Blind source separation utilizing a spatial fourth order cumulant matrix pencil

[0002] The present application claims priority to U.S. Provisional Application Serial No. 60/374,149 filed Apr. 22, 2002, entitled “Blind Source Separation Using A Spatial Fourth Order Cumulant Matrix Pencil” the entirety of which is incorporated herein by reference.

[0003] The present invention is generally related to separating individual source signals from a mixture of source signals, and more specifically related to blind source separation.

[0004] A classic problem in signal processing, often referred to as blind source separation (BSS), involves recovering individual source signals from a composite signal comprising a mixture of those individual signals. An example is the familiar “cocktail party” effect, wherein a person at a party is able to separate a single voice from the combination of all voices in the room. The separation is referred to as “blind” because it is often performed with limited information about the signals and the sources of the signals.

[0005] Blind source separation (BSS) is particularly applicable to cellular and personal wireless communications technologies, wherein many frequency bands have become cluttered with numerous electromagnetic emitters, often co-existing in the same spectrum. The problem of co-channel emitters is expected to only worsen in years to come with the development of low power, unlicensed wireless technologies such as Bluetooth® and other personal area networks. These developments have resulted in the use of multiple sensors and array signal processing techniques to perform spectral monitoring. Such techniques enable the exploitation of spatial information to separate co-channel emitters for detection, classification, and identification. Additionally, many signals designed for a low probability of detection (LPD) or low probability of intercept (LPI) may use ambient background electromagnetic radiation and known co-channel emitters as a means of concealment. Constructing single sensor receiver systems with the required sensitivity to such emitters is generally prohibitive. Thus, many applications utilize BSS and sensor arrays.

[0006] Several techniques have been proposed to solve the BSS problem. These can be classified into two main groups. The first being based on second-order statistics, and the second being based on higher-order statistics, such as those based on independent components analysis (ICA) and other higher-order spectral estimation and spatial filtering techniques.

[0007] One second-order blind source separation technique is a spectral estimation method that exploits the rotational invariance of the signal subspace to estimate the direction of arrival. This technique known as Estimation of Signal Parameters via Rotational Invariance (ESPRIT) employs pairs of calibrated elements and uses a matrix-pencil formed by the spatial correlation and cross-correlation matrices. See, for example, R. Roy, A. Paulraj, T. Kailath, “Direction-of-Arrival Estimation by Subspace Rotation Methods,” Proc. ICASSP86, pp. 2495-2498 and R. Roy and T. Kailath, “ESPRIT—Estimation of Signal Parameters via Rotational Invariance Techniques,” IEEE Trans. on ASSP, Vol. 37, No. 7, July 1989, pp. 984-995, which are each incorporated by reference in their entirety as if presented herein. However, a disadvantage of ESPRIT is that at very low signal-to-noise ratios the signal plus noise subspace and the noise subspace are indistinguishable thus making the conversion of the noise subspace of the spatial correlation matrix impractical. This is due in part to ESPRIT requiring an estimation of the noise variance to convert the noise subspace into a null subspace of the spatial correlation matrix, and assuming that the noise is spatially white.

[0008] Another second order blind source separation technique is known as the Constant Modulus Algorithm (CMA), also referred to as Goddard's algorithm. The CMA is an adaptive spatial filtering technique, which is used to perform source separation by determining a set of spatial filter tap weights that forces the output signal to have a modulus as close to unity as possible. Typically, the CMA is performed sequentially to separate all source signals. The CMA has been suggested for use as a blind equalization technique to reduce inter-symbol interference of constant modulus signals, such as FSK, PSK, and FM modulated signals, on telephone channels (See, for example, D. N. Godard, “Self-recovering Equalization and Carrier Tracking in Two-dimensional Data Communication Systems,” IEEE Trans. Commun., Vol. COMM-28, November 1980, pp. 1867-1875, which is incorporated by reference in its entirety as if presented herein.), and to perform blind equalization to combat multi-path fading and to suppress co-channel interfering signals (See, for example, B. G. Agee, “The Property Restoral Approach to Blind Adaptive Signal Extraction,” Ph.D. Dissertation, Dept. Elect. Eng. Comput. Science, Univ. of Calif., Davis, 1989, which is incorporated by reference in its entirety as if presented herein). However, the CMA technique works only for signals with a constant modulus and is not practicable for most applications. In practice, because signals are filtered to limit their spectral occupancy at the transmitter and to limit the noise bandwidth at the receiver, true constant modulus signals rarely exist. Furthermore, at very low signal-to-noise ratios, noise dominates the input signal thus distorting the spatial filter's output signal's modulus and causing large fluctuation in the error signal used in adaptation.

[0009] Yet another second order blind source separation technique is a spatial filtering technique using second-order cyclostationary statistics with the assumption that the source signals are cyclostationary. This technique was developed as a blind single-input single-output (SISO) channel identification technique for use in blind equalization (See, for example, L. Tong, G. Xu, and T. Kailath, “Blind Identification and Equalization Based on Second-Order Statistics: A Time-Domain Approach,” IEEE Trans. Information Theory, Vol. 40, No. 2, March 1994, pp. 340-349, which is incorporated by reference in its entirety as if presented herein), and was later adapted to perform the blind separation of cyclostationary signals (See, for example, L. Castedo and A. R. Figueiras-Vidal, “An Adaptive Beamforming Technique Based on Cyclostationary Signal Properties,” IEEE Trans. Signal Processing, Vol. 43, No. 7, July 1995, pp. 1637-1650, which is incorporated by reference in its entirety as if presented herein). One disadvantage of this cyclostationary approach is that it requires different symbol rates and/or different carrier frequencies for separating multiple superimposed signals. Another disadvantage is that residual carrier offsets with a random initial phase can cause the signals to become stationary, causing the cyclostationary assumption to become invalid. Other disadvantages include the fact that this approach precludes separating sources that may use non-linear or non-digital modulations, and this approach assusmes the noise vector is temporally and spatially white.

[0010] Still another blind source separation technique based on second-order statistics is referred to as Second-Order Blind Identification (SOBI). See, for example, A. Belouchrani, K. Abed-Meraim, J. F. Cardoso, and E. Moulines, “Blind Source Separation Using Second-Order Statistics,” IEEE Trans. Signal Processing, Vol. 45, No. 2, February 1997, pp. 434-444, for a description of this technique, which is incorporated by reference in its entirety as if presented herein. This technique exploits the time coherence of the source signals and relies on the joint diagonalization of a set of covariance matrices. A disadvantage of this technique is that it requires additive noise to be temporally white and uses the eigenvalues of the zero lag matrix to estimate the noise variance and to spatially whiten the sensor output vector. Another disadvantage is that at low signal to noise rations, the estimation of the noise variance is difficult at best and impossible in most cases. Yet another disadvantage is that the number of sources must be known or estimated. Still, another disadvantage is that the SOBI technique is valid for spatially correlated noise. The estimation of the noise covariance is extremely difficult even at high signal-to-noise ratios, thus making the technique impracticable.

[0011] Another second-order blind source separation technique is based on the generalized eigen decomposition of a spatial covariance matrix-pencil. This technique is related to the ESPRIT algorithm, but does not require 2N sensors to separate up to N-1 signals because time and/or polarization diversity is used to estimate the pair of spatial covariance matrices. This technique uses a dual polarized array and estimates the spatial covariance matrices on the two orthogonal polarizations to form the matrix pencil. See, for example, A. Belouchrani, K. Abed-Meraim, J. F. Cardoso, and E. Moulines, “Blind Source Separation Using Second-Order Statistics,” IEEE Trans. Signal Processing, Vol. 45, No. 2, February 1997, pp. 434-444, which is incorporated by reference in its entirety as if presented herein. This later evolved into using spatial covariance matrices at a zero time lag and a non-zero time lag to form the matrix-pencil. One disadvantage of this technique is that it is limited to separating up to N-1 sources with N sensors. This is due in part to the approach requiring the estimation of the noise variance, similar to ESPRIT, and assuming that the noise is spatially and temporally white.

[0012] Finally, another second order blind source separation technique utilizes two non-zero time lags in the estimation of the spatial covariance matrices. See, for example, C. Chang, Z. Ding, S. F. Yau, and F. H. Y. Chan, “A Matrix-Pencil Approach to Blind Separation of Non-White Sources in White Noise,” Proc. ICASSP98, Vol. IV, pp. 2485-2488, and C. Chang, Z. Ding, S. F. Yau, and F. H. Y. Chan, “A Matrix-Pencil Approach to Blind Separation of Colored Non-Stationary Signals,” IEEE Trans. Signal Processing, Vol. 48, No. 3, March 2000, pp. 900-907, which are each incorporated by reference in their entirety as if presented herein. The non-zero time lags combined with the assumption that the noise vector is temporally white eliminates the need to estimate the noise variance(s) in order to remove the noise subspace and thus allows the technique to separate up to N sources with N sensors. However, disadvantages of the this second-order matrix-pencil technique include the requirement that the noise vector be temporally white and the fact that the estimate of the separation matrix is not done with one of the spatial covariance matrices at a zero time lag, which is when the signal auto-correlations are at their maximum values. These disadvantages are exacerbated by the fact that in many practical applications the noise bandwidth is limited to be on the order of the signal bandwidth making the noise and signal decorrelation times to be approximately equal.

[0013] Higher-order blind source separation techniques include all methods that employ statistics of order greater than two. These include independent component analysis (ICA) methods, spatial filtering methods, and spectral estimation based methods and can use either moments or cumulants of order three or higher.

[0014] The independent components analysis (ICA) methods seek a separating matrix that maximizes the statistical independence of the outputs of the separation process. See, for example, A. K. Nandi, Blind Estimation Using Higher-Order Statistics. (Kluwer Academic, Dordecht, The Netherlands: 1999), A. Hyvärinen, “Survey on Independent Component Analysis,” Neural Computing Surveys, Vol. 2, No. 1, 1999, pp. 94-128, and J. F. Cardoso, “Blind Signal Separation: Statistical Principles'” Proc. of the IEEE, Vol. 9, No. 10, October 1998, pp.2009-2025, which are each incorporated by reference in their entirety as if presented herein. Various ICA methods are known, and are primarily differentiated by the objective/contrast function used to measure statistical independence. Examples of ICA based blind source separation algorithms include the Jütten-Herault algorithm (See, for example, C. Jütten and J. Herault, “Blind Separation of Sources, Part I: An Adaptive Algorithm Based on Neuromimetic Architectures,” Signal Processing, Vol. 24, 1991, pp. 1-10, which is incorporated by reference in its entirety as if presented herein), which attempts to achieve separation via canceling non-linear correlations through the use of a neural network; the higher-order eigenvalue decomposition or HOEVD method (See, for example, P. Comon, “Independent Component Analysis, A New Concept?,” Signal Processing, Vol. 36, No. 3, April 1994, pp. 287-314, which is incorporated by reference in its entirety as if presented herein); the joint approximate diagonalization of eigen matrices (JADE) algorithm (See, for example, J. F. Cardoso and A. Souloumiac, “Blind Beamforming for Non-Gaussian Signals,” IEE Proceedings F, Vol. 140, No. 6, December 1993, pp. 362-370, which is incorporated by reference in its entirety as if presented herein), which exploits the eigen structure of the fourth-order cumulant tensor; the information maximization or infomax technique (See, for example, A. J. Bell and T. J. Sejnowski, “An Information-Maximization Approach to Blind Source Separation and Blind Deconvolution,” Neural Computing, Vol. 7, 1995, pp. 1 129-1159, which is incorporated by reference in its entirety as if presented herein), which seeks a separation matrix that maximizes the output entropy; and the equivariance adaptive source separation or EASI algorithm (See, for example, J. F. Cardoso and B. Hvam Laheld, “Equivariance Adaptive Source Separation,” IEEE Trans. On Signal Processing, Vol. 44, No. 12, December 1996, pp. 3017-3030, which is incorporated by reference in its entirety as if presented herein), in which an estimate of the mixing matrix is chosen such that transformations in the sensor output data produces a similar transformation in the estimated mixing matrix. Disadvantages of these ICA techniques include (1) a pre-whitening step is required, (s), the techniques need to be parametrically tuned to the application, and (3) the ICA based techniques tend to have a slow convergence time.

