Linear Distributed Estimation

This paper considers the problem of implementation of linear centralized estimators in a distributed fashion. In the distributed estimator structure there is a coordinator which contains the model fo the observed process, and an arbitrary number of different observers taking observations of the state process. The measurements of each observer are processed locally. One way communication links exist between each local processor and the coordinator. The results of the local estimation algorithms are communicated to the coordinator. The coordinator reconstructs the statistics of the state process based on all the available measurements, using only the communicated local statistics. The models used to produce the local estimates may be different from the process model and from each other. By constraining the choice of the local models, the derived fusion algorithm reconstructs exactly the characteristics of a centralized estimator. This paper uses a different approach to generalize much of the previous work in distributed estimation with different local models to an arbitrary number of local processing facilities. A simple example of implementation of a centralized estimator in a distributd fashion is presented.