面向语义通信的非线性变换编码  

Nonlinear transform coding for semantic communications

作　　者：张平[1] 戴金晟 张育铭 王思贤 秦晓琦 牛凯 ZHANG Ping;DAI Jincheng;ZHANG Yuming;WANG Sixian;QIN Xiaoqi;NIU Kai(The State Key Laboratory of Networking and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China;The Key Laboratory of Universal Wireless Communications,Ministry of Education,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]北京邮电大学网络与交换技术国家重点实验室,北京100876 [2]北京邮电大学泛网无线通信教育部重点实验室,北京100876

出　　处：《通信学报》2023年第4期1-14,共14页Journal on Communications

基　　金：国家自然科学基金资助项目(No.62293481,No.92067202,No.62001049,No.62071058)。

摘　　要：针对经典通信系统模块设计分离、处理范式受限等因素限制了端到端信息传输性能持续提升的问题,提出了面向语义通信的非线性变换编码传输新框架。首先,基于变分理论推导了语义通信端到端率失真优化准则。据此,设计了非线性变换来提取信源数据在语义隐空间的紧致表征,并通过语义变分熵建模引导实现了变速率非线性联合信源信道编码。实验表明,语义非线性变换编码能显著提升端到端数据传输性能及鲁棒性,是实现语义通信的关键技术之一。The modular design and limited processing mechanism of traditional communication systems limit the continuous improvement of end-to-end data transmission capability.For this reason,a new nonlinear transform coding framework for semantic communications was proposed.First,an end-to-end rate distortion optimization criterion for semantic communication was derived based on variational theory.Based on this,a nonlinear transform was designed to extract the compact representation of source data in the semantic latent space,and variable-rate nonlinear joint source-channel coding was implemented through the guidance of variational entropy model.Experiments show that semantic nonlinear transform coding can significantly improve the end-to-end data transmission performance and robustness,and is one of the key technologies to catalyze future semantic communications.

关 键 词：语义通信 非线性变换 非线性编码 变分熵建模 率失真优化 

分 类 号：TN92[电子电信—通信与信息系统]