GB2506711A

GB2506711A - An adaptive beamformer which uses signal envelopes to correct steering

Info

Publication number: GB2506711A
Application number: GB201308494A
Authority: GB
Inventors: John Edward Hudson
Original assignee: Individual
Current assignee: Individual
Priority date: 2012-10-02
Filing date: 2013-05-13
Publication date: 2014-04-09
Also published as: GB201308494D0; GB201217576D0; GB201219590D0

Abstract

This application addresses the problem of performance degradation caused by steering errors in a steered least squares adaptive beamformer 101 of a receive system. The least squaes beamformer may take the form of a linearly-constrained minimum output power (LCMP) beamformer implemented as a generalised sidelobe canceller (GSLC) using the Griffiths-Jim algorithm. The invention monitors the covariance 6 between the envelope of the signal, d, output from the least squares beamformer and the envelope of the signal at an internal point in the least squares algorithm. Preferably the internal point is at the auxiliary inputs Y for controlling filter weights within the least squares algorithm. Based on the covariance estimates, steering corrections 7 are applied to a beam steering vector, S0. The steering corrections also take account of the statistical dependence of the signals themselves. The invention has applications in the fields of wireless signal transmission, radar, optical systems, speech and sonar.

Description

A METHOD AND SYSTEM OF ADAPTIVE BEAMFORMING

FIELD OF THE INVENTION

[001] The present invention relates to communications in a broad sense, including wireless signal transmission, radar, optical systems, speech and sonar. The present invention particularly relates to a receive system including an adaptive beamforming network and a method of operation thereof to reduce signal degradation due to steering errors.

BACKGROUND OF THE INVENTION

[002] Adaptive beamforming applications in such areas as speech enhancement, sonar, radar, and radio communications operate by using a sensor array with variably weighted, or frequency-filtered, elements to form a directional beam in the direction, assumed known, of a desired signal source while simultaneously detecting, and forming nulls in the unknown directions of, interference sources. Descriptions of such techniques are widely available in the literature [1]. The most-accepted mathematical solution for the sensor element weights is "linearly-constrained minimum output power" (LCMP), which fixes the said array's gain or frequency response in the desired signal direction while minimising its overall output power by varying said weights, the assumption being that interference output power will be preferentially minimised since desired signal output power has been fixed.

[003] The LCMP solution can be implemented, among other choices of configuration, as a generalised sidelobe canceller (GSLC) circuit arrangement using the Griffiths-Jim algorithm defined in [2] and shown at the left side of figure 1, often in conditions such that the array element outputs can be digitised directly and the signal processing is correspondingly fully digital. Bearing a superficial resemblance to a Wiener filter, the LCMP system is often recommended as being the optimum solution for maximising output signal to noise ratio and indeed in desired-signal-absent conditions this potential is more or less achievable, as was analysed by Reed et. al. [3]. p)

[004] However, in signal-present conditions the performance actually achieved by a LCMP circuit arrangement is generally only a fraction of the theoretical limit, a number of factors combining to cause this suboptimum state of affairs. The immediate problem is that under high SNR conditions LCMP solutions becomes excessively sensitive to steering errors in respect of a desired signal as discussed for example in [4]. Generally the physical characteristics, directional patterns and propagation environment of the receiving array elements are imperfectly known and in any case not to the accuracy necessary for optimal results in least squares signal processing.

[005] In the light of this drawback, the LCMP circuit arrangement has been mostly applied in niche applications where it is known that there are only vanishingly weak or transient desired signal sources. One such application is pulsed radar in which the average target signal strength is small whereas applications in areas such as speech enhancement, typically with a strong desired signal source, have been much less successful.

