Title of Invention	A WIRELESS COMMUNICATION APPARATUS AND A METHOD FOR ESTIMATING A COMMUNICATION CHANNEL IN A WIRELESS COMMUNICATION SYSTEM
Abstract	The present invention relates to a wireless communication apparatus adapted to estimate a communication channel, comprising: a correlator operative for estimating a cov~riance matrix representing a link with a transmitter based on signals received from the transmitter; a rank analysis unit coupled to the correlator and operative for estimating a rank of the covariance matrix; and a channel estimation unit coupled to the rank analysis unit and operative for generating a reduced rank channel estimate. The present invention also relates to a method for estimating a communication channel in a wireless communication system.

Title of Invention

A WIRELESS COMMUNICATION APPARATUS AND A METHOD FOR ESTIMATING A COMMUNICATION CHANNEL IN A WIRELESS COMMUNICATION SYSTEM

Abstract

The present invention relates to a wireless communication apparatus adapted to estimate a communication channel, comprising: a correlator operative for estimating a cov~riance matrix representing a link with a transmitter based on signals received from the transmitter; a rank analysis unit coupled to the correlator and operative for estimating a rank of the covariance matrix; and a channel estimation unit coupled to the rank analysis unit and operative for generating a reduced rank channel estimate. The present invention also relates to a method for estimating a communication channel in a wireless communication system.

