Sensor coverage in wireless sensor networks

Distributed and networked embedded systems equipped with sensors and actuators hold promises of making revolutionary changes in the way we observe, understand, and interact with the world. Sensor networks are one class of such networked embedded systems that are designed with the primary objective of observing physical phenomena in an environment. Fundamental design, development, deployment, and utilization issues for any sensor network are first trying to understand and characterize, and second to optimize the "quality of sensing" that the system as a whole can provide, subject to the ever-present resource constraints. In order to approach the quality of sensing, i.e., sensor coverage problem, one needs models of the sensors; the phenomena, and the environments in which the sensors interact with the phenomena. Furthermore, in collaborative settings where multiple distributed sensors participate in the sensing task, the methods and computation algorithms used, also play significant roles in the observability of phenomena and thus, must also be modeled. This thesis is an in-depth investigation of the sensor coverage problem in wireless sensor networks that follows two general directions: (1) coverage by abstraction and (2) data prediction. In the first case, three different methods of sensor abstractions and the way they observe phenomena or detect targets are used to characterize coverage: (a) 0/1 coverage, where sensors are modeled as having known sensing regions; (b) worst- and best-case coverage, where sensing effectiveness is assumed to be a monotonically decreasing function of distance from the sensor, and observability of an event is dictated by the closest, or farthest distance from the event to a sensor; and (c) exposure-based methods, where the time and the path taken by objects impact their observability. While coverage abstractions are the theme in the first case, in the second case, data prediction models are used to evaluate the level of redundancy in the data collected by the sensors. If the measurements of a sensor can be predicted by others, then that sensor is covered. In order to address the sensor prediction problem, we present the details of our Symmetric Monotonic Regression and analyze it extensively using experimental data for indoor environmental monitoring.