Can Large Language Models Automate the Refinement of Cellular Network Specifications?

Abstract: Cellular networks serve billions of users globally, yet concerns about reliability and security persist due to weaknesses in 3GPP standards. However, traditional analysis methods, including manual inspection and automated tools, struggle with increasingly expanding cellular network specifications. This paper investigates the feasibility of LLMs for automated cellular network specification refinement. To advance it, we leverage 200,000+ approved 3GPP Change Requests (CRs) that document specification revisions, constructing a valuable dataset for domain tasks. We introduce CR-eval, a principled evaluation framework, and benchmark 16 state-of-the-art LLMs, demonstrating that top models can discover security-related weaknesses in over 127 out of 200 test cases within five trials. To bridge potential gaps, we explore LLM specialization techniques, including fine-tuning an 8B model to match or surpass advanced LLMs like GPT-4o and DeepSeek-R1. Evaluations on 30 cellular attacks identify open challenges for achieving full automation. These findings confirm that LLMs can automate the refinement of cellular network specifications and provide valuable insights to guide future research in this direction.