Securing Wi-Fi Networks: A Study on RF Fingerprinting and CNN-Based Intrusion Detection

Abstract

Advancements in wireless communication technology not only enhanced seamless connectivity and information exchange but also instigated the intrusion in RF networks, a pivotal challenge to the security of wireless communication networks. This thesis introduces a paradigm shift in wireless communication systems, to prevent unauthorized access attempts and malicious activities by classifying radio signals using Convolutional Neural Networks (CNNs). Diverging from conventional methods, an end-to-end deep learning model is proposed, capable of direct learning from raw time-domain signals, thereby preventing the requirement for manual feature engineering. The model is designed to extract and utilize rich, hierarchical feature representations from various radio signal types with diverse modulation techniques. Tested against a comprehensive radio signal dataset, the model demonstrates significantly enhanced classification accuracy and impressive generalization capabilities with unseen signals. The study explores the influence of different optimization algorithms on model performance, revealing how strategic parameter tuning can improve computational efficiency without compromising classification accuracy. The research not only advances the use of deep learning in radio signal classification, but also lays a foundation for future studies examining CNNs' noise resilience, interference handling, and adaptability to more intricate signals, thereby fostering the evolution of intelligent, autonomous systems in signal processing.