Mixed NMF Blind Source Separation Algorithm

Abstract

The rank one NMF blind source separation algorithm (NMF-R1) was obtained by imposing the sparsity constraint on the fast NMF algorithm based rank one. NMF blind source separation algorithm based on least squares (NMF-LS) was obtained by using pseudo-inverse matrix. NMF-R1 algorithm was superior to the existing blind source separation algorithms based on NMF. NMF-LS algorithm had faster computation speed, but the result of decomposition was not unique. In order to further enhance the signals separated performance, crossover iteration between NMF-R1 and NMF-LS was used to getting the mixing matrix and the signal matrix, and the mixed NMF blind source separation algorithm (NMF-LR) was obtained. Simulation results show that the separation performance of NMF-LR algorithm is better than that of NMF-R1 and NMF-LS, and the computation complexity of NMF-LR algorithm is nearly the same as NMF-R1 algorithm.