DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0034] Adaptive Smart Antenna Processing

[0035] The invention is directed to a processing method for altering the transmit or receive weights used by a communication station to define a transmitted signal or to process a received signal in order to deepen or otherwise manipulate the depth of a null formed to mitigate the effects of one or more known interferers. The interferer may or may not be another remote user sharing the same communication channel with the same base station. The invention may be implemented in a communication station that includes a receiver, an array of antennas and means for adaptive smart antenna processing of received signals. The invention may also be implemented in a communication station that includes a transmitter, an array of antennas, and means for adaptive smart antenna processing of transmitted signals. In a preferred embodiment, the communication station includes a transceiver and the capability of implementing both uplink and downlink adaptive smart antenna processing.

[0036] When receiving a signal from a subscriber (remote) unit, the signals received by each of the antenna array elements are combined by the adaptive smart antenna processing elements to provide an estimate of a signal received from that subscriber unit. In the preferred embodiment, the smart antenna processing comprises linear spatial processing, wherein each of the complex-valued (i.e., including in-phase I and quadrature Q components) signals received from the antenna elements is weighted in amplitude and phase by a weighting factor and the weighted signals are then summed to provide the signal estimate. The adaptive smart antenna processing scheme (i.e., the strategy) can then be described by a set of complex valued weights, one for each of the antenna elements. These complex valued weights can be described as a single complex valued vector of M elements, where M is the number of antenna elements. Thus, in the linear case, the smart antenna processing is designed to determine a set of weights such that the sum of the products of the weights times the antenna element signals provides an estimate of the remote user's transmitted signal which satisfies some prescribed “estimation quality” measure.

[0037] This representation of the adaptive smart antenna processing can be extended to include spatio-temporal processing, where the signal at each antenna element, rather than being weighted in amplitude and phase, is filtered by a complex valued filter, typically for purposes of time equalization. In such a method, each filter can be described by a complex-valued transfer function or convolving function. The adaptive smart antenna processing of all elements can then be described by a complex valued M-vector of M complex valued convolving functions.

[0038] Several methods are known for determining the weighting vectors to be applied when processing received signals. These include methods that determine the directions of arrival of signals from subscriber units, and methods that use the spatial or spatio-temporal characteristics of subscriber units, for example, the spatial or spatio-temporal signatures. See for example U.S. Pat. Nos. 5,515,378 and 5,642,353, entitled “SPATIAL DIVISION MULTIPLE ACCESS WIRELESS COMMUNICATION SYSTEMS”, to Roy, et al., assigned to the assignee of the present invention and the contents of which are incorporated herein by reference, for methods that use directions of arrival. See also the above-referenced U.S. Pat. Nos. 5,592,490 and 5,828,658 for methods that use spatial and spatio-temporal signatures.

[0039] “Blind” methods determine the weights from the signals themselves, but without resorting to a priori knowledge such as training signals or silent periods, that is, without determining what weights can best estimate a known symbol sequence (or in the case of the period silence, the absence of a known sequence). Such blind methods typically use some known characteristic of the signal transmitted by the subscriber unit to determine the best receive weights to use by constraining the estimated signal to have this property, and hence are sometimes referred to as property restoral methods.

[0040] Property restoral methods in turn can be classified into two groups. Simple property restoral methods restore one or more properties of the signal without completely reconstructing the modulated received signal, for example by demodulating and then remodulating. More complex restoral methods typically rely upon reconstruction of the received signal.

[0041] Property restoral methods determine a signal (a “reference signal”) that is constrained to the required property and then determine a set of weights corresponding to the reference signal, such that if the reference signal was transmitted by a remote user, the signals at the antenna elements of the receiving array would be acceptably “close” to the signals actually received. One example of a simple restoral method is the constant modulus (CM) method, which is applicable to communication systems that use a modulation scheme having a constant modulus, including, for example phase modulation (PM), frequency modulation (FM), phase shift keying (PSK) and frequency shift keying (FSK). The CM method has also been shown to be applicable to non-CM signals. Other partial property restoral techniques include techniques that restore the spectral properties of the signal, such as the signal's spectral self-coherence.

[0042] “Decision directed” (DD) methods construct a reference signal by making symbol decisions (e.g.,, demodulating) the received signal. Such decision directed methods use the fact that the modulation scheme of the transmitted subscriber unit signal is known, and then determine a signal (a “reference signal”) that is constrained to have the characteristics of the required modulation scheme. In such a case, the reference signal production process includes making symbol decisions. Weights are determined that produce a reference signal, that if transmitted by a remote user, would produce signals at the antenna elements of the array that are acceptably “close” to the signals actually received. See, for example, U.S. patent application Ser. No. 08/729,390, entitled “METHOD & APPARATUS FOR DECISION DIRECTED DEMODULATION USING ANTENNA ARRAYS & SPATIAL PROCESSING” to Barratt, et al., and serial no. 09/153,110, entitled “METHOD FOR REFERENCE SIGNAL GENERATION IN THE PRESENCE OF FREQUENCY OFFSETS IN A COMMUNICATION STATION WITH SPATIAL PROCESSING” to Petrus, et al., both of which are assigned to the assignee of the present invention and the contents of which are incorporated herein by reference, for descriptions of systems that use decision directed weight determination methods.

