Adaptive Kalman filter enhanced with spectrum analysis to estimate guidance law parameters with unknown prior statistics

Abstract

Adaptive Kalman filtering was proposed to handle unknown prior statistics decades ago. The kernel idea of published mainstream works is to use residual-based adaptive estimation or its improved form to determine unknown noise levels to complete iteration equations inside Kalman filter. In this paper, innovative method of noise level estimation is proposed to enhance adaptive Kalman filter. Based on the observation that signal and noise concentrate on different frequency ranges, cluster analysis in frequency domain is made to determine the partitioning frequency and a digital filter is then applied to separate signal from noise. In the proposed method, the noise level and corresponding covariance matrices for adaptive Kalman filter is estimated from measured or predicted target variables separately, so the result is better than the existing residual-based adaptive estimation, which suffers from interferences between variables. The recognition of the guidance law of a chasing vehicle is highly valuable for predicting its future trajectory, evaluating, and optimizing the evasive guidance law of chased vehicle. It is a hard problem to recognize the guidance law parameters due to lack of prior knowledge about the chasing vehicle. In this paper, enhanced frequency divided adaptive Kalman filter is applied to perform guidance law parameter recognition. Digital simulation results show that in several different conditions, the recognition algorithm can provide better results than Sage-Husa.