Learning-Based Ultra-Wideband Indoor NLOS Identification

This paper focuses on improving the identification accuracy of the line-of-sight (LOS) and non-line-of-sight (NLOS) conditions by channel impulse response (CIR) using machine learning methods. We propose a new definition for soft and hard NLOS based on ranging error percentages. Our research shows that accurately differentiating these two NLOS scenarios significantly enhances the LOS/NLOS identification accuracy. Experiment results demonstrate notable improvements in LOS/NLOS classification accuracy, with the accuracy of LOS detection improving from 72.33% to 93.84% using a convolutional neural network (CNN). In differentiating between hard NLOS and LOS, our accuracy reaches up to 99.82%.