Amelioration in inter-carrier interference in OFDM

Abstract

In an environment where the transmission medium leading to the receiver is time-variant within a block of received signals, and where transmission medium has an finite impulse response extending over a number of symbols intervals, a receiver that includes a filter that, based on an estimate of the transmission channel, is controlled to have filter coefficients that minimize inter-carrier interference due to the intra-block time variations in the transmission medium. The estimate channel coefficients are determined from pilot signals and an interpolation process. When the receiver is an OFDM receiver, the pilot signals are advantageously selected to be in clusters that are equally spaced from each other in the time or the frequency domains. This approach applies to multiple antenna, as well as to single antenna arrangements, and to arrangements that do, or do not use space-time coding.

Description

BACKGROUND OF THE INVENTION

This invention relates to wireless communication and, more particularly, to wireless communication in an environment where transmission channel characteristics change relatively rapidly.

The explosive growth of wireless communications is creating a demand for high-speed, reliable, and spectrally efficient communications. There are several challenges to overcome in attempting to satisfy this growing demand, and one of them relates to the time variations in the transmission of multicarrier-modulated signals.

Multicarrier transmission for wireless channels has been well studied. The main advantage of orthogonal frequency division multiplexing (OFDM) transmission stems from the fact that the Fourier basis forms an eigenbasis for time-invariant channels. This simplifies the receiver, which leads to an inexpensive hardware implementations, since the equalizer is just a single-tap filter in the frequency domain—as long as the channel is time invariant within a transmission block. Combined with multiple antennas, OFDM arrangements are attractive for high data rate wireless communication, as shown, for example, in U.S. patent application Ser. No. 09/213,585, filed 17 Dec. 1998.

Time invariability within a transmission block cannot be guaranteed at all times, however, for example, with the receiving unit moves at a high speed, and that leads to impairments because the Fourier basis at such times no longer forms the eigenbasis, and the loss of orthogonality at the receiver results in inter-carrier interference (ICI). Depending on the Doppler spread in the channel, and the block length, ICI can potentially cause severe deterioration of quality of service

SUMMARY

An advance in the art is realized with an arrangement wherein a receiver that is responsive to transmissions of blocks of symbols includes a filter that, based on knowledge of the transmission channel that is gained through pilot tones, is caused to have filter coefficients that minimize inter-carrier interference due to intra-block time variations in the transmission channels. The channel coefficients are determined through interpolation of the coefficients determined from the pilot tones. In some embodiments, channel coefficients estimates are improved by employing estimates from previous blocks.

In an OFDM system, the pilot signals are advantageously selected to be in clusters that are equally spaced from each other in the time or the frequency domains. This approach applies to multiple antenna, as well as to single antenna arrangements, and to arrangements that do, or do not use space-time coding.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 presents a block diagram of a prior art arrangement with a multi-antenna transmitter and a multi-antenna receiver;

FIG. 2 shows a signal arrangement where transmitted symbol blocks are augmented with a prefix code that is used to remove inter-block interference;

FIG. 3 presents a block diagram of a prior art OFDM arrangement with a multi-antenna transmitter and a multi-antenna receiver; and

FIG. 4 presents a block diagram of an OFDM arrangement with a multi-antenna transmitter and a multi-antenna receiver in accord with the principles of this invention.

DETAILED DESCRIPTION

FIG. 1 presents a general block diagram of a wireless arrangement where a transmitter 10 includes M_t transmitting antennas 12-1, 12-2, . . . 12-M_t, each of which transmits a signal s₁ that is obtained by processing input signal x₁ through a processing unit g₁, shown by units 11-1, 11-2 and 11-M_t. It is assumed that the channel's impulse response function is time-varying. A receiver 20 includes M_r receiving antennas 21-2, 21-2, . . . 21-M_r, and the received signals are applied to processing unit 22.

