Interquadrant row crosstalk
Gert Finger, G. Nicolini
European Southern Observatory
for pdf file ( much better resolution)
Illumination of the complete row
Several images taken during
the SOFI and ISAAC tests have been re-analyzed to further examine one peculiarity
of the array which we call interquadrant row crosstalk. The effect is apparent
when part of the array is exposed to bright illumination and faint objects
have to be detected on the same rows which are exposed to the bright illumination.
One has to account for this effect in spectroscopic applications, when
faint lines have to be detected between bright OH lines or the continuum
to line ratio is of interest.
1 and Figure 2 show
the shading effect with a long slit oriented parallel to the rows of the
detector. The slitwidth is 2 arcseconds on the sky. The image shows the
spectrum of ambient background radiation observed with the narrow band
CO filter and the red grism of SOFI. The dispersion of the grism is 10.22A/pixel.
The peak flux in the center of the slit image is 1374 e/sec/pixel. The
calculated photon flux for the given instrument setup is 1846 photons/sec/pixel.
The rows in quadrants III and IV near row 798 which have increased intensity
but do not receive any photons, are read out at the same time as the rows
close to row 286, which are illuminated by the slit. The intensity of row
798 is 1.5% of the intensity of row 286.
This interquadrant row crosstalk
is a detector anomaly assigned to the multiplexer. The effect is poorly
understood and not negligible at all when a complete row is illuminated.
In imaging mode the Wollaston
prism produces two much brighter images of the slit. The long slit was
tilted along the diagonal of the lower rectangle shown in Figure
3 . The bright stripe produced by interquadrant row crosstalk is marked
by the upper rectangle in Figure
3 and has the same width as the lower rectangle. In each row inside
the lower rectangle a few pixels are illuminated by the slit and result
in the corresponding intensity increase of the row inside the upper rectangle.
Illumination of a small fraction of a row
A second test was performed
to measure interquadrant row crosstalk when only a fraction of the pixels
in a row is illuminated as can be seen in the image shown in Figure
. The echelle slit of SOFI is imaged on 82 pixels in a detector row
generating a peak signal of 26332 electrons/sec/pixel. It is remarkable
that pixels being addressed prior to the brightly illuminated pixels show
crosstalk at the same level as pixels being addressed later than the bright
source. The crosstalk generated in the row of the quadrant containing the
slit image and on rows in the remaining three quadrants being read out
at he same time has a uniform intensity of 30 electrons/sec/pixel which
is 0.113% of the intensity of the slit image.
By scaling the crosstalk to
the number of bright pixels measurements obtained with the short slit (
) and the long slit (
and Figure 2
) yield the same crosstalk. With the echelle slit and with the long slit
the crosstalk is 1.39 10-5 and 1.46 10-5 times the integrated intensity
along the row respectively. The interquadrant row crosstalk is uniform
and depends linearly on the integrated intensity in a row.
A second image of the echelle
slit was taken with increased photon flux saturating the detector pixels.
Since double correlated sampling was used, saturated pixels in the center
of the echelle slit appear dark in Figure
5 . The trace perpendicular to echelle slit shown in Figure
5 and the trace perpendicular to the associated row crosstalk which
is shifted by 512 rows is shown in Figure
6 . Both the image of the slit and the interquadrant row crosstalk
show the saturation of pixels as reduced intensity which is expected for
double correlated sampling.
Illumination of a large area
In ISAAC the dispersion direction
is parallel to the row direction of the detector. In the low resolution
spectroscopic mode a blackbody source at a temperature of 130 C viewed
through the K band filter illuminates a larger area of the detector as
can be seen in the left image of Figure
. Reducing the cut levels in the right image of Figure
makes visible the interquadrant row crosstalk, which appears as darker
and brighter stripes outside the K-band spectrum where the array does not
receive any photons. At the top and at he bottom the array does not receive
any light. The remaining intensity of ~100 electrons at the top and the
bottom of the array is due to interquadrant row crosstalk of the rows in
the corresponding upper or lower quadrants which are brightly illuminated.
Each of the bright rows receives a total integrated intensity of 7.39 106
electrons during the integration time of 1.4 seconds. The rows of the two
darker stripes in the center see crosstalk of the illumination incident
on the row itself but no crosstalk from rows of other quadrants read out
at the same time, since these rows are dark. The bright stripes in the
upper and lower quadrants see interquadrant row crosstalk in addition to
intrarow crosstalk within the row itself. The crosstalk of the bright stripes
has twice the value of the crosstalk of the dark stripes. Since the bias
pattern of the array is nonuniform and the trace perpendicular to rows
as shown in Figure 8
a superposition of bias pattern and row crosstalk, the discontinuity step
at positions 1,2,3 and 4 was used to quantify crosstalk. The step size
is 94.2 electrons yielding a crosstalk of 1.3 10-5.
