Abstract:

The blind separation of convolution mixture signals is a nodus in blind source separation.In this paper,aiming at the independently and identically distributed non-Gaussian signals, a novel approach to the blind decon.volution is presented,namely,the Two-Step Extraction.The first step is to eliminate the delay vectors of all sources and to extract only linear mixture signal vectors.Thus,the problem of convolution is transformed into the linear
one.This step is called convolution linearization.The second step is to separate the estimated linearized signals and to obtain the source signals.Moreover,a method to design the filter of convolution linearization is proved by using the statistic and the matrix theories. The practicability of the proposed method is finally demonstrated by simula-tion.

Key words: blind source separation, convolution mixture, blind deconvolution, convolution linearization