Jan 26, 2014: this dark current measurement is incorrect: it did not take into account variations in the overscan level. See Tech Note 6 for a better measurement of the dark curernt.
This report describes the results of my analysis of some of the Oct 2013 commissioning data. It is not the complete report, but a short note on results so far.
I'll show measurements of readout noise in units of ADU below. Recall that the gain for all 4 amplifiers was measured to be about 1.3 electrons per ADU.
To measure the readout noise, I followed this procedure:
I started with a 4-amplifier dataset, taken UT Oct 19: images c6584 429 - 449. For this set, I find
Quadrant readnoise (ADU)
----------------------------------------------------
1 5.8
2 5.4
3 6.0
4 5.8
----------------------------------------------------
Next, a 4-amp dataset taken UT Oct 21, c6586 004 - 013 .
Quadrant readnoise (ADU)
----------------------------------------------------
1 7.1
2 6.2
3 7.0
4 6.3
----------------------------------------------------
A final 4-amp dataset taken UT Oct 28, c6593 010 - 018 .
Quadrant readnoise (ADU)
----------------------------------------------------
1 8.3
2 7.3
3 8.7
4 7.6
----------------------------------------------------
There appears to be a trend of increasing readout noise with time.
To measure the readout noise in 1-amplifier mode, I used a dataset from UT Nov 2, c6598 230 - 240 . The amplifier used was the "upper right" amplifier. I again subtracted pairs of bias images and then computed the RMS of pixel values within small (100-by-100 pixel) regions.
1-amp readnoise (ADU)
----------------------------------------------------
ur = quad 1 7.4
----------------------------------------------------
The results are shown in the table below. To convert from ADU to electrons, I assumed a gain of 1.3 electrons per ADU.
readnoise(ADU) readnoise(e-)
------------------------------------------------------
1-amp mode 7.4 9.6
4-amp mode
quad 1 5.8, 7.1, 8.3 7.5, 9.2, 10.8
quad 2 5.4, 6.2, 7.3 7.0, 8.1, 9.5
quad 3 6.0, 7.0, 8.7 7.8, 9.1, 11.3
quad 4 5.8, 6.3, 7.6 7.5, 8.2, 9.9
------------------------------------------------------
In order to measure the dark current, I examined two sets of images, both taken in 4-amplifier mode. First, a set of bias frames -- equivalent to dark images of exposure time zero. Next, a set of dark frames with exposure time 1000 seconds. The long exposures were taken immediately after the short ones: UT Oct 19, c6584 429 - 453 .
First, I created a median bias image by taking the pixel-by-pixel median of 20 bias images. Second, I created a median 1000-second dark frame by taking the pixel-by-pixel median of three 1000-second dark frames.
I used two methods to analyze these two median images; we can compare the results in a moment.
Method 1: create histogram of all pixels in each quadrant, fit a gaussian to the histogram. This is a robust way to determine the "typical" pixel value in an image.
typical pixel value in
quadrant 0-sec dark 1000-sec dark difference
(ADU) (ADU) (ADU)
------------------------------------------------------
1 3066 3093 27
2 2895 2917 22
3 2909 2940 31
4 3090 3113 23
-------------------------------------------------------
Method 2: subtract the zero-second image from the 1000-second image. Compute mean of pixel values within 3 100-by-100 pixel boxes in each quadrant.
mean pixel value
quadrant difference image avg difference
(ADU) (ADU)
------------------------------------------------------
1 23, 21, 31 25
2 34, 21, 23 26
3 41, 27, 27 32
4 28, 20, 20 20
-----------------------------------------------------
If we use a gain of 1.3 electrons per ADU to convert these dark current values to electrons, and then divide by the 1000-second exposure time, we derive typical dark current rates of roughly 0.03 electrons/pixel/second.
Crosstalk refers to the phenomenon of a strong signal in one quadrant of the chip causing a fake signal to appear at the corresponding location within other quadrants of the chip.
In order to test for this effect, I pointed the telescope at Jupiter, exposing long enough to saturate the image of Jupiter -- but not so long that long bleed trails appeared. The dataset is UT Oct 20 c6585 225 - 236.
Jupiter was placed in the center of quadrant 1 (upper-right quadrant). A faint "echo" of Jupiter appeared near the center of each other quadrant. I examined the pixels in this "echo" and identified the pixel with the maximum value. In each case, there were only a few pixels with value close to this maximum -- the bulk of the pixels in each "echo" had considerably lower values. In no case did I see evidence for cosmic rays.
I made this measurement in two particular images: 229 and 231.
quadrant max pixel value in "echo"
above the background (ADU)
image 229 image 231
----------------------------------------------
2 29 29
3 66 71
4 70 71
----------------------------------------------
The strong source was in one of the "top" quadrants, but the crosstalk is stronger in the "bottom" half of the chip than in the "top" half.
If we assume that Jupiter was only just past the saturation limit, with pixel values just larger than 65536 ADU, then we can estimate the fraction of this signal which appears in the "echo" images. The background level was about 2940, 2950, 3130 ADU in quadrants 2, 3, 4, respectively. So, for example, in quadrant 2,
max pixel in "echo" above background
crosstalk fraction = -------------------------------------
max pixel in Jupiter above background
29
= ------------------
(65536 - 2940)
= 0.00046
= 0.046 percent
We find values of 0.046%, 0.11%, 0.11% for quadrants 2, 3, 4, respectively.