Blind Separation of Sparse Signals based on Deflation and using the Differential Evolution Algorithm

The aim of this work is the development of a blind source separation method for sparse signals. Our approach is twofold. First, since separation criteria based on the sparsity property often lead to non-convex functions, we address the problem of extracting a single source by performing optimization through a metaheuristic method called differential evolution. Then, a deflation step is set up in order to perform source separation via successive executions of the proposed sparse source extraction algorithm. The resulting method is compared with the gradient descent method by analyzing the existence of local minima in the considered extraction criterion and as well as with respect to the obtained signal-to-interference ratios.