Gaussian mixture modeling for long range radar with higher representational efficiency

Recent advances in radar systems allow targets to be tracked at longer ranges with wider bandwidth waveforms. The combination of these circumstances leads to a well-known effect known as the contact-lens problem in which the measurement error distribution is highly non-Gaussian in Cartesian space. Recently, Gaussian mixture filters have been proposed to model the non-Gaussian measurement error distribution with higher fidelity. However, this approach suffers from inefficiencies in the number of components needed to achieve a useful result, as the measurement must be modeled with finer granularity as the track converges. This work presents a measurement pre-conditioning approach that may be used to improve the quality of the measurement mixture model in the vicinity of the state PDF without increasing the number of components in the model. The pre-conditioned measurement Gaussian mixture filter is then compared with other popular methods of handling the contact-lens problem in terms of estimation performance, covariance consistency, computational complexity, and gating region size. The improvement in measurement-to-track gating performance is studied and illustrated via Monte Carlo simulations.