Estimating the mixing matrix by using less sparsity

In this paper, the nonlinear projection and column masking (NPCM) algorithm is proposed to estimate the mixing matrix for blind source separation. It preserves the samples which are close to the interested direction while suppressing the rest. Compared with the existing approaches, NPCM works efficiently even if the sources are less sparse (i.e., they are not strictly sparse). Finally, we show that NPCM provides considerably accurate estimation of the mixing matrix by simulations.