Comparison of the unscented and cubature Kalman filters for radar tracking applications

Abstract

Among the proposed nonlinear filtering algorithms, the unscented Kalman filter (UKF) has been recommended as a better choice than other algorithms for many applications. Recently, the cubature Kalman filter (CKF) was proposed, which was claimed to be even better. This study compares the two algorithms for two radar tracking applications, namely, high frequency surface wave radar (HFSWR) and passive coherent location (PCL) radar. Monte Carlo simulations are used to fulfill the purpose. It is shown that the UKF outperforms the CKF in both radar applications, using performance measures of root mean square error (RMSE) and normalized estimation error squared (NEES). Results show that the PCL radar's higher nonlinearity provides a challenge for the design of nonlinear filters, and that the CKF is not as well suited as UKF to highly nonlinear systems such as PCL. Sensitivity of the filters becomes a critical design issue. (5 pages)