Title: Part 1: Image Motion Part 2: AO System Optimization Lecture 8
1Part 1 Image MotionPart 2 AO System
OptimizationLecture 8
- Claire Max
- Astro 289C, UC Santa Cruz
- January 31, 2008
2Outline of lecture
- Image motion and its effect on AO system
performance - AO system optimization
3Part One Image motion and its effects on Strehl
ratio
- Sources of image motion
- Wind buffeting of telescope (hard to model a
priori) - Atmospheric turbulence
- Image motion due to turbulence
- Sensitive to inhomogenities telescope diam. D
- Hence reduced if outer scale of turbulence is
D
4Scaling of image motion due to turbulence
- Mean squared angular tilt independent of l and
D-1/3 - But relative to Airy disk (diffraction limit),
image motion gets worse for larger D and smaller
wavelengths
5Typical values of image motion
- Keck Telescope D 10 m, r0 0.2 m, l 2
microns - So in theory at least, rms image motion is 10
times larger than diffraction limit, for these
numbers. - Measurements on Keck I (Gary Chanan) suggested
that actual image motion was smaller than this. - But Keck II seems to have larger image motion
than Keck I. Wind shake? Something else?
6What maximum tilt must the tip-tilt mirror be
able to correct?
- For a Gaussian distribution, probability is 99.4
that the value will be within 2.5 standard
deviations of the mean. - For this condition, the peak excursion of the
angle of arrival is - Note that peak angle is independent of wavelength
7Use Gaussians to model the effects of image motion
- Model the diffraction limited core as a Gaussian
- G(x) exp (-x2 / 2s2) / s (2p)1/2
- For Gaussian, s 0.425 x FWHM
- A Gaussian profile with standard deviation
sA 0.44 l / D has same width as an Airy
function
8Tilt errors spread out the core
- Effect of a random tilt error sa is to spread
each point of the image into a Gaussian profile
with standard deviation sa - If initial profile has width sA then the profile
with tilt has width sT ( sa2 s?2 )1/2
9Image motion reduces peak intensity
- Conserve flux
- Integral under a Gaussian profile with peak
amplitude A0 is equal to 2pA0sA2 - Image motion keeps total energy the same, but
puts it in a new Gaussian with variance sT2
s?2 sa2 - Peak intensity is reduced by the ratio
10Tilt effects on point spread function, continued
- Since sA 0.44 l / D, the peak intensity of the
previously diffraction-limited core is reduced by - Diameter of core is increased by FT-1/2
- Similar calculations for the halo replace D by
r0 - Since D r0 for cases of interest, effect on
halo is modest but effect on core can be large
11Typical values for Keck Telescope, if tip-tilt
is not corrected
- Core is strongly affected at a wavelength of 1
micron - Core diameter is increased by factor of FT-1/2
23 - Halo is much less affected than core
- Halo peak intensity is only reduced by factor of
0.93 - Halo diameter is only increased by factor of 1.04
12Effect of tip-tilt on Strehl ratio
- Define Sc as the peak intensity ratio of the core
alone - Image motion relative to Airy disk size 1.22 l /
D - Example To obtain Strehl of 0.8 from tip-tilt
only (no phase error at all, so sp 0), sa
0.18 (1.22 l / D ) - Residual tilt has to be w/in 18 of Airy disk
diameter
13Image motion main points
- Image motion can be large, if not compensated
- Keck, l 1 micron, sa 0.5 arc sec
- Enters computation of overall Strehl ratio
differently than higher order wavefront errors - Lowers peak intensity of core by Fc-1 1 / 0.002
500 x - Halo is much less affected
- Peak intensity decreased by 0.93
- Halo diameter increased by 1.04
14Part Two Optimization of AO systems
- If you are designing a new AO system
- How many actuators?
- What kind of deformable mirror?
- What type of wavefront sensor?
- How fast a sampling rate and control bandwidth
(peak capacity)? - If you are using an existing AO system
- How long should you integrate on the wavefront
sensor? How fast should the control loop run? - Is it better to use a bright guide star far away,
or a dimmer star close by? - What wavelength should you use to observe?
15Issues for designer of AO systems
- Performance goals
- Sky coverage fraction, observing wavelength,
degree of compensation needed for science program - Parameters of the observatory
- Turbulence characteristics (mean and
variability), telescope and instrument optical
errors, availability of laser guide stars - AO parameters chosen in the design phase
- Number of actuators, wavefront sensor type and
sample rate, servo bandwidth, laser
characteristics - AO parameters adjusted by user integration time
on wavefront sensor, wavelength, guide star mag.
offset
16Factors affecting performance estimates
17Example Keck Observatory AO Blue Book
- Made scientific case for Keck adaptive optics
system - Laid out the technical tradeoffs
- Presented performance estimates for realistic
conditions - First draft of design requirements
The basis for obtaining funding commitment from
the user community and observatory
18What is in the Blue Book?
