Data-Selective Volterra Adaptive Filters

Abstract

This work proposes the use of data-selective schemes in Volterra adaptive filters. By working only with input data that brings novelty to the system, thus avoiding unnecessary parameter updates, one can reduce drastically the high computational burden associated with the use of Volterra series. Considering a nonlinear system identification setup, results show that the proposed data-selective Volterra filters update in only 7.5% of the iterations, whereas in a nonlinear channel equalization setup the update rate is around 27%, without compromising the performance in terms of mean squared error or bit-error rate.