A Generalized Srif Algorithm for Systolic Implementation

Abstract

The standard square-root information filter (SRIF) algorithm for Kalman filtering has been generalized in order to cope with singular covariance matrices and descriptor systems. This generalization relies on a new array for representing statistical information on random vectors. Imposing a suitable structure on the array, i.e. that of the two relevant matrices one is triangular and the other is a non-negative signature matrix, it is possible to single out a novel algorithm with the same dependence-graph as the standard one. Therefore, available systolic designs can be exploited apart from a few modifications at the elementary processor level. Some properties of the estimation algorithm when applied to descriptor systems have been studied.