Research On CSI Amplitude And Phase Based Indoor Positioning Technology

With the advancement of technology and the widespread availability of Wi-Fi infrastructure,mobile devices and wireless networks have penetrated into people’s work and life.Location Based Service(LBS)has become an indispensable part of our life,where location technology is the core of all location services.However,traditional satellite positioning and navigation technologies cannot penetrate the steel and concrete of buildings,making it very difficult to apply in indoor environments.Recently Wi-Fi-based indoor positioning technologies have become the focus of extensive academic attention.However,most of the current angle-based localization systems suffer from large errors in the estimation of angle of arrival,and fingerprint-based localization systems suffer from single fingerprint information and complex models,which seriously affect their pervasiveness.In this thesis,the above problems are investigated separately.In terms of active indoor positioning,in order to solve the current problem of large errors in measuring the angle of arrival using Wi-Fi,this thesis proposes an indoor positioning method based on the Angle of Arrival(Ao A)measurement of Pmusic spectral peak map.The method firstly preprocesses the collected Channel State Information(CSI),and then uses Multiple Signal Classification(MUSIC)algorithm to obtain the Pmusic spectral peak map.Then,we select the Ao A data on the first 3 peaks of the Pmusic spectral peak map and select the Ao A data in the range of the direct path,and use the kmeans clustering method to find the sub-clusters with the largest contour coefficient index.These clusters are weighted and the angle of arrival of the direct path between the transmitter and the receiver is estimated,and finally the estimated angle of arrival is combined with the coordinates of the receiver to estimate the position of the target.The experimental results show that the arrival angle estimation error of the method is smaller than that of the UAT method,which can further improve the indoor positioning accuracy,and the median error of 0.48 m can be obtained in indoor environments,and the method outperforms other indoor positioning methods in 2 scenarios.In terms of passive indoor positioning,in order to solve the problems of single location fingerprint information and complex models of existing CSI fingerprint localization schemes,this thesis proposes an indoor localization method based on Depthwise Separable Convolutional Neural Network(DSCNN).The method is divided into two stages: offline training and online localization.In the offline training phase,the experimental area is firstly divided and the sampling points are set up in the room,and then the CSI data of each sampling point are collected to construct the amplitude difference image and phase difference image,and the fused image of the amplitude difference image and phase difference image is constructed as the CSI feature image using the Laplace pyramid algorithm.Then,the CSI feature images of all training locations are used to train the network model.In the online localization stage,the CSI data of the target location is collected,the CSI feature image of the target location is constructed,and it is fed into the trained model to obtain the predicted location of the target.The experimental results show that the median error of 1.16 m can be obtained in indoor environment,and the method outperforms other indoor localization methods in 2 scenes.