Predictive Control and Low-Latency Mirror Interface for Free-Space Optical Communications

Abstract

Free-Space Optical Communications (FSOC) have recently emerged as a high data rate alternative to traditional radio-frequency links for downlink communication between Low Earth Orbit (LEO) satellites and ground stations. This technology relies on the injection of the transmitted beam into an optical fiber. However, this injection is severely limited by the effect of atmospheric turbulence, motivating the use of Adaptive Optics (AO). In the LEO tracking scenario, the apparent satellite motion produces a very high effective wind speed, which makes the temporal error one of the dominant, and often the dominant, contribution to the AO error budget. This thesis addresses this limitation through two complementary axes, within the AO system de- veloped by AMOS for the SALTO telescope at the ESEC site in Redu. The first, incremental axis reduces the communication latency of the deformable mirror (DM) by replacing the proprietary communication protocol with a custom User Datagram Protocol (UDP) interface. The second, disruptive axis replaces the conventional integrator with two families of predictive controllers: a model-based Linear Quadratic Gaussian controller extended to a VAR(2) model, and a purely data-driven controller based on the Empirical Orthogonal Functions (EOF) formalism. Each is further declined into a self-tuning variant (a periodically refitted LQG VAR(2) and a Recursive EOF based on a Recursive Least Squares update with a forgetting factor) to track the evolution of the atmospheric conditions encountered along a satellite pass. The controllers are evaluated through end-to-end closed-loop simulations in OOPAO, including a Hufnagel–Valley C2n profile, a Bufton wind model, the elevation dependent geometry of a LEO pass, and a split-step Fresnel propagation accounting for scintillation. At high elevation, both predictive controllers improve the mean fiber coupling efficiency by 30 to 47 % over the integrator and reduce the standard deviation of the instantaneous coupling loss by up to a factor of four, the largest gains being obtained under the strongest turbulence (r0 = 2 cm at zenith). Along a simulated satellite pass, the Recursive EOF brings a systematic gain of 0.23 to 0.32 dB on the pass-averaged coupling loss with respect to its batch counterpart, while the self-tuning LQG VAR(2) remains close to its batch version over the 40 s simulated window. The UDP interface is experimentally validated on the AO bench: it yields residual wavefront errors statistically indistinguishable from those of the proprietary communication protocol, demonstrating its functional equivalence and indicating that the host-to-mirror communication is not the dominant contributor to the total loop delay. Together, these contributions provide both the control strategies and the software infrastructure required to deploy predictive AO control within the MOCA architecture, with the on-sky validation on SALTO identified as the next important milestone.