类别不平衡学习识别UWB定位非视距信号  

Class Imbalance Learning for Identifying NLOS in UWB Positioning

作　　者：缪志敏[1] 赵陆文[2] 田世伟[2] 姜劲松[1] 

机构地区：[1]解放军理工大学指挥信息系统学院,江苏南京210007 [2]解放军理工大学通信工程学院,江苏南京210007

出　　处：《信号处理》2016年第1期8-13,共6页Journal of Signal Processing

基　　金：国家863计划(2012AA01A510);中国博士后科学基金(2013M542485)

摘　　要：非视距(Non Line of Sight,NLOS)传播是影响超宽带(Ultra-wide Bandwidth,UWB)定位精度的一个重要因素。针对UWB定位中视距(Line of Sight,LOS)信号数量大于NLOS信号数量所呈现的类别不平衡特点,提出了一种基于类别不平衡学习的NLOS信号识别方法。该方法通过给NLOS信号和LOS信号赋予不同的误分代价来训练一个带野值的支持向量数据描述(Support Vector Data Description,SVDD)学习器,实现对数量少但重要的NLOS信号的识别。仿真结果表明,当LOS信号数量远大于NLOS信号数量时,该方法对NLOS信号的识别性能优于支持向量机(Support Vector Machine,SVM)。Non Line of Sight propagation is an important reason effecting of the positioning accuracy of Ultra-wide Bandwidth system. It＇s ditticult to model and distinguish NLOS signal, as the characteristics of NLOS signal are closely related to the en- vironment. Considering the characteristic that LOS and NLOS signals are very imbalance in UWB positioning system, a NLOS signal recognition method based on the class imbalance learning is proposed. In order to recognize NLOS signals which are rare but important, a Support Vector Data Description learning machine with negative is trained with the NLOS signal and the LOS signal with different misclassification cost. The simulation results show that the performance of this method is better than to Support Vector Machine, while the number of LOS signals and NLOS signals is extreme imbalance.

关 键 词：超宽带定位 非视距 支持向量数据描述 类别不平衡学习 

分 类 号：TP912[自动化与计算机技术]