Detection and Localization in Sensor Networks Using Distributed FDR

Erhan Baki Ermis, Venkatesh Saligrama

Abstract: Wireless sensor networks (SNET) have gained substantial interest for detection and localization of objects, however network energy constraints make it difficult to implement optimal solutions in distributed settings. For object localization with known power, maximum likelihood estimation is the optimal solution, however in many applications it involves performing optimization over a non-convex function, which can be hard to solve even in centralized schemes. Furthermore, the computational complexity of finding this solution is prohibitively large to be implemented in distributed SNETs under energy constraints. In this work we consider the problem of distributed detection and localization of an object that emits a signal with unknown power. Considering the energy constraints of SNETs, we propose a novel technique that makes use of the false discovery rate (FDR) procedure and a belief propagation (BP) like algorithm for detection and localization problems. Inclusion of FDR to the detection process limits the communications necessary to detect the presence of the object as well as the energy consumption to locate it, prolonging the network lifetime. The simulation studies show that this approach is very well suited for detection and localization problems where the signal power of the object decays rapidly with distance.