This paper addresses challenging state estimation problems where the nonlinear signal models are high dimensional along with the unavailability of noise statistics and presents a new
algorithm of variational Bayesian Transformed Cubature filter (VB-TCKF-PR) as a solution method. Conventional CKF for high dimensional signal models suffers from nonlocal sampling where the distance between mean and sample points increases with state dimension. To address this issue, the sample points are transformed using an orthogonal matrix, such that the locations of new sample points are not affected by state dimension. In absence of the foreknowledge of the noise covariance matrices variational Bayesian (VB) approach is followed to adapt the unknown elements of noise covariance matrices which are assumed to follow Inverse Wishart distribution. The superiority of VB-TCKF-PR is demonstrated over its competing algorithms with a radar tracking problem in simulation and a practical estimation problem with real data.

Orthogonal Transformation,Cubature Kalman filter,variational Bayesian,Radar tracking,Noise covariance matrices,Adaptation