Source distribution models for blind source separation

Rai, CS

doi:10.1016/j.neucom.2004.01.003

Source distribution models for blind source separation

2004-03-01

8

PRE

AI

DOI：10.1016/j.neucom.2004.01.003

原文链接

原文求助

摘要

En 中文

Various methods have been proposed to. separate mixtures of sub- and super-Gaussian signals. Effectiveness of a blind source separation algorithm depends upon the source distribution model used for deriving the weight update rule. Different hypothesized source distribution models are used for representing sub- and super-Gaussian sources. In this paper, a more justifiable approach is considered for characterizing the source distributions. For a given density model of super- or sub-Gaussian sources, a symmetric probability density function around a Gaussian distribution is used for representing sub- or super-Gaussian sources. Hence, corresponding to a long tail sharper density function for super-Gaussian signals, a short-tail flatter density function for sub-Gaussian signals is obtained. Optimization of the objective function leads to two different nonlinear functions. This approach leads to appropriate handling of both kinds of signals simultaneously. Simulations results with audio signals have been presented. (C) 2004 Published by Elsevier B.V.

Keyword:

blind source separation (BSS)

objective function

probability density function (pdf)

sub-Gaussian signals

super-Gaussian signals

Source distribution models for blind source separation

摘要

AI总结

期刊

机构

引用论文