Research On Indoor Location Algorithm Based On Geomagnetic Field And CSI Fingerprint

With the popularity of wireless networks and smart devices,indoor positioning has been rapidly developed.Location-based information services have attracted more and more attention and accurate location information has played an important role in practical applications.In indoor positioning,the fingerprinting method based on the geomagnetic field has gradually become a research hotspot because it does not require external facilities and strong anti-interference.However,in the indoor environment,using only the geomagnetic field for positioning often has problems such as high computational complexity and slow convergence speed.Even in some large indoor environments.In this paper,we analyzed the problems of magnetic field fingerprinting and used Channel State Information(CSI)to improve the flaws of magnetic field in indoor positioning.Specifically,the main research contents and contributions of this paper include the following aspects:(1)We studied the technical principles and typical matching algorithms of magnetic field fingerprinting,and analyzed the limitations of magnetic field fingerprinting algorithm in positioning efficiency and positioning accuracy.Aiming at the problem of matching efficiency at the online stage,we proposed an improved line of sight(LOS)/not line of sight(NLOS)recognition algorithm for narrowing the fingerprint matching area.Compared to the original recognition,the improved recognition algorithm using CSI information in multiple APs increases the difference between NLOS and LOS,making the recognition more accurate.Experimental results show that our proposed improved LOS/NLOS algorithm has higher recognition rate and stability than the original recognition algorithm.(2)We combined CSI and magnetic field information to construct a mixed fingerprint database to make full use of sampled data,and proposed a multi-dimensional scaling analysis based on Multi-Dimensional Scaling(MDS)-(k-Nearest Neighbor,KNN)fingerprint matching method.This method converts high-dimensional fingerprints into low-dimensional fingerprints,makes deeper use of the feature information in the fingerprints,reduces the positioning error and reduces the matching computational complexity of the positioning system.The experimental results show that MDS-KNN matching algorithm has a smaller positioning error than traditional methods.(3)We analyzed the offline stage of traditional fingerprint positioning,that is,the fingerprint database construction phase.For most fingerprint construction methods Not fully utilized the fingerprint features.We combined deep learning and used convolutional neural network(CNN)to build a fingerprint library.After collecting the original fingerprint data,we processed the data through a multi-layer CNN network to better display the characteristics of the data,and also avoid the measurement errors and interference that may exist when the original data is used for fingerprinting.The experimental results show that the proposed fingerprint database construction algorithm based on CNN network has higher positioning accuracy than the fingerprint library construction method with no learning and shallow learning.