CNN-BiLSTM-ATTENTION: A Novel Neural Network with Attention Mechanism for NLOS Identification of UWB Signal

Ultra-wideband signals have been widely used in high-precision indoor positioning because of their large bandwidth, low delay and strong penetration. However, the complex indoor environment will hinder the propagation of ultra-wideband signals, resulting in unexpected non-line-of-sight ranging errors and positioning errors. The channel impulse response (CIR) of a signal contains rich feature information, which can be used to classify line-of-sight signals and non-line-of-sight signals. In this paper, we propose a network model combining the ATTENTION mechanism with the bidirectional long short-term memory network and convolutional neural network, which can learn many feature information in signal CIR through the model, so as to realize the model's high accuracy classification. The length of the learned CIR can be adjusted as needed to balance the accuracy and computation of the network. In this paper, the optimal structural design is selected by comparing CNN-BiLSTM-ATTENTION of different structures. In this paper, the DW1000 chip dataset of Decawave company is used for training and testing, and the classification accuracy is as high as 87.91%. At the same time, compared with the existing mainstream deep learning methods, the results show that the proposed model has higher classification accuracy than LSTM, CNN-LSTM and CNN-BiLSTM.