ACA pixel readout anomaly

Summary

In two recent observations, the centroids for fid light ACIS-I-6 show large (0.4 arcsec) step-function deviations at precisely the dither period of 1000 sec. These deviations are not due to a spoiler star dithering through the fid image, but instead appear because the image charge distribution within a single readout column changes as a function of the dither phase. This anomalous behavior is NOT observed in at least 6 other recent observations using ACIS-I-6 at the same position. For the two affected observations, good aspect was recovered by processing without fid ACIS-I-6.

In both affected observations, the discrete jumps in pixel values are precisely correlated with single-pixel shifts in row of the image readout window (due to dither) for one of the guide stars. There is a bad pixel at row,col = (-374,347). (The exact row is uncertain by up to 2 pixels for reasons explained below). When the position of the guide star window is such that pixels from this window are being digitized (and hence slowing down the parallel CCD clocking) just as the fid image is being clocked over the bad pixel, then the fid image is distorted.

Operationally, in the short term we will implement a check in starcheck to prevent having a fid or star which will slowly clock over the bad pixel. In parallel, we need to evaluate the impact on aspect solution accuracy in the normal case when the fid quickly clocks over the bad pixel. If there are noticable problems, we will be forced to de-prioritize ACIS-I-6 by reordering the ACIS-I fid light sets.

Centroids

The plot below shows the Y and Z angle centroids for fid ACIS-I-6 in Obsid 1576.



The large step-function deviations in the Y centroids are obvious; somewhat more subtle are the small positive perturbations in the Z angle centroids occurring each 1000 seconds. This signature does not resemble that of a spoiler star dithering over the fid light. Strangely, the signature looks like perturbations in a star centroid when a warm pixel dithers into the image readout box. However, in this case the fid light is stationary, and the image readout box is fixed during the entire 6000 second observation.

Integrated Counts

Another diagnostic of what could be occurring is the plot of total counts in the image as a function of time, as shown in the top panel of the plot below. Clearly there is no periodic deviation as would occur in the case of a spoiler star. The middle panel shows the sum of three pixels within one readout column near the peak intensity of the fid light image. Again, there is no significant time dependence. In the lower panel, however, we show the sum of just two pixels. Here there is a large periodic change in intensity. In the next section we will show on a pixel-by-pixel basis what is occurring.



Pixel values

Below is a plot showing the pixel readout value for a 4 column by 5 row array of pixels centered on the fid light image. The time has been folded by a 1000 second period, revealing a remarkable pattern in pixel intensity as a function of dither phase. As shown in the plots above, the charge is being redistributed.



From about 500 to 700 seconds, the pixel at (row,col) = (-74,347) is reduced by about ~1600 DN. In that same time range, (-73,347) is increased by ~800 and (-72,347) is increased by ~800. The net change is exactly zero, as if charge is getting trapped and the released into the next row as the CCD is being clocked out. The pixels in rows -73 and -72 also show secondary "trapping" before 500 seconds and after 700 seconds, with no corresponding disturbance in pixel (-74,347). The pixels in the previous column (346) show the same overall pattern, but at a very much lower level.

The other example of this phenomenon is shown below. The same three pixels are affected.



Correlation with guide star image row number

The exact times of the jumps in pixel readout values were compared to the times when the guide star image readout windows shifted in row. For both affected Obsids there was exactly one guide star in which the times matched precisely. For Obsid 1576 the star in slot 6 matched, and in Obsid 2365 the star in slot 5 matched. The distance (in rows) of both stars from the serial readout register is about 300 pixels, versus a row distance of about 438 pixels for the fid ACIS-I-6. Note that in one case (Obsid 1576) the problem star is actually on the opposite side of the CCD, but because the CCD clocking interlaces positive and negative rows, from a timing perspective there is no distinction.

The plot below shows the correlation between image row number for the problem guide star and ACIS-I-6 pixel value. The row number is plotted in red, with the row labeled on the right side of the plot. To remove sign dependence on positive/negative row numbers, the quantity actually plotted is the row distance of the leading edge of the star readout window from the serial register.



The same plots for Obsid 2365 are shown below.



Explanation as a single bad pixel

Normally, as the fid light image is clocked out toward the serial readout register, the parallel clocking speed is fast (24 microsec). However, if another 'upstream' image is currently being digitized, the row clocking period is at least 250 microsec. Empirically what we are seeing is that when the fid light image is slowly clocking over a certain bad pixel because of an upstream image, then charge is lost from the pixel and deposited in the next pixel, without any net loss of charge.

Comparing the plots for 1576 and 2365, there are noticable differences in the pattern of pixel shifts and the exact pixel leading edge distance. This is perplexing until one realizes that in the case of 1576, the problem begins as the leading edge of the guide star readout window shifts forward enough to slow down the clocking. In contrast, in 2365, the problem begins when the trailing edge of the star readout window shifts back enough slow down clocking. This accounts for both the different pattern and the offset by ~10 pixels in the leading-edge distance.

Going through the math for both Obsids, the position of the bad pixel must be near row,col=(-374,347). There is uncertainty of up to 2 pixels in row because of probable "off-by-one" errors, and uncertainty in the precise way the ACA clocks out and digitizes image windows. A dark current measurement should reveal exactly where the bad pixel is located. Previous dark calibrations have not shown at problem at this location.

Nature of the bad pixel

The physical nature of the bad pixel is not completely clear. The main sticking point is that there is apparently never any loss of charge in the fid light image, only redistribution. The net counts for fid ACIS-I-6 is seen to be constant to within ~100 counts, both within the affected Obsids and in non-affected obsids. That appears to rule out the obvious explanation of a charge trap which fills quickly and then releases charge with a timescale of ~100-200 microsec. If this were the case, then in the (normal) fast-clocking case, significant charge would be lost out the back end of the image. In the slow-clocking case, that charge would be released to show the charge tail which we observe, but would result in a net charge gain relative to the normal case.

Having a relatively long filling time constant helps a bit, but would still result in a net charge loss in the case of a trailing-edge slowdown (e.g. obsid 2365).

The other information in the puzzle is that there appears to be a threshold in the charge trapping. The fid image pixel in column 347 with intensity ~2000 DN never suffers from charge loss, despite being clocked slowly over the bad pixel.

Additional examples

Another case of fid ACIS-I-6 being disturbed due to this bad pixel has been observed in Obsid 2835 (2001:399). There is a guide star at Y_ang = -998 which is presumably the culprit. In obsid 1576, a star at Y_ang = -997 caused the trouble.



Aspect Information main page



Comments or questions: Tom Aldcroft

Last modified:12/11/01