BACKGROUND In a digital cellular radio communication system, radio signals which are digitally modulated are used to convey information between radio base stations and mobile stations. The radio base stations transmit downlink signals to the mobile stations and receive uplink signals transmitted by the mobile stations. A common problem that occurs in digital cellular radio communication systems is the loss of information in the uplink and downlink signals as a result of multipath fading and interference which may exist in the radio transmission channel.

With regard to multipath fading, there are basically two effects: fading and time dispersion. When the path length between a mobile station and a base station is relatively short, fading arises from the interaction of the transmitted signal, or main ray, and reflections thereof, or echoes, which arrive at the receiver at approximately the same time. When this occurs, the main ray and echoes add either destructively or constructively. If there are a large number of echoes, the pattern of destructive and constructive addition takes on a Rayleigh distribution, which is why this effect is sometimes called"Rayleigh fading". Where destructive addition results in fading"dips", the received signal generally exhibits a relatively low carrier-to-noise (C/N) characteristic.

The effects of fading dips can be mitigated by having multiple receive antennas and by employing some form of diversity combining, such as selective combining, equal gain combining, or maximal ratio combining, wherein signals from each receive antenna are combined to create a single received signal.

Diversity techniques take advantage of the fact that the fading on the different antennas is not the same, so that when one antenna receives a fading dip, chances are, another antenna is not. Note Mobile Communications Design Fundamentals by William C. Y. Lee, Howard W. Sams & Co., Indiana, USA. In section 3.5.1 of this book, several examples are given describing how signals from two receiver amplifiers with separate antennas can be combined to counteract fading.

For longer path lengths, time dispersion occurs when the echoes are delayed with respect to the main ray. If an echo of sufficient magnitude arrives at the receiver delayed from the main ray by an amount of time on the order of the symbol period, time dispersion gives rise to intersymbol interference (ISI). Time dispersion may be advantageously corrected by using an equalizer. In the case of digital signal modulation, a maximum likelihood sequence estimation (MLSE) equalizer such as described in Digital Communications, 2nd Ed., by John G. Proakis, Mc-Graw Hill Book Company, New York, New York, USA, 1989 may be used. In section 6.7 of this book, various methods are described for detecting signals corrupted by time dispersion, or inter-symbol interference (ISI), using MLSE equalization.

There may also exist signal sources in the radio environment which are not orthogonal to the desired signal. Non-orthogonal signals, or interference, often come from radios operating on the same frequency (i. e., co-channel interference) or from radios operating on neighboring frequency bands (i. e., adjacent-channel interference). Of course, non-orthogonal signals can have a detrimental affect on the carrier-to-interference ration (C/I) of another channel. When the carrier-to- interference ratio (C/I) of a channel is too low, the quality of voice output at the mobile station is poor. Many techniques have been developed in order to minimize interference to tolerable levels including frequency re-use patterns, frequency hopping and adaptive beamforming, wherein the latter is generally used to steer a null in an antenna pattern in the direction of an interferer.

More recently, methods have been proposed that partially solve the problems of multipath fading and interference. In U. S. Patent 5,191,598 to Bäckström, et al., for example, the problem of accurately detecting signals in the presence of fading

and time dispersion is overcome by using a Viterbi-algorithm having a transmission function estimated for each antenna. By reference thereto, U. S. Patent 5,191,598 is incorporated herein in its entirety. Another method of accurately detecting signals in the presence of fading and interference was presented in the IEEE Transactions on Vehicular Technology, Vol. 42, No. 4, Nov. 1993, J. H. Winters:"Signal Acquisition and Tracking with Adaptive Arrays in the Digital Mobile Radio System IS-54 with Flat Fading".

Although the above described conventional techniques can be used to improve signal quality, there remains room for improvement. Thus, in one of the related applications, interference rejection combining (IRC) techniques are described which combat interference, for example, using impairment correlations to improve the maximum likelihood sequence estimation. However, the use of impairment correlations in this way may be mathematically complex and may require significant processor resources. Accordingly, it would be desirable to reduce the calculation complexity while preserving the useful qualities of interference rejection combining.

