STIS Instrument Science Report 97-10 STIS Results from SMOV: CCD Baseline Performance Paul Goudfrooij, Terry Beck, Randy Kimble, and Jennifer A. Christensen May 1997 ABSTRACT We summarize the results from SMOV on the baseline performance of the STIS CCD. Read-out noise, dark current, charge transfer efficiency, and gain are measured. All CCD parameters turn out to be consistent with those measured during Ground Calibration.For in-flight unbinned data taken in CCDGAIN=1, the read-out noise is 3.78 +/- 0.05 electrons, the dark current is 0.0015 +/- 0.0003 electrons per second, and the gain is 0.97 +/0.02 electrons per ADU. 1. Introduction We collect data from the Servicing Mission Orbital Verification (SMOV) of STIS to measure the baseline performance of the CCD system on orbit. Data from the following SMOV proposals are used: 7061 (“CCD Functional”) and 7092 (“CCD Dark Rate and Read Noise”). The following quantities are measured: 1. The Read-out Noise (RON) in electrons, for all available CCDGAIN1 settings; 2. The Gain, defined as the number of electrons per Analog-to-Digital Unit (ADU, this is always the measured unit, sometimes also called Data Number (DN)), for all available CCDGAIN settings; 3. The average Bias level in ADU, its variation over time, and its global variation over the CCD, for all supported CCDGAIN settings; 4. The Dark Current in electrons (taken in the CCDGAIN=1 setting) 5. The Charge Transfer Efficiency (CTE). 1. In this document, words in COURIER font denote keywords in the FITS header of the images 1 2. Results 2.1. Gain and Read-out Noise During the CCD Functional Test (SMOV proposal 7061), two flat field exposures and two bias frames were taken at each CCDGAIN setting to provide a rough confirmation of the gain values measured during ground calibration. Binned data were also taken for CCDGAIN=1. The gain and read-out noise values were derived in the following way: 1. We created a “difference flat” and a “difference bias” for each CCDGAIN setting, e.g., flatdiff = flat1 - flat2 biasdiff = bias1 - bias2 2. Then the gain and read-out noise are measured as follows: gain = ( (mean(flat1) + mean(flat2) - mean(bias1) - mean(bias2) ) / ( (sigma(flatdiff))2 - (sigma(biasdiff))2 ) read-out noise = gain * sigma(biasdiff) / sqrt(2) where the gain is given in electrons per ADU and the read-out noise in electrons. Pairs of bias frames and flat field frames are used to render the effects of non-flat bias frames and/or flat field frames negligible. Only regions free of cosmic rays or dust spots were used in the gain calibration. The results are listed below in Table 1. Table 1. Gain and Read-out Noise. The Ground Calibration measurements (column 3) are taken from GSFC Shift Report #90 (R.S. Hill, March 10, 1997). CCDGAIN setting Binning Gain measured on ground Gain measured Read-out Noise [ADU] Read-out Noise [electrons] 1 1x1 0.994 +/- 0.008 0.97 +/- 0.02 3.91 +/- 0.05 3.78 +/- 0.05 1 1x2 0.995 +/- 0.005 0.98 +/- 0.02 4.26 +/- 0.04 4.18 +/- 0.05 1 2x1 0.995 +/- 0.013 1.00 +/- 0.01 3.65 +/- 0.04 3.65 +/- 0.05 1 2x2 1.001 +/- 0.010 1.00 +/- 0.01 3.76 +/- 0.03 3.76 +/- 0.03 2 1x1 2.008 +/- 0.006 2.06 +/- 0.03 2.58 +/- 0.03 5.32 +/- 0.06 4 1x1 4.096 +/- 0.009 - 1.63 +/- 0.02 - 8 1x1 8.323 +/- 0.027 - 1.43 +/- 0.03 - The gain for settings CCDGAIN=4 and CCDGAIN=8 could not be measured from this in-flight dataset because of saturated flat fields (the flat field lamps turn out to have a 2 higher output than expected). We note that the measured gain values should not yet be regarded as final since they have been derived from only two flat field images per gain setting, which were rather highly illuminated (in the upper 10% of the dynamical range). However, all measured gain values are consistent with the values measured on the ground, and within 5% of the nominal, commanded gain. The gain test will be repeated every 6 months (with adjusted exposure times, and with additional binning settings 1x4 and 4x1). 2.2. Bias Level 2.2.1. Superbias Frames. To construct “superbias” frames, 31 bias frames were taken for each of the following gain & binning combinations: 1. CCDGAIN=1, no binning 2. CCDGAIN=2, no binning 3. CCDGAIN=4, no binning 4. CCDGAIN=8, no binning 5. CCDGAIN=1, BINAXIS1=1, BINAXIS2=2 (i.e., 1x2 binning) 6. CCDGAIN=1, BINAXIS1=2, BINAXIS2=1 (i.e., 2x1 binning) 7. CCDGAIN=1, BINAXIS2=2, BINAXIS2=2 (i.e., 2x2 binning) For the construction of the superbiases, we used the IRAF routine noao.imred.ccdred.zerocombine, with parameters combine=average, reject=minmax, nlow=0, nhigh=1. Overscan level subtraction was done using the IRAF routine noao.imred.ccdred.ccdproc, with parameters function=spline1, order=1, sample=‘‘*’’, low_reject=3., high_reject=3. The image sections used for the overscan subtraction (parameter biassec) and the image section of the final, trimmed image (parameter trimsec) are listed below for the different binning settings. The resulting superbiases were compared with superbiases created by means of the STScI CALSTIS pipeline routines CALSTIS1 and CALSTIS2. No significant differences were found. 