| 摘要: |
| 针对深度联合信源信道编码与正交频分复用(Deep Joint Source-Channel Coding and Orthogonal Frequency Division Multiplexing,DeepJSCC-OFDM)系统在复杂多径条件下面临的定时同步精度受限问题,提出了一种基于通道注意力的学习型定时同步方法。该方法在含定时偏移的DeepJSCC-OFDM接收模型基础上,采用一维卷积神经网络提取接收信号时域特征,并在高层特征处引入压缩激励(Squeeze-and-Excitation,SE)通道注意力模块,对与符号边界判决相关的有效通道进行自适应增强,以提高复杂多径条件下的同步定位精度。围绕同步任务构建训练数据集,并结合定时同步误差概率(Timing Synchronization Error Probability,TSEP)、图像重建峰值信噪比(Peak Signal-to-Noise Ratio,PSNR)和结构相似性(Structural Similarity Index Measure,SSIM)进行性能评估。仿真结果表明,在子载波数为128、循环前缀长度为32的典型设置下,所提方法相较基础方法的TSEP降低8.25%~24.34%;在0 dB信噪比条件下,重建图像PSNR和SSIM分别提高1.10 dB和0.05,且整体更接近理想同步结果。该方法仅将等效复数乘法数由70 240增加至70 321,在提升同步性能的同时保持了较低计算开销。 |
| 关键词: 信源信道编码 正交频分复用 定时同步 通道注意力 |
| DOI:10.20079/j.issn.1001-893x.260505001 |
|
| 基金项目:大熊猫保护研究基金(CGF2024006);大熊猫繁育研究基地开放基金(CAZG2025C04;2024CPB-C08) |
|
| Channel Attention-enhanced Timing Synchronization Method for Deep Joint Source-Channel Coded OFDM Systems |
| YANG Hongsheng,ZHENG Zhilong,LUO Lei,WANG Zhongrong |
| (1.Southwest China Institute of Electronic Technology,Chengdu 610036,China;2.School of Communications and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;3.School of Information and Communication Engineering,University of Electronic Science and Technology of China,Chengdu 611731,China) |
| Abstract: |
| To address the limited timing synchronization accuracy of deep joint source-channel coding and orthogonal frequency division multiplexing(DeepJSCC-OFDM) systems under complex multipath conditions,the authors propose a channel attention-enhanced learning-based timing synchronization method.Based on the DeepJSCC-OFDM receiver model with timing offset,the proposed method employs a one-dimensional convolutional neural network(1D-CNN) to extract temporal features from the received signal.A squeeze-and-excitation(SE) channel attention module is further introduced at the high-level feature stage to adaptively enhance effective channels related to symbol boundary decision,thereby improving timing localization accuracy under complex multipath channels.A training dataset is constructed for the timing synchronization task,and the performance is evaluated in terms of timing synchronization error probability(TSEP),peak signal-to-noise ratio(PSNR),and structural similarity index measure(SSIM) of reconstructed images.Simulation results show that,under the typical configuration with 128 subcarriers and a cyclic prefix length of 32,the proposed SE-1D-CNN method reduces the TSEP by 8.25%~24.34% compared with the conventional 1D-CNN method.At a signal-to-noise ratio(SNR) of 0 dB,the reconstructed image PSNR and SSIM are improved by 1.10 dB and 0.05,respectively,and the performance is closer to that of ideal synchronization.The proposed method only increases the number of equivalent complex multiplications from 70 240 to 70 321,while maintaining low computational overhead. |
| Key words: source-channel coding orthogonal frequency division multiplexing timing synchronization channel attention |