SUMMARY In accordance with one aspect of the present invention, beamforming and signal combining techniques can be used in conjunction with IRC techniques to reduce the overall complexity of processing received signals. According to one exemplary embodiment of the present invention, signals transmitted from mobile stations are received by an array of antenna elements at a base station. The received signals are combined in a beamforming device, for example a Butler matrix. The outputs of the beamforming device correspond to a plurality of beams which cover an entire cell supported by the base station. The number of beams formed can be smaller than the number of antenna elements, which reduces the complexity of processing within the receiver. Each output of the beamforming device is connected to a radio receiver which converts the RF signal to baseband. Other quality improvement processing functions, including locating the training sequence, estimating the channel impulse response and other quality measures, for example

measuring signal strength, noise power and carrier to interference ratio, are then performed. Thereafter, a certain number of received signal branches are selected based upon the estimated quality parameters. This further reduces complexity at the receiver, which performs interference rejection combining on the selected branches.

According to another exemplary embodiment, two or more antenna arrays can be employed to obtain further diversity against fading radio signals. Diversity techniques including, for example, spatial diversity or polarization diversity, can be employed. Like the first described exemplary embodiment, beamforming is also employed to further reduce complexity.

BRIEF DESCRIPTION OF THE DRAWINGS Exemplary embodiments of the present invention will now be described in more detail with reference to the accompanying drawings, in which like descriptive labels are used to refer to similar elements: FIG. 1 illustrates an exemplary cellular radio communication system; FIG. 2 illustrates a base station according to a first exemplary embodiment of the present invention; FIG. 3 depicts how fixed beams can be arranged to cover a cell; and FIG. 4 illustrates a base station according to another exemplary embodiment of the present invention.

DETAILED DESCRIPTION In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular circuits, circuit components, techniques, etc. in order to provide a thorough understanding of the present invention. However, it will be apparent to one skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods, devices, and circuits are omitted so as not to obscure the description of the present invention with unnecessary detail.

An exemplary cellular radio communication system 100 is generally illustrated in FIG. 1. A geographic region served by the system 100 may be subdivided into a number, n, of smaller regions of radio coverage known as cells 1 l0a-n, each cell 110a-n having associated with it a respective radio base station 170a-n. Each radio base station 170a-n has associated with it an antenna system 130a-n where inter alia the transmit and receive antennas are located. The use of hexagonally-shaped cells 110a-n is a graphically convenient way of illustrating areas of radio coverage associated with base stations 170a-n respectively. In actuality, cells 110a-n may be irregularly shaped, overlapping, and not necessarily contiguous. Sectorization within cells 110a-n is also possible and contemplated by the present invention.

Distributed within cells 110a-n are a plurality of mobile stations 120a-m.

Base stations 170a-n provide two-way radio communication with mobile stations 120a-m located within corresponding cells 110a-n respectively. Generally, the number, m, of mobile stations is vastly greater than the number, n, of radio base stations. Radio base stations 170a-n are coupled to the mobile telephone switching office (MTSO) 150 which provides inter alia a connection to the public switched telephone network (PSTN) 160 which, in turn, provides connections to communication devices 180a-c. This basic cellular radiocommunication concept is known in the art and will not be described further.

A base station 130 with an array antenna system, according to an exemplary embodiment of the present invention, is illustrated in FIG. 2. Therein, signals transmitted from mobile stations are received by the array of antenna elements 200.