[006] It is known to experts in the field that the aforesaid difficulties associated with strong desired signals in LCMP can be alleviated by moving away from using linear least squares methods toward more complex signal separation techniques which do not rely entirely on spatial direction. Algorithms such as Independent Component Analysis (ICA) and Higher Order Statistics (HOS) [5] have been developed that explicitly exploit the assumed non-Gaussian nature of many types of signal and these alternative algorithms achieve source separation by achieving statistical independence between multiple beamformer outputs whereas the LCMP solution only achieves decorrelation, i.e. sets to zero the mean product of its beam outputs.

[007] In Leung [09] (Channel estimation in a CDMA wireless communication system), it is noted that a concept described uses post-demodulation data feedback techniques to enhance the accuracy of channel estimation procedures following a demodulation of a digital signal using a known pilot signal. This is a distinct pilot enhancement concept which does not have a high interference rejection capability and relies on prior knowledge both of a pilot signal and the coding parameters of a friendly transmitter. There has developed, accordingly, a need to provide a receiving apparatus which can operate in conditions where transmission parameters are not known.

OBJECT OF THE INVENTION

[009] The present invention seeks to provide an improved least squares beamforming algorithm, method and system, in a broad sense, and particularly seeks to address aspects of performance degradation, especially in a steered least squares adaptive beamformer. The present invention specifically seeks to provide an improved least squares algorithm to address steering error corrections in a telecommunications, radar, acoustic or optical system.

SUMMARY OF THE INVENTION

[010] In accordance with a general aspect of the invention, there exists one or more transmission sources in at least one propagation medium and there is provided one or more multi-element arrays of receiving sensors and a receive-signal processor; wherein the signal transmitter is operable to receive information from at least one source. The target source may be non-cooperative in relation to the receiver so its pilot and training sequences, which normally facilitate reception, are assumed to be unknown. The receiving sensors and receive-signal processor are operable to perform linear filtering operations on the target source signal to maximise signal to noise plus interlerence ratio.

[011] In accordance with a second aspect of the invention, there is provided a receive-signal processor apparatus for forming an optimised least-squares beamformer for receipt of target source signals, the apparatus comprising a multi-element sensor array, a digital sampling arrangement operable to record received signals, a signal processing unit programmed to form a least squares adaptive beam and to determine and monitor signal envelopes and/or statistical dependence of signals at intermediate test points. The test points are conveniently selected such that, when the beamformer is correctly steered with respect to a desired signal, the chosen envelope correlations or statistical dependencies are essentially zero. The statistical dependence metrics are more generalised than the simple cross correlation typically used in the least squares method, the latter known to be frequently inadequate in achieving the goal of source separation [012] The receive-signal processor apparatus can further comprise a sampling unit, programmed with a steering optimisation algorithm and is operable to sample the signal envelopes or statistical dependencies at said intermediate test points and output beamformer control signals to drive the statistical dependence between the envelope or the actual signal waveforms of the least squares beamformer output and the selected test points to a minimal value.

[013] Tn accordance with a third aspect of the invention, there is provided a method to monitor and optimise the steering of a constrained adaptive least squares beamformer, said beamformer comprising a receiving array of sensors arranged to provide input signals to an adaptive signal processing algorithm which is steered to output an estimate of a desired source's waveform, the method comprising the steps: forming a steered beam pointing to a prior estimate of a direction of said source and forming a number of minima in the directional pattern of said steered beam which align to the directions of interference sources whose presence or plurality is not known a-priori; determining the correlations of the envelope fading at the beamformer output with the envelope fading at one or more selected internal points within said adaptive algorithm; Generating corrections for the said beam steering vector, whereby to align the vector to minimise said envelope correlations. By sampling the actual signal waveforms at the said output and one or more selected test points, the statistical dependence calculated using known signal processing algorithms [5] is driven to a minimum value by correcting the said beam steering vector.

[014] The internal test points can be located in such a manner that, when the beam steering vector has been adjusted to correspond with the actual desired signal vector, the desired signal is essentially eliminated from said test points.