[0043] As previously mentioned, weight determining schemes also are known that use training data, that is, data whose symbols are known a priori. The training data (possibly with a timing offset or frequency offset, or both applied) is then used as a reference signal to determine the smart antenna processing strategy (e.g., the weights). Therefore, reference signal based methods include the case in which the reference signal includes training data, the case in which the reference signal includes a signal constrained to have some property of the transmitted signal, and the case in which the reference signal includes constructing a signal based on making symbol decisions.

[0044] Non-linear uplink and downlink processing strategies also are known. In the uplink direction, such methods typically include demodulation and act to determine an estimate of the symbols transmitted by a desired remote user from the set of signals received at the antenna elements of the communication station. One known example of such a processing scheme is based on a Viterbi algorithm using branch metrics. In this regard, it is noted that the present invention is not limited to linear spatial and spatio-temporal processing methods that include weight determining, but also is equally applicable to non-linear methods such as those based on Viterbi algorithms and branch metrics, which may not necessarily include determining weights.

[0045] In theory, adaptive smart antenna processing permits more than one communication link to exist in a single “conventional” communication channel so long as the subscriber units that share the conventional channel can be spatially (or spatio-temporally) resolved. A conventional channel includes a frequency channel in a frequency division multiple access (FDMA) system, a time slot in a time division multiple access (TDMA) system (which usually also includes FDMA, so the conventional channel is a time and frequency slot), and a code in a code division multiple access (CDMA) system. The conventional channel is then said to be divided into one or more “spatial” channels, and when more than one spatial channel exists per conventional channel, the multiplexing is called space division multiple access (SDMA). SDMA is used herein to include the possibility of adaptive smart antenna processing, both with one and with more than one spatial channel per conventional channel.

[0046] Base Station Architecture

[0047] The preferred embodiment of the inventive method and apparatus is implemented in a communication receiver, in particular, a Personal Handyphone System (PHS)-based antenna-array communication station (transceiver) such as that shown in FIG. 1 , with M antenna elements in the antenna array. The PHS standard is described, for example, in the Association of Radio Industries and Businesses (ARIB, Japan) Preliminary Standard, Version 2, RCR STD-28 and variations are described in Technical Standards of the PHS Memorandum of Understanding Group (PHS MoU—see http://www.phsmou.or.jp). The preferred embodiments of the present invention may be implemented in two versions of the communication station of FIG. 1 , one aimed at low-mobility PHS system, with M=4, and another, aimed at a wireless local loop (WLL) system, with a variable number, with typically M=12.

[0048] While systems having some elements similar to that shown in FIG. 1 may be prior art, a system such as that of FIG. 1 with elements 131 and 133 capable of implementing the inventive method is not prior art. Note that the present invention is in no way restricted to using the PHS air interface or to TDMA systems, but may be utilized as part of any communication receiver that includes adaptive smart antenna processing means, including CDMA systems using the IS-95 air interface and systems that use the common GSM air interface.

[0049] In the system of FIG. 1, a transmit/receive (“TR”) switch 107 is connected between an M-antenna array 103 and both transmit electronics 113 (including one or more transmit signal processors 119 and M transmitters 120 ), and receive electronics 121 (including M receivers 122 and one or more receive signal processors 123 ). Switch 107 is used to selectively connect one or more elements of antenna array 103 to the transmit electronics 113 when in the transmit mode and to receive electronics 121 when in the receive mode. Two possible implementations of switch 107 are as a frequency duplexer in a frequency division duplex (FDD) system, and as a time switch in a time division duplex (TDD) system.

[0050] The PHS form of the preferred embodiment of the present invention uses TDD. The transmitters 120 and receivers 122 may be implemented using analog electronics, digital electronics, or a combination of the two. The preferred embodiment of receivers 122 generate digitized signals that are fed to signal processor or processors 123 . Signal processors 119 and 123 incorporate software and/or hardware for implementing the inventive method and may be static (always the same processing stages), dynamic (changing processing depending on desired directivity), or smart (changing processing depending on received signals). In the preferred embodiments of the invention, processors 119 and 123 are adaptive. Signal processors 119 and 123 may be the same DSP device or DSP devices with different programming for the reception and transmission, or different DSP devices, or different devices for some functions, and the same for others. Elements 131 and 133 are for implementing the method of the present invention for downlink and uplink processing, respectively, in this embodiment, and include programming instructions for implementing the processing methods.

[0051] Note that while FIG. 1 shows a transceiver in which the same antenna elements are used for both reception and transmission, it should be clear that separate antennas for receiving and transmitting may also be used, and that antennas capable of only receiving or only transmitting or both receiving and transmitting may be used with adaptive smart antenna processing.

[0052] The PHS system is an 8 slot time division multiple access (TDMA) system with true time division duplex (TDD). Thus, the 8 timeslots are divided into 4 transmit (TX) timeslots and 4 receive (RX) timeslots. This implies that for any particular channel, the receive frequency is the same as the transmit frequency. It also implies reciprocity, i.e., the propagation path for both the downlink (from base station to users' remote terminals) and the uplink (from users' remote terminals to base station) is identical, assuming minimal motion of the subscriber unit between receive timeslots and transmit timeslots. The frequency band of the PHS system used in the preferred embodiment is 1895-1918.1 MHz. Each of the 8 timeslots is 625 microseconds long. The PHS system includes a dedicated frequency and timeslot for a control channel on which call initialization takes place. Once a link is established, the call is handed to a service channel for regular communications. Communication occurs in any channel at the rate of 32 kbits per second (kbps), a rate termed the “full rate”. Less than full rate communication is also possible, and the details of how to modify the embodiments described herein to incorporate less than full rate communication would be clear to those of ordinary skill in the art.

