基于生成对抗网络的MIMO信道估计方法  

GAN-based channel estimation for massive MIMO system

作　　者：华郁秀 李荣鹏[1] 赵志峰[1,2] 吴建军 张宏纲[1] HUA Yuxiu;LI Rongpeng;ZHAO Zhifeng;WU Jianjun;ZHANG Honggang(College of Information Science and Electronic Engineering,Zhejiang University,Hangzhou 310027,China;Zhejiang Lab,Hangzhou 311121,China;Huawei Technologies Co.,Ltd.,Shanghai 200120,China)

机构地区：[1]浙江大学信息与电子工程学院,浙江杭州310027 [2]之江实验室科研发展部,浙江杭州311121 [3]华为技术有限公司未来网络架构实验室,上海200120

出　　处：《电信科学》2021年第6期14-22,共9页Telecommunications Science

基　　金：国家重点研发计划项目(No.2020YFB1804804SCI);国家自然科学基金资助项目(No.61731002,No.62071425);浙江省重点研发计划项目(No.2019C01002,No.2019C03131);华为合作项目:之江实验室资助项目(No.2019LC0AB01);浙江省自然科学基金资助项目(No.LY20F010016)。

摘　　要：作为5G的一项关键技术,大规模多输入多输出(MIMO)系统通过在基站配备大量天线可以显著提高频谱效率和能效。然而,在大规模MIMO系统中,精确的信道估计面临严峻挑战。为了在导频序列长度小于发射天线数量以及信道噪声强烈的情况下进行精确的估计信道,提出了一种基于生成对抗网络的大规模MIMO信道估计方法N2N-GAN。N2N-GAN首先对接收端的导频信号进行去噪,然后使用条件生成对抗网络根据去噪后的导频信号估计信道矩阵。仿真实验证明,与传统的信道估计算法和基于深度学习的算法相比,N2N-GAN对环境噪声具有更高的鲁棒性,而且可以适应更少导频符号和更多天线数量的场景。As a key technology of 5G,massive MIMO system can significantly improve spectrum efficiency and energy efficiency by equipping a large number of antennas in base stations.However,in massive MIMO system,accurate channel estimation faces severe challenges.In order to estimate the channel accurately when the pilot sequence length is smaller than the number of transmitting antennas and the channel noise is strong,the estimation method N2N-GAN was proposed.N2N-GAN firstly denoised the pilot channel at the receiving end,and then used the conditional generative adversarial network to estimate the channel matrix according to the denoised pilot signal.Simulation experiments show that N2N-GAN achieves better robustness against noise compared with traditional channel estimation algorithms and deep learning-based methods.Meanwhile,it can adapt to scenarios with fewer pilot symbols and more antennas.

关 键 词：信道估计 大规模MIMO 信号降噪 生成对抗网络 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]