[0015] Another higher-order blind source separation technique, known as the Kurtosis Maximization Algorithm (KMA), utilizes kurtosis as a separation measure (See, for example, Z. Ding, “A New Algorithm for Automatic Beamforming,” Proc. 25^thAsilomar Conf. Signals, Syst., Comput., Vol. 2, 1991, pp. 689-693, and Z. Ding and T. Nguyen, “Stationary Points of Kurtosis Maximization Algorithm for Blind Signal Separation and Antenna Beamforming,” IEEE Trans. Signal Processing, Vol. 48, No. 6, June 2000, pp. 1587-1596, which is incorporated by reference in its entirety as if presented herein). The KMA is an adaptive spatial filtering technique with a non-deterministic convergence time. One disadvantage of the KMA is that it can not simultaneously separate the source signals. The KMA requires that the sources be separated sequentially one at a time, starting with the signal having the largest kurtosis. Thus, no knowledge of the number of signals is provided by the technique. Other disadvantage of the KMA are that it requires the noise to be spatially white and its convergence with multiple signals has only been proven in the noise free case.

[0016] Finally, another higher-order BSS technique is a higher-order version of the ESPRIT algorithm. As the name implies, higher-order ESPRIT replaces the spatial correlation or covariance matrices with spatial fourth-order cumulant matrices. See H. H. Chiang and C. L. Nikias, “The ESPRIT Algorithm with Higher-Order Statistics,” Proc. Workshop on Higher-Order Spectral Analysis, Vail, CO., June 1989, pp. 163-168, C. L. Nikias and A. P. Petropulu, Higher-Order Spectra Analysis: A Non-Linear Signal Processing Framework. (PTR Prentice-Hall, Upper Saddle River, N.J.: 1993), and M. C. Dogan and J. M. Mendel, “Applications of Cumulants to Array Processing—Part I: Aperture Extension and Array Calibration,” IEEE Trans. Signal Processing, Vol. 43, No. 5, May 1995, pp. 1200-1216, for descriptions of various types of higher-order ESPRIT techniques, which are each incorporated by reference in their entirety as if presented herein. These higher-order ESPRIT techniques possess several disadvantages. Some of these higher-order ESPRIT techniques require the calibration of the N pairs of sensors but can now separate up to N sources since the noise variance no longer needs to be estimated. Other higher-order ESPRIT techniques require the array to be calibrated to guarantee that the pairs of sensors have identical manifolds (similar to the standard ESPRIT). These techniques degrade in performance significantly as the sensor pairs' manifolds deviate from one another.

[0017] Each of the above-mentioned blind source separation techniques have the disadvantages noted. Additionally, none of the above-mentioned blind source separation techniques operate satisfactorily in the situation where there is a low signal-to-noise plus interference ratio. Accordingly, an improved blind source separation technique is desired.

[0018] In one embodiment of the present invention, a method for separating a plurality of signals provided by a respective plurality of sources and received by an array comprising a plurality of elements, includes generating a separation matrix as a function of time differences between receipt of the plurality of signals by the plurality of elements and a spatial fourth order cumulant matrix pencil. The method also includes multiplying the separation matrix by a matrix representation of the plurality of signals.

[0019] In another embodiment of the present invention, a system for separating a plurality of signals provided by a respective plurality of sources includes a receiver for receiving the plurality of signals and for providing received signals. The system also includes a signal processor for receiving the received signals, generating a separation matrix, and multiplying the separation matrix by a matrix representation of the received signals. The separation matrix is a function of time differences between receipt of the plurality of signals by the receiver and a function of a spatial fourth order cumulant matrix pencil.

[0020] In the drawings:

[0021] FIG. 1 is a functional block diagram of a system for performing blind source separation utilizing a spatial fourth order cumulant matrix pencil in accordance with an embodiment of the present invention;

[0022] FIG. 2 is an illustration of signal source, array elements, and a processor for performing array signal processing and BSS processing in accordance with an embodiment of the present invention;

[0023] FIG. 3 is an illustration of a MIMO blind channel estimation scenario showing five unknown sources having distinct radiating patterns and five sensors having distinct receiving patterns;

[0024] FIG. 4 is a graphical illustration of time delays between sensors and sources;

[0025] FIG. 5 is an illustration depicting blind source separation (BSS) showing an input signal mixed with noise provided to the separation process;

[0026] FIG. 6 is an illustration depicting repeating the separation process for a single repeated eigenvalue;

[0027] FIG. 7 is a flow diagram of a process for performing blind source separation using the spatial fourth-order cumulant matrix-pencil in accordance with an embodiment of the present invention; and

[0028] FIG. 8 is a continuation of the flow diagram of FIG. 7.

[0029] A technique for performing blind source separation (BSS) in accordance with the present invention utilizes cumulants in conjunction with spectral estimation of the signal subspace to perform the blind separation of statistically independent signals with low signal-to-noise ratios under a narrowband assumption. This BSS technique makes use of the generalized eigen analysis of a matrix-pencil defined on two similar spatial fourth-order cumulant matrices. The herein described BSS technique utilizes a higher-order statistical method, specifically fourth-order cumulants, with the generalized eigen analysis of a matrix-pencil to blindly separate a linear mixture of unknown, statistically independent, stationary narrowband signals at a low signal-to-noise ratio having the capability to separate signals in spatially and/or temporally correlated Gaussian noise. This BSS technique provides a method to blindly separate signals in situations where no second-order technique has been found to perform the blind separation, for example, at a low signal-to-noise ratio when the number of sources equals the number of sensors.

[0030] To describe this BSS technique, a definition of a spatial fourth-order cumulant matrix suited to blind source separation with non-equal gain and/or directional sensors and a definition of a spatial fourth-order cumulant matrix-pencil using temporal information are provided. The herein description also utilizes the concept of separation power efficiency (SPE) as a measure of the BSS technique's performance, and applies the concept of wide sense equivalence between matrix-pencils to the field of matrix algebra.

[0031] As an overview, the BSS technique described herein utilizes cumulants in conjunction with a spectral estimation technique of the signal subspace to perform blind source separation in the presence of spatially and/or temporally correlated noise at low signal-to-noise ratios. Prior to deriving a separation algorithm based on cumulants, a narrowband array model is developed, all assumptions are stated, four performance measures are defined, and the relevant cumulant properties that allow for the spatial mixing matrix information to be extracted from a spatial cumulant matrix are presented. A novel spatial cumulant matrix definition is then developed and its' relevant matrix properties are derived in order to determine which mathematical methods are valid for extracting the spatial information about the mixing matrix. Additionally, two alternative definitions for the spatial fourth-order cumulant matrix are described and relevant properties are derived. Furthermore, the definitions, properties, and use of a generalized eigen analysis of a matrix-pencil defined on two similar spatial fourth-order cumulant matrices are explored and their applicability to solving the blind source separation problem is investigated. A process is described for performing the blind source separation based on the signal subspace technique using matrix-pencils. In the process the concept of wide sense equivalence between matrix-pencils is developed and then used to show that the generalized eigenvalues of a matrix-pencil defined on two similar spatial fourth-order cumulant matrices are equal to the ratio of fourth-order cumulant of each source at a set of time lags (0,0,0) to the fourth-order cumulant at the set of lags, (τ₁, τ₂, τ₃). Thus the concept of a normalized fourth-order auto-cumulant function is introduced. To further aid in understanding this BSS technique, notation used herein is presented below.

[0032] M≡Number of Sources

[0033] N≡Number of Sensors

[0034] P_j≡Normalized Power of the j^thsource signal

[0035] m_j(t)≡Continous Time Unit Power Modulated Signal from the j^thsource

[0036] s_j(t)≡Continous Time Signal from the j^thsource ={square root over (P_j)}m_j(t)

[0037] r_j(t)≡Delayed version of s_j(t)

[0038] x_i(t)≡Continous Time Signal from the i^thsensor.

[0039] x(t)≡The vector of sensor outputs.

[0040] h_ij(t)≡Continous Time Impulse Response of the channel between the j^thsource and the i^thsensor

[0041] n_i(t)≡Additive Noise Process at the i^thsensor.

[0042] σ_i²≡Variance of the Noise Process at the i^thsensor.

[0043] τ_ij≡Propogation Delay from the j^thsource to the i^thsensor

[0044] Δτ_l,k,j≡“Differential Time Delay”. The difference in propagation delay from the output of the j^thsource to the k^thsensor output and from the output of the j^thsource to the l^thsensor output.

[0045] =τ_lj−τ_kj

[0046] {overscore (τ)}_j≡“Reference Time Delay” from the j^thhsource to some arbitrary array reference point in the vicinity of the array. Nominally this can be the average propagation delay to all N sensors from the j^thhsource.

[0047] Δτ_ij≡“Relative Time Delay”. The difference in propagation time from the j^thsource to the i^thsensor and the array refence point.

[0048] τ≡Time Difference in Correlation of Stationary Processes

[0049] ν_ij≡Complex Weight for the j^thsource at the i^thsensor for the Narrow Band Model. The ij element of the “Mixing Matrix”. The i^thelement of the j^thsteering vector.

[0050] v_j≡The j^th“Steering Vector” for the Narrow Band Model.

[0051] V≡The Narrow Band Model “Mixing Matrix”.

[0052] w_ij≡Complex Weight for the j^thsource at the ih sensor for the Narrow Band Case. The ij element of the “Separating Matrix”. The i^thelement of the j^thsensor weight vector.

[0053] W The “Separation Matrix”.

[0054] α_ij≡Real valued gain(attenuation) of the channel from the i^thsource output to the j^thsensor output.

[0055] BW_NEq[ ]≡Noise Equivalent Bandwidth

[0056] BW_ij^COH≡Coherence bandwidth of the Channel between the j^thhsource and the i^thsensor.

[0057] y_j(t)≡The j^thoutput from the separation process. It is a noisy estimate of the of the j^thdelayed source signal, r_j(t).

[0058] y(t)≡The vector of output signals from the separation process.

[0059] ρ_j≡The j^thsignal loss term. Element of the “loss” matrix.

[0060] S_j≡The separation process output signal power of the j^thsource signal.

[0061] I_j≡The residual interference power in the j^thmseparation process output.

[0062] N_j≡The noise power in the j^thseparation process output.

