Conditionally optimal filtering and recognition of signals in stochastic differential systems (Q1284311)

A partially observed vector signal comes from one of \(n\) systems of Ito stochastic differential equations. The author considers the problem of joint filtration and recognition, i.e., at each time \(t\), providing an optimal estimate for the unobserved component of the signal and deciding which system produced it. The decision criterion is the maximal a posteriori probability. The estimators are based on Bayesian conditionally optimal filtering. The problem is solved in a quite general setting. All coefficients in the Ito equations may be nonlinear, except for the drift terms in the equations for the observed component which should be linear with respect to the unobserved component (but the dependence on the remaining components may be nonlinear). The most laborious computations involve only the model parameters (not the data) and can be done in advance.