| 摘要: |
| 针对智能反射表面辅助多用户通信系统中传统信道估计方法性能下降的问题，将信道估计问题转化为信道去噪问题，利用深度残差学习方法学习残差噪声，从含噪导频信号中恢复信道系数。同时为提升信道估计精度，设计信道估计网络进一步提升去噪性能。网络主体包含两个模块：拼接信息保留模块将每一层卷积输出相融合，防止信道特征丢失，有效提取信道噪声的主体特征；扩张卷积稀疏模块通过扩大感受野范围获得信道的重要结构和细节特征，恢复信道噪声的边缘细节特征。仿真结果表明，归一化均方误差约等于0.45时，所提方法在不明显增加复杂度情况下，相比于线性最小均方误差算法获得3.7 dB的信噪比增益，更为接近最小均方误差信道估计器的性能，表现出了更好的性能和可用性。 |
| 关键词: MU-MISO系统 智能反射面 信道估计 深度残差学习 扩张卷积 |
| DOI：10.20079/j.issn.1001-893x.230329003 |
|
| 基金项目:重庆市自然科学基金面上项目(cstc2021jcyj-msxmX0454) |
|
| Channel Estimation Based on Deep Residual Learning in IRS-assisted MU-MISO Systems |
| CHEN Fatang,ZHU Pengyun,YANG Tao,SUN Chen |
| (1.School of Communication and Information Engineering,Chongqing University of Posts andTelecommunications,Chongqing 400065,China;2.Unit 72506 of PLA,Zhumadian 463200,China) |
| Abstract: |
| To solve the problem of performance declining of the traditional channel estimation method in intelligent reflecting surface-assisted multi-user communication system,the problem of channel estimation is transformed into the problem of channel denoising,and the deep residual learning method is used to learn the residual noise and recover the channel coefficient from the pilot signal with noise.At the same time,in order to improve the accuracy of channel estimation,the channel estimation network is designed to further improve the denoising performance.The main structure of the network consists of two modules.The concatenation information retention module fuses the convolutional output of each layer to prevent the loss of channel features and effectively extract the main features of channel noise.The extended convolutional sparse module can obtain the important structure and detail features of channel by enlarging the range of receptive field and recover the edge detail features of channel noise.The simulation results show that when the normalized mean square error is about 0.45,the proposed method is closer to the performance of the channel estimator than the linear least mean square error algorithm,which can obtain a signal-to-noise ratio gain of 3.7 dB without increasing the complicity obviously,and shows better performance and availability. |
| Key words: MU-MISO system intelligent reflective surface channel estimation deep residual learning extended convolution |
