Homepage » Program » Contributed abstracts » Real-time control » Oral talks » Adapting optimal LQG methods to ELT-sized AO (...)
Adapting optimal LQG methods to ELT-sized AO systems

Abstract 159

Submitted by Carlos CORREIA


C. Correia(1), J.-M. Conan(1), C. Kulcsar(2), H.-F. Raynaud(2), C. Petit(1)


(1) ONERA; (2) L2TI - Univ. Paris XIII.


The optimal Strehl-delivering formulation of reconstruction and control in Adaptive Optics (AO) has been known since a few years to correspond to a discrete-time Linear Quadratic Gaussian (LQG) formulation. In the absence of deformable mirror (DM) dynamics, it combines a Kalman filter for the estimation of the turbulent phase from noisy measurements with a projection step onto the DM space. This approach has proven its potential at delivering superior performance in terms of noise rejection, vibration damping and phase reconstruction in wide-field AO. However, LQG AO control, because it involves an explicit real-time reconstruction of the turbulent phase, is more computationally demanding than the standard gain plus integrator control. With the advent of the Extremely Large Telescopes (ELT) the number of degrees-of-freedom (DoF) is set to increase enormously. Based on the assumptions and computational methods employed so far, the real-time cost of LQG scales as O(DoF^2) multiplications per sample period. This clearly presents a serious obstacle for ELT-sized systems. In this contribution several strategies to render the LQG methods implementable in their new working scenarios in both processing time and memory requirements are outlined. They involve revisiting the bases used to describe the several parameter spaces, namely atmospheric phase, wave-front sensor measurements and deformable-mirror commands together with improved off-line computation of the Kalman filter gain. The potential of these new approaches is evaluated through Monte-Carlo end-to-end simulations for both the single-conjugate and multi-conjugate cases.

Documents to download