基于多通道图像的ECA-CNN WiFi FTM室内定位算法  

An ECA-CNN algorithm based on multi-channel image for WiFi FTM indoor positioning

作　　者：刘林 廖子阳[2] Liu Lin;Liao Ziyang(State Key Laboratory of Intelligent Construction and Maintenance for Geotechnical and Tunnel Engineering under Extreme Environments(FSDI),Xi′an 710043,China;Provincial Key Lab of Information Coding and Transmission,Southwest Jiaotong University,Chengdu 611756,China)

机构地区：[1]极端环境岩土和隧道工程智能建养全国重点实验室(中铁一院),西安710043 [2]西南交通大学信息编码与传输四川省重点实验室,成都611756

出　　处：《仪器仪表学报》2024年第10期323-332,共10页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金高铁联合基金(U2268201)项目资助。

摘　　要：IEEE 802.11-2016定义了精细时间测量(FTM)协议,利用信号往返时间(RTT)实现WiFi室内定位,以期达到米级定位精度。但在非视距或多径环境下,RTT测距精度下降,严重影响定位性能。因此,为了提高RTT定位精度,提出了一种将多个无线接入点(AP)测得的WiFi RTT测距序列转换为多通道图像的方法,基于多通道图像采用有效通道注意力机制卷积神经网络(ECA-CNN)学习测距数据与目标位置之间的关系,实现位置估计。实验结果表明,提出的定位模型与常规深度神经网络(DNN)定位模型、基于单通道图像的卷积神经网络(SCI-CNN)定位模型和基于单通道图像的有效通道注意力机制卷积神经网络(SCI-ECA-CNN)定位模型相比,模型的平均定位误差约为1 m,分别比上述模型降低了31.03%、16.78%和10.68%。IEEE 802.11-2016 defines the fine time measurement protocol,which uses signal round trip time(RTT)to achieve indoor WiFi positioning accuracy at the meter level.However,in non line of sight or multipath environments,the accuracy of RTT ranging decreases,which seriously affects the positioning performance.To improve the accuracy of RTT positioning,this article proposes a method to convert the WiFi RTT ranging sequences measured by multiple access points into the multi-channel image,and uses an efficient channel attention-convolutional neural network to learn the relationship between the ranging data and the target position based on the multi-channel image.The experiments show that the positioning error of the proposed model is about 1 m,and 31.03%,16.78%,and 10.68%less than the conventional deep neural networks positioning,the single-channel-image-based CNN positioning,and the single-channel-image-based ECA-CNN positioning,respectively.IEEE 802.11-2016 defines the Fine Time Measurement(FTM)protocol,which uses signal Round Trip Time(RTT)to achieve indoor WiFi positioning accuracy at the meter level.However,in non line of sight or multipath environments,the accuracy of RTT ranging decreases,which seriously affects the positioning performance.In order to improve the accuracy of RTT positioning,this paper proposes a method to convert the WiFi RTT ranging sequences measured by multiple access points(APs)into multi-channel image,and uses ECA-CNN(Efficient Channel Attention-Convolutional Neural Network)to learn the relationship between the ranging data and the target position based on the multi-channel image.The experiments show that the positioning error of the proposed model is about 1 m,and 31.03%,16.78%,and 10.68%less than the conventional DNN(Deep Neural Networks)positioning,the single-channel-image based CNN positioning,and the single-channel-image based ECA-CNN positioning,respectively.

关 键 词：室内定位 注意力机制 卷积神经网络 精细时间测量 

分 类 号：TH89[机械工程—仪器科学与技术]