Full Text	METHOD AND APPARATUS FOR REDUCED RANK CHANNEL ESTIMATION IN A COMMUNICATIONS SYSTEM HELD The present invention relates to wireless commtinications. More parti BACKGROUND To improve the quality of wireless transmissions, communication systems often employ multiple radiating antenna elements at the transmitter to communicate information to a receiver. The receiver may then have one or more receiver antennas. Multiple antennas are desirable, as wireless communication systems tend to be interference-limited, and the use of multiple antenna elements reduces inter-symbol and co-channel interference introduced during modulation and transmission of radio signals, enhancing the quality of communications. The modeling, and thus design, of such a system, involves estimating several parameters of the space-time channel or link between the transmitter and receiver. The number of estimated channel parameters per transmit-receiver anten;:ia pair is multiplied by the number of permutations of transmitter-receiver antenna pairs, aeating inaeasingly compUcated calculations and decreasing estimation quality. Therefore, it is desirable to have methods of channel estimation that use a reduced set of parameters. Similarly, there is a need ibr an improved method of channel estimation for radio communications systei:ns having multiple transmitter antennas. SUMMARY The presently disclosed embodiments are directed to a novel and improved method and apparatus for estimating channel parameters in a comraunication link in a wireless communication system having multiple tTans;nrutter antennas using a reduced rank estimation method. Each path from a transmitter antenna to the receiver constitutes a channel within the link. The number of channels, therefore, increases with the numbers of transmitter antennas and receiver antennas. The method exploits redundant and/or a priori knowledge within a system to simplify the channel model used as a basis for the estimation calculations and to improve the estimation quality. In one embodiment/ a covariance matrix is calculated and analyzed to deternnine if the numt'er of channel parameters may be reduced for channel estimation. If not, all parameters are estimated, otherwise a reduced rank channel model is used for the calculation of channel parameter estimates. In one aspect, a method for modeling a link in a wireless communication system, the system having a transmitter having N antennas and a receiver having M antermas, each path from one of the N transmitter antennas to the M receiver antennas comprising a channel, includes determining a matrix describing parametric relations of the link; ranking the matrix; determining if the rank is less than NxM; if the rank is less then NxM performing an extraction of a subspace of the matrix; deriving channel impulse responses for each -channel based on the extracted subspace of the matrix; and demodulating a received signal using the channel impulse responses. The matrix may be a covariance matrix describing the link, wherein the covariance matrix represents a plujrality of impulse responses between the transmitter and the receiver. Alternatively, the matrix may be a sample matrix desaibing the link- Further, detennining the matrix may include estimating a plurality of pararrieters describing at least one channel The parameters may include a distance between transnutter antermas. In one embodiment, the parameters include a transmittal angle with respect to a configuration of the trarismitter anternas. In an alternate embodiment, thie matrix describes parametric relations of the link in the frequency domain. Further, ranking the matrix may include determining an eigenvalue for the rratrix. In one embodiment, if the rank is equ^l to (NxM) a set of correlated impulse responses is applied for demodulating. In one aspect, a wireless apparatus is operative to model a Unk in a wijreless communication system by determining a matrix desaibing parametric relations of the link; ranking the matrix; determixung if the rank is less than NxM; if the rank is less then NxM perfoinming an extraction of a subspace of the matrix; deriving channel impulse responses for each channel based on the extracted subspace of the matrix; and demodulating a received signal using the channel impulse responses- In another embodiraeiit, a wireless commuiucation apparatus includes a correlator operative to estimate a covariance matrix representing a hr\k with a transmitter based on signals received from the transmitter; a rank analysis unit coupl«»d to the correlator and operative to estimate a rank of the covariance matrix; and a channel estimation unit coupled to the rank analysis unit and operative to generate a reduced rank channel estimate. The covariance matrix may represent a plurality of impulse responses between the apparatus and the transmitter. In one embodiment, the rank analysis urut is operative to deteririine an eigenvalue corresponding to the covariance matrix and is operative to compare the estimated rank of tl\e covariance matrix to a predetermined full value. In still another embodiment, a method for estimating a link in a wireless communication system includes estimating a covariance matrix for the link; determining if the rank of the covariance matrix is reducible; reducing the rank of the covariance matrix; and estimating a set of impulse responses for the link using the reduced rank covariance matrix. Additionally, the method may include determining a correlation of the channel; ranking the covariance matrix; and performing an extraction of a reduced rank matrix out of the covariance matrix. In one embodiment, a wireless comxriuni<::ation apparatus is operative within a wireless communication system having transmitter n antennas and receiver m each path from one of the to comprising channel- includes first set computer readable instructions determine covariance matrix desaibing link second rank third if less than nxm fourth perform an extraction reduced out then fifth derive chanjiel impulse responses for channel based on sixth demodulate received signal using charuiel responses. may further include equalizer in response oi wherein configuration de-:erinined by nxatrix. embodiment appa:fatus seventh correlated response.> In still another aspect, a wireless communication apparatus includes a charuxel estimation means operative to estimate a covariance matrix representing a link with a transmitter based on signals received from the transiTutter; a rank analysis imit coupled to the correlator and operative to estimate the rank of the covariance matrix; and a charmel estimation means coupled to the rank analysis unit and operative to generate a reduced rank charuiel estimate. Further in another aspect, a wireless communicatiori apparatus includes a correlator operative to estimate a covariance matrbc representing a link with a transmitter based on signals received from the transmitter; a rank analysis unit coupled to the correlator and operative to estimate the rank of the covariance matrix; and a channel estimation means coupled to the rank analysis unit and opera tive to generate a reduced rank channel estimate. In yet another aspect, a method for estimating a link in a wireless communication system includes estimating a covariance matrix for the link; determining if the rank of the covariance matrix is reducible; reducing the rank of the covariance matrix; and estimating a set of impulse responses for the link using the reduced rank covariance matrix. The method may further include determining a correlation of the channel; raiiking the covariance matrix; and performing an extraction of a reduced rank matrix out of the covariance matrix. In another embodiment, a wireless apparatus include channel estimation meariiJ operative to determine significant delays and determine a set of estimates of full dimension channel parameters associated with the sigruficant delayiv wherein each one of the set of estimates corresponds to an instance in time; eigenvalue computation means operative to determine eigenvalues of the set of estimates of the full dimension channel parameters and find any dominant eigenvalues; and channel estimation means operative to determine a set of reduced rai\k channel parameter estimates in response to the dominant eigem^alues. Further, the apparatus may include eigenvector computation mean«i operative to determine at least one eigenvector associated with one of the dominant eigenvalues of the set of estimates; wherein the channel estimation mearis uses the at least one eigenvector to project the set of estimates of the full dimei\sion channel parameters onto the subspace spanned by the at least one eigenvector. BRIEF DESCRIPTION OF THE DRAWINGS The features/ objects^ and advantages of the present invention will become more apparent from the detailed description set forth below when taken in conjunction with the drawings in which like reference characters identify correspondingly throughout and wherein: FIG. 1 illustrates configurations of wireless communication systems including multiple transmitter antennas; FIG. 2 illustrates a model of a wireless connmunication system according to one embodiment; FIG. 3 illustrates a model of a channel between transmitter and receiver in a wireless communication system; FIG. 4 illustrates the physical layout of antennas in a transmitter of a wireless communication system; FIG. 5 illustrates a flow diagram of a method of reduced rank charuiel estimation for a wireless communication system according to one embC'diment; FIG. 6 illustrates a plot of the estimation gain of one embodiment; FIG. 7 illustrates a system configuration according to one embodiment, and FIG. 8 illustrates an exemplary embodiment of a wireless communication system. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Multiple radiating antennas may be used to improve transmission quality in a wireless communications system. In the design of third generation mobile radio systems, for example, various transmitter antenna diversity techniques are presented. Multiple transmitter antennas may be used to communicate information to a receiver using a single or multiple receiver antenaa(s). Multiple antenna systems offer an improvement in quality. However, the improvement is dependent on the accuracy of the channel model used in the receiver to demodulate the transmitted information. Modeling of the transmission channel uses parameter estimates and determines an effective channel impulse response for the channel When multiple antennas are used. the modeling involves estimates of each transnnission channel for all transmitter-receiver antenna pairs. The transnaission channel from transmitter to receiver is a space-time chaniiel described generally by at least one impulse response. Often there is > little change in the channel parameters from one channel to another, such as where the channel impulse responses differ only in phase. In such a case, it may not be necessary to derive estimates of impulse responses independently for e^ich channel^ but rather some information may be reused. When channels are correlated/ a reduced rank representation of the channels may be used. Reduced rank refers to the reduced number of completely imcorrelated channels used to describe the Hnlc between transmitter and receiver. One way to ob;5erve this reduced rank is the rank reduction of the chaimel covariance matrix used to describe the mutual statistical dependencies of the different chauTvel impulse responses. Note that the reduced rank can also be realized by other parameter measures. For example, in one embodiment a sample matrix is formed of columns comprising samples ol channel impulse response estimates over time/ wherein the reduced row rank of such a sample matrix is applied as > described herein. A reduction in rank may result in a less complex filter or demodulator, Le^, reduces the number of filters and/or filter elements and/or demodulation units used In the receiver. Furthermore, the reduction of the number of estimated parameters used to characterize the channel leads to improved accuiracy of the Hiiamtel model antejinas. The wireless communication system 10 includes a transmitter 12 and receiver 16 that communicate via an air interface. A channel model 14 represents the channels for antenna pairs between transmitter 12 and receiver 16. Channel model 14 considers the channels v/ithin a link, such as the MISO link of FIG. 1. Continuing with FIG. 2, let N^^ be the number of antennas used at the transmitter 12 and N^^ the number at the receiver 16^ respectively. In general; for {.^ach significant propagation delay between transmitter and receiver, (Nj.^ • Nff^) transmission charmels exist for the pair, wherein for a significant propagation delay the received signals resemble the known transmitted signals witli high certainty. In other words, define A^^ as the number of significant propagation delays, also referred to as echoes. The (A^^.^- ^/u'^s) channel impulse response samples are then estimated to perform coherent demodulation. When the channels are uncorrelated, the (N^^ • Nf^^^ • N^) channel impulse response samples are modeled as completely uncorrelated random proc€)sses and the estimates of these channel impulse response samples may be derived independently without loss of demodulation performance* However, if the (NT^- NJ^'N^) channel impulse response samples are not uncorrelated random processes the (N^^ - NJU'^E) channel impulse response samples may be modeled as a linear combination of a smaller number N^^ of channel impulse response samples, wherein NcH estimated, and the linear trarxsformation of the NCH channel impulse response samples are resolved into the (Ny^- N^^-Nj^) channel impulse response samples, then modeling may be accomplished with the NCH channel impulse response sample estimations. This reduces the number of parameters to be estimated while increasing estimation quality, yielding an increase in the demodulation performance. Even if the exact representation of the linear transformation of the A^^;;;, channel impulse response samples into the corresponding (N^^ - ^^HC'^E) channel impulse response sanrples is not known, the modeling may still be accomplished with W^^, channel impulse response sample estimations if the subspace spanned by vectors of this linear trans formation is known or can be estimated. This principle is referred to as "Reduced Rank Channel Estimation." The transrormation of the /\^C-A uncorrected channel impulse responses into the i^Tx' ^RX'^E) correlated channel impulse responses can depend on factors including; but not limited to, the antenna configuration, anterma patterns, polar:ization characteristics, propagation conditions and more. In some cases the transformation might be known a priori, in other cases it can be derived or estimated, for example by angle-of-arrival estimation. The subspace spanned by th In one embodiment, the rank reducing ti-ansformation is known a priori or is stimated. In other words, the mapping of the N^^ channels onto the i^Tx'^R.^) channels is ascertainable. The reduced rank channel is then estimated using the ascertained transformation. When desired, an equivalent full dimensional channel model may then be derived from the reduced rank estimate by transforming the reduced rank estimate back to the larger dimension. In an alternate embodiment, the rank reducing transformation is not directly known, but the subspace spanned by the transformahon may be extracted from the dominant eigenvectors of the channel covariance matrix. Note that the subspace may be referred to as the signal subspace or the channel subspace. The process involves first estimating a channel covariance matrix and finding the dominant eigenvalues. By deternuning the associated eigenvectors which span the channel subspace^ the process projects the conventional channel estimate into the channel subspace, yielding a reduced rank channel model with reduced estimation errors* If desired, the reduced rank model may be transformed back into an equivalent full dimension channel model. FIG. 3 illustrates a model 18 of a MIMO channel for continuous time having a linear MIMO filter 20 with N^^ inputs and A^^^ outputs. The linear MIMO filter 20 is defined by the //^^xyv^ matrix H(t) comprising of linear functions /z^.(r), / = I-^T, , j = I'-A'/u • Generally, ft^.(r), / = 1.../^^^,;" = I...A^^are unknown linear functions. The linear MIMO filter 20 represents the (Z/;-^ - Nf^) radio channels through which the N^^^ transmit signals pass to the A^;^^ receiver antennas. These radio channels are characterized by their channel impulse responses hfj(t), i-L..Nj.^, y = 1.,.A^^. The input signal to the model, ic(t), is a (//y.^ >:1) column vector representing the A^^.