Algorithm for Underdetermined Blind Source Separation Based on Least-Mean-Square Error and Sparse Features

An algorithm based on least-mean-square error and sparse features is presented for underdetermined blind source separation (BSS), i.e., situation when the number of observed signals´ is less than that of sources. In this paper, using the sparsity of sources, first, we estimate the mixing matrix using a new potential function based on clustering method. Then use the estimated mixing matrix and the selfcorrelation of sources, by searching the accurate values at the source clustering directions, we can obtain the optimal sub-matrix for separation through least-mean-square error criterion, which overcomes the disadvantages of traditional algorithms in searching the optimal sub-matrix. Simulation results show the separated signals have higher SNR, and compared with the other similar method, the proposed approach has better separation performance.