Deep Learning (DL) based Computer Generated Hologram (CGH) for Beamsteering in Reconfigurable Holographic Switches

Beamsteering application via the dynamic holography with recorded dynamic holograms onto Spatial Light Modulators (SLM), has been used for optical interconnections especially in all-optical networks within the reconfigurable holographic switches. A novel AI based approach using Deep Learning (DL) methodology has been proposed for the generation of a new Computer Generated Hologram (CGH) with improved performance metrics i.e., peak signal-to-noise ratio (PSNR) and fast generation of prediction time, resulting in higher accuracy, controllable and precise predicted beamsteering application. The requirements of a high PSNR, fast generated CGH, low crosstalk, diffraction efficiency and polarization sensitivity for a Liquid Crystal (LC) 2D SLM for reconfigurable beamsteering among others necessitates the integration of the superior features of DL into CGH production i.e., proposed herewith the AI-based DL-CGH. The DL-CGH methodology proposed here has simplified input data channels, residual network and attention mechanism which performed better in comparison with Gerchberg-Saxton (G-S) and Holo-encoder CGH.