^ band-linuted transmit signals, and the output signal from the model, ^(r), is a (iV^j, x 1 )column vector, sampled at r = r, 27..., as illustrated by switch T, where the bandwidth of the transmitted signals is less or equal to l/T, The received signals contain additive perturbation signals represented by the N^^^xl column vector ^(r), introduced due to noise or co-channel interference. The additive perturbation signals are added at summation nodes 22. The relation between the input signals JfCO / the FIG. 4 illustrates the physical configuration of antennas at the transmitter of an exemplary embodiment modeled as in FIG. 2. A reduced rank method is applied to estimate the link represented by channel model 14, having a transmitter 12 configured with the four (4) antenr.as, each spaced at a distance "d." The specifics of the configuration and model are discussed hereinbelow. Note that the estimation procedure is performed at the receiver 16. A reference direcdon is given by the horizontal line. Angles of transmission are measured with respect to this reference. The angle "a" corresponds to an angle of a propagation path with respect to the reference within a 2-D plane as illustrated. A rartge of angles with respect to the reference is also illustrated. The following method is used at the receiver 12 in system 10 to estimate the link. FIG. 5 illustrates a flow diagram of an exemplary method of chaimel estimation used to process signals in a receiver unit in accordance with one embodiment. Process flow begins by searching for significant propagation delays in the channel, le. searching for significant echoes at step 40,_ In one embodiment the process involves a sliding correlation of the received signals with known transmitted signals or known components of the transmitted signals. Correlation refers to the degree with which the received signals are related to the known transmitted signals, wherein a perfect correlation proves a relationship between the signals with high confidence. For time-shifted signals, wherein sliding delays are used to shift the received signals in time, a resultant sliding correlation provides the degree of certainty with which the time-shifted signals resemble the known transmitted, signals. Thus in the wireless system context, sliding correlation relates to the synchronization of known signals transmitted by the Tx antennas with time-shifted versions of the received signals. The exemplary embodiment of reduced rank channel estimation uses sliding correlation of the received signals with known transnutted signals to estimate the number N^ and the values r^f^.K^r^v^ of significant propagation delays, i,e, delays for which the received signals shifted back by these delays in time resemble the known transmitted signals with high certainty. The procedure of sliding correlation in order to find significant propagation delays is alSD known as 'searching" in CDMA systems. The method then estimates parameters for multiple observable channels betv/.een the //^.^ transmitter antennas and the Nj^ receiver antennas at step 42. The cliannels are radio network connection pairs coupling at least a portion of the Nj^ transmitter antennas to at least a portion of the Nf^ receiver antennas. In the exemplary embodiment, there is a connection between each transmitter 12 antenna and each receiver 16 antenna, resulting in (N^.^ • /V^) channels. The parameters describing the multiple charvnels are those characteristics that impact the impulse responses of the channels. Assuming that N^ significant propagation delays (echoes) exist between transmitter and receiver, i^Tx' ^Rs'^e) complex samples of the {N-r^ - N^^^) channel impulse responses could be used as a set of parameters describing the multiple channels. This set of parameters is denoted by a {(Nj.^' Nj^^N^)xl) vector termed n herein. The relation between ^{t), h, ^(/),and ^(r) is developed hereinbelow. Since the output signals ^(/) are sampled at a sampling rate of l/T, vectors containing the discrete time samples can represent segments of a finite duration of the: continuous time signals. For the sake of simplicity, the received signals ^(r) are described herein by a disaete time representation over a firute duration of time t^O,T,K ,{N^-1)T, where N^ is the number of samples taken over time. Therefore, the following abbreviations are used. Each discrete The .second step in the flow diagram in FIG- 5 at step 42, is to repeatedly process estiniates for a set of parameters characterizing the multiple channels between transnutter and receiver. For the above-described mathematical representation of tl\e channel model, this may be equivalent to processing estimates h^"\ n = lK iV^ of the vector li in (14) for N^ different points in time. A conventional method uses the correlation of the received signals, shifted back in time by certain delays, with known transmitted signals, such as pilot signals specific to the transmitter antermas, or predeteriruned training sequences. As the significant propagation delays t^.tiX ,r^^ are already determined in step 40, tl\e exemplary embodiment of reduced rank channel estimation uses the correlation of known transmitted signals v^th versions of the received signals, shifted back in time by r^r^^K ^r^^^, to generate a channel model, such as charu\el model 14 of FIG. 2, characterized by the vector h. If the noise vector K represents spatial and temporal white perturbation, wherein the noise covaiiance matrix is given as R^ =/ftft") = o'^ •I^'^'^', and if the matrix A comprises of the a priori known signals, such as pilot symbols of a CDMA systei-n, channel estimates obtained by correlation can be described by shall hold. As illustrated in FIG. 5, a covariance matrix of the channel parameters is estimated at step 44. Covariance measures the variance of one random variable with respect to another. In this case, the covariance matrix describes the variance of the various channel parameters with respect to each other. According to the above-desaibed mathematical representation of the channel model, step 44 corresponds to processing an estimate R,^ of the chaimel If the MIMO channel has a reduced ranJk wherein N^h As a consequence/ the rank of the channel covariance matrix i?^ is equal to A'^^. Given (20), and assuming that correlation according to (17) is used to derive the channel inipulse response estimates n, the covari yields only N^^^ non-zero eigenvalues, where A is a diagonal matrix containing the eigenvalues and £ is a square matrix containing the eigenvectors of U, Rf, shares the eigenvectors with R^^^^B-R^ •5"/^^'^ Since A is a diagonal matrix with only Np, non-zero elements, (N^j, - Nf^ - A^^)-//^^ eigenvalues of Rf^ are constant, and N^,^ eigenvalues of R^ are larger than the former ones. Tl\es€s larger eigenvalues are termed dominant eigenvalues in the sequel. With a diagonal matrix A^, contairung all dominant eigenvalues of the estimated channel covariance matrix, the matrix E^, containing the corresponding eigenvectors, and with the matrix S^^, containing the remairnng eigenvectors, (23) becomes Therefore, the matrix E^ contains the eigenvectors sparming the channel or signal subspace. The estimated covariance matrix ^^ is then ranked at step 46, meaning that tlie number of dominant eigenvalues is estimated. The rank is compared to I maximum value "MAX" at step 48. MAX is equal to the total number of ^stiirated channel parameters in the vector h . In other words, MAX is equal to i^Tx ^fu'^s)' A^ many of the mechanisms impacting correlation, such as the directionality of the propagation paths, do not change quickly over time, the correlation characteristics may be estimated by averaging over rather long time inten^als in comparison to the inverse fading rate of the channel(s). The rank of the covariance matrix determines whether the i^Tx' ^KX'^E) channel parameters describing the (A'.^^ - N^^) existing transiTiission channels can be modeled as a linear combination of a smaller numt^er A'^^ of equivalent uncorrelated channel parameters If a reduced rank is available, the channel subspace E^ of the estimated covariance matrix Rf^ is derived at step 52, Note that instead of using the estimated covariance matrix R^, the rank of R^ and the channel subspace E^ can also be derived from the matrix of channel parameter estimates effectively projecting the originally estimated channel parameters into the channel subspace. This projection into the channel subspace reduces the estimation error. If a reduced complexity demodulator is used in the receiver, which uses the reduced rank channel, i.e., takes only a reduced number of channel parameters into account for demodulation, the estimates of (26) may be directly used in the demodulator for coherent demodulation. In other words, processing would flow directly from step 54 to step 58, or at a minimum step 58 would used the reduced rank estimates. If a conventional receiver, designed for the full rank channel model, is to be ur>ed, the estimates g^"^ may be transformed back into the full dimensional wherein the factor /?"^^" is used to make the estimate xmbiased. Note that the estimate of the channel subspace E^ may be updated continuously by using a sliding time v^indow for the estimates R^ or X^, respectively. This eliminates the delay of waiting for a new complete sample set, by using a portion of the previous sample set with incrementally time-shifted new values. If rank reduction is not possible, processing continues to use the full rank of the system to model the channel at step 50. In this case the method estinriates the (yVr' ^HX'^E) channel parameters independently from each othei'. Once the system is modeled^ signal demodulation continues at step 58, The MISO path illustrated in FIG. 1 is provided as an exemplary embodiment. As illustrated, the transmitter, Tx, has four (4) radiating anteimas (iVy,^ =4) and the receiver, Rx, has one (1) anterma (iV^ =1). For simplicity, several assumptior\s allow a straightforward analysis demonstrating the applicability of the exemplary embodiment to modeling a system as illustrated in FIG. 1. First, the example assumes that each Tx antenna transn\its a pilot signal specific to that antenna, wherein the antenna-specific pilot signal is time-aligned and orthogonal to the pilot signals of the other Tx antennas. Second, assume the channels are frequency non-selective fa^Jing chaiuiels, each made up of a large number, P, of radio network paths* The paths each have approximately a same nm length and a same attenuation. The secord assumption ensures that the relative propagation delay is smaller than the inverse of the transmission bandwidth. The propagation delay of two radio paths is typically due to differences in run length. Third, the channel model is restricted to 2-D propagation, U. all effective radio paths are located in a 2-D plane. See FIG. 4. Additionally, the geometry of the effective radio paths at the transmitter is assumed to be time-invariant. wherein each path departure angle, measured with respect to a reference direction of Tx, are concentrated around an average angle, a . The radio path angles are Gaussian distributed having mean a and a standard deviation a. For one sinnulation, a is selected randonxly between -60 and +60 degrees. The standard deviation a is assumed to be square root of two degrees* Fourth, the arrival paths at Rx are assumed uniformly distributed between 0 and 360 degrees to consider local scattering. Fifth, no line of sight exists. Sixth, assume a specific phase and Doppler shift for each path. The path-specific phase is selected randonily according to a uniform distribution between 0 and In - Additionally, the path-specific phase is adjusted for each Tx anterma according to the geometrical antenna configuration, i.e., the antenna location with respect to a reference point. For phase adjustment, assume object scattering is considered in the far field- The channel-specific Doppler shifts are generated according to a uniform distribution of the angles of arrival paths at Rx, a carrier frequency and a predetermined R: Given the exeinplary system as detailed, application of the process of FIG. 5 provides a channel model having a time variance according to the classic Doppler spectrum. It is possible to consider an antenna-specific radiation pattern. Witih this charmel model, the charmel impulse responses for the charmels seen through the different transmitter antermas can be generated using the same set of radio paths, thus, introducing realistic correlation in the fading of the different channels. On the receiving side of the air-interface, at the single anteima of Rx, the method derives an impulse response estimate for each of the four transnaission chaimels, le„ the four radio network coimectior\s between Tx antennas and the JU anleima. The estimate is based on the a-priori knowledge of spreading codes used to generate the anterma-specific pilot signals associated with each Tx anteinna. Referring again to FIG. 4, in the geographical configuration of antennas at T>:., the antennas are positioned in a line having constant spacing d between neighboring antennas, wherein d =A, i.e., antennas are spaced one wavelength apart. Note that Rx has a single omnidirectional antenna- A total number of effective radio paths is considered with P = SQ, Channel specific variables, cc^, f^, and C^^ / represent, respectively, the angle measured from the reference line, the Doppler shift and the phase. The equation describing the channel impulse response for a channel between Tx antenna n and the Rx antenna is given as where h(j) is the equivalent channel imptilse response for a equivalent isotropic Tx antenna at the reference point. In this case the channel impulse responses for the different Tx antennas orUy differ by a complex factor, i.e., the channels ai'e completely correlated. The steering vector is then defined as The four (4) channel impulse responses seen from the Tx antennas are then copies of the channel impulse response k(t)r weighted by four different complex factors, which means, the vector n{t) is a linear transformation of the scalar h(t) given by i,e,, the vector ^ in the linear transformation of (19) is in this example equal to the scalar h{t) and the matrix B is equal to the vector ^{a). This means the channel covariance matrix R,, =/ft'A"\ is equal to R^ ==5(ff)5(a)"/\|/i(f)p\ in this example. If the steering vector a{s) is knov^m, such as a-priori knowledge of the antenna configuration and the radio path direction a, it is sufficient to estimeite the scalar h{t) and either calculate an estimate for h{t) using the linear transformation with a{a) or use the estimate of Ht) and ^{a) directly for demodulation. Note that for the case wheni'(a) is known it may be sufficient to estimate h{t) and then compute an estimate of h{t) from the scalar estimate of h{t). If the demodulator is designed such tihat the channel consists of a single scalar, i.e^, the demodulation considers a{a), then it is possible to denriodulate using ^{a) and the scalar channel The antenna-specific pilot signals at the transnutter are termed x^{t), n = IK A^T-^, and the relationship is defined by The pilot signals are made up of segments, each having a duration T^, refen'ed to as the pilot symbol duration, over which the pilot signals are orthogonal, and wherein the following holds The pilot vector is defined as and the receiver noise signal z{t) represents white Gaussian noise. The signal recei\'ed by the single Rx antenna is described as Conventionally/ correlating the received signal with the four (4) pilot sequences derives a set of four (4) channel estimates. Wherein the pilot signals are orthogonal over a pilot symbol period, this estimation is then repeated at the pilot symbol rate. Such a correlation procedure is generally referred to as "integrate and dump" and may be expressed as wherein h^^l is a vector made up of conventional, i.e., integrate and dump, charuiel estimates derived from the n-th pilot symbol. If (34) is transformed into a disaete time representation, by putting Ny^ =TJT samples of the pilot signals x^t) into the columns of the matrix A, N\ samples of the noise signal z{t) :;nto the vector K, and N^ samples of the received signal y[t) into the Fron\ this scalar estimate, using the linear transformation, a new estimate of the channel impulse vector is generated as Igno::ing charmel variations within one pilot symbol (44) becomes If ^(a) is not known a-priori, it may be estimated using the covariance matr::x given by with PA being the average power of the scalar channel impulse response h(t). The covariance matrix R^ can be approximated as whicJi averages the vector with the conventional channel impulse response estimates over a number^ N^, of pilot symbols. For the case without noise and having an angular spread equal to zero, k^ is rairk one (1) and the vector i\a) spax\s K^, Thus (47) reduces to Note that the normalized vector ^(^)/\|\|i?(a\| spans /?