[0063] ζ_j≡The “Interference-to-Signal Ratio” for the j^thseparation process output.

[0064] ISR_avg≡The “Average Interference-to-Signal Ratio”.

[0065] ISR_max≡The “Maximum Interference-to-Signal Ratio”.

[0066] ξ_j≡The “Power Efficiency” of a blind source separation algorithm for the j^thsource.

[0067] ξ_avg≡The “Average Power Efficiency” of a blind source separation algorithm.

[0068] ξ_min≡The “Minimum Power Efficiency” of a blind source separation algorithm.

[0069] C_x⁴(τ₁, τ₂, τ₃)≡N×N “Spatial Fourth-Order Cumulant Matrix 1” with delay lags τ₁, τ₂, τ₃.

[0070] C_x^4′(τ₁, τ₂, τ₃)≡N×N “Spatial Fourth-Order Cumulant Matrix 2” with delay lags τ₁, τ₂, τ₃.

[0071] C_x^4″(τ₁, τ₂, τ₃)≡N×N “Spatial Fourth-Order Cumulant Matrix 3” with delay lags τ₁, τ₂, τ₃.

[0072] Cum [ ]≡Cumulant Operator.

[0073] C_r⁴(τ₁, τ₂, τ₃)≡The fourth-order cumulant of the j^thsource signal with delay lags τ₁, τ₂, τ₃. Also referred to as the fourth-order auto-cumulant.

[0074] {tilde over (V)}≡The “Modified Mixing Matrix”. Defined as the Hadamard Product V⊚V⊚V.

[0075] {overscore (C)}_r⁴(τ₁, τ₂, τ₃)≡The normalized fourth-order cumulant of the j^thsource signal with delay lags τ₁, τ₂, τ₃. Also referred to as the normalized fourth-order auto-cumulant.

[0076] C_r⁴(τ₁, τ₂, τ₃)≡M×M Diagonal “Fourth-Order Signal Cumulant Matrix” with delay lags τ₁, τ₂, τ₃.

[0077] C( )≡“Column Space” of a matrix.

[0078] N_r( )≡The “Rigth Null Space” of a matrix.

[0079] N_l( ) The “Left Null Space” of a matrix.

[0080] I_N≡N×N Identity Matrix.

[0081] tr( )≡The “Trace” of a matrix.

[0082] sp( )≡The “Span” of a sub-space.

[0083] ρ( )≡The “Rank” of a matrix.

[0084] {right arrow over (τ)}≡Vector notation for the set of delay lags, {τ₁, τ₂, τ₃}.

[0085] P_x(λ, {right arrow over (τ)})≡The “Spatial Fourth-Order Cumulant Matrix-Pencil” using a pair of Spatial Fourth-Order Cumulant Matrix 1's.

[0086] P′_x(λ, {right arrow over (τ)})≡The “Spatial Fourth-Order Cumulant Matrix-Pencil” using a pair of Spatial Fourth-Order Cumulant Matrix 2's.

[0087] P″_x(λ, {right arrow over (τ)})≡The “Spatial Fourth-Order Cumulant Matrix-Pencil” using a pair of Spatial Fourth-Order Cumulant Matrix 3's.

[0088] P_r(λ, {right arrow over (τ)})≡The “Fourth-Order Signal Cumulant Matrix-Pencil” using a pair of Diagonal Fourth-Order Signal Cumulant Matrices.

[0089] λ(A, B)≡The “Spectrum” of the pencil defined on the matrices A and B.

[0090] The set of generalized eigenvalues.

[0091] {circumflex over (λ)}(A, B)≡The “Finite Spectrum” of the pencil defined on the matrices A and B.

[0092] The set of non-zero finite generalized eigenvalues.

[0093] λ_j≡The “j^thEigenvalue” of the pencil defined on a pair of spatial fourth-order cumulant matrices. There are M such eigenvalues, counting multiplicities. λ_jtakes on one of the K values of μ_k.

[0094] μ_k≡The “k^thDistinct Eigenvalue” of the pencil defined on a pair of spatial fourth-order cumulant matrices. There are K such values that the set of λ_j's takes on.

[0095] g_k≡The set of indeices, {j}, where λ_j=μ_k.

[0096] ê_j≡The N×1 “j^thEigenvector” of the pencil defined on a pair of spatial fourth-order cumulant matrices associated with the eigenvalue λ_j.

[0097] ε_j≡ê_j^Hv_j.

[0098] γ_j≡The “Normalization Factor” for the j^theigenvector. ${\gamma }_j=\frac{1}{\left|{ϵ}_j\right|}.$

[0099] η_k^geom≡The “Geometric” Multiplicity of an Eigenvalue.

[0100] η_k^alg≡The “Algebraic” Multiplicity of an Eigenvalue.

[0101] η_k≡The “Multiplicity” of an Eigenvalue when η_k^geom=η_k^alg.

[0102] FIG. 1 is a functional block diagram of a system 100 for performing blind source separation utilizing a spatial fourth order cumulant matrix pencil in accordance with an embodiment of the present invention. System 100 comprises a receiver 11 and a signal processor 12. The receiver 11 receives signal s(t), which is indicative of a plurality of signals provided by a respective plurality of sources and provides signal x(t) to the signal processor 12. The receiver 11 may be any appropriate receive configured to receive the signal s(t). For example, the signal s(t) may be an acoustic signal, an optical signal, a seismic signal, an electromagnetic signal, or a combination thereof, and the receiver 11 may be configured to receive the respective type of signal. In one embodiment, the receiver 11 is configured as an array having a plurality of elements. The signal s(t) is received and appropriately processed (e.g., time delayed and multiplexed) and provided to the signal processor 14 in the form of signal x(t).

[0103] The signal processor 12 may be any appropriate processor configured to processor the signal x(t), such a general purpose computer, a laptop computer, a special purpose computer, a hardware implemented processor, or a combination thereof. The signal x(t) may be in any appropriate format, such as an optical signal, and electromagnetic signal, a digital signal, and analog signal, or a combination thereof. As will be explained in more detail below, the signal processor 12 comprises a matrix pencil estimation portion 13, a non-zero finite eigenvalue determination portion 14, a number of distinct eigenvalues determination portion 15, a multiplicity determination portion 16, a linearly independent eigenvector calculation portion 17, a normalization factor calculation 18, a separation vector generation portion 19, a separation matrix generation portion 20, and an optional separation power efficiency calculation portion 21. The matrix pencil estimation portion 13 is configured to estimate the spatial fourth order cumulant matrix pencil as a function of time differences of the arrival of the signal s(t) at the elements of the receiver 11. The non-zero finite eigenvalue determination portion 14 is configured to determine the non-zero finite eigenvalues for the spatial fourth order cumulant matrix pencil. The number of distinct eigenvalues determination portion 15 is configured to determine the number of eigenvalues that are distinct. The multiplicity determination portion 16 is configured to determine the multiplicity of each of the distinct finite eigenvalues. The linearly independent eigenvector calculation portion 17 is configured to calculate linearly independent eigenvectors for each of the distinct finite eigenvalues. The normalization factor portion 18 is configured to calculate, for each eigenvalue having a multiplicity equal to one, a normnalization factor and to generate a respective separation vector as a function of the normalization factor and an eigenvector corresponding to the eigenvalue having a multiplicity equal to one. The separation vector generation portion 19 is configured to generate, for each repeated eigenvalue, a separation vector as a function of an eigenvector corresponding to the repeated eigenvalue. The separation matrix generation portion 20 is configured to generate the separation matrix as a function of the separation vectors. The optional separation power efficiency calculation portion 21 is configured to calculate the efficiency of the separation process in accordance with the following formula: ζ_j≡S_j/P_j, wherein ζ_jis indicative of the separation power efficiency for the j^thsource of the plurality of sources, S_jis indicative of a power of a separated signal from the j^thsource, and P_jis indicative of a normalized power of a signal from the j^thsource.

[0104] FIG. 2 is an illustration of signal source 24, array elements 26, and a processor 22 for performing array signal processing and BSS processing in accordance with an embodiment of the present invention. Array signal processing is a specialization within signal processing concerned with the processing of a set of signals generated by an array of sensors at distinct spatial locations sampling propagating wavefields, such as electromagnetic, seismic, acoustic, optical, mechanical, thermal, or a combination thereof, for example. As shown in FIG. 2, the array samples the j^thwavefield, r_j(t, {right arrow over (z)}_i), generated by the j^thsource 24_jat locations {{right arrow over (z)}₁, {right arrow over (z)}₂, . . . , {right arrow over (z)}_N} (only one location, z_j, shown in FIG. 2) with a set of sensors 26_iwhich generate signals x_i(t) indicative of the wavefield at each location, z_j. The signals x_i(t) may be any appropriate type of signal capable of being processed by the processor 22. Examples of appropriated types of signals x_i(t) include electrical signals, acoustic signals, optical signals, mechanical signals, thermal signals, or a combination thereof. The signal x_i(t) provided by the i^thsensor, 26_i, comprises the sum of the wavefields from all sources 24 at each sensor's location, each weighted with response of the sensor in the signal's r_j(t, {right arrow over (z)}_i) direction of arrival, plus an additive noise term, n_i(t). As described in more detail herein, the processor 22 processes the signals x(t) for enhancing sets of sources signals' individual signal-to-interference-plus-noise ratios by suppressing interfering source signals at different spatial locations without knowledge of the source signal characteristics, the channels between the sources and the array elements, the sources' locations, or array geometry via a blind source separation (BSS) technique in accordance with the present invention.

[0105] A blind source separation technique in accordance with the present invention is described herein by defining underlying assumptions made about the source signals and noise sources. Different multiple input multiple output (MIMO) array channel models are described resulting in a narrowband model, which is utilized in the BSS technique in accordance with present invention.

[0106] Blind source separation (BSS) is applicable to many areas of array signal processing that require the enhancement and characterization of an unknown set of source signals generated by a set of sensors that are each a linear mixture of the original signals. These include, for example, signal intelligence, spectral monitoring, jamming suppression, and interference rejection, location, and recognition. Typically, the mixing transformation, source signal characteristics, and sensor array manifold are unknown. Thus, blind source separation may be viewed as a multiple-input, multiple-output (MIMO) blind channel estimation problem.

[0107] FIG. 3 is an illustration of a MIMO blind channel estimation scenario showing five unknown sources, S₁, S₂, S₃, S₄, S₅, having distinct radiating patterns and five sensors, x₁, x₂, x₃, x₄, x₅, having distinct receiving patterns. The sources, s₁, s₂, s₃, S₄, s₅, may provide and the sensors, x₁, x₂, x₃, x₄, x₅, may correspondingly receive, acoustic energy, electromagnetic energy, optic energy, mechanical energy, thermal energy, or a combination thereof. As shown in FIG. 3, the five unknown sources, s₁, s₂, s₃, s₄, s₅, with distinct radiating patterns are generating a set of wavefields that are impinging on an array of five sensors, x₁, x₂, x₃, x₄, x₅, with an unknown array manifold. Each source, s₁, s₂, s₃, s₄, s₅, provides a respective source signal. A BSS separation technique in accordance with the present invention, jointly extracts the set of source signals from an array of sensors (e.g., x₁, x₂, x₃, x₄, x₅,) sampling the aggregate (composite) of the source signals' propagating wavefields at distinct spatial locations without knowledge of the signal characteristics or knowledge of the array's sensitivity as a function of direction of arrival or geometry.

