Adaptive smart antenna processing method and apparatus

[0001] (Technical field to which the invention belongs) The present invention relates to a wireless communication system, and more particularly to determining a weight for processing an adaptive smart antenna in a wireless communication receiver including an array of antenna elements and processing means of the adaptive smart antenna. [0002] (background) A wireless communication system having a communication station including an antenna array and a processing means for an adaptive smart antenna is well known. Such a communication station may be called a smart antenna communication station. When a signal is received from a subscriber unit, the signal received by each of the antenna array elements is combined with adaptive smart antenna processing means to provide an estimate of the signal received by a particular subscriber unit. With smart antenna processing with linear spatial processing, each complex value (ie, including in-phase I and quadrature Q component) signals received by the antenna element is weighted by a weighting factor in terms of amplitude and phase, and then , The weighted signals are summed to obtain an estimate. Therefore, adaptive smart antenna processing means can be described by a set of complex numerical weights, one for each antenna element. These complex value weights can be described as a single complex value vector of m elements. Here, m is the number of antenna elements. This can also include space-time processing. Here, the signal of each antenna element is not simply weighted by amplitude and phase, but is filtered by some complex value filters, usually for time equalization. Each filter can be described by a complex value transfer function or a convolution function. The adaptive smart antenna processing of all elements can then be described by the complex-valued m-vector of the m-complex-valued convolution function. [0003] Several methods for determining the weight vector of the received signal are well known. These include methods of determining the direction of arrival of signals from the subscriber unit and methods of using the spatial characteristics of the subscriber unit, such as spatial signatures. For example, US Pat. Nos. 5,515,378 and 5,642,353 by Roy et al. For the method of using the arrival direction, "SPATIAL DIVISION MULTIPLE ACCESS WIRELESS COMMUNICATION SYSTEMS" and Barratt et al. US Pat. No. 5,592,490, "SPECTRALLY EFFICIENT HIGH CAPACITY WIRELESS COMMUNICATION SYSTEMS", and US Pat. No. 5,828, by Ottersten et al. See No. 658, "SPECTRALLY EFFICIENT HIGH CAPACITY WIRELESS COMMUNICATION SYSTEMS WITH SPATIO-TEMPORAL PROCESSING". The so-called "blind" method determines the weight from the signal itself, but does not rely on the training signal. That is, it does not determine which weight can best estimate a well-known symbol sequence. Such methods typically use some well-known characteristics of the signal transmitted from the subscriber unit to determine the optimal weight to use by constraining the estimate to have this property. .. Therefore, it is called the property restoration method. Property restoration methods can be divided into two groups. The "partial" property restoration method restores one or more of the normally simple properties of a signal, for example by demodulating and then remodulating, without completely reconstructing the modulated received signal. The "decision directed" (DD) method constructs an exact copy of the signal by symbolizing (eg, demodulating) the received signal. [0004] An example of the first group, that is, the partial restoration method, is the Constant Modulus (CM) method. This can be applied to communication systems that use modulation schemes with CM, including, for example, phase modulation (PM), frequency modulation (FM), phase shift keying (PSK), frequency shift keying (FSK). For example, in April 1985 by JR Treichler and MLLarimore, "New Processing Techniques Based on the Constant Modulus Algorithm" IEEE Transactions on Acoustics, Speech, and Signal Processing, vol.ASSP-33, No.2, pp.420 ~ 431. Please refer. Other partial property restoration techniques include techniques for restoring the spectral properties of a signal, such as spectral self-coherence. The spectral coherence restoration technique uses the well-known spectral coherence properties of the signal received by the antenna array. For example, under certain circumstances, it is assumed that the signal is periodic stationary, i.e., has a periodic autocorrelation function. Other methods include restoring highly ordered statistics, such as moments and cumulants. For example, "Spectral Self-Coherence Restoral: A New Approach to Blind Adaptive Signal Extraction Using Antenna Arrays" by B. Agee, S. Schell, and W. Gardner, IEEE Minutes, vol.78, No. 4, 1990 4 Moon, and US Pat. No. 5,260, by Gardner et al., [0005] The decision directed method takes advantage of the fact that the modulation scheme of the transmitted subscriber unit signal is well known to determine the weights that generate a signal with the required modulation scheme (the "reference signal"). And when transmitted from a remote user, the signal is generated by the antenna elements in the array "close" to the signal actually received. Generating that reference signal involves symbol determination. For a description of the system using decision-based weighting, see, for example, US Patent Application No. 08 / 729,390 by Barratt et al., "METHOD &amp; APPARATUS FOR DECISION DIRECTED DEMODULATION USING ANTENNA ARRAYS &amp; SPATIAL PROCESSING" (October 11, 1996). (Application), and Petrus et al., No. 09 / 153,110, "METHOD FOR REFERENCE SIGNAL GENERATION IN THE PRESENCE OF FREQUENCY OFFSETS IN A COMMUNICATIONS STATION WITH SPATIAL PROCESS IBG" (filed September 15, 1998). [0006] Some iterative methods, including partial restoration methods, such as the CM method, have a low signal-to-noise ratio (SNR) and low SINR (Signal-to-interference-plus-) that may be encountered within a communication system. It is well known that it converges even in noise-ratio) and high fading situations. In this case, the liquidity of the subscriber unit is high. In the present specification, such a method is referred to as a "repetitive weight determination method having good convergence properties". However, a method with good convergence properties iterates over and over to converge. For example, the CM method iterates many times to converge, so that it is slow to converge in an actual system. For example, in a high liquidity system, it is desirable to use a weight vector on the current burst derived from the current burst data. This means that the weight calculation is fast, which would not be possible with the CM method. On the other hand, the decision-based method is a class that converges rapidly when the initial state is, for example, the initial signal-to-noise ratio (SNR) and SINR are high, or the initial weight vector is close enough to the correct value. This is an example of the method of. A method of rapidly converging when the initial weight vector is sufficiently close to the correct value is referred to herein as a "rapid convergence iterative weight determination method". Rapid convergence methods such as the DD method are becoming more and more widely used in smart antenna-based communication stations. If such a method fails, for example in low SINR or high fading situations, this method will not converge. The problem is in communication systems with many users in the presence of high co-channel interference, i.e., when receiving a signal from a particular subscriber unit, from a signal in a conventional channel from another subscriber unit. It is becoming more and more serious when the interference is high. Such other subscriber units include several receiving communication stations, each communicating with a set of subscriber units within that cell, in the case of a cellular system, the same cell or a neighboring cell. From [0007] In theory, adaptive smart antenna processing allows a single "conventional" communication link with more than one, as long as subscriber units sharing the same conventional channel can be spatially (or spatially and temporally) decomposed. Can be in a communication channel. Traditional channels include frequency channels in frequency division multiple access (FDMA) systems and time slots in time division multiple access (TDMA) systems (which usually also include FDMA, to be precise, conventional channels are time. And frequency slots) and codes in code division multiple access (CDMA) systems. Conventional channels are then divided into one or more "spatial" channels, and if there are more than one spatial channel per conventional channel, the multiplex method is referred to as spatial division multiple access (SDMA). As used herein, SDMA is meant to include adaptive smart antenna processing with both