„. For a noisy case with sufficient low noise power and sufficient low angular spread, R^^ is still dominated by one eigenvalue. Therefore, the process perfo::ms an eigenvalue decomposition of k^^ When one eigenvalue is much larger than all other eigenvalues, it is an indication that the angular spread around ^{^) was rather small. Therefore, as ^^^ is the eigenvector corresponding to the largest eigenvalue of R^, the approximation becomes Note that the vector v^^ in this example is equal to the channel subspace E^ - In general, the estimate R^ is used to determine whether the rank of the channel estimation covariance matrix can be reduced. If Rf,is full rank, the channel estimation problem is not reduced to a smaller dimension. According to the exemplary embodiment, orthogonal pilot signals of binary chips have a chip rate of 1.2288 Mcps, and a pilot symbol duration of 64 chips. With this channel model, a received signal, including white Gaussian noise; is generated for 4000 consecutive pilot symbols having a pilot Signal-to-Noise Ratio (SNR). From the received signal, 4000 conventional vector estimates, h^^l, are generated- The thus generated covariance matrix R,^ is avera;5ed over these 4000 consecutive conventional channel estimates. In the exemj?lary embodiment, the process takes approximately 208.3 ms. After extracting the eigenvector corresponding to the maximal eigenvalue of Rf^, the matrix R^ is calculated. Subsequently/4000 ne\v vector estimates hj^^l are produced according to Using the exemplary embodiment, iterations are repeated A/„p = 50 times. Over the 50 iterations the transmitter angles are varied such that a is uniformly distributed within (+/- 60) degrees, while the angular spread remains constant, having a standard deviation of square root of two i^f2) degrees. Additionally, the channel parameters for a given pilot SNR are varied. The varied parameters represent radio path direction(s), path-specific phase, and path-specific Doppler shift, for a certain pilot SNR. An equal number of itera!ions is performed for different pilot SNR values. A comparison of the quality of the set of conventional estimates to the set of new vector estimates, with respect to the reduction factor of the mean squared estimation error that is averaged over time and iterations, is made using the estimation gain given as For the exemplary embodiment, FIG. 6 illustrates the estimation gain in dB as a function of the pilot SNR. Wherein the pilot SNR is defined as the ratio of th(» average energy per pilot chip E^ of one pilot signal received at the single anterna receiver to the received noise power dex\sity I^ in dB. The upper limit for the estimation gain is determined by the number of trans;i\it antennas, which is illustrated in FIG. 6 as 6dB. As illustrated in FIG. 6, the estimation gain approaches the upper limit even though the assumed angular spread is not zero and the received signal is severely corrupted by noise The reduction of the estimation gain with increasing pilot SNR is due to the non-zero angular spread. Although the chaimel impulse responses are not completely correlated, the derivation of the impulse response n^ assumes this property. For larger angular spreads, a smaller estimation gain is expected. For small angular spreads, the estimation gain appears considerable. Note that in general, for residential and suburban environments a standard deviation of one (1) to two (2) degrees is frequently observed. Note also that it is possible to evaluate the perfoimance improvement of the reduced rank channel estimation method using a Monte-Carlo-simulation to derive the reduction of estimation errors of the channel impulse responses as compared to conventional channel estimation using independent correlators. Reduced rank channel estimation for systeiiis using multiple transmitter antertnas allows improvement of the channel estimation quality under certain propagation conditions with limited diversity due to correlated fading. As the mechanisms affecting correlation^ such as the directionality of the radio wave propagation, change relatively slowly over time, the correlation characteristics may l^e estimated by averaging over extended time intervals. This is in contrast to thei time intervals associated with inverse fading rate of the channel and thus allows improved accuracy in estimating the correlation characteristics. Reduced rank channel estimation for multiple transmitter antennas is also applicable to frequency-selective channels by computing either separate estimates of the correlation characteristics or by computing estimates of the correlation characteristics across all propagation delays. Separate estimates A refers to computation of i?^, for each propagation delay. Reduced rank chanriel estimation is then performed taking into accoimt each delay occurring in the frequency-selective channel impulse response. In an alternate embodiment, wherein additional information, such as the antenna config;uration at the transmitter^ is known a-priori, the step of estimating the linear transformation of the reduced number of uncorrected channels into the larger number of correlated channels may be more accurate. Additionally^ the reduced rank estimation process may be extended to cases with more than one receiver antenna. In ttiis case/ the estimation is performed for the MIMO channels/as illustrated in FIG, 1, While the present example involves a system emplcying coherent demodulation, reduced rank channel estimation., as described herein is also applicable to communication systems employing noncoherent demodulation. A receiver 100 according to one embodiment of the present invention is illustrated in FIG. 7. The receiver 100 has a single anteiuia 102 that receives signalr> from a transmitter having multiple antennas. The received signals are first processed by the preprocessor 104. The signals are then provided to a correlator 106, which is used as a sliding correlator for searching and as a correlator for the significant delays for channel estimation. In an alternate embodiment the delays are determined in software without use of a correlator. The outputs of the correlator 106 are used to provide an estimate of the CO variance matrix. In one embodiment, the correlator 106 is made up of fingers to form a rake, having one finger for each combination of transmitter antenna, receiver antenna and significant delay. The estimates are provided to the central processor 112 via bus 116. The processor 112 stores the: channel paraiTieter estimates in memory 114 so that the estimates may be used to derive the charmel covariance matrix averaged over time. From memory 114, the estimated covariance matrix is provided to the rank analysis and subspace estimation unit 108 for eigenvalue decomposition. If orie or more eigenvalues dominate the others, the channel subspace is estirriated by computing the eigenvectors that correspond to the dominant eigenvalues. The eigenvectors spanning the channel subspace are written to memory for further use in the charmel subspace projection unit 109 where redu<:ed rank charmel parameter estimates are produced by computing the projection of a original per estimation time> internal onto the charmel subspace, yielding A^^^ reduced rank charmel parameter estimates per estimation time interval. The results of the channel subspace projection unit 109 are written to memory for use in the demodulator 110. Optionally the channel subspace projection unit 109 could generate equivalent full dimension channel parameter estimates, by re-transforming the A^^^ reduced rank charmel parameter estimates into (A^^^ ^ N^^-N^) equivalent full dimension charmel parameter estimates per estimation time interval. For example in a conventional RAKE-receiver design for the full rank channel model, the number of rake fingers for a full rank demodulator would be {Nj^' ^Rx • ^£)• A full rank demodulator would then use the (A^^^ • Nj^^^N^) original charmel parameter estimates for the finger coefficients. A reduced complexity demodulator could eventually use orUy N^^ RAKE fingers using the A^^;, reduced rank charmel estimates as coefficients. However^ since the receiver would generally be designed in anticipation of a worst case situation, I.e., wherein (A^;.^ • A^^t '^E) f^rigers are implemented, it would be sufficient to compute (/Vrx' ^/u '^E) correlated channel parameter estimates with improved estimation quality over of the iV^^ reduced rank channel parameter estimates. The rank analysis and subspace estimation unit 108 and the subspace projection unit 109 may be implemented in a Digital Signal Processor (DSP), dedicated hardware, software, firmware, or a combination thereof. Modules within receiver 100 may be incorporated together, and are illustrated as separate blocks for clarity based on function. An exemplary configuration of one embodiment is illustrated in FIG. 8 for a system having four (4) transmitter antennas and two (2) receiver antennas. Three (3) transmission paths are illustrated and labeled 1,2 and 3. The points of reflection for paths 1 and 2 are both on a same ellipse, wherein the ellipse is fotmtid such that Tx and Rx are the focal points. Note that the ellipse is superimposed on the illustration of the physical layout of the system. Path 3 falls outside of the illustrated ellipse. Paths 1 and 2 have the same significant delay, TI, with respect to the receiver, while path 3 has a significant delay, X2 different from TJ. The path delay is a function of the corifiguration of the antennas as well as the environment of the system. As illustrated, the four (4) transmitter antennas and the two (2) receiver antennas result in eight (8) chanr..els. Each of the path delays, TI and X2, produce an echo, wherein (iV^ = 2). The dimension of the covariance matrix is given as (JV,.^' Nj^-N^) or sixteen (16) corresponding to the {Nj^ • ^A^-N^) channel impulse response samples. Therefore, the full rank channel parameter vector is a 16-dimension vector. Using the rank reduction methods described herein, the rank of the channel estimation may be reduced to three (3) dimensions/ corresponding to paths 1,2, and 3, wherein {N^^=^2). Note that where the mapping of the Ncs transmission paths to the (A^r^ • N,^N^) channel impulse response samples is not known, the subspace may be extracted from configuration information. If the location and characteristics, such as direction and directionality, of the antennas are known, the information may be used to generate an array response or steering vector. Using the steering vector and path direction information, which is also extractable using subspace algorithms, the angle of tranf-mission, a, is estimated. If the antenna configuration has a fixed deployment the angle of transmission is calculable. A vector is formed including an angle of transmission for each transmitter antenna. Similarly, an arrival angle vector is formed considering the receiver antermas. A linear transformation for the mapping of the NCH transnrdssion paths to the (^7> ^RX'^E) channel impulse response samples is constructed using this information from both the transmitter and receiver configurations- This provides the matrix B as given in (19) hereinabove describing the linear transformation. The covariance matrix is derived therefrom as in (20) hereinabove. The process then proceeds as for the case where corresponding information is obtained from a priori knowledge. VVhile one embodiment has been described herein with respect to the time domain, an alternate embodiment performs a rank reduction of the covaiiance matrix or a sample matrix in the frequency domain. If the parameters and equations are developed in the frequency domain, the process to estimate the chaimel then incorporates the frequency domain values. The previous description of the preferred embodiments is provided to enable any person skilled in the art to make or use the present invention. Tlie various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without the use of the inventive faculty. 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 principles and novel featuies disclosed herein. CLAIMS I (WE) CLAIM: 1. A method for modeling a link in a wireless communication system, the system having a transmitter having N antennas and a receiver having M antennas, each path from one of the N transmitter antennas to the M receiver antennas comprising a channel/ the method comprising: determining a matrix describing parametric relations of the link; ranking the matrix; determining if the rank is less than NxM; if the rank is less then NxM performing an extraction of a subspace of the matrix; deriving channel impulse responses for each channel based on the extracted subspace of the matrix; and demodulating a received signal using the channel impulse responses. 2* The method of claim 1^ wherein the matrix is a covariance matrix descri bing the link, wherein the covariance matrix represents a plurality of impulse responses between the transmitter and the receiver. 3. The method of claim 1, wherein the matrix is a sample matrix describing the link. 4. The method of claim 1, wherein the step of determining the matrix further comprises: estimating a plurality of parameters describing at least one charmel. 5. The method of claim 4, wherein the parameters include a distance between transmitter antennas. 6. The method of claim 4, wherein the parameters include a transmittal angle with respect to a configuration of the transmitter antennas. 7. The method of claim 4, wherein the determining the matrix comprises estimating the matrix, 8. The method of claim 1, wherein the matrix describes parametric relations of the link in the frequency domain. 9. The method of claim 1, wherein the ranking the matrix, further comprises: deternmining an eigenvalue for the matrix, 10. The method of claim 1, wherein if the rank is equal to (NxM) a set of correlated impulse responses is applied for demodulating. 11. A wireless apparatus operative to perform the method of claim 1. 12- A wireless communication apparatus, comprising: a correlator operative to estimate a covariance matrix representing a link with a transmitter based on signals received from the transmitter; a rank analysis unit coupled to the correlator and operative to estimate a rank of the covariance matrix; and a channel estimation unit coupled to the rank analysis unit and operative to generate a reduced rank channel estimate. 13. The apparatus of claim 12, wherein the covariance matrix represents a plurality of impulse responses between the apparatxis and the transmitter. 14. The apparatus of claim 12, wherein correlator is operative to determine a correlation of at least two channels. 15. The apparatus of claim 14, wherein the ranJ: analysis unit is operative to determine an eigenvalue corresponding to the covariance matrix. 16. The apparatus of claim 15, wherein the ranfc. analysis unit is operative to compare the estimated rank of the covariance matrix to a predetermined full value, 17. A method for estimating a link in a wireless communication system, the method comprising; estimating a covariance matrix for the link; deterrxuning if the rank of the covariance matrix is reducible; reducing the rank of the covariance niatrix; and estimating a set of impulse responses for the link using the reduced rai\k covariance matrix. 18. The method of claim 17, further comprising: deternuning a correlation of the channels- ranking the covariance matrix; and perfornrung an extraction of a reduced rank matrix out of the covariance matrix. 19. A wireless communication apparatus operative within a wireless communication system having a transmitter having N antennas and a receiver having M antennas, each path from one of the N transmitter antennas to the M receiver antennas comprising a channel, the apparal-us comprising: a first set of computer readable instructions operative to determine a covariance matrix describing the link; a second set of computer readable irtstructions operative to rank the covariance matrix; a third set of computer readable instructions operative to determine if the rank is less than NxM; a fourth set of computer readable instructions operative to perform an extraction of a reduced rank matrix out of the covariance matrix if the rank is less then NxM; a fifth set of computer readable instructions operative to derive channel impulse responses for each channel based on the reduced rank covariance matrix; a sixth set of computer readable instrucfcons operative to demodulate a received signal using the channel impulse responses. 20. The apparatus of claim 19, further comprising: an equalizer operative in response to the sixth set of computer readable instructions, wherein a configuration of the equalizer is deternruned by the rank of the covariance matrix. 21. The apparatus of claim 19/ further comprising: a seventh set of computer readable instructions operative to derive a correlated channel impulse response. 22. A wireless communication apparatus, comprising: a charmel estimation means operative to estimate a covariance matrix representing a link with a transmitter based on signals received from the transmitter; a rank analysis unit coupled to the correlator and operative to estimate the rank of the covariance matrix; and a channel estimation means coupled to the rank analysis unit and operative to generate a reduced rank channel estimate. 23. A wireless communication apparatus, comprising: a correlator operative to estimate a covariance matrix representing a link with a transmitter based on signals received from the transmitter; a rank analysis unit coupled ^o the correlator and operative to estimate the rank of the covariance matrix; anci a channel estimation means coupled to tl\e rank analysis unit and operative to generate a reduced rank channel estimate. 24. A method for estimating a link in a wireless communication system, the method comprising: estimating a covariance matrix for the link; determining if the rank of the covariance matrix is reducible; reducing the rank of the covariance matrix; and estimating a set of impulse responses for the link using the reduced rank covariance matrix. 25. The method of claim 24, further comprising: determining a correlation of the channel; ranking the covariance matrix; and performing an extraction of a reduced rank matrix out of the covariance matrix; 26. A wireless apparatus, comprising: charmel estimation means operative to determine sigruficant delays and determine a set of estimates of full diinension channel parameters associated with the significant delays, wherein each one of the set of estimates corresponds to an instance in time; eigenvalue computation means operative to deterrxune eigenvalues of the set of estimates of the full dimension channel parameters and find any dominant eigenvalues; and channel estimation means operative to determine a set of reduced rank channel parameter estimates in response to the dominant eigenvalues. 21. The wireless apparatus of claim 26, further comprising: eigenvector computation means operative to determine at least one eigenvector associated with one of the dominant eigenvalues of the set of estimates; wherein the channel estimation means uses the at least one eigenvector to project the set of estimates of the full dimension channel parameters onto tlie subspace spanned by the at least one eigenvector. A method for modelins a link in a wireless communication svstem substantially as herein described with reference to the accompanying drawings. A wireless communication apparatus substantially as herein described with reference to the accompanying drawings.

