| Continuing development in low-power analog and digital electronics and radio transceivers has led to new generation of matchbox-sized and battery-powered computers. These tiny devices, although individually having limited capabilities, when heavily deployed together, can wirelessly organize themselves in powerful ad hoc and sensor networks. Such networks will enable various crucial applications, such as environmental monitoring and controlling, object tracking, and military field surveillance, and may change many aspects of our lives. Since the batteries of the sensor nodes are difficult or impossible to be replenished during the course of a mission, it is essential for sensor networks to operate under energy-efficient architectures and protocols.; This dissertation considers dense wireless sensor networks that are designed for applications with continuous data delivery models such that sensors periodically report to a base-station or a data processing center. Since wireless communication is considerably more energy-expensive than signal processing or computing, we tackle the problem using three different energy-conserving techniques: clustering, which improves the communication efficiency, node scheduling, which obviates wasteful data transmission (and sensing), and distributed data compression, which reduces the wireless data volume. All the algorithms and protocols developed here are distributed and localized , requiring no global knowledge, central control or request-and-acknowledge iterations, and hence scales very well with network sizes.; We first design a clustering protocol, termed the Distributed, Energy-Efficient clustering Protocol (DEEP), in which the sensors are self-arranged in subsets, each with a leader and a group of members. Such an organization turns a flat network architecture to a hierarchical or layered architecture, and constitutes an infrastructure for efficient data routing and aggregation (if required). We next devise a parameterized sleep scheduling protocol, termed the Energy-efficient COordinated Node Scheduling (ECONS), which allows the sensors to take turns to switch off without jeopardizing the surveillance reliability. Two different flavors of ECONS, targeting flat networks and cluster-based networks, respectively, are discussed. Finally, we exploit distributed source coding (DSC) techniques to squeeze out the information redundancy not only within a sensor output but also across neighboring sensor outputs, without expensive inter-sensor communication. The rate-adaptive DSC scheme we developed makes essential use of rate-compatible punctured convolutional (RCPC) codes, has very low encoding complexity (and hence suitable for sensors with limited capability), and can adjust the compression rates to best reflect different data correlation levels. |