[0053] In the PHS used in the preferred embodiment, a burst is defined as the finite duration RF signal that is transmitted or received over the air during a single timeslot. A group is defined as one set of 4 TX and 4 RX timeslots. A group always begins with the first TX timeslot, and its time duration is 8×0.625=5 msec.

[0054] The PHS system uses (π/4 differential quaternary (or quadrature) phase shift keying (π/4 DQPSK) modulation for the baseband signal. The baud rate is 192 kbaud. There are thus 192,000 symbols per second.

[0055] FIG. 2 is a more detailed block diagram of a transceiver that includes a signal processor capable of executing a set of instructions for implementing the method of the present invention. This is the version of the FIG. 1 system suitable for use in a low-mobility PHS system. In FIG. 2, a plurality of M antennas 103 are used, where M=4. More or fewer antenna elements may be used. The outputs of the antennas are connected to a duplexer switch 107 , which in this TDD system is a time switch. When receiving, the antenna outputs are connected via switch 107 to a receiver 205 , and are mixed down in analog by RF receiver modules 205 from the carrier frequency (around 1.9 GHz) to an intermediate frequency (“IF”). This signal is then digitized (sampled) by analog to digital converters (“ADCs”) 209 . The result is then down converted digitally by digital downconverter 213 to produce a four-times oversampled complex valued (in phase I and quadrature Q) sampled signal. Thus, elements 205 , 209 and 213 correspond to elements that might be found in receiver 122 of FIG. 1 . For each of the M receive timeslots, the M downconverted outputs from the M antennas are fed to a digital signal processor (DSP) device 217 (hereinafter “timeslot processor”) for further processing. In the preferred embodiment, commercial DSP devices are used as timeslot processors, one per receive timeslot per spatial channel.

[0056] The timeslot processors 217 perform several functions, which may include the following: received signal power monitoring, frequency offset estimation/correction and timing offset estimation/correction, smart antenna processing (including determining receive weights for each antenna element to determine a signal from a particular remote user, in accordance with the present invention), and demodulation of the determined signal. The version of the uplink processing method of the invention as implemented in each timeslot processor 217 in the embodiment of FIG. 2 is shown as block 241 .

[0057] The output of the timeslot processor 217 is a demodulated data burst for each of the M receive timeslots. This data is sent to host DSP processor 231 whose main function is to control all elements of the system and interface with the higher level processing (i.e., processing which deals with what signals are required for communications in the different control and service communication channels defined in the PHS communication protocol). In the preferred embodiment, host DSP 231 is a commercial DSP device. In one implementation of the present invention, timeslot processors 217 send the determined receive weights to host DSP 231 . Note that if desired, the receive weights may also be determined by software specifically implemented in host DSP 231 .

[0058] RF controller 233 interfaces with the RF transmit elements, shown as block 245 and also produces a number of timing signals that are used by both the transmit elements and the modem. RF controller 233 receives its timing parameters and other settings for each burst from host DSP 231 .

[0059] Transmit controller/modulator 237 receives transmit data from host DSP 231 . Transmit controller 237 uses this data to produce analog IF outputs which are sent to RF transmitter (TX) modules 245 . The specific operations performed by transmit controller/modulator 237 include: converting data bits into a complex valued (π/4 DQPSK) modulated signal; up-converting to an intermediate frequency (IF); weighting by complex valued transmit weights obtained from host DSP 231 ; and, converting the signals to be transmitted using digital to analog converters (“DACs”) to form analog transmit waveforms which are provided to transmit modules 245 .

[0060] The downlink processing method of the invention is implemented in the embodiment of FIG. 2 in host DSP 231 , and is shown as block 243 . In alternate versions, the downlink processing method is implemented in the timeslot processors 217 and in another version, it is implemented in transmit controller/modulator 237 .

[0061] Transmit modules 245 upconvert the signals to the transmission frequency and amplify the signals. The amplified transmission signal outputs are coupled to the M antennas 103 via duplexer/time switch 107 .

[0062] In describing the inventive methods, the following notation is used. Given M antenna elements (M=4 in one implementation, and 12 in another embodiment), let z ₁ (t), z ₂ (t), . . . , z _M (t) be the complex valued responses (that is, with in-phase I and quadrature Q components) of the first, second, . . . , M'th antenna elements, respectively, after down-conversion, that is, in baseband, and after sampling (four-times oversampling in the preferred embodiment). In the above notation, but not necessarily required for the present invention, t is discrete. These M time-sampled quantities can be represented by a single M-vector z(t) with the i'th row of z(t) being z _i (t). For each burst, a finite number of samples, say N, is collected, so that z ₁ (t), z ₂ (t), . . . , Z _M (t) can each be represented as a N-row vector and z(t) can be represented by a M by N matrix Z. In much of the detailed description presented hereinafter, such details of incorporating a finite number of samples are assumed known, and how to include these details would be clear to those of ordinary skill in the art.

