WHO: World Model Approach for Handover Optimization in 5G Networks
Recently, the development and deployment of intelligent controllers for radio access networks (RAN) has attracted significant attention from network operators and international telecommunications organizations, driven by rapid advances in artificial intelligence. Mobility management plays a fundamental role in ensuring seamless connectivity and service quality in 5G RAN. In fact, optimal control in 5G RAN is highly challenging due to its complex, dynamic, and distributed environment. Many approaches have been proposed to address this problem, particularly those based on deep reinforcement learning (DRL). However, contrary to the dense reward assumption in many DRL-based studies, mobility feedback in practical RAN environments is characteristically sparse and delayed. In this paper, we propose WHO (World Model for Handover Optimization), a novel method designed to bridge the gap between sparse feedback and efficient learning in 5G networks. WHO utilizes a world model to convert event-driven rewards into dense predictive signals, facilitating robust multi-agent optimization. Field experiments involving 13 base stations and 39 cells show that the proposed method significantly improves handover performance and network stability compared to conventional DRL approaches, achieving 19–40% higher prediction precision and up to 32% improvement in key performance indicators (KPIs).