from photography to ccd imaging · 2016. 9. 3. · higher efficiency, but more difficult to handle...
TRANSCRIPT
The evolution of detectors:
from photography to CCD’s
Michel Dennefeld
Institut d’Astrophysique de Paris
and
University P. et M. Curie (Paris 6)
and thanks to Henry Mc Cracken ...
Part I : History and progress.....
The human eye: -2.5 cm diameter lens -130 millions pixels -125 M cones (day-time, tri-colour vision) ~ 140˚ FOV, 1’ resolution -5 M rods (night-time, grey vision) -Spectral response: -Daytime: peak at ~ 550 nm -Nightime: slightly bluer
Transparency and energy versus λ …
hν~ GeV-keV ~ 1 eV I=[a(sinωt-φ)]² ?
“Modern” detectors
transform a photon into a (photo)-electron!
What makes the difference, is what happens then to this electron...
hν = Eextr + ½ mv2
Wavelength threshold: λ°(μ) < 1.24/ E(ev)
e.g. Silicon: 1.1 μ; Photo: ~ 0.45 μ
•The principle is the same for all detectors:
-If the electron « escapes » the material, you have
photoelectric systems Requires additional extraction energy (thus less red sensitivity),
but you can then act on the electron (amplification)
-If the electron remains inside, you have
photoconductors Higher efficiency, but more difficult to handle
Photographic plates
• It was the first objective and permanent record of astronomical
phenomena (~ 1880 +...)
• Long exposure times are possible thus faint objects can be detected
• Exists in large format: many sky surveys carried out, as recently as
the 1990s. e.g. the Palomar Sky Surveys (DSS...) in B (103aO) and
R (IIaF); in the South, IIIaJ, etc...
AgBr crystals: Br- + hv gives Br + e- and Ag+ + e- gives (unstable) Ag
The development enhances the quantity of Ag where it exists.
Problems: very non-linear process; impurities and traps compete
for the electrons, hence very low efficiency.
The latter can be improved with sensitising processes, but
the DQE hardly comes over 3%!
But what are QE, DQE, 103aO, etc...??
Fondamental parameters for a detector...
• QE (noted η) = <number of
electrons/photons>, is << 1 !
• DQE = (input S/N)2 / (output S/N)2 thus
DQE is ≤ η (in Poisson statistics, the S/N
is equal to √n )
• Sensitivity: NEP, input power needed to
provide a S/N of 1
• Its dependance with wavelength...
• Linearity, dynamical range, MTF
Parameters of photographic plates
• How to improve sensitivity?
• -Sensitizing (= reduce traps)
•
- Higher energy electrons
(photocathodes + high voltage
= electronographic systems)
Photoelectric detectors
• First application:
Photomultiplier tubes (1950-1990s) = single pixel detectors!
High-voltage (~ 20 kV) acceleration and amplification of electrons.
Initially measured output currents, later used as photon-counting
systems.
• More sensitive than photography (~20% quantum efficiency)
• Extremely stable, linear, broad band devices which made possible
the first highly accurate photometric systems
• Flux calibration of photomultiplier based systems is still more
accurate than modern array imaging detectors!
How to get 2D images with photoelectric systems?
• Electronographic cameras:
• Electrostatic, or magnetic focusing, +photographic plate output
• But photocathodes need to remain in vacuum... (Lallemand, Kron, Spectracon cameras...) •TV tubes, but not very stable, and poor PSF
• Combined systems lead to Photon-Counting devices.
How do Photon Counting systems work ?
• High e- amplification: bright patch of light
at the output
• Centroiding of the light patch, at a fraction
of (output) pixel.
• Coordinates of the event are stored in the
memory: independant of the intensity of
the patch!
• Thus each event has the same weight
• Single photo-electron is detected if
amplification high enough
• No noise, but saturation limit
• Deadtime between events (e.g. decay time
of output phosphor)
Modern photon-counting systems
• Micro-channel plates: more compact, and
lighter than photo-tubes. Various read-
out modes...the best is:
• MAMA (Multi-Anode Multichannel
Arrays)
Photon counting systems:
-Pro’s: - Single photon detected
- Equal weight for each event
- No noise
-Con’s: - Sensitive to high-flux (deadly!!)
