基于AAT模型的毫米波大规模MIMO系统信道估计  

AAT model based channel estimation for mmWave massive MIMO systems

作　　者：于舒娟[1] 刘荣 张昀 谢娜[1] 黄丽亚[1] YU Shujuan;LIU Rong;ZHANG Yun;XIE Na;HUANG Liya(College of Electronic and Optical Engineering&College of Flexible Electronics(Future Technology),Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]南京邮电大学电子与光学工程学院、柔性电子(未来技术学院),江苏南京210023

出　　处：《通信学报》2024年第3期41-49,共9页Journal on Communications

基　　金：国家自然科学基金资助项目(No.61977039)。

摘　　要：针对毫米波大规模多输入多输出信道具有时间相关性、系统易受噪声因素影响导致信道估计精度低的问题,提出了一种基于改进的时序卷积神经网络信道估计方法。该方法将仿真获得的信道矩阵以二维图像数据方式输入系统;利用时间相关性进行特征融合,构建集中注意力机制网络,增强系统模型对信道深层特征的提取能力;将AAN嵌入时序卷积神经网络中进行训练;系统输出去噪后的二维图像,即信道估计矩阵。仿真结果表明,所提信道估计方法在性能和复杂度方面优于传统的信道估计方法,并且当测试场景发生改变时依旧具有鲁棒性。To solve the problems of temporal correlation and susceptibility to noise in millimeter wave massive MIMO channels,which result in decreased channel estimation accuracy,a novel channel estimation method based on an improved temporal convolutional network was proposed.The channel matrices obtained from simulation were feed into the system as two-dimensional image data.The temporal correlation was utilized for feature fusion and an attention in attention network was constructed to enhance the system’s ability to extract deep channel features.Then,AAN was integrated into the temporal convolutional network for training.Finally,the system outputted a denoised two-dimensional image,namely,the channel estimation matrix.Simulation results demonstrate that the proposed method not only exhibits good performance and complexity compared to conventional channel estimation methods but also maintains robustness when the test scenario changes.

关 键 词：大规模多输入多输出信道 时序卷积神经网络 信道估计 集中注意力机制网络 

分 类 号：TN92[电子电信—通信与信息系统]