Discrete stochastic approximation algorithms for design of optimal sensor fusion rules

Abstract

The basic idea of distributed detection is to have a number of independent sensors, each to make a local decision (typically a binary one) and then to combine their decisions at a fusion centre to make a global decision. Fault-tolerance has been considered as one of the main characteristics of wireless sensor networks. A fusion rule in the form of an error correction code has been recently proposed for better fault-tolerance in distributed sensor networks. In this paper, we propose to employ the powerful discrete stochastic approximation techniques to optimise the code matrix, that is, the fusion rule, with the objective of minimising the probability of decision error. We consider both the standard stochastic approximation algorithm and two newly proposed ones for this application. Extensive simulation results are provided to demonstrate the effectiveness of the proposed design paradigm in obtaining optimal fusion rules in distributed wireless sensor networks.