Optimal multisensor integrated navigation through information space approach

Although the centralized Kalman filtering (CKF) solution is widely accepted as providing the globally optimal parameter estimation for multisensor navigation systems, it has inherent defects such as heavy communication and computational load and poor fault tolerance. To address these problems decentralized Kalman filtering (DKF) methods have been proposed. The DKF is configured as a bank of filters instead of the central filter, and aims to achieve the same level of accuracy as the CKF. This CKF-based approach however is found to be too rigorous to limit the further development of DKF algorithms. This paper proposes an alternative framework for resolving the optimal state estimation problem of multisensor integration. The data fusion algorithm is implemented through a series of transformations of vectors from one space into another. In this way, the vectors in the source information spaces are transformed into the estimate information space, where the globally optimal solution is obtained simply by a sum of these transformed vectors. The paper demonstrates how easy it is to derive the conventional DKF algorithms, such as the federated Kalman filter that has been widely applied in the multisensor navigation community. A new global optimal fusion algorithm is derived from the proposed approach. Simulation results demonstrate that the algorithm has higher accuracy than the CKF.