Space Telescope Science Institute
WFC3 Data Handbook 2.1 May 2011
Table of Contents Previous Next Index Print

WFC3 Data Handbook > Chapter 5: WFC3-UVIS Error Sources > 5.6 Generic Detector and Camera Properties

Conceptually, full well depths can be derived by analyzing images of a rich starfield taken at two significantly different exposure times, identifying bright but still unsaturated stars in the short exposure image, calculating which stars will saturate in the longer exposure and then simply recording the peak value reached for each star in electrons (using a gain that samples the full well depth, of course). In practice, as discussed in WFC3 ISR 2010-10, it is also necessary to correct for a ~10% “piling up” effect of higher values being reached at significant levels of over-saturation relative to the value at which saturation and bleeding to neighboring pixels in the column begins.
  Since the full well depth may vary over the CCDs, it is desired to have a rich star field observed at a gain that samples the full well depth (the default WFC3 UVIS gain does that), and for which a large number of stars saturate. Calibration and GO programs have serendipitously supplied the requisite data of rich fields observed at two different exposure times.
There is a real and significant large scale variation of the full well depth on the UVIS CCDs. The variation over the UVIS CCDs is from about 63,000 e to 72,000 e with a typical value of about 68,000 e. There is a significant offset between the two CCDs. The spatial variation may be seen in Figure 5 in the WFC3 ISR 2010-10.
Linearity at low and moderate exposure levels is explored by comparing counts in back-to-back exposures on NGC 1850. Figure 5.6 shows the response of one of the chips, where aperture sums for stars with flux greater than about 2,000 e in a short exposure (central pixel would be at greater than about 350 e) show apparently perfect linear response when compared to the counts in the same aperture in an exposure 50 times as long.  However, below a level of 2,000 e the ratio of long to short exposure counts deviates from a linear response.  By summed counts of 200 e in the short exposure these values are some 5% lower than expected based on scaling from the corresponding long exposure.  These data were acquired in October, 2009 some 5 months after launch of WFC3. The results will be published in a future ISR.
Figure 5.7 shows data from both UVIS CCD for stars yielding short exposure aperture sums of 500 - 2000 e.  A clear signature appears that is consistent with perfect linearity for stars near the readout amplifiers, with linearly growing losses in the short relative to long exposure with distance from the amplifiers.  This is consistent with losses induced by finite charge transfer efficiency in successive parallel shifts in clocking the charge packets off the CCDs.
The extent to which CTE losses are influencing faint object photometry will be an active part of the calibration program. These early results suggest that any intrinsic nonlinearities related merely to accumulated charge levels are small compared to minor-nonlinearities induced by CTE losses at low intensity.
Figure 5.6: The upper panel shows the counts in r = 3 pixel apertures for several thousand stars in back-to-back 500s and 10s exposures plotted against each other. To better illustrate small differences the lower panel shows the same data after binning and plotting as the ratio of long to short, which in this case would result in values of 50.0 for perfect linearity.
Figure 5.7: This shows the data from the previous figure for UVIS2, plus similar data for UVIS1 plotted against y-position in the concatenated detector space. y-values near 2050 correspond to the maximum distance from the readout amplifiers, and hence the most parallel shifts inducing CTE losses.
The response of the WFC3 UVIS CCDs remains linear not only up to, but well beyond, the point of saturation. WFC3 ISR 2010-10 shows the well behaved response of WFC3: electrons are clearly conserved after saturation -- in some locations with the need for a minor calibration, as provided in the ISR, in other regions no correction is needed. This result is similar to that of the STIS CCD (Gilliland et al., 1999, PASP, 111 1009-1020), the WFPC2 camera (Gilliland, R. L. 1994, ApJ, 435, L63-66), and ACS (ACS ISR 2004-01). It is possible to easily perform photometry on point sources that remain isolated simply by summing over all of the pixels into which the charge has bled.