When considering only a single transmitting antenna and a single receiving antenna, the received signal y, at time t, can be expressed byy(t)=∫b(t,τ)s(t−τ)dτ+z(t)  (1)where b(t, τ) is the impulse response of the time-varying channel between the transmitting antenna and the receiving antenna, as a function of τ, at time t, and s(t−τ) is the transmitted signal at time t−τ. When sampled at a sufficiently high rate (above 2W_t+2W_s, where W_t is the input bandwidth and W_s is the bandwidth of the channel's time variation), equation (1) can be written in discrete form; and when the transmission channel between the transmitting antenna and the receiving antenna is represented by a discrete, finite, impulse response corresponding, for example, to a tapped filter having ν samples of memory, then equation (1) can be expressed as

$\begin{array}{cc}y\left(k\right)=y\left(k{\tau }_s\right)=\sum _{l=0}^{v}h\left(k,l\right)x\left(k-l\right)+z\left(k\right)& \left(2\right)\end{array}$ where y(k) and z(k) are the received signal and the received noise at sample time k, h(k,l) represents the sampled (time-varying) channel impulse response that combines the transmit filter g₁ with the channel response b₁, and x(k-l) is the input signal at sample time k-l.

When there are M_r receiving antennas, and M_t transmitting antennas, equation (2) generalizes to

$\begin{array}{cc}y\left(k\right)=\sum _{l=0}^{v}H\left(k,l\right)x\left(k-l\right)+z\left(k\right)& \left(3\right)\end{array}$ where y(k) is an M_r-element vector corresponding to the signals received at the receiving antennas (31, 32, 33) at sample time k, H(k,l) is an M_r by M_t matrix of the l^th tap of the transmission medium filter at sample time k (k is a parameter because the transmission medium varies with time), x(k-l) is an M_t-element input vector at sample time k-l that corresponds to the signals of transmitting antennas 21, 22, 23, and z(k) is the M_r-element noise vector at sample time k.

When considering the transmission of information from transmitter 10 to receiver 20 in blocks, one has to realize that the memory/delay of the transmission medium will cause interference between one block and the next, unless a guard-time interval is provided that corresponds to at least the delay introduced by the transmission medium. One can simply send no signals during this interval, but one can also send a symbol sequence, of any length greater than the memory/delay of the transmission medium, i.e., ν or more symbols. Thus, one can send a block of N symbols from antenna 12-1 with a prefix of ν symbols, as shown in FIG. 2, for example, and a receiving antenna can ignore the first ν receiving signals—because those first ν signals suffer from interference by the previous block—and focus on only the remaining N received signals. This can be expressed, for one transmitting antenna and one receiving antenna, by the equationy=Hx+z  (4)where y is a vector with elements y(k), y(k+1), . . . y(k+N−1), x is a vector with elements x(k−ν), x(k−ν+1), . . . x(k−1), x(k), x(k+1). . . x(k+N−1), and H is a N by N+ν element matrix (channel coefficients between the transmitting antenna and the receiving antenna during the N symbol intervals when the receiving antenna pays attention to the signal). When the prefix is selected so thatx(−ν+i)=x(N−1+i) for i=1,2, . . . ν−1,  (5)then the x vector reduces to an N element vector, and H reduces to a N by N element matrix.

When the full complement of M_t transmitting antennas and M_r receiving antennas are considered, equation (4) holds, but H becomes an N·M_r by N·M_t matrix, y and z become N·M_r-element vectors, and x becomes an N·M_t-element vector.

It may be noted that there is no loss of generality in assuming that M_t=M_r=M, which makes H a square matrix of size NM. In the treatment below, therefore, this assumption is made, but it should be understood that the principles disclosed herein apply to situations where M_t is not necessarily equal to M_r.

At a particular receiving antenna, for example 21-j, the received signal can be expressed by equation (4) where H is an N by NM matrix. H can also be considered as an M element vector, where each element is an N by N matrix of H_1j of transfer coefficients between transmitting antenna i and receiving antenna j, i.e., vector [H_1j, H_2,j, . . . H_Mj].

In an OFDM system like the one shown in FIG. 3, each block of N signals in sequence x_i is generated by block 13-i, which performs the inverse discrete Fourier transform (IDFT) of an information-bearing signal X, i.e.,x=Q^HX,  (6)which in the case of an arrangement where there are M transmitting antennas and M receiving antennas, Q^H is the Hermitian of Q, and Q is an NM by NM matrix with elements {tilde over (Q)} on the diagonal, and 0s elsewhere, where {tilde over (Q)} is the N-point DFT transform matrix

$\begin{array}{cc}\tilde{Q}\left(l,k\right)=\frac{1}{\sqrt{N}}{e}^{-j\frac{2\pi }{N}\mathrm{lk}}\mathrm{for}0\le j,k\le N-1.& \left(7\right)\end{array}$ In the case of a single transmitting antenna and a single receiving antenna, equation (6) is simply x={tilde over (Q)}^HX.

