Learning Spatiotemporal Nonlinearities in Graded-Index Multimode Fibers with Deep Neural Networks

Abstract

In the past few years, graded-index multimode fibers (GIMFs) have become subject to the extensive study of nonlinear optics by featuring novel spatiotemporal interactions. Ultrashort pulse propagation with complex modal interactions revealed new nonlinear dynamics and frequency conversion techniques such as cascaded Raman scattering [1] and supercontinuum generation by exploiting multiple nonlinear effects simultaneously [2,3]. So far, researchers mainly studied the effects of pulse duration and dispersion in GIMFs. Very recently, optimization of spatiotemporal nonlinear effects with initial modal excitation condition is investigated by wavefront shaping and iterative search algorithms [4]. Here, we present the first machine learning approach to investigate the spatiotemporal nonlinear interactions in GIMFs by learning and controlling supercontinuum generation.