[0108] In order to develop a blind source separation technique suitable for separating narrowband signals given a set of outputs from an array of sensors with a relatively small spatial expanse and assess its performance, it is advantageous to develop a multiple-input multiple-output (MIMO) narrowband channel model for the array, state assumptions made, state the problem mathematically, and develop a set of measures to evaluate the technique.

[0109] As such, a narrowband MIMO channel model is developed by starting with the most general convolutional MIMO channel model and then placing restrictions on the signal bandwidth and array size to simplify the problem, resulting in the narrowband model as utilized herein. Signal and noise assumptions are then presented and the blind source separation technique in accordance with the present invention is described mathematically and graphically. Two performance measures to be used in assessing the performance are then described including the novel concept of separation power efficiency (SPE).

[0110] Four multiple-input multiple-output (MIMO) channel models applicable to the blind source separation problem are described herein. These models are the general channel model, the non-dispersive direct path only channel model, the general finite impulse response (GFIR) channel model, and the narrowband channel model. The BSS technique in accordance with the present invention is then described utilizing the narrowband channel model.

[0111] The General Channel Model: In the most general case, the output of each element is modeled as a summation of the M source signals each convolved with the impulse response of the channel between the output of the source and output of the sensor plus the additive Gaussian noise referenced to the sensors input. That is, $\begin{array}{cc}{x}_i\left(t\right)=\sum _{j=1}^M{v}_{\mathrm{ij}}\left(t\right)*s_j\left(t\right)+n_i\left(t\right)& \left(1\right)\end{array}$

[0112] where * denotes convolution. The impulse response, v_ij(t), of the channel between the output of the j^thsource and the i^thsensor output may be time varying and account for such phenomena as multi-path propagation, dispersion, sensor time-varying response, source motion, sensor motion, etc. This can be written in matrix form as the general multiple input multiple output (MIMO) channel model $\begin{array}{cc}\begin{array}{c}x\left(t\right)={\left[x_1\left(t\right)x_2\left(t\right)\dots x_N\left(t\right)\right]}^T\\ =\left[\begin{array}{ccc}{v}_{11}\left(t\right)& \cdots & v_{1M}\left(t\right)\\ ⋮& ⋰& ⋮\\ v_{\mathrm{N1}}\left(t\right)& \cdots & v_{\mathrm{NM}}\left(t\right)\end{array}\right]*\left[\begin{array}{c}{s}_1\left(t\right)\\ ⋮\\ s_M\left(t\right)\end{array}\right]+\left[\begin{array}{c}{n}_1\left(t\right)\\ ⋮\\ n_N\left(t\right)\end{array}\right]\\ =V\left(t\right)*s\left(t\right)+n\left(t\right)\end{array}& \left(2\right)\end{array}$

[0113] where [ ]^Tdenotes transposition.

[0114] The Non-Dispersive, Direct Path Only Channel Model: When there is no multi-path, motion, or dispersion, the channel impulse response can be modeled by a delay and attenuation. That is,

[0115] ν_ij(t)=α_ijδ(t−τ_ij)  (3)

[0116] where α_ijis the cascaded attenuation/gain from the output of j^thsource to the i^thsensor output and τ_ijis the propagation time (delay) from the output of j^thsource to the output of the i^thsensor. Under this model, when the sifting property of the delta function is employed, the output of the i^thsensor (ignoring the noise) becomes $\begin{array}{cc}\begin{array}{c}{x}_i\left(t\right)=\sum _{j=1}^M{v}_{\mathrm{ij}}\left(t\right)*s_j\left(t\right)\\ =\sum _{j=1}^M{\alpha }_{\mathrm{ij}}\delta \left(t-{\tau }_{\mathrm{ij}}\right)*s_j\left(t\right)\\ =\sum _{j=1}^M{\alpha }_{\mathrm{ij}}{s}_j\left(t-{\tau }_{\mathrm{ij}}\right)\end{array}& \left(4\right)\end{array}$

[0117] At this point a “differential” delay is defined as the difference in propagation time from the output of the j^thsource to the output of the k^thsensor and to the output of the l^thsensor.

[0118] Δτ_l,k,j≡τ_lj−τ_kj  (5)

[0119] This differential time delay defines the time difference of arrival between two sensors for a given signal and is a measure of the spatial expanse of the array of sensors. Additionally, to facilitate situations when the minimum propagation delay from the j^thsource to the sensors is much greater than the maximum differential propagation delay, that is $\underset{i}{\min }\left({\tau }_{\mathrm{ij}}\right)|⪢\underset{l,k}{\max }\left|\Delta {\tau }_{l,k,j}\right|,$

[0120] the propagation time τ_ijis decomposed into two components, a “reference” delay, which is defined as the average propagation time from the output of the source to the output of the sensors and denoted as {overscore (τ)}_j, and a “relative” delay, which is defined as the difference in propagation time between the reference time delay and the actual propagation time and denoted as Δτ_ij. The propagation time from the j^thsource to the i^thsensor can then be expressed as

[0121] τ_ij={overscore (τ)}_j+Δτ_ij.  (6)

[0122] FIG. 4 is a graphical illustration of time delays between sensors and sources. The decomposition of the propagation time as depicted in FIG. 4 includes five sources, labeled s₁, s₂, . . . , s₅, with associated reference delays {overscore (τ)}₁, {overscore (τ)}₂, . . . , {overscore (τ)}₅, which are generating a set of wavefields that illuminate a set of five sensors, labeled x₁, x₂, . . . , x₅, and the relative time delay, Δτ₃₁, is shown for the first source, s₁, and the third sensor, x₃. Using the above definitions, the differential time delay can be reformulated as follows: $\begin{array}{cc}\begin{array}{c}{\mathrm{\Delta \tau }}_{l,k,j}\equiv {\tau }_{\mathrm{lj}}-{\tau }_{\mathrm{kj}}\\ =\left({\stackrel{\_}{\tau }}_j+{\mathrm{\Delta \tau }}_{\mathrm{lj}}\right)-\left({\stackrel{\_}{\tau }}_j+{\mathrm{\Delta \tau }}_{\mathrm{kj}}\right)\\ ={\mathrm{\Delta \tau }}_{\mathrm{lj}}-{\mathrm{\Delta \tau }}_{\mathrm{kj}}\end{array}& \left(7\right)\end{array}$

[0123] Both the differential and relative time delays are utilized in the formulation of the narrowband and the general finite impulse response models.

[0124] The General Finite Impulse Response (GFIR) Channel Model: The general model is often simplified by modeling the channel between the output of the j^thsource and the i^thsensor output, ν_ij(t), as a FIR filter or tapped delay line. As with the general model, the GFIR Model may be time varying and can account for such phenomena as multi-path propagation, dispersion, sensor time-varying response, system motion, etc. The FIR filter used to model ν_ij(t) must be long enough to account for the multi-path delay spread of the channel as well as the relative time delay, Δτ_ij, with a “reference” delay, {overscore (τ)}_j, accounted for by defining a delayed version of the source signal as it's input. That is the input to the set of FIR filters used to model the channels between the output of the j^thsource and array of sensors is

[0125] r_j(t)=s_j(t−{overscore (τ)}_j)  (8)

[0126] The FIR filter or tapped delay line model is valid for a fading channel when the coherence bandwidth of such a channel is much less than the noise equivalent bandwidth of the source signal, that is BW_NEq[s_j(t)]<BW_ij^COH, where the coherence bandwidth is defined as the reciprocal of the multi-path delay spread. In this situation the multi-path components in the channel separated by a delay of at least 2π/BW_NEq[s_j(t)] are resolvable and the fading phenomenon is referred to as being “frequency selective”. Thus the channel impulse response can be represented as $\begin{array}{cc}{v}_{\mathrm{ij}}\left(t\right)=\sum _{l=0}^{L_{\mathrm{ij}}-1}{v}_{\mathrm{ij}}^{\left(l\right)}\left(t\right)\delta \left(t-2\pi l/{\mathrm{BW}}_{\mathrm{NEq}}\left[s_j\left(t\right)\right]\right)& \left(9\right)\end{array}$

[0127] where the time varying complex-valued channel gain of the l^thcomponent can be represented as

[0128] ν_ij^(l)(t)=α_ij^(l)(t)e^jφ(t).  (10)

[0129] The length of the model, L_ij, is the number of resolvable multi-path components which is

[0130] L_ij=┌BW_NEq[s_j(t)]/BW_ij^COH┐  (11)

[0131] where ┌ ┐ denotes the ceiling function. For the GFIR channel model, the length of the FIR filter has to not only accommodate the multi-path delay spread but also the relative time delay, Δτ_ij. That is equation (11) becomes

[0132] L_ij=┌BW_NEq[s_j(t)]·[(|Δτ_ij|/2π)+(1/Bw_ij^COH)]┐.  (12)

[0133] In practice, the length of all the FIR filters are set to a common value, L, which is defined as $\begin{array}{cc}L=\underset{i,j}{\max }\left(L_{\mathrm{ij}}\right).& \left(13\right)\end{array}$

[0134] When the coherence bandwidth is greater than the noise equivalent bandwidth of the source signal, that is BW_NEq[s_j(t)]<BW_ij^COH, the fading is referred to as “frequency non-selective” and the fading model reduces to a single time varying complex weight. That is L_ij=1, and thus $\begin{array}{cc}\begin{array}{c}{v}_{\mathrm{ij}}\left(t\right)=v_{\mathrm{ij}}^{\left(0\right)}\left(t\right)\\ ={\alpha }_{\mathrm{ij}}^{\left(0\right)}\left(t\right)e^{{\mathrm{j\phi }}_{\mathrm{ij}}^{\left(0\right)}\left(t\right)}\\ ={\alpha }_{\mathrm{ij}}\left(t\right)e^{{\mathrm{j\phi }}_{\mathrm{ij}}\left(t\right)}\end{array}& \left(14\right)\end{array}$

[0135] which begins to look like a time-varying narrowband model. However, for the above simplification to a single complex weight to hold in array signal processing, the source signal must have a noise equivalent bandwidth much less then the center frequency and the array of sensors must have a relatively small spatial expanse, that is $\begin{array}{cc}{\mathrm{BW}}_{\mathrm{NEq}}\left[s_j\left(t\right)\right]⪡{\omega }_j& \left(15\right)\\ \underset{i}{\max }\left|{\mathrm{\Delta \tau }}_{\mathrm{ij}}\right|⪡\pi /{\mathrm{BW}}_{\mathrm{NEq}}\left[s_j\left(t\right)\right].& \left(16\right)\end{array}$

[0136] The Narrowband Channel Model: A measure of the spectral support of a signal is the noise equivalent bandwidth, denoted as BW_NEq[ ]. By the duality principle of time and frequency, the inverse noise equivalent bandwidth can be used as a measure of the temporal support of the signal, in other words it is can be used as an indication of the decorrelation time of the signal. When the signal noise equivalent bandwidth is much less then the center frequency, that is

[0137] BW_NEq[s_j(t)]<<ω_j  (17)

[0138] where ω_jis the center frequency of the j^thsource, then the propagation delay, or relative propagation delay, can be modeled as a phase shift. In this situation, when there is no dispersion or multi-path, the channel model is referred to as the narrowband model.

