3. Removing the Effects of the Charge Transfer Traps

Figure 5 shows that in most cases where two subexposures have been taken, performing the standard cosmic ray removal procedures removes the majority of the tails. Exceptions are:

the dark tail just above the actual trap, which can be treated with wfixup or fixpix,
cases where real objects cross the bad column, such as the object just above 2-637 in Figure 5,
the tails caused by cosmic rays which are too faint to be caught by the cosmic ray removal program (e.g., a very faint artifact can still be seen in Figure 3 above trap
2-337), and
cases where only a single exposure is available.

The same region as shown in Figure 1, but after cosmic ray removal. The bright tails caused by the cosmic rays are now gone, since they only appear in one of the subexposures and are therefore removed along with the cosmic rays. The dark tail above 2-337 is still present, and can be removed via the wfixup or fixpix task. The bright tail in the object above trap 2-637 is still present, and will affect the photometric, astrometric, and size determinations.

The standard technique for removing the effects of traps (i.e., "bad columns") is to use the .c1h (data quality) image which is included on the data tape, and the STSDAS task wfixup or the IRAF task fixpix. At present, the correction begins at the location of the trap and replaces the data in the column with an interpolation from either side of the affected column, up to a value of Ystop (old) (see Table 1). While this provides a cosmetically cleaner image, it may occasionally affect your results, and hence may not always be recommended. For example, if a bad column falls precisely on the peak of a bright star, such as the object in column 2-637 of Figure 1, the central peak, which should be about 254 counts, is degraded to 123 counts. As will be discussed below, this can result in an error of several tenths of a magnitude when small aperture photometry is performed, and an increase of about a pixel in the FWHM.

Currently (before August, 1995), the bad pixels flagged by the .c1h file only cover the region immediately above the trap (i.e., are for the dark tails seen when sufficient background is present), except in the cases of 2-337 and 4-574 which are so bad, that they can generally be seen all the way to the top of the chip. However, any pixel clocked through the bad pixel during the readout is affected. These columns should be considered suspect all the way to the top of the chip. We are therefore planning to modify the .r0h reference files, and the corresponding .c1h data quality files that are sent to observers on their data tapes, to flag both the portion just above the trap with a value of 2 ("Defect") and the rest of the column above the trap with a value of 256 ("Questionable Pixel"). This is planned for August 1995, after wfixup has been modified to allow a switch to be set for the types of bad pixels to be modified.

Another problem with the .c1h image is that the length of the tail which is masked out is often much larger than actually required. We have therefore updated the .r0h and .c1h images, as described above. The only exception are traps 2-337 and 4-574, which will be flagged as 2 to the top of the column. We will also add the three new traps which have not been identified previously in the .r0h and .c1h images.

Since the tails follow a simple exponential decay, it should be a relatively straightforward task to reconstruct the original image using the measurements of a listed in Table 1.

The formula for this procedure is:

where C*J is the corrected number of counts in pixel J, CJ and CJ-1 are the observed number of counts in pixels J and J-1, T is the transfer efficiency from Table 1, and B is the background. Figure 6 shows the results for an object just above trap 4-441. (Note that trap 2-637 from Figure 1 and Figure 5 appears to be the only trap which leaks a small amount of charge in an "upstream" direction, thus we chose a more typical trap for this example.)

This is the result of using the reconstruction formula shown above on an object from an archival image just above trap 4-441. By chance, the center of the star was located at about column 441.5, hence the corrected profile for column 441 should match column 442.

The long tail is now gone and the central profile looks relatively similar to the unaffected column 442 profile. A potential problem with the procedure is the amplification of noise. This is most clearly seen by the behavior of the solid line from pixels 7 to 11.

[Previous] [Next] [Up]

Generated with CERN WebMaker