- Low QE (limited by photocathode, ~20%)
- DQE ~ 7% (e.g. HST/STIS)
How to improve the Quantum Efficiency?
and its wavelength dependance?
• ~20% QE limit for photo-emissivity,
• and poor red-sensitivity
•Semi-conductors have better QE,
but electrons trapped inside
•One thus needs to find ways
to « read » the electrons...
How a semi-conductor works...
• Exemple of a photo-diode
- Energy diagram
-Read-out mode
Many capacitors in parallel,
thus high read-out noise
(NEC ~Cv/q √[4kTBR1] )
14
How to do it on a 2D image: CCD Readout
Thomson 2048 X 2048 CCD
Horizontal register = Serial register
Digitize via Analog to digital (A/D) converter
Charge Coupled Devices (CCD’s)
•First astronomical use: TI 800 x 800 for HST (~ 1980)
• To reduce Cv: no external connections, and no
physical limit between pixels! Only one connection at
the extremity.
• The free photo-electrons in the conducting band are
then ‘shuffled’ to the output electrodes by applying
electrical pulses.
• The output signal is amplified and converted from
electron units to digital units with a conversion factor
or Gain, “g” (measured in electrons per analogue-to-
digital unit, e-/ADU).
• The Gain is simply a multiplicative factor.
• Number of electrons = Pixel ADU value x Gain
How CCD’s work (II)
• Reading out a CCD can take time,
especially for the largest arrays
with many pixels: the mode is
different for video or astro
• The r.o.n. is ~ 1/ √t
• You can use several amplifiers in
parallel, but it is then difficult to
have the same gain everywhere!
• In drift scanning CCDs are read out
continuously (for example like the
SDSS camera); no read-out time
but integration time is fixed!
• Shuffling charge on CCDs can be a
way to compensate for atmospheric
seeing -- see the orthogonal charge
transfer chips used in the Pann
Starrs project
The positive points of CCD cameras...
• Extremely good quantum
efficiency at redder
wavebands – close to 100%!
• Limiting magnitudes increased
by four to five magnitudes!
• Sensitivity of telescopes is now
limited by light collecting area
and not the detector area.
• CCD’s are linear over a very wide flux range: calibrations are easier!
• Output is digital!
Some problems with CCD detectors
• Poor blue (< ~4500 Å) response (too small
penetration depth of blue photons)
• Megacam (Canada-France Hawaii
Telescope) is one of the few wide-field
detectors with good U* response
• Adequate surface treatment can help solve
this problem, but at the cost of fringing!
• It’s hard to make large monolithic CCD
detectors: make mosaics from ‘buttable’
detectors, but problem of curved focal planes.
• Charge transfer efficiency between wells is
not 100% and degrades over time especially in
space-based detectors.
• Sensitive to Cosmic Rays
• Cryogenic cooling (normally
liquid nitrogen is used) is
necessary to reduce dark
current
• Reading out CCDs can take
time!
Examples of unwanted effects...
Charge transfer bleeding in STIS detector
Diffraction spike (secondary
mirror) Not specific to
CCD’s !
Cosmic ray hits in megacam-u* image
Some real image defects (Hubble/WFPC2):
Ghost
Bleeding
Cosmic ray
Charge Transfer (In)efficiency
CCDs are read out by
clocking charge along
registers. These transfers are
impeded by radiation damage
to the chips.
This effect gets worse with
time and is worse in space,
This image is from the STIS
CCD on Hubble. Note the
vertical tails on stars.
Can degrade photometry and
astrometry
More unwanted effects...
• Fringing : due to « thin » CCD...
• Stronger in red than in blue
• Dead columns
• Non uniform dark current
• (e.g. heating by the amplifier)
All that can (needs to...!) be corrected
during data reductions.
(Flat Fielding!)
