An accelerated fast Newton algorithm for the identification of time-varying systems with noisy outputs - application to acoustic echo cancellation

In this work, we present a new version of the fast Newton transversal filter algorithm (FNTF). This version, which we call the accelerated one (AFNTF), is based on a simple modification of its filtering part. This modification increases the dual Kalman gain which can be provided by the prediction part of any fast RLS algorithm. The accelerated version of the FNTF algorithm improves the convergence speed as well as the tracking ability of time-varying acoustic systems. The performance properties of the proposed version and of the original one are compared and presented in this paper.