基于AI的端到端语义编码传输方案设计  

Design of an AI-Based End-to-End Semantic Encoding Transmission Scheme

作　　者：李立华[1] 杨琳琳[1] 任欣然 LI Lihua;YANG Linlin;REN Xinran(Beijing University of Posts and Telecommunications,Bejing 100876,China)

机构地区：[1]北京邮电大学网络与交换技术全国重点实验室,北京100876

出　　处：《移动通信》2024年第5期1-7,14,共8页Mobile Communications

基　　金：国家自然科学基金“智慧车间复杂传播环境感知、信道重构与资源配置理论研究”(92167202)。

摘　　要：面向6G多样化场景下对高性能和低复杂度的传输需求,通过将AI架构引入无线传输收发方案的联合优化,打破基于模块化物理层设计方法和传统信息论的局限,一种基于AI的端到端语义编码传输方案被提出。首先,为了克服未知衰落信道下无法通过反向传播联合训练发射机的问题,设计了一种基于CGAN的两子网架构及分阶段训练方法,有效消除衰落信道的影响。其次,进一步提出了语义信道联合编码的系统架构,在端到端的联合优化方面具有优势。仿真结果表明,所提方案通过对语义编码和传输的联合优化提升了系统性能,并适用于实际通信中信道未知场景,具备灵活性、智能性和高效性。In response to the diverse scenarios of 6G,which demand high performance and low complexity in transmission,this paper proposes an Al-based end-to-end semantic encoding transmission scheme by integrating AI architectures into the joint optimization of wireless transmission protocols.This approach transcends the limitations of traditional modular physical layer design methodologies and classical information theory.Initially,to address the challenge of training transmitters via backpropagation under unknown fading channels,a two-subnetwork architecture based on conditional generative adversarial networks(CGANs)is designed with a phased training methodology,effectively mitigating the effects of fading channels.Furthermore,a system architecture for joint semantic channel coding is introduced,which shows significant advantages in end-to-end joint optimization.Simulation results demonstrate that the proposed scheme enhances system performance through the joint optimization of semantic encoding and transmission,proving its adaptability,intelligence,and efficiency in practical communication settings where channel conditions are unknown.

关 键 词：端到端传输 信源信道联合编码 智能通信 语义通信 

分 类 号：TN92[电子电信—通信与信息系统]