AI-Powered Dynamic Spectrum Sharing In 5G Networks: A Federated Learning Approach

Dynamic Spectrum Sharing is a critical technology in 5G networks, enabling real-time allocation of spectrum resources based on traffic demand, interference, and network conditions. Traditional spectrum management approaches, often centralized and static, struggle to adapt to the heterogeneous and rapidly changing environments of modern wireless networks. Moreover, they involve high communication overhead and pose risks to data privacy. To address these limitations, this study proposes an Artificial Intelligence-powered dynamic spectral sharing framework utilizing Federated Learning, where fifth-generation (5G) base stations collaboratively train a global model without sharing raw data. Each base station performs local training based on real-time spectrum usage and transmits only model updates to a central aggregator, which combines them using the Federated Averaging algorithm. This decentralized approach preserves privacy, reduces latency, and ensures scalable, region-aware spectrum optimization. Simulation results demonstrate significant improvements in throughput, reduced latency, and enhanced spectrum efficiency compared to traditional methods. The model achieves faster convergence and adapts well to diverse environments, proving the effectiveness of FL in dynamic spectrum sharing. The proposed approach establishes a solid basis for intelligent, privacy-preserving, and scalable spectrum management in emerging network architectures.