Characterizing the tradeoffs between different sensor allocation and management algorithms

Recent work on distributed algorithms for Sensor Allocation and Management has focused on heuristic methods to address the challenges posed by limited processing and communication resources. Heuristic methods tend to have good real-time performance, but are often lacking in system optimality. This paper explores the tradeoff by proposing a spectrum of algorithms starting with centralized math programming based algorithms having provable optimality and convergence characteristics, followed by increasingly distributed math programming based algorithms. The proposed algorithms tend to use differing mixes of math programming vs. heuristic techniques. We provide an important characterization of the proposed sensor allocation and management algorithms along two important dimensions: degree of centralization, and degree of heuristic content. Applying the algorithms to a multi-sensor multi-target tracking problem, we demonstrate that different algorithms tend to outperform others as operating conditions, such as sensor target ratio and target clustering, change. The results provide a strong indication that tracking systems need to maintain a family of alternate SAM algorithms employing different ones based on operating conditions.