1:00, MULT-P3.1
A COMBINED KALMAN FILTER AND NATURAL GRADIENT ALGORITHM APPROACH FOR BLIND SEPARATION OF BINARY DISTRIBUTED SOURCES IN TIME-VARYING CHANNELS
M. JAFARI, H. SEAH, J. CHAMBERS
A combined Kalman filter (KF) and natural gradient algorithm (NGA) approach is proposed to address the problem of blind source separation (BSS) in time-varying environments, in particular for binary distributed signals. In situations where the mixing channel is non-stationary, the performance of NGA is often poor. Typically, in such cases, an adaptive learning rate is used to help NGA track the changes in the environment. The Kalman filter, on the other hand, is the optimal minimum mean square error method for tracking certain non-stationarity. Experimental results are presented, and suggest that the combined approach performs significantly better than NGA in the presence of both continuous and abrupt non-stationarities.

1:00, MULT-P3.2
JOINT DIAGONALIZATION VIA SUBSPACE FITTING TECHNIQUES
A. VAN DER VEEN
Joint diagonalization problems of Hermitian or non-Hermitian matrices occur as the final parameter estimation step in several blind source separation problems such as ACMA, JADE, PARAFAC, and SOBI. Previous approaches have been Jacobi iteration schemes and alternating projections. Here we show how the joint diagonalization problem can be formulated as a (weighted) subspace fitting problem so that it can be solved using the efficient Gauss-Newton optimization algorithm proposed for that problem. Since a good initial point is usually available, the algorithm converges very fast.

1:00, MULT-P3.3
ON OPTIMAL AND UNIVERSAL NONLINEARITIES FOR BLIND SIGNAL SEPARATION
H. MATHIS, S. DOUGLAS
The search for universally applicable nonlinearities in blind signal separation has produced nonlinearities that are optimal for a given distribution, but also nonlinearities which are most robust against model mismatch. This paper shows yet another justification for the score function, which is in some sense a very robust nonlinearity. It also shows that among the class of parameterizable nonlinearities, the threshold nonlinearity with the threshold as a parameter is able to separate any non-Gaussian distribution, which is also proven in this paper.

1:00, MULT-P3.4
NONLINEAR BLIND SOURCE SEPARATION BY SPLINE NEURAL NETWORKS
M. SOLAZZI, A. UNCINI, F. PIAZZA
In this paper a new neural network model for blind demixing of nonlinear mixtures is proposed. We address the use of the Adaptive Spline Neural Network recently introduced for supervised and unsupervised neural networks. These networks are built using neurons with flexible B-spline activation functions and in order to separate signals from mixtures, a gradient-ascending algorithm which maximize the outputs entropy is derived. In particular a suitable architecture composed by two layers of flexible nonlinear functions for the separation of nonlinear mixtures is proposed. Some experimental results that demonstrate the effectiveness of the proposed neural architecture are presented.

1:00, MULT-P3.5
INDEPENDENT COMPONENT ANALYSIS WITH SINUSOIDAL FOURTH-ORDER CONTRASTS
J. MURILLO-FUENTES, F. GONZALEZ-SERRANO
The authors propose a new solution to the Independent Component Analysis (ICA) problem. In the two-dimensional case, we prove that under the whiteness constraint some fourth-order contrasts may be approximated by a sinusoid. Thus, the minimization of the contrast reduces to computing its phase. The novel approach, called SICA (Sinusoidal ICA), uses the 'Jacobi optimization' to cope with higher dimensions. The method presented has a good performance along with a low computational cost. Some experiments with blind separation of audio and synthetic sources are included to compare the algorithm to other well-known approaches. http://alcaudon.tsc.uc3m.es/~murillo/

1:00, MULT-P3.6
JOINT ANTI-DIAGONALIZATION FOR BLIND SOURCE SEPARATION
A. BELOUCHRANI, K. ABED-MERAIM, M. AMIN, A. ZOUBIR
We address the problem of blind source separation of non-stationary signals of which only instantaneous linear mixtures are observed. A blind source separation approach exploiting both auto-terms and cross-terms of the time-frequency (TF) distributions of the sources is considered. The approach is based on the simultaneous diagonalization and anti-diagonalization of spatial TF distribution matrices made up of, respectively, auto-terms and cross-terms. Numerical simulations are provided to demonstrate the effectiveness of the proposed approach and compare its performances with existing TF-based methods.

1:00, MULT-P3.7
ELEMENTARY COST FUNCTIONS FOR BLIND SEPARATION OF NON-STATIONARY SOURCE SIGNALS
M. JOHO, R. LAMBERT, H. MATHIS
Blind source separation (BSS) is a problem found in many applications related to acoustics or communications. This paper addresses the blind source separation problem for the case where the source signals are non-stationary and the sensors are noisy. To this end, we propose several usefulelementary cost functions which can be combined to an overall cost function. The elementary cost functions might have different objectives, such as uncorrelated output signals or power normalization of the output signals. Additionally, the corresponding gradients with respect to the adjustable parameters are given. We discuss the design of an overall cost function and also give a simulation example.

