An Expectation-Maximization-Based Estimation Algorithm for AOA Target Tracking With Non-Gaussian Measurement Noises

This paper considers angle-of-arrival (AOA) target tracking in the two-dimensional plane with non-Gaussian noise. Practical situations arise where sensor measurement noise is non-Gaussian and the standard extended Kalman filter (EKF) based tracker may not be robust, resulting in large tracking error or even lack of convergence. This motivated development of a new tracker, designed for white non-Gaussian noise cases. The noise is decomposed into a sum of Gaussian components using expectation maximization (EM), and the target state is estimated by adapting an extended Kalman filter (EKF). Adaptations are necessary to combine the EM with the iterative EKF. From the analysis of the estimation error, a closed-form expression was derived revealing the presence of a bias in the estimate. On investigation, a sensor path optimization scheme is found that can eliminate bias. Furthermore, a compensation algorithm is presented, which reduces bias for the case of static sensors. The new tracker is suitable for real-time implementation and simulation results demonstrate significant improvements for the target tracking with non-Gaussian noise.