LLM-assisted gNB Parameter Configuration for Radio Access Network

This paper presents a framework leveraging a large language model (LLM) for automatic parameter configuration of gNB in radio access networks (RANs). By fine-tuning the LLM using synthetic training data derived from gNB error logs, the system achieves a significant accuracy improvement in correcting misconfigurations, from 13.8% to 85.4%, and up to 92.7% with retrieval-augmented generation (RAG) on an OpenAirInterface testbed. This advancement is crucial for practitioners as it facilitates scalable, autonomous operations in RANs, reducing reliance on manual configurations and enhancing system reliability.