Integrated intensity in row
is 1.324E6 ADU = 7.388E6 e
Illumination with a point source
If all the light is concentrated
in a single point source as in the 30 Doradus image shown in Figure
, interquadrant row crosstalk can still be seen. The intensity of
the point source on row 384 is shown in Figure
. The integrated intensity of the point source in row 384 is 64170e/sec/pixel.
A trace perpendicular to the rows affected by interrow crosstalk is plotted
in Figure 11
. To increase
the signal to noise ratio of the trace, columns 210 to 385 have been averaged.
The intensity in row 896 is 0.87 e/sec/pixel. This yields a measured crosstalk
of a=1.35 10-5. This reconfirms that the row crosstalk is proportional
to the integrated intensity along a row.
Swapping of fast and slow clocks
A further effort has been
made to reduce the crosstalk by clocking the fast column shift register
slowly and the slow shift register (row) fast. Figure
shows the image of a grid of holes applying standard clocking. The
column shift register is running with fast clocks and the row shift register
is running with slow clocks. The image in Figure
was taken under identical conditions but the clocks were swapped.
The row shift register is running with fast clocks and the column shift
register is running with slow clocks. The Holes at position 1,2,3,4 and
5 have trails shifted by one pixel in direction of slow clock. The trails
at position 3 and 4 can hardly be seen in Figure
due to poor reproduction quality. The trails have uniform intensity
and the crosstalk a is 2 10-4. They are shifted by one pixel in the direction
of the slow clocks. Not all point sources have a trails suggesting a kind
of subthreshold effect of the addressing FET's. The standard clocking does
not exhibit any trail at this level. Therefore swapping clocks is not recommended.
Model for row crosstalk
The row crosstalk is uniform
within one row and does not depend on column index j. Let
be the intensity of the pixel at row i and column j. Due to row crosstalk
the observed intensity
is modified by
the row crosstalk as described by equation
The row crosstalk
consists of two terms, namely the intraquadrant row crosstalk
and the interquadrant row crosstalk
. Both terms depend linearly on the integrated intensity of row number
i and row number as described by the
coefficient in equation
2 . The plus sign applies for indices
, the minus sign for :
can be derived from the observed intensity
by subtracting the row crosstalk as shown in equation
The result of this correction
algorithm is demonstrated in Figure
14 and Figure 15 by
comparison of the raw K-band spectrum of Figure
14 with the K-band spectrum modified by the correction algorithm using
equation 3 . The row crosstalk
can be well removed. The correction algorithm was also applied to the 30
Doradus image. Row crosstalk can be removed without any degradation of
image quality as shown by Figure
16 and Figure 17 with
the raw and the corrected images of 30 Doradus.
Even though interquadrant
row crosstalk is not yet understood the following behavior of the effect
has been observed:.
the crosstalk is uniform for all pixels in a
The crosstalk affects all rows in the other
quadrants being read out at the same time with equal intensity. MUXSUB,
CELLWELL and DSUB are the only bond pads which are shared by all four quadrants.
Pixels being addressed prior to the brightly
illuminated pixels show crosstalk at the same level as pixels being addressed
later than the bright source.
The intensity of the crosstalk is 1.3 10-5 times
the integrated intensity along one row
The effect is 1.5% if a slit is oriented parallel
to the row and all pixels are illuminated. This case is encountered when
the dispersion direction is perpendicular to the rows and bright atmospheric
lines illuminate the whole length of the slit.
Saturation of brightly illuminated pixels is
reproduced by interquadrant row crosstalk
If the slow shift register is clocked fast and
fast shift register is clocked slow the point source images exhibit strong
trails confined to one quadrant, an effect not apparent with standard clocking.
Spectroscopic applications will be adversely
If an explanation can be found design modifications
on the 2Kx2K multiplexer should be made to suppress this effect
An efficient and simple algorithm has been developed
which removes the effect of row crosstalk without any degradation of image
Further discussions should
clarify whether the dispersion direction should be perpendicular or parallel
to the rows of the array. If the dispersion direction is parallel to the
rows all bright spectral lines of a point source add to row crosstalk.
If the dispersion direction is perpendicular to the rows bright skylines
(OH lines) which fill the complete length of the slit add to row crosstalk.
A simple alternative is
offered by the algorithm described above which efficiently removes the
effect of row crosstalk.