基于CSI反馈的语义图像传输

摘要：语义通信通过构建传输背景知识库、感知传输内容，大大节省传输带宽，提升传输质量。该方面的研究逐步与无线通信物理层融合，从而构建完整的语义传输系统，但现存的端到端语义编码设计无法感知不断变化的无线信道。提出了基于信道状态信息反馈的图像语义分割编码（SS-CSI）方案，根据不同子信道上的信噪比（SNR）对传输的关键特征信息进行分割和编码，并根据实际分类任务需求保护相关语义特征。仿真测试表明，SS-CSI 根据实际信道环境，在低SNR 下大幅提升了物体传输质量和分类性能，在高信噪比上则进一步传输更多背景信息，提升了整体的图像均方误差性能。

关键词：语义分割；无线通信；图像传输

Abstract: Semantic communications greatly saves transmission bandwidth and improves transmission quality by constructing transmission background knowledge base and sensing transmission content. At present, this research is gradually integrated with the wireless communication physical layer to build a complete semantic transmission system. However, the existing end-to-end semantic coding design is unable to perceive the changing wireless channel. An image semantic segmentation coding based on CSI feedback (SS-CSI) is proposed. According to the signal-to-noise ratio (SNR) on different sub-channels, the key feature information transmitted is segmented and coded. And the relevant semantic features are protected according to the actual classification requirements. Simulation results show that SS-CSI greatly improves the object transmission quality and classification performance at low SNR, and further transmits more background information at high SNR, which improves the overall image mean square error performance.

Keywords: semantic segmentation; wireless communication; image transmission