A distributed expectation-maximization algorithm for OTHR multipath target tracking

Consider the problem of joint state estimation and mode identification for over-the-horizon radar (OTHR) multipath target tracking, where multiple resolvable detections corresponding to the different propagation modes come from the same target, a consensus based distributed expectation-maximization (EM) framework is proposed in this paper. By regarding the OTHR as a sensor network where each propagation mode corresponds to a sensor node, the proposed distributed EM deals with the multiple data association and state estimation jointly based on the network consensus theory. In the E-step, each mode independently calculates local state estimate by using its associated measurement. A consensus filter is used to exchange its localized estimate with its neighbors and then fuse them. In the M-step, each mode uses the estimated global state to find the local optimal measurement in the nearest neighbor sense. Based on the above iterative and consensus mechanism, the global optimal state estimate is obtained in the E-step while the local optimal measurement for each mode is chosen in the M-step. The proposed distributed EM algorithm is more accurate and computational effective than the conventional EM algorithm via a numerical example of OTHR target tracking.