A quantitative convergence and tracking comparison of two adaptive echo-cancellation algorithms

Abstract

The sign algorithm (SA) and the signed regressor algorithm (SRA) have become appealing in high data rate adaptive filtering applications such as speech processing and echo cancellation. The paper provides a quantitative comparison of the two algorithm convergence times for adaptive echo cancellation. The paper also compares the steady-state algorithm performances when the echo-path parameters are randomly time varying and the far-end signal is fading. The comparison is done for Gaussian near-end and far-end signals. An operating range is determined where the SA performs nearly as well as the SRA for a stationary echo path. It is also shown that the SA outperforms the SRA for a randomly time-varying echo path and a fading far-end signal.