3 Table 2. bias and trim image sections for ifferent binning settings. The image sections are given in IRAF format for “raw” (i.e., non-processed) frames. Columns 5 and 6 give the mean and standard deviation of the remaining signal of the bias frame after subtraction of the overscan. CCDGAIN Binning Overscan section (parameter biassec) Trimming section (parameter trimsec) Mean Signal RMS 1 1x1 [2:16,21:1044] [20:1043,21:1044] 0.2 1.0 1 1x2 [2:9,11:522] [12:1035,11:512] -0.3 1.3 1 2x1 [2:9,11:1034] [12:522,11:1034] 0.5 0.9 1 2x2 [2:9,11:522] [12:522,11:522] 0.5 1.2 2 1x1 [2:16,21:1044] [20:1043,21:1044] 3.0 1.3 4 1x1 [2:16,21:1044] [20:1043,21:1044] 1.5 0.8 8 1x1 [2:16,21:1044] [20:1043,21:1044] 0.7 0.4 The overscan sections were chosen by taking into account the following considerations: (1) avoid column #1 which always shows a somewhat elevated count level above the other columns in the (trailing) overscan region; (2) avoid the 3 columns which are nearest the sensitive region, since Charge Transfer (non-)Efficiency effects will tend to show up in those (see Section 2.4). 2.2.2. Time Evolution of Overscan Level. Using bias frames taken in CCDGAIN=1 and CCDGAIN=4 modes during SMOV proposals 7061 and 7092, the stability of the level of the trailing serial overscan region with time has been studied during the month of March, 1997. The results are depicted in Figures 1 and 2. The overscan level turns out to vary in time by about 8-10 electrons (full range). This is consistent with what was found during Ground Calibration, and is most probably principally related to variations in CCD temperature. 4 Figure 1: Overscan level vs. Modified Julian Date for Bias frames taken in CCDGAIN=1 during SMOV proposals 7061 and 7092. 5 Figure 2: As Figure 1, but now for bias frames taken in CCDGAIN=4. The filled symbols represent the overscan levels, while the open symbols represent the levels in the acquisition subarray used for target acquisitions. 2.2.3. Stability of Bias Pattern over Time. We have studied the stability of the global pattern of the bias level over time as follows. For all bias frames taken in CCDGAIN=1 during the month of March, 1997, we determined the bias level in 25 sub-areas of 80 x 80 pixels, uniformly distributed over the CCD. The bias value in the trailing serial overscan region was also measured. Linear regressions were performed between the levels of the sub-areas, after subtraction of the bias level in the overscan region. The RMS of the 300 independent regression fits is 0.13 +/- 0.04 ADU, i.e., any residual change of bias pattern over the CCD chip stays within that amount. A plot of all RMS values of the regression fits is shown in Figure 3. The bulk of the scatter is systematic, in the sense that the scatter is higher for regressions between levels in areas that are further apart in the AXIS1 direction (i.e. along rows). This is due to a 6 ramp in the structure of the bias level along rows (e.g., Goudfrooij & Walsh 1997, ISR 9709). Figure 3: A plot of all RMS errors of linear regressions between the bias levels in different areas on the CCD. The mean RMS is 0.13 +/- 0.04 ADU. 2.3. Dark Current To measure the darkcurrent, 37 dark frames taken in CCDGAIN=1 with an exposure time of 1800 s were averaged together using the STScI CALSTIS pipeline routines CALSTIS1 and CALSTIS2. Prior to cosmic ray rejection in CALSTIS2, the in-flight superbias image (cf. Section 2.2) was subtracted from each input dark frame. In Figs. 4 and 5, we show cross-cuts along rows and columns of the superdark image, respectively. Due to the large number of hot pixels on the superdark image, we only averaged 10 rows (resp. columns) for this exercise. The average darkcurrent (excluding hot pixels) turns out to be 0.0015 +/- 0.0003 electrons/s, which is slightly lower than that measured during Ground Calibration. The hot pixels and cosmic ray rates will be discussed in a separate, forthcoming Instrument Science Report. 7 Figure 4: The structure of the superdark image along rows. Ten columns of the superdark image were averaged together. Figure 5: As Figure 4, but along columns. Ten rows of the superdark image were averaged together. 