The array 200 includes m number of elements. In the example illustrated in FIG. 2, m is five. The received signals are then forwarded to and combined in a fixed beamforming device 202, e. g., a Butler matrix. The outputs of the beamformer 202 correspond to r narrow beams that cover the entire cell supported by base station 130. For example, FIG. 3 illustrates three beams 204,206 and 208 which provide coverage of cell 210. The principle behind beamforming, as it is used in the context of signal reception, is that transmissions from a particular mobile station will be

received in less than all of a base station's beams. For example, a mobile located at point M in FIG. 3 may contribute relevant information received on beams 206 and 208, but not on beam 204. Thus, the processing of the received signals can be simplified vis-a-vis mobile M by disregarding beam 204. The number of formed beams, r, can be smaller than the number of antenna elements, m, which is another way to reduce the complexity of the receiver and to adjust the coverage area associated with the cell 210. Note that although the fixed beamforming is illustrated in FIG. 2 as being performed at RF, it can instead be performed at baseband.

Each beam's output from the beamformer 202 is connected to a radio receiver 220 which converts the RF signal to baseband. The estimation blocks 230 locate the training sequence, estimate the channel impulse response and perform other quality measurements, e. g., measurements of signal strength, noise power and C/I. Thereafter, the selector 240 selects the p best branches with respect to some of the estimated quality measures. The choice of p is a tradeoff between complexity and performance. In this example, p is chosen to be two. The selection of the p best branches can be memoryless or can take into account previous measurements as well.

The symbol detection is then performed with a multidimensional detector, e. g., a multidimensional Maximum Likelihood Sequence Estimator, that suppresses the interferers. The suppression process starts with estimation of the noise cross- correlation matrix Q at block 250. The noise includes, for example, thermal noise and co-channel interference. If the cellular system conforms to the GSM standard and the base stations are synchronized, then it is sufficient to estimate the noise cross-correlation matrix during the training sequence. Otherwise, if the base stations are not synchronized or the IS-136 (or PDC) standard is used, the noise cross-correlation matrix Q should be periodically estimated during the time slot.

In GSM, the radio channel will stay nearly constant during one time slot for all normal mobile station velocities. This implies that channel estimates H will be constant during a time slot and that the noise cross-correlation matrix Q will not

vary during a time slot. Therefore, H and Q need not be updated during the time slot which leads to a drastic reduction in complexity.

This should be compared with DAMPS in which the radio channels vary considerably during a time slot even for moderate velocities of the mobile station.

The gain from the reduced number of signals will also be greater for GSM systems due to the slower channel variation. In GSM, one can gain from not selecting a branch in which the desired signal is faded out during a whole time slot.

This is a rare occurrence in DAMPS systems.

As described above, the selection of p of the r branches consists generally of selecting a number of branches from a first criterion and then selecting a smaller number of branches from the first selection of branches using a second criterion.

Interference rejection detection is then performed at block 260 to whiten the received signals using the inverse of the noise cross-correlation matrix Q. This corresponds conceptually to pointing spectral nulls towards interfering mobile stations as described in more detail below. Instead of updating the noise cross- correlation matrix, it is also possible to directly estimate and track the inverse of the noise cross-correlation matrix.

In another exemplary embodiment of the present invention, shown in FIG. 4, two antenna arrays 400 and 402 are used to achieve diversity for mitigating the effects of radio signal fading. In this figure, similar beamforming, radio receiver, estimation, selection, noise cross-correlation and IRC blocks are again shown.

Since the function of these blocks has already been described above with respect to FIG. 2, reference is made to that description. The diversity antenna arrangement used could, e. g., be spatial diversity or polarization diversity which techniques are well known to those skilled in the art. Thus in the example of FIG. 4, m=10, r=6 and p=2. Furthermore, if the beamforming devices associated with the first and the second antenna arrays 400 and 402 are designed to form beams such that one beam from one of the beamforming devices always covers an area next to areas covered by beams from the other beamforming device, i. e., the beams from the two antenna

arrays are interleaved, then simulations have shown that it is best to choose the same number of beams from the two antenna arrays.

Having provided a general overview of exemplary base station architectures according to the present invention, a more detailed description of the signal processing according to the present invention will now be provided, wherein the signal transmitted by a mobile station is denoted as y (t), and the radio channel from the mobile station's antenna to an antenna element of the base station antenna array is modeled as a k tap FIR-filter: hi [hll h 12... hlk] Then, let the antenna array have m elements and model the radio channel to the other antenna elements in a similar way as the first, and collect them all in the channel matrix: <BR> <BR> <BR> <BR> fTrrlr<BR> H=&,...&,J.