Full Text

METHOD AND APPARATUS FOR REDUCED RANK CHANNEL ESTIMATION IN A COMMUNICATIONS SYSTEM
HELD
The present invention relates to wireless commtinications. More parti BACKGROUND
To improve the quality of wireless transmissions, communication systems often employ multiple radiating antenna elements at the transmitter to communicate information to a receiver. The receiver may then have one or more receiver antennas. Multiple antennas are desirable, as wireless communication systems tend to be interference-limited, and the use of multiple antenna elements reduces inter-symbol and co-channel interference introduced during modulation and transmission of radio signals, enhancing the quality of communications. The modeling, and thus design, of such a system, involves estimating several parameters of the space-time channel or link between the transmitter and receiver.
The number of estimated channel parameters per transmit-receiver anten;:ia pair is multiplied by the number of permutations of transmitter-receiver antenna pairs, aeating inaeasingly compUcated calculations and decreasing estimation quality. Therefore, it is desirable to have methods of channel estimation that use a reduced set of parameters. Similarly, there is a need ibr an improved method of channel estimation for radio communications systei:ns having multiple transmitter antennas.

SUMMARY
The presently disclosed embodiments are directed to a novel and improved method and apparatus for estimating channel parameters in a comraunication link in a wireless communication system having multiple tTans;nrutter antennas using a reduced rank estimation method. Each path from a transmitter antenna to the receiver constitutes a channel within the link. The number of channels, therefore, increases with the numbers of transmitter antennas and receiver antennas. The method exploits redundant and/or a priori knowledge within a system to simplify the channel model used as a basis for the estimation calculations and to improve the estimation quality. In one embodiment/ a covariance matrix is calculated and analyzed to deternnine if the numt'er of channel parameters may be reduced for channel estimation. If not, all parameters are estimated, otherwise a reduced rank channel model is used for the calculation of channel parameter estimates.
In one aspect, a method for modeling a link in a wireless communication system, the system having a transmitter having N antennas and a receiver having M antermas, each path from one of the N transmitter antennas to the M receiver antennas comprising a channel, includes determining a matrix describing parametric relations of the link; ranking the matrix; determining if the rank is less than NxM; if the rank is less then NxM performing an extraction of a subspace of the matrix; deriving channel impulse responses for each -channel based on the extracted subspace of the matrix; and demodulating a received signal using the channel impulse responses. The matrix may be a covariance matrix describing the link, wherein the covariance matrix represents a plujrality of impulse responses between the transmitter and the receiver. Alternatively, the matrix may be a sample matrix desaibing the link-
Further, detennining the matrix may include estimating a plurality of pararrieters describing at least one channel The parameters may include a distance between transnutter antermas. In one embodiment, the parameters include a transmittal angle with respect to a configuration of the trarismitter

anternas. In an alternate embodiment, thie matrix describes parametric relations of the link in the frequency domain.
Further, ranking the matrix may include determining an eigenvalue for the rratrix. In one embodiment, if the rank is equ^l to (NxM) a set of correlated impulse responses is applied for demodulating. In one aspect, a wireless apparatus is operative to model a Unk in a wijreless communication system by determining a matrix desaibing parametric relations of the link; ranking the matrix; determixung if the rank is less than NxM; if the rank is less then NxM perfoinming an extraction of a subspace of the matrix; deriving channel impulse responses for each channel based on the extracted subspace of the matrix; and demodulating a received signal using the channel impulse responses-
In another embodiraeiit, a wireless commuiucation apparatus includes a correlator operative to estimate a covariance matrix representing a hr\k with a transmitter based on signals received from the transmitter; a rank analysis unit coupl«»d to the correlator and operative to estimate a rank of the covariance matrix; and a channel estimation unit coupled to the rank analysis unit and operative to generate a reduced rank channel estimate. The covariance matrix may represent a plurality of impulse responses between the apparatus and the transmitter. In one embodiment, the rank analysis urut is operative to deteririine an eigenvalue corresponding to the covariance matrix and is operative to compare the estimated rank of tl\e covariance matrix to a predetermined full value.
In still another embodiment, a method for estimating a link in a wireless communication system includes estimating a covariance matrix for the link; determining if the rank of the covariance matrix is reducible; reducing the rank of the covariance matrix; and estimating a set of impulse responses for the link using the reduced rank covariance matrix. Additionally, the method may include determining a correlation of the channel; ranking the covariance matrix; and performing an extraction of a reduced rank matrix out of the covariance matrix.

In one embodiment, a wireless comxriuni<::ation apparatus is operative within a wireless communication system having transmitter n antennas and receiver m each path from one of the to comprising channel- includes first set computer readable instructions determine covariance matrix desaibing link second rank third if less than nxm fourth perform an extraction reduced out then fifth derive chanjiel impulse responses for channel based on sixth demodulate received signal using charuiel responses. may further include equalizer in response oi wherein configuration de-:erinined by nxatrix. embodiment appa:fatus seventh correlated response.> In still another aspect, a wireless communication apparatus includes a charuxel estimation means operative to estimate a covariance matrix representing a link with a transmitter based on signals received from the transiTutter; a rank analysis imit coupled to the correlator and operative to estimate the rank of the covariance matrix; and a charmel estimation means coupled to the rank analysis unit and operative to generate a reduced rank charuiel estimate.
Further in another aspect, a wireless communicatiori apparatus includes a correlator operative to estimate a covariance matrbc representing a link with a transmitter based on signals received from the transmitter; a rank analysis unit coupled to the correlator and operative to estimate the rank of the covariance

matrix; and a channel estimation means coupled to the rank analysis unit and opera tive to generate a reduced rank channel estimate.
In yet another aspect, a method for estimating a link in a wireless communication system includes estimating a covariance matrix for the link; determining if the rank of the covariance matrix is reducible; reducing the rank of the covariance matrix; and estimating a set of impulse responses for the link using the reduced rank covariance matrix. The method may further include determining a correlation of the channel; raiiking the covariance matrix; and performing an extraction of a reduced rank matrix out of the covariance matrix.
In another embodiment, a wireless apparatus include channel estimation meariiJ operative to determine significant delays and determine a set of estimates of full dimension channel parameters associated with the sigruficant delayiv wherein each one of the set of estimates corresponds to an instance in time; eigenvalue computation means operative to determine eigenvalues of the set of estimates of the full dimension channel parameters and find any dominant eigenvalues; and channel estimation means operative to determine a set of reduced rai\k channel parameter estimates in response to the dominant eigem^alues. Further, the apparatus may include eigenvector computation mean«i operative to determine at least one eigenvector associated with one of the dominant eigenvalues of the set of estimates; wherein the channel estimation mearis uses the at least one eigenvector to project the set of estimates of the full dimei\sion channel parameters onto the subspace spanned by the at least one eigenvector.
BRIEF DESCRIPTION OF THE DRAWINGS
The features/ objects^ and advantages of the present invention will become more apparent from the detailed description set forth below when taken in conjunction with the drawings in which like reference characters identify correspondingly throughout and wherein:

FIG. 1 illustrates configurations of wireless communication systems including multiple transmitter antennas;
FIG. 2 illustrates a model of a wireless connmunication system according to one embodiment;
FIG. 3 illustrates a model of a channel between transmitter and receiver in a wireless communication system;
FIG. 4 illustrates the physical layout of antennas in a transmitter of a wireless communication system;
FIG. 5 illustrates a flow diagram of a method of reduced rank charuiel estimation for a wireless communication system according to one embC'diment;
FIG. 6 illustrates a plot of the estimation gain of one embodiment;
FIG. 7 illustrates a system configuration according to one embodiment, and
FIG. 8 illustrates an exemplary embodiment of a wireless communication system.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
Multiple radiating antennas may be used to improve transmission quality in a wireless communications system. In the design of third generation mobile radio systems, for example, various transmitter antenna diversity techniques are presented. Multiple transmitter antennas may be used to communicate information to a receiver using a single or multiple receiver antenaa(s). Multiple antenna systems offer an improvement in quality. However, the improvement is dependent on the accuracy of the channel model used in the receiver to demodulate the transmitted information. Modeling of the transmission channel uses parameter estimates and determines an effective channel impulse response for the channel When multiple antennas are used.

the modeling involves estimates of each transnnission channel for all transmitter-receiver antenna pairs.
The transnaission channel from transmitter to receiver is a space-time
chaniiel described generally by at least one impulse response. Often there is
>
little change in the channel parameters from one channel to another, such as where the channel impulse responses differ only in phase. In such a case, it may not be necessary to derive estimates of impulse responses independently for e^ich channel^ but rather some information may be reused. When channels are correlated/ a reduced rank representation of the channels may be used. Reduced rank refers to the reduced number of completely imcorrelated channels used to describe the Hnlc between transmitter and receiver. One way to ob;5erve this reduced rank is the rank reduction of the chaimel covariance matrix used to describe the mutual statistical dependencies of the different chauTvel impulse responses. Note that the reduced rank can also be realized by other parameter measures. For example, in one embodiment a sample matrix is formed of columns comprising samples ol channel impulse response estimates
over time/ wherein the reduced row rank of such a sample matrix is applied as
>
described herein. A reduction in rank may result in a less complex filter or demodulator, Le^, reduces the number of filters and/or filter elements and/or demodulation units used In the receiver. Furthermore, the reduction of the number of estimated parameters used to characterize the channel leads to improved accuiracy of the Hiiamtel model

antejinas. The wireless communication system 10 includes a transmitter 12 and receiver 16 that communicate via an air interface. A channel model 14 represents the channels for antenna pairs between transmitter 12 and receiver 16. Channel model 14 considers the channels v/ithin a link, such as the MISO link of FIG. 1.
Continuing with FIG. 2, let N^^ be the number of antennas used at the
transmitter 12 and N^^ the number at the receiver 16^ respectively. In general; for {.^ach significant propagation delay between transmitter and receiver, (Nj.^ • Nff^) transmission charmels exist for the pair, wherein for a significant
propagation delay the received signals resemble the known transmitted signals witli high certainty. In other words, define A^^ as the number of significant
propagation delays, also referred to as echoes. The (A^^.^- ^/u'^s) channel impulse response samples are then estimated to perform coherent demodulation. When the channels are uncorrelated, the (N^^ • Nf^^^ • N^) channel
impulse response samples are modeled as completely uncorrelated random proc€)sses and the estimates of these channel impulse response samples may be derived independently without loss of demodulation performance* However, if the (NT^- NJ^'N^) channel impulse response samples are not uncorrelated
random processes the (N^^ - NJU'^E) channel impulse response samples may be modeled as a linear combination of a smaller number N^^ of channel impulse response samples, wherein NcH estimated, and the linear trarxsformation of the NCH channel impulse response samples are resolved into the (Ny^- N^^-Nj^) channel impulse response samples, then modeling may be accomplished with the NCH channel impulse response sample estimations. This reduces the number of parameters to be estimated while increasing estimation quality, yielding an increase in the demodulation performance. Even if the exact representation of the linear

transformation of the A^^;;;, channel impulse response samples into the
corresponding (N^^ - ^^HC'^E) channel impulse response sanrples is not known,
the modeling may still be accomplished with W^^, channel impulse response
sample estimations if the subspace spanned by vectors of this linear trans formation is known or can be estimated.
This principle is referred to as "Reduced Rank Channel Estimation." The transrormation of the /\^C-A uncorrected channel impulse responses into the
i^Tx' ^RX'^E) correlated channel impulse responses can depend on factors including; but not limited to, the antenna configuration, anterma patterns, polar:ization characteristics, propagation conditions and more. In some cases the transformation might be known a priori, in other cases it can be derived or estimated, for example by angle-of-arrival estimation. The subspace spanned by th In one embodiment, the rank reducing ti-ansformation is known a priori or is stimated. In other words, the mapping of the N^^ channels onto the
i^Tx'^R.^) channels is ascertainable. The reduced rank channel is then estimated using the ascertained transformation. When desired, an equivalent full dimensional channel model may then be derived from the reduced rank estimate by transforming the reduced rank estimate back to the larger dimension.

