Direct decoding of nonlinear OFDM-QAM signals using convolutional neural network.

Area of Science:

• Optical Communications Engineering
• Signal Processing
• Machine Learning Applications

Background:

• Fiber optic communication systems face capacity limitations.
• Nonlinear Fourier transform (NFT) offers a potential solution but suffers from speed and accuracy bottlenecks.
• Machine learning, particularly convolutional neural networks (CNNs), shows promise for enhancing NFT applications.

Purpose of the Study:

• To develop and evaluate a CNN for decoding information in NFT-based optical communication systems.
• To compare the performance of the developed CNN against a fast NFT algorithm.
• To assess the potential of CNNs to replace traditional NFT calculations.

Main Methods:

• Development of a convolutional neural network (CNN) model.
• Numerical simulation and demonstration of the CNN's performance.
• Comparative analysis against a fast nonlinear Fourier transform (NFT) algorithm.

Main Results:

• The developed CNN was numerically demonstrated for information decoding in NFT-based systems.
• Performance comparison indicated the CNN's viability.
• The study highlights the potential for CNNs to substitute complex NFT computations.

Conclusions:

• Convolutional neural networks show significant potential for overcoming practical limitations in NFT.
• CNNs can effectively decode information in NFT-based optical communication systems.
• This machine learning approach offers a promising alternative to conventional NFT algorithms for future high-capacity communication.