An Efficient Recognition Method for Orbital Angular Momentum via Adaptive Deep ELM

Area of Science:

• Optics and Photonics
• Machine Learning
• Atmospheric Science

Background:

• Orbital angular momentum (OAM) recognition in atmospheric turbulence is challenging.
• Existing methods often require manual intervention and struggle with dynamic environments.

Purpose of the Study:

• To develop a self-adapted deep learning framework for robust OAM recognition.
• To enable automatic matching of different OAM modes without user intervention.

Main Methods:

• Utilized a derived extreme learning machine (ELM) for automatic OAM mode matching.
• Employed a multilayer ELM to quantify laser spot characteristics.
• Applied a fast iterative shrinkage-thresholding algorithm for parameter optimization and analytic expression derivation.
• Developed a relationship between laser spot features and OAM modes for stable neural network architecture.

Main Results:

• The proposed analytical machine learning model automatically matches OAM modes.
• The framework successfully extracts features from intensity distributions.
• Numerical simulations on experimental datasets show improved OAM recognition capacity.

Conclusions:

• The developed deep learning framework offers an efficient and self-adapted solution for OAM recognition in turbulent atmospheres.
• The method is suitable for time-varying environments, avoiding trial-and-error adjustments.
• The approach demonstrates superior performance in OAM recognition tasks.