The Robustness ot the Fixed Point Smoothing Algorithm

Abstract

Methods of fixed point smoothing by filtering an augmented state vector are shown to be analogous to the estimation of the state of an unobservable system. Moreover, such a!gorithn s are identical to some of tbose previously developed for fixed point smoothing. An investigation reveals that tbe effective-ness of smoothing techniqunes, which are based on forward time computation, depend heavily on the accuracy of tbe covariance of error matrix and consequently lack robustness, Results of simulations of tbe continuous time fixed point smoothing algorithm are included to illustrate its characteristics. An analysis of the noise sources, when using a variable step-size integration method, is provided to validate their suitability.