BC-AWFedAvg: blockchain-assisted adaptive federated learning for secure RAN Slicing in beyond-5G networks

The evolution of beyond-5G networks introduces new challenges for radio resource management, particularly for heterogeneous service requirements across multiple virtual network operators. This work presents BC-AWFedAvg, a layered framework for federated deep reinforcement learning in O-RAN network slicing that integrates adaptive aggregation, blockchain-based governance, secure aggregation, and differential privacy. The proposed design separates learning, governance, and storage functions to support coordinated training while preserving privacy and limiting exposure of individual updates. Simulation results in the considered setting indicate that the proposed framework improved robustness in the considered setting under several adversarial scenarios while maintaining acceptable quality-of-service performance. These findings suggest that combining complementary mechanisms may be a promising direction for secure federated learning in next-generation wireless networks.