Network echo canceller with active taps stochastic localization

Abstract

Normalized least mean square (NLMS) is the most popular adaptive algorithm used in echo cancellation. However, the convergence rate of NLMS is rather slow in the case of large taps number. This paper explores stochastic taps NLMS (STNLMS), which accelerate the convergence of echo canceller by randomly detecting the significant part in echo path. In the proposed algorithm, an auxiliary adaptive filter is adopted and assigned with active taps different from the primary filter. Via periodical evaluation, the active taps of these two filters is found and used to refresh the primary filter. Then a new is specified in a stochastic way base on the above better taps position and assigned to the auxiliary filter for further evaluation. Using the echo path models of ITU-T G.168, experiments results prove that the convergence rate and tracking rate of STNLMS are much faster than NLMS with full taps update and are very close to NLMS with exact active taps update.