Energy efficiency, delay, and decision fusion in clustered wireless sensor networks

Wireless sensor networks are the focus of many current research efforts because they should provide significant benefits in a wide variety of applications. These applications include smart buildings, environmental monitoring, defense, and health monitoring. Before these become reality, though, there are challenging issues that must be addressed concerning the design of these networks. These include: energy efficiency---the nodes are battery-powered yet must function for months at a time; delay---the network must gather and forward information in a timely fashion; and decision fusion---information from many nodes must be appropriately fused to make accurate decisions. These issues are addressed in this thesis. We first propose a random access scheme that is more energy efficient and incurs lower delay than existing schemes. It achieves these gains by reducing the number of collisions on the channel and the time the channel is idle by exploiting the small physical scale of these networks and the presence of sink node that is gathering the information. These characteristics allow the network to be synchronized and the retransmission strategy to be optimized through the use of a Synchronized, Shared Contention Window (SSCW). The resulting protocol performs better than traditional Binary-Exponential-Backoff (BEB) based 802.11 DCF (WiFi) and 802.15.4 (ZigBee). We next quantify the improvements that can be obtained in a clustered sensor network\u27s energy efficiency and delay behavior when directional antennas are used. A trade-off between spatial reuse and connectivity arises, and we develop an approach that yields the main lobe beamwidth that maximizes the spatio-temporal sampling rate. We also show that the more dense the network, for a fixed transmission radius, the greater the benefits of using directional antennas. The final topic is distributed detection in multi-hop, clustered sensor networks. The low-cost of sensor nodes implies that their communications suffer high bit error rates. Data gathered from a node that is many hops away is thus less reliable than data from nearby nodes. This less reliable data can still be used to improve decisions and estimates at the clusterhead when weighted appropriately. We derive the optimal weights in a MAP decision context and determine the trade-offs in decision error probability, time to reach a decision, and energy efficiency