A prefiltering version of the Kalman filter with new numerical integration formulas for Riccati equations

Abstract

A prefiltering version of the Kalman filter is derived for both discrete and continuous measurements. The derivation consists of determining a single discrete measurement that is equivalent to either a time segment of continuous measurements or a set of discrete measurements. This prefiltering version of the Kalman filter easily handles numerical problems associated with rapid transients and ill-conditioned Riccati matrices. Therefore, the derived technique for extrapolating the Riccati matrix from one time to the next constitutes a new set of integration formulas which alleviate ill-conditioning problems associated with continuous Riccati equations. The prefilter extends Potter's square root filtering, with all its advantages, to continuous measurement problems. Prior to this work, no technique existed for numerically integrating the general ill-conditioned, time-varying and continuous Riccati equation.