A unified algorithm for blind separation of independent sources

This paper presents a unified algorithm of blind source separation based on the "Independent Component Analysis"(ICA) principle. The algorithm can separate all sources provided there is at most one Gaussian distributed source. The key point is to find a matrix by which the estimates of the original signals are pairwise independent in the absence of noises. If the observed signals are corrupted by noises, minimum-variance unbiased estimates are obtained. In comparison with the algorithm proposed previously by the authors, this algorithm has a parallel pipeline structure, and will not need a preset threshold if both of the 3rd- and 4th-order cumulants of any non-Gaussian distributed source are not zero. This algorithm has significant advantages over existing algorithms.