Machine learning-based EDFA Gain Model Generalizable to Multiple Physical Devices
This work addresses the need for accurate and generalizable gain models in optical communication systems, though it appears incremental as it applies existing machine learning methods to a specific domain.
The researchers tackled the problem of predicting gain in erbium-doped fiber amplifiers by developing a neural-network model from experimental data, achieving low prediction errors with MSE ≤ 0.04 dB² for the training device and ≤ 0.06 dB² for generalization to different units.
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 $\leq$ 0.04 dB$^2$) and different physical units of the same make (generalization MSE $\leq$ 0.06 dB$^2$).