At the receiver, the signal of each antenna 21-j is applied to element 24-j, which performs a N-point DFT, generating an N-element vector Y_j. This can be expressed by

$\begin{array}{cc}{Y}_j=\tilde{Q}\left(\sum _{i=1}^M{H}_{\mathrm{ij}}{\tilde{Q}}^H{X}_i+z\right).& \left(8\right)\end{array}$ where H_1j is the N by N matrix of coefficients between transmitting antenna I and receiving antenna j, and X_l is an N-element vector applied to inverse FFT element 13-i (advantageously, N is a power of 2 integer). The signal at the output of element 24-i at clock interval p (within a block) can be written as

$\begin{array}{cc}{Y}_j\left(p\right)=\sum _{l=1}^M\left(\sum _{q=1}^N{G}_{\mathrm{lj}}\left(p,q\right)X_l\left(q\right)\right)+Z\left(p\right)& \left(9\right)\end{array}$ where Z(p) is the transformed noise at clock interval p, X_l(q) is the signal at clock interval q of the block applied to inverse FFT element 13-l, and G_lj(p,q) is the (p,q)^th element of matrixG_lj={tilde over (Q)}H_lj{tilde over (Q)}  (10)Equation (9) can also be written as

$\begin{array}{cc}{Y}_j\left(p\right)=\sum _{l=1}^M\left(G_{\mathrm{lj}}\left(p,p\right)+\sum _{\underset{q\ne p}{q=1}}^N{G}_{\mathrm{lj}}\left(p,q\right)X_l\left(q\right)\right)+Z\left(p\right),& \left(11\right)\end{array}$ and generalizing to the multiple receiving antenna case, equation (11) can be expressed by

$\begin{array}{cc}\begin{array}{c}Y\left(p\right)=G\left(p,p\right)+\sum _{\begin{array}{c}q=1\\ q\ne p\end{array}}^N{G}_{\mathrm{lj}}\left(p,q\right)X_l\left(q\right)+Z\left(p\right)\\ =G\left(p,p\right)+Z_{\mathrm{ICI}}\left(p\right)+Z\left(p\right).\end{array}& \left(12\right)\end{array}$

When the transmission medium coefficients do not vary with time, matrix H of equation (4) becomes a circulant matrix, and it can be shown that when H is circulant, {tilde over (Q)}H{tilde over (Q)}^H is a diagonal matrix. However, when H does vary with time, {tilde over (Q)}H{tilde over (Q)}^H is no longer a diagonal matrix, and consequently, signal Z_ICI(p) of equation (12) is non-zero. Stated in other words, when the transmission coefficients do vary with time, the received signal contains inter-carrier interference. If the Z_ICI(p) were to be eliminated, however, then a conventional decision circuit can be used to arrive at the N elements of Y(p) that form the received block of information signals. Obviously, therefore, it is desirable to eliminate—the inter-carrier interference signals, but in order to do that one must know the values of all of the coefficients of matrix H.

In the time span of N symbol intervals, i.e., in a block, it is possible for each coefficient to change with each sample interval. The matrix H_lj between any transmitting antenna i and any receiving antenna j (in the time span of a block) is an N by N matrix (as demonstrated by above), but only ν terms in each row are non-zero. Consequently, only Nv coefficients need to be ascertained for each H matrix between a transmitting antenna and a receiving antenna (rather than N² coefficients). Nevertheless, this number is still much too large to ascertain, because only N signals are transmitted in a block and, therefore, it is not possible to estimate Nv coefficients, even if all N sample intervals in a block were devoted to known (pilot) signals—which, of course, one would not want to do because it would leave no capacity for communicating any information.

The disclosure below presents a novel approach for developing the necessary coefficients of H_lj, but alas, some of those coefficients are likely to be inexact. Consequently, one can only reduce the inter-carrier interference to some minimum level, rather than completely eliminate it. Still, while recognizing that the channel coefficients that are available in receiver 20 are not all totally accurate, in the treatment below it is assumed that all coefficients of matrix H_lj are known.

