Signal Classification for Detecting Primary User Emulation Attack in Centralized Cognitive Radio Networks

This paper presents a Cognitive Radio Network (CRN) consisting of
$N$
Secondary Users (SUs),
$M$
real Primary Users (PUs), and one attacker where the main threat in such a network is the Primary User Emulation (PUE) attack. It is assumed that
$K$
active PUs coexist in the communication range of CRN where parameter
$K\leq M$
is unknown for the PUE detection system. We aim to detect malicious PUE attackers, where the CR base station performs the feature extraction on the PU transmissions in the frequency domain to recognize SUs' and PUs' (either real PUs or PUE attacker) signals, independently. Then, the Bayesian nonparametric clustering approach based on the Dirichlet Process Mixture Model (DPMM) is used for clustering and determining the number of detected PUs. To reduce the convergence time of the Gibbs sampling in our scheme, we employ the Chinese Restaurant Process (CRP) with the uniform and nonuniform sampling of unknown clusters' parameters. We investigate the effect of choosing suitable initial assignments of clusters on the convergence time of the DPMM algorithm. To separate the real PU from the PUE attacker, the CRN performs an identification mechanism sequentially in some spectrum sensing periods to investigate the behavior of PU transmitters. The performance of the proposed PUEA identification method in terms of the PU detection rate and the PUEA detection accuracy is evaluated numerically for various sequential (e.g., CRP-sequential) and random sampling approaches.