Windowing and random weighting-based adaptive unscented Kalman filter

The conventional unscented Kalman filter (UKF) requires prior knowledge on system noise statistics. If the statistical characteristics of system noise are not known exactly, the filtering solution will be biased or even divergent. This paper presents an adaptive UKF by combining the windowing and random weighting concepts to address this problem. It extends the windowing concept from the linear Kalman filter to the nonlinear UKF for estimation of system noise statistics. Subsequently, the random weighting concept is adopted to refine the obtained windowing estimation by adjusting random weights of each window. The proposed adaptive UKF overcomes the limitation of the conventional UKF by online estimating and adjusting system noise statistics. Experimental results and comparison analysis demonstrate that the proposed adaptive UKF outperforms the conventional UKF and adaptive robust UKF under the condition without precise knowledge on system noise statistics