High-speed PAM4 transmission with a GeSi electro-absorption modulator and dual-path neural-network-based equalization

Area of Science:

• Optical Communications
• Artificial Intelligence
• Signal Processing

Background:

• Nonlinearity mitigation is crucial for high-speed optical communication systems.
• Artificial neural networks (NNs) offer effective nonlinearity mitigation but can be computationally intensive.

Purpose of the Study:

• To propose and validate a novel dual-path neural network (DP-NN) for equalization.
• To reduce computational cost compared to conventional NN equalizers.

Main Methods:

• Developed a DP-NN combining a linear equalizer with an input-pruned NN equalizer.
• Tested the DP-NN on 4-ary pulse amplitude modulation (PAM4) transmission at 160 Gb/s (133.3 Gb/s net bitrate).
• Utilized a GeSi electro-absorption modulator operating at C-band over 2 km transmission.

Main Results:

• Achieved a bit error rate below the 1.5×10^-2 threshold for 20% hard-decision forward-error-correction.
• Demonstrated superior performance over Volterra equalization.
• Showed comparable performance to conventional NN-based equalization with reduced computation.

Conclusions:

• The DP-NN is a feasible and efficient method for nonlinearity mitigation in optical systems.
• DP-NN offers a practical approach to balancing performance and computational cost.
• This technique advances equalization strategies for next-generation optical networks.