Combined Step-size Affine Projection M-estimator-like Algorithm

To improve the standard Affine Projection (AP) algorithm, which suffers from high steady-state error and poor convergence performance when dealing with unsteady impulse noise, a Combined Step-size Affine Projection M-estimator-like Algorithm (CS-APMLA) is proposed. CS-APMLA uses a convex combination method to combine multiple APMLA algorithms and adjusts the combination step according to the saturation characteristics to resolve the conflict between convergence speed and steady-state error, which can prevent the divergence of weight coefficients triggered by large-value impulse noise. From the theoretical derivation and simulation experiments, it can be shown that the convergence speed and noise reduction performance of the CS-APMLA algorithm is greatly improved compared with the APSA algorithm, VSS-APSA, and APMLA algorithm.