nap antennas 634a through 634ap and processed by Nap receiver units 632a through 632ap to obtain received symbols for the uplink. An RX spatial processor 644 performs receiver spatial processing on the received symbols and provides detected data symbols, which are further processed by an RX data processor 646 to obtain decoded data for the uplink.
100821 Processors 638 and 678 perform channel estimation and spatial filter matrix computation for the access point and user terminal, respectively. Controllers 640 and 6S0 control the operation of various processing units at the access point and user terminal, respectively. Memory units 642 and 682 store data and program codes used by controllers 630 and 680, respectively.
100831 FIG. 7 shows an embodiment of processor 678, which performs channel estimation and spatial filter matrix computation for user terminal 650. A
channel estimator 712 obtains received pilot symbols and derives a channel response matrix for each transmission span in which received pilot symbols are available. A filter 714 may perform time-domain filtering of the channel response matrices for the current and prior transmission spans to obtain a higher quality channel response matrix 11(m) .
A unit 716 then computes an initial spatial Ilter matrix M õõ
[0084] For fully correlated 1-1(m1), the initial spatial filter matrix M
may be (1) a base spatial filter matrix M, b, computed based on ll(m) and the selected receiver processing technique or (2) a spatial filter matrix Mx(1) for transmission span 1 computed based on H(1). V(1), and the selected receiver processing technique.
For partially correlated U(m), the initial spatial filter matrix N'I may be an initial guess or 1C1(nz) that is obtained based on a spatial filter matrix M() computed for another transmission span L. A unit 718 computes the spatial filter matrix M(m) for transmission span in based on the initial spatial filter matrix 1µ4,õ and the steering matrix V(M) used for that transmission span. For partially correlated H(iii) , unit 718 may implement an iterative procedure to compute for M (in) based on the initial spatial filter matrix, which is an initial guess of M(m).
[0085] Processor 638 performs channel estimation and spatial filter matrix computation for access point 610 and may be implemented in similar manner as processor 678.
5 [0086] The spatial filter matrix computation techniques described herein may be implemented by various means. For example, these techniques may be implemented in hardware, software, or a combination thereof. For a hardware implementation, the processing units for spatial filter matrix computation may be implemented within one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing 10 devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, micro-controllers, microprocessors, other electronic units designed to perform the functions described herein, or a combination thereof.
[0087] For a software implementation, the spatial filter matrix computation may be performed with modules (e.g., procedures, functions, and so on). The software codes may be 15 stored in memory units (e.g., memory units 642 and 682 in FIG. 6) and executed by processors (e.g., controllers 640 and 680 in FIG. 6). The memory unit may be implemented within the processor or external to the processor, in which case it can be communicatively coupled to the processor via various means as is known in the art.
[0088] Headings are included herein for reference and to aid in locating certain sections.
20 These headings are not intended to limit the scope of the concepts described therein under, and these concepts may have applicability in other sections throughout the entire specification.
[0089] The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the teachings defined herein 25 may be applied to other embodiments without departing from the scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the teachings and novel features disclosed herein.

Claims (14)

CLAIMS:

1. A method of deriving spatial filter matrices in a wireless multiple-input multiple-output (MIMO) communication system, comprising:
deriving a first spatial filter matrix for a first transmission span, the first transmission span comprising a transmission span in time, frequency, or time and frequency;
determining a first initial spatial filter matrix for a second transmission span based on the first spatial filter matrix and on at least one steering matrix, the second transmission span comprising a transmission span in time, frequency, or time and frequency, and deriving a second spatial filter matrix for the second transmission span based on the first initial spatial filter matrix.

2. The method of claim 1, wherein the first spatial filter matrix is derived based on a channel response matrix obtained for a MIMO channel in the first transmission span and further in accordance with a receiver spatial processing technique.

3. The method of claim 1, wherein the determining the first initial spatial filter matrix comprises:
processing the first spatial filter matrix to remove a first steering matrix used for the first transmission span, and wherein the first initial spatial filter matrix is equal to the first spatial filter matrix with the first steering matrix removed.

4. The method of claim 1, wherein determining the first initial spatial filter matrix comprises:
processing the first spatial filter matrix to remove a first steering matrix used for the first transmission span and to include a second steering matrix used for the second transmission span, and wherein the first initial spatial filter matrix is equal to the first spatial filter matrix with the first steering matrix removed and the second steering matrix included.

5. The method of claim 1, wherein the second spatial filter matrix is derived using an iterative procedure that iteratively performs a set of computations on the first initial spatial filter matrix to obtain a final solution for the second spatial filter matrix.

6. The method of claim 1, further comprising:
determining a second initial spatial filter matrix for a third transmission span based on the second spatial filter matrix, the third transmission span comprising a transmission span in time, frequency, or time and frequency; and deriving a third spatial filter matrix for the third transmission span based on the second initial spatial filter matrix.

7. The method of claim 1, wherein the first and second transmission spans correspond to two different symbol periods.

8. The method of claim 1, wherein the first and second transmission spans correspond to two different frequency subbands.

9. An apparatus in a wireless multiple-input multiple-output (MIMO) communication system, comprising:
a processor operative to derive a first spatial filter matrix for a first transmission span, the first transmission span comprising a transmission span in time, frequency, or time and frequency, determine a first initial spatial filter matrix for a second transmission span based on the first spatial filter matrix and on at least one steering matrix, the second transmission span comprising a transmission span in time, frequency, or time and frequency, and derive a second spatial filter matrix for the second transmission span based on the first initial spatial filter matrix.

10. The apparatus of claim 9, wherein the processor is operative to process the first spatial filter matrix to remove a first steering matrix used for the first transmission span, and wherein the first initial spatial filter matrix is equal to the first spatial filter matrix with the first steering matrix removed.

11. The apparatus of claim 9, wherein the processor is further operative to determine a second initial spatial filter matrix for a third transmission span based on the second spatial filter matrix, the third transmission span comprising a transmission span in time, frequency, or time and frequency, and to derive a third spatial filter matrix for the third transmission span based on the second initial spatial filter matrix.

12. An apparatus in a wireless multiple-input multiple-output (MIMO) communication system, comprising:
means for deriving a first spatial filter matrix for a first transmission span, the first transmission span comprising a transmission span in time, frequency, or time and frequency;
means for determining a first initial spatial filter matrix for a second transmission span based on the first spatial filter matrix and on at least one steering matrix, the second transmission span comprising a transmission span in time, frequency, or time and frequency; and means for deriving a second spatial filter matrix for the second transmission span based on the first initial spatial filter matrix.

13. The apparatus of claim 12, wherein the means for determining the first initial spatial filter matrix comprises:
means for processing the first spatial filter matrix to remove a first steering matrix used for the first transmission span, and wherein the first initial spatial filter matrix is equal to the first spatial filter matrix with the first steering matrix removed.

14. The apparatus of claim 12, further comprising:
means for determining a second initial spatial filter matrix for a third transmission span based on the second spatial filter matrix, the third transmission span comprising a transmission span in time, frequency, or time and frequency; and means for deriving a third spatial filter matrix for the third transmission span based on the second initial spatial filter matrix.