基于无监督学习的卫星NLOS信号检测方法  

Unsupervised learning based satellite NLOS signal detection method

作　　者：夏炎 潘树国[1] 赵鹏飞 赵庆[2] 叶飞 Xia Yan;Pan Shuguo;Zhao Pengfei;Zhao Qing;Ye Fei(School of Instrument Science and Engineering,Southeast University,Nanjing 210096,China;School of Transportation,Southeast University,Nanjing 210096,China)

机构地区：[1]东南大学仪器科学与工程学院,南京210096 [2]东南大学交通学院,南京210096

出　　处：《东南大学学报（自然科学版）》2019年第3期565-572,共8页Journal of Southeast University：Natural Science Edition

基　　金：“十三五”国家重点研发计划资助项目(2016YFB0502101);国家自然科学基金资助项目(41574026,41774027)

摘　　要：针对复杂环境下NLOS信号接收造成的GNSS定位精度恶化问题,提出了一种基于无监督学习的卫星NLOS信号检测方法.综合考虑了信号载噪比、伪距残差和卫星高度角对于GNSS接收信号的影响,采用k-means++聚类算法将观测数据划分为LOS、多径和NLOS三类,并对NLOS信号进行分离.使用GPS/BDS双系统伪距单点定位对信号分类效果进行了验证.结果表明,采用该方法剔除NLOS信号后定位精度得到了显著的提升.静态实验中,对1h的数据样本进行聚类,事后定位精度提高了约30%,实时定位精度提高约12%.动态实验中,城市峡谷路段东、北、天3个方向的定位精度分别提高了27.98%、8.06%和3.66%.相较于有监督学习的分类方法,该方法简单有效、易于实现,且无需使用先验信息,能显著降低运算负荷和GNSS设备成本.与传统的阈值法以及RAIM算法相比较,该方法在改善定位的精度方面具有一定的优势.Aiming at the problem of global navigation satellite system(GNSS) positioning accuracy deterioration caused by non-line-of-sight (NLOS) signal reception in complex environments, a detection method for satellite NLOS signal based on unsupervised learning was proposed. The effects of carrier-to-noise ratio, pseudorange residual and satellite elevation angle on received GNSS signals were overall considered. The observation data were divided into three categories of line of sight (LOS), multipath and NLOS using k-means++ clustering algorithm, and then NLOS signal was separated from the dataset. GPS/BDS dual-system single point positioning was used to verify the signal classification effect. The results show that the positioning accuracy is improved after removing the NLOS signal by the proposed method. In the static experiment, 1 h data samples are clustered, and the post-positioning accuracy is improved by about 30%, while the real-time positioning accuracy is improved by about 12%. In the dynamic experiment, the positioning accuracies in the east, north, and up directions of the urban canyon section are improved by 27.98%, 8.06%, and 3.66%, respectively. Compared with the supervised learning classification method, the proposed method is simple and effective, easy to implement, and does not require the use of prior information, thus reducing the computational load and GNSS equipment cost. Compared with the traditional threshold method and receiver autonomous integrity monitoring(RAIM) algorithm, the method can improve the positioning accuracy.

关 键 词：无监督学习 NLOS k-means++聚类算法 双系统 伪距单点定位 

分 类 号：P228.1[天文地球—大地测量学与测量工程]