UWB Wireless Positioning Method Based on LightGBM

Abstract

In the ultra-wideband indoor positioning sceneraio, the non-line of sight (NLOS) propagation may be caused by obstacles, which may lead to the deviation of ranging value and affect the positioning precision. Therefore, we propose a NLOS identification and error regression positioning algorithm based on light gradient boosting machine (LightGBM). Firstly, ReliefF algorithm combined with Spearman correlation coefficient is used to analyze the feature correlation, and eight channel features such as total channel impulse response power and standard deviation of noise are selected as NLOS identification features. Then, we adopt genetic algorithm to optimize the hyperparameters of LightGBM for NLOS identification. On this basis, the proposed error regression model based on convolutional neural network (CNN) combined with LightGBM is used to correct the ranging results, so as to achieve highprecision positioning. Through the verification on the public dataset, the NLOS identification accuracy reached 91.8%, and the positioning precision is improved by 45cm after correcting the ranging results.