[015] A Griffiths-Jim configuration conveniently enables these conditions to be achieved and can be used to implement the least squares beamforming and can provide main input and at least one auxiliary input into the least squares adaptive process. The envelope-sensing or statistical test points can be provided by the auxiliary signal lines of said configuration. Moreover, a "sphering" operation, well known to signal separation experts [5], can be optionally applied to the said auxiliary signal lines to equalize the levels of different principle components and boost a weak desired signal to a usable level when interference is strong.

[016] In a first option, where the determined signal envelopes are of interest, the Griffiths-Jim configuration can be sampled at said test points, where the envelopes are further filtered to remove unwanted carrier components. A matrix of numerical values, indicating the statistical dependence between the envelopes at the said test points and said least squares signal output, can be determined by averaging with a window over some time period. The dependency measure can comprise a simple correlation matrix of a power law function of the fading envelopes; a single-valued performance metric can be determined from the dependence matrix which is an increasing function of the degree of departure of the least squares beamformer steering away from the true desired signal direction.

[017] Alternatively, in a second option, the statistical dependence of the original sampled waveforms at said test points can be estimated by standard estimation techniques such as higher order statistics (HOS) measures [5] and mutual information measures [6], [7].

[018] These two options are not mutually exclusive: statistical dependence can be used in place of or can augment envelope correlation measures and assist in driving steering errors to zero. Statistical measures of the kurtosis [5] (i.e. sidelobe levels of the probability distribution) of the desired signal estimate can be used to estimate a degree of signal purity in the case of digital communication systems.

[019] The optimisation techniques can be further applied to correction of the steering of a wide-band least-squares beamformer applied to propagating carrier signals and dispersive channels, whereby the sensor weights are frequency filters according to techniques known in the state of the art in wideband adaptive radio communication and acoustic receiving arrays.

[020] The step of optimising the least squares beamformer steering can be applied to acoustic, radio, millimetric, terahertz, infrared, optical, or other electromagnetic propagating scenarios which contain fluctuating information sources or possess non-Gaussian waveforms where statistical dependence can be measured. Further there may be simultaneous propagation in different channels or different media.

[021] A variable linear beamformer can be provided to operate with regard to a vector of sensor signals, said matrix producing at least one beamformer output signal which is an approximation to the waveforms of the one or more distinct signal sources illuminating said sensor array from said different directions; the method further comprising the use of algorithms which determine the envelope amplitudes of the signals at selected test points or sample waveforms to enable statistical dependencies to be adequately estimated.

[022] The signals can be further analysed, by a spectrum analyser such that the incoming sensor signals are analysed into sets of Fourier components or bins, where separate beamformers are used in each bin and the envelopes of the signals can be measured separately in each bin.

[023] Thus, the present invention provides a system and method, using an algorithm that approaches a goal of source separation in a more efficient way by modelling fluctuating signal sources as the product of a "carrier" component which propagates in the medium and an envelope modulation component which said components can be separately extracted and measured independently in a receiver and analysed statistically. If decorrelation of the carrier and modulation components can be determined simultaneously, then separation of signal sources can be more or less assured.

[024] In a further aspect of the invention, the source decoupling goal can be achieved by using the digital samples of the signal waveforms to measure a statistical dependence between the optimised output and the auxiliary lines, and driving the dependence to a minimum state by control feedback to the beam steering. This further aspect is not exclusive of the first aspect and both can be used together. In either case the introduction of a sphering operation into the auxiliary lines raises the level of a weak signal and facilitates measurement of envelope correlation and statistical dependence. In a still further aspect of the invention, a front-end signal projection operation onto the measured signal space, can help stabilize the operation of the steering correction algorithm.

[025] Such signal processing methods are equally applicable for combined equalization and directional optimisation of arrays in dispersive channels.

The wideband equivalent to the LCMP array is the classical "Frost" processor [8] in which the frequency response of the sensor array in the desired source's direction is fixed.

[026] In optical or terahertz-radio signal processing the weights applied to the "carrier" may be physically analogue in nature but operating under digital control. Envelope measurements and statistical dependence measures can also be extracted by analogue methods and sampled digitally at a convenient rate, allowing similar algorithms for optimisation under these circumstances.

