Channel Non-Line-of-Sight Identification Based on Convolutional Neural Networks
Qingbi Zheng, Ruisi He, Bo Ai, Chen Huang, Wei Chen, Zhangdui Zhong, Haoxiang Zhang
Abstract
The distinction between line-of-sight (LOS) and non-line-of-sight (NLOS) channels is important for location awareness related technologies and wireless channel modeling. So far, most of the existing methods identify the LOS and NLOS channels based on the characteristics of radio propagation, e.g., using the Ricean <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">K</i> factor. However, the Ricean <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">K</i> factor is sensitive to the propagation environment, and it is thus difficult to find a proper threshold for NLOS identification. In this letter, we propose a novel NLOS identification method based on the convolutional neural network (CNN). Evaluated by channel measurement data, the proposed algorithm achieves better performance compared with the existing conventional method. Firstly, the CNN network is trained by using the pre-labeled LOS and NLOS data collected from channel measurements. The network parameters are set based on the feedback of training. Then, the method is validated by using different datasets. Compared with the Ricean <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">K</i> factor based identification method, the accuracy of which is 0.86, the proposed method shows higher accuracy of 0.99 for the NLOS channel identification.