- Chapter titles
- 1. Introduction
- 2. Scientific Rationale and Objectives
- 3. Characteristics of Sky, Atmosphere, and
Telescope - 4. Limitations and Expected Performance of
Adaptive Optics at Keck - 5. Facility Design Requirements
- Appendices Technical details and overall error
budget
19Other telescope projects have similar Books
- Keck Telescope (10 m)
- Had a Blue Book for the telescope concept
itself - CELT Telescope (30 m)
- Green Book
- Includes a chapter on adaptive optics
These Books are the kick-off point for work on
the Preliminary Design
20First, look at individual terms in error budget
one by one
21Dependence of Strehl on l and number of DM
degrees of freedom
- Assume bright natural guide star
- No meast error or iso-planatism or bandwidth
error
Deformable mirror fitting error only
22Dependence of Strehl on l and number of DM
degrees of freedom
- Assume bright natural guide star
- No meast error or iso-planatism or bandwidth
error
Deformable mirror fitting error only
23Strehl vs l and seeing (r0)
- Assume bright natural guide star
- No meast error or iso-planatism or bandwidth
error
Decreasing fitting error
Deformable mirror fitting error only
24Strehl vs l and guide star magnitude (measurement
error)
Assumes no fitting error or other error terms
25Strehl vs l and guide star magnitude (measurement
error)
Assumes no fitting error or other error terms
bright star
Decreasing measurement error
dim star
26Strehl vs l and guide star angular separation
(anisoplanatism)
27Strehl vs l and guide star angular separation
(anisoplanatism)
0 arcsec
4 arcsec
10 arcsec
20 arcsec
28Strehl degradation due to anisoplanatism as a
function of l
29Sky coverage accounting for guide star densities
LGS coverage 80
Tip/tilt sensor magnitude limit
Hartmann sensor magnitude limit
Galactic latitude
NGS coverage 0.1
Isokinetic angle qk
Isoplanatic angle q0
30Point-spread function for different no. of DM
degrees of freedom
PSF, l2.2 mm, D/r0 8.5
1
218 DOF
50 DOF
0.1
24 DOF
12 DOF
2 DOF
0.01
Peak intensity relative to diffraction limit
uncorrected
0.001
0.0001
0
0.1
0.2
0.3
0.4
0.5
Radius (arcsec)
31Encircled energy curves for various l
2.2 m
1.65 m
1.25 m
uncorrected
0.88 m
0.7 m
32Background-limited spectroscopy optimal slit
width
As Strehl is increased (by increasing DOF), the
optimal slit width becomes smaller
Question give a qualitative explanation of why
this is so
33Overall system optimization
34Error model mean square wavefront error is sum
of squares of component errors
- Mean square error in wavefront phase
Measurement error
How to evaluate signal to noise ratio SNR?
35(Long) Digression on signal to noise ratio
- Based on a lecture by Bruce Macintosh and on Ian
McLeans book Electronic Imaging in Astronomy - Detector technology
- Basic detector concepts
- Modern detectors CCDs and IR arrays
- Signal-to-Noise Ratio (SNR)
- Introduction to noise sources
- Expressions for signal-to-noise
- Terminology is not standardized
- Two Keys 1) Write out what youre measuring.
2) Be careful about units! - Work directly in photo-electrons where possible
36References for detectors and signal to noise ratio
- Excerpt from Electronic imaging in astronomy,
Ian. S. McLean (1997 Wiley/Praxis) - Excerpt from Astronomy Methods, Hale Bradt
(Cambridge University Press) - Both are in the Reader and on the web
37Primary properties of detectors
- Quantum Efficiency QE Probability of detecting a
single photon incident on the detector - Spectral range (QE as a function of wavelength)
- Dark Current Detector signal in the absence of
light - Read noise Random variations in output signal
when you read out a detector - Gain G Conversion factor between internal
voltages and computer Data Numbers DNs or
Analog-to-Digital Units ADUs
38Secondary detector characteristics
- Pixel size (e.g. in microns)
- Total detector size (e.g. 1024 x 1024 pixels)
- Readout rate (in either frames per sec or pixels
per sec) - Well depth (the maximum number of photons that a
pixel can record without saturating or going
nonlinear) - Cosmetic quality Uniformity of response across
pixels, dead pixels - Stability does the pixel response vary with time?
