A solution to reduce noise enhancement in pre-whitened LMS-type algorithms: the double direction adaptation

摘要

The LMS algorithm suffers from its slow rate of convergence, especially for high correlated input signal. The input pre-whitening based algorithms provide better convergence rate with the price of noise enhancement. To mitigate this noise enhancement, we present in this paper a technique, which consists on exciting simultaneously the adaptive filter at two directions: the input and the pre-whitened input directions. The proposed algorithm improves the rate of convergence without enhancing the noise. An analytical analysis of both convergence rate and steady state performances is presented. Simulation results are also presented to support the analysis and to compare the proposed algorithm with classical ones.