LLM-POWERED RAN AUTOMATION USING RETRIEVAL-AUGMENTED GENERATION (RAG)

Abstract

The increasing complexity of Radio Access Network (RAN) environments, especially 5G and future 6G infrastructures, has prompted the development of smarter and more flexible network automation infrastructures. As a more advanced form of context-driven decision-making and process automation in wireless networks, Large Language Models (LLMs) have recently been refined using Retrieval-Augmented Generation (RAG). This paper reviews current developments in applying RAG-augmented LLMs to RAN automation, including spectrum and power allocation, fault detection in distributed RANs, and secure 5G/6G multi-agent automation. It also presents comparative studies with more conventional approaches, such as Deep Reinforcement Learning (DRL), and discusses multi-agent systems, graph-based retrieval mechanisms, and agentic AI systems. The review highlights potential limitations, including safety concerns, data management challenges, and scalability issues, as well as future research and implementation directions. The discussion demonstrates the disruptive potential of RAG-enhanced LLMs in reshaping automation and intelligence in next-generation wireless networks.