IISER MOHALI

Abstract: Structured light beams, particularly those carrying orbital angular momentum (OAM), offer a powerful platform for high-dimensional optical communication. Yet, traditional demultiplexing methods often reliant on interference or diffraction struggle with alignment sensitivity, limited scalability, and inefficacy in handling intensity-degenerate modes.

This presentation introduces a machine learning driven approach to structured light demultiplexing based on speckle pattern recognition. By harnessing the random scattering of light through diffusers and analysing the resulting speckle patterns with convolutional neural networks, we achieve robust, single-shot classification of OAM modes. This framework operates efficiently across detector types: 2D cameras, 1D line arrays, and even 0D single-pixel detectors, enabling compact, low-cost, and scalable implementations.

We further demonstrate spatial-to-temporal mapping techniques, converting complex spatial modal information into temporal signals for high-speed processing with reduced data overhead. Crucially, the approach extends seamlessly into the quantum domain, allowing for the classification of entangled OAM photon pairs.

The result is a versatile, alignment-tolerant, and reference-free demultiplexing architecture that works from classical to quantum domain. It enables high-throughput, non-line-of-sight, and fiber-compatible optical systems with minimal computational demand advancing the frontiers of singular optics and intelligent photonic signal processing.

We look forward to your participation in what promises to be an engaging and insightful session.