Adaptive step-size control in blind source separation

The behavior of the classic algorithm for blind source separation is reviewed for a fixed step-size. This analysis is extended to the case where the step-size decreases proportionally to 1/ t. Although such a step-size sequence guarantees error-free convergence, mismatches and perturbations make it unrealistic for most practical implementations. To ameliorate these difficulties an error-dependent step-size must be employed. The coefficients of the estimating function provide an appropriate “measure of error” and serve as the basis for a self-adjusting time-varying step-size. Extensive simulations show the proposed approach tracks a time-varying mixing environment and performs error-free convergence in a time-invariant environment.