The colored noise that the antenna elements sense can be denoted as: <BR> <BR> <BR> <BR> <BR> <BR> z (t) = [ZI (t) z(t)...<t)'K<BR> while a vector with time delayed copies of the transmitted signal can be denoted as: y (t) = [y (t) y(t-1) ... y (t-k+1)]T.

The received signals for all m antenna elements can now be written as follows: <BR> <BR> <BR> <BR> <BR> <BR> x(t) x2(t)...xm(t)]T=.Hy(t)+z(t)[x1(t) <BR> <BR> The beamformermatrix of block 202 rconstructs narrow beams, e.g., beams 204,206 and 208 illustrated in FIG. 3, which can be described with an r x m matrix B. The output signals generated by the beamformer 202 are: r(t) = [r1(t) rr(t)]T=Bx(t)=BHy(t)+Bz(t)...

The signals r (t) are the input signals to the radio receiver blocks 220, which signals are converted from RF to baseband. The baseband r (t) signals are then used to estimate the training sequence position and the channel estimate (i. e., the product BH) for each of the r branches at block 230. Some other quantities, like the signal strength, noise power and C/I, are also estimated for each branch using known techniques.

The training sequence position can, e. g., be estimated by performing a sliding correlation between the received signal and the training sequence. The channel = BH can, e. g., be estimated with a least square estimate, #= RyrRyy-1 where Ryy is the training sequence correlation matrix and Ryr is the cross- correlation between the received signal and the training sequence.

To further reduce the complexity of the receiver, p of r branches are selected at block 240. The selection is made based on one or more of the estimated values in the previous estimation blocks 230, e. g., the estimated C/I. The selection could also be based in part on previously stored estimation values.

In a first exemplary embodiment, the C/I is used to select the p branches by ordering the r branches according to their estimates of the C/I and to choose the p branches with the highest C/I.

In another exemplary embodiment, the p + a branches with the highest signal strength (from the estimate of C) are selected first and then from the remaining p + a branches, the p branches with the highest C/I are selected. Here, a is a natural number (i. e., 1,2,..., n) such that p + a < r.

Letting the p x k matrix G consist of the p selected rows of the BH matrix, and letting thep x I vector n (t) consist of the p selected elements of the Bz (t) vector, the p selected signals can then be described as: p (t) = [PI (t) P2 (t)... Pp (t) I = Gy (t) +n (t).

The new channel estimate matrix G contains the p selected row of the previous channel estimate >.

The next step is the estimate the noise cross-correlation matrix: Q = Efn (t) n (t)} This can be done by calculating the sample error cross-correlation matrix: A significant benefit of the present invention can be seen in the foregoing equations. By selecting only p of the r branches, the vector p becomes smaller.

This, in turn, reduces the noise cross-correlation matrix Q, e. g., from an 8x8 matrix to a 3x3 or 4x4 matrix. Since Q may need to be updated very frequently, this results in significant processing reduction. If the noise is Gaussian distributed, then the MLSE solution (Maximum Likelihood Sequence Estimator) can be described as: This can be efficiently implemented using the Viterbi algorithm. Those skilled in the art will be familiar with Viterbi implementations which are not described in further detail here. The metric used in this case will be: J = (p (t)-ây (t)) Q~1 (p (t)-ây (t)) g Exemplary embodiments with two (or more) antenna arrays, e. g., that are shown in FIG. 4, can be described using the same mathematical framework as above, but with the difference that the beamforming matrix B now is a block matrix.

For example, with two antenna arrays the beamforming matrix B may take the form: where B1 and B2 can be of different sizes.

Having provided some additional details regarding the signal processing occurring in, for example, blocks 202,220,230,240 and 250 of FIG. 2, a brief summary of IRC techniques performed at block 260 is provided below. For more information the interested reader is referred to the parent IRC applications. IRC techniques expand conventional diversity combining techniques to exploit the above- described correlation, whereby significant gains in the quality of the received signal are realized. The branch metrics Mh (n) formed according to IRC techniques can be described by the following equation.

