On the Local Sensor Signaling for Inference Centered Wireless Sensor Networks

Biao Chen

Abstract:Emerging wireless sensor network (WSN) technology has attracted steady interest among researchers from various fields. The fundamental task, ubiquitous to most WSN applications, is to convey useful information collected at distributed sensors to a decision maker. As such, much of the current work in WSN employs a data-centric approach – the objective there is to recover the original sensor observations under a fidelity measure. While appropriate for applications where the sensor observations are of primary interest (e.g., 2-D and 3-D measurement of temperature, CO2 density, etc.), this datacentric approach may waste significant resources for inference-centered WSN where the primary goal is successful assessment of certain situation (e.g., detection of a target or a hazardous event). In this work, we consider a distributed detection problem and examine the local sensor signaling schemes to achieve optimal detection performance. Taking into account the omnipresent transmission channels for any wireless systems, a central theme is the integration of processing and transmission: the design of local processing should be channel informed. The present work indeed can be considered a dual problem to that studied in [1, 2] where channel-aware decision fusion algorithms at the fusion center were developed. Two cases are presented below. Assuming perfect channel knowledge (in terms of conditional probability function), we first establish optimality condition for local sensor decision rules under a Bayesian criterion. In the absence of channel knowledge, robust sensing through integrated processing/transmission design is investigated to combat potential link failure.