Research On NLOS Recognition And Suppression Algorithm Based On Deep Learning

With the gradual arrival of the era of interconnection of all things,location information has become more and more important in work and life.The positioning technology closely related to location information is also developing.Different from outdoor positioning,the complex indoor environment makes it difficult for satellites to play a role,so there are many indoor positioning schemes.Among many schemes,the positioning algorithm based on UWB has the best accuracy,and the error caused by NLOS is the main factor affecting the positioning accuracy of UWB.Based on this,the main content of this thesis focuses on NLOS,and improves the positioning accuracy by identifying and suppressing NLOS.In recent years,the application of deep learning in indoor location is more and more.This thesis will also use deep learning to realize NLOS recognition and suppression algorithm.Collecting data and making data sets is the premise of applying deep learning algorithm.This thesis improves the original UWB positioning equipment in the laboratory,and realizes the collection of signal strength,CIR and other data by modifying the underlying code.This thesis collects a large number of NLOS and LOS data in different indoor scenes,compares and analyzes the distribution characteristics of different signal strength parameters in NLOS and LOS environments,selects the signal strength parameters that have a positive impact on NLOS recognition,and forms the data set required by this deep learning algorithm together with CIR.In this thesis,convolutional neural network is used to form NLOS recognition network.Because each sample data is one-dimensional data,the NLOS recognition network uses onedimensional convolutional neural network.NLOS identification network predicts the probability that the ranging data comes from NLOS environment according to each sample data.This probability is called NLOS factor in this thesis.In the location calculation,the ranging data of the three base stations with the lowest NLOS factor are used for location estimation.After static and dynamic experimental tests,the results show that the accuracy of localization after NLOS recognition network has been improved compared with trilateral localizationIn terms of NLOS suppression,this thesis realizes the design of ranging error correction network RECNet through residual neural network.RECNet is a regression model of ranging error,which predicts the ranging error of the base station according to the signal strength and CIR information of the base station.The predicted ranging error is used to correct the original ranging value of the base station,and the corrected ranging value is obtained for position estimation.In this thesis,the performance of RECNet in predicting base station ranging error is tested.The experiment shows that RECNet can effectively predict ranging error in NLOS environment.After static and dynamic experimental tests,the results show that the accuracy of position solution using the corrected ranging value has been improved compared with trilateral positioning.The trilateral positioning algorithm,the positioning algorithm based on NLOS recognition and the positioning algorithm based on RECNet are compared.The positioning algorithm based on RECNet has better positioning accuracy.