Generalization of the theory of constrained optimal filtering to the case of nonwhite noise in observations (Q1070207)

The problem of estimating a state process which is only partially observed is of great importance; the nonlinear filtering theory generally consists of computing (or approximating) expectations conditioned by a filtration which describes the available information (the filtration of the observation process). On the other hand, in the constrained filtering theory introduced by \textit{V. S. Pugachev} [ibid 45, 212-218 (1984; Zbl 0547.93065)], one is given a class of stochastic differential equations driven by the observation process; this class is indexed by some unknown parameters which are involved linearly, and one wants to select them so that the solution of the corresponding equation approximates the state as well as possible. The problem of choosing these parameters is a somewhat complicated regression problem but its solution is only based on a priori information. Pugachev has considered the case of white noise in the observation and in this work, the author extends the study to the case of colored noise solution of a differential equation; the observation is differentiated until white noise is obtained, so that one is reduced to Pugachev's case; two examples are given. The advantage of constrained filtering is its easy implementation in real-time applications, its main disadvantage lies in the choice of the class of equations which may look arbitrary; however, it may be used for improving some classical filters.