Differential Evolution on Mixed Process Signal Separation

In this paper, a novel heuristic method which is applied to solving the mixed process signal separation problem is presented. Based on a stochastic searching technique, the proposed method performs simultaneously the estimation of the demixing and corresponding second moment (covariance) matrices so as to separate mixed process signals arising from several original sources. It hinges on one of state-of-theart evolutionary algorithms, termed differential evolution (DE) that is particularly devised for continues optimization. In the context of signal separation, the presented method could seek efficiently and reliably the optimal estimate of the demixing matrix for the mixed process signals, such as including autoregressive (AR) series, step change in process mean, and Gaussian noises. Following the proposed approach, the blind signal separation framework can be posed as a nonlinear mixed signal problem, where only the demixing matrix appear as unknowns. Several simulations involving linear mixtures of various practical process signals with different statistical characteristics are addressed to illustrate the new method.