one per conventional channel and more than one spatial channel. [0008] Rapid convergence methods, such as decision-based methods, fail when high co-channel interference is present in an SDMA system with more than one spatial channel per conventional channel. [0009] Therefore, there is a low signal-to-interference plus noise in the art for SDMA systems with one spatial channel per conventional channel and for SDMA systems with multiple spatial channels per conventional channel. There is a need for adaptive smart antenna processing methods that efficiently determine the weight of adaptive smart antenna processing in an environment) or high fading environment. [0010] Therefore, there is a need in the art for weighting methods that are well implemented under low SINR and high fading situations and that converge rapidly, i.e., with a small number of iterations. [0011] Therefore, there is a need in the art for methods that combine good convergence properties with rapid convergence properties. [0012] Therefore, there is a need in the art for a "blind" method (ie, a method that does not use training data) that combines good convergence properties (convergence when SINR is low) with rapid convergence. [0013] [Patent Document 1] U.S. Pat. No. 5,515,378 [Patent Document 2] U.S. Pat. No. 5,642,353 [Patent Document 3] U.S. Pat. No. 5,592,490 [Patent Document 4] U.S. Pat. No. 5,828,658 [Patent Document 5] U.S. Patent Application No. 08 / 729,390 [Patent Document 6] U.S. Patent Application No. 09 / 153,110 [Non-Patent Document 1] "New Processing Techniques Based on the Constant Modulus Algorithm" by JRTreichler and MLLarimore, IEEE Transactions on Acoustics, Speech, and Signal Processing, vol.ASSP-33, No.2, pp.420 ~ 431 [Non-Patent Document 2] 1988, "application of constant modulus adaptive beamformer to constant and non-constant modulus signals" by J. Lundell and B. Widrow, 1988 Asilomar Conference on Signals, Systems and Computers (ACSSC-1988), pp.432-436. Minutes [0014] (Summary) One object of the present invention is to provide a weighting method that combines the advantages of a method with good convergence properties with the advantages of a method of rapid convergence. [0015] Another object of the present invention is to provide a "blind" weighting method and apparatus that is well performed under low SINR and high fading situations and that converges rapidly, i.e., with a low number of iterations. [0016] Another purpose is adaptive smart antenna processing, which efficiently weights adaptive smart antenna processing in low SINR or high fading environments for SDMA systems with one spatial channel per conventional channel. To provide the law and equipment. [0017] Another purpose is for adaptive smart antennas that efficiently determine the weight of adaptive smart antenna processing in low SIN or high fading environments for SDMA systems with multiple spatial channels per conventional channel. It is to provide a processing method and an apparatus. [0018] Another object of the present invention is to provide an adaptive smart antenna processing method and apparatus for determining the weight of adaptive smart antenna processing used in a current burst of data, the weight of which is the data from the current burst. Fits the current burst of data by being determined by. [0019] The present invention will be better understood from the detailed preferred embodiments of the present invention, but the present invention should not be construed as limiting to a particular embodiment, these are for illustration purposes only. It is intended to facilitate understanding. The embodiment will be described with reference to the following figures. [0020] (Detailed description of preferred embodiments) Base station architecture The methods and devices of the preferred embodiment are implemented in a communication receiver having an m-antenna element in the antenna array, particularly in a PHS-based antenna array communication station (transceiver) as shown in FIG. In certain embodiments, m = 4. A system similar to that shown in FIG. 1 may be in the prior art, but the system shown in FIG. 1 with a programming or hard-wired element that implements aspects of the invention is not in the prior art. Also, the present invention is by no means limited to the use of PHS air interfaces or TDMA systems, and can be used with any communication receiver equipped with adaptive smart antenna processing means. In FIG. 1, one or more elements of the antenna array 103 are selectively connected to transmit electronics 113 in transmit mode and to receive electronics 121 in receive mode, thus transmitting / The receive (TR) switch 107 is the m-antenna array 103 and the transmit electronics 113 (including one or more transmit signal processors 119 and the m transmitter 120) and the receive electronics 121 (m receiver 122). Is connected to both (including one or more received signal processors 123). Two possible implementations of Switch 107 are as a frequency duplexer in a frequency division duplex (FDD) system and also as a time division duplex (TDD). time division duplex) A time switch in the system. In a preferred embodiment of PHS according to the invention, TDD is used. The transmitter 120 and the receiver 122 can be implemented using analog electronic devices, digital electronic devices, or a combination of the two. The receiver 122 of the preferred embodiment generates a digitized signal to be sent to one or more signal processors 123. Signal processors 119 and 123 are static (always the same), dynamic (changed by the desired direction), or smart (changed by the received signal) and are adaptive in preferred embodiments. Signal processors 119 and 123 are one or more of the same DSP device, or different DSP devices, or different devices for some features and the same device for other features, programmed differently for reception and transmission. Is. [0021] [0021] Figure 1 shows a transceiver that uses the same antenna elements for reception and transmission, but it is possible to have separate antennas for reception and transmission, or just for reception or transmission. It should be noted that it is also clear that both reception and transmission can have adaptive smart antenna processing. [0022] For example, PHS (Personal HandyPhone System) described in Radio Industry Association (ARIB Japan) Tentative Standard, Version 2, RCR STD28, and PHS Memorandum of Understanging Group (PHS MoU, http://www.phsmou. The variant described in the technical standard (see or.jp) is an 8-slot time division multiple access (TDMA) system with true time division multiplexing (TDD). Therefore, the eight time slots are divided into four transmit (TX) time slots and four receive (RX) time slots. This means that the reception frequency is the same as the transmission frequency for any particular channel. Also, assuming that the movement of the subscriber unit between the receive time slot and the transmit time slot is minimal, the downlink (from the base station to the user's remote terminal) and the uplink (from the user's remote terminal to the base station) ) Both interactions, that is, the same propagation path. The frequency band of the PHS system used in the preferred embodiment is 1895 to 1918.1 MHz. The length of each of the eight time slots is 625 microseconds. The PHS system has a dedicated frequency and time slot for the control channel that initiates the call. Once the link is established, the call is passed to the service channel for normal communication. Communication occurs on any channel of 32 kilobits per second (kbps), called full rate. Communication below full rate is also possible, and details of how to modify the embodiments described herein to communication below full rate will be apparent to those skilled in the art. [0023] In a PHS used in a preferred embodiment, a burst is defined as a finite time RF signal transmitted or received over radio waves in a single time slot. A group is defined as a set of 4TX and 4RX time slots. The group always starts from the first TX time slot and its duration is 8 x 0.625 = 5 msec. [0024] PHS systems use π / 4 differential quality (or quad ratcha) phase shift keying (/ 4DQPSK) modulation for baseband signals. The baud rate is 192 kilobaud, which means there are 192,000 symbols per second. [0025] FIG. 2 is a more detailed but still simplified block diagram showing a PHS base station that has adaptive smart antenna processing and implements one embodiment of the present invention. Again, there may be a system in the prior art with an architecture similar to that shown in FIG. 2, but the system shown in FIG. 2 with elements programmed or hard-wired to implement aspects of the invention. Is not in the prior art. In FIG. 2, a plurality of m antennas 103 are used, where m = 4. It is also possible to use fewer or more antenna elements. The output of the antenna is connected to the