GAIN TRANSFER: AN ALGORITHM FOR DECENTRALIZED HIERARCHICAL ESTIMATION

Large scale systems that are composed of multiple sensors require the utilization of innovative methods for real-time system state estimator implementation. In fact, the classical approach to centralized estimation cannot provide satisfactory performance, since it requires the transmission of all sensor observations to a central location at large communication bandwidth. Therefore, a decentralized estimation system that distributes the processing is currently the focus of much attention.; The real-time implementation of a decentralized hierarchical estimator allows the goals of a centralized estimation system to be met, that is, the estimation of the global system parameters, yet provides reduced data communication bandwidth, improved reliability and surviveability. The Gain Transfer algorithm developed herein provides a method for decentralized hierarchical estimator implementation. Localized estimators process measurement observations locally within their own microprocessors. A global estimator independently receives information in the form of a Kalman gain transfer from these local systems. This global estimator models the overall system and thereby formulates the state estimates and state error covariances of global interest.; The Gain transfer algorithm has been developed and successfully applied to a problem in navigation. Alternative approaches to decentralized estimation have also been applied to the identical problem. The results indicate that the Gain Transfer algorithm compares favorably to other methods for decentralized estimation. The Gain Transfer method reduces the processing time considerably, with only minor sacrifice in filter performance.