SDN-Driven Bandwidth Allocation for Federated Learning of Driving Policies Across Autonomous Vehicle Swarms

Abstract

This paper presents a joint bandwidth allocation and federated learning optimization framework for Software-Defined Networking (SDN)-assisted autonomous vehicular networks. The framework addresses challenges such as dynamic channel conditions, mobility-induced fading, and heterogeneous data by minimizing convergence bounds, latency, and energy consumption while maximizing fairness. Using a multi-objective optimization approach and Gaussian Process Regression for mobility prediction, the model enables proactive, channel-aware resource management. Three solvers—Sequential Quadratic Programming (SQP), Interior Point, and Active Set—are evaluated, with the Interior Point method showing the best performance in convergence and efficiency. This work lays the foundation for future research into more complex resource management strategies in SDN-enabled vehicular networks.