Crosstalk for commissioning and SV data

An example for negative crosstalk can be seen on CCD N1 (#32, extension #4) of DECam_com5_00133787 Amplifier A (sources are marked in green, victims in red): Amplifier B (victims marked red): Victims on Amplifier A show positive crosstalk (victim area is brighter), while victims on Amplifier B show negative crosstalk (victim area is darker).

The crosstalk is determined by doing a linear regression to the counts of the victim pixels vs. the source pixels. Dead columns or sources nearby (bright objects as well as cosmics) can produce a bias. Examples for this are show below. Colors of symbols indicate to which source the pixel/count information belongs to.

The first two can be nicely cut by using the linear correlation coefficient used by Bill. The last two are not as straight forward.

Source amplifier A is marked with + while amplifier B is marked with triangles.

For most of the source amplifiers, 3 good xtalk fits were available, but a few only had one or two. Errors show are: weighted errors from the linear regression (black) and standard deviation of the xtalk coefficients for a given source amp. First plot shows the average (weighted mean) crosstalk coefficients including errors. Second plot is just the errors described above. The third plot shows the scatter of the coefficients around the weighted mean (for the given amp). The second and third plot will be more meaningful as more exposures are used.

For intra-ccd crosstalk only

A robust fit is done using l1-norm, i.e. sum(|model(ix)-iy|) is minimized where model = a + b*ix for a fit over all available source/victim pixel pairs. To discard amp pairs with outliers or to large scatter between sources (i.e. the offset between the sources, i.e. a_{i_source} is too large).

cuts will remove a rather large number of amplifier pairs for a given exposure (can be on the order of 50%), but will make sure not too many outliers will affect the xtalk measurement.

• only use source pixels with counts 10000 to 0.9*saturate
• source level (source but not entire amp pairing will be discarded)
  • at least 10 source/victim pixel pairs
  • a range of at least 20000 counts per source
  • nmad (robust estimate of scatter) of source pixels > 0.2*(0.9*saturation-10000) to discard sources that are mostly clustered around a very narrow range. Example:
• amplifier pair level cuts
  • at least 3 different sources (have not tested the impact of this cut, so maybe this can be relaxed)
  • at least 80 data points (source and victim pixel pairs)
  • if scatter of offset between sources is small enough (scatter of offset < 1.5*median scatter of residuals for individual source fits)

results that pass the cut overall show low scatter, but a lot of amplifier pairs don't pass the the cuts. White area means there are no values for the crosstalk (either due to being the same amplifier or there being no exposures for the amp pair that passed the cut).

x and y axis show the ccd number: 0-61 is amplifier A, 62-123 is amplifier B. Hence the main diagonal is all white as those are the elements for the same amplifier. The colored off-diagonals are the intra-ccd crosstalk.

Used exposures 133787-133806 Figure shows the xtalk coefficients, the standard deviation of the coefficients and the number of xtalk measurements used in the average (that passed the above cuts). Color scale of standard deviation is cut at 1e-4

The crosstalk matrix can be found here, columns are victim and source amp, averaged xtalk coeff, std deviation of individual measurements, number of individual measurements used. Only coefficients > 1e-6 are included. if no xtalk coeff could be determined, all values are not included

Used exposures 136085-136100 Color scale of standard deviation is cut at 1e-4!!!

The crosstalk matrix can be found here.

Noise is a lot larger - not sure why yet.

• 83 exposures
• Color scale of standard deviation is cut at 1e-4
• still no crosstalk for ccd46…?

The crosstalk matrix (with absolute values larger 1e-6) can be found here.

• 191 exposures
• full matrix is here
•