ANALYSIS OF ERBIUM-DOPED FIBER AMPLIFIERS USED IN RING BACKBONE OPTICAL NETWORKS USING MACHINE LEARNING

This study investigates erbium-doped fiber amplifier placement in wavelength-division multiplexing ring backbone optical networks and evaluates whether machine-learning-guided screened search can improve feasibility-aware planning. Ring-8 to Ring-16 topologies were simulated under standard and heterogeneous scenarios using a physical layer model that includes received power, amplified spontaneous emission (ASE), nonlinear interference, and an enhanced Gaussian noise (EGN) inspired correction. Candidate amplifier masks were ranked with a no-leak screening model and refined through bounded local search. Random Forest delivered the best screening performance, reaching receiver operating characteristic area under the curve (ROC-AUC) values of 0.9971 in the standard dataset and 0.9379 in the harder dataset, while leave-one-topology-out validation showed robust generalization across unseen ring sizes. In scaling experiments, the improved screened strategy matched the exhaustive optimum feasible ratio through Ring-16 and became 7.44 times faster than exhaustive search at Ring-16. These findings show that machine learning can serve as an efficient and interpretable decision support layer for impairment-aware EDFA placement without replacing full physical layer simulation.