Research On WiFi Fingerprint Based Wireless Indoor Localization Algorithms

With the widespread use of smartphones and various wireless devices,as well as the rapid development of Io T technology and smart city systems,the demand for location-based services has been increasing.Among many indoor positioning technologies,WiFi signalbased fingerprint positioning technology stands out for its high positioning accuracy and low equipment cost.In WiFi signals,Channel State Information(CSI)is the attribute information describing the signal channel of a communication link,which can describe the propagation properties of wireless signals in multiple channels and subcarriers in both frequency domain and time domain,and has attracted widespread attention in indoor positioning technology.In CSIbased fingerprint indoor localization techniques,deep learning approaches can achieve high accuracy rates,where Convolutional Neural Network(CNN)have outstanding advantages in processing tensor data,so that CSI can be processed into a multi-dimensional tensor and then trained on the processed tensor using CNN,which exhibits high accuracy.However,deep learning methods require a large amount of sample data to have better localization accuracy,which leads to a large amount of manpower and time consumed in data collection to build a more reliable fingerprint database.In addition,the signal features in the considered real-world scenario change with the indoor environment,which requires re-collection of data from reference points to build the database to adapt to the changing environment.Therefore,the CNN-based fingerprint indoor localization algorithm cannot meet the localization needs arising from small sample scenes and changes in indoor environment.To address the problem of reduced accuracy due to environmental changes in indoor localization,this thesis proposes the Domain Adversarial Neural Network Based Unsupervised CSIFingerprint Indoor Localization Algorithm(DiFi).DiFi is divided into the offline phase and the online phase.In the offline phase,the CSI datas of each location is first processed into CSI tensors.GAN expands the CSI tensors for each location to form a fingerprint database,which is used as the source domain sample set,and then the CNN algorithm is used to train a classification model for indoor localization.The online phase addresses the problem that the fingerprint database in the offline phase no longer matches the changed environment due to the change of the indoor environment.Firstly,the unlabeled CSI tensor is collected and constitutes the target domain sample set,then the source and target domain samples are trained with domain adversarial training to eliminate the differences between the domains,update the classification model in the offline phase,and finally estimate the position of the target in the indoor environment after the change of the environment.The experimental scenario for collecting CSI data in this thesis is a section of a corridor in the first section of the main building of the North Campus of Xidian University,where the localization accuracy of the DiFi algorithm and the CNN-based fingerprint indoor localization algorithm are compared.The experimental results show that the DiFi algorithm can significantly improve the localization accuracy of target domain samples compared with the CNN-based fingerprint indoor localization algorithm in a small sample scenario.