Large Language Models for Intelligent Communication Networks: Architectures, Applications, and Future Directions

Traditional rule-based systems cannot meet the requirements of dynamic environments, diverse data, and real-time optimization as communication networks grow more intelligent and complex. This paper offers a thorough analysis of the ways in which large models, particularly large language models and generative artificial intelligence, are changing the communication environment. This paper examines how these models enhance the capabilities of communication systems at every OSI model layer, from physical signal processing to application-layer service orchestration, starting with the classic Transformer architecture and newer alternatives like Mamba, KAN, and RWKV. In addition, this paper discusses significant technical mechanisms for privacy-preserving model training, including data pre-processing, multi-task learning, edge deployment, model compression, and federated training. Large models offer notable benefits in terms of flexibility, cross-task generalization, and astute decision-making; however, they also faced real-world difficulties with regard to computational expense, latency, robustness, and explainability. Future directions for customized models, hybrid large-small model frameworks, green Artificial Intelligence (Al), and interdisciplinary integration are discussed in the paper's conclusion. It is anticipated that these developments will usher in a new era of intelligent and sustainable global communication by transforming 5G and 6G networks into more intelligent, effective, and user-centric systems.