基于改进FSRCNN模型的OFDM图像传输系统信道估计方法OA
Channel Estimation Method for OFDM Image Transmission Systems Based on an Improved FSRCNN Model
正交频分复用(Orthogonal Frequency Division Multiplexing,OFDM)技术广泛应用于现代通信系统中,但在0～10 dB低信噪比(Signal to Noise Ratio,SNR)环境下的端到端图像传输中面临信道估计精度不足和本地部署内存溢出问题.针对以上问题,在对图像进行预处理时,通过无重叠分块解决了单块处理的内存溢出问题,提升了数据集多样性.同时,提出了一种融合残差网络(Residual Network,ResNet)和坐标注意力(Coordinate Attention,CA)机制的改进快速超分辨率卷积神经网络(Fast Super-Resolution Convolutional Neural Network,FSRCNN)信道估计模型——RC-FSRCNN,该模型在增强特征提取能力的同时,缓解了深层网络出现的梯度消失问题,提升了图像传输的信道估计性能.实验结果表明,与采用了线性插值的最小二乘法(Least Squares,LS)、线性最小均方误差(Linear Minimum Mean Squared Error,LMMSE)法和原始 FSRCNN 模型相比,RC-FS-RCNN在低SNR区域表现出更低的均方误差(Mean Squared Error,MSE).该研究为OFDM图像传输系统提供了一种高效的信道估计方案.
Orthogonal Frequency Division Multiplexing(OFDM)technology is widely used in modern communication systems.However,in end-to-end image transmission under low Signal to Noise Ratio(SNR)conditions from 0 to 10 dB,OFDM faces problems of insufficient channel estimation accuracy and memory overflow in local deployment.To address these problems,non-overlapping block partitioning is adopted for image preprocessing,which solves the memory overflow problem caused by single-block processing and improves dataset diversity.Meanwhile,an improved Fast Super-Resolution Convolutional Neural Network(FSRCNN)channel estimation model fusing Residual Network(ResNet)and Coordinate Attention(CA)mechanism is proposed,denoted as ResNet CA Fast Super-Resolution Convolutional Neural Network(RC-FSRCNN).The proposed model enhances feature extraction capability and alleviates the gradient vanishing problem in deep networks,thereby improving channel estimation performance for image transmission.Experimental results show that,compared with Least Square(LS)method with linear interpolation,Linear Minimum Mean Squared Error(LMMSE)method,and the original FSRCNN model,RC-FSRCNN achieves lower Mean Squared Error(MSE)in the low-SNR region.This study provides an efficient channel estimation scheme for OFDM image transmission systems.
张学刚;王涛;马爱玲;李姝慧;王昱博;李思憶
青海民族大学智能科学与工程学院,青海西宁 810007青海民族大学通信工程国家级实验教学示范中心,青海西宁 810007||青海民族大学通信工程重点实验室,青海西宁 810007青海建筑职业技术学院信息技术学院,青海西宁 810013青海民族大学智能科学与工程学院,青海西宁 810007青海民族大学智能科学与工程学院,青海西宁 810007青海民族大学智能科学与工程学院,青海西宁 810007
信息技术与安全科学
正交频分复用快速超分辨率卷积神经网络信道估计残差网络坐标注意力机制
OFDMFSRCNNchannel estimationResNetCA mechanism
《无线电工程》 2026 (4)
668-674,7
2022年度青海省重点研发与转化计划科技援青合作专项项目(2022-QY-205)2022 Key R&D and Transformation Program-Science & Technology Aid-Qinghai Cooperation Special Project(2022-QY-205)
评论