基于CNN的毫米波无蜂窝大规模MIMO信道估计

针对小区间干扰导致蜂窝边缘无法满足不断增长的数据速率需求问题，毫米波无蜂窝大规模多输入多输出(Multiple-Input Multiple-Output, MIMO)系统被认为是一种很有前途的解决方案。然而，毫米波的高频率、大带宽以及接入点配置的大量天线给信道估计带来了较大挑战。将毫米波大规模MIMO信道矩阵视为二维图像，结合图像去噪方法提出一种基于改进去噪卷积神经网络(Improved-Denoising Convolutional Neural Network, I-DnCNN)的信道估计算法。通过具有注意力机制的压缩与激励(Squeeze-and-Excitation, SE)模块，自适应调整提取的全局特征以增强对信道噪声特征的学习，根据接收信号估计出噪声等级图且增添为输入，提升对噪声的鲁棒性。最后，采用残差学习的方式获得估计信道矩阵。利用理论信道模型和基于波束追踪的信道数据集进行的仿真实验结果表明，与去噪卷积神经网络(Denoising Convolutional Neural Network, DnCNN)算法相比，所提算法在两个数据集下的信道估计精度可分别平均提升2.27...

CNN Based Channel Estimation in Millimeter Wave Cell-free Massive MIMO Systems

The edge of the cell cannot meet the increasing data rate requirements because of inter-cell interference.Millimeter wave cell-free massive multiple-input multiple-output(MIMO) system is regarded as a promising solution.However,the high frequency and large bandwidth of millimeter wave and a large number of antennas at the access point bring great challenges to channel estimation.The millimeter wave massive MIMO channel matrix is regarded as a two-dimensional image,and combined with image denoising method,a channel estimation algorithm based on improved denoising convolutional neural network(I-DNCNN) is proposed.The extracted global features are adaptively adjusted to enhance the learning of channel noise features through the squeeze-and-excitation(SE) module with attention mechanism.The noise level graph is estimated according to the received signals and added as input to improve the robustness to noise.Finally,residual learning is used to obtain the estimated channel matrix.Simulation results by using both the theoretical channel model and the ray-tracing based channel dataset show that,compared with that of the denoising convolutional neural network(DnCNN) algorithm,the channel estimation accuracy of the proposed algorithm can be improved by 2.27 dB and 2.60 dB on average in the two datasets,respectively.