In an alternate embodiment, the rank reducing transformation is not directly known, but the subspace spanned by the transformahon may be extracted from the dominant eigenvectors of the channel covariance matrix. Note that the subspace may be referred to as the signal subspace or the channel subspace. The process involves first estimating a channel covariance matrix and finding the dominant eigenvalues. By deternuning the associated eigenvectors which span the channel subspace^ the process projects the conventional channel estimate into the channel subspace, yielding a reduced rank channel model with reduced estimation errors* If desired, the reduced rank model may be transformed back into an equivalent full dimension channel model.
FIG. 3 illustrates a model 18 of a MIMO channel for continuous time having a linear MIMO filter 20 with N^^ inputs and A^^^ outputs. The linear
MIMO filter 20 is defined by the //^^xyv^ matrix H(t) comprising of linear
functions /z^.(r), / = I-^T, , j = I'-A'/u • Generally, ft^.(r), / = 1.../^^^,;" = I...A^^are
unknown linear functions. The linear MIMO filter 20 represents the (Z/;-^ - Nf^)
radio channels through which the N^^^ transmit signals pass to the A^;^^ receiver
antennas. These radio channels are characterized by their channel impulse responses hfj(t), i-L..Nj.^, y = 1.,.A^^. The input signal to the model, ic(t), is a
(//y.^ >:1) column vector representing the A^^.^ band-linuted transmit signals, and the output signal from the model, ^(r), is a (iV^j, x 1 )column vector, sampled at r = r, 27..., as illustrated by switch T, where the bandwidth of the transmitted signals is less or equal to l/T, The received signals contain additive perturbation signals represented by the N^^^xl column vector ^(r), introduced
due to noise or co-channel interference. The additive perturbation signals are added at summation nodes 22. The relation between the input signals JfCO / the

FIG. 4 illustrates the physical configuration of antennas at the transmitter of an exemplary embodiment modeled as in FIG. 2. A reduced rank method is applied to estimate the link represented by channel model 14, having a transmitter 12 configured with the four (4) antenr.as, each spaced at a distance "d." The specifics of the configuration and model are discussed hereinbelow. Note that the estimation procedure is performed at the receiver 16. A reference direcdon is given by the horizontal line. Angles of transmission are measured with respect to this reference. The angle "a" corresponds to an angle of a propagation path with respect to the reference within a 2-D plane as illustrated. A rartge of angles with respect to the reference is also illustrated. The following method is used at the receiver 12 in system 10 to estimate the link.
FIG. 5 illustrates a flow diagram of an exemplary method of chaimel estimation used to process signals in a receiver unit in accordance with one embodiment. Process flow begins by searching for significant propagation delays in the channel, le. searching for significant echoes at step 40,_ In one embodiment the process involves a sliding correlation of the received signals with known transmitted signals or known components of the transmitted signals. Correlation refers to the degree with which the received signals are related to the known transmitted signals, wherein a perfect correlation proves a relationship between the signals with high confidence. For time-shifted signals, wherein sliding delays are used to shift the received signals in time, a resultant sliding correlation provides the degree of certainty with which the time-shifted signals resemble the known transmitted, signals. Thus in the wireless system context, sliding correlation relates to the synchronization of known signals transmitted by the Tx antennas with time-shifted versions of the received signals. The exemplary embodiment of reduced rank channel estimation uses sliding correlation of the received signals with known transnutted signals to estimate the number N^ and the values r^f^.K^r^v^ of significant propagation delays, i,e, delays for which the received signals shifted back by these delays in time resemble the known transmitted signals with high certainty. The

procedure of sliding correlation in order to find significant propagation delays is alSD known as 'searching" in CDMA systems.
The method then estimates parameters for multiple observable channels betv/.een the //^.^ transmitter antennas and the Nj^ receiver antennas at step 42.
The cliannels are radio network connection pairs coupling at least a portion of the Nj^ transmitter antennas to at least a portion of the Nf^ receiver antennas.
In the exemplary embodiment, there is a connection between each transmitter 12 antenna and each receiver 16 antenna, resulting in (N^.^ • /V^) channels. The
parameters describing the multiple charvnels are those characteristics that impact the impulse responses of the channels. Assuming that N^ significant propagation delays (echoes) exist between transmitter and receiver, i^Tx' ^Rs'^e) complex samples of the {N-r^ - N^^^) channel impulse responses could be used as a set of parameters describing the multiple channels. This set of parameters is denoted by a {(Nj.^' Nj^^N^)xl) vector termed n herein. The
relation between ^{t), h, ^(/),and ^(r) is developed hereinbelow.

Since the output signals ^(/) are sampled at a sampling rate of l/T, vectors containing the discrete time samples can represent segments of a finite duration of the: continuous time signals. For the sake of simplicity, the received signals ^(r) are described herein by a disaete time representation over a firute
duration of time t^O,T,K ,{N^-1)T, where N^ is the number of samples taken over time. Therefore, the following abbreviations are used. Each discrete

The .second step in the flow diagram in FIG- 5 at step 42, is to repeatedly process estiniates for a set of parameters characterizing the multiple channels between transnutter and receiver. For the above-described mathematical representation of tl\e channel model, this may be equivalent to processing estimates
h^"\ n = lK iV^ of the vector li in (14) for N^ different points in time. A
conventional method uses the correlation of the received signals, shifted back in time by certain delays, with known transmitted signals, such as pilot signals specific to the transmitter antermas, or predeteriruned training sequences. As the significant propagation delays t^.tiX ,r^^ are already determined in step
40, tl\e exemplary embodiment of reduced rank channel estimation uses the correlation of known transmitted signals v^th versions of the received signals, shifted back in time by r^r^^K ^r^^^, to generate a channel model, such as
charu\el model 14 of FIG. 2, characterized by the vector h. If the noise vector K represents spatial and temporal white perturbation, wherein the noise
covaiiance matrix is given as R^ =/ftft") = o'^ •I^'^*'^'*, and if the matrix A
comprises of the a priori known signals, such as pilot symbols of a CDMA systei-n, channel estimates obtained by correlation can be described by

shall hold.
As illustrated in FIG. 5, a covariance matrix of the channel parameters is estimated at step 44. Covariance measures the variance of one random variable with respect to another. In this case, the covariance matrix describes the variance of the various channel parameters with respect to each other. According to the above-desaibed mathematical representation of the channel
model, step 44 corresponds to processing an estimate R,^ of the chaimel

If the MIMO channel has a reduced ranJk wherein N^h

As a consequence/ the rank of the channel covariance matrix i?^ is equal to A'^^. Given (20), and assuming that correlation according to (17) is used to derive the
channel inipulse response estimates n, the covari
yields only N^^^ non-zero eigenvalues, where A is a diagonal matrix containing the eigenvalues and £ is a square matrix containing the eigenvectors of

U, Rf, shares the eigenvectors with R^^^^B-R^ •5"/^^'^ Since A is a diagonal matrix with only Np, non-zero elements, (N^j, - Nf^ - A^^)-//^^ eigenvalues of
Rf^ are constant, and N^,^ eigenvalues of R^ are larger than the former ones. Tl\es€s larger eigenvalues are termed dominant eigenvalues in the sequel. With a diagonal matrix A^, contairung all dominant eigenvalues of the estimated channel covariance matrix, the matrix E^, containing the corresponding eigenvectors, and with the matrix S^^, containing the remairnng eigenvectors, (23) becomes

Therefore, the matrix E^ contains the eigenvectors sparming the channel or signal subspace.
The estimated covariance matrix ^^ is then ranked at step 46, meaning that tlie number of dominant eigenvalues is estimated. The rank is compared to

I maximum value "MAX" at step 48. MAX is equal to the total number of
^stiirated channel parameters in the vector h . In other words, MAX is equal to i^Tx ^fu'^s)' A^ many of the mechanisms impacting correlation, such as the directionality of the propagation paths, do not change quickly over time, the correlation characteristics may be estimated by averaging over rather long time inten^als in comparison to the inverse fading rate of the channel(s).
The rank of the covariance matrix determines whether the i^Tx' ^KX'^E) channel parameters describing the (A'.^^ - N^^) existing transiTiission channels can be modeled as a linear combination of a smaller numt^er A'^^ of equivalent uncorrelated channel parameters If a reduced rank
is available, the channel subspace E^ of the estimated covariance matrix Rf^ is derived at step 52, Note that instead of using the estimated covariance matrix R^, the rank of R^ and the channel subspace E^ can also be derived from the matrix of channel parameter estimates

effectively projecting the originally estimated channel parameters into the channel subspace. This projection into the channel subspace reduces the estimation error. If a reduced complexity demodulator is used in the receiver, which uses the reduced rank channel, i.e., takes only a reduced number of channel parameters into account for demodulation, the estimates of (26) may be directly used in the demodulator for coherent demodulation. In other words, processing would flow directly from step 54 to step 58, or at a minimum step 58 would used the reduced rank estimates.

If a conventional receiver, designed for the full rank channel model, is to be ur>ed, the estimates g^"^ may be transformed back into the full dimensional
wherein the factor /?"*^^" is used to make the estimate xmbiased. Note that the estimate of the channel subspace E^ may be updated continuously by using a
sliding time v^indow for the estimates R^ or X^, respectively. This eliminates
the delay of waiting for a new complete sample set, by using a portion of the previous sample set with incrementally time-shifted new values.
If rank reduction is not possible, processing continues to use the full rank of the system to model the channel at step 50. In this case the method estinriates the (yVr*' ^HX'^E) channel parameters independently from each othei'. Once the system is modeled^ signal demodulation continues at step 58,
The MISO path illustrated in FIG. 1 is provided as an exemplary embodiment. As illustrated, the transmitter, Tx, has four (4) radiating anteimas (iVy,^ =4) and the receiver, Rx, has one (1) anterma (iV^ =1). For simplicity,
several assumptior\s allow a straightforward analysis demonstrating the applicability of the exemplary embodiment to modeling a system as illustrated in FIG. 1. First, the example assumes that each Tx antenna transn\its a pilot signal specific to that antenna, wherein the antenna-specific pilot signal is time-aligned and orthogonal to the pilot signals of the other Tx antennas.
Second, assume the channels are frequency non-selective fa^Jing chaiuiels, each made up of a large number, P, of radio network paths* The paths each have approximately a same nm length and a same attenuation. The secord assumption ensures that the relative propagation delay is smaller than the inverse of the transmission bandwidth. The propagation delay of two radio paths is typically due to differences in run length.
Third, the channel model is restricted to 2-D propagation, U. all effective radio paths are located in a 2-D plane. See FIG. 4. Additionally, the geometry of the effective radio paths at the transmitter is assumed to be time-invariant.