DECRIPTION OF THE DRAWINGS

[027] Reference shall now be made to the Figures, wherein: Figure 1 schematically illustrates an adaptive beamformer where envelope correlation methods are applied in accordance with the present invention; Figure 2 is a flowchart illustrating one method of providing source waveform estimates in accordance with the present invention; Figure 3 is a flowchart illustrating the operation of a frequency domain receiving processor in accordance with the present invention; and, Figure 4 shows a beamformer similar to figure 1 where waveform sampling is used and statistical sampling methods are applied.

DETAILED DESCRIPTION OF THE INVENTION

[028] In order to provide a better understanding of the present invention an embodiment of the invention will now be described. It will be apparent, however, to one skilled in the art, that the present invention may be practised without these specific details. This should not be construed to limit the present invention, but should be viewed merely as an example of a specific way in which the invention can be implemented. Well known features have not been described in detail so as not to obscure the present invention.

[029] With reference to Figure 1, there is shown an adaptive beamformer system. In the area defined by box 101, a known generalized narrow-band sidelobe canceller (GSLC) signal processing unit is provided. An implementation of the present invention, being an envelope sensing adaptive beamformer, is provided in box 102. On the left side at 101 is a generalized narrow-band sidelobe canceller (GSLC), according to the state of the art, often called a Griffiths-Jim system, in which an input signal vector X at 1, optionally preceded by subspace projection and dimensionality reduction operation at 10, is separated into two channels.

In a first branch, the said vector is beamformed conventionally using steering vector S to form the "main" adaptive input y at 2 while in a second branch the input vector is conditioned to remove desired signal components by a mathematical projection operation P0 orthogonal to S at 3 giving vector Y of N-i "auxiliary" inputs. The main input signal y and auxiliary input signals Y form the inputs for the least squares adaptive canceller. In the least squares (LCMP) solution, power output P of signal d at box 4 is minimised by variation of weights W. It can be shown that this is equivalent to a decorrelation of the final output signal d from each of the auxiliary inputs Y. Since the desired signal is removed from the auxiliary inputs Y, it is not possible for the adaptive weights to change the gain of the array toward the desired signal, which consequently remains fixed and achieves the desired signal constraint.

[030] Tn figure i the optimal weighting vector W0 is given by WIPT = R4C (eqn. 1) where is the covariance matrix of the auxiliary data vectors Y obtained over T time samples R = EXXH (eqn. 2) and C is the covariance between the main beam output Yt and the auxiliary lines Y c=yH (equ.3) The dimensionality reducing matrix at iO is defined on the sampled array covariance matrixR çy,H. This has an eigenvalue decomposition R =QFQ" where Q=[QIQ2...QNj is a matrix whose columns are the N orthogonal eigenvectors and F is an ordered real diagonal matrix of eigenvalues F = diag(y,y, **7N) with ui /2 »= Beyond a certain index Al, the smaller eigenvalues correspond to non-useful noise rather than signals and can be discarded. The associated pre-processor is thus the Al rows by N columns matrix defined as A = [QIQ2...QM]H and, when pre-processor A is used, the reduced dimensionality processor inputs become X = AX. The use of the pre-processor also modifies the steering vector which transforms to S0 = AS0.

[03i] Tn accordance with one aspect of the invention, there is provided an envelope detector, shown in box 102 "Novel Envelope Detection Section".

The envelope detector performs envelope detection and low-pass filtering procedures to the auxiliary signals at 5 and estimates a correlation matrix P at box 6. The correlations so determined function in the preferred

ENV

implementation as follows: If steering S is correct, then the desired signal has been completely removed from the auxiliary inputs Y. Furthermore, after a least squares adaptation of weights W, interference is effectively removed from the optimal output d. Consequently there are no common signal sources present in d and Y under such conditions, whereby both decorrelation of signals carriers and near-independence of their envelopes is provided. On the other hand, when there is steering misalignment, S has a vector error relative to the desired signal direction, desired signal components appear in the auxiliary channels Y which have the effect of introducing a partial correlation with the main LS output d. Linear correlation measurements between Y and d are insensitive to this unwanted correlation error due to the guaranteed overall decorrelation of d and Y, in the LCMP solution.

