Semiparametric Estimation of Target Location in Wireless Sensor Network.

Abstract

Wireless sensor networks are widely used for monitoring natural phenomena in space and over time, as well as for target detection and tracking. In a target detection setting, sensors acquire signals emitted from the target, corrupted by background noise, and decisions are made on the presence and exact location of the target. Often, the signal propagation model is unknown in practice, as it depends on the type of target present in the monitored region. There is a rich literature on detection/tracking frameworks based on parametric signal propagation models, which are not particularly robust to misspecification. In this thesis, we introduce a few semiparametric methods of estimating the target location which does not rely on the form of the signal propagation model, and hence overcomes such issues. Further, the proposed framework for isotropic signal models incorporates a two-stage signal acquisition design that enables the utilization of only a small number of available sensors, thus reducing energy and communications costs. Both simulation studies and data examples demonstrate the utility and robustness of the proposed methods for isotropic and anisotropic signal models.