NP-LLM: A Unified Large-Language-Model-Assisted Framework of 6G Network-Layer Planning

The emerging 6G techniques have facilitated the broadening and diversification of network coverage, connection types, and applications. They nevertheless have posed great challenges for conducting network-layer planning on large-scale 6G-driven production networks. Network-layer planning is a process of optimizing the network architecture and resources to meet certain objectives based on current traffic applications and network conditions. The state-of-the-art artificial intelligence (AI)-based solutions may suffer from scalability issues, and they could not be well-generalized to different network or traffic settings. In this work, we propose a framework that leverages large language models (LLMs), an emerging type of foundation model, to conduct traffic optimization and allocate the network resources afterward. We fully exploit LLMs' generality to apply scalable per-flow allocation to better generalize to various settings. The concatenated prompt of neural network/traffic representation and planning objective further improves the model reusability. Experiments on real-world and simulated networks have validated the superiority of our proposed framework.