Stealthy-Attacker Detection With a Multidimensional Feature Vector for Collaborative Spectrum Sensing

Abstract

Byzantine attackers are serious threats to collaborative spectrum sensing (CSS) systems of cognitive radio networks (CRNs). When the attackers eavesdrop on the sensing reports of honest users and fabricate their local reports according to an undetectable criterion, they can even evolve into stealthy attackers, which cannot be detected by conventional detection approaches. However, in existing works, the significant threats of stealthy attacks have yet to be well considered. In this paper, we focus on the issue of detecting stealthy attackers in CSS systems. To analyze the detectability of stealthy attackers, we propose a multidimensional metric. We prove that dissimilarity in the metrics between different subsets of cognitive radios (CRs) reveals the existence of attackers, and when the attackers fabricate their local reports to eliminate the dissimilarity, they can avoid being detected. Based on these analyses, we propose a multidimensional feature vector (FV) and its empirical form to indicate the identity (an honest user or an attacker) of a CR. By classifying the CRs according to their empirical FVs (EFVs), attackers are distinguished from honest users. The effectiveness of the EFV-based attacker detection scheme is validated by both mathematical proof and numerical experiments.