Federated Learning–Assisted Slot-Specific SB-SPS Scheduling for Distributed C-V2X Networks*

Abstract

Cellular vehicle-to-everything (C-V2X) sidelink Mode 4 relies on sensing-based semi-persistent scheduling (SB-SPS) to enable distributed resource allocation without infrastructure support. However, conventional SB-SPS suffers from resource collisions and suboptimal slot reuse under high vehicle density, particularly in multi-lane highway environments with dynamic topology changes. This paper proposes a federated learning–assisted slot-specific SB-SPS framework that enhances distributed scheduling efficiency while preserving decentralized operation. Instead of centralized optimization, vehicles locally learn slot occupancy patterns and collaboratively update a lightweight global model through federated aggregation. The proposed approach integrates slot-level sensing statistics with adaptive candidate resource selection, enabling improved collision avoidance and resource reuse efficiency. Simulation results under representative multi-lane highway scenarios demonstrate that the proposed method significantly reduces packet collision probability and improves packet reception ratio compared with conventional SB-SPS and heuristic-based approaches while maintaining scalable and communication-efficient model updates. The results indicate that federated edge intelligence can effectively enhance distributed sidelink resource management in future B5G and 6G vehicular networks.