Text-Only Version
Adaptive Optics, Center for Home Page CfAO Logo
Search
The CenterAdaptive OpticsResearchEducation/HRMembersCalendar of Events
PublicationsSoftwareEmploymentPicture GalleryLinksWhat's New
Research Projects
ELT Analysis
About
Project Status
5-Year Roadmap
Publications

Printable Version
 
1. Modeling and Analysis
2. Simulations
3. Fast Reconstructor Algorithms
4. Closed Loop Control





Analysis, Modeling, and Simulation of Adaptive Optics Systems for Extremely Large Telescopes

Project Status

1. Modeling and Analysis - The problem of designing adaptive optics system for extremely large telescopes can be considered one of identifying and quantifying the sources of wavefront correction error. The result of such a process forms the basis of a system error budget. Over the past year there have been a series of workshops and meetings where we have listed the known error sources and discussed their possible solutions. The collaborative research activity has made progress in modeling the effects and how design parameters and atmospheric conditions affect them. Here is a list of error sources, with links to research activity related to them:

  • Wavefront fitting error due to finite number of actuators on multiple deformable mirrors
  • Tomographic measurement error due to finite number of guidestars at limited angles and their limited brightness
  • Anisoplanatism - the error due to a finite number of DMs sampling a continuous atmosphere
  • Cone effect - the error due to the cone beams of laser guide star sampling the column of cylinder beam of the science light
  • Laser guidestar elongation, resulting in additional wavefront measurement error, caused by the projection of the laser through the finite thickness of the sodium layer
  • Bandwidth error due to limited frame rate of the wavefront sensor and computational delays.

Up to top

2. Simulations - In addition to the error budget modeling work there is a parallel effort to outright model point-design MCAO systems on 30 meter class telescopes. Brent Ellerbroek is using an existing code he developed for the Gemini South MCAO system and has generated some preliminary performance predictions.

Effort continues on the Arroyo AO modeling and simulation code. Matthew Britton gave presentations describing the current status and validation tests on the atmospheric propagation components.

Rapid turn around of ELT-sized simulations requires that the codes be capable of running on massively parallel supercomputers. Current "serial" coding requires 10^6 seconds (11 days) of compute time to simulate one second of real-time. With a 1000 parallel processors, this could be potentially reduced to 10^3 (17 minutes) of compute time. Jose Milovich has ported the Keck AO simulator and also run some 30-meter aperture test cases to show dramatic speedup on a massively parallel processing architecture. This code will be combined with Arroyo over the next few months. Aron Ahmadia has studied ways of further parallelizing MCAO simulations to take advantage of thousands of processors.

Up to top

3. Fast Reconstructor Algorithms - Significant progress has been made on algorithms for wavefront reconstruction, i.e. how one determines the commands to the multiple DMs' actuators given the measurements from the multiple guidestar wavefront sensors. The challenge is to keep these real-time computations tenable. A "standard" least-squares solution takes on the order of n^2 operations, which, for ELT sized AO systems goes beyond state of the art for present day computers. Algorithms have now been developed that have close to order n operations. These are the hierarchical decomposition methods, iterative sparse matrix methods, and FFT methods, along with hybrids (like the FFT-preconditioned-iterative-hierarchical sparse method).

Up to top

4. Closed Loop Control - Given knowledge of the statistics of the seeing and the measurement noise, it is possible to derive a controller that gives the minimum wavefront error, resulting in maximum AO-corrected Strehl. Such a statistical optimum is known as a Kramer-Rao lower bound and represents the best possible mean performance. We can quantitatively compare the performance of any other (possibly simpler to implement) controller against this baseline.

The control law that results from statistical considerations is often called a "regularized" controller. Brent Ellerbroek has developed ideas for rapid calculation of an approximate regularized closed-loop controller.

Up to top

Last Modified: Apr 14, 2003 

You are here: Research Projects > ELT Analysis | In this section: About | Project Status | 5-Year Roadmap | Publications
Center for Adaptive Optics | Search | Sitemap | The Center | Adaptive Optics | Research | Education/HR
Members | Calendar of Events | Publications | Software | Employment | Picture Gallery | Links | What's New

Design Copyright © 2002-2003 University of California Regents