Transmission in Multimode fiber with a deep learning network

Spatio-temporal control of femtosecond pulse-delivery through multimode fibers (MMF) can be used to achieve two photon photo-polymerization. Transmission-matrix method is used in linear domain to solve the scrambling effect at... [ view full abstract ]

Spatio-temporal control of femtosecond pulse-delivery through multimode fibers (MMF) can be used to achieve two photon photo-polymerization. Transmission-matrix method is used in linear domain to solve the scrambling effect at the fiber output and generate focused spots. However, multimode fibers suffer from non-linear effects at high peak intensities. The method is not effective to describe light propagation in the non-linear regime. Here we propose a deep learning network which can learn the relationship between inputs and outputs of the MMF. We show that once the network is properly trained, it can directly calculate the inverse of the transmission matrix.