Blind source separation based on immune algorithm with four-order mutual cumulates

Abstract

A blind source separation method based on immune algorithm (IA) is proposed. The first step performs initialization of mixed signals, which estimates the dimension of signals and abstracts principal component information by eigenvalue decomposition. The second step performs separation of sources, where a separate matrix is estimated. The separate matrix is updated by IA and high order statistics (high order cumulates), where contrast function is based on the transformation of four-order mutual cumulates. The effectiveness of proposed method for blind source separation is demonstrated by simulation.