Research On Indoor High Precision Positioning Method Based On CSI

With the continuous development of Wi-Fi technology,Multiple Input Multiple Output(MIMO)and Orthogonal Frequency Division Multiplexing(OFDM)technologies are applied in IEEE 802.11 n and later versions,which makes Wi-Fi channel characteristics more abundant.The channel characteristics are accurately described in the form of Channel State Information(CSI).The channel fading factor,multipath effect,scattering,signal interference and other information can be used in indoor Wi-Fi CSI positioning.Although there have been a lot of researches,there are still some problems,such as low efficiency of fingerprint data acquisition,low efficiency of fingerprint database matching and low positioning accuracy.Therefore,an indoor high-precision positioning method is proposed in this thesis based on CSI.The main works are as follows.(1)For the low efficiency problem of indoor CSI fingerprint acquisition,a data preprocessing method based on noise elimination and improved PCA dimension reduction is proposed.Firstly,the idea of edge stripping clustering is adopted to remove the interference points in CSI information,so as to reduce the noise on CSI data.Secondly,a CSI data preprocessing method based on improved PCA,YJ-MICPCA,is proposed.The Yeo-Johnson transform is employed to convert non-Gaussian data into Gaussian distribution data.Then,the covariance matrix of PCA is replaced by the Maximum Information Coefficient(MIC)to evaluate the nonlinear relationship between CSI features.Finally,the data dimension reduction is completed by PCA.This method can be applied to CSI data with non-Gaussian distribution in reality,suitable for nonlinear CSI features,and improve the clustering effect of CSI fingerprints.(2)Regarding the problem that CSI indoor positioning accuracy is affected by many factors,two CSI indoor positioning schemes are proposed.Firstly,a fingerprint database construction method based on smooth constraint low rank theory is presented,assisting the CSI data channel characteristics between each fingerprint more continuous,reducing the number of fingerprint collection,and facilitates the implementation of the matching algorithm.On this basis,a high-precision positioning scheme based on Radial Basis Function(RBF)-K-Nearest Neighbor(KNN)is offered,which has good applicability.Secondly,according to the problems caused by unbalanced data,a new Bootstrap random forest algorithm is designed.Multiple training data sets are generated by resampling,so as to construct multiple models and conduct integrated prediction to improve the overall positioning accuracy and positioning performance.Experiment results show that the preprocessing method improves the positioning accuracy of RBF-KNN,and the new random forest method has lower positioning error.