An algorithm for the indentification of a class of nonstationary systems

The paper is concerned with the problem of identification of the innovations model of a class of non-stationary stochastic processes. It is proposed that the time varying state transition matrix in the companion form be estimated first from the equivalent ARMA model. The result may then be used for estimating the time varying gain matrix and the innovations variance. The results of application of the proposed algorithm to a segment of a real speech signal are presented in order to illustrate the results.