Abstract

Laser induced filamentation is a mature field of study with multiple applications [1], while Machine Learning has currently a strong presence in the field of Photonics [2,3]. Combining techniques from both fields has led to the demonstration of reconstructing images from speckle patterns after the light propagation inside multimode fibers, optical waveguides, scattering, and turbulent media using neural networks [4]. Here we demonstrate that Machine Learning can also be used to even more complex and chaotic systems (both in space and time), like in the nonlinear propagation of intense ultrafast laser beams (filamentation). A mixture of nonlinear optical phenomena (Kerr self-focusing and plasma defocusing, space-time coupling, etc.) combined with nonlinear absorption induced thermal turbulence in gas and liquid media, leads to a chaotic distortion of the beam profile in the form of speckle patterns. These random intensity and phase profiles make impossible the transfer of information.

© 2023 IEEE