Comparison of blind source separation algorithms
Yan Li, David M W Powers, James T. Peach · 2001
Abstract:- A set of experiments are designed to evaluate and compare the performances of three well known blind source separation algorithms in this paper. The specific algorithms studied are two group of neural networks algorithms, Bell and Sejnowski’s infomax algorithm and Hyvärinen’s fixed-point family, and J. F. Cardoso’s joint approxomate diagonalization of eigen-matrices algorithm. In this paper, the algorithms are quantitively evaluated and compared using the three measures, MATLAB flops (floating point operations), the difference between the mixing and separating matrices and the signal-to-noise ratios of the separated signals in this paper. Key-Words:- blind souce separation, information maximisation, fixed-point algorithm,JADE 1