面向语义通信的率-失真-感知平衡理论  

The Rate-Distortion-Perception Tradeoff for Semantic Communication

作　　者：柴靖轩 肖泳 石光明 CHAI Jingxuan;XIAO Yong;SHI Guangming(School of AI,Xidian University,Xi’an 710126,China;Pazhou Laboratory(Huangpu),Guangzhou 510335,China;School of Electronic Information and Communications,Huazhong University of Science and Technology,Wuhan 430074,China;Peng Cheng Laboratory,Shenzhen 518055,China)

机构地区：[1]西安电子科技大学人工智能学院,陕西西安710126 [2]琶洲实验室(黄埔),广东广州510335 [3]华中科技大学电信学院,湖北武汉430074 [4]鹏城实验室,广东深圳518055

出　　处：《移动通信》2025年第2期70-76,共7页Mobile Communications

基　　金：国家科技重大专项“面向语义通信的语义知识库技术研究”(2024ZD1300700)。

摘　　要：语义通信因其能够满足快速发展的用户定义和以人为中心的通信服务需求而引起了广泛关注。然而,最近的研究表明,传统的香农信息理论主要关注传递并准确恢复与语义无关的符号,这不足以研究接收端恢复的消息在语义层面的感知质量。针对上述挑战,提出了一种全新的面向语义通信的率-失真-感知理论,旨在解决现有体系中未能充分考虑的语义的感知层面的问题。该研究关注两种约束:符号级别的失真控制和分布级别的感知质量约束。考虑到语义信息往往蕴含着编码器难以直接捕捉到的丰富内在知识,引入了一个间接感知机制,允许编码器通过间接方式获取语义信息。为了验证所提出的理论框架的有效性,采用真实的图像数据集进行了一系列仿真实验。实验结果表明,新的语义通信架构不仅能够在保持传统通信质量的基础上显著提升语义层面上的信息传递效率,而且还能更好地满足不同应用场景下的多样化需求。Semantic communication has attracted extensive interests recently due to its capability to meet the fast growing demand on userdefined and human-oriented communication services.However,recent study suggests that the traditional Shannon information theory,focusing mainly on delivering and accurately recovering semantic-agnostic symbols,will be insufficient to investigate the semantic-level perceptual quality of the recovered messages at the receiver.For the above challenges,a novel rate-distortionperception theory for semantic communication is proposed to address the issue of semantic perceptual quality that is not fully considered in existing systems,where two constraints of symbol-level distortion control and distribution-level perceptual quality are considered.Since the semantic information generally involves rich intrinsic knowledge that cannot always be directly observed by the encoder,an indirect perception mechanism is introduced to allow the encoder to obtain semantic information indirectly.In order to verify the effectiveness of the proposed theoretical framework,a series of simulation experiments are carried out using real image datasets.Experimental results show that the novel semantic communication architecture can not only significantly improve the efficiency of information transmission at the semantic level while maintaining the quality of traditional communication,but also better meet the diverse requirements of different application scenarios.

关 键 词：语义通信 感知质量 率-失真-感知 

分 类 号：TN929.5[电子电信—通信与信息系统]