The pre-whitened NLMS: a promising algorithm for acoustic echo cancellation

The rate of convergence of adaptive filters depends on the correlation of the input signal. In this paper, we present a simple pre-whitened NLMS (normalized least mean square) algorithm, suitable for highly correlated inputs such as speech. It uses both input and error pre-whitening. The adaptive filter is excited at the two directions of the input and the pre-whitened input. The updating rate is reduced (every two iterations). When compared to classical algorithms, the proposed method provides a better convergence rate with low complexity. Furthermore, the robustness with respect to noise is improved.