To reduce this inter-carrier interference in accord with the principles disclosed herein, a filter element 25 with transfer function W is interposed between the receiving antennas and the FFT elements 24-j (where index j ranges from 1 to M_r), as shown in the FIG. 4 embodiment. The signal developed at the outputs of the FFT elements is

$\begin{array}{cc}\begin{array}{c}Y={\mathrm{QWHQ}}^{H}X+\mathrm{QWz}\\ =\stackrel{\_}{G}X+\stackrel{\_}{Z},\end{array}& \left(13\right)\end{array}$ where Q, W, and H are NM by NM matrices, and X and z are NM-element vectors. Defining e_m, as an NM-element vector with a 1 in the m^th element and zeros elsewhere, then vector q_m=Q^He_m, represents the m^th column of matrix Q^H, which is an NM element vector, comprising M concatenated sets of values

$\frac{1}{\sqrt{N}}{e}^{j\frac{2\pi }{N}\mathrm{mk}}$ for 0≦k≦N−1.

Defining h_m=Hq_m, w_m=W^Hq_m, and R_m=HH^H−h_mh_m^H and further, assuming that w_m^Hw_m=1 for 0≦m≦N−1, it can be shown that the optimum vector at symbol interval m, (i.e., for frequency bin m), w_m, is one that results from solving the optimization problem

max_m w_m^Hh_mh_m^Hw_m subject to

$\begin{array}{cc}{w}_m^H\left(\frac{1}{\mathrm{SNR}}I+R_m\right)w_m=1\mathrm{and}{w}_m^H{w}_m=1.& \left(14\right)\end{array}$ It can be shown that by defining

$R_{\mathrm{yy}}=\frac{1}{\mathrm{SNR}}I+{\mathrm{HH}}^H$ and computing the inverse matrix R_yy⁻¹, the optimum vector for frequency bin m is

$\begin{array}{cc}{w}_{m,\mathrm{optimum}}=\frac{R_{\mathrm{yy}}^{-1}{h}_m}{R_{\mathrm{yy}}^{-1}{h}_m}.& \left(15\right)\end{array}$

Repeating the computations leading to equation 15 for all values of m=0,1, . . . NM yields the various vectors that correspond to the columns of matrix W^HQ^H, by definitions of the relationships w_m=W^Hq_m and q_m=Q^He_m. Forming the matrix, post-multiplying it by Q and taking the Hermitian of the result yields the matrix W.

In the FIG. 4 arrangement, controller 50 is responsive to all of the signals acquired by antennas 21-j (via bus 51), computing the values of the channel coefficients (elements of H), and also computing the coefficients of matrix W as disclosed above. The coefficients of W are applied to filter 25, and the coefficients of H are applied to detection element 23. Computation of channel coefficients from received pilot signals is well known in the art. An innovative approach for estimating channel coefficients in an OFDM system is disclosed, for example, in a patent application by Ariyavisitakul et al, titled “Channel Estimation for OFDM Systems with Transmitter Diversity” which was filed on Dec. 18, 1998 and bears the Ser. No. 09/215,074. This application is incorporated by reference herein.

As indicated above, computation of the optimum filter that is placed following each antenna requires knowledge of the channel coefficients. The following discloses a method and corresponding apparatus that ascertains a selected number of coefficients of H through the use of pilot signals, and obtains the remaining coefficients of H through interpolation. For sake of simplicity of the mathematical treatment, it is assumed that M=1 because the generality of the treatment is not diminished by this assumption. Also, in accord with the principles disclosed herein, it is assumed that the H matrix coefficients in the course of transmitting a number of adjacent symbols within a block do not vary significantly and that, therefore, if two rows of coefficients of matrix H that are fairly close to each other are known, then the coefficient rows between them can be obtained through linear interpolation of the known rows.

A row of coefficients effectively defines the channel at the clock interval corresponding to the row, and in the treatment below it is designated by h(n,l), where index n corresponds to the row within matrix H (i.e., an integer between 1 and N, inclusively) and index l corresponds to the ν potentially non-zero coefficients on a row of H.

