摘要
针对多径效应影响指纹定位算法中定位精度的问题,提出了一种基于聚类的主成分分析(principal component analysis,PCA)和广义回归神经网络(generalized regression neural network,GRNN)的信道状态信息(channel state information,CSI)指纹定位算法。离线训练阶段,利用中值滤波对CSI幅值进行去噪,并利用线性变换校准CSI相位,将处理后的幅值和相位作为原始联合指纹,利用改进的K-means算法将各个参考点的联合指纹集划分成多个子数据集来描述位置的多径特性,通过高维数据的PCA算法提取子数据集的特征以减少冗余信息、提高不同位置指纹的区分性,最后利用特征指纹训练GRNN模型。在线阶段,利用训练好的GRNN模型对在线测量的CSI数据进行目标对象的位置预测。实验结果表明,该算法可有效反映出位置的多径信息,且与CSI-MIMO,DeepFi和CSI-PCA相比,在定位精度方面有明显的提升。
Aiming at the problem of location accuracy in the multi-path effect of fingerprint location,this paper proposes a CSI fingerprint localization algorithm based on clustering principal component analysis(PCA)and generalized regression neural network(GRNN).In the offline phase,it uses median filtering method to reduce the noise of the amplitude,and calibrates the phase by linear transformation method.Then it takes the processed amplitude and phase as the original joint fingerprint.To describe the multi-path characteristics of the position,it uses the improved K-means algorithm to divide the joint fingerprint set of each reference point into multiple sub-datasets.Then,it extracts the feature of the sub-dataset by the PCA algorithm of high-dimensional data to reduce redundant information and improve fingerprint discrimination at different locations.Finally,the GRNN model is trained by using the feature fingerprint.In the online phase,it uses the trained GRNN model to predict the position of the target object based on the online measured CSI data.Experimental results show that this algorithm can not only reflect the multi-path information of position effectively,but also improve the positioning accuracy significantly compared with CSI-MIMO,DeepFi and CSI-PCA algorithm.
作者
李新春
黄朝晖
LI Xinchun;HUANG Zhaohui(College of Electrics and Information Engineering,Liaoning Technical University,Huludao 125105,P.R.China;College of Graduate Studies,Liaoning Technical University,Huludao 125105,P.R.China)
出处
《重庆邮电大学学报（自然科学版）》
CSCD
北大核心
2021年第3期449-457,共9页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
基金
国家自然科学基金(61372058)。
关键词
室内定位
信道状态信息
聚类
主成分分析
广义回归神经网络
indoor positioning
channel state information
clustering
principal component analysis
generalized regression neural network
