基于元学习的短波MIMO信号检测  

Meta learning based HF MIMO signal detection

作　　者：杨小石 卜方玲[1] 王智勇 陈宁 YANG Xiao-shi;PU Fang-ling;WANG Zhi-yong;CHEN Ning(School of Electronic Information,Wuhan University,Wuhan 430072,China)

机构地区：[1]武汉大学电子信息学院,湖北武汉430072

出　　处：《计算机工程与设计》2022年第1期43-49,共7页Computer Engineering and Design

基　　金：国家重点研发计划项目课题基金项目(2018YFB2100503)。

摘　　要：针对短波场景下已有的MIMO检测算法性能不佳或复杂度太高的问题,提出一种基于元学习的MIMO检测方法MetaMMNet。将不同的信道矩阵下的MIMO检测看作为独立的任务,通过对大量不同任务的学习使得模型获得对于不同信道矩阵的泛化能力,能够快速适应变化的信道矩阵。在4QAM、16QAM和64QAM调制方式下进行仿真实验,实验结果表明,在短波空间相关信道下MetaMMNet的误符号率性能要优于MMNet约1 dB,优于OAMPNet约2 dB。且MetaMMNet与在线学习的MMNet相比能够更快地适应快速变化的信道,与OAMPNet相比具有更低的计算复杂度。Aiming at the problems of poor performance and high complexity of existing MIMO detection algorithms in short-wave scenarios,a meta-learning based MIMO detection method MetaMMNet was proposed.MIMO detection under different channel matrices was regarded as an independent task,and the generalization ability of model for different channel matrices was obtained by learning a large number of different tasks.Simulation experiments were carried out under 4QAM,16QAM and 64QAM modulation.Experimental results show that MetaMMNet is better than MMNet(1 dB)and OAMPNet(2 dB)on short wave spatially-correlated channels.MetaMMNet can adapt to the variant channel more quickly than online learning MMNet,and has lower computational complexity compared with OAMPNet.

关 键 词：短波通信 MIMO检测 迭代算法 神经网络 元学习 

分 类 号：TN911.23[电子电信—通信与信息系统]