Files in this item



application/pdf3086005.pdf (5MB)Restricted to U of Illinois
(no description provided)PDF


Title:Energy -Efficient Sensor Networks for Detection Applications
Author(s):Appadwedula, Swaroop
Doctoral Committee Chair(s):Jones, Douglas L.; Veeravalli, Venugopal V.
Department / Program:Electrical Engineering
Discipline:Electrical Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Engineering, Electronics and Electrical
Abstract:Sensor networks powered by batteries or charged by energy-scavenging devices are to be used in a wide variety of applications where detection is the primary step. Faced with tight constraints on energy, bandwidth, and other system resources, the new paradigm of decentralized detection under resource constraints has emerged. To address some of the system-level costs in a sensor network, we formulate detection problems with constraints on the cost arising from transmission (sensor nodes to a fusion center) and measurement (at each sensor node). For transmission, a send/nosend scheme effectively reduces how often sensor nodes must communicate while maintaining small error and false-alarm probabilities. We find that randomization over the choice of measurement and rate of transmission can meet resource constraints and maximize the detection performance. In practice, the distribution of the observations is time varying and only partially known. Invariant tests with thresholds or parameters that can be determined independently at each sensor node are necessary to avoid frequent communication between nodes. For distance metrics and for sufficiently small communication rates in Bayesian and Neyman-Pearson problems, we find that the censoring regions can be determined independently at each sensor node, eliminating the joint optimization over the sensor nodes. The sensor decision rule is simply to transmit the likelihood ratio when it is above some threshold, chosen to meet the communication-rate constraint. Under the censoring scenario, we show that robust, locally optimum, and maximum-likelihood approaches result in simple tests and provide parameter invariance for parametric distributions. Using power measurements for existing testbeds and acoustic data collected outdoors with microphones, our approach is demonstrated in the problem of detecting machine sounds.
Issue Date:2003
Description:97 p.
Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 2003.
Other Identifier(s):(MiAaPQ)AAI3086005
Date Available in IDEALS:2015-09-25
Date Deposited:2003

This item appears in the following Collection(s)

Item Statistics