A Design of Multi-Head Attention Neural Network for UWB NLOS Identification in Outdoor

Abstract

In this paper, we introduce a method of classifying UWB CIR data into LOS and NLOS environments by applying the multi-head attention algorithm. The 1016 UWB CIR values sampled at 100 ms intervals are divided into 100 segments. By comparing the classification time and accuracy of the LSTM-CNN algorithm and the multi-head attention algorithm, it is shown that the latter achieved a classification accuracy of 94.41% for LOS/NLOS environments, outperforming the LSTM-CNN model.