Noise adaptation and correlated maneuver gating of an extended Kalman filter

Abstract

Extended Kalman filtering is used to provide estimates of the position and velocity of a target based upon observations of the target's bearing and range. Non-stationary noise is shown to degrade the performance of the filter and cause filter divergence. By estimating the noise power from the variance of the filter's residual we adapt the filter to compensate for varying noise power. This thesis also introduces the method of correlated maneuver gating to adapt the Kalman filter to target dynamics. By spatially and temporally correlating the Mahalanobis Distance of the residual, the Kalman filter's performance is increased while tracking tangentially accelerating targets. Monte Carlo simulations are run for three different sets of target dynamics: stationary, moving linearly, and accelerating tangentially. Results for the simulation show significant performance advantages of using correlated maneuver gating in conjunction with noise adaptation. These results should generalize to other applications of the extended Kalman filter whose state and observation spaces enjoy a one-to-one mapping