Abstract
In this paper, a new filter called Gaussian unscented filter (GUF) is proposed to address state filtering problems for nonlinear non-Gaussian state-space models. The GUF is similar to the unscented Kalman filter (UKF) in terms of both using the unscented transformation (UT) technique, however, the GUF is not a Kalman-type filter which provides linear minimum mean state estimates. The GUF is a sigma point realization of a Bayesian filter which sequentially estimates the prior and posterior probability density functions of states. It is therefore capable of handling non-Gaussian noises while the UKF is not. The GUF is also similar to the Gaussian particle filter (GPF) but the number of samples called sigma points necessary in the GUF algorithm is far less than the number of particles required in the Gaussian particle filter (GPF). The GUF is therefore expected to be superior in terms of the calculation cost viewpoint. We compare the state estimation accuracy and calculation time of the GUF with those for the UKF, GPF and the other particle filter using a numerical example.
