NOTCam Array Electronics upgrade Dec 2007

The purpose of upgrading the array electronics was to get rid of the "dark current" problem, i.e. the fact that the measured dark counts are much higher than expected, and the strange behaviour where the dark images have less counts the longer(!) the integration time is ...

The electronic upgrade consisted in exchanging the printed circuit board (PCB) to which the array is connected with an upgraded PCB with recommended voltage settings (from Rockwell) for the Hawaii arrays. The main difference between the old and the new electronic configuration is that while the old setup had no "biasgate" voltages set, the new setup allows the user to adjust these around a default recommended value of 3.5V.

On the 3rd of December 2007 the new PCB was tested out with the engineering grade array cold. The QC for the ramp-sampling mode was run for two different values of the biasgate voltage (default=3.5V and max=3.8V), but it was not possible to do further tests because of a serious problem with the vacuum of the cryostat. (Also, it was not possible to have the QC analysis updating the web because of the web server swap at these dates. We can re-run this and get the plots on the web.)

No major difference was seen between the old and the new PCB from the QC results. The gain was around 3.2 e-/adu and the ron was at 13 e- for the ramp-sampling mode. The 1% linearity limit was at 17000-24000 adu. Also, no obvious undesirable effects were found with the new PCB. But apparently the upgrade did not cure the dark current problem: Still the the measured counts on darks were too high and the behaviour vs. exposure time as before, i.e. many counts on short darks and less counts on longer darks.

In the following we describe the differences between the old and the new PCB as seen on the engineering grade array.

We also compare the behaviour of the new science array with the old arrays.

Darks

• The vertical cuts show a high intensity ramp in the first readout lines per quadrant. This effect, called the "dc-gradient", has its origin in the "reset anomaly", the fact that the integration ramp has a nonlinear initial part after a "reset" of the detector.
  
  
• For the old PCB the ramp starts at 300 adu and levels off at about 50 adu on a "dark 0", while for the new PCB it starts at 600 adu and levels off at 50.
  
  
• The vertical cut shows the same type of profile in every 512 pixel quadrant. Apart from the steep ramp from pixel Y=1, there are also smaller jumps at Y = 156, 264, 304, 337 and 460. The jump positions are exactly the same for each quadrant. The jump positions do not change with the new PCB. These are probably inherent to the array (the readout multiplexer).
  
  
• There is no apparent difference in the vertical profile for the two different values of the biasgate voltage for the new PCB.
  
  

Reset images

• The faint broad bands, tilted about 20 degrees, are the interference features that are believed to be due to pick-up noise.
  
  
• We notice the presence of narrow horizontal stripes of alternatingly higher and lower intensity. The stripes are stronger with the new PCB (peak-to-peak amplitude 11%) than with the old PCB (peak to peak amplitude 7%).
  
  
  1. The stripes normally subtract out well, and there is no presence of such stripes on the reset-subtracted images. For low-background cases, such as darks, there are two exceptions, however: 1) where there is strong interference pattern (due to pick-up noise?), and 2) along the first few columns of each quadrant. Here the stripes don't cancel well.
     
     
  2. The stripes are relatively constant in time. The ratio of one reset image with respect to the subsequent reset image makes the stripes go away and the typical peak-to-peak noise amplitude is the same vertically as horizontally: about 0.3 to 0.4 %.
  
  
  

The reset level

• The "dc offset" voltages were adjusted for each of the quadrants of the new science array when commissioned 13/12-2007 to give around 5000 adu on the reset image. On the first night after the vacuum improvements, 16/4-2008, the average level was found to have dropped to around 3500. The "dc offset" voltages were increased by 0.02 V in each quadrant on 21/5-2008 to reach a count level of around 6000 adu.
  
  
• The reset level is found to drop to very low counts on several occasions throughout a night. If the level is very low it gives rise to "crap" images similar to the very first images obtained after the array has been passive. A typical feature is that each quadrant has a horizontal gradiant, and that there is a distinct jump between the right and the left quadrants.
  
  
• A particularly bad situation was seen on 13/11-2008 during a daytime domeflat session during about 3 hours. The domeflats that have a very low reset level are useless.
  
  
• Comparison with old PCB and old arrays shows that these instantaneous drops also did occur earlier, but more seldomly or with smaller amplitude. See plots for 14/1-2006, 28/8-2006, 29/4-2007, 4/5-2007, and 16/11-2007.
  
  

Skyflats - comparison between the 3 arrays

• For the new science array (SWIR3) we note a pronounced structure outlining quadrant borders, especially the horizontal border between lower and upper quadrants. The above image is from December 2007. By April 2008 this effect is even stronger.
  
  
• While the differential skyflat procedure (short exposures exp 5) used to work well for the two previous arrays, it is not a useful strategy for the new array. The jump across lower and upper quadrants has nothing to do with the detector flat field. It is an effect related to the "reset anomaly" which is now stronger with the new PCB. Because the "reset anomaly" has to do with the detector's stabilization to a given intensity level, it does not subtract out well on differential flats (bright - faint).

Skyflats - SWIR3

• The plot shows all columns [1:1024] averaged. The dip around line 512 is the effect mentioned above, the remnant of the "dc-gradient" visible as a horizontal bar of lower brightness.
  
  
• On the commissioning night 13/12 2007 this effect was relatively modest (a dip of about 2%) and did not catch our attention. The feature was discovered first on the next NOTCam tech night in April 2008. The dip was then around 5%. Later is was realized that the difference between bright and faint was exceptionally low on the 13/12 (due to a late dome opening caused by high humidity). On the 17/4, the difference was typical with bright around 15000-20000 adu and faint around 2000 adu.
  
  

Domeflats

• The lamp ON images had 17000 adu and the lamp OFF images 2400 adu.
  
  
• Flat fields obtained with multiple reads using the ramp-sampling mode are better than flat fields obtained with single reset-read-read exposures in the sense that the structure of the 4 quadrants - which is related to the "reset-anomaly" and has nothing to do with the flat field - slowly disappears the larger the value of N in the command frames t N is.
  
  
• It is clear that differential flats, even when taken with multiple reads, don't work well. The reason is that the OFF images are of a much lower intensity level and therefore the ramp behaviour is different ("reset-anomaly") and does not subtract out well. In fact, the subtraction adds structure to the flat. See the ratio image.

Conclusions

A consequence of this is that differential flats don't work well any longer for NOTCam. Due to the "reset anomaly" the "dc-gradient" is different for different intensity levels, and therefore the "bright minus faint" subtraction leaves traces (most notably a horizontal bar of lower intensity across the centre of the array).

It should be noted that the reset anomaly for the NICS/TNG instrument is quite similar and makes it impossible for them to obtain accurate differential flats.

Multiple reads gets rid of the "dc-gradient", but multiple reads require of the order of 15-20 seconds exptime, not feasible on the twilight sky. Domeflats are considered, but from earlier experience we know that they are not optimal. This is still under study. Another option is to try to get rid of the remnants of the "dc-gradient" in the differential flats by collapsing a median column and subtract it.

Results of "dc-gradient" suppression in the mkflat.cl script

Differential twilight flat for WF camera and Ks band with "dc-gradient suppression" (left) and without (right).

Comments to Amanda