Fast Fixed-Point Algorithms for Bayesian Blind Source Separation

摘要

In this paper, various fast algorithms for linear blind source separation(BSS)211u001eare developed. The new point of view opens way for developing several new fast 211u001efixed-point algorithms for extracting signals with various properties. A Bayesian 211u001eversion of the ordinary independent component analysis and a version which takes 211u001einto account both the time-domain behavior and non-Gaussianity of the source 211u001esignals are studied in more depth. The Bayesian version can be used for 211u001eoptimizing model structure and comparing different hypotheses. Natural signals 211u001ehave typically non-Gaussian distributions and time-dependencies. There has, 211u001etherefore, been a demand for an algorithm which can utilize both types of 211u001einformation.