Blind separation of delayed sources based on information maximization

Bell and Sejnowski (see Neural Computation, vol.7, no.6, p.1004-34, 1995) have presented an approach to blind source separation based on the information maximization principle. We extend this approach into more general cases where the sources may have been delayed with respect to each other. We present a network architecture capable of coping with such sources, and we derive the adaptation equations for the delays and the weights in the network by maximizing the information transferred through the network. Examples using wideband sources such as speech are presented to illustrate the algorithm.