Reliable data fusion in wireless sensor networks: A dynamic Bayesian game approach

Abstract

In this paper, we consider the reliable data fusion problem in a tactical environment in the presence of adversary. First, we characterize malicious behavior of compromised sensors assuming probabilistic models. Performance of the fusion process, in the presence of malicious sensors, is then quantified. The performance analysis shows that malicious sensors incur significant degradation on the fusion process. Our goal is to substantially improve reliability and robustness of data fusion in sensor networks. To this end, we introduce monitoring nodes (MN) that are distributed across a sensor network to detect, identify and isolate malicious sensors. We formulate the interaction between sensors and MN as a dynamic game with incomplete information, which provides a platform for designing reputation based data fusion. The reputation system ensures reliable data fusion by confining the fusion process to trustworthy sensors. We evaluate performance of the reputation system both by analysis and simulation.