Extending this thought, if the channel coefficients are known at P clock intervals, where P is any selected number, i.e., if P rows of H are known, then the remaining rows of H can be obtained by interpolation of the P known rows. Intuitively it is apparent that the error in estimating the coefficients of H decreases as the value of P is increased (i.e., more rows of H are known). Stating the interpolation mathematically, generally, a row h(n,l) can be obtained from

$\begin{array}{cc}h\left(n,l\right)=\sum _{i=1}^P{a}_n^{l}h\left(m_1,l\right)& \left(17\right)\end{array}$ where coefficients α_n^l are members of a set of coefficients α_n and h(m_i,l) is the i^th known set of channel coefficients. In vector notation,h(n,l)=a_nP  (18)where P is a P-element vector P=[h(m₁,l)h(m₂,l). . . h(mp,l)]^T, and a_n is a vector with elements α_n^l. If H_C(i) designates the N by N H matrix if it were not time variant and had the coefficients of the i^th channel that is known, i.e., channel h(m₁,l), then, the channel estimate, {tilde over (H)}, can be expressed by

$\begin{array}{cc}\tilde{H}=\sum _{i=1}^P{A}_{m\left(i\right)}{H}_c\left(m_1\right)& \left(19\right)\end{array}$ where A_l is an N by N diagonal matrix with elements

${\left[A_{m\left(i\right)}\right]}_{n,n}$ that is equal to 1 when n corresponds to the known channel h(m₁,l), i.e., when n equals m(i), is equal to 0 when n corresponds to the other P−1 known channels, and is equal to α_n¹ otherwise.

To illustrate, suppose N=5, and rows 1, 3 and 5 of H are known through detection of pilot signals at that are send during clock intervals 1, 3 and 5. That is, index i has values 1, 3, and 3, and m(i) has values 1, 3 and 5. From the known rows we can then construct H_C(m₁) from h(1,l), H_C(m₂) from h(3,l), and H_C(m₃) from h(5,l). For the two missing rows, we have vector a₂ that has three elements, α₂¹, α₂², α₂³, for example [0.2, 0.3, 0.5] and vector a₄ that has three elements, for example, [0.1, 0.2, 0.7]. According to the above,

$\begin{array}{cc}{A}_5=\left[\begin{array}{ccccc}1& 0& 0& 0& 0\\ 0& a_2^1& 0& 0& 0\\ 0& 0& 0& 0& 0\\ 0& 0& 0& a_4^1& 0\\ 0& 0& 0& 0& 0\end{array}\right]=\left[\begin{array}{ccccc}1& 0& 0& 0& 0\\ 0& 0.2& 0& 0& 0\\ 0& 0& 0& 0& 0\\ 0& 0& 0& 0.1& 0\\ 0& 0& 0& 0& 0\end{array}\right],& \left(20\right)\end{array}$

$\begin{array}{cc}{A}_3=\left[\begin{array}{ccccc}0& 0& 0& 0& 0\\ 0& a_2^2& 0& 0& 0\\ 0& 0& 1& 0& 0\\ 0& 0& 0& a_4^2& 0\\ 0& 0& 0& 0& 0\end{array}\right]=\left[\begin{array}{ccccc}0& 0& 0& 0& 0\\ 0& 0.3& 0& 0& 0\\ 0& 0& 1& 0& 0\\ 0& 0& 0& 0.3& 0\\ 0& 0& 0& 0& 0\end{array}\right],\mathrm{and}& \left(21\right)\end{array}$

$\begin{array}{cc}{A}_5=\left[\begin{array}{ccccc}0& 0& 0& 0& 0\\ 0& a_2^3& 0& 0& 0\\ 0& 0& 0& 0& 0\\ 0& 0& 0& a_4^3& 0\\ 0& 0& 0& 0& 1\end{array}\right]=\left[\begin{array}{ccccc}0& 0& 0& 0& 0\\ 0& 0.5& 0& 0& 0\\ 0& 0& 0& 0& 0\\ 0& 0& 0& 0.7& 0\\ 0& 0& 0& 0& 1\end{array}\right],& \left(22\right)\end{array}$ and{tilde over (H)}=A₁H_c(1)+A₃H_c(3)+A₅H_c(5).  (23)

Two remaining considerations are the placement of the pilot tones, and the values employed in the a_n vectors, for n=1, 2, . . . ,P.

