Machine learning assisted speckle and OAM spectrum analysis for enhanced turbulence characterisation

TL;DR

This paper presents a deep learning framework that combines speckle patterns and OAM spectral data to accurately characterize atmospheric turbulence, improving over traditional single-modality methods for environmental sensing and FSO system optimization.

Contribution

A novel multimodal deep learning approach integrating speckle and OAM spectral data for turbulence parameter inference, outperforming single-modality techniques.

Findings

Validation accuracy exceeds 80%

High inference accuracy across various turbulent conditions

Enhanced training stability and data efficiency

Abstract

Atmospheric turbulence degrades the performance of free-space optical (FSO) communication and remote sensing systems by introducing phase and intensity distortions. While a majority of research focuses on mitigating these effects to ensure robust signal transmission, an underexplored alternative is to leverage the transformation of structured light to characterize the turbulent medium itself. Here, we introduce a deep learning framework that fuses post-propagation intensity speckle patterns and orbital angular momentum (OAM) spectral data for atmospheric turbulence parameter inference. Our architecture, based on a modified InceptionNet backbone, is optimized to extract and integrate multi-scale features from these distinct optical modalities. This multimodal approach achieves validation accuracies exceeding 80%, substantially outperforming conventional single-modality baselines. The framework demonstrates high inference accuracy and enhanced training stability across a broad range of simulated turbulent conditions, quantified by varying Fried parameters (r0) and Reynolds numbers (Re). This work presents a scalable and data-efficient method for turbulence characterization, offering a pathway toward robust environmental sensing and the optimization of dynamic FSO systems.