An Adaptive Approach for Ultra-Wideband Positioning in Complex Environment

Abstract

The machine learning methods for ultra-wideband (UWB) positioning in non-line-of-sight (NLOS) environment either mitigates the NLOS ranging errors after identifying the NLOS signals (indirect mitigation methods) or directly mitigates the errors (direct mitigation methods). Despite their positioning accuracy, the indirect mitigation methods face two problems: the positioning system faces a high computing load, for lots of samples are needed to train the classification model and the regression model, respectively; the uneven distribution of the NLOS signal samples is often ignored, reducing the generalization ability of the regression model. To solve the two problems, this paper designs an adaptive approach to reduce the complexity and improve the positioning accuracy of UWB system in complex environment. Under this approach, the moment-based imbalanced binary classification (MIBC) is firstly adopted to identify the NLOS signal samples, and divides the samples into mild and severe obstruction propagation signals, according to the magnitude of NLOS signal ranging errors; then, the fuzzy comprehensive evaluation (FCE) and Gaussian process regression (GPR) were combined into the F-GPR to mitigate the ranging errors of the two types of the signals. The excellence of the proposed adaptive approach was fully proved through simulations, in comparison with the hybrid method and the global GPR.