Massive MIMO User Positioning Based On Deep Learning

As human society enters the 5G era,location-based services are increasing rapidly,automatic driving and unmanned factories in daily life all need to obtain the accurate location information of equipment,Therefore,positioning technology has attracted more and more attention.The research direction of traditional location technology is mainly divided into parameter-based location technology and fingerprint-based location technology,but the traditional parameter-based location method needs high computational complexity to estimate the channel parameters accurately,For the fingerprint location technology based on deep learning,it is necessary to collect the fingerprint data in the experimental environment to form a fingerprint database and complete the location through fingerprint matching,The cost is very high and it is highly dependent on the experimental environment,the positioning accuracy fluctuates greatly when the environment changes.In addition,due to the wide application of satellite positioning,the current research on positioning technology mainly focuses on the indoor environment,but the error of satellite positioning is large in the non line of sight(NLOS)scene.To solve this problem,we propose a fingerprint-free terminal location method based on deep learning,the problem of user localization in large-scale multiple input multiple output(MIMO)system is studied and a fingerprint-free neural network localization method based on 3D Convolutional Neural Networks(3D CNN)is proposed.Different from traditional location methods,this thesis develops the research direction of current location technology,which integrates the two parts of channel feature extraction and feature processing into the neural network model,and creatively fits the universal mapping relationship between channel features and terminal location through a large number of high-dimensional channel state information(CSI)in different environments to reduce the impact of environmental changes on the performance of the algorithm.Firstly,the scene discriminator and the positioning model under different user scenes are designed and trained in the offline stage.In the online stage,the CSI of the user needs to be collected and input into the scene discriminator,and then the position estimation can be realized by inputting the data into the corresponding trained network models according to the user scene estimated by the discriminator.Therefore,the scheme also includes the NLOS scene positioning model.On the other hand,in order to make full use of the advantages of Massive MIMO technology,the base station needs to estimate the CSI from each terminal to the base station,and the base station can also use these CSI to locate the user terminal;In this way,this method can be applied into the existing communication system to reduce the positioning cost.At the same time,it does not need to estimate the channel parameters and has low algorithm complexity.In our thesis,The COST 2100 channel model is used to simulate a large number of environments to verify the performance of the proposed scheme.In each environment,the information capacity is improved by collecting high-dimensional CSI.As the corresponding channel characteristics in this environment,the higher dimension of CSI used in this method includes the antenna,frequency and time.Simulation results show that the scheme can maintain high positioning accuracy and achieve high positioning performance for the new environment that does not participate in the training phase.