Approximate stochastic realization and robust prediction: algorithms for iterative solution

The related problems of (finite-length) robust prediction and maximum-entropy approximate stochastic realization are considered. Such problems are of interest in situations where there is uncertainty in the finite-length covariance data of an observed signal or time series. General properties of iterative solutions of these problems are developed, and two iterative algorithms that converge monotonically to such solutions are presented for the situation in which the uncertainty class is a simplex.<>