CNN-Based NLOS Detection in UWB Systems

This technical report presents a comparative analysis of three approaches for Non-Line-of-Sight (NLOS) detection in Ultra-Wideband (UWB) systems, focusing on the application of a one-dimensional Convolutional Neural Network (1D-CNN). The report outlines the challenges posed by NLOS conditions, which can lead to significant localization errors due to obstructions in the direct signal path. The proposed 1D-CNN method is evaluated against two established detection algorithms: a statistical approach and a feature-based neural network. The evaluation is based on real-world measurements using the Decawave DW1000 hardware, demonstrating the effectiveness of the CNN-based method with a detection ratio of 99.91%, compared to 60.81% for the statistical approach and 95.58% for the feature-based neural network. The report details the methodologies employed, including preprocessing of received signals and the implementation of the detection approaches, and discusses the implications of the findings for improving indoor localization accuracy.