Localization Algorithm Based On 5G Synchronous Signal Measurements

5G is large-scale deployed and developing in China.With the increasing demand for location-based services of mobile terminals,we can use 5G for positioning without laying lots of special positioning hardware.Using 5G for positioning can avoid the passivity of outdoor positioning of GNSS,and make up for the deficiency that GNSS cannot achieve indoor accurate positioning.This paper studies the location algorithm based on fingerprint and path loss model of physical layer synchronous signal measurements to improve the positioning accuracy of 5G,and conducts indoor and outdoor experiments.The main results are as follows:(1)Due to the single variable fingerprints are similar at different locations,which leads to poor positioning accuracy.A convolutional neural network location classification model(CNN-LCM)algorithm is proposed.Use the rich parameters contained in 5G synchronization signal to build multivariable fingerprints(MVF).MVF is used as the input of CNN-LCM,and the model transforms the positioning into image classification.In three indoor experimental scenarios,the positioning accuracy of CNN-LCM and KNN with MVF is higher than that with single variable fingerprints.The positioning error of CNN-LCM in best scenario is 1.96 m when CDF is 80%,and the average positioning error is 1.41 m,which is 24.60% higher than that of KNN.(2)Some reference points have similar features but their actual positions scatter everywhere when using fingerprint algorithm.A hybrid positioning model of resource block number constrained path loss model and CNN-LCM is proposed.The path loss model is trained by selecting reference points that are less affected by the environment,and the weight of each reference point in the CNN-LCM output layer is adaptively adjusted to improve the positioning accuracy.In two indoor LOS experimental scenarios,the positioning error of hybrid positioning model in best scenario is 1.67 m when CDF is 80%,and the average positioning error is 1.29 m,which is 14.57% higher than that of CNN-LCM.(3)In the outdoor environment,the number of 5G base stations is sufficient.However,some base stations are interfered by noise,which seriously reduces the positioning accuracy when using range-based methods.A base station selection based on distance variance(BSDV)algorithm is proposed.Select the base station combination with the minimum distance variance and the terminal located inside or at the outer edge of the polygon formed by the base station.The positioning accuracy of BSDV is similar to that of the horizontal dilution of precision(HDOP)base station selection algorithm,but the computational complexity of BSDV is43.86% lower than that of HDOP algorithm,effectively reducing the instability of matrix inversion.The average positioning accuracy of BSDV is 20.92% higher than that of the algorithm without base station selection.(4)Fingerprint location model based on neural network cannot accurately fit the large outdoor area.A dimension reduction fingerprints KNN(DRF-KNN)algorithm is proposed.The 5G ultra dense networking is used to build multi base stations fingerprints.Reduce the dimension of the original multi base stations fingerprints based on the physical cell identifier,and use the KNN algorithm to obtain the location.The positioning error of DRF-KNN algorithm is 29.72 m when CDF is80%,and the average positioning error is 22.59 m.DRF-KNN algorithm reduces the times of fingerprint matching,and improves the average positioning accuracy by 14.60% compared with KNN.