Dynamic Challenge Based Federated Learning Scenario for Network Digital Twin Aided O-RAN

This paper provides a robust and privacy-preserved federated learning (FL) framework in the open radio access network (O-RAN). The proposed solution uses the network digital twin (NDT) as an rApp in O-RAN to generate dynamic challenge scenarios specific to the network configuration prior to the local model training process in FL. The generated challenge scenario and a predetermined model success threshold are made available to the FL clients to prompt them to subsequently join the model aggregation process. The solution prevents free-riding and issues with global model accuracy in FL, while preserving local data privacy and improving compliance in a multi-vendor environment. This study enables fair and robust FL in a heterogeneous O-RAN environment and makes O-RAN networks effective and adaptive.