Edge-Intelligence-Based Dynamic Spectrum Allocation for 6G Wireless Networks

Abstract

The rapid proliferation of ultra-dense 6G networks has intensified the challenges of real-time spectrum management, as traditional centralized or static allocation methods struggle to achieve a balance among responsiveness, energy efficiency, and scalability. Existing approaches that rely on global coordination incur considerable signaling overhead and fail to adapt effectively to non-stationary wireless environments. To address these limitations, this study introduces an Edge-Intelligence-Based Dynamic Spectrum Allocation (EI-DSA) framework that integrates deep reinforcement learning (DRL) with federated learning (FL) for localized spectrum prediction and distributed decision-making. Utilizing empirical parameters derived from the NTT Docomo Tokyo 6G Testbed and Huawei Futian CBD Field Trials, the proposed framework achieves notable improvements-8.8% higher spectrum utilization, 26% lower latency, and 43% better energy efficiency-compared with centralized RL and proportional fairness baselines. The findings validate that embedding edge intelligence within radio access networks enables real-time, energy-aware, and privacy-preserving control. Theoretically, this research bridges communication engineering and intelligent optimization, presenting a scalable paradigm for AI-native 6G systems. Practically, it provides design guidance for developing green, autonomous, and adaptive wireless infrastructures that align with next-generation communication and electronic engineering advancements.