Mb (n) =[r(n)-c (n) a (n)] H A (n) [r (n)-C (n) eh (n)] =oh (n) A (n) oh (n) where: n is a time index; r (n) [ra (n), rb (n)], are the signal samples received on each antenna; C(n) = are the channel tap estimates of the form Cx (s) where C is the delay, i. e., T=0 is the main ray, # =1 is the first echo, etc.; sh (n) = [sh (n), Sh (n-l)...] T, are the hypothesized signal samples; z (n) = [za (n), Zb (n)] T, are the signal impairments received on each antenna; A (n) = Rzz (n)-', or a related quantity, where Rzz is the impairment correlation matrix which equals the expected value E (z (n) zH (n)); eh (n) = r (n)-C (n) sh (n), is an estimate of the impairment for a given hypothesis.

The A (n) matrix (i. e., the A-matrix) is the inverse of the R (n) matrix, or a related quantity such as the adjoint or pseudo-inverse. As will be apparent to a person skilled in the art reading this application, Rzz (n) and A (n) are specific examples of impairment correlation properties of which other forms are known.

Throughout the following, the term A-matrix is used generically to refer to any estimate of the impairment correlation properties.

Determination of the A-matrix for use in the present invention can be performed in a number of ways depending upon the specific application and the required performance. The simplest approach is to use a fixed set of values for the A-matrix, stored in memory, that are never updated. These values depend primarily on the configuration of the receive antennas and on the carrier frequencies being employed. An alternative approach is to determine the A-matrix from synchronization information and to keep the A-matrix values constant between synchronization fields, or other known fields. At each new occurrence of the synchronization field, the A-matrix can be recomputed, with or without use of the previous A-matrix values. Another alternative approach is to use synchronization fields to initialize, or improve, the A-matrix values and then to use decisions made on the data field symbols to track the A-matrix values.

Also, consideration is given for the method used to track the A-matrix values. Since the A-matrix comprises information regarding the impairment correlation properties between the antennas in the antenna array, standard estimation methods for estimating correlation or inverse correlation matrices can be applied.

Using either known or detected symbol values, impairment values can be obtained by taking the differences between the received signal sample streams and the hypothesized received signal sample streams. At time n, this gives a vector of impairment values, denoted z (n); one value for each antenna. A straightforward way of forming the A-matrix is given by: A (n) =Ris (In) R (n) =AR (n-l) +Kz (n) zH (n)

K is a scaling constant, typically 1 or Because R (n) is a Hermitian matrix, only a portion of the matrix elements need be computed.

Such a straightforward approach is fairly high in complexity. One way to reduce complexity is to apply the matrix inversion lemma and update the A-matrix directly as: where: p (n) =A (n-l) z (n) Because the A-matrix is Hermitian, it is only necessary to compute those elements on the diagonal and either those elements above or below the diagonal.

These techniques for estimating and tracking the A-matrix are given only for purposes of illustration. In general, the A-matrix can be expressed and estimated in a variety of ways, as will be appreciated by a person skilled in the art who is reading this application. The present invention may also be applied to the blind equalization problem in which known synchronization sequences are absent. In this case, the A-matrix is estimated in a manner similar to how the channel is estimated.

While the present invention has been described with respect to a particular embodiment, those skilled in the art will recognize that the present invention is not limited to the specific embodiments described and illustrated herein. Different embodiments and adaptations besides those shown and described as well as many variations, modifications and equivalent arrangements will now be reasonably suggested by the foregoing specification and drawings without departing from the substance or scope of the invention. While the present invention has been described herein in detail in relation to its preferred embodiments, it is to be understood that this disclosure is only illustrative and exemplary of the present invention and is merely for the purposes of providing a full and enabling disclosure of the invention.

Accordingly, it is intended that the invention be limited only by the spirit and scope of the claims appended hereto.