duplexer switch 107. Switch 107 of this TDD system is a time switch. When receiving, the antenna output is connected to receiver 205 via switch 107 and is analog mixed down from carrier frequency (approximately 1.9GHz) to intermediate frequency (IF) by RF receiver coefficient 205. .. This signal is then digitized (sampled) by an analog-to-digital converter (ADC) 209. It is then digitally downconverted by the digital downconverter 213 to produce a signal sampled to 4x oversampled complex values (common mode I and quadrature Q). Therefore, elements 205, 209, and 213 correspond to receiver 122 in FIG. For each of the m receive time slots, the m down-converted output from the m antenna is sent to digital signal processor (DSP) device 217 (hereafter referred to as the "time slot processor") for further processing. Will be. In a preferred embodiment, a commercial DSP device is used as a timeslot processor, one for each receive timeslot. [0026] Timeslot processor 217 performs several functions, including: That is, received signal power monitoring, frequency offset estimation / correction and timing offset estimation / correction, weights for each antenna element for determining a signal from a particular remote user using the method according to one aspect of the invention. Smart antenna processing including decision, demodulation of decided signal. [0027] The output of the time slot processor 217 is a demodulated data burst for each of the m (= 4) received time slots. This data is sent to the host DSP processor 231. The main function of this processor is to control all the elements of the system and to interface with higher levels of processing. This high-level processing is the processing defined in the PHS communication protocol that deals with which signals are required for communication within all the different control and service communication channels. In a preferred embodiment, the host DSP231 is also a commercial DSP device. In addition, the time slot processor sends the determined receive weight to the host DSP231. [0028] It is shown as block 245 that the RF control means 233 interfaces with the RF system, which also produces some timing signals used by the RF system and modem. RF control means 233 receives its timing parameters and other settings from host DSP231 for each burst. [0029] The transmission control means / modulator 237 receives transmission data from the host DSP 231. The transmission control means uses this data to generate an analog IF output that is transmitted to the RF transmitter (TX) module 245. Specific operations performed by the transmit control means / modulator 237 include conversion of data bits to a complex-valued π / 4DQPSK modulated signal, up-conversion to IF frequency, weighting with a complex-valued transmit weight obtained from the host DSP231. Includes the conversion of signals to analog transmit waveforms transmitted to transmit module 245 using a digital / analog converter (DAC). The transmission module 245 upconverts the signal to the transmission frequency and amplifies the signal. The amplified transmit signal output is coupled to the m-antenna 103 via the duplexer / time switch 107. [0030] Notation Use the following notation. There is an m-antenna element (m = 4 in the preferred embodiment) and after down-conversion, i.e., baseband, and after sampling (4x oversampling in the preferred embodiment), the first, second. The complex numerical response of the m-th antenna element (that is, having in-phase I and orthogonal Q components) is z.₁(t), z₂(t) z_mLet it be (t). In the above notation, t is a count value, although it is not always necessary in the present invention. These m sampling quantities are represented by a single m vector z (t), where the i-th row of z (t) is z._iIt can be represented by (t). For each burst, collect a finite number of samples, eg N, and z₁(t), z₂(t) z_mEach (t) can be represented by an N-row vector, and z (t) can be represented by m by the N matrix Z. Hereinafter, the detailed description of the invention will omit most of the details of incorporating a finite number of samples, but those details will be apparent to those skilled in the art. [0031] Some signals, some, for example N_sIt is assumed that it is transmitted from a remote user of the above to a base station. Specifically, it is assumed that the subscriber unit transmits a signal s (t). For adaptive smart antenna processing, the received signal z is used to extract the estimated value of the transmitted signal s (t).₁(t), z₂(t) z_mContains certain combinations of I and Q values in (t). Such a weight is the i-th element w_riComplex number weight vector w containing_rIt can be represented by the receive weight vector for this particular subscriber unit displayed by. The estimated value of the transmitted signal is then [Number 1]<img file="JP4674030B2_D0001.tif" />In the above formula, w'_riIs w_riComplex conjugate of, w_r^HIs the receive weight vector w_rHermitian transpose (ie transpose and complex conjugate). In embodiments that include space-time processing, each element in the received weight vector is a function of time, so the weight vector is the i-th element w._riw with (t)_rDisplayed as (t). Then, the estimated value of the signal is expressed as follows. [Number 2]<img file="JP4674030B2_D0002.tif" />In the above equation, the operator "*" is a convolution operation. Spatio-temporal processing, for example, combines temporal equalization and spatial processing and is particularly useful for wideband signals. The formation of signal estimates using space-time processing can also be done within the frequency (Fourier transform) domain. S ^ (k) (Note: "^" in this specification means that it is attached to the character before it), Z_i(k), W_i(k) and s ^ (t), z, respectively_i(t), w_riThe frequency domain display with (t) is [Number 3]<img file="JP4674030B2_D0003.tif" />In the above equation, k is the counting frequency value. [0032] For spatiotemporal processing, the convolution operation of formula (2) is usually finite, performed on sampled data, and uses a time domain equalizer with a finite number of equalizer taps to perform spatial processing such as time. Equal to the one combined with the conversion. That is, w_riEach of (t) has a finite number of values of t, and likewise in the frequency domain, w_iEach of (k) has a finite number of k values. Then the convolution function w_riIf the length of (t) is n, the complex number m weight vector w_rInstead of deciding, the column is w_rComplex number m × n matrix w, which is the n value of (t)_rTo determine. [0033] In the rest of the explanation, the complex number received weight vector w_rOr whenever describing that element, the weight matrix w_rAs mentioned above, it should be understood that the decision is generalized to incorporate spatial or spatiotemporal processing. Therefore, spatial processing and space-time processing are referred to herein as adaptive smart antenna processing. [0034] Determining spatial weights The "blind" method for determining weights for adaptive smart antenna processing is a method that does not require training data to be reconstructed. The method according to the invention, like most blind methods, uses knowledge of the original format of the transmitted signal and gives the output signal one or more well-known input signal properties. Its properties can be amplitude characteristics, such as entropy or cycro-stationaryity, correct modulation scheme, or statistical characteristics such as exact replica reconstruction. Such a method is sometimes called "property restoration". [0035] How to have good convergence properties Methods with good convergence properties include partial property restoration methods. That is, a method of replaying one or more properties by determining the bitstream and reconstructing the signal without attempting to replay the exact replica. This type includes methods of preserving signal amplitude (coefficients), entropy, and spectral coherence (eg, periodic stationarity). [0036] The constant modulus (CM) method is a very simple and effective technique that can be applied to signals modulated by a method that results in a constant amplitude signal. These include all forms of phase and frequency modulation, including differential phase shift keying modulation of PHS systems used in preferred embodiments. The CM method is also applicable to non-CM signals, as described below. In the CM method, a) the constant amplitude (CM) property of the signal is restored, and b) the signal that generates the signal to the antenna elements of the array "close" to the actually received signal when transmitted by a remote user. Generate and determine the weight. Amplitude fluctuations are introduced by interference, fading, and timing offsets in these modulation schemes, and the CM properties depend on accurate timing offset correction, including, for