A Parallel Deep Learning Based NLOS Identification Method Using CIR Signal

With the rapid development of indoor localization technology, Ultra-Wide Band (UWB) technology stands out for their good ability of noise-resistant, strong penetration and high localization accuracy. However, the increasingly complex indoor environment leads to the non-line-of-sight (NLOS) propagation of UWB localization signals, which seriously affects the accuracy of ranging-based localization algorithm. To mitigate the effects of NLOS, it is necessary to identify NLOS propagation first. In this paper, a novel NLOS identification method based on multi-inputs parallel deep learning model and Gramian Angular Field (GAF) is proposed. It is the first to utilize GAF to transform 1-dimision Channel Impulse Response (CIR) signal into 2-dimision colored images, which adds additional high-level abstract features to the CIR signals. In the model training phrase, the original CIR signals are used to extract temporal features by Convolutional Neural Network (CNN), and the GAF encoding images is used to extract visual features by Resnet. The performance of proposed method using open-source real-time measured dataset is compared with exiting popular NLOS identification methods. The experimental results show that our method can effectively improve the accuracy of traditional LOS/NLOS binary classification as well as multi-NLOS scenario classification.