Geometric
Distortion
HST/ACS/WFC - a
severe case of distortion
- more than 200 pixels at
the corners. Large skew.
Cameras normally have
some distortion, typically a
few pixels towards the
edges,
It is important to understand
and characterise it to allow it
to be removed if necessary,
particular when combining
multiple images.
Distortion may be a function
of time, filter and colour.
Due to optics, not CCD!
Characterising CCD detectors
• Quantum efficiency (QE): Can be as high as
90% in visible wavelengths for the best detectors
today. Specific treatments to improve blue (or
red) response.
•Resolution/pixel size: typical pixel sizes from
10 to 30 microns. For the images to be properly
sampled, there must be at least two resolution
elements per PSF (point spread function), and
better three... otherwise the images are
undersampled.
Characteristics of CCD’s (II)
• Dynamic range/full well capacity/saturation level.
• Around 105e-. Close to saturation level, non-linearity effects start to show up.
• It’s important to choose an exposure time short enough so that objects of interest are not saturated! (But long enough so you are sky-noise limited).
• Multiply exposures to be able to remove CR !
• Even in the absence of illumination, CCDs produce a thermal signal or ‘dark’ current.
Measure during bad weather...
Sources of noise in CCD observations
• Dark current
• Even in the absence of illumination, CCDs produce a signal or ‘dark’ current. Typical values are ≤ 10 e- /hr if cooled to ~ -140° (LN)
• Readout noise
• This is an electronic noise (amplifier, or else), usually quoted in electrons per pixel. It does not depend on exposure time.
• Photon noise (object, or sky)
• Photon noise follows a Poissonian distribution (counting statistics): σ = √n
• For bright objects, object photon noise dominates
• For faint objects, the photon noise from the sky dominates. Sky background increases at longer and longer wavelengths.
• In order to avoid being dominated by read noise, CCD exposures should
be long enough to accumulate enough counts in the sky background to
be sky noise dominated. Usually difficult in spectroscopy !!
Sky background
Night sky from Paranal • Moonlight is yellow, so better to observe in redder
bandpasses near the full moon
• At longer wavelengths airglow lines cause significant
variations in sky brightness
The CCD equation
• Object Signal: Sobj = E* D2 Δλ η τ t
• E in ergs/cm2/s/Å or W/m2/Å, flux from object
• D Diameter of telescope, Δλ (bandpass), η (quantum efficiency), τ (transmission), t exposure time
• Total signal: Stot = (E* + (2)Esα2) D2Δλητt + dark. t + n2R2
• (α seeing, n number of pixels corresponding to α)
Astronomical
Source
Sky
background Dark current
Read noise
Total counts
per pixel, in
electrons
N(oise) = square root of total signal
Carefull! if calculation made in ADU or in electrons, or in photons...
per pixel, or per total area of object!
A rough calculation of signal to noise
• Usually, Dark Current is negligible
• We assume the read-out noise is negligible also
• IF object brighter than sky, then:
• S/N = [ (E* . D2Δλητt ]1/2 (~ independant of seeing)
• and Limiting magnitude Elim is ~ (S/N)2 1/ [D2 Δλητt]
• goes as 1/D2 or 1/t for a given S/N
• IF object fainter than sky, AND r.o.n. negligible, then:
• S/N = E* . D/α. [Δλ η τ t /Ec]1/2 or:
• Limiting magnitude: E* = S/N . α/D . [ES / (Δλ η τ t)]1/2
• Goes as 1/D, or 1/√t only
• and Seeing is as important as diameter of telescope...!
• This explains the performance of the HST despite its only 2.4m diameter.
“How long should I integrate for” ??
• You use the above equation...but you have to know the parameters!!
• Most observatories will offer “exposure time calculators” which will allow you to
estimate how long to integrate to reach a given S/N
• But you need some critical eye...input parameters can be wrong, software have
bugs, etc...so always have a rough check with the above equation yourself.
• But if you have (or can get) real data taken with the same instrumental setup, it’s
almost always better to use this, and to scale it up to your particular application.