example, the DQPSK modulation scheme of preferred embodiments, and CM properties. Holds only at the bow point. In the presence of co-channel interferers, the CM method tends to pick the stronger signal of either the desired signal or the co-channel interferer. If the desired signal strength is greater than 0.5 dB above the strength of any interferer, the CM method correctly selects the stronger signal, i.e. the desired signal. That is, the CM method has very good convergence properties. [0037] The CM method has many variants. These usually minimize the cost function of the general expression. That is, [Number 4]<img file="JP4674030B2_D0004.tif" />In the above equation, E (.) Indicates the statistical expected value operation, and p and q are positive integers, usually 1 or 2. In practice, the statistical operation is some form of sample averaging or accumulation (eg, by summing a set of samples, which in a preferred embodiment is a subset of all the samples in a burst. Will be apparent to those skilled in the art. It is also clear that additional terms are added within the cost function of equation (4), which falls within the scope of the present invention. For example, you can add a term that limits the size of the weight vector. See US Application 08 / 729,390 above for examples of cost functions (not CM cost functions) that include such additional terms. Signal s_ref(t) is a normalized copy signal (called a "reference signal") used in the cost function. That is, the reference signal that determines the weight is the sum of the weights of the received antenna signals that are then normalized. The weight determination determines a set of weights that minimizes the cost function in equation (4). [0038] The CM method is also applicable to non-CM signals. For example, 1988, "application of constant modulus adaptive beamformer to constant and non-constant modulus signals" by J. Lundell and B. Widrow, 1988 Asilomar Conference on Signals, Systems and See Computers (ACSSC-1988), Minutes of pp.432-436. Lundell and Widrow use a cost function similar to formula (4) with p = q = 2 (which is called the 2-2CM method), and both CM and non-CM signals are of the second moment. As long as the ratio of the fourth moment to the square (this ratio is called kurtosis) is less than 2, it is shown that such a 2-2CM method can be used for recovery. For example, M quadrature amplitude modulated signals (M-QAM) are well known to have a kurtosis of approximately 1.4 to 1.2 / (M-1), so the kurtosis of any QAM signal is always less than 1.4. .. Therefore, the CM method is applicable to such signals. [0039] At least one iteration of a particular method, CM method, with good convergence properties is used in the preferred embodiment. In the implementation of the CM method, a value of 1 is used for p and a value of 2 is used for q in the formula (4). When the present invention is applied to non-CM signals and the CM method is used, other values for p and q, such as p = q = 2, can be used. A preferred implementation is also a block-based method. That is, a block of antenna received signals is weighted and the weight is determined using this block of data. A block is a subset of samples in a burst. In particular, it is preferable to use 75 samples of 120 PHS burst symbols. Here, the 75 symbols are in the central payload of the PHS burst. Using the data from the payload, you can conveniently ensure that the data used to calculate the weights for one remote user is not the same as the data for another subscriber unit. There are up to 88 such payload samples in a PHS burst. [0040] This method with p = 1 and q = 2 is called the least squares CM method and includes the following steps. 1. Start the weight vector. For example, w_{r, initial}= Use [100 0]'. Where x'indicates the transpose of x. In the improved embodiment, R corresponds to the maximum singular value of Z._zz= ZZ^HUse the maximum eigenvectors of. Yet another embodiment uses the weight vector from the previous burst. 2. For the sample, execute a copy signal and normalize it. [Number 5]<img file="JP4674030B2_D0005.tif" />3. Weight vector w using least squares procedure_rTo calculate. In other words [Number 6]<img file="JP4674030B2_D0006.tif" />In the above equation, N is the number of samples used in the calculation. The solution of formula (6) is [Number 7]<img file="JP4674030B2_D0007.tif" />Is. In the above formula, R_zz= ZZ^H, [Number 8]<img file="JP4674030B2_D0008.tif" />And N is the number of samples used. 4. Repeat steps 2 and 3 until convergence is reached. [0041] Note that in the calculation of step 3, the overall scale factor is not really important. It is preferred that all scale factors for weights be applied in combination as gains in the system. [0042] Note that the CM method can be extended to space-time processing. One well-known method uses the 2-2CM method and shows that the CM method for spatial and temporal weighting (ie, weight matrix determination) in certain assumptions that usually occur in practice always converges. .. 1996, "A space-time constant modulus algorithm for SDMA" by CB Papadias and A. Paulraj on pages 86-90 of the minutes of the 46th IEEE Vehicle Technology Conference. See systems. However, the method by Papadias et al. Is not based on block data. However, the spatial weighting method can be easily modified to spatiotemporal processing according to the weighting matrix by redisplaying the problem with matrices and vectors of different sizes. Throughout this description, m is the number of antenna elements and N is the number of samples. n is the number of time equalizer taps for each antenna element. (m × N) Each row vector of the N sample of the received signal matrix Z is rewritten as n rows of the shifted version of the first row to generate a received signal matrix Z of size (mn × N) and size. When pre-multiplied by Hermitian transposition of the weight vector of (mn × 1), an estimated received signal row vector of N samples is generated. Therefore, the spatial and temporal problem is re-expressed as the problem of weight vector determination. For the CM method, in formula (7), for example, the weight vector is a "long" weight vector of size (mn x 1), R._ZZIs a matrix of size (mn x mn), r_zsIs a long vector of size (mn × 1). By rearranging the terms, the required (m × n) weight matrix is obtained. [0043] In a preferred embodiment, the sampled data needs to be approximately on-baud in order to retain the CM properties, so if step 2 is performed, tie. is performed and correction timing offset estimation In this case, a sample of the received signal from each antenna element of the antenna array 103 may be oversampled and may include some timing offsets, thus including decimation and interpolation over time. Therefore, the variables t in equations (5) and (7) roughly represent the on-board time for that sample. Obviously, and as described in US Patent Application 09 / 153,110 above, timing offset estimation / correction (which can include decimation / interpolation) precedes the signal copy operation. Performed on the m signal, or after a signal copy operation. [0044] Run the simulation to determine the accuracy level of the timing offset / baud point estimation. FIG. 8 is a diagram showing the result. In the simulation, the signal is not exactly the baud point, but the offset from the ideal baud point due to the timing offset that changes from -1/2 to +1/2 baud in every 1/8 baud step. When sampled, the output SINR obtained using the CM weights is calculated. The results show that the output SINR is reduced by 0.6 dB, even for signals offset by ± 1/8 minutes of 1 baud. This number is specific to the test cases shown, but the conclusion is that the accuracy of timing offset correction for the CM method does not need to be high. Therefore, a simple method can be used for offset correction / decimation / interpolation to generate an approximate baud-aligned sample. [0045] Also note that timing offset correction (including decimation / interpolation) of the oversampled signal is not always necessary for all modulation schemes that have CM properties. For example, DECT (the The Digital European Cordless Telecommunications () standard and the GSM (Global System for Mobile Communications) standard use Gaussian minimum shift keying (GMSK) signals, which always have a CM, and therefore, in these cases, the CM method. However, timing offset correction is not necessary. [0046] The least squares CM weighting method is fairly straightforward to implement. Frequency offset estimation and correction, and demodulation are not required as demodulation does not occur. In one embodiment according to the present invention, when the CM method is carried out, frequency offset correction is performed, although it is not necessary. A further feature of simple property restoration methods such as the CM method is that convergence occurs even