|
This is an attempt to list the issues that could be significantly limiting
Keck AO performance on bright stars. It's based on the current level of
knowledge at LLNL, and may be obsolete in some areas - ie, there may be
suggestions here that Keck staff have studied and know much more about
than I do; any comments on such things would be welcome. It also
emphasizes (as our CfAO project currently does) bright-star performance
rather than issues related to improving the limiting magnitude or
dim-star performance. I've listed some explanations that are
unlikely (controller behaviour, tip/tilt) but that might be worth
one last experiment to verify; and some that are more fundamental
(edge effects, quad-cell spot-size effects.)
- The Keck AO controller is not rejecting atmospheric disturbences
sufficiently well
ie, due to a flaw in the control software or the DM, it is
failing to keep up with atmospheric turbulence.
This hypothesis was pretty much refuted by Erik Johansson's study of the
controller performance (Johansson et al 2000 Prof. SPIE Vol 4007) and by
the error-budget analysis we did after that; the controller is behaving
like a controller, with a good amount of rejection and decent
bandwidth. One uncertainty in these measurements was the size of the spots
across the wavefront sensor (which is needed to convert WFS measurements
into physical units), although several independent techniques gave similar
answers, we usually only looked at the average size rather than individual
sizes.
actions :
- verify the error budget/controller behaviour of both AO systems
- remeasure WFS spot sizes on both AO systems, preferably on a spot-by-spot
basis and using the science camera as a verification tool.
- Residual tip/tilt motion
The same error budget analysis showed very little residual tip/tilt as
measured at the wavefront sensor. It is marginally possible that there
is some vibration making tip/tilt on the science leg. Some people have
suggested there is a slow drift in image position even with the loop closed
(in addition to differential refraction); David was measuring this last time
I was at Keck
actions :
- user NIRC2 rapid readout/subarrays to measure residual fast tip/tilt
in the science plane
- look at David's recent measurements of image position stability
- Residual vibration (tip/tilt or higher order)
Our analysis and Scott Actons showed that while vibration can be high
(sometimes as much as 100 nm) it didn't dominate even 2 years ago.
actions :
- quantify vibration as part of a study of
the error budget/controller behaviour of both AO systems
- Primary mirror edge effects
The AO system is calibrated without a pupil mask in place that reproduces
the sharp edges of the real Keck primary. This could result in a displacement
of the reference centroids from the true location if there are imperfections
in the WFS optics - Scott Acton used to worry about this a fair amount.
Don Gavel studied this a little early in the design stage. Naively, this
should only mess up the edge subapertures, but it is possible for effects
to propogate inwards through the reconstructor. One interesting experiment
would be to use NIRC2's inscribed-circle pupil and see what the Strehl
ratio is; this could be done with a modified control matrix that zeroes
out the edge subapertures. Another experiment is to calibrate with a pupil
mask in place over the DM, or even with the telescope simulator.
Finally, NIRC2's Lyot-viewing mode can provide some diagnostics.
This is Don Gavel's favourite suspect, and Scott did worry about it too.
We have someone at LLNL working on a simulation of this.
actions :
- dust off Don's old reports
- verify optical quality of WFS optics
- NIRC2 inscribed-pupil tests
- calibration and image quality tests with pupil mask in AO system
- Internal calibration and static errors
Calculations of internal Strehl ratio are a little uncertain due to the
finite size of the reference fibre source and worries about the uniformity
of its illumination; some odd effects have been observed e.g. changing to/from
narrowband filters on SCAM. Probably this is a small effect but it should
be explored. Variations in internal Strehl ratio with different pupil
sizes should also be studied. I don't fully understand the current
calibration procedure, so it would be worth talking through it at some
point.
- Spot size/quad cell/non-common-path interaction
(This is my bet for what the main problem is.)
The response of a quad cell centroider to a given wavefront distortion
depends on the size of the spot in each subaperture; 0.5 waves of astigmatism
will produce twice as large a reading in pixel units with a 0.5" spot as
with a 1" spot. If the AO system is controlling to a flat wavefront
this isn't an issue. However, due to non-common-path errors, the goal
for each spot will be displaced relative to the quad cell centers;
and the size of the spots on the sky will be different than the size of the
reference spot used to calibrate. as a result, the AO system controls to
slightly the wrong static shape. This effect can be seen clearlyw ith big
reference objects (e.g. Uranus, which sometimes produces a double-humped
PSF.) It could be signficant even for stars - supposedly Scott once
calculated that if the WFS reference spots were diffraction-limited (0.3")
compared to the seeing-limtied spots on the sky (0.8") the error would
be enormous. What saves us is that the WFS spots aren't diffraction limited
due to the quality of the WFS optics (especially lenslets, I believe.)
We have measured that the WFS spots both on the reference and the sky are
~0.8", although the sky measurements are much less certain. This could
still be a major effect and will be much more severe for the laser.
actions :
- evaulate magnitude of non-common-path errors from the reference centroid
files
- model the effects of this for different spot sizes
- measure WFS spot sizes on calibration source and on the sky very carefully
on a spot-by-spot basis
- experiment with adjusting/rescaling reference centroids
- experiment with behaviour when all reference centroids are set to zero
(this will degrade the internal calibration image, but removes the
uncertainty about the spot size effect, so the performance on the sky
becomes easier to model)
- CCD background subtraction effects
Peter and David once showed me an odd behaviour of the wavefront sensor:
when one sets up reference centroids on the light source at high intensity,
and then turns down the light source and closes the loop, the DM moves
into a shape with 4 clear columns aligned along the CCD output tap.
This is very odd; if you look at how the centroider and controller are
supposed to work, this shouldn't happen (gain differences between the 4
outputs should cancel out between the brigth and dim centroid measurements,
and bias/dark current differences should be removed by the background
subtraction.) It could indicate that the bias/dark current isn't very
stable; or that there's a bug in the centroider; or that the behaviour
is related to the thresholds in the centroider; or to the modifications
Chris Shelton made to the centroider. This probably isn't affecting bright-star
behaviour but might affect dim behaviour - there should be no need for
different dim and bright-star centroids, for example. This ought to be easy
to debug during daytime by taking WFS image data and centroid data and
studying the changes as the source gets brighter and dimmer.
- Primary mirror phase quality:
The original error budget assumes a relatively good primary mirror phase,
but I've been on AO runs when it has been 3 or more months since
the telescope has been phased, and even stackign doesn't take place on AO
nights. Original simulations Don worked on show
that the Hartmann sensor doesn't have problems with phase discontinuities,
but we should evaluate the magnitude of this effect for realistic levels of
primary mirror error; possibly experiment with using old and new snapshot
files, and the edge-minimizing phase algorithm Gary Chanan devised, to see
if it has any noticeable effect on strehl, PSF shape, and Lyot mode.
General experiments:
- NIRC2 Lyot mode:
using NIRC2's pupil-viewing optics together with
different sizes of focal plane spot provides a way to measure the
locations on the primary with the greatest phase errors at different
spatial scales. I've tried this with a large focal plane stop to see
the causes of the radial spikes in the deep PSF (pretty much confirming
the hypothesis that it is the center-of-mirror dimples.) This could be
extended with smaller focal plane stops - the focal plane stop acts like a
low-pass filter in the pupil image - to look for AO system effects.
The smaller stop will reduce the contrast of the images, though, so the
experiment must be performed carefully.
- 3-5 micron phase retrival:
the Strehl ratio at 3-5 microns is probably
high enough to run a phase retrival code on actual starlight on the sky.
Getting far enough out of focus (presumably by moving the WFS and hence
putting focus on the DM) is an issue, but it could be possible to devise an
experiment.
|