Enhancing Classification Accuracy of Intensity Degenerate Orbital Angular Momentum modes through Astigmatism Induced by Tilted Spherical Lens

Abstract

The remarkable abilities and properties of orbital angular momentum (OAM) beams have gained significant attention within the realm of optical communication [1]. However, fully utilizing the entire OAM mode spectrum remains challenging. A groundbreaking solution is found in AI-based OAM mode classification, using a speckle-learned [2-3] de-multiplexing method to address beam wandering and alignment issues. This paper introduces a tilted spherical convex lens to induce astigmatism [4] in far-field speckle patterns. The astigmatic transformed speckle patterns are simulated using the Fresnel diffraction integral. Classification of the astigmatic transformed speckle patterns employs a trained Convolutional Neural Network (CNN) achieving an impressive 99.25% accuracy at β=45 deg. Classification of intensity degenerate far-field speckles of LG modes by introducing astigmatism using a tilted spherical lens offers several advantages over the method discussed in [3] where a cylindrical lens is used for introducing astigmatism. Firstly, the tilted spherical lens provides greater control over accuracy and astigmatism. By varying the tilt angle, we observe a direct correlation between accuracy and the angle of tilt. This allows for precise adjustments to optimize classification performance. Secondly, the method employing a tilted spherical lens offers enhanced dynamic tunability. The ability to dynamically control astigmatism allows for real-time adjustments to adapt to different experimental conditions or specific classification requirements. This ensures reproducibility and ease of application in various optical systems.The remarkable abilities and properties of orbital angular momentum (OAM) beams have gained significant attention within the realm of optical communication [1]. However, fully utilizing the entire OAM mode spectrum remains challenging. A groundbreaking solution is found in AI-based OAM mode classification, using a speckle-learned [2-3] de-multiplexing method to address beam wandering and alignment issues. This paper introduces a tilted spherical convex lens to induce astigmatism [4] in far-field speckle patterns. The astigmatic transformed speckle patterns are simulated using the Fresnel diffraction integral. Classification of the astigmatic transformed speckle patterns employs a trained Convolutional Neural Network (CNN) achieving an impressive 99.25% accuracy at β=45 deg. Classification of intensity degenerate far-field speckles of LG modes by introducing astigmatism using a tilted spherical lens offers several advantages over the method discussed in [3] where a cylindrical lens is used for introducing astigmatism. Firstly, the tilted spherical lens provides greater control over accuracy and astigmatism. By varying the tilt angle, we observe a direct correlation between accuracy and the angle of tilt. This allows for precise adjustments to optimize classification performance. Secondly, the method employing a tilted spherical lens offers enhanced dynamic tunability. The ability to dynamically control astigmatism allows for real-time adjustments to adapt to different experimental conditions or specific classification requirements. This ensures reproducibility and ease of application in various optical systems.