Particle Swarm Optimization-Based SVM for Cooperative Spectrum Sensing Against Byzantine Attack in Cognitive Wireless Sensor Networks

• Published in: IEEE sensors journal Vol. 25; no. 3; pp. 5584 - 5594
• Main Authors: Zheng, Kaifan, Wu, Jun, Cai, Fangbing, Xia, Jiahao, Xu, Xiaorong, Bao, Jianrong
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.02.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Algorithms; Byzantine attack; Channels; Classification algorithms; Cognitive radio; cognitive radio (CR); cooperative spectrum sensing (CSS); Genetic algorithms; Optimization; Particle swarm optimization; particle swarm optimization (PSO); Radio spectrum management; Sensors; Sequences; Signal detection; Signal processing; Simulation; support vector machine (SVM); Support vector machines; Training; Wireless sensor networks
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2024.3519176

In order to meet the increasing frequency demand of sensors, cognitive radio (CR) technology assists multiple sensors in detecting available channels being prioritized by the primary user (PU) through cooperative spectrum sensing (CSS) and accessing the channels underutilized by the PU. However, this paradigm provides those malicious sensor nodes (MSNs) with an opportunity to launch Byzantine attacks, severely reducing the performance and efficiency of CSS. Therefore, we characterize Byzantine behaviors as independent and joint attacks in a cognitive wireless sensor network (CWSN). To enhance the PU signal detection accuracy in the presence of a Byzantine attack, we exploit the particle swarm optimization (PSO) to optimize support vector machine (SVM) by fine-tuning penalty and kernel parameters, that is, PSO-based SVM (PSO-SVM). In addition, considering that existing algorithms require a large number of training sequences, we propose optimal PSO-SVM, which requires fewer training samples and fewer iterations to achieve more stable and accurate results about the PU's status. Finally, numerical simulation results show that the performance of the optimal PSO-SVM outperforms other PSO-SVM and genetic algorithm (GA)-based SVM (GA-SVM) and boasts 83.8% detection accuracy, 2.25-s processing time 2% more accurate than grid search-based SVM, with just 6% of its processing time.