[0063] Assume signals are transmitted to the base station from N _S remote users all operating on the same (conventional) channel. In particular, assume that one of these, a particular subscriber unit of interest, transmits a signal s(t). Linear adaptive smart antenna processing, which is used in the preferred embodiment of the invention, includes taking a particular combination of the I values and the Q values of the received antenna element signals z ₁ (t), z ₂ (t), . . . , z _M (t) in order to extract an estimate of the transmitted signal s(t). Such complex valued weights may be represented by the receive weight vector for this particular subscriber unit, denoted by a complex valued weight vector w _r , with i ^th element w _ri . The estimate of the transmitted signal from the remote unit may then be represented as: 1 $\begin{array}{cc}s\left(t\right)=\sum _{i=1}^Mw_{\mathrm{ri}}^{\prime }z_i\left(t\right)=w_r^Hz\left(t\right)& \left(1\right)\end{array}$

[0064] where w′ _ri is the complex conjugate of w _ri and w _r ^H is the Hermitian transpose (that is, the transpose and complex conjugate) of receive weight vector w _r . Eq. 1 is called a copy signal operation, and the signal estimate s(t) thus obtained is called a copy signal.

[0065] The spatial processing described by Eq. 1 may be re-written in vector form for the case of N samples of M-vector signals z(t) and N samples of the transmitted signal s(t) being estimated. In such a case, let s be a (1 by N) row vector of the N samples of s(t). The copy signal operation of Eq. 1 may then be re-written as s=w _r ^H Z.

[0066] In embodiments which include spatio-temporal processing, each element in the receive weight vector is a function of time, so that the weight vector may be denoted as w _r (t), with ith element w _ri (t). The estimate of the signal may then be expressed as: 2 $\begin{array}{cc}s\left(t\right)=\sum _{i=1}^Mw_{\mathrm{ri}}^{\prime }\left(t\right)*z_i\left(t\right)& \left(2\right)\end{array}$

[0067] where the operator “*” represents the convolution operation. Spatio-temporal processing may combine time equalization with spatial processing, and is particularly useful for wideband signals. Forming the estimate of the signal using spatio-temporal processing may equivalently be carried out in the frequency (Fourier transform) domain. Denoting the frequency domain representations of s(t), z _i (t), and w _ri (t) by S(k), Z _i (k), and W _i (k), respectively, where k is the discrete frequency value: 3 $\begin{array}{cc}S\left(k\right)=\sum _{i=1}^MW_{\mathrm{ri}}^{\prime }\left(k\right)Z_i\left(k\right).& \left(3\right)\end{array}$

[0068] With spatio temporal processing, the convolution operation of Equation (2) is usually finite and when performed on sampled data, equivalent to combining the spatial processing with time equalization using a time-domain equalizer with a finite number of equalizer taps. That is, each of the w _ri (t) has a finite number of values of t and equivalently, in the frequency domain, each of the W _i (k) has a finite number of k values. If the length of the convolving functions w _ri (t) is K, then rather than determining a complex valued M-weight vector w _r , one determines a complex valued M by K matrix W _r whose columns are the K values of w _r (t).

[0069] Alternatively, a spatial weight determining method can be modified for spatio-temporal processing according to a weight matrix by re-expressing the problem in terms of matrices and vectors of different sizes. As throughout this description, let M be the number of antenna elements, and N be the number of samples. Let K be the number of time equalizer taps per antenna element. Each row vector of N samples of the (M by N) received signal matrix Z can be rewritten as K rows of shifted versions of the first row to produce a received signal matrix Z of size (MK by N), which when pre-multiplied by the Hermitian transpose of a weight vector of size (MK by 1), produces an estimated received signal row vector of N samples. The spatio-temporal problem can thus be re-expressed as a weight vector determining problem.

[0070] For example, for covariance based methods, the weight vector is a “long” weight vector of size (MK by 1), the covariance matrix R _zz =ZZ ^H is a matrix of size (MK by MK), and the correlation of the antenna signals Z with some signal represented by a (1 by N) row vector s is r _zs =Zs ^H , a long vector of size (MK by 1). Rearranging terms in the “long” weight vector provides the required (M by K) weight matrix.

[0071] A downlink (i.e., transmit) processing strategy using adaptive smart antenna processing includes transmitting a signal, denoted in the finite sampled case by a (1 by N) vector s, from the communication station to a particular remote user by forming a set of antenna signals (typically, but not necessarily in baseband). Linear smart antenna processing determines the antenna signals as:

Z=w _t s,

[0072] where w _t is the downlink (or transmit) weight vector. When SDMA is used to transmit to several remote users on the same (conventional) channel, the sum of w _ti for different signals s _i aimed at different remote users is formed, to be transmitted by the M antenna elements.

[0073] Note that a downlink strategy, for example including determining the downlink weights w _t , may be implemented by basing it on an uplink strategy, for example uplink weights, together with calibration data. In this situation, the calibration accounts for differences in the receive and transmit electronic paths for the different antenna elements. The downlink strategy may also be found from the uplink strategy by using the transmit spatial signatures of the remote users, or by other known methods. Again, the weight vector formulation may be used for both linear spatial processing and linear spatio-temporal processing.

