Research On Indoor Ultra-Wideband Ranging And Positioning And Error Suppression Technology

In recent years,indoor positioning technology has entered a stage of rapid development.Location-based services were originally limited to outdoor scenes,and gradually expanded to buildings.From large public buildings such as airports,shopping malls and hospitals to many industrial scenes,many indoor environments require positioning services.Among many indoor positioning technologies,indoor positioning technologies based on ultra-wideband signals are highly favored by scientific researchers and equipment manufacturers due to their high positioning accuracy and strong anti-interference ability.Suppressing positioning error,as a hot research direction of ultra-wideband indoor positioning technology,will be the main research content of this thesis.Under the premise that the positioning data comes from online open source data sets,and the positioning method adopts the indoor positioning method based on the time of arrival,this article mainly starts from the two aspects of non-line-of-sight ultra-wideband positioning signal recognition and ranging error correction to solve the positioning error problem.First,in order to improve the positioning accuracy of the position estimation algorithm,the weighted least squares algorithm with lower complexity is selected,in which the weight factor is set to the mixed calculation result of the prediction error and the measured distance according to the positioning method used in this paper.The simulation experiment compares the cumulative distribution probability and root mean square error of the two algorithms,and proves that the weighted least squares algorithm can effectively improve the accuracy of position estimation.The existing non-line-of-sight ultra-wideband signal recognition method mainly distinguishes non-line-of-sight signals based on time-domain features.In order to improve the accuracy of signal recognition,wavelet transform is used to facilitate the extraction of time-frequency domain features at the same time.Non-line-of-sight signal recognition method based on wavelet transform and convolutional neural network.Experiments show that for 6 different indoor scenes in the dataset,the non-line-of-sight positioning signal recognition models all produce effective recognition results,and the performance in specified scenes is better than that of traditional recognition methods.Finally,in view of the problem of low training efficiency of the non-line-of-sight signal recognition model,the power delay profile in the form of down-sampling,the energy of the channel impulse response and the measured distance are used as the input characteristics of the network,instead of the traditional channel impulse response,a branch with branch is proposed.The structure of the convolutional neural network,and use the non-line-of-sight signal recognition model and the ranging error regression prediction model based on this network structure to realize the positioning error suppression in the simulated scene.During the experiment,the problem of weak generalization ability of the network model was solved by mixing the data in a variety of indoor scene positioning data.At the same time,according to the measurement method of positioning data,five positioning strategies are simulated to test the performance of this method in restraining positioning errors.It can be seen from various comparative experiments that the proposed positioning data processing method can greatly accelerate the training speed of the non-line-of-sight recognition model without affecting the recognition performance.At the same time,five positioning strategies comprehensively tested the positioning error suppression performance of the non-line-of-sight signal recognition model,the ranging error regression prediction model and the two position estimation algorithms.After identifying and removing the non-line-of-sight propagation in the positioning information,the measured distance is corrected by the ranging error regression prediction model,and finally the prediction error and the measured distance are weighted to the least square algorithm.This positioning strategy produces the best positioning effect.The average error after a large number of repeated simulation experiments can reach ten centimeters.