A fixed point algorithm for blind identification of underdetermined mixtures

Abstract

In this paper, we consider the problem of blind identification in the underdetermined instantaneous mixtures case, where there are more sources than sensors. Recently, a simultaneous matrix diagonalization (SMD)-based method has been proposed without any need of sparsity assumption on sources. Instead of using SMD technique, we propose a fixed point algorithm, which estimates the columns of the mixing matrix sequentially one by one. Computer simulations are presented in order to demonstrate that the proposed algorithm has the ability to identify the mixing matrix, and its performance is comparable to that of SMD-based algorithm.