at very low SINR values. [0047] The main drawback of methods with good convergence properties, such as the simple property restoration method, is the high number of iterations to converge. In a typical system, the processing power, such as DSP processing power, is very limited, so the convergence when using the CM method is constant, for example, to use the weight vector in a current burst. It may not happen in time. [0048] Rapid convergence method Unlike the partial property restoration method, rapid convergence methods such as decision-based methods converge very quickly. In a decision-based method, the restored property is a perfect replica of the original transmitted signal with the correct modulation scheme. That is, a signal copy operation such as equation (1) estimates the received signal, demodulates the signal, and builds a reference signal with the correct bit stream. For it to work well, it is necessary to correct the frequency and timing offset when constructing the reference signal. The correct weight is a weight that produces a reference signal that is close to the transmitted signal. This method makes one or more iterations to get the "optimal" weight. Timing offset correction (including any decimation) and frequency offset correction are shown below as occurring after the signal copy operation, but do so once or multiple times before the signal copy operation. It is clear that it can be done. See co-owned U.S. Patent Applications 08 / 729,390 and 09 / 153,110 above for examples of these operations that occur before and after signal copying, as well as a detailed description of the decision-based method. When using the least squares criterion, the general method involves the following steps: [0049] 1. Start the weight vector. For example, w_{r, initial}= Use [100 0]'. Where x'indicates the transpose of x. In the improved embodiment, the R corresponding to the maximum singular value_zz= ZZ^HUse the singular vector of. Yet another embodiment uses the weight vector from the previous burst. As described below, one aspect of the invention comprises using a decision-based method after using the partial property method. In such cases, the last obtained weight vector (ie, using the partial property restoration method) is used when implementing any of the embodiments according to the invention. [0050] 2. Perform signal copy. [Number 9]<img file="JP4674030B2_D0009.tif" />If the sample was originally oversampled, it is followed by decimation / interpolation (in alternative forms, decimation / interpolation can occur before the copy signal operation). [0051] 3. Estimate the timing and frequency offset to generate a signal with the correct timing and frequency offset. [0052] 4. Reference signal s by symbol determination (ie demodulation)_refDetermine (t) and s_refMake sure that (t) has the same modulation scheme as the correct bitstream and the same timing and frequency offset as the signal sent from a particular user to the receiver. [0053] 5.w_rCalculate the weight vector by least squares minimization greater than. In other words [Number 10]<img file="JP4674030B2_D0010.tif" />There is a solution to this, [Number 11]<img file="JP4674030B2_D0011.tif" />In the above formula, R_zz= ZZ^H,and [Number 12]<img file="JP4674030B2_D0012.tif" />[0054] 6. Repeat steps 2, 3, 4 and 5 until convergence is reached. [0055] Steps 2, 3 and 4 correct the frequency and timing offset of the signal so that the correct demodulation decision is made in step 4, and step 5 generally reintroduces the correct frequency and timing offset. Note that the reference and copy signals of the cost function need to have the same timing and frequency offset. Also, as described above for CM and as described in US Application No. 08 / 729,390 above, the minimization of equation (9) imposes constraints on the norm of the weight vector weight vector term weights. Also note that other terms such as norms are also included. See also U.S. Patent Applications 08 / 729,390 and 09/153, 110 co-owned above for a detailed description of how the reference signal (step (4)) is determined. This will be described below with reference to FIG. [0056] A method of following the determination is to determine the weight matrix for spatiotemporal processing, for example by rearranging the terms as described herein for the CM method, and by other methods apparent to those skilled in the art. Note that it can be easily extended. Therefore, the present invention also includes a method of determining a weight vector and a weight matrix for space-time processing. [0057] Therefore, the decision-based method reproduces an exact replica of the signal that is supposed to have been sent to the receiver, and the partial property restoration method reproduces one or more simple properties, such as the correct amplitude. The decision-based system works very well and converges with very few iterations in a reasonably high SINR environment. However, these methods are sensitive to the initial state and may not converge if the initial SINR is low. This situation is common in highly fluid cellular systems and other systems exhibiting fading. [0058] [0058] It should be noted that iterating over the CM method is generally less computationally expensive than iterating over the decision-based method and does not require frequency offset correction or demodulation. [0059] Preferred method: single user One aspect of the present invention is the number of iterations N of an iterative weighting method with good convergence properties, such as a partial property restoration method, preferably the CM method.₁The second number of iterations of the rapid convergence test, such as a decision-based method N₂A weighting method that involves combining both and in this order to obtain the benefits of a good convergence property with rapid convergence. N₁CM iteration is a decision-based method of N₂It provides a starting condition for iteration in areas where rapid convergence of decision-based methods is most certain. In a preferred embodiment, one iteration of the decision-based method (N)₂Using = 1), in the alternative form, 2 iterations (N)₂= 2) is used. This method iterates over the iterative weighting method with good convergence tests until it meets the switching criteria, then starts with the weights obtained with the method with good convergence properties and repeats several iterations of the rapid convergence test. It can be rephrased as what you do. In some cases, the switching criterion is the explicit definition number of iterations N₁Is. In another preferred embodiment, N₁Is not explicitly specified. Rather, the switching criterion is the SINR threshold of the copy signal, and switching to a decision-based method occurs when the SINR estimate is equal to or greater than the threshold. With this method, N₂Sufficient number of CM iterations N to obtain sufficient SINR to ensure convergence of the decision-based method with only the iterations of₁Is used. [0060] Many methods can be used to determine SINR estimates. In a preferred embodiment, the method of use is described in U.S. Patent Application No. 09 / 020,049 by Yun, "POWER CONTROL WITH SIGNAL QUALITY ESTIMATION FOR SMART ANTENNA COMMUNICATION SYSTEMS" (filed February 6, 1998). The implementation form of the estimation method is described below. [0061] The number of burst samples used for the estimate is indicated by N. The sampled coefficient information is first extracted by forming the sum of the in-phase squares and the quadrature phase signals (the real and imaginary parts of the signal s (t)). The average power and mean square power are then used to determine the average power and mean square power using the average number of samples for the expected operation. [Number 13]<img file="JP4674030B2_D0013.tif" />[Number 14]<img file="JP4674030B2_D0014.tif" />[0062] Once instantaneous power R²(t) = I²(t) + Q²Once (t) is determined, the squared power R⁴(t) = [R²(t)]²Only a single additional multiplication is required per sample to determine, preferably [Number 15]<img file="JP4674030B2_D0015.tif" />The estimated SINR is determined by at most one square root operation using. Percentage, [Number 16]<img file="JP4674030B2_D0016.tif" />And both quantity A are sometimes called kurtosis. This preferred method of signal quality estimation is not sensitive to frequency offsets and is therefore particularly attractive for use with the CM method, which is also not sensitive to frequency offsets. [0063] Other methods of determining the quality of a post-copy operation signal can also be used in alternative embodiments according to the invention. [0064] A weighting method for a single user is shown in the flowchart of FIG. The initial weight vector is formed at 303. This is [100 ... 