[0139] However, since the phase shift is modulo 2π with respect to the center frequency, the requirement that the bandwidth be much less than the center frequency is itself insufficient in order for the time delay to be modeled as a phase shift and preserve the waveform, i.e. negligible inter-symbol interference (ISI) is induced in a digital communications signal. Therefore, for the narrowband model to hold, the array of sensors must also have a relatively small spatial expanse. That is $\begin{array}{cc}\underset{l,k}{\max }\left|{\mathrm{\Delta \tau }}_{l,k,j}\right|⪡2\pi /{\mathrm{BW}}_{\mathrm{NEq}}\left[s_j\left(t\right)\right]& \left(18\right)\\ \mathrm{Since}\Delta {\tau }_{l,k,j}={\mathrm{\Delta \tau }}_{\mathrm{lj}}-{\mathrm{\Delta \tau }}_{\mathrm{kj}},\mathrm{requiring}\underset{i}{\max }\left|{\mathrm{\Delta \tau }}_{\mathrm{ij}}\right|⪡\pi /{\mathrm{BW}}_{\mathrm{NEq}}\left[s_j\left(t\right)\right]& \left(19\right)\end{array}$

[0140] is a sufficient condition to guarantee (18) holds, via the triangle inequality. When the Narrowband conditions defined in (17) and (19) hold, the relative time delay is negligible in comparison to the decorrelation time of the signal and thus

[0141] s_j(t−{overscore (τ)}_j−Δτ_ij)≅s_j(t−{overscore (τ)}_j)  (20)

[0142] which says the waveform is preserved (within a phase shift). Thus, the relative time delay can be modeled as a phase shift, $\begin{array}{cc}\begin{array}{c}{v}_{\mathrm{ij}}\left(t\right)*s_j\left(t\right)={\alpha }_{\mathrm{ij}}\delta \left(t-{\tau }_{\mathrm{ij}}\right)*s_j\left(t\right)\\ ={\alpha }_{\mathrm{ij}}{s}_j\left(t-{\tau }_{\mathrm{ij}}\right)\\ ={\alpha }_{\mathrm{ij}}{s}_j\left(t-{\stackrel{\_}{\tau }}_j-{\mathrm{\Delta \tau }}_{\mathrm{ij}}\right)\\ ={\alpha }_{\mathrm{ij}}{e}^{-{\mathrm{j\omega }}_j{\mathrm{\Delta \tau }}_{\mathrm{ij}}}{s}_j\left(t-{\stackrel{\_}{\tau }}_j\right)\\ ={\alpha }_{\mathrm{ij}}{e}^{-{\mathrm{j\phi }}_{\mathrm{ij}}}{s}_j\left(t-{\stackrel{\_}{\tau }}_j\right)\equiv v_{\mathrm{ij}}{r}_j\left(t\right)\end{array}& \left(21\right)\end{array}$

[0143] where r_j(t)=s_j(t−{overscore (τ)}_j), φ_ij=ω_jΔτ_ij, and a complex weight, ν_ij, is defined as

[0144] ν_ij=α_ije^−jφ.  (22)

[0145] This complex weight together with the other N−1 weights associated with the j^thsignal form the j^thsteering vector.

[0146] v_j=[ν_1jν_2j. . . ν_Nj]^T  (23)

[0147] The output of the i^thsensor is then $\begin{array}{cc}{x}_i\left(t\right)=\sum _{j=1}^M{v}_{\mathrm{ij}}{r}_j\left(t\right)+n_i\left(t\right)& \left(24\right)\end{array}$

[0148] As done for the general m j^thmodel, this can be re-formulated in matrix form for the vector of sensor outputs as $\begin{array}{cc}\begin{array}{c}x\left(t\right)={\left[x_1\left(t\right)x_2\left(t\right)\dots x_N\left(t\right)\right]}^T\\ =\left[\begin{array}{ccc}{v}_{11}& \cdots & v_{1M}\\ ⋮& ⋰& ⋮\\ v_{\mathrm{N1}}& \cdots & v_{\mathrm{NM}}\end{array}\right]\left[\begin{array}{c}{r}_1\left(t\right)\\ ⋮\\ r_M\left(t\right)\end{array}\right]+\left[\begin{array}{c}{n}_1\left(t\right)\\ ⋮\\ n_N\left(t\right)\end{array}\right]\\ =\left[v_1\dots v_M\right]\left[\begin{array}{c}{r}_1\left(t\right)\\ ⋮\\ r_M\left(t\right)\end{array}\right]+\left[\begin{array}{c}{n}_1\left(t\right)\\ ⋮\\ n_N\left(t\right)\end{array}\right]\\ =\mathrm{Vr}\left(t\right)+n\left(t\right)\end{array}& \left(25\right)\end{array}$

[0149] Due to conservation of energy, the total average signal power from the j^thsource illuminating the array can never exceed P_j. Since the signal-to-noise ratio is established at the input of the sensor, in the total array gain can be viewed as being normalized. Thus for the narrowband model, the inner product of the j^thcolumn of the mixing matrix V is, $\begin{array}{cc}\begin{array}{c}{v}_j^H{v}_j=\sum _{i=1}^N{v}_{\mathrm{ij}}^{*}{v}_{\mathrm{ij}}\\ =\sum _{i=1}^N{\alpha }_{\mathrm{ij}}^2{e}^{+{\mathrm{j\phi }}_{\mathrm{ij}}}{e}^{-{\mathrm{j\phi }}_{\mathrm{ij}}}\\ =\sum _{i=1}^N{\alpha }_{\mathrm{ij}}^2\\ =1\end{array}& \left(26\right)\end{array}$

[0150] where [ ]^Hdenotes the Hermitian transpose.

[0151] Signal and Noise Assumptions: The following assumptions are made about the source signals and noise vector. These assumptions are made to allow the use of the fourth-order cumulant and to ensure a sufficient number of degrees of freedom for the separation technique to exploit. Assumptions A1 and A2 ensure that the fourth-order cumulant of the source signals exist. The zero-mean assumption is not necessary for the use of cumulants but is assumed since practical propagating electromagnetic signals have a zero-mean. Assumptions A3 and A4 are particularly useful to the use of cumulants for the BSS problem. Without them, the noise sources would need to be treated as a signal source thus requiring additional degrees of freedom in the array. Note that the noise sources are not assumed to be temporally or spatially white. This is contrast to assumptions made in other second order techniques. The final assumption on the number of source signals helps to ensure that the there are enough degrees of freedom to perform the separation using the matrix-pencil approach.

[0152] The first assumption (A1): The M source signals illuminating the array are statistically independent non-Gaussian stationary random processes. Assumption A1 is represented mathematically as follows. $\begin{array}{cc}{f}_{r_1,r_2,\dots ,r_M}\left(r_1,r_2\dots ,r_M\right)=\prod _{j=1}^M{f}_{r_j}\left(r_j\right)& \left(27\right)\end{array}$

[0153] The source signals are assumed stationary to order four over the estimation period of the spatial fourth-order cumulant matrix.

[0154] The second assumption (A2): The M source signals illuminating the array have zero mean with power P^jand a non-zero fourth-order moment. Assumption A2 is represented mathematically as follows.

[0155] E[m_j(t)]=0  (28)

[0156] E[r_j(t)]=E[{square root over (Pm_j(t)]=0  (29)

[0157] E[m_j(t)m*_j(t)]=1  (30)

[0158] E[r_j(t)r*_j(t)]=E[{square root over (Pt){square root over (P_j)}m*_j(t)]=P_jE[m_j(t)m*_j(t)]=P_j  (31)

[0159] E[r_j(t)r*_j(t)r_j(t)r*_j(t)]=P_j²E[m_j(t)m*_j(t)m_j(t)m*_j(t)]  (32)

[0160] E[m_j(t)m*_j(t)m_j(t)m*_j(t)]≠0.  (33)

[0161] The third assumption (A3): The set of source signals (processes) and set of noise processes are statistically independent. Assumption A3 is represented mathematically as follows.

[0162] $\begin{array}{cc}{f}_{r_1,r_2,\dots ,r_M,n_1,n_2,\dots ,n_N}\left(r_1,r_2,\dots ,r_M,n_1,n_2,\dots ,n_N\right)=f_{n_1,n_2,\dots ,n_N}\left(n_1,n_2,\dots ,n_N\right)\prod _{j=1}^M{f}_{r_j}\left(r_j\right)& \left(34\right)\end{array}$

[0163] The fourth assumption (A4): The noise processes are stationary zero-mean Gaussian random processes. They are not assumed to be spatially or temporally independent. Assumption A3 is mathematically represented as follows.

[0164] n_i(t)□N(0,σ_i²)  (35)

[0165] n(t)≡[n₁(t),n₂(t), . . . , n_N(t)]^T□N(0,K_n)  (36)

[0166] $\begin{array}{cc}{f}_{n_1,n_2,\dots ,n_N}\left(n_1,n_2,\dots ,n_N\right)=\frac{1}{{\left(2\pi \right)}^{1/2}{\left|\det K_n\right|}^{1/2}}{e}^{-n^T{K}_n^{-1}n}& \left(37\right)\end{array}$

[0167] The fifth assumption (A5): The number of sources is less than or equal to the number of sensors, i.e. M≦N.

[0168] FIG. 5 is a functional block diagram of an apparatus for performing blind source separation (BSS) of M statistically independent narrowband source signals given a set of outputs from an array of sensors with a relatively small spatial expanse and with an arbitrary and unknown array geometry, in accordance with an embodiment of the present invention. The BSS technique as described herein determines a separation matrix W that will diagonalize the mixing matrix V. This involves finding a N×M separation matrix W, with complex elements w_ij, $\begin{array}{cc}W=\left[\begin{array}{ccc}{w}_{11}& \cdots & w_{1M}\\ ⋮& ⋰& ⋮\\ w_{\mathrm{N1}}& \cdots & w_{\mathrm{NM}}\end{array}\right]& \left(38\right)\end{array}$

[0169] that will diagonalize the mixing matrix, V. That is, a separation matrix W is desired such that the product W^HV results in a M×M diagonal “loss” matrix with elements ρ_j. $\begin{array}{cc}{W}^{H}V=\left[\begin{array}{cccc}{\rho }_1& 0& \cdots & 0\\ 0& {\rho }_2& ⋰& ⋮\\ ⋮& ⋰& ⋰& 0\\ 0& \cdots & 0& {\rho }_M\end{array}\right]& \left(39\right)\end{array}$

[0170] When the separation matrix, W, is applied to the vector of sensor outputs, the result is $\begin{array}{cc}\begin{array}{c}y\left(t\right)=W^{H}x\left(t\right)=W^H\left\{\mathrm{Vr}\left(t\right)+n\left(t\right)\right\}\\ =W^H\mathrm{Vr}\left(t\right)+W^{H}n\left(t\right)\\ =\left[\begin{array}{cccc}{\rho }_1& 0& \cdots & 0\\ 0& {\rho }_2& ⋰& ⋮\\ ⋮& ⋰& ⋰& 0\\ 0& \cdots & 0& {\rho }_M\end{array}\right]\left[\begin{array}{c}{r}_1\left(t\right)\\ ⋮\\ r_M\left(t\right)\end{array}\right]+W^{H}n\left(t\right)\\ =\left[\begin{array}{c}{\rho }_1{r}_1\left(t\right)\\ ⋮\\ {\rho }_M{r}_M\left(t\right)\end{array}\right]+W^{H}n\left(t\right)\end{array}& \left(40\right)\end{array}$

[0171] and the source signals are separated. For mathematical completeness. note that the vector r(t)εC^M, the vectors x(t),n(t)εC^N, and the matrices V,WεC^N×M. If the loss matrix is the identity matrix, the separation process has captured all of the signal energy illuminating the array thus guaranteeing that the separated output signal has reached the maximum achievable signal to interference plus noise ratio.

