The Need

Current federated learning (FL) systems face significant challenges in efficiently aggregating model updates over wireless networks, especially in environments with diverse data and varying channel conditions. There is a critical need for a solution that enhances bandwidth efficiency, personalization, and adaptability to support the growing demands of future wireless communication systems, such as 6G.

The Technology

OSU engineers have developed Over-The-Air Federated Learning (OTA-FL) with Personalized Reconfigurable Intelligent Surfaces (RIS). OTA-FL allows simultaneous transmission of model updates, while RIS optimizes signal quality by adjusting phase shifts. The proposed framework, PROAR-PFed, dynamically manages power control, local training, and RIS configurations to improve both global and personalized model performance in time-varying channels.

Commercial Applications

• Smartphone and IoT Device Networks: Enhancing federated learning for personalized services
• Autonomous Vehicles: Improving communication efficiency and model accuracy for real-time decision-making
• Healthcare: Enabling secure and efficient federated learning for personalized medical treatments
• Smart Cities: Optimizing data aggregation and model training for urban infrastructure management
• Industrial IoT: Enhancing predictive maintenance and operational efficiency through improved federated learning

Benefits/Advantages

• Bandwidth Efficiency: OTA aggregation reduces bandwidth usage
• Personalization: Tailors models to individual devices, improving performance
• Adaptability: Handles dynamic wireless conditions and imperfect channel information
• Scalability: Suitable for large-scale deployments in future 6G networks
• Robustness: Enhances model accuracy and reliability in diverse data environments