Maximum entropy principle in problems of identification of a class of nonstationary signals (Q1898574)

An identification problem for a certain class of nonstationary stochastic signals is investigated by applying the maximum entropy principle. The analytic form of maximum entropy spectral estimator is derived, and the criteria for evaluation of the lengths of homogeneity time intervals, within which the process under investigation fits well (up to a given degree) an adopted model, are presented for a class of autoregressive processes using a priori information about the process. Possible extensions of the results obtained are pointed out.