Low-complexity characterized-long-short-term-memory-aided channel modeling for optical fiber communications

Area of Science:

• Optical Communications Engineering
• Machine Learning Applications
• Signal Processing

Background:

• Optical single-mode fiber (SMF) communications face challenges with intersymbol interference (ISI) due to dispersion.
• Traditional channel modeling methods like the split-step Fourier method (SSFM) can be computationally intensive.
• Long short-term memory (LSTM) networks offer sequence correlation learning but can suffer from gradient explosion problems.

Purpose of the Study:

• To propose a low-complexity, data-driven channel modeling technique for SMF communications.
• To enhance traditional LSTM networks by incorporating feature information to better characterize dispersion-induced ISI.
• To improve modeling accuracy and computational efficiency compared to existing methods.

Main Methods:

• Development of a characterized-long-short-term-memory (C-LSTM) model.
• Integration of feature information into the LSTM input layer to capture sequence correlations.
• Comparative analysis with SSFM and conditional generative adversarial networks (CGAN) based on mean square error (MSE) and computational complexity.

Main Results:

• The proposed C-LSTM effectively alleviates the gradient explosion problem.
• C-LSTM achieves a stable and lower mean square error (MSE) compared to traditional LSTM.
• C-LSTM demonstrates superior computational complexity over SSFM and CGAN.
• The C-LSTM-aided technique exhibits higher modeling accuracy than traditional LSTM.

Conclusions:

• The C-LSTM model offers a computationally efficient and accurate solution for SMF channel modeling.
• The technique's ability to handle sequence correlations and feature selection makes it highly effective.
• The C-LSTM approach is adaptable for other channel modeling applications with significant sequence correlations.