兰州大学机构库
UWB Localization Based on Dual-Channel Neural Network and Total Least Square Method
H. Lv; J. Feng; H. Shou; J. Zhang; T. Cui; Z. Mei
2023-12-25
Source PublicationIEEE Sensors Journal   Impact Factor & Quartile Of Published Year  The Latest Impact Factor & Quartile
ISSN1558-1748
Volume24Issue:3Pages:3477-3487
AbstractThe ultrawideband (UWB) positioning system has been extensively used in positioning. Nevertheless, there are many obstacles in complicated environments, which will lead to the non-line-of-sight (NLOS) propagation of UWB signals. In order to achieve accurate localization, it is imperative to identify the line-of-sight (LOS) and NLOS signals accurately. Therefore, we propose a text dual-channel neural network for NLOS/LOS recognition, and the text time-domain and text time–frequency domain characteristics of the signals are considered comprehensively. In the first channel, we employ a convolutional neural network (CNN), time convolutional network (TCN), and self-attention (CNN-TCN-Attention) to process the UWB channel impulse response (CIR). In the second channel, we first convert CIR signals into images using the complex Gaussian wavelet transform (CGWT) and then use CNN to process the images. Finally, the results of the dual channels are integrated to realize signal classification. In the experiment, open-source datasets, local datasets, and mixed dataset are used for signal classification. Experimental results exhibit that the proposed text dual-channel neural network achieves the classification accuracy of 89.24% on open-source datasets, 95.33%, 87.50%, 98.83%, and 97.83% on datasets collected in four local complex environments, and 90.49% on mixed dataset. Finally, we input the actual sample sequences collected locally into the model, and after NLOS/LOS classification is realized, the total least square (TLS) method is used for positioning. The experimental results demonstrate that the proposed algorithm has excellent performance in intricate scenarios.
KeywordConvolutional neural network (CNN)-time convolutional network (TCN)-Attention complex Gaussian wavelet transform (CGWT) non-line-of-sight (NLOS) identification total least square (TLS) ultrawideband (UWB) positioning
PublisherIEEE
DOI10.1109/JSEN.2023.3344288
Indexed ByIEEE
Citation statistics