Underdetermined Blind Source Separation Based on Fuzzy C-Means Clustering and Sparse Representation

摘要

Traditional blind source separation is based on over- determined, but the underdetermined is more consistent with actual situation, based on sparse representation, Bofill proposed "two step" method to solve the problem under some assumptions. The accuracy of the mixture affects the recovery of sources, avoiding the subjectivity of choosing parameter, using the fuzzy C-means clustering to get the mixing matrix estimation; at the same time, to lessen the requirement of sparsity, combining ICA with SCA, based on the criterion of negentropy, sources can be separated. The test shows that the algorithm proposed here get a good result.