Control of a nonlinear optical phase modulator via a neural network.

Stephane Barland

Controlling the spatial profile of an light beam is of crucial importance for many tasks such as adaptive optics or ultrashort light pulse generation. Many approaches exist and all basically consist in delaying locally the light beam propagation in order to obtain the desired phase landscape. Our approach is based on a liquid crystal element whose local refractive index depends (nonlinearly) on the local temperature. Thus, to achieve spatial control of an incoming beam, we need to locally heat the liquid crystal. This is obtained by applying another light beam (the “control beam”) with a spatially modulated intensity profile, which will heat the gold plate on which the liquid crystal is attached. In the forward direction, the physics is reasonably simple: a locally high intensity of the control beam heats locally the gold plate (with some diffusion), which transfers heat to the liquid crystal (again with some diffusion), which delays locally (of some amount) the phase of the beam to be controlled…

Although in principle manageable, the system appears to be very hard to control since the question we need to answer is “what control beam profile do I need to achieve the desired phase profile?”. To answer this question, we design and train a neural network with the available experimental data and assess the performance of this approach in the context of optical aberrations correction.

I am research director at Institut de Physique de Nice and am interested in nonlinear photonics, complex systems and (photonic or computer) neural networks.