• In practice, CCDs have finite “full well capacity” and the non-zero sky background
means that integrating infinitely long is impractical (images would saturate on the
sky). Also detectors have bad pixels / columns, so it is in general better to take
several shorter exposures which are then dithered or offsetted on the sky and
which can then be combined afterwards. This also helps to remove C.R.
Important to choose large enough dithers to overcome detector defects.
• This is however only practicable for imaging, not always for spectroscopy......
CCDs and CCD mosaics
The next generation of large mosaic detectors...
Image Area Image Area (Architecture unchanged)
Serial register Serial register {
Gain register
On-Chip
Amplifier
On-Chip
Amplifier
The Gain Register can be added to any existing design
Conventional CCD EMCCD
Electron Multiplying CCDs.
On-Chip
Amplifier
CCD60 128 x 128 CCD87/97 512x512 CCD201 1K x 1K
Image Area Store Area
Nor
mal
Seri
al r
egi
ster
Multiplication register
Multiplication
registe
r
Multiplication register
Standard MOSFET amplifier
1e- in
1000e- signal out
Avalanche multiplication takes
place in Multiplication Register,
using an HV clock (40-45Volts).
Devices used :
Electron Multiplying CCDs.
52
0 m
ultiplication stages
Multiplication Noise
EMCCD Photon Transfer Plot
The EMCCD multiplication process increases the Poissonian noise in the image
Electron Multiplying CCDs.
Benefits : 1) Negligible read noise, therefore capable of seeing single photons 2) Read noise independent of read-out rate Disadvantages : 1) At present only available in small formats. 2) Suffer from ‘Multiplication Noise’ that limits performance at high signals.
Applications : Low signal regimes where detector read noise is the limiting factor. This could be an intrinsically faint source or a brighter source observed at very high frame rates. -> Adaptive optics wave front sensing -> High time resolution spectroscopy
QUCAM2 : an EMCCD camera on ISIS.
1k x 1k pixel image area. 3.2second full frame read out. Frame Transfer design
Superconductive Tunnel Junction
• ESA tests (M. Perryman, 2009)
• Charge proportionnal to the
incoming photon energy
• R (spectral resolution) limited
by the statistics of the generated electrons (R ~ √E , a few 10)
• Working at very low
temperature, in photon-counting
mode
• Tests made at the WHT
• Developing a 2D array, and
searching for “ higher” T
semiconductor material:
S-Cam, 10 x 12, ESA 1m, R~15
(Martin et al. 2006)
Photon Energy
R
40
Image courtesy of J. Allington-Smith (CfAI - U. of Durham)
Photonic Multi-object spectroscopy ?
Use of photonic lanterns
to transfer multi– into single-- mode fibers (p=V²/4, V=kₒ a sinθ) (kₒ=2π/λ)
41
The PIMMS instrument concept
J. Bland-Hawthorn et al,
Proc. SPIE 7735, 77350N (2010)
The photonic integrated multimode
micro-spectrograph (PIMMS)
Optical/NIR Microwave Kinetic
Inductance Detectors
Rupert Dodkins University of Oxford
DARKNESS 10,000 pixel
MKID array ARCONS 2024 pixel
MKID array
MKID’s
Operating Principle
Incident photons briefly change the total
inductance of superconductors through the
Kinetic Inductance effect
1 MKID pixel
(microresonator)
Optical/NIR MKIDs
MKIDs background: 2/3
Resonant frequency
(and amplitude) shifts
P. Day et al., Nature 2003
Specifics of IR detectors
• The thermal background could rapidly
saturate the detector…thus separate
the detection area from the reading
area.
