Non-Line-of-Sight Identification and Localization Using UWB Based on Deep Learning

Due to its characteristics of high precision, high reliability, and low power consumption, UWB technology is widely used in indoor positioning. However, the performance of UWB positioning systems can be affected by issues such as multipath effects and signal attenuation in both non-line-of-sight (NLOS) and line-of-sight (LOS) environments. This paper proposes an algorithm for NLOS and LOS state identification and localization using UWB, based on deep learning. First, a recognition model suitable for NLOS and LOS environments is established by analyzing UWB channel models and signal characteristics. Then, a method for identifying NLOS and LOS states is designed using deep learning techniques, with classification accuracy metrics introduced to evaluate model performance. To validate the proposed positioning algorithm, a UWB node positioning system is designed and tested in real environments. Experiments were conducted in two different environments—a basement and a lounge—to verify the positioning performance. Subsequently, the deep learning-based NLOS error identification and mitigation algorithm proposed in this paper was tested. The experiments compared the effects of different deep learning algorithms on NLOS error identification and mitigation. Results show that the positioning algorithm using the Convolutional Neural Network Long Short-Term Memory (CNN-LSTM) model performs best in identifying and mitigating NLOS errors, achieving the highest positioning accuracy. The study results demonstrate that the proposed algorithm not only has theoretical value but also shows great potential for practical applications.