Neighbor dependency-based dynamic fusion tree for a multi-radar system

Abstract

A fusion process, which eliminates duplicate tracks, is required to create a single integrated air picture (SIAP) with air tracks collected from a dynamic multi-radar system (DMRS). A DMRS consists of numerous vehicle-mounted radar nodes that are connected via a wireless ad hoc network. Meanwhile, two-tier fusion process is a suitable approach to create a SIAP in a DMRS. Since, the two-tier fusion process is divided into local and global parts, so that the number of tracks to be processed at the central server can be reduced, dramatically. Local fusion nodes execute the local fusion process to create the local SIAP, and the central server executes the global fusion process to create the global SIAP with the tracks from local fusion nodes. This hierarchical structure can be modeled as a fusion tree: Each radar node, local fusion node, and the central server is a leaf, internode, and the root, respectively. This dissertation addresses the proposal that the number of processed air tracks of a two-tier fusion process can be increased by applying a balanced fusion tree, which can balance tracks across local fusion nodes. In this dissertation, fusion tree generation (FTG) algorithms, based on clustering approach, are proposed for a DMRS. Basic FTG (B-FTG) algorithm, which can generate a balanced fusion tree by considering the distribution of radar nodes, is proposed. The performance of B-FTG is evaluated on the OPNET (Optimized Network Engineering Tool) network simulator

and the simulation result shows that B-FTG outperforms a na？ve method, which randomly select local fusion nodes by considering the position of radar nodes, when used to generate balanced fusion trees. This result reveals that the effectiveness of clustering approach in generating fusion trees. In addition, Non-uniform FTG (NU-FTG) algorithm, which can generate a balanced fusion tree in the non-uniform distribution of targets by considering the distribution of targets, is proposed. NU-FTG outperforms B-FTG and clustering methods for the wireless sensor networks (WSNs) when used to generate balanced fusion trees in the non-uniform distribution of targets. However, it is observed that the performance of B-FTG and NU-FTG are degraded in a dynamic environment where the distribution of radar nodes and targets changes dynamically. This performance degradation of B-FTG and NU-FTG might be induced from the property that those algorithms consider the initial distribution of radar nodes and targets only. In this dissertation, the dynamic FTG (D-FTG), which can be used in the dynamic environment, is proposed. D-FTG consists of initial clustering and pruning-and-rejoining stages. The neighbor dependency-based scoring method is devised to improve the performance of the initial clustering. Besides, the pruning-and-rejoining strategy is proposed to prevent some of local fusion nodes from becoming bottleneck nodes.