A distributed and kernel-based scheme for location verification in wireless sensor networks

A technical problem central to wireless sensor networks is location verification: how to accurately detect sensors' location information that are damaged or deliberately compromised. The need for this is evident in, e.g., battleground surveillance, where compromised sensors may incur severe damage. Previous approaches to tackling this problem are mostly centralized: a sensor node is designated as the sink, which collects location information from all nodes, conducts analysis and detects possible attacks. These, however, have to bear with prolonged response time, high energy consumption and low reliability. This paper proposes a novel location verification scheme named LPV. In contrast to previous approaches, LPV is distributed: the falsified locations of sensors are detected by conducting evaluation analysis on information of their neighboring nodes.We show that in LPV, it suffices to detect malicious nodes when each sensor only collects the information of neighboring nodes within one hops. Another advantage of LPV is, it consider the impact of localization error. We proposed a kernel-based approach that approximates the distribution of the localization results. We evaluate the correctness of distribution to detect localization attack.We verify, via formal analysis and experimental study in various settings, that LPV has performance comparable to centralized approaches in accurately identifying falsified sensors with higher reliability.