DACNN-aided nonlinear equalizer for a probabilistic shaping coherent optical communication system.

Area of Science:

• Optical Communications
• Signal Processing
• Machine Learning

Background:

• Probabilistic shaping (PS) is crucial for fiber optic systems to approach the Shannon limit.
• Nonlinear equalizers struggle with the non-uniform distributions inherent in PS systems.
• Existing methods face challenges in effectively equalizing PS signals due to distribution mismatch.

Purpose of the Study:

• To propose a novel nonlinear equalizer for probabilistic shaping optical communication systems.
• To address the ineffectiveness of traditional nonlinear equalizers with non-uniform PS signal distributions.
• To reduce training complexity and enhance equalizer performance simultaneously.

Main Methods:

• A distribution alignment convolutional neural network (DACNN) aided nonlinear equalizer was developed.
• The DACNN calibrates the equalizer using the prior distribution of probabilistic shaping.
• This approach aligns the equalizer's training with the signal's non-uniform distribution.

Main Results:

• The DACNN equalizer demonstrated nonlinear equalization for 120 Gb/s PS 64QAM signals over 375 km.
• It improved receiver sensitivity by 2.6 dB compared to Volterra equalizers.
• It showed a 1.1 dB improvement over standard convolutional neural network (CNN) equalizers and faster convergence.

Conclusions:

• The proposed DACNN equalizer effectively mitigates nonlinear distortion in probabilistic shaping fiber optic systems.
• DACNN offers superior performance and reduced training complexity compared to existing methods.
• This technique holds significant potential for advancing high-capacity optical communication systems.