Blind source separation and channel equalization: an epsilon-approximation algorithm and finite alphabet approach
Qingyu Li, Er‐Wei Bai · 2003
In this thesis, we shoe, that a blind separator/equalizer can be obtained without using any high order including second order statistical information on the input by formulating the blind source separation/channel equalization problem into a quadratic optimization with binary constraints. Then, efficient e-approximation algorithms are presented and applied to find the separator/equalizer coefficients. Identifiability conditions are derived and simulations verify the theoretical results. The method is extended to slowly time-varying as well as nonlinear systems. Different implementation for OFDM systems is also provided. The proposed approach applies to non-minimum phase channels and converges to an e neighborhood of the global optimum. A discussion for Blind Source Separation when second order statistics are available is also presented.