A nonlinear acoustic echo canceller with improved tracking capabilities

This paper introduces the use of a variable step size for a functional link adaptive filter (FLAF). We consider a split FLAF architecture, in which linear and nonlinear filterings are performed in two separate paths , thus resulting well-suited for online filtering applications, like the nonlinear acoustic echo cancellation (NAEC). We focus our attention on the nonlinear path to improve the overall NAEC performance. To this end, we derive a variable step size for the filter on the nonlinear path that shows reliance not only on the nonlinear path, but on the whole split FLAF architecture. The introduction of the variable step size for the nonlinear filter aims at improving the modeling of nonlinear speech signals, thus yielding superior performance in NAEC problems. Experimental results prove the effectiveness of the proposed method with respect to the standard split FLAF involving a fixed step size.