Hybrid Deep Learning and Multi-domain Feature Extraction for Detecting Primary User Emulation Attacks In Cognitive Radio Networks
Year: 2026
Type: article
Abstract: Primary User Emulation Attacks (PUEAs) are attacks that cause a serious threat to Cognitive Radio Networks (CRNs). In this type of attack, malicious users mimic a legitimate primary signal that led to a severe degradation of spectrum utilization. Most of the PUEA detection techniques often rely on prior knowledge of attack characteristics, which exhibit poor performance at low signal to noise ratios (SNR) and fail to identify previously unseen attack modulations. To mitigate these challenges, this paper has proposed a hybrid deep learning framework that combines the Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and Extreme Value Theory (EVT)-based OpenMax for open-set PUEA detection under Rayleigh fading conditions. Multi-domain signal features were extracted that enable reliable identification of known and unknown attackers, which is proposed in this method. This approach does not require prior knowledge of the attack. The dataset used for the evaluation of the model is simulated data that contain 20,000 signal realizations with four classes generated over an SNR that ranges from −10 to +10 dB under Rayleigh fading with additive white Gaussian noise. The results show that the proposed approach outperforms the conventional energy detection and cyclostationary methods by achieving an accuracy of 99.96% and 92.40% for both the closed and open sets, respectively, with an AUROC of 0.9803.
Source: Nigerian Journal of Engineering
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7215106957