wherein each path departure angle, measured with respect to a reference direction of Tx, are concentrated around an average angle, a . The radio path angles are Gaussian distributed having mean a and a standard deviation a. For one sinnulation, a is selected randonxly between -60 and +60 degrees. The standard deviation a is assumed to be square root of two degrees* Fourth, the arrival paths at Rx are assumed uniformly distributed between 0 and 360 degrees to consider local scattering. Fifth, no line of sight exists.
Sixth, assume a specific phase and Doppler shift for each path. The path-specific phase is selected randonily according to a uniform distribution between 0 and In - Additionally, the path-specific phase is adjusted for each Tx anterma according to the geometrical antenna configuration, i.e., the antenna location with respect to a reference point. For phase adjustment, assume object scattering is considered in the far field- The channel-specific Doppler shifts are generated according to a uniform distribution of the angles of arrival paths at Rx, a carrier frequency and a predetermined R: Given the exeinplary system as detailed, application of the process of FIG. 5 provides a channel model having a time variance according to the classic Doppler spectrum. It is possible to consider an antenna-specific radiation pattern. Witih this charmel model, the charmel impulse responses for the charmels seen through the different transmitter antermas can be generated using the same set of radio paths, thus, introducing realistic correlation in the fading of the different channels.
On the receiving side of the air-interface, at the single anteima of Rx, the method derives an impulse response estimate for each of the four transnaission chaimels, le„ the four radio network coimectior\s between Tx antennas and the JU anleima. The estimate is based on the a-priori knowledge of spreading codes

used to generate the anterma-specific pilot signals associated with each Tx anteinna.
Referring again to FIG. 4, in the geographical configuration of antennas at T>:., the antennas are positioned in a line having constant spacing d between neighboring antennas, wherein d =A, i.e., antennas are spaced one wavelength apart. Note that Rx has a single omnidirectional antenna- A total number of effective radio paths is considered with P = SQ, Channel specific variables, cc^,
f^, and C^^ / represent, respectively, the angle measured from the reference line,
the Doppler shift and the phase. The equation describing the channel impulse response for a channel between Tx antenna n and the Rx antenna is given as

where h(j) is the equivalent channel imptilse response for a equivalent isotropic Tx antenna at the reference point.
In this case the channel impulse responses for the different Tx antennas orUy differ by a complex factor, i.e., the channels ai'e completely correlated. The steering vector is then defined as

The four (4) channel impulse responses seen from the Tx antennas are then copies of the channel impulse response k(t)r weighted by four different
complex factors, which means, the vector n{t) is a linear transformation of the scalar h(t) given by
i,e,, the vector ^ in the linear transformation of (19) is in this example equal to the scalar h{t) and the matrix B is equal to the vector ^{a). This means the
channel covariance matrix R,, =/ft'A"\ is equal to R^ ==5(ff)5(a)"/|/i(f)p\ in
this example. If the steering vector a{s) is knov^m, such as a-priori knowledge of the antenna configuration and the radio path direction a, it is sufficient to estimeite the scalar h{t) and either calculate an estimate for h{t) using the linear transformation with a{a) or use the estimate of Ht) and ^{a) directly for demodulation.
Note that for the case wheni'(a) is known it may be sufficient to
estimate h{t) and then compute an estimate of h{t) from the scalar estimate of h{t). If the demodulator is designed such tihat the channel consists of a single

scalar, i.e^, the demodulation considers a{a), then it is possible to denriodulate using ^{a) and the scalar channel
The antenna-specific pilot signals at the transnutter are termed x^{t), n = IK A^T-^, and the relationship is defined by

The pilot signals are made up of segments, each having a duration T^,
refen'ed to as the pilot symbol duration, over which the pilot signals are orthogonal, and wherein the following holds

The pilot vector is defined as

and the receiver noise signal z{t) represents white Gaussian noise. The signal recei\'ed by the single Rx antenna is described as

Conventionally/ correlating the received signal with the four (4) pilot sequences derives a set of four (4) channel estimates. Wherein the pilot signals are orthogonal over a pilot symbol period, this estimation is then repeated at the pilot symbol rate. Such a correlation procedure is generally referred to as "integrate and dump" and may be expressed as

wherein h^^l is a vector made up of conventional, i.e., integrate and dump, charuiel estimates derived from the n-th pilot symbol. If (34) is transformed into a disaete time representation, by putting Ny^ =TJT samples of the pilot signals x^t) into the columns of the matrix A, N\ samples of the noise signal z{t) :;nto the vector K, and N^ samples of the received signal y[t) into the

Fron\ this scalar estimate, using the linear transformation, a new estimate of the channel impulse vector is generated as

Igno::ing charmel variations within one pilot symbol (44) becomes

If ^(a) is not known a-priori, it may be estimated using the covariance matr::x given by

with PA being the average power of the scalar channel impulse response h(t). The covariance matrix R^ can be approximated as

whicJi averages the vector with the conventional channel impulse response estimates over a number^ N^, of pilot symbols.
For the case without noise and having an angular spread equal to zero,
k^ is rairk one (1) and the vector i\a) spax\s K^, Thus (47) reduces to

Note that the normalized vector ^(^)/||i?(a| spans /?„.
For a noisy case with sufficient low noise power and sufficient low angular spread, R^^ is still dominated by one eigenvalue. Therefore, the process
perfo::ms an eigenvalue decomposition of k^^ When one eigenvalue is much

larger than all other eigenvalues, it is an indication that the angular spread around ^{^) was rather small. Therefore, as ^^^ is the eigenvector
corresponding to the largest eigenvalue of R^, the approximation becomes

Note that the vector v^^ in this example is equal to the channel subspace E^ -
In general, the estimate R^ is used to determine whether the rank of the channel
estimation covariance matrix can be reduced. If Rf,is full rank, the channel estimation problem is not reduced to a smaller dimension.
According to the exemplary embodiment, orthogonal pilot signals of binary chips have a chip rate of 1.2288 Mcps, and a pilot symbol duration of 64 chips. With this channel model, a received signal, including white Gaussian noise; is generated for 4000 consecutive pilot symbols having a pilot Signal-to-Noise Ratio (SNR). From the received signal, 4000 conventional vector
estimates, h^^l, are generated- The thus generated covariance matrix R,^ is avera;5ed over these 4000 consecutive conventional channel estimates. In the exemj?lary embodiment, the process takes approximately 208.3 ms. After extracting the eigenvector corresponding to the maximal eigenvalue of Rf^, the
matrix R^ is calculated. Subsequently/4000 ne\v vector estimates hj^^l are produced according to

Using the exemplary embodiment, iterations are repeated A/„p = 50 times. Over the 50 iterations the transmitter angles are varied such that a is uniformly distributed within (+/- 60) degrees, while the angular spread remains constant, having a standard deviation of square root of two i^f2) degrees. Additionally, the channel parameters for a given pilot SNR are varied.

The varied parameters represent radio path direction(s), path-specific phase, and path-specific Doppler shift, for a certain pilot SNR. An equal number of itera!ions is performed for different pilot SNR values. A comparison of the quality of the set of conventional estimates to the set of new vector estimates, with respect to the reduction factor of the mean squared estimation error that is averaged over time and iterations, is made using the estimation gain given as

For the exemplary embodiment, FIG. 6 illustrates the estimation gain in dB as a function of the pilot SNR. Wherein the pilot SNR is defined as the ratio of th(» average energy per pilot chip E^ of one pilot signal received at the single
anterna receiver to the received noise power dex\sity I^ in dB.
The upper limit for the estimation gain is determined by the number of trans;i\it antennas, which is illustrated in FIG. 6 as 6dB. As illustrated in FIG. 6, the estimation gain approaches the upper limit even though the assumed angular spread is not zero and the received signal is severely corrupted by noise The reduction of the estimation gain with increasing pilot SNR is due to the non-zero angular spread.
Although the chaimel impulse responses are not completely correlated,
the derivation of the impulse response n^ assumes this property. For larger
angular spreads, a smaller estimation gain is expected. For small angular spreads, the estimation gain appears considerable. Note that in general, for residential and suburban environments a standard deviation of one (1) to two (2) degrees is frequently observed. Note also that it is possible to evaluate the perfoimance improvement of the reduced rank channel estimation method using a Monte-Carlo-simulation to derive the reduction of estimation errors of the channel impulse responses as compared to conventional channel estimation using independent correlators.

Reduced rank channel estimation for systeiiis using multiple transmitter antertnas allows improvement of the channel estimation quality under certain propagation conditions with limited diversity due to correlated fading. As the mechanisms affecting correlation^ such as the directionality of the radio wave propagation, change relatively slowly over time, the correlation characteristics may l^e estimated by averaging over extended time intervals. This is in contrast to thei time intervals associated with inverse fading rate of the channel and thus allows improved accuracy in estimating the correlation characteristics.
Reduced rank channel estimation for multiple transmitter antennas is also applicable to frequency-selective channels by computing either separate estimates of the correlation characteristics or by computing estimates of the correlation characteristics across all propagation delays. Separate estimates
A
refers to computation of i?^, for each propagation delay. Reduced rank
chanriel estimation is then performed taking into accoimt each delay occurring in the frequency-selective channel impulse response. In an alternate embodiment, wherein additional information, such as the antenna config;uration at the transmitter^ is known a-priori, the step of estimating the linear transformation of the reduced number of uncorrected channels into the larger number of correlated channels may be more accurate. Additionally^ the reduced rank estimation process may be extended to cases with more than one receiver antenna. In ttiis case/ the estimation is performed for the MIMO channels/as illustrated in FIG, 1, While the present example involves a system emplcying coherent demodulation, reduced rank channel estimation., as described herein is also applicable to communication systems employing noncoherent demodulation.
A receiver 100 according to one embodiment of the present invention is illustrated in FIG. 7. The receiver 100 has a single anteiuia 102 that receives signalr> from a transmitter having multiple antennas. The received signals are first processed by the preprocessor 104. The signals are then provided to a correlator 106, which is used as a sliding correlator for searching and as a correlator for the significant delays for channel estimation. In an alternate

embodiment the delays are determined in software without use of a correlator. The outputs of the correlator 106 are used to provide an estimate of the CO variance matrix. In one embodiment, the correlator 106 is made up of fingers to form a rake, having one finger for each combination of transmitter antenna, receiver antenna and significant delay. The estimates are provided to the central processor 112 via bus 116. The processor 112 stores the: channel paraiTieter estimates in memory 114 so that the estimates may be used to derive the charmel covariance matrix averaged over time.
From memory 114, the estimated covariance matrix is provided to the rank analysis and subspace estimation unit 108 for eigenvalue decomposition. If orie or more eigenvalues dominate the others, the channel subspace is estirriated by computing the eigenvectors that correspond to the dominant eigenvalues. The eigenvectors spanning the channel subspace are written to memory for further use in the charmel subspace projection unit 109 where redu<:ed rank charmel parameter estimates are produced by computing the projection of a original per estimation time> internal onto the charmel subspace, yielding A^^^ reduced rank charmel
parameter estimates per estimation time interval. The results of the channel subspace projection unit 109 are written to memory for use in the demodulator 110. Optionally the channel subspace projection unit 109 could generate equivalent full dimension channel parameter estimates, by re-transforming the A^^^ reduced rank charmel parameter estimates into (A^^^ ^ N^^-N^) equivalent
full dimension charmel parameter estimates per estimation time interval. For example in a conventional RAKE-receiver design for the full rank channel model, the number of rake fingers for a full rank demodulator would be {Nj^' ^Rx • ^£)• A full rank demodulator would then use the (A^^^ • Nj^^^N^) original charmel parameter estimates for the finger coefficients. A reduced complexity demodulator could eventually use orUy N^^ RAKE fingers using the
A^^;, reduced rank charmel estimates as coefficients. However^ since the receiver would generally be designed in anticipation of a worst case situation,

