Blind source separation with nonlinear autocorrelation and non-Gaussianity (Q1023329)

The authors study blind source separation (BBS) problem in this paper when source signals have non-Gaussianity and temporal structure with nonlinear autocorrelation. By maximizing an objective function that combines both of non-Gaussianity and nonlinear autocorrelation of the desired signals, a gradient ascent source separation algorithm is proposed. Its stability conditions are analyzed. By computer simulations, the authors show that the proposed algorithm can separate the source signals with square temporal autocorrelation and non-Gaussianity.