An Blind Source Separation Algorithm Based on Constrained NMF
Chunyun Xu · Piezoelectrics and Acoustooptics · 2010
The nonnegative matrix factorization(NMF) is another new approach for resolving the problem of the blind source separation(BSS) with the non-independent source signals mixed.In this paper,an algorithm for BSS based on the constrained NMF is proposed.Through imposing the least correlated component constraints on the source signals estimated by NMF,the determinant criterion is applied to constrain the estimated mixing matrix,so that the unique factorization of NMF has been realized.Comparing with the present algorithm,the sparsity of the mixing matrix is relaxed and the quality of the recovered source signals has been improved significantly.The proposed algorithm is also suitable for separating the independent source signals.