Improving performance for adaptive filtering with voice applications

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] Adaptive filtering is a common solution to many signal processing applications in which the environments are time varying. Its performance is limited by the misadjustment. A method to reduce the misadjustment when the additive noise in the desired signal is correlated is investigated in this work. The proposed method consists of two adaptive components: a modeling filter and a noise whitening filter. The proposed method performs better than the conventional LMS and RLS algorithms both in convergence speed and misadjustment factor. The performance of the adaptive filter when updated using LMS is observed. The improvement factor is proportional to the ratio of the noise power to white innovation power in the desired signal. The reduction in gradient noise allows larger step size, thus increasing the overall performance of the system. Both theoretical and simulation results proved that the proposed method gives a smaller misadjustment and better tracking capability than existing methods.