Abstract
An efficient, federated Kalman filtering method is presented, based on rigorous information-sharing principles. The method applies to decentralized navigation systems in which one or more sensor-dedicated local filters feed a larger master filter. The local filters operate in parallel, processing unique data from their local sensors, and common data from a shared inertial navigation system. The master filter combines local filter outputs at a selectable reduced rate, and yields estimates that are globally optimal or subset-optimal. The method provides major improvements in throughput (speed) and fault tolerance, and is well suited to real-time implementation. Practical federated filter examples are presented, and discussed in terms of structure, accuracy, fault tolerance, throughput, data compression, and other real-time issues.

This publication has 7 references indexed in Scilit: