Federated Learning-Based Resource Allocation for V2X Communications

Abstract

In federated learning (FL), devices contribute to global training by uploading only the local model gradients (outcomes), providing connected devices with the ability to learn while preserving privacy. FL-based resource allocation for V2X communications is proposed, referred to as FL-RA-V2X, which optimizes and maximizes the throughput of all vehicle users within the constraints of maximum power and the signal-to-interference-plus-noise ratio (SINR). It ensures fairness in resource allocation, meeting the minimum SINR requirements for cellular users and outage probability constraints for vehicle users. An approximate expression for vehicle users’ throughput is derived, eliminating the non-convexity associated with the SINR expression through iterative calculations of auxiliary variables. The resource allocation is designed to allow each vehicle user to share uplink resources with cellular users, maximizing the number of vehicle users while utilizing their maximum power transmission capability. Simulation results demonstrate the fairness and enhanced throughput efficiency of the proposed approach compared to contemporary algorithms, considering vehicle outage ratio, computational complexity, computing time, maximum transmitting power, cumulative distribution function of achievable sum rates, and convergence metrics. Furthermore, the proposed approach addresses critical aspects, including high mobility and distributed V2V communications, asynchronous training issues in cellular V2X networks, and the convergence analysis under different conditions such as varied vehicle densities and mobility patterns. These considerations broaden the applicability and robustness of the FL-RA-V2X method across diverse scenarios. The analysis also explores the impact of vehicle speed, auxiliary variables, interference effects of cellular users, and the dependence of throughput on FL iterations.