In this letter, fully convolutional denoising approximate message passing (FCDAMP) algorithm is proposed by combining fully convolutional denoising networks with learned approximate message passing networks in millimeter-wave massive MIMO system. In particular, an asymmetric neural network architecture is considered that can learn channel structure and extract noise characteristics. Simulation and analysis show that the proposed FCDAMP algorithm satisfies the lower estimation error and the higher achievable sum rate especially in the low SNR. Moreover, the performance can be further improved by increasing the antenna array in massive MIMO system.

中文翻译：

---

毫米波大规模 MIMO 系统中基于深度学习的波束空间信道估计

在这封信中，通过将全卷积去噪网络与毫米波大规模 MIMO 系统中的学习近似消息传递网络相结合，提出了全卷积去噪近似消息传递（FCDAMP）算法。特别地，考虑了可以学习通道结构并提取噪声特征的非对称神经网络架构。仿真和分析表明，所提出的FCDAMP算法满足较低的估计误差和较高的可实现和率，尤其是在低信噪比的情况下。此外，在大规模 MIMO 系统中，通过增加天线阵列可以进一步提高性能。