BLIND SEPARATION OF MIXTURE OF INDEPENDENT SOURCES THROUGH A QUASI-MAXIMUM LIKELIHOOD APPROACH

In this paper, we propose two methods fur separating mixtures of indep endent sources without any precise knowledge of their probability dist ribution, They are obtained by considering a maximum likelihood (ML) s olution corresponding to some given distributions of the sources and r elaxing this assumption afterward, The first method is specially adapt ed to temporally independent non-Gaussian sources and is based on the use of nonlinear separating functions. The second method is specially adapted to correlated sources with distinct spectra and is based ore t he use of linear separating filters, A theoretical analysis of the per formance of the methods has been made, A simple procedure for optimall y choosing the separating functions from a given linear space of funct ions is proposed, Further, in the second method, a simple implementati on based on the simultaneous diagonalization of two symmetric matrices is provided, Finally, same numerical and simulation results are given , illustrating the performance of the method and the good agreement be tween the experiments and the theory.