Research On Indoor Fingerprint Localization Algorithm Based On Channel State Information

With the development of the Internet of Things,the demand for location-based services is becoming more and more urgent.The fingerprint localization system based on channel state information has become the key research direction of location-based services with its low cost and high accuracy.This paper will investigate the indoor fingerprint localization algorithm based on channel state information.To enhance the robustness of the algorithm in an indoor localization system,this paper proposes a fingerprint localization algorithm fusing weighted the locally linear embedding and the gradient boosting decision tree.In the offline stage,firstly,the filtered and phase-corrected original channel state information data are regarded as joint fingerprints,and then weighted using the elastic network algorithm and downscaled using the locally linear embedding algorithm.Finally,the fruit fly optimization algorithm is used to find the globally optimal values of all the super-parameters in the gradient boosting decision tree,which purpose to train offline fingerprint information,meanwhile,build a fingerprint database containing location and fingerprint information.In the positioning stage,the fingerprints of the test points are fed into the ENLLE+GBDT algorithm.The offline fingerprint information with the highest similarity to the fingerprint information of the test point can be predicted,and thus its corresponding actual location can be obtained.It is shown experimentally that in the presence of 2% missing value dataset.The localization accuracy of the algorithm is consistently above 95%,which has high robustness compared to other comparative algorithms.To address the problem of limited mapping capability of positioning algorithms in multi-view indoor positioning systems,this paper proposes a fingerprint localization algorithm fusing the multi-view canonical correlation analysis and the convolutional neural network.In the offline phase,firstly,the original channel state information of the reference points is collected using three access points,meanwhile,converted into feature images of pixel points separately,and pre-processed.The image of each access point is regarded as a view,and then the multi-view canonical correlation analysis is used to learn the discriminative images between and within views,which are regarded as the input of the convolutional neural network.Finally,the mapping relationship between the fingerprint and the location information is saved as an offline fingerprint library.In the positioning stage,the multi-view canonical correlation analysis algorithm is used to learn the views data of the test points,and then the convolutional neural network algorithm is used to perform coarse classification of the test points,finally,the location information of the test points is matched using euclidean distance.The experiments show that the MCCA-CNN algorithm has a higher localization accuracy compared with other comparative multi-view algorithms and has a certain application value.The paper has 36 pictures,13 tables,and 52 references.