Machine-learning-assisted orbital angular momentum recognition using nanostructures.

Area of Science:

• Optics and Photonics
• Optical Communication
• Machine Learning for Optics

Background:

• Orbital angular momentum (OAM) recognition is crucial for optical communication.
• Existing OAM recognition methods require precise alignment, limiting practical applications.
• Previous speckle-based methods offer transverse alignment freedom but are limited to the far-field.

Purpose of the Study:

• To develop an alignment-free OAM recognition technique using speckle patterns generated by micro- and nanostructures.
• To extend OAM recognition capabilities to both near-field and far-field regimes.
• To demonstrate the effectiveness of machine learning in analyzing OAM speckle fields.

Main Methods:

• Investigated the dynamic evolution of OAM speckle fields from near to far field using micron-scale ground glass diffusers.
• Utilized randomly grown ZnO nanosheets on an aluminum substrate as nanostructures.
• Trained and tested a three-layer convolutional neural network (CNN) on OAM speckle fields for recognition.

Main Results:

• Speckle patterns generated by micro- and nanostructures enable OAM recognition with relaxed alignment requirements in both transverse and propagation directions.
• OAM speckle field evolution was analyzed from near-field to far-field.
• The CNN achieved over 92% accuracy in OAM recognition across all tested regimes.

Conclusions:

• Speckle-learned OAM recognition with micro- to nanostructures offers a robust and versatile solution for optical communication.
• The technique significantly reduces alignment sensitivity, enhancing practical implementation.
• This approach broadens the applicability of OAM recognition across near-field to far-field propagation distances.