(A) study on the decentralized estimation using gain transfer for multi

Abstract

The estimation problem may posed in terms of a single sensor making measurements on a single process or, more generally, in terms of multiple sensors and multiple processes. The latter case is referred to as a multisensor system. In a multisensor system in which each individual sensor has its own "built-in`` Kalman filter, one is interested in combining the estimates from these independent data sources (i.e., the built-in Kalman filter) to generate a global estimate that will, ideally, be optimal. The study on the algorithm for integrating the multisensor measurement data begins by surveying existing theories and examining the characterization of the decentralized estimation algorithm without feedback and with feedback, and the federated filtering algorithm. All of the previous works result in the global estimate of the state in a decentralized system, but they all still require extensive calculations of the local and global inverse covariances. The decentralized estimation algorithm using gain transfer can eliminate this costly computational requirement: instead of the inverse covariances, a global processor receives the information in the form of a Kalman gain transfer from local systems and formulates the global estimate. A new decentralized estimation algorithm using gain transfer introduced in this dissertation with parallel processing capability and with more general form for the decentralized estimation. In the proposed algorithm developed here, there is no need to calculate the measurement update of the local covariances in order to obtain the global estimates. In addition, the computational burden is significantly reduced as compared to currently proposed methods that require the inversion of potentially large matrices. These characteristics make the proposed algorithm very attractive for real time multi-sensor integration. Some outstanding examples of multisensor system occur in the field of navigation and tracking. Navigation is the estimation of ...