An algorithm for under-determined blind source separation based on the least-mean-square error and sparse features
Guoxia Sun · Journal of Shandong University · 2008
An algorithm was presented based on the least-mean-square error and sparse features for under-determined blind source separation,i.e.,observed signal numbers are less than sources numbers.Based on the clustering method,the mixing matrix was first estimated by a new potential function using the sparseness of sources.By using the estimated mixing matrix and the self-correlation of sources and searching the accurate values at the source clustering directions,the optimal sub-matrix for separation was obtained according to the least-mean-square error criterion.This can overcome the disadvantages of traditional algorithm in searching the optimal sub-matrix.Simulation results show that the separated signals have higher SNR,and the proposed approach has better separation performance compared with other similar methods.