Research On Indoor Localization Technology Based On CSI Fingerprint

With the development of wireless communication technologies and the rapid spread of smart terminal devices,more and more location-based services are integrated into various products and applications.Since satellite signals are blocked by buildings,satellite localization systems cannot be directly applied to indoor localization.To solve this problem,researchers have developed solutions based on wireless communications,geomagnetism,and computer vision.The Wi-Fi-based indoor localization has become an important branch of indoor localization because of the pervasive availability of Wi-Fi infrastructure and the richness of mobile terminal.Traditional Wi-Fi-based indoor localization uses Received Signal Strength(RSS)as a feature parameter.However,RSS is an energy superposition of multipath signals with poor time stability.It cannot accurately describe the characteristics and variations of the indoor environment,resulting in large localization errors.In recent years,hardware manufacturers have gradually opened up internal information of network cards.Researchers can easily obtain fine-grained channel state information(CSI),which brings research of Wi-Fi-based indoor localization to a new stage.Compared with RSS,CSI has stronger time stability and greater positional difference.IEEE 802.1 In,which uses Multiple-Input Multiple-Output(MIMO)and Orthogonal Frequency Division Multiplexing(OFDM),can provide CSI on subcarriers and multiple antennas at different fr-equencies,further enriching the source of localization feature parameters.Most of the existing CSI-based researches use the CSI amplitude of a single antenna or a single moment to construct a localization model.The mining of multi-antenna and multi-time is not enough,and the localization accuracy still has room for improvement.After introducing the theory and basic technologies of indoor localization,this thesis verifies the premise that CSI can be used as the localization characteristic parameter through experiments.This thesis designs a fingerprint matrix with rich localization features by combining CSI of subcarriers at different frequencies,different receiving antennas,and time span.By means of the characteristics of non-full connection and local weight sharing of convolutional neural networks,the features of high-dimensional fingerprint matrix are extracted efficiently and the localization model is constructed.A device-free localization method based on CSI amplitude fingerprint and convolutional neural network is designed for the needs of home security and home care.This thesis proposes a fingerprint feature enhancement method based on time domain filtering to reduce the interference of environmental background information,which improves the proportion of the human body in the localization features.The method successfully improves the localization accuracy of the edge position in the area of interest.In a typical office,the localization accuracy is 97.19%,which is better than the localization method using other classifiers or artificially designing features.Motivated by the needs of indoor navigation and mobile social networks,this thesis designs an active localization method based on CSI amplitude and phase fimgerprint and convolutional neural network.The phase error compensation method based on linear transformation is used to eliminate the obstacle that the original phase of CSI cannot be applied to the localization due to large error.This thesis combines the amplitude with the corrected phase to form an active localization fingerprint matrix with richer features.In the composite scene of the office and the corridor,the average localization error is 1.13 m,which is better than the existing localization methods based on RSS and CSI fingerprints.