AI-based limited data spectrum sensing for cognitive radio networks

• Main Author: Liu, Jianqiao
• Other Authors: Teh Kah Chan
• Format: Thesis-Master by Research
• Language: English
• Published: Nanyang Technological University 2025
• Subjects: Engineering
• Online Access: https://hdl.handle.net/10356/182462
• Institution: Nanyang Technological University

This dissertation presents an advanced approach to spectrum sensing in cognitive radio networks by integrating deep learning techniques, specifically Convolutional Neural Networks (CNNs). Through simulations of quadrature phase-shift keying (QPSK) and 8 phase-shift keying (8PSK) signals in MATLAB and subsequent training of CNN models in Python using PyTorch, the study demonstrates the superior performance of deep learning-based methods over traditional spectrum sensing techniques. The proposed feature extraction method and the use of diverse signal features lead to enhanced detection and reduced false alarm probabilities, laying the groundwork for future research in optimizing spectrum sensing models with more sophisticated CNN architectures. By simulating the entire signal transmission process multiple times, characteristic values corresponding to energy, power and cycle stability are collected, combined for different frequency domain environments and used to train targeted AI models, which can improve the recognition accuracy of the idle frequency domain with limited data. The findings underscore the potential of deep learning to revolutionize spectrum management and utilization.