A robust adaptive procedure for solving a non Gaussian identification problem

Abstract

Consider an unknown linear time-invariant system S without control, drived by random variables with known law. We are interested in the identification of S from the output. The usual results work only under the major assumption : S is minimum phase. In our case the system S is non minimum phase and the litterature gives only a negative result : the identification is impossible for a gaussian driving noise. For a large class of other input laws we give here a solution to this problem and present some numerical results for a concrete case, origin of our study: the blind settling phase of an equalizer in data communication.