[0074] Consider a communications station communicating with N _s remote users on any channel. On the uplink, for an M antenna system, the M received signals at the base station can be stacked into an (M by 1) received signal vector z(t) which may be modeled as 4 $\begin{array}{cc}z\left(t\right)=\sum _{i=1}^{\mathrm{Ns}}a_{i}s_i\left(t\right)+v\left(t\right),& \left(4\right)\end{array}$

[0075] where t is time, which in the preferred embodiment is discrete, s _i (t), i=1, . . . , N _s , denotes the signal (preferably in baseband) at time t transmitted by the ith of N _s remote users, a _i is the spatial (or spatio-temporal) signature (a complex valued M-vector) of the ith remote co-channel user, and v(t) denotes additive noise (which may include other interfering signals, viewed as noise for any remote user). Note that the signature of any user is the signal received at the M antenna elements in the absence of noise and the absence of any other interfering users when the user transmits a unit impulse signal (see above referenced above-referenced U.S. Pat. Nos. 5,592,490 and 5,828,658). In matrix form the above signal model can be represented as

z ( t )= As ( t )+ v ( t ). (5)

[0076] where

A=[a ₁ a ₂ . . . a _Ns ], and s ( t )=[ s ₁ ( t ) s ₂ ( t ) . . . s _Ns ( t )] ^T . (6)

[0077] with the superscript T denoting the matrix transpose.

[0078] Processing Strategy Computation Methods

[0079] The preferred embodiment of the invention improves a method for computing an uplink or downlink processing strategy that uses as inputs the received antenna signal data and typically a reference signal, and that takes into account the interference environment present in the received antenna signal data for interference mitigation. The improvement is to deepen or otherwise modify the null depth the strategy provides for mitigating the effects of any one or more known interferers. As would be known to those in the art, interference mitigating strategy determining methods explicitly or implicitly use one or more characteristic features of the received antenna data. Depending on the known strategy computation method, the input signal data may be explicitly reduced to one or more particular characteristic feature that the known method explicitly uses for its computation. For example, for methods that use the spatial or spatio-temporal covariance matrix of the input, the data may be reduced to the spatial or spatio-temporal covariance matrix of the data. Other methods may be based on other properties and in such cases the input signal data may be reduced to the particular feature or property which the known strategy method utilizes. Yet other methods, while implicitly dependent on some property (e.g., spatio-temporal covariance) of the received signal, do not require explicit estimation of the property such as the covariance.

[0080] Since some of the discussion below applies for both receive and transmit strategy, the “r” or “t” subscripts are omitted in such quantities as the weight vectors w. Such subscripts may be used explicitly to identify uplink or downlink processing, and their addition will be clear from the context to those of ordinary skill in the art.

[0081] Let Z be the matrix of received antenna array signals, preferably but not necessarily in baseband. Let s ₁ , be a (1 by N ₁ ) reference signal vector of N ₁ samples. A reference signal s ₁ , may be a known training sequence, or, in decision directed methods, a signal constructed to have the same known modulation structure as the signal transmitted by the subscriber unit, or for property restoral methods, a signal that is constrained to have the required property.

[0082] The invention may be applied to deepen the nulls resulting from uplink or downlink strategies determined with any strategy computation method. Thus, in all of the embodiments of the invention to be described herein, a known strategy (e.g., weight) determining process is used that computes an uplink or a downlink strategy based on received signal inputs Z from the antenna array elements. Many such methods also use reference signal, denoted s ₁ determined for the remote user signal of interest from the received signal data Z. Such a strategy determining method is shown in FIG. 3 , which is marked prior-art, but is only prior art without the modifications described herein. The reference signal is typically extracted from a set Z of received data as shown in FIG. 4 . FIG. 5 is one case of the system of FIG. 3 wherein the method explicitly uses as an input at least one of the characteristic features of the set of received signal data Z, so in FIG. 5 , the set of input data is further operated upon to obtain one or more characteristic features (e.g., the covariance, the principal components of the covariance matrix, a particular feature of the data, etc.) of the data, which is then explicitly used in the known strategy (e.g., weight) determining method.

[0083] One example of a weight determining method that has the structure shown in FIG. 5 is the well known least squares (MSE) technique computes the uplink or downlink weights by solving the minimization problem:

w =argmin _w ∥w ^H Z−s ₁ ∥ ² =( R _zz ) ⁻¹ r _zs =( ZZ ^H ) ⁻¹ Zs ₁ ^H , (7A)

[0084] where R _zz =ZZ ^H is the spatial (or spatio-temporal) covariance matrix of the antenna signals, ∥ ∥ denotes the vector norm, and r _zs =Zs ₁ ^H is the cross correlation between the antenna signals and the reference signal. The method of Eq. (7A) is called the minimum mean squared error (MMSE) method herein, and uses the covariance R _zz of the input signal data as the characteristic feature for weight determining. Thus, calculation of the weights according to Eq. (7A) requires having data corresponding to the signals received at the antenna elements and a reference signal. In practice, this typically entails identifying the reference signal data and extracting it from a data burst, forming the covariance matrix for the received signals in the data burst, forming the cross correlation term, and solving for the weights.

