An on-line reconfigurable four-ary tree-based network on chip for distributed particle filters

This paper presents an on-line reconfigurable four-ary tree-based network on chip for distributed particle filters. The proposed architecture model provides the capability of fast prototyping a large scale of particle filters system basing on the idea of reusability. Furthermore, it supports dynamic reconfiguration by regrouping PEs so as to reduce the overall traffic loads on communication infrastructure. After formulating the regrouping into number partitioning problem, we apply two heuristic algorithms which are Balanced Largest Differencing Method (BLDM) and Balanced Farthest Binding Method (BFBM) as reconfigure strategies. Finally, system level evaluation indicates that a huge decrease of traffic loads can be achieved with help of reconfigure feature.