Horizontal Striping
Vertical Striping
Superdark has bias “hot column” residuals in it. These
don’t scale linearly with exposure time and print onto
science data.
Superbias images below with hot pixels removed, then smoothed to show structure
bias stripes removed
with bias stripes
Pipeline bi-weekly superdarks show similar bias striping
Other macro-scale structures in the dark are likely amplifier bias shape
differences between the 1000s dark exposure and the 0s bias subtracted from it.
The 1200s science exposures will have similar (but not exactly the same)
residuals.
Low/hi corners and
vertical/horizontal
gradients across the
quads are most likely
bias, not dark current
Glint
(internal optical reflection)
Optical glint, possibly from an off-chip
source crosses the field diagonally. Would
like to characterize its behaviour (motion and
intensity) so it can be subtracted from the data
before drizzling.
The glint in each chip was
measured at 3 points along
its length to get its slope and
amplitude
Fitting lines to the glint across both chips revealed that
its is pivoting about a point near the upper left edge of
chip2.
A model glint image
was build by shifting
and rotating all the
images to the same glint
orientation and filtering
sources. This model,
masked to minimum
extent then rotated,
shifted and scaled, was
then subtracted from
each input image before
drizzling.
Chip 2
Chip 1
Close-up of Chip 1 Glint
Residual “sky” Subtraction
Since many of the residual
bias/dark structures
discussed previously have
not been corrected by the
reference files, a catch-all
sigma-clipped average,
source-masked sky made
from all of the input images
is subtracted from each
image before drizzling.
Some glint residual is visible
here and that will
oversubtract from the data,
leaving a faint dark residual
in the final image.
This is an image of all the systematic noise
remaining after the pipeline calibration
Amplifier Ringing?
Moderately bright star (but unsaturated in 1200s) leaves a long
decaying, very faint trail in the serial read direction. The decay is
slow enough to wrap around to the next row. This is similar to
behaviours seen in WFPC2 and NICMOS, but much, much
fainter.
Its likely this is
present for all
bright, sharp
sources, but
only visible
above the noise
with deep
imaging like
UDF
Correctable with a sum of exponentials in the serial read direction
Electronic Crosstalk Ghosts
Sources produce dark ghosts of themselves in the opposite quadrant
and even in the quads of the other WFC CCDs. This is some form of
electronic crosstalk (see ACS ISR 4-12, 4-13, Giavalisco 2004 for details).
Even in this GAIN=2 data, the ghosts are quite prominent. In the
drizzled z-band image, the ghosts are smeared by the dithering.
Due to the orientations of the readout directions of the 4 WFC
quadrants, the ghosts move in the opposite direction of the sources
in the serial readout direction (X), but they move together in the
parallel readout direction (Y).
Ghosts and their sources
Serial
Parallel
C
Serial
Parallel
A
Serial
The same logic
applies to all 4
quads.
D
Y
Parallel
If a source in
quad D moves in
+X,+Y, its ghost
in quad C moves
in -X,+Y.
Chip2
Parallel
ACS WFC
readout
orientations.
Sources and
ghosts follow the
arrows.
B
Serial
Chip1
X
Flip -X,Y then copy
Serial
Parallel
C
D
Y
A
Serial
Parallel
Parallel
Three
permutations -3
drizzles - gives
you all 12 ghost
quads. (3 ghosts
per 4 source
quads)
Serial
Parallel
Use a trick to
make drizzled
images of the
ghosts: Flip the
quads in
appropriate
directions, then
drizzle!
Example for l-r interchip permutation
B
Serial
Flip -X,Y then copy
X
Once you have high S/N images of ghosts you can either
subtract the image directly, or try to build a model of the ghost
behavior as a function of source - that’s the next step…
Side-by-Side Comparison
Original
(same stretch)
Cleaned
Still to do: E-ghosts, better glint subtraction, amp ringing correction
end transmission