Large Language Model (LLM)-based Indoor Wireless Network Planning

Efficient indoor wireless network (IWN) planning is crucial for providing high-quality 5G in-building services. However, existing meta-heuristic and artificial intelligence (AI)-based planning methods, such as deep reinforcement learning (DRL) suffer from enormous computational complexity, training overhead, heavy manual pre-design, and susceptibility to local optima. Fortunately, large language models (LLMs) have emerged with tremendous attention, which are distinguished by their ability to generate human-like contextually appropriate responses due to their extensive parameters, trained on vast text datasets covering diverse domains of human life. Recent works also enable LLMs to act as optimizers through meta-prompts and solve black-box problems. Therefore, in this thesis, we aim to investigate and explore the LLM-based framework design, methodology proposal, and simulation validation in three typical IWN planning optimization problems.
The first part aims to simultaneously optimize indoor wireless and daylight performance by adjusting the positions of windows and the beam directions of window-deployed reconfigurable intelligent surfaces (RISs) for RIS-aided outdoor-to-indoor (O2I) networks utilizing LLMs as optimizers. In the second work, we are the first to investigate the joint deployment optimization of multiple small-cell base stations (SBSs) and RISs by adjusting their quantity and locations for indoor millimetre wave (mmWave) networks via LLMs. The third focus point proposes two general LLM-based frameworks to provide efficient IWN planning for high-quality 5G in building services. Traditional meta-heuristic and artificial intelligence-based planning methods face significant challenges due to the intricate interplay between indoor environments (IEs) and IWN demands. Therefore, we present an indoor wireless network Planning with large LANguage models (iPLAN) framework, which integrates multi-modal IE representations into LLM-powered optimizers to improve
IWN planning. Finally, we conclude the effectiveness and future works of LLM optimizers, inaugurating a new class of IWN planning tools.