There is a trend to develop blind or semi-blind source extraction algorithms based on

second-order statistics, due to its low computation load and fast processing speed.

An important and primary work is done by Barros and Cichocki, who propose an

extraction algorithm based on a time delay. The algorithm is simple and fast, but its

performance is not satisfying. The paper extends their work and proposes a robust

algorithm based on eigenvalue decomposition of several delayed covariance matrices.

It is faster and has better performance, which is conˉrmed by theoretical analysis

and computer simulations.