However, under these same conditions, envelope demodulation at d and Y will continue to provide significant positive correlations provided of course there was significant envelope fluctuation or fading of the desired source in the first place.

[032] The sphering device at box 8, acts to enhance weak signals by equalizing the principle components of the data, thus increasing the sensitivity of detection of desired signal leakage into the auxiliary lines.

That is to say, the sphering device receives auxiliary signals Y and applies the sphering method, known from, for example, the method in Blind Source Separation [5]. It can be shown mathematically that desired signal leakage into the auxiliary channels is mostly associated with only a single principle component. The sphering operation is a matrix-valued operator S derived from an eigenvalue decomposition. Let R defined in equation (1) above have eigenvalue expansionR=ZAZ" where Z is a matrix whose columns are the orthogonal eigenvectors and A is a positive real diagonal matrix of its eigenvectors. Then the sphering matrix S is defined as S =A_*ZH and the "sphered" auxiliary inputs are defined byY'=SY.

[033] The overall adaptive procedure continues by way of adjusting steering vector S via line 7 until envelope correlations between the LS output d and the auxiliary inputs Y become small, conveniently vanishingly small. As S is adjusted, the LCMP solution is continually updated at 101, though this updating is not an onerous requirement because of the numerical efficiency of the LCMP. One simple strategy to minimise all correlations is to sum the N-i envelope correlations, known to be always positive-valued, and minimise the resulting sum. There are numerous efficient mathematical methods for minimising such positive sums, known as multivariate optimization, which are described in the relevant literature and are commonly provided in computer mathematics packages. By squaring the values of the envelope amplitudes before correlation, the concept of power-linearity of the sums independent sources' powers can also be invoked.

[034] With reference to Figure 2, there is shown an alternative implementation in which source-separating beamformer at 22 exploits the same step of decorrelating both the carrier and modulation components (as described above and as known from the prior art, with reference to the ICA and HOS algorithms) at a beamformer output 23. Sensor inputs 21 arising from spatial signal sources 20 are linearly combined in mathematical beamformer matrix device 22 producing estimates of the source waveforms at output 23. The propagating carrier correlations are estimated at 24 while the output envelopes are measured at processor 25 and statistical dependence is estimated at processor 26, the simplest estimate of said dependence is the cross correlation of the envelope powers. Processor 27 performs a mathematical optimisation algorithm, whereby to set both sets of correlations to zero by providing correction signals via line 28 to the beamformer matrix 22. The adapted beamformer matrix B corresponds closely to the one achieved in the standard Blind Source Separation method [5] but with the difference that a combination of envelope detection and carrier decorrelation has been used in place of the higher order statistics conventionally used in BSS.

