Dielectric Metasurface Diffraction Neural Network-Based Mode, Wavelength, Polarization Division Demultiplexer

Light has three inherent degrees of freedom (DoFs), including spatial mode, wavelength and polarization state. To meet the demand for higher communication capacity in the era of big data, wavelength and polarization division multiplexing have been widely adopted over the past decades [1]. In addition to these intrinsic characteristics, mode division multiplexing (MDM) has recently attracted considerable attention [2]. Taking vortex beam carrying orbital angular momentum (OAM) as the example, its unlimited topological charge (TC) provides a promising approach to breaking the current optical communication capacity bottleneck. However, traditional OAM modes demultiplexers, based on vortex blazed grating (VGG), Daman grating or optical coordinate transformation (OCT) suffers from issues such as spatial distortion, high loss and lack of flexibility. Diffractive neural networks (DNNs), composed of cascaded diffractive layers, have garnered considerable attention for its low-loss, scalability and versatility, as they can realize the unitary transformation between arbitrary set of orthogonal modes [3]–[5]. However, when multiple multiplexing technologies are employed simultaneously, both the volume and complexity of the system increase significantly, making it challenging to be applied in practical scenarios.