Indoor Wireless Fingerprinting Localization Based On Machine Learning

In recent years,fingerprinting localization has become a research hotspot of indoor localization.It has the advantages of low cost,simple deployment,high accuracy,and strong anti-interference ability.With the rapid development and wide application of Orthogonal Frequency Division Multiplexing(OFDM)technology in the field of communication,the feature of fingerprinting localization gradually changed from Received Signal Strength(RSS)to Channel State Information(CSI).CSI is a kind of more fine-grained data compared with RSS,which contains channel state information on each subcarrier.It can provide richer information as a fingerprint feature.At the same time,more and more mobile communication base-stations are widely deployed with the rapid popularization of 4G and 5G applications.The cost of indoor localization system can be reduced by using these ready-made infrastructures.This thesis studies the Indoor Wireless Fingerprinting Localization based on machine learning algorithms with the signal features from LTE-CSI.The main work of this thesis is as follows:Firstly,the physical layer structure of the downlink of the LTE system and the method of channel estimation using Cell-specific Reference Signal(CRS)are studied.A LTE downlink channel estimation and CSI extraction system is built using software radio equipment on the basis of theoretical research.The characteristics of LTE-CSI data are analyzed from the position difference of the receiver and the space difference of the transmitting antenna port,which provides a basis for indoor fingerprinting localization based on LTE-CSI.Secondly,a LTE-CSI data smoothing method based on Density-Based Spatial Clustering of Applications with Noise(DBSCAN)is proposed.This method can effectively identify abnormal data in the samples.The mean value of the samples will be used to replace the abnormal data in order to provide stable and effective fingerprint data for the subsequent localization algorithm.Thirdly,an indoor fingerprinting localization method based on LTE-CSI descriptor and XGBoost(e Xtreme Gradient Boosting)algorithm is proposed.This method can increase the positioning speed and meet the user's requirements for real-time positioning.Experiment shows that this method has lower time complexity than the KNN based algorithm.Finally,a deep neural network-based LTE-CSI fingerprinting localization method is studied,which makes full use of the channel state information of each subcarrier contained in the CSI data.A data dimensionality reduction method based on Stacked Automatic Encoder(SAE)is studied in order to solve the problem of high dimensionality of LTE-CSI data.And a new feature extraction method is proposed,which uses the Pearson correlation coefficient between the CSI data of different channel as the fingerprint feature.Experiment shows that this method can greatly reduce the dimension of the fingerprint feature while ensuring the positioning accuracy...