• In the near-IR (λ <~ 2.5 µ), one uses
Hg-Cd-Te (variable proportions)
• For longer wavelengths, material
changes (InSb, SiGa, GeGa, etc…)
• In the thermal IR, necessity of rapid
read-out, and wobbling (background
subtraction: the background is much
more important than the signal, and
can change rapidly)
Various IR detectors
Hg (1-x) Cd (x)Te:
-0.85-2.5μ, 2k x 2k (e.g. VISTA)
20μ pix, 78 ᵒK
-0.6-5.3μ, 18μ, 37 ᵒK, (JWST)
InSb
0.6-5.5μ, 2k x 2k, 25μ, 32 ᵒK
SiAs (IBC)
5-28μ, 1k x 1k, 18μ, 8 ᵒK, (WISE)
5-28μ, 1k x 1k, 25μ, 7 ᵒK (JWST)
JWST Detectors
Hawaii 2k x 2k MIRI 1k x 1k
End of Part I
• Questions?
Part-II: Reducing CCD observations
CCD calibration frames
• Raw images from CCD detectors are not immediately
usable for scientific exploitation but are instead
contaminated by several instrumental effects. which need to
be removed.
• Bias frame: There is a fixed offset level applied to all
data. There may also be structure in the bias. It can be
measured by examining the “overscan” region of each
CCD, or take a zero-exposure (= bias) frame.
• Dark frame: Even without any signal, a “dark current” of
a few electrons may be present. Ideally, one should take
a dark frame of the same exposure time as the real data.
• Flat field: the pixel-to-pixel sensitivity of the detector is
not constant. A uniformly (Flat...) illuminated exposure
can measure this. Flat-field is wavelength dependent.
• Fringe frame: Thinned CCDs do produce interference
fringes at longer wavelengths. The amplitude and
separation of the fringes can be time-dependent, making
it difficult to remove them (night-sky variations...)
Megacam data with badly
removed fringe pattern
Typical CCD pre-reduction steps
• Subtract bias or overscan
• Master bias can be computed from median of many frames. The “overscan”
region can be used to compute the bias level
• Subtract dark frame
• Master dark frame should be computed from many observations of several
tens of minutes scaled up to the appropriate exposure time
• Divide by twilight or dome flat-field (filter dependent; imaging only)
• Dome flat-fields are almost always not flat enough for wide-field cameras.
• Sky flats are time limited...the best is to use “super-flats” (imaging only)
• Make sure that there are enough counts in the flat-fields so they are
effectively sky-noise limited.
• Subtract sky flat / fringe frame
• At each stage in the reductions you should verify that the noise per pixel in the
individual frames decreases. It should approach the Poisson limit.
• If the reduction steps add noise, then you are doing something wrong!
Other calibration frames you may need
(for imaging)
• A bad pixel mask is computed from a
single image and identifies where the
stuck pixels, bad columns and cosmic rays
and satellite trails are in each image.
• A weight map or coverage map which is
computed from the normalised flat-field.
• The weight map is necessary to compute
magnitude errors and detection thresholds
correctly in a stacked image. Weight maps
are less important for single detector
cameras.
• Not all sections of the image will receive
equal exposure time (vignetting..., shutter
problems, etc...)
Megacam single image weight map (flat-field +
bad pixel mask + cosmic ray mask)
FITS: the standard for astronomical images
• FITS=flexible image transport system
• FITS is an architecture independent way to store and
transfer astronomical data
• Simplest FITS file consists of a primary header
comprising of n(36x80) ASCII characters followed by
binary data
• FITS files can also contain tables or spectra.
• In ‘multi-extension-FITS’ there are several extensions,
which can be useful for mosaic camera data.
• topcat, ds9, fv can be used to manipulate and inspect
fits tables
Complete reduction step for CCD imaging data
• Carry out the pre-reductions described previously.
• If you have a series of images on the same part of the sky you will need to separately flux calibrate and astrometrically calibrate them.
• Images which are dithered around the sky will need to be re-aligned. For small detectors, linear shifts are sufficient; for wide-field imaging detectors interpolation and reprojection will have to be carried out
• Next the images need to be combined to produce a single, stacked image.
• The last step involves the extraction of catalogues.
• Spectroscopy: next talk !!
CCDPROC (iraf)
SCAMP (Terapix)
SWARP (Terapix) or
IMSHIFT (IRAF)
SWARP (Terapix) or
IMCOMBINE (IRAF)
Sextractor or PHOT