Optimization of digital filtration algorithms for mixed Markov processes under guidance of a maneuvering object (Q1386925)

Consider the continuous model of a maneuvering object and a measuring device \[ dx/dt= F\bigl(a_j (t_k)\bigr) x(t)+ G_x\bigl(t,a_j (t_k)\bigr) \xi(t) \] \[ v(t)=H \bigl(t,b_m (t_k)\bigr)x(t) +G_y \bigl(t,b_m (t_k)\bigr) \zeta(t), \] where \(x\) and \(v\) are phase vector and measurement vector respectively; \(F,G_x,H\), and \(G_y\) are known matrices; \(\xi(t)\) and \(\zeta (t)\) are mutually independent white noises; all matrices have compatible dimensions, and \(s(t_k)= [a_j(t_k), b_m(t_k)]^T\) is a two-component discrete finite-dimensional conditional Markov process. By using the methods of conditional Markov processes [cf. \textit{R. L. Stratonovich}, Conditional Markov processes and their application to the theory of optimal control (Russian) (1966); Engl. transl. (1968; Zbl 0159.46804)], the equation for describing the joint process \([x^T,y^T]^T\), where \(dy/dt =v\), in discrete form is used as a representation of the problem. Particularly, a practically realizable digital filter is proposed, in which the algorithm is described by a system of recurrent equations and the quasi-optimization is obtained by taking into account a Gaussian approximation of the conditional a posteriori probability.