UWB (N)LOS Identification Based on Deep Learning and Transfer Learning
Jianglong Li, Songlin Liu, Yipin Gao, Yunzhu Lv, Hua Wei
Abstract
In the field of ultra-wideband (UWB) communications, the ability to identify line-of-sight (LOS) and non-line-of-sight (NLOS) signals is of critical importance for the precise positioning and communication performance of devices. Conventional machine learning (ML) and deep learning (DL) methods have proven to be insufficient for achieving satisfactory accuracy levels in unmeasured or variable environments, and the computational cost of these methods is also high. In this letter, we propose a method based on convolutional neural networks (CNN), bidirectional long short-term memory networks (BiLSTM), and transfer learning (TL). Experimental results indicate that when utilizing the open-source dataset eWINE, the CNN-BiLSTM model achieves an average recognition accuracy of 84.38% in a single scene. This represents a 10% improvement in accuracy compared to other methods. In the case of small samples, CNN-BiLSTM+TL reduces training time by 68% and required training volume by 50% compared to CNN-BiLSTM, with no more than a 1% difference in identification performance.