Location estimation in sensor networks.

Abstract

In wireless sensor networks, self-localizing sensors are required in a wide variety of applications, from environmental monitoring and manufacturing logistics to geographic routing. In sensor networks which measure high-dimensional data, data localization is also a means to visualize the relationships between sensors' high-dimensional data in a low-dimensional display. This thesis considers both to be part of the general problem of estimating the coordinates of networked sensors. Sensor network localization is 'cooperative' in the sense that sensors work locally, with neighboring sensors in the network, to measure relative location, and then estimate a global map of the network. The choice of sensor measurement technology plays a major role in the network's localization accuracy, energy and bandwidth efficiency, and device cost. This thesis considers measurements of time-of-arrival (TOA), received signal strength (RSS), quantized received signal strength (QRSS), and connectivity. Extensive RF measurement campaigns were conducted, and the statistical characterization and models which resulted from them are reported. From these models, Cramer-Rao lower bounds on the variance possible from unbiased location estimators are derived and studied. Next, several cooperative location estimation algorithms are developed and presented, for both centralized and distributed implementations. Manifold learning-based algorithms are shown to be particularly effective, in particular, when combined with adaptive neighbor selection methods. Finally, these cooperative localization algorithms are shown to be useful in internet traffic visualization to help show when an anomaly event (such as a port or network scan) is occurring, and to help answer questions about the place and features affected by the anomalous event.