Air Interface Federated Learning

Technical Explanation

Collecting raw channel measurements from thousands of user devices to a central server incurs prohibitive uplink bandwidth and violates user privacy. Release 18 explores Federated Learning (FL) for 5G-Advanced, where UEs train local neural models on observed multi-path conditions and upload only model weight updates, which the gNodeB aggregates into an optimized global model.

Key Functions

• Trains radio AI models across thousands of distributed devices without uploading raw data
• Preserves user location and behavioral privacy during machine learning workflows
• Reduces network data transport overhead compared to centralized dataset aggregation
• Improves model robustness across diverse real-world device form-factors