The ramp starting points on side B (left) of the 70 µm array exhibit a distinctive pattern in data from Campaigns F and G. The pattern seems to be related to crosstalk with the address lines for each readout, and is completely synchronous with the readout period for the array (i.e. the pattern does not change from read to read, or DCE to DCE). Readout 4,4 of the 70um array, which appeared to be dead in Campaign A, but appeared to be fine in Campaigns B - E, is also screwed up again. Now, however, a signal from some pixels IS visible after a cosmic ray hit, indicating that the problem is NOT an open output line, but instead a large DC offset to the data ramps.
Three data sets were analyzed:
The signal at the first read (DN) for the data ramps on side B of the 70µm
array (left 1/2) showed a strong pattern in campaign F, with some
ramps starting below the bottom rail of the A/D, and others starting
up to about 10000 DN above the bottom rail (nominal would be for all
to start ~3000 DN above the bottom rail). The pattern was strong,
distincitive, identical for all 16 readouts on that side, and was not
a function of time (i.e. read # w/in a DCE or DCE in an exposure).
Figure 1. DN at the first read for several reads in a 70 µm DCE in Campaign F. CSMM was in SED/160 Dark position. Image scaling is from -500 to 5000 DN. Readout 4,4 looks dead in these 4 images, but in later reads it did show a signal for pixels that were hit by a cosmic ray, bringing the signal on scale for the A/D.
Because the DN at the first read in a DCE was frequently below the
bottom rail of the A/D converter, it was difficult to estimate the
magnitude of the offset for some pixels. To address that, the fitted
slopes were used to compute the DN at time zero in the DCE (ramp
zero points). The zero points are correlated with the voltage on the
address lines for the readouts, as shown in the figure below.
Figure 2. Correlation of ramp zero point with signal on the address lines to the CRC-696 readouts. This correlation indicates some kind of cross-talk between the address line and the output.
The data ramps on Side B also displayed evidence of cross-talk in
response to signals generated at the array (as opposed to an applied
signal from the address lines). In particular, some cosmic rays were
seen to affect the output from pixels in separate readouts and
modules, but which were being read out nearly simultaneously.
Figure 3. Cross-talk of cosmic ray hits to pixels in 4 readouts on 4 different modules. See, for example, the plots for pixels [1,1], [2,1], [3,1], [4,1], [1,2], [3,7], [1,8], for example.
For campaign G the Voffset voltage was changed to try to move the ramp starting points up by about 5000 DN. The DAC was changed from the nominal value of 0xC8 to 0xA0 by changing the irs-mips_pconst ier run at startup of the CE. The change is implemented for all subsequent campaigns, and so will need to be changed and propagated if we find a way to mitigate the problem. Within campaign G the change to Voffset applied to all tasks up through mips-126, the 70um Voffset test, which returned the offsets to the nominal value at its termination. Because of that the data from 4 campaign G tasks, 111, 920, 917, and 992, were actually taken with the nominal Voffset, and not with the new offset...
The figures below show the DN at the first read in a dark DCE for
Campaign G and F. The character of the pattern of the offsets to
the data ramps is different, but retains some of the same character.
In particular, the contrast between columns 1 and 4 within a 4x8 readout
is lower. The change to Voffset did increase the values by
about 5000 DN, as desired.
Figure 4. DN at the first read in DCEs from Campaign G (left) and campaign F (right). The pattern changed and the contrast within the pattern is lower. Scaling for both images is between -500 DN and +8000 DN.
The data from the above images is plotted below, showing the shift
in ramp offset due to the change in the side-B Voffset.
Just the DN from Side-B are shown, with raw pixel # (i.e. from 0 - 512).
Figure 5.DN at first read in a dark DCE from Campaign F (right panel in Fig. 4).
Figure 6.DN at first read in a dark DCE from Campaign G (left panel in Fig. 4).
The figures below show the Y-intercept from the fitted ramps from a
stim DCE for Campaign G and F. As seen with the DN at the first read, the
character of the pattern of the offsets to the data ramps is different,
but retains some of the same character. The change to Voffset
did increase the values by about 5000 DN, as desired. Note that the pattern
visible on the A side (right 1/2) is normal, and caused by the variation
in effective exposure time for each pixel when the first read is done.