Here the extent to which accurate photometry can be extracted for point sources in which one or more pixels have exceeded the physical full well depth is explored. Ideal data for these tests consist of multiple exposures taken back-to-back on a moderate-to-rich star field with a broad range of exposure times resulting in both unsaturated and saturated data for many stars.
Figure 5.8 and Figure 5.9 show results for UVIS. Over a range of nearly 7 magnitudes beyond saturation, photometry remains linear to ~ 1% after a simple calibration. For Amp C of UVIS2 the response is sufficiently linear beyond saturation that no correction is required.
All of the above linearity results are based upon comparisons of FLT images. The conservation of flux property of drizzle leads to equally good results for linearity beyond saturation comparing long and short drz images.
Figure 5.8: Analysis of linearity beyond saturation for Amp A on UVIS1. The upper panel shows the ratio of counts in the long exposure divided by the counts in the short exposure multiplied by the relative exposure time; linearity would thus result in a value of unity. The x-axis shows the multiplicative degree by which a star is over-saturated in the long exposure.  The middle panel shows the long exposure aperture sums versus short exposure aperture sums. The lower panel shows the peak data value in the long exposure relative to the short exposure value.  The response is linear up to the value of 68,000 e (y-axis) where the long exposure encounters saturation. Amp A data show significant deviations from a linear response for over-saturations near and beyond 10.
Figure 5.9: Upper panel shows data from the upper panel of Figure 5.8 for Amp A, plus similar data for Amp B.  The lower panel shows the same data after applying the corrections given in detail in WFC3 ISR 2010-10. Not only is the mean level appropriately restored independent of degree of over-saturation, but the star-to-star scatter is much reduced.
The WFC3-UVIS shutter is a circular, rotating blade divided into two open and two closed quadrants (See Section 2.3.3 of the Instrument Handbook for details). Operationally, the shutter mechanism has two distinct modes, based on commanded exposure times. At the shortest commanded exposure time of 0.5 seconds, the shutter motion is continuous during the exposure, rotating from the closed position through the open position and on to the next closed position. For commanded exposure times of 0.7 seconds and longer (0.6 seconds is not allowed), the shutter rotates into the open position, stops and waits for an appropriate amount of time, and then rotates to the closed position.
For short exposure times, detector position dependent exposure time (shutter shading), A versus B blade shutter dependence, stability, and timing accuracy were assessed using data taken during SMOV. For a full discussion of the analysis of shutter behavior from on-orbit data see WFC3 ISR 2009-25.
No systematic difference in shutter behavior (exposure time, repeatability, etc) is found when comparing the A and B blades of the shutter. Even at the shortest exposures, measured shutter shading does not exceed ~0.2% across the detector. The small magnitude of this effect means that no correction for shutter shading is necessary in calwf3.
Stability of shutter timing is a bit more problematic. Results are based on 11 pairs of back-to-back exposures at each commanded exposure time. For exposure times of 1.0 seconds or shorter, the rms variation in exposure time for a series of images is 1% or greater, implying possible trouble in achieving 1% photometric accuracy. For a commanded exposure time of 0.5 seconds, the rms variation is 1.9%. For commanded exposure times of 0.7 and 0.8 seconds, the true exposure times vary by 1.5% and 1.4% respectively. At an exposure time of 1.0 second, the rms variation falls to 1.0%.
While rms variations were all less than 2%, we observed individual exposures at each commanded exposure time that deviated by larger amounts. For the 0.5, 0.7, 0.8 and 1.0 second exposures, we found individual exposures with measured errors of 4.0%, 4.0%, 3.0%, and 2.0% respectively. This implies that exposures of 1.0 seconds or shorter may experience timing fluctuations that could compromise a goal of 1 or even 2.0% accuracy. This conclusion regarding shutter stability is not regarded as robust, but is offered as that most consistent with a simple and conservative interpretation of the test data.
