Research On Indoor Location Algorithm Based On Channel State Information Fingerprint

With the popularization of the Internet of Things(IoT)and mobile devices,there have been more devices access into the IoT.During this period,the Location Based Service(LBS)is increasingly important for meeting the needs of locating things in large-scale application scenarios like smart city,industrial automation,intelligent transportation and digital healthcare.As a fundamental technique,LBS is the antecedent demand in the smart city's construction and the improvement of positioning accuracy will exponentially force the corresponding service.For example,smart home needs to provide home appliances control,lighting control,burglar alarm and environmental monitoring.For now,in the field of indoor positioning,there exist widespread and mature localization systems based on Bluetooth,WiFi and infrared.However,bluetooth is characterised by slow transmission rate and low transmission range.In the actual life it's difficult to deal with requirements of multi-terminal,high-density and low power.Products using infrared technology have directivity,which means the transmitter must be aligned with the receiver with no obstruction in the middle[1].In this context,WLAN gradually emerged and highlighted its superiority.This thesis investigates the development and current situation of indoor positioning technology research at home and abroad,systematically studies the indoor positioning technology based on WiFi,and chooses to use channel state information(CSI)as the research object.CSI can be directly extracted from WiFi.It can avoid the effects of multipath and noise as much as possible and improve the positioning stability and accuracy.In view of the existing problems in the current indoor CSI fingerprinting technology,based on the existing research,two indoor positioning methods based on channel state information(CSI)fingerprints are proposed:(1)A multidimensional scaling(MDS)analysis based CSI fingerprinting method is proposed.Firstly in the offline stage,we utilize CSI amplitudes as fingerprints and store them in the database.In the online stage,we employ MDS analysis to reduce the dimension of CSI and convert the similarity into relative coordinates.After dimension reduction,we choose a deterministic algorithm K-Nearest Neighbor(KNN)to match target point position with fingerprints.By KNN algorithm,we obtain the reference point closest to the target point and take the average position as the target estimation position.Moreover,we select a small part of reference points to obtain another target position to be estimated via an optimized triangular centroid algorithm based on CSI propagation model.Finally,the average of MDS positioning result and centroid algorithm positioning result is used as the final target position.(2)A deep learning method is proposed to implement a CSI-based indoor positioning method.In the off-line phase,we use a stacked automatic encoder(SAE)network to train the CSI data,explore its characteristics,and then use the deep network weights as fingerprints to store in the database.In the online phase,we use the Bayesian probability model based on radial basis neural networks to estimate the location of the user.Both of these two methods are conducted experiments and compared with other typical algorithms.Experiments show that both algorithms are effective.Compared with other existing typical CSI fingerprint-based indoor location methods,these two methods achieve better accuracy.Finally,all the work of this thesis is summarized,and the fellow step of the work is prospected.