Intelligent constellation diagram analyzer using convolutional neural network-based deep learning.

Area of Science:

• Optical communications engineering
• Artificial intelligence in signal processing
• Deep learning applications

Background:

• Constellation diagrams are crucial for visualizing and analyzing optical signal quality.
• Accurate modulation format recognition (MFR) and optical signal-to-noise ratio (OSNR) estimation are vital for optical network performance.
• Traditional methods for analyzing constellation diagrams can be complex and require manual intervention.

Purpose of the Study:

• To develop an intelligent analyzer for modulation format recognition (MFR) and optical signal-to-noise ratio (OSNR) estimation.
• To leverage Convolutional Neural Network (CNN)-based deep learning for automated constellation diagram analysis.
• To evaluate the performance of the CNN-based approach against traditional machine learning algorithms.

Main Methods:

• Utilized a CNN-based deep learning technique to process raw constellation diagram images.
• Generated constellation diagram images for six modulation formats across various OSNR ranges.
• Conducted both simulation and experimental validation of the proposed CNN model.

Main Results:

• CNN achieved superior accuracy for both MFR and OSNR estimation compared to four traditional machine learning algorithms.
• OSNR estimation reached high accuracies (95-99%) within 200 epochs, with experimental errors below 0.7 dB.
• MFR achieved 100% accuracy even with limited training data and fewer epochs.

Conclusions:

• The proposed CNN-based intelligent analyzer effectively performs MFR and OSNR estimation with high accuracy and efficiency.
• The technique demonstrates robustness against factors like training data size, image resolution, and network architecture.
• This approach holds significant potential for integration into optical test instruments and optical performance monitoring systems.