It can be shown that for time-selective channels, pilot tones should be grouped together. On the other hand, in frequency selective time-invariant channels, placing the pilot tones equally spaced on the FFT grid is the optimal scheme. Therefore, for purposes of the FIG. 4 arrangement, it is advantageous to partition the pilot tones into equally spaced groups on the FFT grid. The pilot tones are generated in element 15 of FIG. 4, and are applied to coding unit 13-0, to be applied to elements 13-j as disclosed above. It can be shown that the pilot tones can be coupled directly following elements 13-j, since dispersed clusters in the frequency domain transform to dispersed clusters in the time domain. Of course, in the frequency bins where pilot tones are placed (or time intervals where pilot tones are placed), no information that is to be communicated can be sent.

As for the values employed in the a_n vectors, without imposing any assumptions on the underlying channel variations, linear interpolation appears to be the simplest method for choosing the weight vectors. On the other hand, if apriori knowledge about the underlying channel model is available, more sophisticated channel interpolation schemes can be devised.

In the case of the linear interpolation, each a_n vector consists of two non-zero terms that correspond to the two closest known rows of H (one on either side), and the values of the two non-zero terms reflect the relative distance of the row corresponding to n to the two known rows. For example, if H contains 48 rows and rows 1, 2, 3, 16, 17, 18, 31, 32, 33, 46, 47 and 48 are known, the a₄ vector is {0,0, ½, 11/12, 0,0,0,0,0,0,0,0}

To give an example of a situation where apriori knowledge bout the channel is available assume, for example that the channels follow the Jakes model (see W. C. Jakes, Microwave Mobile Communications, John Wiley & Sons, Inc. 1994) where E[(h(m,l)h^H(n,l)]=J₀(2πf_d(m−n)T) with f_d denoting the Doppler frequency, and T denoting the symbol period, then the calculation of the interpolation weights is straightforward. For example if we fix rows h₁, h_N/2, h_N, the set of weights a_n=[α_n(1), α_n(N/2), an α_n(N)] that minimizes E[|h(n,l)−a_n^H{tilde over (h)}(l)|²] where {tilde over (h)}(l)=[h(1,l), h(N/2,l), h(N,l)]^T can be obtained using the orthogonality principleα_n^H=R_hnĥR_ĥĥ⁻¹ whereR_hnĥ=[J₀[1−n],J₀[N/2−n],J₀[N−n]]  (24)and

$\begin{array}{cc}{R}_{\hat{h}\hat{h}}=\left[\begin{array}{ccc}1& J_0\left[N\text{/}2-1\right]& J_0\left[N\right]\\ J_0\left[N\text{/}2-1\right]& 1& J_0\left[N\text{/}2\right]\\ J_0\left[N\right]& J_0\left[N\text{/}2\right]& 1\end{array}\right]& \left(25\right)\end{array}$ with J₀(n) being equal to J₀(2πf_dnT). Typically, however, for Doppler values of practical importance, there is little to be gained by adopting the Jakes-based estimator in place of the linear interpolator. Hence, from an implementation point of view, the linear estimator appears to be an attractive solution, as it dispenses with the estimation of the Doppler frequency, without sacrificing performance.

An additional enhancement is achieved through channel tracking. In channel tracking, it is assumed that matrix H of one block is related to matrix H of the previous blocks and, therefore, given a matrix Ĥ_μ-1 that is used during block μ−1, and an estimate of the H matrix derived from the pilot signals for block μ, {tilde over (H)}_μ, a matrix to be employed during block 1 is obtained fromĤ_μ=α{tilde over (H)}_μ+(1−α)Ĥ_μ-1  (26)where α is a preselected constant less than 1.

It should be noted that the above-disclosed approach could be used in conjunction with any coding technique in coder 13-0 of FIG. 4, including space-time coding as described, for example, in U.S. Pat. No. 6,185,258.

It should also be realized that the receiver embodiment shown in FIG. 4 intends to clearly demonstrate the signal flow in the receiver and that the actual, physical, embodiment will likely have a somewhat different block diagram. Specifically, the coefficients of W for block μ of the received signals (the block being the M_r received signals during N symbol intervals) need to be developed from the coefficients of the channel transmission matrix, as best estimated by processor 50. In some applications, as disclosed above, this matrix is Ĥ_μ, which is developed pursuant to equation (26). Equation (26) needs to have access to Ĥ_μ-1 and to {tilde over (H)}_μ. The former implies a memory within controller 50, and the latter implies access to the received signals of block μ and processing time that is necessary to develop {tilde over (H)}_μ. To obtain this processing time, filter W (or an element between antennas 21-j and filter W might advantageously include memory that can store at least one block's work of received signals. In an embodiment where the functions of controller 50, filter 25, FFT elements 24-j, and detector & decoder element 23 are implemented with a stored program controlled processor, more than one block's work of memory is needed, though the precise amount is dependent of the specific code that the artisan will write to implement to method disclosed herein.

