Mode-matched filtering via the EM algorithm

We show that a generalization of the EM algorithm, the alternating expectation conditional maximization (AECM) algorithm, can be used to derive a mode matched filtering algorithm called the MMAECM. Mode-matched filtering methods are used for state estimation of jump Markov linear systems. Such models are used in a wide variety of areas in which the system switches between different modes of operation, as in target tracking. The optimal conditional mean estimator for jump Markov linear systems is of exponential complexity, hence algorithms are necessarily suboptimal. We derive the MMAECM according to the maximum a posteriori criterion. Performance of an online version of the MMAECM algorithm is compared to existing mode-matched filtering algorithms such as the interacting multiple model algorithm and generalized pseudo Bayesian methods