Adaptive Rate/Power Control With ML-Based Channel Prediction for Optical Satellite Systems

Free-space optical (FSO)-based satellite communications, owing to extremely high data rates and global coverage capability, have recently drawn substantial research attention. The adverse issues on FSO-based satellite links, including atmospheric turbulence and pointing error, pose various challenges in designing and deploying such systems. Nevertheless, recent efforts in error-control design focusing on adaptation-based mitigation techniques face practical restrictions due to the outdated feedback channel state information (CSI) caused by long-distance/high-latency satellite links. This paper addresses the design of an adaptive rate/power control scheme using machine learning (ML)-aided channel prediction for FSO-based satellite systems. Notably, we employ the echo state network (ESN) model, an efficient form of recurrent neural network (RNN), for channel prediction. The design proposal facilitates concurrent control of data rate and satellite's transmitted power for each equal-duration channel state, leveraging accurately predicted CSIs. The average required transmitted power and energy efficiency performance metrics are analytically derived. Numerical results demonstrate the severe impact of outdated CSIs on the performance of FSO-based satellite systems and highlight the necessity of our design proposal. Moreover, we confirm the effectiveness of the ESN model for FSO channel prediction by comparing its performance with other ML approaches in terms of predicted accuracy and complexity.