Figure 7. Y-intercept of fitted slopes to stim DCEs from Campaign G (left) and campaign F (right). The pattern changed and the contrast within the pattern is lower. Scaling for both images is between -5000 DN and +15000 DN.
The data from the above images is plotted below, showing the shift
in ramp offset due to the change in the side-B Voffset.
Just the DN from Side-B are shown, with raw pixel # (i.e. from 0 - 512).
Figure 8.Y-Intercept from fitted slope in a stim DCE from Campaign F (right panel in Fig. 7).
Figure 9.Y-Intercept from fitted slope in a stim DCE from Campaign G (left panel in Fig. 7).
After mips-126 was run in campaign G the 70um side-B Voffset was
reset to the same value as used in campaign F. However, the change did not
return the array behavior to what was seen in campaign F. The images below
(Figure 10) compare the DN at the first read for the two Voffset
values. Figure 11 shows plots of the same data. Note that even after
Voffset was returned to the campaign F value the pattern of the
offsets did not return to that seen in campaign F, and instead remained
the same seen before the change in campaign G. The effects are clearer
in the Y-Intercept figures (13-15) below.
Figure 10. DN at the first read in Campaign G. On the left a dark DCE taken with the patched value of Voffset (0xA0), On the right a low-backgground sky DCE taken with the same Voffset used in Campaign F. The pattern and its contrast is essentially unchanged by changing Voffset within Campaign G, although the average level does shift by about 5000 DN. Scaling for both images is between -500 DN and +8000 DN.
Figure 11.DN at first read in a dark DCE from Campaign G (left panel in Fig. 10).
Figure 12.DN at first read in a dark DCE from Campaign F (right panel in Fig. 10).
The figures below show the Y-intercept from DCEs taken before and after the Voffset change.
Figure 13. Y-intercept of fitted slopes to stim DCEs from Campaign G before (left) and after (right) the Voffset change. The pattern and its contrast are unchanged. Scaling for both images is between -5000 DN and +15000 DN.
The data from the above images is plotted below, showing the shift
in ramp offset due to the change in the side-B Voffset.
Just the DN from Side-B are shown, with raw pixel # (i.e. from 0 - 512).
Figure 14.Y-Intercept from fitted slope in a stim DCE taken before the Voffset change in Campaign G (left panel in Fig. 13).
Figure 15.Y-Intercept from fitted slope in a stim DCE taken after the Voffset change in Campaign G (right panel in Fig. 13).
The figures below show time series of the Y-intercepts calculated from the fitted slopes for stim DCEs taken in Campaign F (Fig. 16), and Campaign G beforeVoffset was changed back to the value it had in Campaign F (Fig. 17), and Campaign G after the Voffset change (Fig. 18). The data are ordered according to the order in which the pixels are addressed during readout of the array. I.E. the first sixteen points all have the same address bits within a readout and are the first 16 pixels sampled, etc. The variation of the Y-intercept offset with pixel address is clear, especially in the campaign F data (Fig. 16). Equally clear is that the pattern in G was quite different in F, indicating that the details of the effect are not stable and would be particularly difficult to deal with from a calibration standpoint. For example, this anomaly would make it nearly impossible to derive or use an electronic non-linearity correction on side-B. The repeatability of the pattern within campaign G (compare Fig. 17 and Fig. 18) is, well, eerie...
Figure 16. Time series of Y-intercept values (DN) from fitted slopes in Campaign F. The order of the data is the order in which the pixels are addressed as they are read-out. Blocks of 16 points correspond to a particular pixel address within each of the 16 readouts, i.e. each has the same address bits. As the pixel addressing changed, so did the offset of the data ramps.
Figure 17. Time series of Y-intercept values (DN) from fitted slopes in Campaign G. These data were taken at the initial Voffset value used in G, which added about 5000 DN to the ramps (as noted in the figure title). The pattern of the offsets in G is clearly quite different from that in F.
Figure 18. Time series of Y-intercept values (DN) from fitted slopes in Campaign G. These data were taken at the same Voffset as in Campaign F, i.e. after mips-126 was run. The character of the offset pattern is indistinguishable from earlier in Campaign G, and unlike the pattern in F.
