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

Abstract: Atmospheric turbulence presents a major obstacle to the performance of free-space optical (FSO) communication and environmental sensing. While most studies focused on robust detection and recognition of transmitted structured light in turbulent media, a critical, yet often underexplored, avenue involves leveraging the structure of light itself to detect and characterize the medium itself. Here we introduce a deep learning framework that synergistically leverages post-transmission intensity speckle patterns and orbital angular momentum (OAM) spectral data to enhance turbulence state classification, essential for applications such as energy delivery, effective adaptive optics and communications. The proposed architecture builds upon an enhanced InceptionNet backbone optimized for multi-scale feature extraction from complex optical inputs, where we demonstrate that multiple degree of freedom analysis outperforms the conventional single parameter approaches, reaching validation accuracies in excess of 80%. Our approach is benchmarked against standard implementations and found to be superior in handling diverse turbulence scenarios, demonstrating enhanced stability across varying turbulence conditions defined by different Reynolds numbers and Fried parameters. Our method offers a scalable, data-efficient solution for precise atmospheric turbulence detection, with strong potential for deployment in real-world optical sensing applications.