Claims

1. A wireless receiver comprising: a reception module including Mr antennas for acquiring a block of Mr input signals in each of N symbols intervals, where each input signal is a sum of Mt input signals from a transmitter, received at said receiver after passing through a channel having a memory of ν symbols intervals, where Mr, and Mt are integers greater than zero and ν is an integer greater than 1, which channel is characterized by signal transmission coefficients that vary from each of said symbol intervals to a next of said symbol intervals, said channel expressible by a matrix H, where a row of said matrix comprises signal transmission coefficients of said channel as experienced by one of said Mr input signals at one of said N symbol intervals, and a next row of said matrix comprises signal transmission coefficients of said channel as experienced by said one of said Mr input signals at a next one of said N symbol intervals;a filter module with modifiable coefficients responsive to said reception module;a detector module responsive to said filter module; anda processor, responsive to estimates of transmission coefficients of said channel, said processor developing coefficients for said filter and applying said developed coefficients to said filter.

2. The receiver of claim 1 where said channel coefficients are selected to minimize effects of time variability of said channel.

3. The receiver of claim 1 where said reception module includes Mr receiving antennas, and said filter is an Mr input—Mr output element, where Mr is greater than 1.

4. The receiver of claim 1 where, in the course of N symbol intervals said reception module receives N symbols at each of Mr receiving antennas, thus acquiring a block of NMr signals, expressible by vector Y of length NMr, andsaid processor(1) computes, from signals received by said reception module, estimates of transmission coefficients between a transmitter having Mt transmitting antennas and said receiver, and constructing a matrix {tilde over (H)} from said transmission coefficients,(2) determines signal to noise ratio of said receiver during said block of N symbol intervals;(3) computes

5. The receiver of claim 1 further comprising: a channel estimation module, responsive to said Mr input signals received at said N symbol intervals that, based on preselected pilot signals received and detected during A of said N symbols intervals, A being an integer, estimates said channel transmission coefficients and assigns them to corresponding A rows of said matrix {tilde over (H)}, matrix {tilde over (H)} being an estimate of matrix H; andan interpolation module that computes remaining rows of said matrix {tilde over (H)} from at least some of said A rows of said matrix {tilde over (H)}.

6. The receiver of claim 5 where said interpolator applies coefficients of matrix {tilde over (H)} to said reception module to enable said reception module to detect information symbols contained in said Mr input signals received at said N symbol intervals.

7. The receiver of claim 5 where said A of said N symbol intervals are dispersed throughout said N symbol intervals.

8. The receiver of claim 5 where said A of said N symbol intervals comprise B clusters of C consecutive symbol intervals, B and C being integers, and said B clusters are dispersed throughout said N symbols intervals.

9. The receiver of claim 8 where said B clusters are essentially equally spaced throughout said N symbols intervals.

10. The receiver of claim 5 where said interpolation module computes a row n of coefficients of said matrix {tilde over (H)}, where 1<n<N by linearly interpolating between a first row of said matrix {tilde over (H)}, j, with coefficients determined from detection of pilot signals, that is closest to n but smaller than n, and a second row of said matrix {tilde over (H)}, k, with coefficients determined from detection of pilot signals, that is closest to n but larger than n.

11. The receiver of claim 5 where said interpolation module computes said each of said remaining rows of said matrix {tilde over (H)} from more than two of said A rows.

12. The receiver of claim 5 where said interpolation module employs apriori knowledge about characteristics of said transmission channel.

13. The receiver of claim 5 where said interpolation module employs linear interpolation.

14. The receiver of claim 5 where said interpolator develops a new matrix {tilde over (H)} with each successive block of N symbol intervals, {tilde over (H)} for each developed {tilde over (H)}, creates coefficients matrix Ĥ, =αĤr-1+(1−α){tilde over (H)}r, where α is a selected constant less than 1, stores matrix Ĥr, applied Ĥr to said reception module to enable said reception module to detect information symbols contained in said Mr input signals received at said N symbol intervals, and advances integer index r by 1.