Blind source separation by new M-WARP algorithm

A new independent component analysis technique is presented, which is based on the information-theoretic approach and implemented by the functional-li nk network, that allows mixed independent sub-Gaussian and super-Gaussian s ource signals to be separated out. To assess the theory, the results of com puter simulations performed both on synthetic and real-world data are prese nted, and the performances of the new algorithm compared with those exhibit ed by the 'mixture of densities' based algorithm of Xu ct al.