Differential Privacy Federated Edge Learning-assisted for Securing RAN Intelligent Controller in O-RAN 6G Communications

Privacy Preserving technique becomes an important factor especially in the Federated Learning for Open-RAN’s Radio Intelligent Controller (RICs). Existing method that integrates Differential Privacy in the present AI/ML Model training faces privacy protection challenges. This paper proposes a hybrid privacy-preserving Federated Learning (FL) framework for O-RAN RICs, integrating Secure Multi-Party Computation (SMC) for secure gradient aggregation and distributed Differential Privacy (DP) for local model update perturbation. Adaptive parameter tuning balances privacy (ϵ, δ) and model accuracy, particularly in non-IID scenarios. Our experiments on MNIST demonstrate that Gaussian DP outperforms Laplace (O(T) vs. O(T2) noise), and the hybrid approach achieves 6-12% higher accuracy than pure local DP under equivalent privacy guarantees. Our proposed framework offers a practical solution for privacy-assured and effective FL in distributed O-RAN environments.