Design and evaluation of an intelligent flat-gain<i>O</i>+<i>E</i> band hybrid optical fiber amplifier: a machine learning approach

This study presents the design of an <i>O</i>+<i>E</i>-band hybrid optical amplifier combining praseodymium-doped fiber amplifiers (PDFAs), thulium-doped fiber amplifiers (TDFAs), and a Raman amplifier (RA). The PDFAs and TDFAs are arranged in parallel to amplify signals between 1280 and 1470 nm, but there is a low gain (16 dB) in the 1360-1410 nm region. This gap is addressed by integrating a RA to enhance the gain to an average of 35 dB across the targeted range. To optimize the RA's performance, a machine learning technique called P2RAnet is developed to predict optimal pump parameters. The P2RAnet model, trained with neural networks, minimizes errors in gain prediction, achieving a mean absolute error of 0.5 dB and a mean squared error of <0.4<i>d</i><i>B</i>². A SHapley Additive exPlanations analysis identifies the pump wavelength as the most influential factor. The model's predictions, validated using OptiSystem software, show a gain efficiency of 96.9% and a flatness of 6.85 dB. With optimized parameters, the amplifier achieves over 29.43 dB gain and a noise figure under 4.1 dB from 1290 to 1470 nm.