Experimental validation of machine-learning based spectral-spatial power evolution shaping using Raman amplifiers.

Area of Science:

• Optical Engineering
• Machine Learning Applications
• Fiber Optic Amplifiers

Background:

• Raman amplifiers are crucial for signal amplification in optical communication systems.
• Precise control over signal power evolution across frequency and distance is essential for performance.
• Existing methods for controlling amplifier parameters can be complex and time-consuming.

Purpose of the Study:

• To develop and experimentally validate a real-time machine learning framework for controlling Raman amplifier pump power.
• To shape signal power evolution in two dimensions: frequency and fiber distance.
• To automatically achieve target 2D power profiles with high accuracy.

Main Methods:

• Utilized a framework combining a convolutional neural network (CNN) and differential evolution (DE) for pump power optimization.
• Applied the optimization framework online to a Raman amplifier setup.
• Optimized power values of four first-order counter-propagating pumps.

Main Results:

• Achieved very low maximum absolute error (MAE) (<0.5 dB) between target and obtained 2D power profiles.
• Demonstrated successful multi-objective optimization for flat gain levels and minimum spectral excursion.
• DE achieved less than 1 dB maximum gain deviation for target flat gain levels under unconstrained pump power.

Conclusions:

• The real-time machine learning framework effectively controls Raman amplifier pump power for precise 2D signal shaping.
• The system offers high accuracy and robustness in achieving desired optical amplifier performance.
• The framework shows potential for advanced optical communication system design and optimization.