Study Of Distributed Detection In Wireless Sensor Networks Under Byzantine Attacks

Wireless sensor networks consist of a large of number of tiny power-limited sensors than are densely and spatially deployed to monitor physical phenomena. When detecting a target in the region of interest, all the sensors report their local decisions to the fusion center where a global-decision is made. For the advantage of easy of deployment and fast self-organization, WSNs have been widely used in target detection. Because of its increasing importance, it’s imperative to incorporate secure target detection into WSNs. The most often appeared threaten is Byzantine attacks where some authenticated sensor nodes have been fully controlled by an intelligent adversary. These compromised sensors are reprogrammed and then forced to send false information to the FC in order to confuse it.In this thesis, distributed detection in wireless sensor networks under Byzantine attacks is studied. A network contains both honest and Byzantine sensors. First, in the perspective of Byzantine attackers, independent malicious byzantine attacks and collaborative malicious byzantine attacks have been considered and analyzed. Based on these two attacking strategies’ merits and drawbacks, a new kind of attacking strategy is proposed in our dissertation, which is named after neighborhood malicious byzantine attacks. Blind attacking power is Byzantine attacking power when the fusion center is blinded completely. We then analyze the blind attacking power of the three attacking strategy through Kullback-Leibler divergence and a closed-form expressions are derived for them respectively and the condition of optimal Byzantine attacking is got. We see that the attacking performance of neighborhood malicious byzantine attacks is very close to that of collaborative malicious byzantine attacks and always outperforms that of independent malicious byzantine attacks. Numerical results are presented to consolidate our conclusions. One of the efficient ways to depress the effect of Byzantine attacking is to find Byzantine sensors out. Simple voting is studied and a more reasonable threshold is given. We prove that voting based on threshold proposed by us works well. However, it loses its power when there are Byzantine sensors in wireless sensor networks. Last, in the perspective of fusion center, we propose two kinds of distance discrepancy to help find Byzantine sensors out. We see that the Byzantine identification is efficient. After that we construct a robust decision fusion rule which is based on Byzantine identification to combat with Byzantine attacks.