Classification of Single Photons in Higher-Order Spatial Modes via Convolutional Neural Networks

TL;DR

This paper demonstrates that convolutional neural networks can accurately classify single-photon spatial modes, even with atmospheric turbulence, enhancing quantum optical communication by leveraging intensity-based analysis.

Contribution

It introduces a CNN-based method for classifying single-photon spatial modes, including a denoising autoencoder to correct turbulence effects, achieving high accuracy in a practical setting.

Findings

99.2% overall classification accuracy

Hermite-Gauss modes had the highest individual accuracy

CNN relies solely on intensity, enabling efficient single-photon mode classification

Abstract

Spatial modes are a promising candidate for encoding information for classical and quantum optical communication due to their potential high information capacity. Unfortunately, compensation of the wavefront upon propagation through the atmosphere is necessary to benefit from advantages spatial modes offer. In this work, we leverage the success of convolutional networks in denoising and classifying images to improve information transfer of spatial modes. Hermite-Gauss, Laguerre-Gauss, and Ince-Gauss modes are experimentally generated using single photons and imaged. A denoising autoencoder corrects for turbulence effects on the wavefront, followed by a convolutional neural network to classify mode orders. The model achieves a 99.2% classification accuracy across all modes, and Hermite-Gauss modes exhibited the highest individual mode accuracy. As the convolutional networks rely solely on intensity, they offer an efficient and cost-effective tool for optical communication systems in the single photon limit.