Summary
2007 International Symposium on Nonlinear Theory and its Applications
2007
Session Number:19AM1-C
Session:
Number:19AM1-C-1
Noise-assisted detection in sensor network with suboptimal fusion of optimal detections
Shin Mizutani, Kenichi Arai, Peter Davis, Naoki Wakamiya, Masayuki Murata,
pp.377-380
Publication Date:2007/9/16
Online ISSN:2188-5079
DOI:10.34385/proc.41.19AM1-C-1
Summary:
We analyze the performance of a distributed sensor network which fuses the detection results from multiple binary detectors at a single data fusion center in the presence of noise. We show the property of noise-assisted detection, whereby the detection correctness probability can be improved by adding noise. We point out that this property can be observed when the data fusion is not optimal, even when each detector has an optimal threshold. We also show the nonmonotonic behavior of noise-assisted detection for increasing added noise. Here we consider a simple model of distributed detection (DD) in which the detection result depends on the outputs of multiple identical signal detectors with common signal and independent noise. DD decides its output based on the rule of data fusion executed at the data fusion center which receives outputs from all detectors. Noise-assisted detection means that the correctness probability can be larger with noise compared to without noise. It also means that it may be possible to optimize correctness probability by adjusting the intensity of noise. Noise-assisted detection is known to occur for suboptimal detectors, and so can also be expected to occur for DD with non-optimal detectors. However, we show that noise-assisted detection in DD can occur even with optimal detectors if the rule of data fusion is suboptimal. This result is significant from the point of view of optimizing the whole system including noise levels to optimize detection.