基于卷积神经网络的大规模MIMO-D2D的导频复用  

PILOT MULTIPLEXING OF MASSIVE MIMO-D2D BASED ON CONVOLUTIONAL NEURAL NETWORK

作　　者：程智超 赵峰[1,2] Cheng Zhichao;Zhao Feng(School of Information and Communication,Guilin University of Electronic Technology,Guilin 541004,Guangxi,China;School of Electronics and Communication Engineering,Yulin Normal University,Yulin 537000,Guangxi,China)

机构地区：[1]桂林电子科技大学信息与通信学院,广西桂林541004 [2]玉林师范学院电子与通信工程学院,广西玉林537000

出　　处：《计算机应用与软件》2022年第1期89-93,175,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(61871466);广西自然科学基金创新研究团队项目(2016GXNSFGA380002);桂林电子科技大学研究生创新项目(2018YJCX40)。

摘　　要：大规模MIMO-D2D异构网络中,可以通过在蜂窝用户和D2D用户之间使用相同的频谱资源来提高频谱效率,但是在信道估计中,共享相同导频序列的用户之间会产生严重干扰。为了解决该问题,利用卷积神经网络,通过学习最优的导频分配来推断导频分配结果以减轻导频污染的影响。将用户在小区中的位置和相应的导频分配作为输入和输出标签,通过穷举法得到用户位置的最佳导频分配作为训练数据。经卷积神经网络导频分配系统(CNN-PAS)分析训练数据,利用所产生的推断函数提供近似最优的导频分配结果。仿真结果表明,该方案实现了近98.78%的理论上限性能。In massive MIMO-D2D heterogeneous networks,the spectral efficiency can be significantly improved by using the same spectrum resources between cellular users and D2D users.However,severe interference can occur between users who share the same pilot sequence in channel estimation.To solve the problem,we used a convolutional neural network to infer the pilot allocation results by learning the optimal pilot allocation to mitigate the effects of pilot pollution.The user’s position in the cell and the corresponding pilot allocation were used as input and output labels,and the optimal pilot allocation of the user position was obtained as training data by exhaustive method.The proposed convolutional neural network pilot distribution system(CNN-PAS)analyzed the training data and used the inferred function to provide an approximately optimal pilot allocation result.Simulation results show that this scheme achieves nearly 98.78%of the theoretical upper limit performance.

关 键 词：D2D 大规模多输入多输出(MIMO) 导频污染 导频复用 卷积神经网络 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]