8 2.4. Charge Transfer Efficiency Three sets of three flat field frames were taken to check the CTE of the CCD, and compare it to the pre-flight Ground Calibration results. The flat fields were taken using the tungsten lamp. The optical element was grating G750M and the aperture was the 52x2 slit. Each set of three flat fields had the same exposure time (and, thus, intensity level), and they were averaged together using the IRAF routine imred.ccdred.flatcombine using parameters combine=median, scale=mode, zero=none, weight=none, reject=crreject, mclip=yes, nkeep=1, rdnoise=4, gain=1., snoise=0., hsigma=3, blank=1.The flat field frames were interleaved with bias frames, which were averaged together using the IRAF routine imred.ccdred.zerocombine as before (cf. Section 2.2), and subtracted from the averaged flat fields. The effect of imperfect CTE is to slightly smooth out the sharp drop in charge as one moves from real (sensitive) pixels into the trailing overscan region. Unfortunately however, this edge response method cannot (generally) be used to determine the absolute value of the CTE. This is so because the sharp drop in charge encountered when moving from the sensitive part of the CCD into the overscan region is also affected by imperfect electronics, creating a slight dip in the charge recorded in the overscan region. The main purpose of the test reported here is to compare the sharpness of the edge with that of similar images taken during Ground Calibration, for which the CTE was determined properly (using the “Sparse Field Test”, which involves reading out narrow-slit measurements by two different amplifiers at opposite ends of the CCD. The CTE then follows from the relation of the ratio of slit intensity with the position of the slit on the CCD. This CTE test is however too difficult to pursue as in-flight programme, as offsetting positions of the narrow slits require real-time commanding)2. The results of the edge response test are depicted in Figures 1 - 3. The (~ 400) rows of the flat field images that are in between the occulting bars have been averaged together to produce the plots. Figure 1 shows the average intensities in the 19 rows of the trailing overscan region for the on-flight (solid line) and the pre-flight (dashed line) average flat field images with the highest intensity (~10,000 ADU/pix). Figure 2 and 3 are analogous to Figure 1, but for averaged flat field images with ~1000 ADU/pix and ~100 ADU/pix, respectively. The main result is that there is no significant change in CTE since the pre-flight data. The only notable difference is that the first overscan pixel of the highest-flux flat field image seems to have ~2 ADU more intensity than that of the corresponding pre-flight flat field image. As only the first overscan pixel shows this effect, it is difficult to quantify this 2. In the course of Cycle 7 calibrations, we will provide a more proper measure of the CTE by performing CCD imagery of a field in Omega Centauri (a globular cluster) at 2 telescope roll angles different by 180 degrees, so that the CTE will follow from the comparison of the measured fluxes of the stars in the two images. 9 result in terms of CTE, since the imperfect electronics cause an uncertainty in the intensity of the first overscan pixel, even with a CTE of 100%. The behavior of the overscan region in the on-flight flat field images with lower flux are indistinguishable from their pre-flight equivalents. For reference, the CTE measured on the pre-flight data was 0.999995 (E. Malamuth, STIS Analysis Report #55, Aug. 14, 1996), implying that only ~0.5% of the flux will be lost for an object at the top of the CCD. Figure 6: Trailing overscan region of highly illuminated flat fields (~10,000 ADU/pix). The solid line represents the on-flight flat field, and the dashed line represents the preflight flat field obtained during Ground Calibration. The sensitive region starts at pixel 20. 10 Figure 7: As Figure 1, but now for flat fields with an intensity level of ~ 1000 ADU/pix. Figure 8: As Figure 1, but now for flat fields with an intensity level of ~ 100 ADU/pix. 11 3. Conclusion The STIS SITe CCD is performing very well on orbit. Gain, Read-out noise, dark current and CTE are entirely according to specifications. Below we summarize all main results in Table 3. Table 3. Gain, Read-out Noise, and Dark Current of STIS CCD. CCDGAIN setting Binning Gain measured Read-out Noise [electrons] Dark Current [electrons/sec] 1 1x1 0.97 +/- 0.02 3.78 +/- 0.05 0..0015 +/- 0.0003 1 1x2 0.98 +/- 0.02 4.18 +/- 0.05 - 1 2x1 1.00 +/- 0.01 3.65 +/- 0.05 - 1 2x2 1.00 +/- 0.01 3.76 +/- 0.03 - 2 1x1 2.06 +/- 0.03 5.32 +/- 0.06 - 4 1x1 - - - 8 1x1 - - - 12