Finally, our investigation of measured versus commanded exposure times indicated that for exposures commanded to be 0.5 seconds, the shutter was actually open for 0.48 seconds. Similarly, for exposures commanded to be 0.7 seconds, the measured exposure time was in fact 0.695 seconds. In order to support accurate photometry, we account for these exposure time differences by updating the header information in all 0.5 and 0.7 second exposures. For these data, the EXPTIME header keyword value is updated to the shorter, measured values during initial OPUS data processing.
At wavelengths longer than about 650 nm, silicon becomes transparent enough that multiple internal reflections in the UVIS detector can create patterns of constructive and destructive interference, or fringing. Fringing produces wood-grain patterns in response to narrow-band illumination at long wavelengths, see Figure 5.10.
Figure 5.10: Quadrant B of two ground flat fields: affected by fringing (right) and not affected (left). Black region of the FQ906N flat is masked to avoid areas affected by the quad filter edges; the same region of F673N is masked for consistency.
The amplitude and phase of the fringes is a strong function of the silicon detector layer thickness and the spectral energy distribution of the illumination. Fringe amplitude--the contrast between constructive and destructive interference--is greatest at the longest wavelengths (where the high transparency allows more internal reflections) and for the narrowest spectral energy distributions. For broad SEDs, interference is averaged over phase, so that the amplitude of the fringing is reduced. Thus fringing is significant for UVIS imaging data only if narrow-band red filters are used, or if sources with red line emission are observed.
Flat fields from ground tests (see WFC3 ISR 2008-46) have been used to estimate the magnitude of fringing effects, for a continuum light source, in the narrow-band red filters (see Table 5.4 and WFC3 ISR-2010-04). Each column lists a different metric of fringe amplitude, for a control filter (F606W) and for the filters in which fringing effects could be detected in the flat-field data. These metrics can best be understood by examining the histograms (Figure 5.11) of the flat fields shown in Figure 5.10.
Table 5.4: Metrics of fringe amplitude based on ground flat fields. Values are given in units of percentage of the normalized flat-field signal level. Each metric is described in the text and graphically represented in Figure 5.11.
Full Width at 20% maximum
Distance between histogram peaks (percent)
Figure 5.11: Histograms of the two flat-field samples shown in Figure 5.10. Symbols correspond to fringe amplitude metrics listed in Table 5.4: rms deviation (triangles with error bars), full width at 20% maximum (circles with error bars), and bimodal histogram peaks (squares).
The first data column in the table is simply the root mean square deviation from the mean of the sample, and is indicated by triangles with horizontal error bars in the histograms. Filters/quadrants with rms deviations greater than corresponding values for the control filter (F606W) may be influenced by fringing. The second column is full width at 20% maximum, rather than full width at 50% maximum, because this metric is more effective for bimodal pixel brightness distributions in filters with strong fringing, such as FQ906N (pictured). The third data column gives the separation between histogram peaks, which can be detected in flat-field data for only the five reddest of the twelve filters affected by fringing. Squares in Figure 5.11 mark the histogram peaks. Adjacent fringes were also manually sampled, and the results reported in the final data column.
For estimating photometric uncertainties in data taken with these filters, it may be useful to consider how much the rms deviation in the flat fields departs from that of the control filter F606W. For exposure time calculations of targets affected by fringing, the full width at 20% maximum or the distance between histogram peaks may be the most useful metric. To understand how small dithers might affect photometry of targets that happen to fall on adjacent positive and negative fringes, the manually determined amplitude might be most appropriate. For sources with SEDs very similar to the calibration lamp, application of the pipeline flat fields should considerably reduce the effect of fringing on the data.
Eventually, tools will be created so that users of UVIS data will be able to generate “fringe flats” for any combination of source SED and long-wavelength filter. Monochromatic ground test data have been used to create thickness maps of the UVIS detector (WFC3 ISR 2010-05), and these maps can be used to model the expected fringing response to an arbitrary SED. On orbit calibration data (Programs 11922 and 12091) are being taken in 2010 with the narrow band filters listed in Table 5.4, and these data will be analyzed to evaluate the fringe model solutions and thickness maps.

Table of Contents Previous Next Index Print