On the robustness of adaptive predictive scheme for tracking randomly time-varying channels

This paper deals with the use of adaptive predictive algorithms in order to improve tracking performances of classical identification scheme. The performances of coupled structure depend on the prediction device. In this paper, we present a new robust algorithm for the predictive scheme. The robustness deals with insensibility of proposed algorithm to the input statistics variations, namely the power and fourth order statistics. We illustrate our contribution by performances evaluation of the proposed structure in double non stationary context: on tracking random walk channel with speech input signal.