Novel Variable Step Size EASI Algorithm for Blind Source Separation

This paper presents a new variable step size equivariant adaptive source separation via independence (VSS-EASI) for online blind separation of independent sources. By building a nonlinear relationship between the step size and the difference among adjacent separating matrices, the proposed algorithm makes the step size adaptively decrease with the change of separating matrices until the step size reduces to zero. Simulation results show that the convergence and steady state error performance of the proposed algorithm outperforms the fixed, step size algorithms in both stationary and nonstationary environments.