Fourier-Based Predictive AO Schemes for Tomographic AO systems

AO4ELT3 May 31, 2013

Fourier-Based Predictive AO Schemes for Tomographic AO systems

S. Mark AmmonsLawrence Livermore National Laboratory

May 31, 2013

Thanks to:Lisa Poyneer

Alex RudyDon GavelClaire Max

Luke JohnsonBenoit Neichel

Andrés GuesalagaAngela Cortes


1. A final note on GEMS astrometry – the on detector guide window

2. The mechanics of Fourier predictive control

3. Implementing multi-layer predictive control on Shane AO

4. Extending predictive control to tomographic systems: Simulations

5. Application to GEMS telemetry

Outline


1-2 mas Astrometry on Bright Stars (V > 4) with the GEMS On-Detector

Guide Window

GEMS On-Detector Guide Windows

GSAOI on-detector guide windows can be read out quickly and saved to prevent target star saturation

1-2 mas stability on central star position (over short term) for V > 4 stars

Performance limited by flat-fielding and detector artifacts on ODGW


We will Implement Predictive Control on the Upgraded Shane AO

System Project funded by the University of California Office of the President

Goals: Develop 4-dimensional LQG tomographic predictive control for MCAO

and MOAO systems Implement single-conjugate predictive control on the upgraded Shane

LGS AO system at Lick Observatory

Optimal control schemes use a model:

• We need a model of both the control system and theoperating conditions (e.g. noise level and atmosphere)

• Many methods assign parametersMVU/Keck Bayesian - assign star brightness and r0 estimate

• For best performance, we need to be “data-driven” and useactual measurements to populate our model

• Getting wind velocity a little wrong is OK• Getting wind direction a lot wrong is very bad




ShaneAO: Adaptive Optics System at the Shane 3-meter Telescope

(LGS mode, new fiber laser)

ShaneAO is a diffraction-limited imager, spectrograph, and polarimeter for the visible and near-infrared science bands.

7

0 2 4 6 8 10 12 140.0

0.2

0.4

0.6

0.8

1.0

Field Angle , arcsec

Strehl

Z

J

H

K

Band

0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0

0.02

0.05

0.10

0.20

0.50

1.00

Wavelength , microns

FWHM,arcseconds

Pixel

DiffLim

Seeing


Shane AO will Deliver Improved Strehl over Existing Lick System

8

Airy core forming

LGS mode, new fiber laser


ShaneAO instrument characteristics

9

Detector sampling 0.035 arcsec/pixelField of view 20 arcsec squareScience detector: Hawaii2RG Hawaii2RGScience wavelength coverage: 0.7 to 2.2 microns 0.7 to 2.2 micronsSpectral resolution R = 500Slit width: 0.1 arcseconds 0.1 arcsecSlit decker: 10 arcseconds (?) 10 arcsecSlit angle on sky adjustable 0-360°Long-exposure stability hold to the diffraction-limit for one hour

hold to ½ slit width for 4 hoursPolarimitry mode: polarization analyzer and variable angle waveplateDelta magnitude within seeing disk DmK=10Minimum brightness tip/tilt star: mV=18Tip/tilt star selection field 120 arcsecSky coverage ~90% LGS modeMinimum brightness natural guide star mV=13Camera readout modes Correlated double-sampling (CDS)

up the ramp (UTR)sub-frame region of interest (ROI)quick take

Exposure support: Multiple frame co-addedautomated nod and expose coordinated with telescope (snap-i-diff, box-4, box-5)automated darks sequence based of history of science exposures

Observations support automatic data loggingautomatic data archiving


ShaneAO is being Asssembled with First Light in Fall 2013

10

Deformable Mirror

Wavefront Sensor Science Detector


1. A final note on GEMS astrometry – the on detector guide window

2. The mechanics of Fourier predictive control

3. Implementing multi-layer predictive control on Shane AO

4. Extending predictive control to tomographic systems: Simulations

5. Application to GEMS telemetry

Outline


Simulation Design

LGSLGS

LGS

Three-Layer Atmosphere We simulate an 8-m MOAO system with tomography:

3 LGSs over 120” diameter

3-layer Taylor frozen-flow atmospheric model assumed at 0, 5, 10 km (55%, 30%, 15%)

Wind velocities randomized, 0-15 m/s

200 realizations of 1 second length, 1 kHz operation

To isolate the effect of prediction on tomographic error, we add no WFS noise


Back-Projection Tomography

• 25 iterations per time step, alternating pre-conditioned conjugate gradient / linear steps

Minimum Variance Back-Projection Tomography (Gavel 2004)


Correction Scheme – Shift & Average Multi-sampled Voxels

• Wind Identification and Estimation not simulated

• For each layer, replace voxels in downwind direction with shifted and averaged voxels from tomographic time history

• Only shift voxels originating in multi-sampled region, where height can be effectively determined

• Wind vectors assumed to be known perfectly

Multi-sampled regions

Phase height cannot be constrained in sparsely-sampled regions from

tomography alone!


Prediction Improves RMS Errors on Layer Estimates

• After 1 second, on average, the layer estimates improve 3-13%

• Downwind regions improve 10-30%, especially for high altitude layers

Fractional Improvement in Layer Estimates

Ground layer

5 km layer

10 km layer


Prediction Improves RMS Errors on Layer Estimates

• With downwind layers better determined, the tomographic error improves beyond the radius of the guide stars

Tomographic Error vs. Field Radius

Maps of Improvement in Layer Estimates

LGS radius

With shift & average

Without shift & average


Application to GEMS Telemetry

• Method #1: Apply wind identification step to pseudo open loop slopes provided by A. Guesalaga and B. Neichel

• Method #2: 1. Perform tomographic reconstruction on pseudo open-

loop slopes to estimate true volumetric phase2. Apply wind identification step to layers of different

atmospheric heights

We expect that wind layers separated by height will have different temporal properties.


Tomographic Reconstruction + Wind Identification Cleanly

Separates Layers

Pseudo Open-Loop slopes Tomographically-reconstructed ground layer

Tomographically-reconstructed 4.5 km layer

The temporal properties of layers are cleanly distinguished by a tomographic reconstruction (using only geometric information!)


From Identification to Correction

For a tomographic system:

Apply a prior that wind vectors in layers separated by geometric tomography should be uncorrelated.

If a wind peak is seen with a certain vector that corresponds to a stronger detection in another layer, reject those frequencies.

For faster tomography for on-line use: Use Guesalaga & Cortes autocorrelation technique


Summary

1. We will Implement Predictive Control on the Upgraded Shane AO System

2. In simulation, shifting and averaging predictive control provides 3-13% benefits in tomographic wavefront estimation quality at all layers

3. From GEMS telemetry: Adding a tomographic reconstruction to pseudo-open loop slopes, using geometric information only, cleanly separates the temporal properties of the wind flow.

4. This can be exploited to design filters that reject multiple common detections across layers, which are likely to result from tomographic reconstruction error.

Documents

Fourier-Based Predictive AO Schemes for Tomographic AO systems