Optimizing Application Performance through Learning and Cooperation in a Wireless Sensor Network

A wireless sensor network performing surveillance in time-critical missions involving event or target tracking demands accurate ground information be delivered within a delay guarantee. Present methods solve this by using in-network fusion across all packets to reduce network load in the hope of achieving the delay guarantee. In this paper, we aim to maximize data quality from sensor fusion, while still respecting delay guarantees. The proposed method makes admission control and routing decisions using a fully distributed algorithm based on constrained Markov Decision Processes (MDPs). Cooperation is enforced through well-defined rewards and leading nodes. Assessment of data quality is derived from likelihood ratio, which is a commonly used metric in sensor fusion. We study the performance of the proposed algorithm through extensive simulations, and show that it can achieve soft delay guarantees and good data quality compared to other schemes.