Gaussian mixture approach to long range radar tracking with high range resolution

For precise, long range radars, the distribution of the polar measurements becomes highly non-Gaussian as the range resolution is increased [1]. Tracking performance with conventional Extended Kalman Filter (EKF) techniques degrades as the resolution is increased. Attempts to fit this distribution with an inflated Gaussian result in a larger-than-necessary region in which measurements may pass traditional Chi-square distance gating metrics. In this paper, a method of parameterizing the radar contact lens distribution for a given measurement is developed such that parameters for a closely fitted Gaussian mixture distribution may be derived by simple lookup and interpolation. This mixture distribution is then used to update the EKF in a way similar to the PDAF approach by combining the different measurement mixture components into a single update according to their posterior probabilities, while maintaining a tight gating region which admits a larger number of true measurements, in proportion to its Euclidian size, than traditional methods. Simulation results that compare the new approach to other popular non-linear filtering techniques are presented with respect to accuracy, consistency, and size of the gating region.