Adaptive Sparse Factorization for Even-Determined and Over-Determined Blind Source Separation

In this paper, we present an adaptive sparse factorization method for even-determined and over-determined blind source separation, where the sources are assumed to be sparse. The objective of our method is to find a demixing matrix to make the output signal as sparse as possible. First, a cost function measuring the sparsity of the output signals is introduced. Then an adaptive algorithm for the learning of the demixing matrix is proposed by the nature gradient. Compared to Independent Component Analysis, the new method can deal with the mixing cases that the sources are mutually correlated. Simulation results show the effectiveness of the new method.