[035] With reference to Figure 3 there is shown a wideband implementation wherein the basic method of Figure 2 is extended for use in dispersive (convolutional) channels for optimized directional beamforming. This situation, referred to by experts in the field as a "convolutional channel problem", can be solved by operating the receiver in a multiplicity of narrow bandwidths over an envelope observation interval TE. That is to say, the system can be considered in a simplistic fashion such that, in respect of an observation interval, narrow bands are created by dividing the incoming data into a series of N slots of limited duration indicated by reference numeral 30 in Figure 3, then performing a discrete Fourier Transform upon each slot indicated by reference numeral 31, to form a number of frequency bins F, and subsequently reorganizing the numbered series of Fourier spectrum bins as a number F of different narrow band separation problems with N samples in time, a procedure known from, for example, references 9 and 10, respectively U56167417 and U56898612. In this manner the wide band separation problem is transformed into a corresponding number F, of narrow band separation problems. It is noted that in accordance with the system described in Figure 2 there are as many beamformer matrices 22 as there are frequency bins, F. [036] In the frequency domain solution the discrete complex Fourier Transform can be applied to the time series Xr sampled at a particular array element input with a time window t = 0,...,T-1. The Fourier component at frequency k is given by 71k = sMx exp(-2k/T), Wt being a standard windowing function such as raised cosine. By repetition of this procedure at a series of consecutive partially-overlapping windows, with typically a SO% overlap, the sampling times (assuming even-valued T) are t={0...T-iJ,CT+f..2T--IJ,(2T...2T-i-*-IJ... etc. We thus obtain, for each array element, a series of L Fourier components XkI for overlapping windows indexed i = 0,..L-1. The narrow-band array processor operating at frequency index k then sees a sequence of L narrow-band input samples from the array.

[037] In this further aspect of the invention, the time-frequency signal envelopes within each of these Fourier spectrum bins is determined and used as a measure of source separation for that bin, in a manner similar to that previously described, above.

[038] Numeral 32 indicates, in Figure 3, the spectrum envelopes in the Fourier system that are measured in a representational system by observing the instantaneous (single slot) Fourier bin envelopes over the duration of N slots. In a further feature of the envelope measurement method, the envelope may also be averaged over a number of parallel frequency bins thereby taking advantage of a common characteristic of fluctuating sources, namely, their envelope fading is coherent not only in time but also across a finite spectrum bandwidth as is typical in speech spectrograms. The joint time and frequency envelope averaging windows are adjusted to achieve best system performance given the characteristics of the sources of interest.

[039] It is to be noted that measurement of envelope levels in wideband source separation has previously been used to solve the spectrum permutation problem wherein the identities of multiple sources are ambiguous in the different frequency bins [12]. This is a different concept in which the use of envelopes plays no part in the actual separation of source signals and is applied purely to reconcile the identities of signal sources across the frequency bins in the post source separation phase.

[040] In accordance with a second aspect of the invention, Figure 4 shows the principle component outputs of said sphering operation at box 8 acting on the auxiliary channels, together with the optimised LCMP output, being passed to an algorithm at box 9 capable of measuring generalised statistical dependence between signals. A measure of the dependence between the LCMP output and at least one principle component output from the sphering system is then driven to a minimum value by adjustment of the steering vector via the feedback control line at 7. This technique is particularly, though not exclusively, favourable for non-fading digital communications signals using QAM (quadrature amplitude) modulation and a good, though not the only, measure of statistical dependence is mutual information as described in [6], [7]. In the case of fading communications signals, Higher Order Statistics [5] can be used to measure dependence of the sampled waveform.

[041] The said different aspects of the invention i.e. Fourier analysis, sphering, envelope decorrelation and statistical waveform examination are not exclusive and can be applied either separately or in any combination, to best match the perceived circumstances of the unpredictable non-cooperative received desired and interference signals, typically in a surveillance scenario.

REFERENCES CITED

[1] B. Allen and M. Ghavami: "Adaptive Array Systems", Wiley-Blackwell 2005 [2] L 3 Griffiths and C W Jim. "An Alternative Approach to Lineary Constrained Adaptive Beamforming", IEEE Trans. Antennas and Propagation, AP-26, pp. 27-34. Jan. 1982 [3] I S Reed, 3 D Mallett, and L E Brennan, "Rapid convergence rate in adaptive arrays", IEEE Trans Aerospace, AES-lO, 1974, pp. 853-863 [4] 3 E Hudson, "Adaptive Array Principles" Peter Peregrinus 1982 [5] A. Chichocki and S. Amari. "Adaptive Blind Signal and Image Processing", Wiley, 2002, Chapter 4.

[6] Georges A. Darbellay and Petr Tichavsky: "Independent Component Analysis Through Direct Estimation of The Mutual Information", Proceedings ICA2000: Second International Workshop on Independent Component Analysis and Blind Separation, Helsinki, 19-22 June 2000, pp. 69-75.

