Computer Science · Computer Networks and Communications
Technical Notes and Correspondence Decentralized Structures for Parallel Kalman Filtering
Hamid R. Hashemipour, Sumit Roy, Alan, J. Laub
See on the mapAbstract
Abstract-This note describes various multisensor network scenarios whose signal processing tasks are amenable to multiprocessor implemen-tation. The natural origins of such multitasking are emphasized and new parallel structures for state estimation using the Kalman filter are proposed that extend existing results in several directions. In particular, hierarchical network structures are developed which have the property that the optimal global estimate based on all the available information can be reconstructed from estimates computed by local processor nodes based solely on their own local information and transmitted to a central processor. The algorithms potentially yield an approximately linear speed-up rate, are reasonably failure-resistant, and are optimized with respect to communication bandwidth and memory requirements at the various processors. I.