Related Experiment Video
Updated: May 14, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
Domain-Adversarial Neural Network for UWB NLOS Identification in Multiple Environments
Suying Jiang1,2, Jiachun Li3, Yadong Xu1
1School of Electrical and Control Engineering, Shaanxi University of Science and Technology, Xi'an 710021, China.
Sensors (Basel, Switzerland)
|May 13, 2026
Summary
This study introduces a new method for identifying Non-Line-of-Sight (NLOS) signals in Ultra-Wideband (UWB) localization systems. The approach enhances accuracy and generalization across different environments, improving positioning performance.
Area of Science:
- Signal Processing
- Machine Learning
- Localization Systems
Background:
- Accurate Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) identification is vital for Ultra-Wideband (UWB) localization systems.
- Existing NLOS identification methods lack cross-domain adaptability and fail to generalize to new environments.
- Traditional feature extraction methods struggle with the nonlinear characteristics of Channel Impulse Response (CIR) data.
Purpose of the Study:
- To propose a novel NLOS identification strategy with enhanced cross-domain generalization capabilities for UWB systems.
- To develop a robust feature extraction model capable of capturing deep nonlinear characteristics from CIR data.
- To integrate a hybrid feature extraction model with a Domain-Adversarial Neural Network (DANN) for improved NLOS identification.
Main Methods:
- A CNN-DAE-MLP-Attention (CDM) hybrid model was developed for high-quality channel feature extraction from raw CIR data and handcrafted features.
- The CDM model was integrated into the DANN framework, creating the CDMD algorithm for robust feature representation and domain adaptation.
- The CDMD algorithm was evaluated using measured data from diverse real-world scenarios.
Main Results:
- The proposed CDMD algorithm demonstrated strong generalization ability in cross-domain NLOS identification.
- Accuracies of 77.00% and 72.84% were achieved for cross-domain NLOS recognition from underground parking garage to corridor and underground parking garage to lobby, respectively.
- The study confirmed that limited target-domain samples are sufficient for accurate cross-domain transfer using the proposed model.
Conclusions:
- The CDMD algorithm significantly enhances cross-domain NLOS identification performance in UWB localization systems.
- The hybrid feature extraction and domain-adversarial approach overcomes limitations of traditional methods.
- The proposed strategy offers a promising solution for robust and adaptable NLOS identification in complex environments.