Topological charge recognition of vortex beams based on a convolutional neural network.

Area of Science:

• Optical physics and communications engineering.
• Machine learning applications in optical systems.

Background:

• Vortex beams are crucial for optical communications but are degraded by atmospheric turbulence and system aberrations like spherical and coma aberrations.
• Accurate identification of vortex beam properties is essential for maintaining communication integrity.

Purpose of the Study:

• To develop an advanced deep learning model for robust topological charge identification of vortex beams.
• To simultaneously predict topological charge, spherical aberration, and coma aberration coefficients from distorted intensity patterns.

Main Methods:

• An enhanced IResNet18 model was proposed and trained on distorted vortex beam intensity patterns.
• Experimental configurations included vortex beams with spherical aberration, coma aberration, and combined aberrations.
• Model performance was evaluated under varying propagation distances and atmospheric turbulence intensities.

Main Results:

• The enhanced IResNet18 model demonstrated superior accuracy in predicting topological charge and aberration coefficients.
• The model exhibited improved training efficiency compared to existing methods.
• Robust performance was observed across different aberration types and turbulence conditions.

Conclusions:

• The proposed model offers a reliable framework for aberration-aware vortex beam recognition.
• This research provides valuable insights for enhancing the performance and reliability of free-space optical communication systems.