[0172] As developed previously in the narrowband model, the output of each sensor is a weighted linear mixture of the independent source signals plus noise. $\begin{array}{cc}{x}_i\left(t\right)=\sum _{l=1}^M{v}_{\mathrm{il}}{r}_l\left(t\right)+n_i\left(t\right)& \left(41\right)\end{array}$

[0173] Applying the separation matrix to the vector of sensor outputs separates the sources. $\begin{array}{cc}\begin{array}{c}y\left(t\right)=W^{H}x\left(t\right)\\ =W^H\left\{\mathrm{Vr}\left(t\right)+n\left(t\right)\right\}\\ =W^H\mathrm{Vr}\left(t\right)+W^{H}n\left(t\right)\end{array}& \left(42\right)\end{array}$

[0174] The j^thelement of the separation process output vector, y_j(t), is an estimate of the j^thsource signal, r_j(t) and is the inner product of the j^thcolumn of the separation matrix and the vector of sensor outputs. $\begin{array}{cc}\begin{array}{c}{y}_j\left(t\right)=w_j^{H}x\left(t\right)\\ =\sum _{i=1}^N{w}_{\mathrm{ij}}^{*}{x}_i\left(t\right)\end{array}& \left(43\right)\end{array}$

[0175] Substituting equation (41) into equation (43) yields $\begin{array}{cc}\begin{array}{c}{y}_j\left(t\right)=\sum _{i=1}^N{w}_{\mathrm{ij}}^{*}\left\{\sum _{l=1}^M{v}_{\mathrm{il}}{r}_l\left(t\right)+n_i\left(t\right)\right\}\\ =\sum _{i=1}^N{w}_{\mathrm{ij}}^{*}\sum _{l=1}^M{v}_{\mathrm{il}}{r}_l\left(t\right)+\sum _{i=1}^N{w}_{\mathrm{ij}}^{*}{n}_i\left(t\right)\\ =\sum _{i=1}^N{w}_{\mathrm{ij}}^{*}{v}_{\mathrm{ij}}{r}_j\left(t\right)+\sum _{i=1}^N\sum _{\underset{l\ne j}{i=1}}^M{w}_{\mathrm{ij}}^{*}{v}_{\mathrm{il}}{r}_l\left(t\right)+\sum _{i=1}^N{w}_{\mathrm{ij}}^{*}{n}_i\left(t\right)\end{array}& \left(44\right)\end{array}$

[0176] where it is clear there are three distinct terms corresponding to the desired signal, the residual interference, and the output noise. Of particular interest in evaluating the performance of communications and signal intelligence systems is the second-order moment of each of these terms. The second-order moment of the first term is the desired signal output power and is defined as $\begin{array}{cc}\begin{array}{c}{S}_j=E\left[{\left|\sum _{i=1}^N{w}_{\mathrm{ij}}^{*}{v}_{\mathrm{ij}}{r}_j\left(t\right)\right|}^2\right]\\ =E\left[\sum _{i=1}^N{w}_{\mathrm{ij}}^{*}{v}_{\mathrm{ij}}{r}_j\left(t\right)\right\}{\left\{\sum _{k=1}^N{w}_{\mathrm{kj}}^{*}{v}_{\mathrm{kj}}{r}_j\left(t\right)\right\}}^{*}]\\ =\sum _{i=1}^N\sum _{k=1}^N{w}_{\mathrm{ij}}^{*}{v}_{\mathrm{ij}}{v}_{\mathrm{kj}}^{*}{w}_{\mathrm{kj}}E\left[r_j\left(t\right)r_j^{*}\left(t\right)\right]\end{array}& \left(45\right)\end{array}$

[0177] Applying assumptions A1 and A2,

[0178] E[r_j(t)r*_j(t)]=P_j  (46)

[0179] and thus equation (45) becomes $\begin{array}{cc}{S}_j=P_j\sum _{i=1}^N\sum _{k=1}^N{w}_{\mathrm{ij}}^{*}{v}_{\mathrm{ij}}{v}_{\mathrm{kj}}^{*}{w}_{\mathrm{kj}}& \left(47\right)\end{array}$

[0180] which can be represented using vector notation as

[0181] S_j=P_jw_j^Hv_jv_j^Hw_j.  (48)

[0182] The second-order moment of the second term in (44) is the residual interference power and is given by $\begin{array}{cc}\begin{array}{c}{I}_j=E\left[{\left|\sum _{i=1}^N\sum _{\underset{l\ne j}{l=1}}^M{w}_{\mathrm{ij}}^{*}{v}_{\mathrm{ij}}{r}_l\left(t\right)\right|}^2\right]\\ =E\left[\left\{\sum _{i=1}^N\sum _{\underset{l\ne j}{l=1}}^M{w}_{\mathrm{ij}}^{*}{v}_{\mathrm{il}}{r}_l\left(t\right)\right\}{\left\{\sum _{k=1}^N\sum _{\underset{m\ne j}{m=1}}^M{w}_{\mathrm{kj}}^{*}{v}_{\mathrm{km}}{r}_m\left(t\right)\right\}}^{*}\right]\\ =\sum _{\underset{l\ne j}{l=1}}^M\sum _{\underset{m\ne j}{m=1}}^{M}E\left[r_l\left(t\right)r_m^{*}\left(t\right)\right]\sum _{i=1}^N\sum _{k=1}^N{w}_{\mathrm{ij}}^{*}{v}_{\mathrm{il}}{v}_{\mathrm{km}}^{*}{w}_{\mathrm{kj}}\end{array}& \left(49\right)\end{array}$

[0183] However, by assumption A1 the signals are statistically independent and therefore

[0184] E[r_l(t)r*_m(t)]=0, for m≠l.  (50)

[0185] Additionally, applying the stationarity of assumption A1 and assumption A2,

[0186] E[r_l(t)r*_l(t)]=P_l.  (51)

[0187] Using (50) and substituting equation (51) into equation (49), the residual interference power reduces to $\begin{array}{cc}{I}_j=\sum _{\underset{l\ne j}{l=1}}^M{P}_l\sum _{i=1}^N\sum _{k=1}^N{w}_{\mathrm{ij}}^{*}{v}_{\mathrm{il}}{v}_{\mathrm{kl}}^{*}{w}_{\mathrm{kj}}& \left(52\right)\end{array}$

[0188] which can be represented using vector notation as $\begin{array}{cc}{I}_j=\sum _{\underset{l\ne j}{l=1}}^M{P}_l{w}_j^H{v}_l{v}_l^H{w}_j.& \left(53\right)\end{array}$

[0189] The second-order moment of the third term in (44) is the output noise power and is given by $\begin{array}{cc}\begin{array}{c}{N}_j\equiv E\left[{\left|\sum _{i=1}^N{w}_{\mathrm{ij}}^{*}{n}_i\left(t\right)\right|}^2\right]\\ =E\left[\left\{\sum _{i=1}^N{w}_{\mathrm{ij}}^{*}{n}_i\left(t\right)\right\}{\left\{\sum _{k=1}^N{w}_{\mathrm{kj}}^{*}{n}_k\left(t\right)\right\}}^{*}\right]\\ =\sum _{i=1}^N\sum _{k=1}^N{w}_{\mathrm{ij}}^{*}E\left[n_i\left(t\right)n_k^{*}\left(t\right)\right]w_{\mathrm{kj}}\end{array}& \left(54\right)\end{array}$

[0190] which can be represented using vector notation as

[0191] N_j=w_j^HE[n(t)n^H(t)]w_j.  (55)

[0192] By definition and assumption A4, the expectation of the outer product of the noise vector is the noise covariance matrix,

[0193] E[n(t)n^H(t)]=K_n  (56)

[0194] and thus the output noise power is

[0195] N_j=w_j^HK_nw_j.  (57)

[0196] To evaluate the effectiveness of a blind source separation technique, a measure of the quality of the separation is utilized. As previously described, the blind source separation technique as described herein determines a separation matrix W that will diagonalize the mixing matrix V. Two measures to assess the quality of a blind source separation algorithm are developed herein. Performance of the BSS technique may be measured in terms of residual interference and in terms of the efficiency of the algorithm in “capturing” all available signal power illuminating the array of sensors.

[0197] One measure of the quality of separation is the amount of residual interference found in a signal output after the separation matrix has been applied. Specifically, the power of the residual interference relative to the desired signal in the estimate of the desired source averaged over all sources as well as the peak or maximum residual interference-to-signal ratio to assess the separation technique in terms of suppressing co-channel interference are proposed for use. This measure is of significance because, if the separation matrix does not perfectly diagonalize the mixing matrix, the off diagonal terms of the resultant matrix will permit residual interference in the signal outputs.

[0198] In most communications applications, the common measure of the amount of interference is the signal-to-interference ratio, which is the ratio of the desired signal power to the combined power of all interfering signals. However, as the goal of the blind source separation is to completely eliminate all interference, this ratio could become extremely large. As a result, the Interference-to-Signal ratio (ISR), which quantifies the residual power of the interference that a blind source separation algorithm or technique fails to suppress relative to a particular desired signal power, is proposed. The better an algorithm is the smaller this ratio will become.

[0199] The ISR of a particular desired signal is defined as $\begin{array}{cc}{\zeta }_j\equiv \frac{I_j}{S_j}.& \left(58\right)\end{array}$

[0200] Substituting (53) and (48) into (58), the ISR for a particular signal is $\begin{array}{cc}{\zeta }_j=\frac{\sum _{\underset{l\ne j}{l=1}}^M{P}_l{w}_j^H{v}_l{v}_l^H{w}_j}{P_j{w}_j^H{v}_l{v}_j^H{w}_j}.& \left(59\right)\end{array}$

[0201] This value is also known as the rejection rate.

[0202] The overall quality of separation of a blind source separation technique may be measured by looking at the mean value of the individual source signal ISR's, ζ_j, over all j.

[0203] Thus the primary measure to be used to evaluate the performance of a blind source separation algorithm in terms of residual interference will be the average ISR given by $\begin{array}{cc}{\mathrm{ISR}}_{\mathrm{avg}}\equiv \frac{1}{M}\sum _{j=1}^M{\zeta }_j.& \left(60\right)\end{array}$

[0204] The secondary measure in terms of residual interference will be the peak or maximum ISR, which is defined as $\begin{array}{cc}{\mathrm{ISR}}_{m\mathrm{ax}}\equiv \underset{j}{\max }\left[{\zeta }_j\right]& \left(61\right)\end{array}$

[0205] This secondary measure ensures that all source signals are effectively separated with an ISR no worst than ISR_max.

