| 摘要: |
| 在户外光线追踪场景下，针对毫米波大规模多输入多输出（Multiple-Input Multiple-Output，MIMO）系统受户外环境噪声干扰导致估计精度低的问题，提出了一种融合残差挤压激励网络(Squeeze-and-Excitation Network，SENet)的条件生成对抗网络的信道估计方法。该方法采用条件生成对抗网络将低分辨率接收信号重建为高分辨率的原始信号完成信道估计，同时在生成器网络中引入SENet网络模块来抑制户外场景下显著性噪声干扰，提高估计精度；最后将残差网络中的残差块添加到SENet的放缩操作后，提高条件生成对抗网络的收敛速度。仿真结果表明，相较于正交匹配追踪算法、卷积神经网络、去噪卷积神经网络和条件生成对抗网络算法，所提方法在户外噪声环境下估计精度平均提高了约2.2 dB,且在高噪声强度下估计精度的提高更为显著。 |
| 关键词: 毫米波大规模MIMO 信道估计 条件生成对抗网络(CGAN) 残差挤压激励网络(SENet) |
| DOI：10.20079/j.issn.1001-893x.220808001 |
|
| 基金项目:国家自然科学基金资助项目（61931004） |
|
| mmWave Massive MIMO Channel Estimation Fused with Residual SENet |
| LIU Qingli,b,YANG Guoqiang |
| (a.School of Information Engineering;b.Communication and Network Laboratory,Dalian University,Dalian 116622,China) |
| Abstract: |
| In the outdoor ray tracing scene,to solve the problem of low estimation accuracy of channel estimation caused by outdoor ambient noise interference in mmWave massive multiple-input multiple-output(MIMO) systems,a channel estimation method based on the Conditional Generative Adversarial Network(CGAN) of fused residual Squeeze-and-Excitation Network(SENet) is proposed.The method uses the CGAN to reconstruct the low resolution received signal into the original signal with high resolution to complete channel estimation.At the same time,SENet is introduced into the generator network to suppress the significant noise interference in the outdoor scene and improve the estimation accuracy.Finally,the residual block in the residual network is added to the scaling operation of the SENet to improve the convergence speed of the CGAN.The simulation results show that compared with that of Orthogonal Matching Pursuit(OMP) algorithm,Convolutional Neural Network(CNN) algorithm,Denoising Convolutional Neural Network(DnCNN) algorithm and CGAN algorithm,the estimation accuracy of the proposed method in outdoor noise environment is improved by about 2.2 dB on average,and the improvement of estimation accuracy is more significant under high noise intensity. |
| Key words: millimeter wave massive MIMO channel estimation conditional generative adversarial network(CGAN) squeeze-and-excitation network(SENet) |
