大规模MIMO系统领域自适应定位  

Domain Adaptive Localization in Large-scale MIMO

作　　者：蒋志函 王斌 潘志文[1,2] 刘楠 JIANG Zhihan;WANG Bin;PAN Zhiwen;LIU Nan(National Mobile Communications Research Laboratory,Southeast University,Nanjing 210096,China;Purple Mountain Laboratories,Nanjing 211100,China;Science and Technology on Communication Networks Laboratory,Shijiazhuang 050081,China)

机构地区：[1]东南大学移动通信国家重点实验室,江苏南京210096 [2]网络通信与安全紫金山实验室,江苏南京211100 [3]通信网信息传输与分发技术重点实验室,河北石家庄050081

出　　处：《无线电通信技术》2022年第4期665-672,共8页Radio Communications Technology

基　　金：国家重点研发计划(2020YFB1806805);通信网信息传输与分发技术重点实验室项目。

摘　　要：环境变化引起的信道状态改变使得传统的基于信道状态信息(Channel State Information,CSI)的指纹定位方法定位精度显著下降。近年来的研究表明,领域适应是克服上述缺点的一种合适策略。领域适应将环境改变前后分别看成是两个不同分布下的源域和目标域,能够通过缩小两个领域之间的差异得到适用于目标域的模型。基于此,提出了一个基于分歧差异的深度卷积对抗网络(Disparity Discrepancy Based Deep Convolutional Adversarial Network,DDCAN)领域自适应指纹定位方法,该方法在大规模多输入多输出(Massive Multiple Input Multiple Output,M-MIMO)系统中将CSI重构得到的角度时延信道幅度矩阵(Angle Delay Channel Amplitude Matrix,ADCAM)作为指纹,在拥有源域的有标签样本和目标域的无标签样本的情况下,通过最小化源域误差和分歧差异可以有效训练出适用于目标域的定位网络。在此基础上,分别构建了分类和回归定位模型,其中针对分类定位模型,提出网格中心重定位的方法,以增加定位精度。仿真结果表明,该方法可以有效缓解环境变化对定位方法的影响,并达到较高的定位精度。The Channel state changes caused by environmental changes make the positioning accuracy of the traditional fingerprint positioning method based on channel state information(CSI)degrades significantly.Recent studies reveal that domain adaptation is a suitable strategy to overcome above drawbacks.By considering the state before and after the environment changes as the source domain and the target domain respectively,domain adaptation can narrow the differences between the two domains to construct a model suitable for the target domain.In this paper,a domain adaptation fingerprint positioning method utilizing disparity discrepancy based deep convolutional adversarial network(DDCAN)is proposed.In massive multiple input multiple output(M-MIMO)system,the angle delay channel amplitude matrix(ADCAM)reconstructed by CSI is used as fingerprint.The localization network suitable for the target domain can be effectively trained by minimizing the errors and differences in the source domain with labeled samples in source domain and unlabeled samples in target domain.Classification and regression positioning models are constructed respectively.For classification positioning model,a grid center repositioning method to increase the positioning accuracy is proposed.Simulation results show that this method can effectively alleviate the influence of environmental changes on the positioning method,and achieve high positioning accuracy.

关 键 词：指纹定位 领域适应 ADCAM DDCAN 网格中心重定位 

分 类 号：TN929.5[电子电信—通信与信息系统]