[0085] In practice, for example in a mobile PHS, because of limited computational power, only a small number of the 960 samples making up a received signal burst are used to determine R _zz ⁻¹ . In addition, the samples are determined at the baud points rather than being oversampled. Using only a small number of samples to determine weights may cause what is called “overtraining” herein: the weights performing well on the data containing the reference signal but not on new data, i.e., while the weights extract the desired signal energy and reject interference, the weights may perform poorly on new data. An improvement to the least squares technique which can assist in this situation includes a process termed “diagonal loading” (see, for example, B. L. Carlson: “Covariance matrix estimation errors and diagonal loading in adaptive arrays,” IEEE Transactions on Aerospace and Electronic Systems , vol. 24, no. 4, July 1988), by which a diagonal adjustment is added as follows:

w =( ZZ ^H +γI ) ⁻¹ Zs ₁ ^H , (7B)

[0086] where γ is a small adjustable factor used to improve the performance of the least squares solution by reducing sensitivity to statistical fluctuations in Z.

[0087] The methods described by Eq. (7B) also is of the structure shown in FIG. 5 , again with the characteristic feature being the spatial (or spatio-temporal) covariance R _zz =ZZ ^H . Other uplink processing methods are known that use the noise-plus-interference covariance matrix, denoted R _vv , rather than the noise-plus-interference-plus-signal covariance matrix R _zz . Such methods include some noise plus interference covariance estimator which uses the received data (containing the signal plus interference plus noise) and the reference signal for the desired user to determine R _vv .

[0088] The invention includes two main aspects: (i) modifying the uplink or downlink strategy to realize an improved null, and (ii) accurately estimating the signature in the direction where a modified (e.g., deeper) null is desired followed by null deepening based on the estimate.

[0089] Null depth is generally limited by two effects. The first effect is the “natural” null depth of the uplink or downlink strategy. For example, when the signatures of the remote user and interferer are highly correlated, when the interferer power is not too strong, and when the signal-to-noise ratio (SNR) is modest, an uplink strategy based on maximizing the signal-to-interference-plus-noise ratio (SINR) will not direct particularly deep nulls towards the interferer. One aim of the present invention is to deepen the nulls formed under these conditions.

[0090] While a preferred embodiment is described in detail for deepening a null based on known or estimated signatures towards which the nulls are to be deepened, various embodiments are possible. They include using: synthetic signal injection, synthetic signal injection only for the “noise plus interference” estimation block, low-rank update of the noise plus interference estimate, and strategy orthogonalization. In the latter case, the uplink weights for two remote users that share a spatial channel are combined to construct weights with improved nulling performance, while again substantially preserving the rest of the null and gain pattern. These embodiments are detailed hereinunder.

[0091] The described methods for null deepening use a known or estimated signature. Known signature estimation methods such as simple (e.g., maximum likelihood) signature estimation of the interfering remote user may not produce a sufficiently accurate signature estimate for effective null deepening in the direction of the interfering remote user. This is because the estimate may suffer from contamination by the signal of the remote user to which delivered power is maximized. In the preferred embodiment, the method used includes joint signature estimation using the reference signals of both the remote user and any interferer(s). Geometric methods based, for example, on angle of arrival, may alternatively be used to estimate signatures which then are used for null deepening according to any of the signature-based null deepening embodiments described herein.

[0092] Another effect that limits null depth is the accuracy to which signatures or covariance matrices are estimated, which in turn is limited by the number of samples, the SNR of the remote user whose signature is to be estimated, and the power of other remote co-channel users. An additional aspect of the present invention is improving the signature estimate by combining data gathered over several bursts with reference signals of multiple remote users computed over the same bursts. This aspect of the invention is motivated by the recognition that the signature estimate of the interfering remote user must be accurate to a degree. To improve the null depth of a least-squares solution over a single burst, for example, the signature estimate typically may need be based on estimates derived from several bursts.

[0093] Null Deepening

[0094] Using the invention can deepen null towards one remote user or more than one users based on knowledge of the signatures of the one or more users. The signatures may be known or estimated, and one aspect of the invention is a method for estimating the signatures applicable to null deepening. For best performance, the signature estimate should be substantially free of contamination from signals to which beams are directed. The invention uses the signature (known or estimated) of one or more known interferers to modify a known strategy or known strategy determining method to produce a modified strategy with improved null deepening. The knowledge of the signatures may be, for example, from historical signature records.

[0095] For the description below, the desired remote user is denoted SU _i , and the weight vector for communicating with SU _i is denoted by w _i . When communicating with SU _i , all other users, denoted SU _j where j is not i, are interferers, and the signature for such interferers are denoted a _j , where j is not i.

[0096] The Signal Injection Method

[0097] A first embodiment of the invention applicable to one null-deepening based on the known or estimated signature of one interferer denoted SU _j with signature (estimate) a _j is shown in FIG. 6 . A signal formed from the estimated or known interferer signal is scaled and added to the received signal data Z before determining the strategy, that is, modified received signals described by

{circumflex over (Z)}=Z+α _j â _j s _synth ,

[0098] with α _j a tunable scale factor for any interferer j,s _synth is any signal that has the same temporal structure as a typical communication waveform used in the system, or preferably, a temporal structure that would be identified by the particular strategy computation method as not that of the user. For example, for a constant modulus property restoral method, using a non-constant modulus structure for s _synth ensures that the strategy computation method recognizes this as an interferer. In the preferred embodiment, however, s _synth comprises N random noise samples. In another embodiment, this is a constant signal, that is, the signal [1 1 . . . 1]. The scale factor α _j for any interferer j can be set to a large fixed value to force a deep accurate null towards that undesired subscriber unit signature. The modified input signal {circumflex over (Z)} is then applied to any strategy determining scheme, including the ones shown in FIG. 3 and FIG. 5 .

