MODELING OF NONSTATIONARY SIGNALS

In the first part of this paper we introduce a measurement of nonstati onarity based upon the correlation of an AR-model coefficients at two successive instants. In the second part we propose a new method for th e tracking of rapidly time-varying processes with a complementary mode lling. This method is based on the filtering of the signal with two co mplementary filters (high and low), both of these signals are modeled with an AR-structure. AR coefficients are determined by a version of K alman filter adapted to time-varying processes. AR-model of rapidly no nstationary signal is obtained by a synthesis of the two AR-models of filtered signals and the expression of the filters.