Machine learning meets Singular Optics: Speckle-based Structured light demultiplexing

In this paper, the advancements in structured light beams recognition using speckle-based convolutional neural networks (CNNs) have been presented. Speckle fields, generated by the interference of multiple wavefronts diffracted and scattered through a diffuser, project a random distribution. The generated random distribution of phase and intensity correlates to the structured light beam of the corresponding speckle field. This unique distribution of phase and intensity offers an additional dimension for recognizing the encoded information in structured light. The CNNs are well-suited for harnessing this unique ability to recognize the speckle field by learning hidden patterns within data. One notable advantage of speckle-based recognition is their ability to identify structured light beams from a small portion of the speckle field, even in high-noise environments. The diffractive nature of the speckle field enables off-axis recognition, showcasing its capability in information broadcasting employing structured light beams. This is a significant departure from direct-mode detection-based models to alignment-free speckle-based detection models, which are no longer constrained by the directionality of laser beams.