Low-Complexity Channel Prediction Based on Retroreflection of Auxiliary Beam and Deep Learning for Free-Space Optical Communication Systems

In this paper, a low-complexity channel prediction scheme for adaptive FSO systems based on auxiliary beam retroreflection and deep learning is proposed. The scheme enables direct channel prediction at the transmitter without requiring channel information feedback links. Simulation results show an improvement of prediction accuracy using the method predicting downlink CSI from the retroreflective beam compared to the method predicting downlink CSI from the uplink CSI. Additionally, we explore the use of prediction model based on the Conv-GRU network, which incorporates both spatial and temporal domain optical field features as the input data to further enhance prediction accuracy.