Combined non-orthogonal joint zero-diagonalization and joint diagonalization for sources separation

This communication is concerned with blind separation of instantaneous mixtures of source signals based on the use of spatial quadratic time-frequency (spectrum) distributions. First, we propose a new algorithm to perform the non orthogonal joint zero-diagonalization and joint-diagonalization of given sets of matrices. We also present a selection procedure of useful time-frequency points in order to automatically determine which matrices have to be joint (zero-) diagonalized. One advantage of the proposed approaches is that they do not require any whitening stage and thus they are intended to work even with a class of correlated signals, rather than the classical independent one. Finally, computer simulations are provided in order to illustrate the effectiveness of the proposed methods and to compare them with more classical ones