Semantic Communication-Enhanced U-Shaped Split Federated Learning With Adaptive Compression for Vehicular Networks

Federated learning (FL) in vehicular networks offers the advantage of leveraging extensive distributed vehicle data while addressing data privacy and communication overhead concerns. However, conventional FL approaches encounter substantial difficulties due to the limited communication bandwidth and network dynamics in vehicular environments. To mitigate these challenges, this paper introduces a Semantic Communication-Enhanced U-Shaped Federated Split Learning (SC-USFL) framework tailored for image classification tasks in vehicular networks. The proposed framework incorporates a task-oriented semantic communication module, comprising semantic encoder/decoder and channel encoder/decoder components, to efficiently compress and transmit task-relevant features, thereby significantly reducing communication overhead. Additionally, a network status monitor (NSM) module is developed to dynamically adjust the compression ratio (CR) based on real-time network conditions, ensuring an effective balance between communication efficiency and classification performance. Extensive simulations on the CI-FAR10 dataset demonstrate that the proposed SC-USFL framework effectively manages the trade-off between communication overhead and classification accuracy. Furthermore, the results indicate robust performance over both additive white Gaussian noise (AWGN) and Rayleigh fading channel conditions, validating the effectiveness and adaptability of the proposed framework.