A variable step size pre-whitened sign algorithm with quantized normalizing factor. Application to acoustic echo cancellation

In this paper, a variable step size pre-whitened sign algorithm with a quantized normalizing factor is proposed and applied for acoustic echo cancellation. The normalizing factor, introduced in adaptive algorithms, makes filter insensitive to variations in filter input power. The Quantization of this factor to the nearest power of two, leads to a low implementation cost. Furthermore, the concept of input pre-whitening is used to drive the adaptive filter, yielding to a faster convergence rate. To make the acoustic echo canceller robust, a new way of varying the step size is proposed. The main idea of varying the step size is very simple and depends on the state of the algorithm: transient state, steady state, or “double talk” state. It was finding that it improves significantly the ability of the algorithm to tackle the problem of double talk presence, and contributes in increasing the convergence rate by maintaining a good steady state performance. Simulation results are presented to support the proposed algorithm.