[0206] A second measure of the quality of separation is utilized to determine the efficiency of the source separation matrix in terms of its ability to make use of the available signal power. A BSS technique is considered more efficient if the output signal-to-interference-plus-noise ratio is maximized, thus having greater sensitivity in terms of being able to capture smaller signals, than a BSS technique not maximizing the output signal-to-interference-plus-noise ratio.

[0207] The efficiency of a blind source separation algorithm in using all of a source's signal power illuminating the array of sensors is yet another important measure of its quality of separation. This measure determines how much of the available signal power from a particular source is wasted or lost in the separation process. This loss results in a lower signal-to-noise-plus-interference ratio then would otherwise be theoretically achievable and thus a loss in system sensitivity. The Separation Power Efficiency (SPE) for a particular separation process output relative to the desired source signal's available normalized power is defined as $\begin{array}{cc}{\xi }_j\equiv \frac{S_j}{P_j},\mathrm{where}& \left(62\right)\end{array}$

[0208] ζ_jis indicative of the separation power efficiency for the j^thsource of the plurality of sources, S_jis indicative of a power of a separated signal from the j^thsource, and P_jis indicative of a normalize power of a signal from the j^thsource.

[0209] Substituting equation (48) in for the separation process output power reveals that the particular SPE $\begin{array}{cc}\begin{array}{c}{\xi }_j=\frac{P_j{w}_j^H{v}_j{v}_j^H{w}_j}{P_j}\\ =w_j^H{v}_j{v}_j^H{w}_j\end{array}& \left(63\right)\end{array}$

[0210] depends only on the steering vector for the j^thsource and the j^thcolumn of the separation matrix. As with ISR, both the average SPE and the minimum SPE, defined as $\begin{array}{cc}{\xi }_{\mathrm{avg}}\equiv \frac{1}{M}\sum _{j=1}^M{\xi }_j& \left(64\right)\end{array}$

[0211] and $\begin{array}{cc}{\xi }_{\min }=\underset{j}{\min }\left[{\xi }_j\right]& \left(65\right)\end{array}$

[0212] respectively, will be used to evaluate the separation power efficiency.

[0213] Note that by the definition of the illuminating source signal power, P_j, that the maximum value the SPE can achieve is one. Thus the maximum achievable average SPE is also one. A separation algorithm that achieves an SPE of one is guaranteed to have maximized the source signal power in the corresponding separation process output. The minimum SPE provides a measure of ensuring that all sources are successfully separated with a minimum separation power efficiency.

[0214] A BSS technique in accordance with an embodiment of the present invention utilizes cumulants, specifically spatial fourth order cumulant matrices. To better understand the use of cumulants in performing blind source separation, a cumulant definition and associated properties are provided below.

[0215] The joint cumulant, also known as a semi-invariant, of order N of the set of random variables {s₁, s₂, . . . , s_N} is defined as the N^th-order coefficient of the Taylor series expansion about the origin of the second characteristic function. See, for example, C. L. Nikias and A. P. Petropulu, Higher-Order Spectra Analysis: A Non-Linear Signal Processing Framework. (PTR Prentice-Hall, Upper Saddle River, N.J.: 1993) and M. Rosenblatt, Stationary Sequences and Random Fields(Birkhauser, Boston, Mass.: 1985), which are hereby incorporated by reference in their entirety as if presented herein. The second characteristic function is defined as the natural logarithm of the characteristic function,

[0216] Ψ_s(ω₁,ω₂, . . . ,ω_N)≡ln[Φ_s(ω₁,ω₂, . . . ,ω_N)]  (66)

[0217] where the characteristic function is defined as

[0218] Φ_s(ω₁,ω₂, . . . ,ω_N)≡E[e^j(ωs^+ωs^{+ . . . +ωsN})].  (67)

[0219] The joint N^th-order cumulant is then $\begin{array}{cc}\mathrm{Cum}\left[s_1,s_2,\dots ,s_N\right]={\left(-j\right)}^N\frac{{\partial }^N{\Psi }_s\left({\omega }_1,{\omega }_2,\dots ,{\omega }_N\right)}{\partial {\omega }_1\partial {\omega }_2\dots \partial {\omega }_N}{|}_{{\omega }_1={\omega }_2=\dots ={\omega }_N=0}.& \left(68\right)\end{array}$

[0220] Cumulants, unlike moments, cannot be directly estimated from the data. See, for example, A.K. Nandi, Blind Estimation Using Higher-Order Statistics(Kluwer Academic, Dordecht, The Netherlands: 1999), which is hereby incorporated by reference in its entirety as if presented herein. However, cumulants can be found through their relationship to moments and can thus be estimated indirectly by first estimating the required moments. The relationship of cumulants to moments is described in M. Rosenblatt, Stationary Sequences and Random Fields(Birkhauser, Boston, Mass.: 1985) for the N^th-order joint cumulant of the set of random variables {s₁,s₂, . . . ,s_N} as $\begin{array}{cc}\mathrm{Cum}\left[s_1,s_2,\dots ,s_N\right]=\sum _{p=1}^N{\left(-1\right)}^{p-1}\left(p-1\right)!\left\{\sum _{n=1}^{N\left(p\right)}\prod _{l=1}^{p}E\left[\prod _{i\in g_{l,p,n}}{s}_i\right]\right\}& \left(69\right)\end{array}$

[0221] where there are N (p) ways of partitioning the set of integers {1, 2, . . . , N} into p groups, each denoted as g_l,p,n, such that $\begin{array}{cc}\bigcap _{l=1}^p{g}_{l,p,n}=\left\{\varnothing \right\}\bigcup _{l=1}^p{g}_{l,p,n}=\left\{1,2,\dots ,N\right\}& \left(70\right)\end{array}$

[0222] As an example, for the case N=4, the partitioning is defined on the set of integers {1,2,3,4} and is given in Table 1.0 below.

[0223] The 4^th-order joint cumulant as a function of the moments is then $\begin{array}{cc}\mathrm{Cum}\left[s_1,s_2,s_3,s_4\right]=E\left[s_1{s}_2{s}_3{s}_4\right]-E\left[s_1\right]·E\left[s_2{s}_3{s}_4\right]-E\left[s_2\right]·E\left[s_1{s}_3{s}_4\right]-E\left[s_3\right]·E\left[s_1{s}_2{s}_4\right]-E\left[s_4\right]·E\left[s_1{s}_2{s}_3\right]-E\left[s_1{s}_2\right]·E\left[s_3{s}_4\right]-E\left[s_1{s}_3\right]·E\left[s_2{s}_4\right]-E\left[s_1{s}_4\right]·E\left[s_2{s}_3\right]+2E\left[s_1{s}_2\right]·E\left[s_3\right]·E\left[s_4\right]+2E\left[s_1{s}_3\right]·E\left[s_2\right]·E\left[s_4\right]+2E\left[s_1{s}_4\right]·E\left[s_2\right]·E\left[s_3\right]+2E\left[s_2{s}_3\right]·E\left[s_1\right]·E\left[s_4\right]+2E\left[s_2{s}_4\right]·E\left[s_1\right]·E\left[s_3\right]+2E\left[s_3{s}_4\right]·E\left[s_1\right]·E\left[s_2\right]-6E\left[s_1\right]·E\left[s_2\right]·E\left[s_3\right]·E\left[s_4\right]& \left(71\right)\end{array}$

[0224] Note that equation (71) shows that computation of the N^th-order joint cumulant requires knowledge of all moments up to order N.

[0225] Cumulants possess several properties that make them attractive for use in the blind separation of a linear mixture of unknown statistically independent signals in spatially and/or temporally correlated Gaussian noise, especially at a low signal-to-noise ratio.

[0226] One property that makes cumulants attractive for use in blind source separation is that if the set of random variables {s₁, s₂, . . . , s_N} can be divided in to two or more groups that are statistically independent, then their N^th-order joint cumulant is zero. Thus, the cumulant operator in the blind separation of statistically independent sources will suppress all cross-source signal cumulant terms. In general, this is not the case for higher-order moments. Another property that makes cumulants attractive for use in BSS is that the Cum[s₁+n₁, s₂+n₂, . . . s_N+n_N]=Cum[s₁, s₂, . . . s_N]+Cum[n₁, n₂, . . . , n_N]. Because in general the set ofsignal terms {s₁, s₂, . . . , s_N} and the set of noise terms {n₁, n₂, . . . , n_N} are statistically independent from each other, the N^th-order joint cumulant of the terms of their vector sum, {s₁+n₁, s₂+n₂, . . . , s_N+n_N}, is the sum of their individual joint cumulants. Therefore, the cross cumulants between the noise terms and signal terms will be zero. This property is important in guaranteeing that the spatial fourth-order cumulant matrix can be decomposed into the sum of two matrices, one corresponding to the signals and the other corresponding to noise vector.

[0227] Yet another property that makes cumulants attractive for use in BSS is that the joint cumulant of order N>2 of a Gaussian random variable is zero. Because the noise vector is a multi-variate Gaussian random process, n=┌n₁, n₂, . . . , n_N┐^T˜N(μ_n, K_n), its joint cumulant of order three or higher will be zero. That is Cum[n₁, n₂, . . . , n_N]=0. This last property results in the spatial fourth-order cumulant matrix not having a noise subspace and the only non-zero elements of the matrix are associated with and only with the source signals. This is true even if the noise vector is spatially or temporally correlated.

[0228] Finally, cumulants of order higher than two preserve phase information that is lost by the use of second-order statistics, such as correlation. For example, auto-correlation destroys the information necessary to distinguish between minimum phase and non-minimum phase signals. Thus, two signals may have identical second-order statistics yet have different higher-order statistics. This property is of particular interest in handling signals with identical auto-correlation functions and adds additional degrees of freedom for finding a set of time lags where a group of source signals will have different higher-order cumulants. This property is particularly advantageous to a BSS technique in accordance with the present invention because a condition of identifiability of this BSS technique is that all signals have a unique normalized fourth-order auto-cumulant. Note that the fourth-order cumulant is used because odd order cumulants of a process with a symmetric distribution will be zero.

[0229] Four properties of cumulants utilized in the BSS technique in accordance with the present invention are described below. Proofs of these cumulant properties may be found in C. L. Nikias and A. P. Petropulu, Higher-Order Spectra Analysis: A Non-Linear Signal Processing Framework. (PTR Prentice-Hall, Upper Saddle River, N.J.: 1993) and M. Rosenblatt, Stationary Sequences and Random Fields(Birkhauser, Boston, Mass.: 1985).

[0230] The N^thorder joint cumulant of the set of random variables {a₁s₁, a₂s₂, . . . , a_Ns_N} is $\begin{array}{cc}\mathrm{Cum}\left[a_1{s}_1,a_1{s}_2,\dots ,a_N{s}_N\right]=\left\{\prod _{i=1}^N{a}_i\right\}\mathrm{Cum}\left[s_1,s_2,\dots ,s_N\right]& \left(69\right)\\ \mathrm{where}\left\{a_1,a_2,\dots ,a_N\right\}\mathrm{are}\mathrm{constants}.& \end{array}$

[0231] If the set of random variables {s₁, s₂, . . . s_N} can be divided in to two or more groups that are statistically independent, then their N^th-order joint cumulant is zero.