[7] 3 E Hudson. "Signal Processing Using Mutual Information", IEEE Signal Processing Magazine, Vol 23, No. 6, Nov 2006, pp. 50-54 [8] 0. L. Frost "An Algorithm for Linearly Constrained Adaptive Array Processing", Proceedings IEEE, Vol 60, 1971, pp.661-675.

[9] US patent US6452917 "Channel estimation in a CDMA wireless communication system", Leung, Issued: September 17, 2002 [10] Patent number: U56167417. "Convolutive blind source separation using a multiple decorrelation method", Inventors: Lucas Parra, Clay Douglas Spence.

[11] Patent number: U56898612. "Method and system for on-line blind source separation". Inventors: Lucas Parra, Clay Douglas Spence.

[12] S Ikeda and N Murata. "A method of ICA in time-frequency domain" Proceedings of International Workshop on Independent Component Analysis and Blind Signal Separation (ICA'YY), pp.365-371, Aussions, France, January 1999

Claims

CLAIMS1. A receive-signal processor apparatus, for forming an optimised least-squares beamformer for receipt of target source signals, the apparatus comprising a multi-element sensor array, a digital sampling arrangement operable to record received signals, a signal processing unit programmed to form a least squares adaptive beam and to monitor signal samples and signal envelopes at intermediate test points.
2. The receive-signal processor apparatus of claim 1, wherein the functionality is achieved implicitly by integrating one or more of the separate components of the processing chain into a more complex digital algorithm.
3. The receive-signal processor apparatus of claim 1 or 2, wherein the test points are selected so that when the said beamformer is correctly steered for the desired signal, the envelope correlations and/or statistical dependencies are essentially zero at the test points.
4. The receive-signal processor apparatus according to any one of claims 1 -3, wherein the processor further comprises a sampling unit, programmed with a steering optimisation algorithm and operable to sample the signals and signal envelopes at said intermediate test points and output signals to drive the statistical dependence between the signals and the signal envelopes of the least squares beamformer output and the test points to a minimal value.
5. The receive-signal processor apparatus according to any one of claims 1 -4, wherein the signal and signal envelope sampling points and the optimal beam forming function are implemented wholly or partly by techniques involving analogue electromagnetic physical devices and filters.
6. The receive-signal processor apparatus according to any one of claims 1 -5, wherein the signals propagate in two or more propagation media.
7. An adaptive signal transmission transfer system comprising at least one signal transmitter and a receive apparatus comprising a multi-element sensor array, a digital sampling arrangement operable to record received signals, a signal processing unit programmed to form a least squares adaptive beam and to monitor signal samples and signal envelopes at intermediate test points; wherein the signal transmitter is operable to transmit, through at least one propagation medium, an information-bearing modulating signal upon a carrier wave to the receiving sensors; wherein the receiving apparatus is operable to perform linear filtering operations on the carrier component and to sample and demodulate the information-bearing modulating signal; whereby to provide an estimate of the waveform of at least one desired signal.
8. An adaptive signal transmission transfer system according to claim 7, wherein the modulating signal is arranged to be sampled and demodulated at the test points simultaneously with the performance of linear filtering operations on the carrier component.
9. A method of monitoring and optimising the steering of a constrained adaptive least squares beamformer, said beamformer comprising a receive array of sensors operable to provide signals to an adaptive signal processor, the method comprising the steps: forming a steered beam pointing to a prior estimate of the direction of said source and the further step of forming a number of minima in the directional pattern of said steered beam which align to the directions of interference sources whose presence or plurality is not known a-priori; determining the statistical dependence of the signals and their envelope fading at the beamformer output with the signals and their envelope fading at one or more selected internal points within said adaptive algorithm; and, generating corrections for the said beam steering vector, whereby to align the vector and to minimise said signal and signal envelope statistical dependence.
10. A method according to claim 9, wherein said internal test points are located in such a manner that, when the beam steering vector has been adjusted to correspond with the actual desired signal vector, the desired signal is essentially eliminated from said test points.
