Pre-monitoring-assisted deep cascaded network for nonlinear equalization in coherent optical communication systems

Area of Science:

• Optical Communications
• Machine Learning
• Signal Processing

Background:

• Machine learning (ML) offers potential for nonlinear equalization in coherent optical systems.
• Conventional ML schemes lack adaptability to dynamic link states like varying launch power and transmission distances.

Purpose of the Study:

• To propose a pre-monitoring-assisted deep cascaded network (DCN) for adaptive nonlinear equalization.
• To enhance the adaptability of ML-based equalization in optical communication systems.

Main Methods:

• A lightweight convolutional neural network (CNN) analyzes signal spectrum for link state identification.
• The CNN selects an optimal deep neural network (DNN) model for nonlinear equalization.
• The proposed DCN was demonstrated in a 28-GBaud PDM-16QAM system.

Main Results:

• The DCN achieved a Q-factor gain of 2.01 dB over linear equalization at optimal launch power for 250 km transmission.
• The scheme demonstrated a computational complexity of 2885 real multiplications per symbol (RMpS).

Conclusions:

• The pre-monitoring-assisted DCN effectively adapts to dynamic link states for nonlinear equalization.
• This approach offers improved performance and manageable computational complexity in coherent optical systems.