Machine learning-based EDFA Gain Model Generalizable to Multiple Physical Devices

We report a neural-network based erbium-doped fiber amplifier (EDFA) gain model built from experimental measurements. The model shows low gain-prediction error for both the same device used for training ( MSE ≤ 0.04 dB
²
) and different physical units of the same make (generalization MSE ≤ 0.06 dB
²
).