rag replacements USING CASCADED NEURAL NET FILTER IN NON-LINEAR IDENTIFICATION OF ACOUSTIC ECHO PATH

A linear acoustic echo canceller (AEC) is unable to effectively identify an acoustic echo path because the echo path has both linear and non-linear attributes. The acoustic echo channel is physically formed by a non-linear distortion system concatenated with a linear faded echo path. Motivated by this physical reality, a cascaded neural net filter is used as the filter structure in a non-linear AEC. It consists of a feedforward neural net filter, in series with a linear FIR filter. Different adaptive algorithms are used to estimate the linear and non-linear parts of the filter. The adaptive cascaded filter converges to a higher echo return loss enhancement (ERLE) than the linear adaptive FIR filter. 1