Digital holography and physics-informed neural networks for mode decomposition of multimode fibers towards classical and quantum communication

Analysis of the transmission characteristics of multimode fibers (MMF) plays an increasing role in communications, biomedicine, metrology, endo-microscopy, fiber sensors, displays, and nondestructive testing. With prior knowledge of the spatial modes of MMF, mode decomposition (MD) is an effective technique that can analyze the light field at the output end of the fiber in the mode domain to obtain complete mode weights from the speckle patterns. Digital holography as standard approach retrieves both the amplitude and phase of coherent multimode beams. However, the necessary reference beam represents a major challenge, especially for long fibers, since phase noise in fibers increases with length. A number of reference-less methods have been proposed, including the Gerchberg-Saxton optimization. We propose the merging of physical models with AI, a pre-training-based physics-informed neural network for MD of MMF. Paradigm shifts for secure data transmission by classical and quantum communication with MMF are promising as well as advancements are possible for lensless MMF endoscopy, photonics chips using MMF links, neuromorphic computing, quantum computing, and reservoir computing.