[0099] A generalization of this embodiment, called “signal injection” herein is illustrated in FIG. 7 . Here, nulls are directed to several interferers denoted SU _j where j is not i. Signals are formed from the estimated (or known) signatures of the undesired subscriber units, then are scaled and added to received signal data Z before determining the strategy, that is, modified received signals described by 5 $\begin{array}{cc}\stackrel{⋒}{Z}=Z+\sum _{j\ne i}{\alpha }_j{\hat{a}}_js_{\mathrm{synth}\_j},& \left(8\right)\end{array}$

[0100] with α _j a tunable scale factor for any interferer j, are used in place of Z and applied to the strategy determining method, say that of FIG. 3 or that of FIG. 5 . for processing SUi signals. The scale factor α _j for any interferer j can be set to a large fixed value to force a deep accurate null towards each undesired subscriber unit signature. Each of the s _{synth — j} is any signal that has the same temporal structure as a typical communication waveform used in the system and that would be recognized a non signal of interest, and s _{synth — j} for any interferer j must be non-collinear with any s _{synth — j} for any other interferer j′. Again, in the preferred embodiment, each s _{synth — j} is a set of random samples, with the sets for two distinct interferers being statistically independent.

[0101] An improved embodiment makes α _j a function of the relative power levels transmitted to SU _i and jth subscriber units, or to any similar measure derived from received signal strengths, thus forming deep nulls in a controlled way only where needed. That is, the nulls are strong where strong nulls are needed and weak where such strong nulls are not needed. Thus forming weak nulls where allowable would reserve degrees of freedom for use elsewhere, for example to null additional noise sources.

[0102] The parameter setting method is described in more detail later in this document.

[0103] FIG. 8 shows an embodiment of the signal injection method for the case of null-deepening for one interferer for a covariance based strategy determining method. FIG. 9 shows an embodiment of the signal injection method for the case of null-deepening for one interferer for an interference-plus-noise covariance based strategy determining method.

[0104] FIG. 10 is a block diagram showing an embodiment of the inventive method applied to a non-linear strategy generator, where the strategy generator includes a demodulation stage. This may be a Viterbi-algorithm-based decoder which operates based on branch metrics, for example. The invention is shown supplying a modified noise-plus-interference covariance to a demodulator, with the modification based on the signal injection method. The resulting strategy is then applied to some new input data. The demodulator, in this case, may be a Viterbi-algorithm based demodulator in which covariance information is used to “whiten” (i.e., decorrelate) the input data, in this case the new input data. In an alternative, the reference data might also be used to estimate the channel, and provided in the form needed for the Viterbi-method in the demodulator.

[0105] Note that while the signal injection methods preferably involve adding a fraction of the “synthetic” signal generated from the one or more known or estimated interferer signatures, other, non-additive, methods of combining also are within the scope of the present invention. This is shown in FIG. 11 . For example, one method for combining includes performing a matrix factorization of the input data and the synthetic data into factors, and then combining the resulting factors to form a combined signal. The factorization may be a generalized singular value decomposition.

[0106] FIG. 12 is a block diagram showing a general application of the inventive method as used to process data in the downlink data. As shown in FIG. 12 , an uplink strategy generator (e.g., a weight determining method) is based on a reference signal (obtained from the primary data) and a set of data. In accordance with the present invention, the data input to the strategy generator is a combination of the input data and a signal generated from one or more known or estimated interferer signatures. The output of the strategy generator is a set of parameters which is used in the processing of the downlink data. In the linear spatial processing case, the parameters may be a set of weights. Note that the parameters produced by the strategy generator may be combined with calibration data to produce the required downlink strategy parameters for processing of downlink data.

[0107] Strategy Orthogonalization Method

[0108] According to this implementation, for the ith remote user, a strategy (e.g., a weight vector ŵ _i ) is determined that is orthogonal to the to estimated (or known) interferer signature(s) a _j . Moreover, the strategy is determined from the strategy (e.g., a weight vector w _i ) for the desired user SU _i and the strategy (e.g., a weight vector w _j ) used to communicate with the interferer j. The strategies for the user and interferer, respectively, for example weight vectors w _i and w _j , respectively, may be obtained using some known strategy determining method, for example that of FIG. 3 or FIG. 5 (including using Eq. (7A) or Eq. (7B)).