[0232] If the sets of random variables {s₁, s₂, . . . , s_N} and {n₁, n₂, . . . , n_N} are statistically independent, i.e. f_s,n(s₁, s₂, . . . , s_N, n₁, n₂, . . . , n_N)=f_s(s₁, s₂, . . . , s_N)·f_n(n₁, n₂, . . . , n_N) then the N^th-order joint cumulant of the pair-wise sum is

[0233] Cum[s₁+n₁,s₂+n₂, . . . , s_N+n_N]=Cum[s₁, s₂, . . . , s_N]+Cum[n₁, n₂, . . . , n_N].

[0234] If the set of random variables {n₁, n₂, . . . , n_N} are jointly Gaussian, then the joint cumulants of order N>2 are identically zero. That is, if n=[n₁, n₂, . . . , n_N]^T□N(μ_n, K_n), then Cum[n₁, n₂, . . . , n_N]=0.

[0235] A BSS technique in accordance with the present invention utilizes a fourth order spatial cumulant matrix. Three definitions of the spatial fourth-order cumulant matrix and associated properties are provided below.

[0236] The spatial fourth-order cumulant matrix is used as a basis for estimating a separation matrix at low signal-to-noise ratios and in the presence of spatially and temporally correlated noise since it theoretically has no noise subspace, even if the noise is correlated. This eliminates the need to use either degrees of freedom and/or secondary sensor data to estimate the noise subspace, which must be removed in order for the matrix-pencil to be formed. As described below, the absence of the noise subspace is a direct result of using a higher-order cumulant, i.e. order>2, and is particularly advantageous to a blind source separation technique in accordance with the present invention.

[0237] The three spatial fourth-order cumulant matrix definitions and their properties are presented herein with consideration of the fact that the sensors are in reality never omni-directional, never have identical manifolds, and that different sets of time lags are needed to estimate a pair of spatial fourth-order cumulant matrices to form the matrix-pencil. These considerations are a clear distinction from previous treatments of the spatial fourth-order cumulant matrix. See, for example, H. H. Chiang and C. L. Nikias, “The ESPRIT Algorithm with Higher-Order Statistics,” Proc. Workshop on Higher-Order Spectral Analysis, Vail, Colo., June 1989, pp. 163-168, C. L. Nikias, C. L. Nikias and A. P. Petropulu, Higher-Order Spectra Analysis: A Non-Linear Signal Processing Framework(PTR Prentice-Hall, Upper Saddle River, N.J.: 1993), M. C. Dogan and J. M. Mendel, “Applications of Cumulants to Array Processing—Part I: Aperture Extension and Array Calibration,” IEEE Trans. Signal Processing, Vol. 43, No. 5, May 1995, pp. 1200-1216, and N. Yuen and B. Friedlander, “Asymptotic Performance Analysis of ESPRIT, Higher-order ESPRIT, and Virtual ESPRIT Algorithms,” IEEE Trans. Signal Processing, Vol. 44, No. 10, October 1996, pp. 2537-2550. Understanding the properties of the spatial fourth-order cumulant matrix such as its rank, null spaces, etc., and its relationship to the mixing matrix are beneficial to developing a signal subspace blind separation technique using fourth-order cumulants and a matrix-pencil in accordance with the present invention.

[0238] A brief review of the spatial correlation matrix and its properties are provided below to aid in understand its use in a BSS technique in accordance with the present invention. The spatial correlation matrix of the sensor array output is defined in D. H. Johnson and D. E. Dudgeon, Array Signal Processing: Concepts and Techniques. (PTR Prentice-Hall, Englewood Cliffs, N.J.: 1993), which is hereby incorporated by reference in its entirety as if presented herein, as:

[0239] R_x(τ)=E[x(t)x^H(t−τ)]  (72)

[0240] Substituting (25) for x(t) in to equation (72) and applying assumptions A1 and A3, the spatial correlation matrix becomes $\begin{array}{cc}\begin{array}{c}{R}_x\left(\tau \right)=E\left[\left\{\mathrm{Vr}\left(t\right)+n\left(t\right)\right\}{\left\{\mathrm{Vr}\left(t-\tau \right)+n\left(t-\tau \right)\right\}}^H\right]\\ =E\left[\mathrm{Vr}\left(t\right)r^H\left(t-\tau \right)V^H\right]+E\left[\mathrm{Vr}\left(t\right)n^H\left(t-\tau \right)\right]\\ +E\left[n\left(t\right)r^H\left(t-\tau \right)V^H\right]+E\left[n\left(t\right)n^H\left(t-\tau \right)\right]\\ =\mathrm{VE}\left[r\left(t\right)r^H\left(t-\tau \right)\right]V^H+E\left[n\left(t\right)n^H\left(t-\tau \right)\right]\\ ={\mathrm{VR}}_r\left(\tau \right)V^H+R_n\left(\tau \right)\end{array}& \text{(73)}\end{array}$

[0241] which has elements $\begin{array}{cc}{\left[R_x\left(\tau \right)\right]}_{\mathrm{rc}}=\sum _{j=1}^M{v}_{\mathrm{rj}}{v}_{\mathrm{cj}}^{*}E\left[r_j\left(t\right)r_j^{*}\left(t-\tau \right)\right]+E\left[n_r\left(t\right)n_c^{*}\left(t-\tau \right)\right]& \text{(74)}\end{array}$

[0242] where the subscript rc indicates the element is in the r^throw and c^thcolumn. Since the signal and noise processes are assumed to be zero mean, assumptions A2 and A4, the spatial correlation matrix defined in equation (72) is equivalent to the spatial covariance matrix, and thus the terms are used interchangeably.

[0243] In general, most second-order techniques make use of the spatial correlation or covariance matrix only at a delay lag of zero, {τ=0}. In such a case the spatial correlation matrix is Hermitian and non-negative definite. See for example D. H. Johnson and D. E. Dudgeon, Array Signal Processing: Concepts and Techniques. (PTR Prentice-Hall, Englewood Cliffs, N.J.: 1993), C. L. Nikias and A. P. Petropulu, Higher-Order Spectra Analysis: A Non-Linear Signal Processing Framework. (PTR Prentice-Hall, Upper Saddle River, N.J.: 1993), and A. Papoulis, Probability, Random Variables, and Stochastic Processes. (WCB/McGraw-Hill, Boston, Mass.: 1991), for example. Further, if the sensor outputs are linearly independent, that is E[{a^Tx(t)}{a^Tx(t)}*]>0 for any a=[a₁, a₂, . . . , a_N]^T≠0, then the spatial correlation matrix is positive definite. As a consequence of the spatial correlation matrix being non-negative definite for τ=0, its determinant will be real and non-negative, and will be strictly positive if and only if the sensor outputs are linearly independent. However, if τ≠0 then the spatial covariance matrix is indefinite and non-Hermitian.

[0244] The first definition of a spatial fourth-order cumulant matrix presented takes advantage of the steering vectors having a norm of one. This is stated mathematically in equation (26). As will be shown, this is utilized to factor the spatial fourth-order cumulant matrix into Hermitian form when the sensors are not omni-directional with identical manifolds. The first spatial fourth-order cumulant matrix is defined at the set of time lags (τ₁, τ₂, τ₃) as $\begin{array}{cc}{C}_x^4\left({\tau }_1,{\tau }_2,{\tau }_3\right)\equiv \sum _{i=1}^N\mathrm{Cum}\left[x_i^{*}\left(t-{\tau }_1\right)x_i\left(t-{\tau }_2\right)x\left(t\right)x^H\left(t-{\tau }_3\right)\right]& \text{(75)}\end{array}$

[0245] and is referred to as spatialfourth-order cumulant matrix 1.

[0246] The spatial fourth-order cumulant matrix 1 as defined in (75) is in general a complex N×N matrix with the element in the r^throw and c^thcolumn given by $\begin{array}{cc}{\left[C_x^4\left({\tau }_1,{\tau }_2,{\tau }_3\right)\right]}_{\mathrm{rc}}=\sum _{i=1}^N\mathrm{Cum}\left[x_i^{*}\left(t-{\tau }_1\right)x_i\left(t-{\tau }_2\right)x_r\left(t\right)x_c^{*}\left(t-{\tau }_3\right)\right]& \text{(76)}\end{array}$

[0247] where { }* denotes complex conjugation. Substituting equation (24) into (76), element rc becomes $\begin{array}{cc}{\left[C_x^4\left({\tau }_1,{\tau }_2,{\tau }_3\right)\right]}_{\mathrm{rc}}=\sum _{i=1}^N\mathrm{Cum}[\left(\sum _{j=1}^M{v}_{\mathrm{rj}}{r}_j\left(t\right)+n_r\left(t\right)\right){\left(\sum _{m=1}^M{v}_{\mathrm{cm}}{r}_m\left(t-{\tau }_3\right)+n_c\left(t-{\tau }_3\right)\right)}^{*}·{\left(\sum _{k=1}^M{v}_{\mathrm{ik}}{r}_k\left(t-{\tau }_1\right)+n_i\left(t-{\tau }_1\right)\right)}^{*}\left(\sum _{l=1}^M{v}_{\mathrm{il}}{r}_l(t-{\tau }_2)+n_i(t-{\tau }_2)\right)]& \text{(77)}\end{array}$

[0248] Then, by Cumulant Property 3 and assumption A3, (77) becomes $\begin{array}{cc}{\left[C_x^4\left({\tau }_1,{\tau }_2,{\tau }_3\right)\right]}_{\mathrm{rc}}=\sum _{i=1}^N\mathrm{Cum}\left[\sum _{j=1}^M{v}_{\mathrm{rj}}{r}_j\left(t\right)\sum _{k=1}^M{v}_{\mathrm{ik}}^{*}{r}_k^{*}\left(t-{\tau }_1\right)\sum _{i=1}^M{v}_{\mathrm{il}}{r}_l\left(t-{\tau }_2\right)·\sum _{m=1}^M{v}_{\mathrm{cm}}^{*}{r}_m^{*}(t-{\tau }_3)\right]+\sum _{i=1}^N\mathrm{Cum}[n_r(t)n_i^{*}(t-{\tau }_1)n_i(t-{\tau }_2)n_c^{*}(t-{\tau }_3)]& \text{(78)}\end{array}$

[0249] where the terrns have been re-ordered. However, by assumption A4 and Cumulant Property 4, $\begin{array}{cc}\sum _{i=1}^N\mathrm{Cum}\left[n_r\left(t\right)n_i^{*}\left(t-{\tau }_1\right)n_i\left(t-{\tau }_2\right)n_c^{*}\left(t-{\tau }_3\right)\right]=0& \text{(79)}\end{array}$

[0250] and thus (78) becomes $\begin{array}{cc}{\left[C_x^4\left({\tau }_1,{\tau }_2,{\tau }_3\right)\right]}_{\mathrm{rc}}=\sum _{i=1}^N\mathrm{Cum}\left[\sum _{j=1}^M{v}_{\mathrm{rj}}{r}_j(t)\sum _{k=1}^M{v}_{\mathrm{ik}}^{*}{r}_k^{*}\left(t-{\tau }_1\right)\sum _{l=1}^M{v}_{\mathrm{il}}{r}_l\left(t-{\tau }_2\right)·\sum _{m=1}^M{v}_{\mathrm{cm}}^{*}{r}_m^{*}\left(t-{\tau }_3\right)\right]& \left(80\right)\end{array}$