39Early detectors Eyes, photographic plates, and
photomultipliers
- Eyes
- Photographic plates
- very low QE (1-4)
- non-linear response
- very large areas, very small pixels (grains of
silver compounds) - hard to digitize
- Photomultiplier tubes
- low QE (10)
- no noise each photon produces cascade
- linear at low signal rates
- easily coupled to digital outputs
40Modern detectors are based on semiconductors
- Semiconductors have partially-filled valence
band and empty conduction band energy levels - Energetic photon can create an electron (and
corresponding hole) and kick it into
conduction band
41Bandgap energies for commonly used detectors
42Charge-Coupled-Devices, or CCDs
- CCD consists of an array of storage wells
defining pixels - Charge-shifting readout architecture allows large
numbers of pixels to couple to single output - Large arrays can be manufactured
- CCDs provided an order of magnitude increase in
sensitivity relative to photmultipliers - The most significant development in astronomy in
the 1980s
43CCD readout process charge transfer
- Adjusting voltages on electrodes connects wells
and allow charge to move - Charge shuffles up columns of the CCD and then is
read out along the top - Charge on output amplifier (capacitor) produces
voltage
44CCD phase space
- CCDs dominate inside and outside astronomy
- Even used for x-rays
- Large formats available (4096x4096)
- High quantum efficiency 80
- Dark current from thermal processes
- Long-exposure astronomy CCDs are cooled to reduce
dark current - Readout noise can be several electrons per pixel
each time a CCD is read out - Trade high readout speed vs added noise
45Main sources of detector noise for wavefront
sensors in common use
- Poisson noise or photon statistics
- Noise due to statistics of the detected photons
themselves - Read-noise
- Electronic noise (from amplifiers) each time CCD
is read out - Other noise sources (less important for wavefront
sensors, but important for other imaging
applications) - Sky background
- Dark current
46Photon statistics Poisson distribution
- CCDs are sensitive enough that they care about
individual photons - Light is quantum in nature. There is a natural
variability in how many photons will arrive in a
specific time interval T , even when the average
flux F (photons/sec) is fixed. - We cant assume that in a given pixel, for two
consecutive observations of length T, the same
number of photons will be counted. - The probability distribution for N photons to be
counted in an observation time T is
47Properties of Poisson distribution
- Average value FT
- Standard deviation (FT)1/2
- Approaches a Gaussian distribution as N becomes
large
Horizontal axis FT
48Properties of Poisson distribution
Horizontal axis FT
49Properties of Poisson distribution
- When is large, Poisson distribution
approaches Gaussian - Standard deviations of independent Poisson and
Gaussian processes can be added in quadrature
Horizontal axis FT
50Look at all the various noise sources
- Wisest to calculate SNR in electrons rather than
ADU or magnitudes - Noise comes from Poisson noise in the object,
Gaussian-like readout noise RN per pixel, Poisson
noise in the sky background, and dark current
noise D - Readout noise
- where npix is the number of pixels and RN is the
readout noise - Photon noise
- Sky background for RSky e-/pix/sec from the sky,
- Dark current noise for dark current D
(e-/pix/sec)
51Total signal to noise ratio
where F is the average photon flux, T is the
time interval of the measurement, RSky is the
electrons per pixel per sec from the sky
background, D is the electrons per pixel per sec
due to dark current, and RN is the readout noise
per pixel.
52Some special cases
- Poisson statistics If detector has very low read
noise, sky background is low, dark current is
low, SNR is - Read-noise dominated If there are lots of
photons but read noise is high, SNR is - If you add multiple images, SNR ( Nimages
)1/2
53Now, back to calculating measurement error for
Shack-Hartmann sensor
- Assume we are read-noise limited (usually the
case with todays CCDs). Then
54Error model mean square wavefront error is sum
of squares of component errors
- where Tcontrol is the closed-loop control
timescale, typically 10 times the integration
time Tint
55Question
- For a fixed subaperture diameter d, approximately
how does each term scale with r0? - What are the implications for designing a new AO
system to work at a shorter wavelength?
56Integration time trades temporal error against
measurement error
- Modified from Hardy, Figure 9.23
Measurement error r0 0.1 m
Temporal error 1 / t0 39 Hz
Optimum integration time
57First exercise in optimization
- Minimize the sum of read-noise and temporal
errors by finding optimal integration time - Sanity check optimum Tint larger for long t0,
large read noise RN, and lower photon flux F
58Similarly, subaperture size d trades fitting
error against measurement error
- Hardy, Figure 9.25
- Smaller d
- better fitting error, worse measurement error
59Keck AO error budget (as of a couple of years
ago)
Assumptions NGS is 10 arcsec off-axis NGS is
very bright 30 degree zenith angle Closed-loop
bandwidth 90Hz Chose to have relatively large
fitting error, in order to be able to use fainter
guide stars (lower measurement error) Note that
uncorrectable errors in telescope itself are
significant
Just an example
60Summary What can you really optimize?
Once telescope is built on a particular site,
dont have control over t0, ?0 , r0 But when you
build your AO system, you CAN optimize choice of
subaperture size d , maximum speed of AO system,
field of view, etc. Even when you are observing
with an existing AO system, you can optimize
things wavelength of observations (changes
fitting error) integration time of wavefront
sensor Tint tip-tilt bandwidth brightness of
guide star
61Summary, part two
- Image motion
- Broadens core of AO PSF
- Contributes to Strehl degradation differently
than high-order aberrations - Crucial to correct tip-tilt