A PARTITIONED DATA COMPRESSION ALGORITHM

This thesis introduces a new approach to the design of estimation algorithms. It is applicable to systems which have a special structure and the measurements involve only subsets of the total state vector.

The approach used is to partition the system into subsystems, each containing subsets of the state associated with a particular measurement type. Reduced-order filters are designed for each of these subsystems. A separate full-order filter interacts with these filters to ensure that they maintain close to unbiased (state and covariance) estimates. This full-order filter, in turn, receives compressed information from the reduced-order filters. The combination of the full-order filter and reduced-order filters has a smaller computational requirement compared to that of the optimal filter. In fact, a systematic design approach is provided wherein a trade-off between accuracy and computational requirements can be made. Computer simulation of the algorithm for typical systems has verified the predicted performance of the algorithms.

The design approach introduced in this thesis also provides a hierarchical structure in the estimation process whereby information is assimilated at different rates at different levels. As such the algorithm provides valuable insight for the design of multi-level and multi-rate estimation algorithms for systems of large dimension.