0]', or in improved embodiments, the R corresponding to the maximum singular value._zz= ZZ^HSingularity vector of is used. In yet another embodiment, the weight vector from the previous burst is used. Here, copy operation 305 is performed according to equation (1), but in a preferred embodiment only the central portion of the burst is used, preferably only 75 symbols (300 samples) from the payload portion in the center of the burst. use. Correct the output for timing offset 307. Any timing offset correction method can be used. As mentioned herein, the timing offset correction need not be so accurate. The preferred method is as described in US Patent Application No. 09 / 153,110 above. Copy operations and timing offset correction operations can be combined. Specific to the timing offset correction operation, but not explicitly shown in Figure 3, is the required decimation and interpolation, so after step 307, the data will be approximately 75 from the center of the current burst. The symbol's baud point contains 75 complex number (I and Q) samples. The SINR of the copy signal is estimated to be 309, preferably using the kurtosis, as described herein above. At step 311 it is determined if the SINR exceeds the threshold SNR. If not, the CM method iteration is performed in step 313 using the least squares cost function criterion as described in equations (6) and (7) above. This method then returns to the copy operation of step 305 for another iteration. On the other hand, if it is determined in step 311 that the SINR threshold is exceeded, then in steps 315 and 317 N of the decision-based method, including frequency offset correction 315.₂The iteration is performed. Any frequency offset correction method can be used, the preferred method of which is described in U.S. Patent Application No. 09 / 153,110 described above. Similarly, any method can be used for decision-based adaptation involving the generation of a reference signal, and in a preferred embodiment, the method described in U.S. Patent Application No. 09 / 153,110 described above is used. ing. In a preferred embodiment, once the weights are determined, only a subset of the samples within each burst is used. Therefore, the last determined weight vector is used in copy operation and demodulation step 318 over all bursts. In this embodiment, preferably, step 318 includes timing and frequency offset determination and correction, demodulation, using the architecture described below with reference to FIG. The output of the adaptation based on the decision is signal 319. [0065] A preferred embodiment for reference signal generation that is part of decision-based adaptation step 317 (using some of the burst data) and used to demolish all burst data in step 318 is reference. Forming the phase of the reference signal at the sample point by using a signal generation architecture and relaxing the phase of the signal that ideally travels from the previous reference signal sample towards the phase of the copy signal at the same sample point. It is preferable to use a method including a tracking mechanism for each sample. The copy signal is formed from the receiving antenna signal. The reference signal is constructed at each sample point by constructing an ideal signal sample from the copy signal at the same sample point, and the ideal signal sample is determined from the copy signal at the sample point. Having a phase, the phase of the ideal signal sample at the initial symbol point is set to the initial ideal signal phase, relaxing the phase of the ideal signal sample towards the copy signal sample phase. Generate the phase of the reference signal. The phase of an ideal signal is determined from the phase of the reference signal at the previous sample point to determine that phase, and from the determination based on the copy signal. In one implementation, the reference signal is determined in the forward time direction, and in the other implementation, the reference signal sample is determined in the reverse time direction. In some versions, copy signal b_NThe step of relaxing the phase of the ideal signal sample towards the phase of (n) corresponds to adding a filtered version of the difference between the copy signal phase and the ideal signal phase. In other versions, the step of relaxing the phase of the ideal signal sample towards the phase of the copy signal is to add a filtered version of the difference between the copy signal and the ideal signal to the ideal signal sample. Corresponds to forming a reference signal sample. [0066] As an example, 4DQPSK with reference to Figure 4. We will explain in detail using PHS signals. Modifications to other modulation schemes will be apparent to those skilled in the art. The phase detector unit 403 detects a phase difference 405 between the copy signal 401 (corrected for timing and frequency offset) and the previous reference signal 417. The phase difference signal 405 is sent to the slicer 407 to generate a determined phase difference 419. The correct phase difference for / 4DQPSK is (2i-1) / 4, i = 1, 2, 3 or 4, which is the phase difference between the previous reference signal sample and the ideal signal. This is subtracted from the actual phase difference 405 at block 409 to generate the error signal 411. This error signal is filtered by the filter 413 to generate the filtered error signal 415. This is the filtered error signal used to adjust the phase difference 419 to be close to the actual phase difference 405. The corrected phase difference 421 is then used within the frequency synthesizer / phase accumulator 423 to generate the reference signal 429. This is the sample value 417 before the reference signal 429 used by the phase detector 403, and therefore a unit time delay 425 is shown between these signals. Signal 430 (N₂The symbol of the iterative signal 319) is determined in block 427. Mathematically, b_RIf (t) displays the reference signal complex sample value at bow point t and displays the phase, the input to the phase accumulator 423, b_R(t) -b_R(t-1) ,, filter [d_ideal(n)-decide [d_ideal(n)]] + decide [d_ideal(n)] In the above formula, decide {d_ideal(n)} is the output of slicer 407, equal to (2i-1) / 4, and i = 1, 2, 3 or 4 for / 4 DQPSK. Here, the "ideal" complex number sample point b_ideal(t) is defined as follows. b_ideal(0) = b_R(0) = b (0) In the above equation, b (t) is a sample of the input signal 401, d_ideal(t) is the phase difference between the current input sample and the previous reference signal sample. [Number 17]<img file="JP4674030B2_D0017.tif" />In the above equation, * represents the complex conjugate. An "ideal" signal is a reference signal with a phase advanced by an ideal amount, the amount of which is d._idealIt depends on the decision according to (t). In other words b_ideal(t) = b_R(t-1) + (2i-1) / 4, i = 1, 2, 3, or 4 To get the reference signal, b_idealThe phase of (t) is the quantity [b (t) -b_ideal(t)], (b (t) and b_ideal(T) phase error) is filtered, and the filtered quantity is b_idealBy adding to the phase of (t), it relaxes toward the phase of b (t). In an alternative embodiment, the quantity (b (t) -b) rather than the phase error_ideal(t)) is filtered. The filter is preferably a constant of proportionality. Higher-order filters may be required. Mathematically, in one embodiment, b_R(t) = b_ideal(t) + filter [b (t) -b_ideal(t)] And in other embodiments, the architecture of FIG. b_R(t) = b_ideal(t) + filter [b (t) -b_ideal(t)] May be modified slightly due to the use of. [0067] In a preferred embodiment including a tracking reference signal generator, the method shown in the flowchart of FIG. 3 is performed as a set of instructions for the time slot processor 217, which is a signal processor (DSP) device. [0068] [0068] The system described the simulation of the method shown in Fig. 3 is performed, and the initial weight vector is R corresponding to the maximum eigenvalue._zz= ZZ^HAs an eigenvector of. A simulation was performed on the PHS base station shown in Fig. 2, which has four antenna elements. The input signal of each antenna had a signal-to-noise ratio (SNR) of 11.9 dB. The input signal-to-interference ratio (CIR) is 1.1 dB, which corresponds to the initial copy signal SINR of 0.8 dB. A typical PHS burst has 120 symbols, but only the central 75 symbols were used in the weight calculation. All calculations were done offline using a MATLAB environment (The Mathworks, Inc., Natick, MA). The results are shown in Figures 5A, 5B, 5C, where the determination-based method, the least-squares CM method, and the mixing method according to the invention under low SINR conditions (in this case, with the initial weight vector). The convergence characteristics of the post-copy SINR) of about 0.8 dB after the first copy operation are compared. The output SINR (dB) after each iteration is measured and plotted using the SINR estimation method. The first SINR value shown is the initial weight vector (R)_zzThis is the time when SINR was estimated after performing a copy operation with (the first singular vector) of. This is the same for all three methods. As shown in Figure 5A, the decision-based method does not converge even after 10 iterations. Figure 5B shows that the CM method converges slowly and the output (estimated) SINR steadily rises as the iterations progress. The optimum SINR is 18 dB, and the CM method repeats 10 times or more until it converges to this optimum value. FIG. 5C is a diagram showing a method according to the invention operating at a switching output SINR threshold of 7.5 dB. Note that the results start similar to those shown in Figure 5B, but then diverge after switching to a decision-based method (results shown in Figure 5B are shown by the dashed lines after switching). After the decision-based method has started, this method reaches the optimum SINR by repeating the decision-based method only twice, but already the optimum value when the decision-based method is repeated only once. Very close to. Overall, the method shown in the flowchart of FIG. 3 converges within 5 iterations, whereas when only the CM method is used, it repeats more than 10 iterations. [0069] Multiport architecture Within the same traditional channel, both inside and outside the cell, some, for example N_sIf there are subscriber units (ie, co-channel users), a preferred embodiment according to the invention uses a multiport architecture. Each "port" forms a separate copy signal, N_sTrack only one of the subscriber units. Therefore, the subscriber unit on any port is the other N_sA co-channel interferer for the subscriber unit and its corresponding port. Only co-channel users with signal components received by the antenna element above a certain noise lower limit are subject to such tracking. The number of such users can be estimated. If there is any burst (matrix Z), then R_zz= ZZ^HYou can perform order estimation by examining the eigenvalues of. Any order estimation method can be used. For example, Rissanen's MDL (Rissanen minimum descriptive length) standard or Akaike's information theory (Akaike information theoretic) standard is well known. Research on technologies for determining the number of effective co-channel users, January 1996, Virginia Polytechnic Institute, Bradley Department of Electrical engineering, Mobile and Mobile Radio Research Section 3.8 of "Direction of arrival estimation using antenna arrays" by Mobile and Portable Radio Research Group, Technical Report MPRG-TR-96-03, Rias Muhamed and TSRappaport (also Rias) See Muhamed, 24061, Virginia, USA, Blacksburg, Virginia Polytechnic Institute and State University, Bradley School of Electrical Engineering, Master's Thesis, Direction of arrival estimation using antenna arrays. I want to be. In a preferred embodiment, the MDL standard is used. [0070] Preferred embodiments include estimation of all "effective" co-channel users, followed by tracking. In other embodiments, a "good" wireless design environment is envisioned. That is, co-channel users who are not communicating with the same base station are assumed to be far away, so only valid co-channel users are users who share the same conventional channel and communicate with the base station. That is, a subscriber unit that is a different spatial channel within a conventional channel. In such a case, N_sIs well known. [0071] For example, two well-known co-channel users (ie, N) in the environment (because they are estimated or aware)._sConsider the case where = 2) exists. When tracking one of the subscriber units, the other subscriber units are interferers. Therefore, in this architecture, the desired signal and the effective interferers communicating within the same conventional communication channel are tracked at the same time. In the fading environment encountered, for example, if the subscriber unit moves rapidly, the CIR is very low and the instantaneous CIR fluctuates over a wide range. Thus, at any given time, at any given burst, the desired signal is weaker than any of its interferers, and the port of the desired signal can be locked by the interfering waves. That is, this may initiate tracking of the interferer from the desired remote user. [0072] FIG. 6 is a block diagram showing a multi-port adaptive smart antenna processing device of a preferred embodiment. In each port, the oversampled output 605 of the receiver 122 from the antenna element 103 has the initial weight vectors 631-i, i = 1 ... N_sIs first used and combined within signal copy operation 607. These initial weights are provided by the weight initializer 621. The resulting copy signal is the timing offset corrected by the timing offset corrector unit 609. This unit decimates / interpolates to produce a set of nearly baud-aligned samples (for iterations of the CM method) or substantially baud-aligned samples (for iterations of the decision-based method). The baud-aligned sample is sent to the SINR estimator 613 and its output is sent to the weight calculator and decoder 615. Here, the baud-point aligned sample and / or antenna signal 605 is used to determine the reference signal and pair of weights according to the method according to the invention described herein. Its output is demodulated signal 617, as at least one iteration of the decision-based method is used within the weight calculator and decoder 615. With this method, N_sN for subscriber units_sThe demodulated signal is determined. The multiple ports allow both the desired subscriber unit transmit signal to be tracked and either co-channel interferer to be tracked. The adaptive method described further below has the ability to switch between the desired user and the interferer in a fading environment. Therefore, N_sBy using the port, N_sIf users are tracked at the same time and one user's signal jumps from one port to another, this can happen in a fading environment, but the output of the port is categorized by the user classifier 623, Isolate the desired user from the interference wave, and N_sThe demodulated signal 625 is output correctly. [0073] Other multiport architectures are known but are not used in the adaptive methods described herein. For example, in 1989, IEEE, New York, BG Agee, "Blind separation and capture of communication signals using a multitarget constant modulus beamformer", 1989 IEEE Military Communications Conference. ("MILCOM 89 ), vol.2, pp.340 ~ 346. Agee's method differs from the method described herein in several respects, for example: 1) weighting start method, 2) what kind of calculation is performed on each port. Agee's method jointly orthogonals all weight vectors at each stage of the iteration, which is computationally intensive, while in a preferred embodiment according to the invention, each port is jointly initiated. Later, each port can independently adapt to 3) different methods of determining the weight vector. Note that in alternative embodiments, the weight vectors for each port are orthogonal, as it is expected that computational power will be more easily available in the future. [0074] Preferred weighting method: multi-user FIG. 7 is a flowchart showing a preferred method for determining the weight and output signal 625. First of all, R_ZZPerform a start copy using the eigenvectors of the matrix. In step 703, the eigenvector 631-1 corresponding to the maximum singular value, the port labeled "# 1", the next eigenvector 631-2, the second port ... N_sSecond eigenvector 631-N_sIn N_sStart the port. These eigenvectors are guaranteed to be linearly independent and are generally also the first preferred values. Alternatively, either substantially independent initial weight vector can be used. For example, in an alternative embodiment, port # 1 starts with a vector [100 ... 0]', port # 2 starts with a vector [0100 ... 