Memorized error and varying error bound for extending adaptiveness for normalized subband adaptive filtering

Abstract

This paper extends the adaptive normalized sub-band adaptive filtering (NSAF) by introducing variable error bound and memorizing the error convergence. The variable error bound attempts to vary the updating point of the filter coefficients. The error memory aids in updating the point based on the history of error rather than the previous error. The extended adaptiveness significantly improved NSAF in terms of convergence, complexity and noise robustness. The algorithm is also proved for its stability though the step-size is varied. The characteristics of the step-size are also investigated to determine its significance and nature on minimizing the error. The superiority of the MVS-SNSAF algorithm is proved against conventional algorithm using the aforesaid analysis.