I.e., wherein (A^;.^ • A^^t '^E) f^rigers are implemented, it would be sufficient to compute (/Vrx' ^/u '^E) correlated channel parameter estimates with improved estimation quality over of the iV^^ reduced rank channel parameter
estimates.
The rank analysis and subspace estimation unit 108 and the subspace projection unit 109 may be implemented in a Digital Signal Processor (DSP), dedicated hardware, software, firmware, or a combination thereof. Modules within receiver 100 may be incorporated together, and are illustrated as separate blocks for clarity based on function.
An exemplary configuration of one embodiment is illustrated in FIG. 8 for a system having four (4) transmitter antennas and two (2) receiver antennas. Three (3) transmission paths are illustrated and labeled 1,2 and 3. The points of reflection for paths 1 and 2 are both on a same ellipse, wherein the ellipse is fotmtid such that Tx and Rx are the focal points. Note that the ellipse is superimposed on the illustration of the physical layout of the system. Path 3 falls outside of the illustrated ellipse. Paths 1 and 2 have the same significant delay, TI, with respect to the receiver, while path 3 has a significant delay, X2 different from TJ. The path delay is a function of the corifiguration of the antennas as well as the environment of the system. As illustrated, the four (4) transmitter antennas and the two (2) receiver antennas result in eight (8) chanr..els. Each of the path delays, TI and X2, produce an echo, wherein (iV^ = 2). The dimension of the covariance matrix is given as (JV,.^' Nj^-N^) or sixteen (16) corresponding to the {Nj^ • ^A^-N^) channel impulse response samples.
Therefore, the full rank channel parameter vector is a 16-dimension vector. Using the rank reduction methods described herein, the rank of the channel estimation may be reduced to three (3) dimensions/ corresponding to paths 1,2, and 3, wherein {N^^=^2). Note that where the mapping of the Ncs transmission paths to the (A^r^ • N,^*N^) channel impulse response samples is
not known, the subspace may be extracted from configuration information. If the location and characteristics, such as direction and directionality, of the

antennas are known, the information may be used to generate an array response or steering vector. Using the steering vector and path direction information, which is also extractable using subspace algorithms, the angle of tranf-mission, a, is estimated. If the antenna configuration has a fixed deployment the angle of transmission is calculable. A vector is formed including an angle of transmission for each transmitter antenna. Similarly, an arrival angle vector is formed considering the receiver antermas. A linear transformation for the mapping of the NCH transnrdssion paths to the
(^7>* ^RX'^E) channel impulse response samples is constructed using this information from both the transmitter and receiver configurations- This provides the matrix B as given in (19) hereinabove describing the linear transformation. The covariance matrix is derived therefrom as in (20) hereinabove. The process then proceeds as for the case where corresponding information is obtained from a priori knowledge.
VVhile one embodiment has been described herein with respect to the time domain, an alternate embodiment performs a rank reduction of the covaiiance matrix or a sample matrix in the frequency domain. If the parameters and equations are developed in the frequency domain, the process to estimate the chaimel then incorporates the frequency domain values.
The previous description of the preferred embodiments is provided to enable any person skilled in the art to make or use the present invention. Tlie various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without the use of the inventive faculty. 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 principles and novel featuies disclosed herein.

CLAIMS
I (WE) CLAIM:
1. A method for modeling a link in a wireless communication system, the system having a transmitter having N antennas and a receiver having M antennas, each path from one of the N transmitter antennas to the M receiver antennas comprising a channel/ the method comprising:
determining a matrix describing parametric relations of the link;
ranking the matrix;
determining if the rank is less than NxM;
if the rank is less then NxM performing an extraction of a subspace of the matrix;
deriving channel impulse responses for each channel based on the extracted subspace of the matrix; and
demodulating a received signal using the channel impulse responses.
2* The method of claim 1^ wherein the matrix is a covariance matrix descri bing the link, wherein the covariance matrix represents a plurality of impulse responses between the transmitter and the receiver.
3. The method of claim 1, wherein the matrix is a sample matrix describing the link.
4. The method of claim 1, wherein the step of determining the matrix further comprises:
estimating a plurality of parameters describing at least one charmel.
5. The method of claim 4, wherein the parameters include a distance
between transmitter antennas.

6. The method of claim 4, wherein the parameters include a transmittal angle with respect to a configuration of the transmitter antennas.
7. The method of claim 4, wherein the determining the matrix comprises
estimating the matrix,
8. The method of claim 1, wherein the matrix describes parametric
relations of the link in the frequency domain.
9. The method of claim 1, wherein the ranking the matrix, further
comprises:
deternmining an eigenvalue for the matrix,
10. The method of claim 1, wherein if the rank is equal to (NxM) a set of correlated impulse responses is applied for demodulating.
11. A wireless apparatus operative to perform the method of claim 1.
12- A wireless communication apparatus, comprising:
a correlator operative to estimate a covariance matrix representing a link
with a transmitter based on signals received from the transmitter; a rank analysis unit coupled to the correlator and operative to estimate a
rank of the covariance matrix; and a channel estimation unit coupled to the rank analysis unit and operative
to generate a reduced rank channel estimate.
13. The apparatus of claim 12, wherein the covariance matrix represents a
plurality of impulse responses between the apparatxis and the transmitter.
14. The apparatus of claim 12, wherein correlator is operative to determine a
correlation of at least two channels.

15. The apparatus of claim 14, wherein the ranJ: analysis unit is operative to determine an eigenvalue corresponding to the covariance matrix.
16. The apparatus of claim 15, wherein the ranfc. analysis unit is operative to compare the estimated rank of the covariance matrix to a predetermined full value,
17. A method for estimating a link in a wireless communication system, the method comprising;
estimating a covariance matrix for the link; deterrxuning if the rank of the covariance matrix is reducible; reducing the rank of the covariance niatrix; and
estimating a set of impulse responses for the link using the reduced rai\k covariance matrix.
18. The method of claim 17, further comprising:
deternuning a correlation of the channels-
ranking the covariance matrix; and
perfornrung an extraction of a reduced rank matrix out of the covariance matrix.
19. A wireless communication apparatus operative within a wireless
communication system having a transmitter having N antennas and a receiver
having M antennas, each path from one of the N transmitter antennas to the M
receiver antennas comprising a channel, the apparal-us comprising:
a first set of computer readable instructions operative to determine a
covariance matrix describing the link; a second set of computer readable irtstructions operative to rank the
covariance matrix; a third set of computer readable instructions operative to determine if
the rank is less than NxM;

a fourth set of computer readable instructions operative to perform an
extraction of a reduced rank matrix out of the covariance matrix
if the rank is less then NxM; a fifth set of computer readable instructions operative to derive channel
impulse responses for each channel based on the reduced rank
covariance matrix; a sixth set of computer readable instrucfcons operative to demodulate a
received signal using the channel impulse responses.
20. The apparatus of claim 19, further comprising:
an equalizer operative in response to the sixth set of computer readable instructions, wherein a configuration of the equalizer is deternruned by the rank of the covariance matrix.
21. The apparatus of claim 19/ further comprising:
a seventh set of computer readable instructions operative to derive a correlated channel impulse response.
22. A wireless communication apparatus, comprising:
a charmel estimation means operative to estimate a covariance matrix
representing a link with a transmitter based on signals received
from the transmitter; a rank analysis unit coupled to the correlator and operative to estimate
the rank of the covariance matrix; and a channel estimation means coupled to the rank analysis unit and
operative to generate a reduced rank channel estimate.
23. A wireless communication apparatus, comprising:
a correlator operative to estimate a covariance matrix representing a link with a transmitter based on signals received from the transmitter;

a rank analysis unit coupled ^o the correlator and operative to estimate
the rank of the covariance matrix; anci a channel estimation means coupled to tl\e rank analysis unit and
operative to generate a reduced rank channel estimate.
24. A method for estimating a link in a wireless communication system, the
method comprising:
estimating a covariance matrix for the link; determining if the rank of the covariance matrix is reducible; reducing the rank of the covariance matrix; and
estimating a set of impulse responses for the link using the reduced rank covariance matrix.
25. The method of claim 24, further comprising:
determining a correlation of the channel;
ranking the covariance matrix; and
performing an extraction of a reduced rank matrix out of the covariance matrix;
26. A wireless apparatus, comprising:
charmel estimation means operative to determine sigruficant delays and determine a set of estimates of full diinension channel parameters associated with the significant delays, wherein each one of the set of estimates corresponds to an instance in time;
eigenvalue computation means operative to deterrxune eigenvalues of the set of estimates of the full dimension channel parameters and find any dominant eigenvalues; and
channel estimation means operative to determine a set of reduced rank channel parameter estimates in response to the dominant eigenvalues.

21. The wireless apparatus of claim 26, further comprising:
eigenvector computation means operative to determine at least one eigenvector associated with one of the dominant eigenvalues of the set of estimates; wherein the channel estimation means uses the at least one eigenvector to project the set of estimates of the full dimension channel parameters onto tlie subspace spanned by the at least one eigenvector.

A method for modelins a link in a wireless communication svstem substantially as herein described with reference to the accompanying drawings.
A wireless communication apparatus substantially as herein described with reference to the accompanying drawings.

Documents:

597-chenp-2003-abstract.pdf

597-chenp-2003-claims duplicate.pdf

597-chenp-2003-claims original.pdf

597-chenp-2003-correspondnece-others.pdf

597-chenp-2003-correspondnece-po.pdf

597-chenp-2003-description(complete) duplicate.pdf

597-chenp-2003-description(complete) original.pdf

597-chenp-2003-drawings.pdf

597-chenp-2003-form 1.pdf

597-chenp-2003-form 13.pdf

597-chenp-2003-form 26.pdf

597-chenp-2003-form 3.pdf

597-chenp-2003-form 5.pdf

597-chenp-2003-pct.pdf

« Previous Patent

Next Patent »

Patent Number

209642

Indian Patent Application Number

597/CHENP/2003

PG Journal Number

50/2007

Publication Date

14-Dec-2007

Grant Date

05-Sep-2007

Date of Filing

22-Apr-2003

Name of Patentee

M/S. QUALCOMM INCORPORATED

Applicant Address

5775 Morhouse Drive, San Diego, California 92121-1714

Inventors:

#	Inventor's Name	Inventor's Address
1	JOSEF J. BLANZ	Weinstrasse 1, Hofgut 10, 67157 Wachenheim,

PCT International Classification Number

H04L 25/02

PCT International Application Number

PCT/US2001/050068

PCT International Filing date

2001-10-19

PCT Conventions:

#	PCT Application Number	Date of Convention	Priority Country
1	09/694,432	2000-10-23	U.S.A.