Research On Indoor Positioning Method Based On Channel State Information Of Multimodal Deep Networ

Location Based Services(LBS)is playing a vital role in many fields and has provided great convenience to people’s lives.Therefore,relevant scholars at home and abroad have successively carried out analysis and research on it.With the popularization and application of wireless network technology,indoor positioning systems based on channel state information(CSI)contain rich multipath information and have high robustness,which has attracted the attention of many researchers.How to make full use of the information contained in the CSI to build a low-cost,highprecision,and high-stability indoor positioning system has extremely high research significance and research value.In recent years,due to the rapid development of machine learning,it has become a new research trend to construct original CSI information into images and use deep networks to train and learn them in the existing CSI-based indoor positioning methods.However,there are still some problems and challenges,mainly as follows:(1)How to extract more stable and reliable features from CSI information and reduce the positioning cost of the system.(2)How to process multi-dimensional CSI data to ensure high data utilization.(3)How to choose different deep networks to ensure the positioning accuracy and stability of the system in a complex indoor environment.Based on the above background,this paper proposes two indoor positioning systems based on CSI.The main research contents are as follows:(1)This paper proposes a CSI indoor positioning system based on multi-modal deep convolutional neural network.The first to fourth central moments extracted from the channel impulse response amplitude as feature images are used to increase the data dimension,and this paper preprocess the amplitude information and phase information.Then,three CSI features are input into the corresponding convolutional neural network,and system analyzes the correlation of the three images through the canonical correlation analysis algorithm,and performs feature fusion to build a fingerprint library,and finally realizes the positioning function.(2)This paper proposes a CSI indoor positioning system based on multi-modal confrontation generation network.In order to solve the problem that traditional canonical correlation analysis algorithms cannot perform data mining on nonlinear data,this paper proposes a CSI multi-dimensional image construction method,which will extract the amplitude information,phase information and the first to fourth central moments of the channel impulse response amplitude.The three CSI information is analyzed and fused to construct a CSI multi-dimensional image through the core canonical correlation analysis algorithm.Then,the fused image is input into the confrontation generation network for training,and finally the positioning function is realized.(3)In order to evaluate the performance of the proposed method in the actual environment,this paper builds a positioning system,collects relevant data in different measured scenarios,and combines the positioning scheme proposed in this paper to verify its positioning performance in the actual scene.