Content area
Vulnerability detection is crucial for defending against cyber threats and protecting wireless communication systems. Despite advancements in robust detection methods, such as machine learning and scalable cloud-based vulnerability detection, existing approaches to automatic vulnerability detection still face several limitations: the lack of fully automated protocol-based vulnerability detection, heavy dependence on computational resources for detecting implementation vulnerabilities, and the inability to update learned attack patterns during runtime.
This dissertation presents an advanced vulnerability detection framework that addresses these gaps through three key contributions. First, we developed a pretrained large language model-based extractor for formal properties, enabling the automatic translation of wireless protocols into formal verification formats. This method achieved over 97% classification accuracy on the 3GPP RRC protocol, supporting effective formal verification. Second, we designed a formal-guided fuzz testing framework that integrates protocol analysis with a digital twin testing platform, allowing efficient detection of high-risk vulnerabilities. Third, we introduced a probability-based strategy that reduces the exponential growth of time complexity in vulnerability testing to a linear process, significantly minimizing computational overhead.
Together, these contributions form a unified, automated vulnerability detection system that combines formal methods, dynamic analysis, and adaptive runtime pattern recognition to enhance cybersecurity in wireless systems.