Recent Advances in Resource Allocation and Beam Prediction for Large Language Models Empowered ISAC Systems

Integrated sensing and communication (ISAC) stands as a key enabler for next-generation communication networks, designed to synergistically optimize spectral efficiency and hardware utilization through infrastructure sharing. Despite its promise, ISAC systems face challenges in balancing dynamic resource competition and performing real-time beam prediction within highly dynamic environments. This article comprehensively reviews recent advances in large language models (LLMs) assisted ISAC systems, with a specific focus on resource allocation and beam prediction. LLMs can create a unified semantic representation for sensing and communication, thereby enabling efficient resource allocation. Leveraging multimodal fusion and cross-modal alignment mechanisms, LLMs also allow for high-accuracy beam prediction with minimal latency. This article systematically dissects the fundamental mechanisms and performance benefits of LLMs-enhanced methodologies. We also identify key challenges, such as the scarcity of high-quality multimodal data and security vulnerabilities, and outline valuable future research directions for the development of future intelligent ISAC systems.