Reliable Convolution in Point-Mass Filter for a Class of Nonlinear Models

This paper is devoted to the Bayesian state estimation of the nonlinear stochastic dynamic systems. The stress is laid on the numerical solution to the Bayesian recursive relations by the point-mass filter for a class of state-space models with linear dynamics and nonlinear measurement. In particular, a novel reliable technique for convolution computation is proposed. The technique combines the standard point-mass-based convolution with a density-weighted integration to provide accurate results even for systems with small state noise. Several implementations of the technique are developed, theoretically analysed, and evaluated in a numerical study.