基于SAMP-Net的MIMO信号检测算法  

Signal detection based on SAMP-Net method for MIMO systems

作　　者：胡钟秀 王鸿 宋荣方[1] HU Zhongxiu;WANG Hong;SONG Rongfang(College of Teleconnnunications&Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003

出　　处：《南京邮电大学学报（自然科学版）》2020年第6期36-41,共6页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：江苏省自然科学基金(BK20181392)资助项目。

摘　　要：文中提出了一种用于多输入多输出(MIMO)检测的模型驱动型深度学习网络。首先,对MIMO系统信号检测问题进行了建模,并且介绍了几种传统检测算法;然后,将简化近似消息传递(SAMP)迭代检测算法与深度学习结合,通过展开原算法得到网络结构,提出了新型模型驱动的深度学习网络SAMP-Net,通过学习得到最优可训练的参数,提高检测性能;最后,将其与最小均方误差(MMSE)算法、Richardson算法、SAMP算法进行检测性能比较。仿真结果表明,SAMP-Net算法可以在瑞利MIMO信道中以较低计算复杂度逼近MMSE算法的检测性能。A model-driven deep learning network for multiple-input multiple-output(MIMO)detection is proposed.Firstly,the signal detection problem of MIMO system is modeled and several traditional iterative detection algorithms are introduced.Secondly,a new model-driven deep learning network,called the SAMP-Net,is proposed by combining the simplified approximate messaging(SAMP)iterative detection algorithm with deep learning.Finally,the detection performance of SAMP-Net algorithm is compared with that of the minimum mean square error(MMSE)algorithm,Richardson algorithm and SAMP algorithm.Simulation results show that the SAMP-Net algorithm can approach the detection performance of the MMSE algorithm with low computational complexity in Rayleigh MIMO channels.

关 键 词：深度学习 迭代算法 模型驱动 MIMO检测 

分 类 号：TN929.5[电子电信—通信与信息系统]