Detection of non-uniform byzantine attacks in collaborative spectrum sensing

In most of existing attacker detection algorithms for collaborative spectrum sensing (CSS) systems, attacking strategies are assumed to be uniform (identical). However, when non-uniform attacking strategies are adopt, the diversity of attacking strategies make it difficult to isolate the attackers. In this paper, to detect the non-uniform strategy attackers, we use the dissimilarity metric of a cognitive user (CU) to describe its identity (an honest user or an attacker), and propose a new detection algorithm which based on kernel clustering mechanism. By extensive numerical experiments, the favourable performance of the proposed algorithm is validated.