High-resolution recognition of low-coherence fractional OAM modes with deep learning-based methods

Area of Science:

• Optics and Photonics
• Machine Learning
• Optical Communications

Background:

• Partially coherent fractional vortex beams carry orbital angular momentum (OAM).
• Accurate recognition of topological charge (TC) is crucial for OAM-based applications.
• Existing methods struggle with high-resolution TC detection and small OAM mode intervals.

Purpose of the Study:

• To develop a high-resolution method for recognizing the topological charge (TC) in partially coherent fractional vortex beams.
• To achieve accurate TC detection with a minimal orbital angular momentum (OAM) mode interval of 0.01.
• To leverage deep learning for enhanced TC recognition.

Main Methods:

• Utilized DenseNet-based deep learning frameworks for TC recognition.
• Analyzed the cross-spectral density (CSD) function distribution for feature extraction.
• Simulated applications in free-space optical transmission systems for image transfer.

Main Results:

• Achieved a recognition accuracy of up to 99.99% for TC detection.
• Demonstrated significant improvement over traditional intensity-based recognition methods.
• Successfully encoded information using the CSD's correlation structure in optical transmission.

Conclusions:

• The proposed deep learning approach enables high-resolution TC recognition in partially coherent fractional vortex beams.
• The method significantly enhances accuracy compared to existing techniques.
• The findings indicate potential for improving optical communication capacity and security.