0]', and so on. At each port, after the start, the method proceeds at each port as opposed to the single user case shown in the flowchart of FIG. That is, step 305 is the first copy operation performed by default. The resulting signal is a corrected timing offset (including decimation / interpolation if first oversampled) in step 307, producing a substantially baud-aligned sample, which is a signal quality estimate. It is sent to the vessel and in step 309 estimates the SINR substantially at the bow point. In step 311 it is decided whether to choose the CM method or the decision-based method for weight adaptation. If the SINR is lower than the predefined SINR threshold, optimization based on the partial property restoration method (preferably the CM method) is performed in step 313, which returns to step 305 for another iteration and for this port. Start with the last determined weight vector. If the SINR is higher than the threshold, then frequency offset correction is performed in step 315 and a decision-based adaptive iteration is performed once in step 317. If the timing offset correction in step 307 is only an approximation, a more accurate correction to the decision-based method will be needed, but such corrections will be apparent to those skilled in the art. In a preferred embodiment, the decision-based iteration is performed only once. Alternatively, the decision-based iteration may be performed more than once. Determining the weights, in the preferred embodiment, using only a subset of the samples within each burst, the final determined weight vector for each port spans the entire burst. Used in copy operations and demodulation steps. In a preferred embodiment, this copy operation includes timing and frequency offset determination and correction, demodulation, preferably demodulation using the architecture shown in FIG. The result for each port is the demodulated signal 617. [0075] Note that one feature of the invention in both single-user and multi-user cases uses the weights obtained using the current burst data to determine the signal for the current burst. Using the weight vector from the previous burst may not give good results if the subscriber unit is moving around and there are other fading and low SINR environments. [0076] The final step is the classification of these outputs to determine if any of the output ports are locked by the interfering waves. A PHS burst, for example, a traffic channel burst, contains a field for the payload, a unique word (UW) known to all subscriber units, and an error detection cycle redundancy check (CRC) field. Other protocols include slightly different fields, such as those used to determine if a particular message is from a particular subscriber unit or to a particular base station. In a preferred embodiment, the idea for determining interfering locking is from a desired subscriber unit that is transmitting a valid waveform to the system, from an interfering subscriber unit that is also transmitting a valid waveform to the system. To distinguish. That is, there is a means for defining a "valid" subscriber unit waveform, eg, a key-scrambled waveform with the required data and modulation format for that subscriber unit. The interfering unit also includes a means of defining a waveform that has its own effectiveness, eg, the required data and modulation format for its subscriber unit, and is scrambled with a different specific key. Is done. In a preferred PHS implementation, one method for detecting interferer locking includes both unique word (UW) and CRC monitoring. It is specified that the data bits in each burst are scrambled with a bit pattern generated using the lower 9-bit cell station identification code (CSID). This 9-bit word used to encrypt both the burst payload and the associated CRC is called the scrambling key. When designing a communication system, for example a cellular system, it is desirable to ensure that each proximity station (base station) has a different scrambling key. Note that in the PHS specification, a base station or communication station is called a cell station. [0077] It is well known that the interferer locking detection method based on the preferred embodiment includes the following steps. [0078] For a particular port, that is, a particular subscriber unit 1 Demodulate the received signal using the receive weight determined for the subscriber unit. It then reverse-scrambles the burst payload using the CSID-based key for that subscriber unit. [0079] 2 Compare the received CRC with the CRC calculated from the demodulated and inversely scrambled bit sequence of the burst payload. [0080] [0080] 3 Show that the transmission error or the key is incorrect, determine if both are significantly different, while the UW indicates no error. And if the conditions are met, the counter is triggered. If the weight "tracking" is included and the conditions are not met, the communication is considered unlocked to the interfering wave and is used to receive from the subscriber unit (or subscriber unit spatial signature). The weight is stored ("tracked") as a "good" value for that subscriber unit. [0081] 4 If a certain number of consecutive bursts meet the conditions described in step 3, it is determined that the port has been determined to be locked by the interferer, as determined by the counter. [0082] The current PHS specification uses unique words and CRC to determine that interference locking does not work if the co-channel user is all different spatial channels within the same traditional channel. This is because the CSID is the same for all subscriber units on the same traditional channel. Interferer locking determinations when a co-channel user is a spatial channel of the same traditional channel can be made by maintaining a spatial signature history for the co-channel user. [0083] After classifying the outputs, the result is a set of output signals from each port. [0084] apparatus FIG. 9 is a block diagram showing an apparatus that implements one aspect of the present invention. The device that determines the weight vector for receiving a particular signal transmitted by a particular subscriber unit minimizes the starting means 902 to start with the first initial vector value and the first cost function. It includes a first iterative means 905 for iteratively modifying the weight vector according to the first iterative method. The first iterative method is an iterative weighting method with good convergence properties, preferably a CM method implemented as described herein. The device also includes a second iterative means 907 for iteratively modifying the weight vector according to a second adaptive method that minimizes the second cost function. The second adaptive method is a rapidly converging iterative weighting method, preferably a method based on the decisions described herein. The initializer 902, the first repeating means 905, and the second repeating means 907 are under the control of the controlling means 911. The control means starts with the first initial vector value supplied by the starting means 902 until the final weight vector after the final iteration of the first adaptive method meets the switching criterion, which is the second vector value. It is programmed to drive the first iterative means 905 and the second iterative means 907 starting from the second vector value to determine the weight vector 909. When the weight vector 909 is used by the spatial processor and demodulator 915 and is under the control of control means 911, it produces a demodulated signal and the spatial processor receives the signal on the antenna array 103 via the receiver 122. use. Each iteration method involves determining the copy signal. The device preferably comprises a SINR estimator 913 that estimates the posterior copy SINR in the first iterative means copy signal using the weight vector determined by the first iterative means 905, and switching criteria are preferred. Is a SINR estimate that exceeds the SINR threshold. [0085] The weighting device preferably has at least one digital signal processor (DSP) device at the base station, with elements, 902, 905, 907, 909, 911, 913, 915, preferably one or more. It is executed as a program in the DSP. As will be appreciated by those skilled in the art, experts will be able to create various modifications of the methods and devices described above without departing from the spirit and scope of the invention. .. For example, the communication station that implemented this method can use one of many protocols. In addition, the architecture of these stations is also possible. Many variants are also possible. The true purpose and scope of the present invention shall be limited to the contents described in the following claims. [Simple explanation of drawings] FIG. 1 is a functional block diagram showing a multi-antenna transceiver system that may include a receive weight determiner according to aspects of the invention. FIG. 2 is a detailed block diagram showing a transceiver including a signal processor that executes a receive weight determiner when traveling a set of instructions according to aspects of the invention. FIG. 3 is a flowchart showing an embodiment of a weight determination method according to the present invention. FIG. 4 is a block diagram showing a tracking reference signal generator and decoder used in a preferred embodiment of the present invention. FIG. 5 is a diagram showing the performance of executing the CM method, the determination-based method, and the mixing method, respectively, according to the aspect of the present invention. FIG. 6 is a block diagram showing a multiport weight determiner and a spatial processor according to a preferred embodiment of the present invention. FIG. 7 is a flowchart showing a preferred embodiment of the multi-user weight determination method according to the present invention. FIG. 8 is a diagram showing the effect of timing offset on the performance of the CM method. FIG. 9 is a block diagram showing an apparatus for executing one aspect of the present invention.