| Some attributes of a wireless sensor network (WSN) are its large scale of deployment, possible harsh ambient conditions, and limited resource budgets. Therefore, minimizing the demand placed on resource utilization is vital. WSN applications that require binary or quantized information are of interest owing to the resulting savings in bandwidth and power. Two very important applications of binary WSNs are target localization and target detection.;Optimizing a multi-modal surface to estimate a target's location and power is a highly complex problem. Deterministic and stochastic approaches are computationally intensive and sluggish or converge to local minima, causing large estimation errors. This dissertation presents a simple and effective two-stage estimator that can be used to localize a target via the optimization of a complex maximum likelihood (ML) function. The two-stage estimator allows the estimation error variance to approach the Cramer Rao lower bound (CRLB).;Although algorithms can be designed to optimize resource consumption in WSNs and result in longer lifetimes, resources are bound to be depleted over time. Events like node failures cause performance degradation. Performance-guided reconfiguration (PGR) is an algorithm proposed in this dissertation that employs redundant resources to reconfigure distributed networks in order to maintain the desired localization performance. Simulations show that after reconfiguration the localization errors are within a specified limit that PGR uses to make reconfiguration decisions.;Performance measures to characterize the detection performance of two-tier networks that use a large number of sensors are available. However, the two-tier approach does not provide enough degrees of freedom to design sensor networks to achieve a desired level of performance. Moreover, it is impractical to require a large number of sensors to transmit decisions to one fusion center. A distributed network consisting of multiple clusters helps manage a large number of sensors. A distributed network is also less susceptible to catastrophic failures, because the failure of one cluster does not imply a failure of the entire field. All of the above factors provide the necessary motivation for the development of hierarchical distributed binary sensor networks and relevant target detection performance measures. The work in this dissertation applies repeated trials to hierarchical networks, and grants the degrees of freedom necessary to design sensor networks that have an arbitrary field level performance. |
