Statistical models for energy-efficient selective communications in sensor networks

An inherent characteristic of Wireless Sensor Networks is their ability to operate with autonomy when sensor node devices are resource-constrained. Optimizing energy consumption with the goal of achieving longer sensor network lifetime is a major challenge. This thesis focuses on energy-efficient strategies based on the reduction of communication processes, the most energy expensive tasks by far. In particular, we analyze selective communication policies that allow sensor nodes to save energy resources at the same time that can assure the quantity and quality of the transmitted information. This thesis proposes selective communication strategies for energy-constrained Wireless Sensor Networks, which are based on statistical models of the information flowing through the nodes. Assuming that messages are graded according to an importance/priority value (and whose traffic can be statistically modeled) and that the energy consumption patterns of each individual node are known (or can be estimated), the design and evaluation of optimal selective communication policies that maximize the quality of the information arriving to destination along the network lifetime are analyzed. The problem is initially stated from a decision theory perspective and later reformulated as a dynamic programming problem (based on Markov Decision Processes). The total importance sum of the transmitted, forwarded or finally delivered messages are used as performance measures to design optimal transmission policies. The proposed solutions are fairly simple and based on forwarding thresholds whose values can be adaptively estimated. Simulated numerical tests, including a target tracking scenario, corroborate the analytical claims and reveal that significant energy saving can be obtained to enlarge sensor network lifetime when implementing the proposed schemes