Low-order discrete linear filters: Their optimal structure (Q1883803)

The right side of the difference equation describing a nonlinear filter of order equal to the number of estimated components of the Markov vector of the state of the observed object is determined. The presented modification of the nonlinear filter is based on the determination of an optimal filter in the class of filters of arbitrary structure -- this class being wider than the class used in the conditionally optimal nonlinear filtering method. The minimum of the mean-square estimation error as the optimality criterion is used. Relations for exactly determining the mean-square optimal structure of the filter, the joint probability distributions of the object and filter states, and the measurer output are derived. They are used in developing approximate methods of synthesizing suboptimal, low-order, nonlinear filters. An example is given.