The noise test data are comprised by the sequence of images 349821.fits to 349867.fits, where a series of pairs exposures with ascending and then descending integration times were obtained, with 9 bias images interspersed through the test (see Data Collected). All these images are a 400x400 pixel window centered on the imaging area of the CCD. An LED light source, driven by the CCD controller shutter driver, was used to obtain the images.
A median stack of the 9 bias images was obtained. The resulting image is shown below. Note the rising left edge, which is an artifact of the settling time of the CCD readout electronics on beginning each row. The bias overscan is divided into two regimes - the first 20 pixels are the on-chip bias overscan pixels and the second 20 pixels are true bias overscan generated by reading beyond the physical edge of the CCD. The last on-chip overscan pixel shows anomolously high counts and should not be used in data reduction. The imaging area of the CCD shows slightly higher bias counts than the overscan region, the cause of this is unknown at this time (see Biases).
To perform the noise test, the above median-stacked bias image was subtracted from each of the images exposed to light. The average counts in each image, and one-half the variance of the difference of each pair of images was obtained. The window (104:345,99:344) on each image was used for the calculation. The results are shown in the table below:
| Exposure time (sec) | Mean counts (img1) | Mean counts (img2) | variance/2 |
| 2.50000 | 3378.44 | 3378.24 | 759.685 |
| 5.00000 | 6757.97 | 6757.43 | 1507.20 |
| 7.50000 | 10136.0 | 10136.3 | 2263.25 |
| 10.0000 | 13514.5 | 13514.2 | 2995.36 |
| 12.5000 | 16891.2 | 16890.7 | 3728.83 |
| 15.0000 | 20266.6 | 20266.4 | 4511.36 |
| 17.5000 | 23641.2 | 23641.6 | 5221.13 |
| 20.0000 | 27015.6 | 27015.1 | 5943.50 |
| 19.0000 | 25668.5 | 25667.4 | 5615.86 |
| 16.5000 | 22290.9 | 22291.2 | 4907.51 |
| 14.0000 | 18915.6 | 18914.7 | 4144.62 |
| 11.5000 | 15538.8 | 15538.6 | 3439.13 |
| 9.00000 | 12161.6 | 12161.9 | 2702.05 |
| 6.50000 | 8784.20 | 8783.79 | 1941.60 |
| 4.00000 | 5405.13 | 5405.86 | 1203.83 |
| 1.50000 | 2026.51 | 2026.80 | 455.312 |
The linearity curve is obtained by plotting the mean counts versus the fractional count rate and the result is shown in the figure below. Note that the maximum deviation from linearity (a horizontal straight line) is ~0.04%. The scatter around the polynomial fits shown is due to slight variations in the light source as can been seen by the systematic difference between ascending and descending branches of the test.
The transfer curve is obtained by plotting the mean counts versus the (flat-fielded) variance of the counts in an image. This is shown in the figure below:
The straight line fit has an inverse slope equal to the conversion
factor for this CCD of 4.56 electrons/ADU. The standard deviation
of counts (readout noise in ADU) in a single bias image is 1.9
ADU, which yields a read noise of 8.7 electrons/pixel for
STIS2
Using image 349841.fits (an unsaturated image with high counts), the number of times each bit was set and not set was counted. The result is shown in the plot below:
The dotted and solid lines are the number of times the a bit has
value 1 and 0 respectively. Note the slight bias to set bits
in the lower significant bits. The statistics at the 9th
bit and higher are insufficient to draw any reliable conclusions.