Abstract:
The normalized least-mean-square (NLMS) algorithm and the affine projection algorithm (APA) are the most common choices for echo cancellation. In this type of application, an adaptive algorithm with a constant step-size has to compromise between several performance criteria (e.g., high convergence rate versus low misadjustment). In this paper we present a class of variable step-size NLMS and APAs, which are designed to recover the near-end signal in the error of the adaptive filter. The simulation results indicate a robust behaviour of these algorithms against different types of near-end signal variations, including double-talk.
