Detail of publication

Abstract

The paper deals with the design of active detector and input signal generator for stochastic discrete-time systems. The active detection problem is formulated for the jump Markov linear Gaussian model. Since the optimal solution is intractable, some simplifications and approximations are considered. Firstly, the problem of discrimination between several models and its optimal solution are obtained as a special case of the active detection problem for jump Markov linear Gaussian model. Then, two different approximations of the optimal solution are carried out to obtain feasible design techniques of the active detector and input signal generator. The first approximation is based on interchanging the minimization and expectation operators. It is shown that this approximation leads to a known approach that uses the Bayesian risk as design criterion. The second approximation is based on l-step lookahead policy and rollout algorithm. Both feasible design techniques are compared in the numerical example.