Federated Knowledge Distillation for Hybrid Quantum-Classical End-to-End Communications: An Initial Study

End-to-end communication systems have advanced significantly with the evolution of machine learning, enabling applications such as joint source-channel coding and semantic communication (SemCom). To bring these capabilities to privacy-sensitive, distributed edge environments, federated learning (FL) provides a promising training framework. However, challenges such as limited data availability, out-of-distribution issues, and model complexity persist—especially in wireless networks with dynamic multi-user interference. To address these, we propose Fed-SKD, a federated knowledge distillation framework that leverages a static teacher model to enhance generalization. Additionally, we explore the integration of hybrid quantum-classical (HQC) networks in the channel encoder-decoder to assess the potential of quantum neural networks. Simulation results demonstrate that Fed-SKD achieves robust performance under unseen interference and limited data conditions.