Physics-Informed Neural Network-based Indoor Non-Line-of-Sight Optical Wireless Communication

Indoor optical wireless communication (OWC) offers high-speed and secure transmission, but is vulnerable to blockage under line-of-sight (LOS) conditions. Non-line-of-sight (NLOS) OWC mitigates this by leveraging diffused optical paths, yet introduces a challenging channel environment with severe multipath dispersion, signal attenuation, device nonlinearity, and inter-symbol interference (ISI)—especially at high data rates. These factors significantly degrade performance and limit the effectiveness of traditional equalizers. This letter proposes a physics-informed neural network (PINN)-based equalizer designed to address the challenges of NLOS indoor OWC. Compared to conventional equalizers, the PINN equalizer improves the bit error rate (BER) by 24.6% to 70.87% under varying data rates and silicon photomultiplier (SiPM) bias currents. Compared to an equalizer based on a vanilla recurrent neural network (RNN), our proposed PINN equalizer reduces the training time by 8.21% to 52.22%. Experimental results confirm the potential of PINN for efficient and robust NLOS indoor OWC.