Decentralized estimation algorithms for a backward pass fixed-interval smoother

Decentralized smoothing algorithms are described for parallel-processing of the multisensor data obtained through linear discrete-time systems. The global fixed-interval smoother of backward-pass in time is used, modified so as to use the U-D factorization. Two cases are considered for the problems of decentralized smoothing and smoothing update: When the local forward-pass information filtered estimates are available, and when the local-smoothed estimates are available. It is then shown that the resulting algorithms are the dual versions of algorithms in a forward-pass realization derived by authors. The situation where the data at each local processor are to be time-sequential is also examined.