Defending dominant cooperative probabilistic attack in CRNs by JS-divergence-based improved reputation algorithm

Abstract

Rapid advances in wireless communication services has made limited spectrum resources increasingly scarce. One promising solution for enhancing spectrum utilization is cooperative spectrum sensing (CSS) in cognitive radio networks (CRNs). However CSS is vulnerable to Byzantine attack. Current researches show that Byzantine attack is easily defended for their fixed attack probability. In this context, we propose an improved attack model called the dominated cooperative probabilistic attack (DCPA) model in the actual situation, building upon the generalized collaborative probabilistic Byzantine attack model. This DCPA model contains auxiliary cooperative attackers (ACAs) who launch attacks and a dominant attacker (DA) who determines ACAs’ attack probability intervals based on their respective credibility. The DCPA model allows ACAs to flexibly launch attacks, without being identified by the traditional reputation defense algorithm, significantly compromising the sensing performance of CSS. To successfully resist the threat posed by the DCPA model to CSS, we propose a JS-divergence-based improved reputation algorithm that can distinguish honest users (HUs) from attackers. This algorithm analyzes two consecutive sensing reports to identify differences in sensing behavior between HUs and attackers. Through Python simulation analysis, we demonstrate that, compared to the generalized cooperative probabilistic attack (CPA) model and the attack-free CSS (AFC) model, the proposed DCPA model is more concealed and significantly more disruptive to the performance of traditional reputation defense algorithms. Furthermore, our approach greatly enhances the performance of CSS by promoting the participation of HUs and suppressing attackers during the final data fusion. And also compared with the PAM2 algorithm, the conventional voting rule (CVR) algorithm and the traditional reputation defense algorithm, our proposed algorithm improves the detection performance by at least 7%, 15% and 50%.