Multiclass Identification of NLOS Conditions in UWB Localization Based Machine Learning Methods

This paper explores the critical task of accurately distinguishing line-of-sight (LOS) and non-line-of-sight (NLOS) propagation channels in ultra-wideband (UWB) indoor localization systems. Using a comprehensive dataset of real measurements from a Qorvo DW1000 UWB device, the study evaluates the performance of multiple machine learning classifiers, including Decision Tree (DT), Random Forest (RF), XGBoost, Support Vector Machine (SVM), Naive Bayes (NB), AdaBoost (ADA), Stochastic Gradient Descent (SGD), Logistic Regression (LR), and K-Nearest Neighbors (KNN), under various indoor conditions. The results demonstrate that machine learning techniques can effectively classify LOS, Multipath (MP), soft-NLOS, and hard-NLOS propagation channels within UWB ranging systems. XGBoost achieved the highest overall accuracy of 99.47%, followed by DT at 98.97% and RF at 98.59%. Notably, classifiers showed significant improvement with our optimized dataset, underscoring the importance of feature selection and data balancing. Even with a reduced feature set, classification accuracies remained high, validating the robustness of the models. These findings underscore the potential of machine learning to significantly enhance the precision of UWB indoor localization systems, paving the way for more reliable and accurate indoor positioning technologies.