基于深度学习的大规模MIMO检测算法研究  

Deep learning for massive MIMO detection algorithm

作　　者：张晓羽 贺光辉[1] ZHANG Xiao-yu;HE Guang-hui(School of Electronics,Shanghai Jiaotong University,Shanghai 200240,China)

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240

出　　处：《信息技术》2020年第6期10-14,共5页Information Technology

基　　金：国家自然科学基金项目(61674103)。

摘　　要：大规模MIMO是5G的关键技术之一,提升性能、降低复杂度是大规模MIMO检测面临的巨大难题。深度学习具有良好的并行性和鲁棒性,使其能够应用于大规模MIMO检测,目前已有的深度学习MIMO检测方法复杂度很高,且不能支持高阶调制。文中提出了一种去噪稀疏连接网络(Denoising Sparsely connected detection network,DSNet),通过将网络全连接结构修改为稀疏连接简化网络模型,进一步减小了训练参数。另外,为了使得检测网络能够适用于高阶调制,文中提出一种基于高斯去噪的激活函数。仿真结果表明该网络在64×16规模MIMO系统下能获得1dB左右的增益,且复杂度和训练所需的参数量明显降低。Massive MIMO is one key technology for fifth generation wireless systems.A challenge for massive Multiple Input Multiple Output(MIMO)detection is to improve performance and reduce complexity with the increasing number of antennas for massive MIMO system.There are already some deep learning aided detection models.Unfortunately,they have high complexity and unfriendly to higher-order modulation,so a Denoising Sparsely connected detection network(DSNet)is proposed.The sparsely connection takes place of the full connection in the network structure to reduce the trainable parameters.Moreover,an active function based on gaussian denoising is proposed to make the detection network suitable for high-order modulation.The simulation results show that the proposed network reduces the complexity efficiently with about 1dB performance gain compared with existing detection network for 64×16 MIMO systems.

关 键 词：大规模MIMO 深度学习 高阶调制 检测算法 

分 类 号：TN929[电子电信—通信与信息系统]