A blind source separation method based on diagonalization of correlation matrices and genetic algorithm

Abstract

A new cost function based on diagonalization of the correlation matrices is proposed to measure the independency of output signals in this paper. In order to expand the search space and decrease the cross-correlation among the sub-sources, we propose to perform nonlinear transformation for the cost function. The real coded genetic algorithm is also proposed to search the optimum solution, which can overcome the drawbacks of traditional gradient search technique being likely tend to fall into local minimums. This novel method can be applicable to instantaneous or convolutive mixture models with stationary or non-stationary input signals. Simulation results demonstrate the algorithm not only has fast convergence performance and high accuracy, but also can improve the output SNR greatly.