Crosstalk analysis of the SV data indicates non-linear crosstalk behavior for negative crosstalk which consists of two components (more details below):
1) "saturation" of the crosstalk at about 40 - 45 k counts (above overscan level)
2) for very large, bright sources which show an almost flat source profile (and hence should have a flat crosstalk signal), the corresponding crosstalk is larger than expected at the edges of the source.
The crosstalk correction is working reasonably well for bright (but not too bright) sources. But for very bright sources (~60000 cts), the corrections get bad, at least for some part of the victim area. I'll illustrate the effect for ccd 3 in exposure 146986 on nite 20121104 (which has negative crosstalk). The left plot is the uncorrected exposure, the one on the right is the corrected image - and strongly overcorrected. Less bright sources are corrected fine (i.e. http://www.usm.uni-muenchen.de/people/paech/files/xtalk_neg_example4.png). I have also seen this in later nights (20121119).
Using a smaller coefficient will result in the following correction, which will do a better job on the central part, but does not get the outer parts right (and it will leave less bright sources undercorrected http://www.usm.uni-muenchen.de/people/paech/files/xtalk_neg_example5.png).
This seems to indicate that there may be some nonlinear effect. So I used several exposures from nite 20121102 and 20121104, overscan corrected them (as well as subtracting the median sky level) and stacked source vs. victim pixel counts for all exposures of those nights with a sky noise < 7 (leaving about 50 exposures) and created binned averages (black symbols) and median (red symbols) to illustrate the effect (lines are fits to the binned values):
For very high source counts, the relation deviates from a linear relationship (this is also true if I use only exposures from one of the nights). Not sure if we would expect such a behavior taking nonlinearities into account. I see this for a lot of amplifier pairs with negative crosstalk that show this behaviour.
The same is true for other amplifier pairs with negative crosstalk. The upper plot shows the median binned crosstalk for all amplifier pairs (same ccd) with negatvie crosstalk. Also have fitted a straight line to those binned crosstalk measurements up to 35000 cts where the crosstalk ist mostly linear and subtracted this linear crosstalk from the binned crosstalk (lower plot) indicating where the deviation from the linear regime occurs.
However, using a linear interpolation between those average data points, I corrected the original exposure (with my own python script) for crosstalk.
Looking at the source profile, I don't see why I'd see these kind of residuals. This led me to look at the profile of the source and victim counts (using the median of +/- 10 rows for a given column for the victim area). First row is the source profile trough the center of the source. Second is the victim pixel counts (corrected for average background level and overscan) in blue and the crosstalk correction that is applied according to a linear interpolation for the above binned medians in green. Third row shows the victim area after the correction is applied.
Moving up/down 20 rows does not change the picture:
It seems crosstalk also may be depending on the location of the CCD:
Positive Intra-CCD crosstalk (averaged over multiple exposures) (Updated March 4, 2013)¶
Positive crosstalk also shows some non-linearities, though not as strong (noise makes hard to tell, will try averaging over more exposures)
Inter-CCD crosstalk averaged over multiple exposures¶
Using average over exposures from 6 different nights (225 exposures with low sky noise) crosstalk coefficients are determined. As there are a lot of pairs to check, start with intra-DA board amp pairs and concentrate on neighboring DA's for inter-DA crosstalk. First matrix exculdes crosstalk coefficients for intra-ccd crosstalk.
Most of the Inter-CCD crosstalk is positive (only a few pairs with small negative crosstalk), no strong non-linearities are visible (hard separate the noise from the signal). Some examples for strong Inter-CCD crosstalk are below.
Some examples of very large inter-ccd crosstalk (including intra-ccd crosstalk for source for comparison):¶