Construction and Applications of Discrete-Time Smoothing Error Models.

摘要

We present a unified perspective on techniques for constructing both forward and backward markovian realizations of the error process associated with discrete-time fixed-interval smoothing. Several alternative approaches are presented, each of which provides additional insight and connections with other results in linear estimation theory. In addition, two applications of such models are described. The first is a problem in the validation of an error model for a measurement system, while the second concerns the problems of updating and combining smoothed estimates. We also present an example of the use of our updating solution for the mapping of random fields based on several sets of data tracks. Reprints.