Performance Enhancement of Satellite-to-Ground FSO System using Deep Learning-Based Detection

This paper proposes to use deep learning to improve the performance of satellite-to-ground free-space optical (FSO) communication system under the negative impact of many physical impairments such as atmospheric turbulence, atmospheric attenuation, and free-space path loss. We first implement a simulation of the satellite-to-ground FSO system over Gamma-Gamma turbulence channel using a conventional detection to obtain the binary data at the receiver. This data is subsequently utilized for training the deep learning (DL)-based detection model via a convolutional neural network (CNN) model. Simulation results show that bit-error rate (BER) of the satellite-to-ground FSO communication system is significantly lower than that of the traditional systems. Moreover, we demonstrate that the proposed DL detection model is especially effective in term of the performance gain.