Minimax filtering in the presence of parameter uncertainties

A discrete time Minimax Filter is presented in this paper and its steady state performance is compared with the performance of other Kalman based filters. Given uncertainty in the system model, this filter is designed such that it minimizes the ma;maximum value of the cost, i.e. the trace of the steady state estimation error covariance matric, over the ranges of uncertainty. The existence of a saddle point is pointed out for uncertainties in the noise characteristics, but no longer exists for plant dynamics uncertainties.