Data-driven gain equalization for few-mode erbium-doped waveguide amplifiers using deep neural networks

Area of Science:

• Photonics and Optical Engineering
• Materials Science
• Computational Electromagnetics

Background:

• Few-mode erbium-doped waveguide amplifiers (FM-EDWAs) are crucial for high-capacity optical communication.
• Optimizing FM-EDWAs is challenging due to complex interdependencies between structural parameters and performance metrics.
• Existing methods struggle to integrate gain equalization and crosstalk suppression into a unified design framework.

Purpose of the Study:

• To propose a joint multi-objective optimization method for designing six-mode FM-EDWAs.
• To establish a precise mapping between waveguide structural parameters and key performance indicators.
• To simultaneously optimize average gain, differential mode gain, and crosstalk.

Main Methods:

• Integration of a Transformer model with an improved Non-dominated Sorting Genetic Algorithm II (NSGA-II).
• Development of a unified optimization framework for waveguide structural parameters and performance metrics.
• Simulation and analysis of a six-mode amplification device under single LP01 pumping.

Main Results:

• Achieved balanced amplification across six spatial modes (LP01, LP11a, LP11b, LP21a, LP21b, LP02).
• Optimized device exhibits an average gain of 21.84 dB and a differential mode gain of 0.84 dB.
• Demonstrated crosstalk suppression below -22.6 dB at 1550 nm, confirming effective mode crosstalk mitigation.

Conclusions:

• The proposed method effectively designs FM-EDWAs with simultaneous optimization of gain and crosstalk.
• This work provides significant advancements for on-chip multimode amplifier design.
• The findings pave the way for enhanced high-capacity mode-division multiplexing systems.