Clustered decentralized binary detection: an information-theoretic approach

This paper presents an information-theoretic approach to decentralized binary detection in sensor networks. In particular, we consider a Bayesian approach for the minimization of the probability of decision error. Two scenarios are considered: (i) a scenario where clusters are identical (uniform clustering) and (ii) a scenario where clusters are different (non-uniform clustering). The performance analysis obtained with a classical “communication-theoretic” approach is extended to the “information-theoretic” realm using the concept of mutual information. We then propose a simplified binary symmetric channel (BSC) model to analyze the clustered schemes, and we show that it allows to accurately predict their realistic performance. Our results show that uniform clustering leads to negligible performance degradation beyond the first clustering subdivision. Moreover, we show that for a given value of the system mutual information, the probability of decision error is uniquely determined. The results predicted by the analytical framework are confirmed by simulations.