Abstract

Line-of-sight (LOS)/Non-line-of-sight (NLOS) identification in wireless communication systems is crucial for positioning, mobile communication, and wireless sensing. Conventional LOS/NLOS identification approaches generally employ channel features, such as the Rice factor, kurtosis of the received power, etc. However, these approaches have limited identification accuracy and cannot function efficiently in a dynamic environment. The deep learning approach can show better performance; however, it has high computational complexity issues. In this study, we first demonstrate that the channel impulse response (CIR) of the 5G channel outperforms channel frequency response employing convolutional neural networks (CNNs) for LOS/NLOS identification. Then, we propose a simple efficient lightweight CNN-based LOS/NLOS identification approach. The proposed one-dimensional CNN network can effectively extract CIR features and has cross-scenario adaptability. Additionally, for the first time, 5G experimental data were employed for performance verification. The experimental results demonstrate that the proposed approach can achieve an identification accuracy of 93.31% at a computational cost of 1.35 M FLOPs and has higher identification accuracy and speed than existing MWT-CNN deep learning approaches. The code is available at https://github.com/boa2004plaust/SEL-CNN.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call