11. A method according to claims 9 and 10, wherein a Griffiths-Jim configuration, is used to implement the least squares beamforming and wherein the Griffiths-Jim configuration provides a main input and at least one auxiliary input into the least squares adaptive process, operable to provide the said signal and signal envelope test points.
12. A method according to claim 11, whereby the auxiliary inputs are prefiltered by a sphering operation, such that the principle statistical components of the auxiliary data are equalized to a uniform level prior to sampling.
13. A method according to claim 10, wherein the signal envelopes are sampled at said test points, and their envelopes are further filtered to remove unwanted carrier components and wherein a matrix of numerical values, indicating the statistical dependence between the envelopes at the said test points and said least squares signal output, is formed by averaging with a window over some time period.
14. A method according to claim 13, wherein a multivariate optimisation search technique is applied to the dependence matrix whereby to drive a beam steering performance metric to a best value by adjustment of the beam steering vector.
15. A method according to claim 13, wherein the dependency measure is a simple correlation matrix of a power law function of the fading envelopes.
16. A method according to claim 13, further comprising the formation of a single-valued performance metric from the dependence matrix which is an increasing function of the degree of departure of the least squares beamformer steering away from the true desired signal direction.
17. A method according to any one of claims 4 -12 wherein the cross-correlation of the envelopes of the signals at test points is augmented or replaced by a measurement of statistical dependence between the same raw wideband signals (i.e. without envelope detection) using any competent measure of dependence applicable to the types of signal occurring.
18. A method according to claim 15, comprising the step of optimising the least squares beamformer steering when applied to acoustic, radio, millimetric, terahertz, infrared, optical, or other electromagnetic propagating scenarios which contain fluctuating information sources.
19. A method according to claim 18, wherein said optimisation techniques are applied in wideband adaptive radio communication and acoustic receiving arrays, to correct steering signals in a wide-band least-squares beamformer applied to propagating carrier signals and dispersive channels, wherein the sensor weights comprise frequency filters.
20. A method according to claim 18, wherein said information sources and beamforming array simultaneously use channels operative in more than one propagation medium.
21. A method according to any one of claims 18 -20, wherein a variable linear beamformer 22, operating on a vector of sensor signals, said matrix producing at least one beamformer output signal which is an approximation to the waveforms of the one or more distinct signal sources illuminating said sensor array from said different directions; the method further comprising the use of an array of envelope detectors 25 which determine the envelope amplitudes of the signals at outputs 23.
22. The method of claim 21, further comprising the step of estimating the statistical dependence both of said envelopes and of the underlying carrier components of said signals, employing said statistical dependence estimates in a numerical optimisation search algorithm, which adjusts variable coefficients in matrix 22 in a fashion so as to jointly eliminate statistical dependence between the envelope and carrier signal outputs.
23. The method of claim 21, wherein the analysis of signals further comprises the following steps: i) analysing incoming sensor signals into sets of Fourier components; ii) separating incoming sources out within individual Fourier frequency bins by applying different adaptive beamformers in each frequency bin; iii) measuring the envelopes of beamformer output signals over a time frequency window designed to fit the known source envelope characteristics, commonly known as the spectrogram to experts in speech analysis, in a maximally advantageous manner; and, iv) further adapting all the multiple beamformers to such a state that a desired signal estimate, reconstructed from its frequency components, is maximally independent of the interference in the auxiliary signals, such dependence being estimated by whatever statistical techniques are most advantageous.
24. A receive-signal processor apparatus, substantially as disclosed herein, with reference to any one or more of the figures as shown in the accompanying drawing sheets
25. A method of use of a receive-signal processor apparatus substantially as disclosed herein, with reference to any one or more of the figures as shown in the accompanying drawing sheets