1:00, MULT-P3.8
A SECOND-ORDER METHOD FOR BLIND SEPARATION OF NON-STATIONARY SOURCES
R. ZHANG, M. TSATSANIS
The question addressed in this paper is whether and under what conditions blind source separation is possible using only second-order statistics. It is well known that for stationary,i.i.d. sources the answer is negative due to the inherent unitary matrix ambiguity of output second-order information. It is shown in this paper however, that if the sources' power is allowed to vary with time, unique identifiability can be achieved without resorting to higher order statistics. In many applications the sources' power does change with time (e.g., speech or fading communication signals), and therefore the result has practical relevance. A novel second-order source separation method is proposed based on a generalized eigen-decomposition of appropriate correlation matrices and the identifiability conditions are investigated. Asymptotic performance results for the output SIR are developed.

1:00, MULT-P3.9
OVER-COMPLETE BLIND SOURCE SEPARATION BY APPLYING SPARSE DECOMPOSITION AND INFORMATION THEORETIC BASED PROBABILISTIC APPROACH
S. KADAMBE, A. OSSADTCHI
Both in the case of ceullar communication and in the case of spoken dialogue based information retrieval systems on the mobile platform there exist a number of interference signals. Therefore, it is essential to separate these interference signals from the intended signal(s) in order to have clear communication in the case of cellular phone and to improve the speech recognition accuracy in the case of spoken dialogue based information retreival systems. Since the number and nature of source signals (intended + interference signals) change, it is not practical to know them a priori. Therefore, it is not always practical to apply signal separation techniques that work well when the number of source signals is equal to the number of sensors. In addition, since how the signals get mixed is unknown, we need to apply blind techniques for the separation. This paper is concerned with a blind source separation (BSS) technique for the over-complete case (# of sources > # of sensors) that is based on the sparse decomposition and, the joint estimation of mixing matrix and the separated source signals by applying information theoretic based probabilistic approach. Experimental results of signal separation using various real speech and noise signals indicate that the quality of separated source signals is 4 dB better than the current techniques.

1:00, MULT-P3.10
AN ANY ORDER GENERALIZATION OF JADE FOR COMPLEX SOURCE SIGNALS
E. MOREAU
In this paper, considering the complex case, we extend some results leading to the popular JADE algorithm to cumulants of any order greater than or equal to three. We first exhibit a new contrast function which constitutes a generalization for the underlying contrast of JADE which thus appears as a particular case. Then we generalize the link between this new contrast and a joint-diagonalization criterion of a set of matrices. Moreover, in the two sources case, we show that the generalized contrast can be written as a simple quadratic form whatever the cumulant order. Finally, some computer simulations illustrate the potential advantage one can take of considering statistics of different orders for the joint-diagonalization of cumulant matrices.

1:00, MULT-P3.11
AN ADAPTIVE BLIND SIGNAL SEPARATION BASED ON THE JOINT OPTIMIZATION OF GIVENS ROTATION
N. BIENATI, U. SPAGNOLINI, M. ZECCA
Blind signal separation (BSS) is a recurrent problem in many multi-sensors applications where observations can be modelled as mixtures of N statistical independent source signals. In this paper we propose the estimation of the orthonormal transformation matrix Q in the case of whitened observations and a cost function based on the fourth-order moments. Q is described as combination of elementary Givens rotations and the optimization is carried out jointly for all the rotations. When sub-sets of angles are optimized separately the method reduces to the deflation approach which has been proved to be globally convergent. The joint estimation of Givens rotation matrices has a computational complexity O(7N^2) and, compared to other adaptive BSS, simulations demonstrate that it converges faster and achieves a better attenuation of cross-talks.

1:00, MULT-P3.12
A NEW TECHNIQUE FOR BLIND SOURCE SEPARATION USING SUBBAND SUBSPACE ANALYSIS IN CORRELATED MULTICHANNEL SIGNAL ENVIRONMENTS
K. OWEISS, D. ANDERSON
We investigate a new framework for the problem of blind source identification in multichannel signal processing. Inspired by a neurophysiological data environment, where an array of closely spaced recording electrodes is surrounded by multiple neural cell sources, significant spatial correlation of source signals motivated the need for an efficient technique for reliable multichannel blind source identification. In a previous work [1], we adopted a new approach for noise suppression based on thresholding an Array Discrete Wavelet Transform (ADWT) representation of the multichannel data. We extend our previous work to identify sources from the observation mixtures. The technique relies on separating sources with highest spatial energy distribution in each frequency subband spanned by the corresponding wavelet basis. Accordingly, the best basis selection criterion we propose benefits from the additional degree of freedom offered by the space domain. The transform magnitude-invariance property revealed by this technique makes it very efficient to track spatial source variations sometimes encountered in multichannel neural recordings. Results from real multichannel multiunit neural data are presented and the overall performance is evaluated. [1]http://www.eecs.umich.edu/~koweiss/Publications.html