Consider a random process whose state is observed by A’ distributed sensors. Each sensor's measurements are supplied to a nearby local station. Each station processes its observation history to produce a local conditional density function. A coordinator must reconstruct the centralized (global) conditional density of the state process, conditioned on the distributed noise corrupted observation histories of all the stations. The coordinator can only access the local conditional densities, not the observation histories themselves. The local processors’ models can differ from the coordinator's model of the distributed observation dynamics. By constraining the choice of the local models, the coordinator reconstructs exactly the centralized conditional density (as if it had access to all the measurement histories). This note solves a nonlinear distributed estimation problem using reduced-order local models. The use of local models with lower dimensions than the observed process model will lessen the local processors’ complexities or computational loads. © 1990 IEEE
