摘要
高效的深度信源信道联合编码(deep joint source-channel coding,DeepJSCC)是实现带宽受限场景下语义通信的关键技术,然而在车联网或无人机等终端资源受限的场景中,现有方法难以适应多输入多输出(multiple-input multiple-output,MIMO)信道的动态变化,且模型庞大难以部署。为此,提出一种轻量化DeepJSCC框架(VxLJSCC)。首先,基于扩展长短期记忆网络的语义提取网络实现轻量化的高质量语义特征提取;然后,采用MIMO-信道状态信息(channel state information,CSI)预测来解决语义通信系统使用过时CSI而导致系统性能下降的问题;最后,为使语义信息充分适应时变MIMO信道质量,设计了基于信道预测的特征分配与自适应模块,结合语义特征的重要性,为不同特征分配合适的传输信道和时隙,并对特征进行调整,从而提升图像重建的语义精度。实验表明,相较于先进的DeepJSCC-MIMO方法,VxLJSCC在节省最多61.67%模型存储和77.86%计算量的情况下,仍能提供高达2.972 dB的信道增益。
Efficient deep joint source-channel coding(DeepJSCC)is a key technology for enabling semantic communication in band-width-constrained scenarios.However,in resource-limited environments such as vehicular networks or unmanned aerial vehicles,existing methods struggle to adapt to the dynamic characteristics of multiple-input multiple-output(MIMO)channels and are challenging to deploy due to their large model size.To address these issues,a lightweight DeepJSCC framework(VxLJSCC)was proposed.A semantic extraction network based on extended long short-term memory was proposed to achieve a lightweight,high quality semantic feature extraction.Then,MIMO-channel state information(CSI)prediction was employed to address the performance degradation caused by CSI aging in semantic communication systems.Finally,to adapt the semantic information to the timevarying quality of MIMO channels,a channel adaptive module was designed.This module assigns appropriate transmission subchannels and time slots to different features based on their importance,thereby enhancing the semantic accuracy of image reconstruction.Simulation results show that,compared to the best method DeepJSCC-MIMO,Vx‐LJSCC saves up to 61.67%in model storage and 77.86%in computational cost,while still providing up to 2.972 dB channel gain.
作者
于创宇
徐彦彦
潘少明
YU Chuangyu;XU Yanyan;PAN Shaoming(State Key Laboratory of Information Engineering in Surveying,Mapping and Remote Sensing,Wuhan University,Wuhan 430079,China)
出处
《电信科学》
北大核心
2025年第6期29-47,共19页
Telecommunications Science
基金
国家重点研发计划项目(No.2022YFB3903404)
湖北省技术创新计划科技重大项目(No.2024BAA011)。
关键词
信源信道联合编码
语义通信
信道状态信息
多输入多输出
图像传输
特征分配
joint source-channel coding
semantic communication
channel state information
MIMO
image transmission
feature allocation
