Graph-Aware Hierarchical Federated Intelligence with Adaptive Differential Privacy for Robust Non-IID Spectrum Sensing in Next-Generation Wireless Networks

The rapid evolution of AI-native wireless networks has created an increasing demand for scalable, privacy-preserving, and intelligent spectrum sensing frameworks capable of operating effectively under highly heterogeneous and Non-Independent and Identically Distributed (Non-IID) data environments. Conventional federated learning-based spectrum sensing approaches often experience model divergence, excessive communication overhead, privacy-utility trade-offs, and reduced sensing accuracy in dynamic wireless scenarios. To address these challenges, this paper proposes GHFI-Net (Graph-Aware Hierarchical Federated Intelligence Network), a novel collaborative spectrum sensing framework that integrates Graph Neural Networks (GNNs), hierarchical federated learning, adaptive differential privacy, and multi-agent intelligence to enhance distributed spectrum analytics. The proposed framework introduces a graph-aware client clustering mechanism that dynamically captures spatial, temporal, and spectral relationships among distributed wireless nodes, enabling Graph Attention Networks (GATs) and GraphSAGE models to learn robust representations of spectrum occupancy. An adaptive privacy management module allocates personalized differential privacy budgets according to client reliability, data quality, and heterogeneity, thereby improving the privacy–utility balance while minimizing unnecessary perturbations. In addition, hierarchical aggregation, knowledge distillation, and multi-agent collaborative decision-making significantly reduce communication costs, accelerate model convergence, and improve scalability across edge, fog, and cloud computing layers. Experimental evaluation demonstrates that GHFI-Net achieves superior spectrum sensing accuracy, faster convergence, enhanced communication efficiency, and stronger privacy preservation compared with existing federated learning methods. The proposed framework provides a scalable, explainable, and privacy-aware foundation for autonomous spectrum management, cognitive radio intelligence, and next-generation AI-native 6G wireless communication systems.