Fast single pixel modal wavefront sensing using neural networks

Abstract

Dynamic wavefront aberrations negatively impact a wide range of optical applications including astronomy, optical free-space telecommunications and bio-imaging. Wavefront errors can be compensated by an adaptive optics system comprised of a deformable mirror and wavefront sensor connected by a control loop. For satellite optical communications (SatCom), wavefront sensing is particularly challenging due to the rapid wavefront fluctuations induced by strong turbulence and movement of the transmitting satellite across the sky. Existing wavefront sensing techniques require fast cameras (>kHz) that are not widely available at wavelengths suitable for SatCom (e.g., 1550nm and mid-to-long wave infrared). Here, we propose a new wavefront sensing technique that uses a single photodiode and a fast mirror to make phase-diverse intensity measurements of the incoming wavefront. We train neural networks to accurately estimate the input phase given this phase-diverse sub-millisecond intensity trace. Our simulations show that our technique is robust in cases of strong turbulence where previous modal wavefront sensors fail due to modal crosstalk, achieving 99% of the optimal Strehl ratio from a 50-mode correction at a sensing rate of 2kHz. We explore typical cases of turbulence magnitude, sensing speed and noise that might be encountered by such a system.