Variational Bayesian-based Adaptive Unscented Particle Filter for BDS/INS Integration

This paper presents a variational Bayesian-based adaptive unscented particle filter (VB-AUPF) to improve the positioning accuracy of BDS/INS integrated system under the condition without accurate measurement noise covariance. This method approximates the joint posterior distribution of states and measurement noise by using variational Bayesian approach on each time step. Based on this, the measurement noise covariance is estimated with a fixed point iteration and subsequently fed back to the UPF procedure to improve the filtering accuracy. Compared with the traditional UPF, the proposed method can simultaneously estimate state and inaccurate measurement noise variance. The efficacy of the proposed method is demonstrated through the simulation of BDS/INS integrated system.