Table 8.

Summary of the applications of ML for FSO communications.

References	ML Model Architecture	Contributions	Remarks
[129]	CNN and SVM algorithms	Channel prediction	CNN outperformed the SVM, predicted channels with ASE noise well, and provided an accurate prediction for turbulence and pointing error in low-speed transmission
[130]	DCNN algorithm	Atmospheric turbulence problems detection	Achieved the optimum performance with low complexity, 2×, 3×, and 7.5× faster for 16, 64, and 256 modulation orders, respectively
[131]	Unsupervised-based technique	Estimated the number of concurrently transmitting users sharing time, bandwidth, and space resources	Achieved over 92% accuracy in differentiating simultaneously transmitting users, even in moderate atmospheric turbulence
[132]	Supervised learning-based ML method	Transmission quality estimation	SVM achieved the highest accuracy of 92%
[133]	Combined GNN and CNN schemes	Transmission quality estimation	Efficiently received improved signals that had deteriorated and showed better classification accuracy