New blind source separation algorithm based on approximate maximum likelihood and its application

Blind source separation (BSS) is aiming at recovering the source signals from the observed mixtures, exploiting only the assumption of mutual independence between the source signals. A new solution to the BSS is proposed based on the traditional approximate maximum likelihood method. This new method have initialized four-order moments matrix. By Jacobi optimization, the blind source signals are separated through the rotation matrix acting on the each pair of signals. The results of real speech signal experiments show that this method can separate the mixture signals of sub-Gaussian and sup-Gaussian. The method is compared with FastICA algorithm and JADE algorithm. The results show that the method has more excellent separation performance than those two methods, and the computation efficiency of the new method is obviously higher than those two methods especially JADE algorithm. This method can be extended to other non-stationary signals and fault diagnosis.