[0109] In the linear strategy case, assume a linear strategy w _i (spatial or spatio-temporal) exists for receiving from or transmitting to a desired remote user, say the ith, and another linear strategy w _j for receiving from or transmitting to an interferer, say the jth user, where j≠i, where w _i and w _j have substantially the same null and gain patterns, except of course that w _i nulls the interferer and w _j nulls the desired remote user i because when communicating with the interferer, the remote user is an interferer. According to this embodiment of the invention, the improved strategy is

ŵ _i =w _i +α _j w _j , (9)

[0110] where the scalar tunable factor α _j is chosen to make ŵ _i orthogonal to the interferer signature a _j . That is, α _j is chosen so that a _j ^H w _i =0. This is obtained with 6 ${\alpha }_j=-\frac{a_j^Hw_i}{a_j^Hw_j}.$

[0111] A generalization when one has a number of interferers to which nulls are to be deepened is, given a strategy w _i to a desired user i, and several weight vectors w _j , j≠i, for interferers, then the strategy (i.e., weight vector) to use is 7 ${\hat{w}}_i=w_i+\sum _{j\ne i}{\alpha }_jw_j,$

[0112] where the α _j are chosen to ensure that the modified strategy ŵ _i is orthogonal to the interferer signatures a _j , j≠i.

[0113] Another particular orthogonalization implementation that modifies the strategy w _i to be orthogonal to all known interferer signatures is, 8 ${\hat{w}}_i=w_i-\sum _{j\ne 1}{\gamma }_j{\hat{a}}_j,$

[0114] where factor γ _j is chosen to make ŵ _i orthogonal to all the interferer signatures a _j , j≠i. The factors γ _j may be computed using Gram-Schmidt orthogonalization, which in the case of a single interferer j leads to 9 ${\gamma }_j=\frac{w_i^H{\hat{a}}_j}{{{\hat{a}}_j}^2}.$

[0115] A disadvantage of the this forced signature orthogonalization method, however, is that it may disturb nulls which were formed towards “coherent” noise sources (e.g., from neighboring cells in a cellular system) or towards other incoherent interferers, and may have an effect on the “main lobe” towards the desired SU. Therefore, the preferred strategy orthogonalization method is that of Eq. (9) or its extension to multiple interferers.

[0116] Covariance Modification

[0117] A third embodiment is applicable to strategy determining methods that use an estimate of the spatial or spatio-temporal covariance (including the interference-plus-noise covariance) determined from the input signal (and possible a reference signal) to determine the strategy. The methods of Eqs. (7A) and (7B) so use the covariance to determine a weight vector. One embodiment of the invention adds some information from the interferer signatures to data used by the covariance estimator of the known weight determining method. For example, if the known weight computation method of Eq. (7B) is used, according to this embodiment of the invention, the weights to use are determined as: 10 $\begin{array}{cc}\hat{w}={\left({\mathrm{ZZ}}^H+\gamma I+\sum _{j\ne i}{\alpha }_ja_{j}a_j^H\right)}^{-1}{\mathrm{Zs}}_i^H,& \left(10\right)\end{array}$

[0118] where Z is the received data, and α _j again a tunable scale factor for any interferer j, set as described above for the signal injection method.

[0119] Many modifications are possible within the scope of the invention. In general, any uplink (i.e., receive) processing method which takes as input a spatial covariance matrix R _zz =ZZ ^H computed from input data Z can be modified by altering the spatial covariance matrix R _zz to incorporate the effects of the interferers towards which the nulls are desired to be deepened, the modification being: 11 $\begin{array}{cc}{\tilde{R}}_{\mathrm{zz}}=\left({\mathrm{ZZ}}^H+\sum _{j\ne i}{\alpha }_ja_ja_j^H\right).& \left(11\right)\end{array}$

[0120] Note that in the preferred embodiment, the weight determining method to which the invention is applied takes as input a spatial covariance matrix R _zz and a reference signal cross correlation r _zs =Zs _i ^H .

[0121] FIG. 13 shows one version of this embodiment. The weight strategy computation method is one that uses covariances to determine spatial (or spatio-temporal) weights. These weights might be uplink weights, or may be downlink weights. The known scheme is modified using an aspect of the present invention to determine weights that include interference mitigation based on signature data on known interferers. The computation required includes estimating the known interferer spatial signatures when such spatial signature estimates are unavailable, and also estimating the covariance of those signature estimates and adding a tunable fraction of these interferer covariances to the covariance estimate determined from the received data.

[0122] Other covariance matrix modifications also are possible within the scope of the invention. In yet another alternative shown in FIG. 14 , rather than the combining of the interferer data being by adding part of the interferer covariance matrix to the input signal covariance matrix, other methods of combining could be used. For example, the method for combining may includes performing a matrix factorization of the input data and of the signatures into factors, and then combining the resulting factors by one of several methods to form a combined covariance matrix.

[0123] This embodiment can also be applied to processing methods that use the noise-plus-interference covariance matrix, denoted R _vv , rather than the noise-plus-interference-plus-signal covariance matrix R _zz . Such methods include some noise plus interference covariance estimator which uses the received data (containing the signal plus interference plus noise) and the reference signal for the desired user to determine R _vv . An embodiment of the invention used in such a method is shown in FIG. 15 , which shows one embodiment of the method of the invention applied to such R _vv -based techniques, with R _vv substituted for R _zz . That is, for R _vv obtained from the signal data, a modification 12 $\begin{array}{cc}{\tilde{R}}_{\mathrm{zz}}=\left(R_{\mathrm{vv}}+\sum _{j\ne i}{\alpha }_ja_ja_j^H\right).& \left(12\right)\end{array}$