De-cumulant based approaches for convolutive blind source separation

This paper studies the blind separation of signal sources (BSS) based on the approach of de-cumulant. It considers the cases where independent signal sources are mixed through convolutive mixing system with unity autochannel frequency responses and causal cross-channel FIR filters. Firstly, it tries to show that the de-cumulant is sufficient for separation. Secondly, novel algorithms are developed based on zero-forcing of cross-cumulant pairs. These algorithms are developed in the time-domain and so there is not the frequency permutation ambiguity problem usually suffered by most of the frequency-domain algorithms. Simulation results are presented to support the validity of the proposed algorithms.