Adaptive Optics Model Anita Enmark Lund Observatory Outline
Adaptive Optics Model Anita Enmark Lund Observatory
Outline • Adaptive Optics Model • Current Work Generalization Model Improvment Parallelization
Euro 50 Model Overview SCAO on a NGS DM 3169 Actuators
Adaptive Optics Model Test Vehicle for AO Design together with ELT • SCAO / NGS Atmosphere Deformable mirror • • • Geometrical or Fresnel Propagation SVD Various influence functions and geometries DM 1: 2 nd order systems for every actuator Delay/non-linearities from WFS Geometrical or physical model of SH-WFS Wavefront sensor Controller Reconstruction
Current Work • Generalization • Improved adaptive optics model • Parallelization
Generalization and improvment • Adaptive optics model tested for VTT • Improved model (atmosphere, noise etc)
Movie of VTT model with modes 1, 2 (tip/tilt) and 30 -35 set to zero
Parallelization Simulation environment for first order model Beowulf cluster with Matlab+Matlab. WS Memory capacity limiting factor Full matrix with DM influence function 6 GB for 3169 actuators 64 bit Matlab needed– more primary memory Parts of code too slow Simple first order model takes many days/sec Network too slow Currently evaluating other simulation environments Cluster – LUNARC, Lund Shared memory - Galway, Ireland
ODE Multirate Solvers for Systems with Mixed Dynamics 1 million State Variables Slow Subsystem Fast Subsystem • Modern work on multi-rate solvers (eg. Anne Kværnø et al) • Andrus: Mixed 4 th order Runge-Kutta • Simpler multirate solvers with extra/interpolation • Execution time reduction 5 -10 x
Integated Model Bottle Neck Atmosphere Deformable mirror Wavefront sensor Controller Reconstruction
Shack-Hartman Wavefont Sensor Model Main drivers for execution time: • One image for each subaperture -> For every subaperture: exponential and 2 D IFFT • Must give FOV as close to nominal as possible -> interpolation of wavefront
Image formation Every subaperture is propagated to the image plane with Fraunhofer propagation. is then The image of the subaperture wavefront in angular coordinates 2 where u(x, y) is the complex amplitude of the wavefront in is the wavefront phase. the spatial coordinate system (x, y) and denotes the inverse Fourier transform. => exponential and IFFT
Field of View A large local tilt (for example wind residuals) in the wavefront gives a subaperture PSF far away from the center. A large FOV is needed. Fourier transforms: Higher resolution in one domain gives larger format in the other domain • Dense sampling in the spatial domain gives a large band width • Dense sampling of the wavefront gives a large FOV The characteristics of the atmosphere gives the number of samples for the wavefront, but in order to simulate a given FOV for the SH-WFS a denser grid can be necessary => interpolation
Impact on AO-model Good News: Both have outer loops with an independent variable=> Suitable for parallelization. Bad News: Matlabs parallel computing tools not good for our needs. Matlab. WS not ported to shared memory machine. New C-code needed. More good news: The bottle necks are within the same Matlab subfunction, only a limited part of the model must be coded in C. This gives fast execution, but keeps good structure. Model groves with more sensors – can be parallelized
Status for C-code development in cooperation with Michael Browne National University of Ireland, Galway Tilted wavefront CCD detector 15 x 15 subapertures Matlab-function result CCD detector 4 x 4 pixels/subaperture C mex-function result Matlab-function result C mex-function result
Test results for one call to WFS Matlab on one CPU machine Sequential C-code on the same machine Sequential C-code on Itanium machine Expected for multi-processor Itanium ~ 100 sec ~ 14 sec ~ 0. 5 sec
Integrated Modeling Conclusions • • A full Euro 50 model with AO in place Generalization work in progress AO model improved Parallelization in progress – faster network and more memory needed – under test • First tests promising
Full first order model simulation Atmosphere correction
- Slides: 19