Deep learning-enhanced NLOS NOMA-VLC system based on chirp spread spectrum modulation

With the rapid advancement of smart cities, visible light communication (VLC) is increasingly viewed as a promising complement to traditional radio frequency (RF) wireless networks. Non-orthogonal multiple access (NOMA) has been recognized as an effective technique to enhance spectral efficiency in VLC systems. However, practical line-of-sight (LOS) VLC is vulnerable to blockages, while non-line-of-sight (NLOS) transmission suffers from significant attenuation and multipath effects. To address these challenges, we propose an NLOS NOMA-VLC system employing chirp spread spectrum (CSS) modulation to maintain reliable links under obstructions while enhancing transmission rates. Additionally, obtaining accurate channel state information remains challenging in mobile VLC environments. In this work, we propose a denoising-attention-graph neural network combined with a successive interference cancellation network (DAG-SICNet) based demodulator for NOMA-VLC, jointly realizing signal compensation and recovery. Both simulation and experimental results show that the proposed demodulator can effectively mitigate both linear and nonlinear transmission impairments introduced by NLOS propagation and eliminate interference between multiple users. It achieves transmission rates of 40.7 and 24.3 Mbps over a 1.5 m communication distance for User1 and User2, respectively, demonstrating higher reliability and robustness.