Polarized deep diffractive neural network for sorting, generation, multiplexing, and de-multiplexing of orbital...

Area of Science:

• Optics and Photonics
• Metamaterials
• Machine Learning

Background:

• Orbital angular momentum (OAM) beam multiplexing/de-multiplexing is crucial for optical communication.
• Optical diffractive neural networks (ODNNs) can perform OAM beam manipulation.
• Conventional ODNNs struggle with OAM modes featuring spatially varying polarization.

Purpose of the Study:

• To propose a novel polarized optical deep diffractive neural network (PODNN) capable of handling polarized OAM beams.
• To design the PODNN using dielectric rectangular micro-structure meta-materials.
• To optimize the PODNN for sorting, generation, multiplexing, and de-multiplexing of polarized OAM beams.

Main Methods:

• Design of a polarized optical deep diffractive neural network utilizing dielectric rectangular micro-structure meta-materials.
• Optimization of the network architecture for precise control over polarized OAM modes.
• Simulation-based validation of the network's performance in various OAM manipulation tasks.

Main Results:

• Successful sorting of 14 types of orthogonally polarized vortex beams.
• Effective de-multiplexing of hybrid OAM beams into Gauss beams at multiple spatial locations.
• High-quality generation of hybrid OAM beams and multiplexing of linear beams into cylinder vector beams.

Conclusions:

• The proposed PODNN framework demonstrates robust capabilities for manipulating polarized OAM beams.
• This technology offers a significant advancement for optical communication systems requiring complex OAM mode control.
• The meta-material-based design provides a scalable and efficient solution for future photonic integrated circuits.