Research On Indoor Fingerprint Localization Algorithm Based On Channel State Information

With the development of wireless devices,WiFi localization technology has become a hot trend in the research field of indoor localization because of its high universality.However,due to the multipath effect and problem of a single wireless access point,signal metric and positioning methods have become the key factors affecting positioning performance.Considering the granularity and stability of the former,and the cost-effectiveness and robustness of the latter,this paper studies the fingerprint localization algorithm based on Channel State Information(CSI),the main contents are as follows:(1)In order to improve the performance of fingerprint feature extraction and enhance the degree of map fitting,this paper proposes a fingerprint localization algorithm based on convolutional autoencoder(CAE)and deep support vector machine.The algorithm adjusts the CSI amplitude to the image form as a fingerprint.In the offline phase,CAE is trained using fingerprint data at first.Then the weight of the coding part of CAE is saved,which is combined with the deep support vector machine into a deep positioning algorithm model(CAE-DSVM)by using the fully connected layer,and then the complex mapping relationship between the original CSI image data and the location is learned by the model;in the positioning phase,the fingerprint data of test point is taken as the input of CAE-DSVM,and the location information of the point can be directly obtained.The results of simulation experiments show that under the verification of two kinds of positioning environments data,the average positioning error of this algorithm is 0.9899 m and 1.3452 m respectively,and its accuracy is better than comparison algorithms.(2)In order to solve the problem that localization accuracy is deteriorated by multipath effect in the complex indoor environment,based on full analysis of CSI data,this paper proposes a fingerprint localization algorithm based on multipath division and 3D convolutional neural network(3DCNN).The algorithm aggregates amplitude image and phase image of CSI.In the offline phase,the clustering method is first used to analyze the number of clusters of fingerprint data,this value represents the extent to which the reference point is affected by the multipath effect.Combined with the threshold principle,the fingerprint library is divided into two kinds of sublibraries with different extent of multipath,which are both deeply studied by 3DCNN;in the positioning stage,the number of cluster classes of test point is first calculated,and according to the calibration algorithm to determine sub-library which the point belongs,and finally use the corresponding 3DCNN model to estimate its position.The experimental results show that the proposed algorithm can achieve a good positioning efficiency in a complex environment,and be able to control the positioning error of 78% measurement results within 1.25 m,which is at least 28% higher than related algorithms.The paper has 38 pictures,19 tables,and 68 references.