Blind source separation of convolutive mixtures

When a priori information about the propagation or the geometry of the array are not available, the model can be generalized to a blind source separation problem. It supposes the statistical independence of the sources and their non-Gaussianity. The observed signals are assumed to be convolutive mixtures of wide-band sources. Several criteria of source separation are studied which are based on the cancellation of different fourth-order cross-cumulants. For these criteria, we show in which conditions the separation is achieved. Results on real data illustrate the proposed methods