基于条件变分自编码器的射线样本生成算法  

Ray Sample Generation Algorithm Based on Conditional Variational Auto-encoder

作　　者：朱军[1] 杨军 李凯 于文欣 ZHU Jun;YANG Jun;LI Kai;YU Wenxin(School of Electronic and Information Engineering,Anhui University,Hefei Anhui 230601,China;ShanghaiTech University,Shanghai 201210,China;Huawei Shanghai Research Institute,Shanghai 201206,China)

机构地区：[1]安徽大学电子信息工程学院,安徽合肥230601 [2]上海科技大学,上海201210 [3]华为上海研究所,上海201206

出　　处：《通信技术》2022年第4期409-414,共6页Communications Technology

基　　金：国家自然科学基金项目(62071002)。

摘　　要：射线追踪数据样本的缺失是造成大规模多输入多输出(Massive Multiple-Input MultipleOutput,Massive MIMO)信道特征预测出现较多预测误差较高的用户的主要原因。为了降低高误差用户数及预测误差,提出了一种基于条件变分自编码器(Conditional Variational AutoEncoder,CVAE)的射线样本生成算法来增添缺失区间的射线样本。仿真结果表明,基于所提出的算法在原有射线样本集中扩充新样本后,可将高预测误差用户数降低到原来的46.4%;完善训练集后的神经网络在降低得到信道幅值的时间开销的同时,将信道幅值预测精度提升了6.2%。The lack of ray-tracing-data samples can cause more high-prediction-error users to appear in the Massive MIMO(Multiple-Input Multiple-Output) channel feature prediction. In order to reduce the number of high-prediction-error users and prediction errors, a ray sample generation algorithm based on CVAE(Conditional Variational Auto-Encoder) is proposed. This algorithm can add ray samples in the missing intervals. The simulation results indicate that the number of high-prediction-error users can be reduced to 46.4% by expanding new samples in the original ray sample set based on the proposed method. Moreover,the extended training set improves the channel amplitude prediction accuracy by 6.2% while obtaining a reduction in the time overhead of predicting the channel amplitude.

关 键 词：大规模多输入多输出 三维信道模型 条件变分自编码器 射线追踪 

分 类 号：TN911[电子电信—通信与信息系统]