Time-ahead forecasting of fiber-coupled signals using temporal convolutional networks for free-space optical links [Invited]

Free-space optical communication (FSO-Com) systems are strongly affected by turbulence-induced signal fading, which limits link performance and reliability. This paper presents an approach for time-ahead forecasting of received fiber-coupled laser power signals at FSO-Com terminals using temporal convolutional networks (TCNs), aimed at proactive mitigation of atmospheric turbulence effects. The study utilized atmospheric field measurements obtained over a 7 km propagation path and synthetic time-series datasets generated using wave-optics numerical modeling and simulation. The developed TCN architecture demonstrated the capability to accurately forecast received optical power fluctuations 10–20 timesteps ahead with average inference time of approximately 1.2 ms per forecast. The results demonstrate the potential of TCN-based received signal processing for enhancing FSO-Com link performance and reliability.