-Henry C. Ferguson 5/20/96
The WFPC2 CCD's have a small but significant charge-transfer efficiency (CTE) problem which causes some signal to be lost when charge is transferred down the chip during readout. This has the effect of making objects at higher row numbers (more charge transfers) appear fainter than they would be if they were at low row numbers.
Because the effect depends on the number and location of traps inside the individual potential wells, it probably depends on background levels (which cause some filling of the wells, and therefore less charge trapping) and on the brightness of the stars. Observations of a field in Omega Cen were taken in Cycle 5 to try to calibrate these dependencies. The data are not yet completely analyzed, but it was deemed useful to put some preliminary results on the Web. A more complete analysis will be completed within the next month.
The bottom line so far is as follows:
- The CTE effect seems to decrease linearly with the log of the background.
- At 160e- preflash it is likely (but not proven) that the variation of measured flux with row number is negligible (less than 1%).
- Preflash has less than 2% effect for stars with more than 2500 DN.
- For stars with less than 250 DN, adding 160e- preflash can brighten the measured magnitudes by more than 0.05.
- There is a dependence on row number. Preflash makes little difference (less than 3%) at row numbers less than 200, but makes up to 5\% difference at high row numbers.
- There is marginal evidence for a change in PSF as a function of preflash level.
- With low background and low stellar flux (less than 10 DN), photometric errors can be as large as 0.2 mag.
Now for the details...
The crowded Omega Cen field was observed for 40 sec through F555W with gain 7, each with a preflash of 0, 5, 10, 20, 40, 80 and 160 electrons. As a gain check and calibration, it was observed at the same orientation with gain 15 twice at preflash levels of 0 and 160 electrons. The sequence was repeated with filter F814W. The first sequence was repeated again for F555W but began with a VISFLAT observation. Next, a whole orbit was filled with 1 second exposures in order to investigate the effect of CTE on low signal level stars (but with high accumulated signal to noise). This orbit was done again (using F555W) but each subsequent image was shifted in increments of 5,5 PC pixels. The last orbit was the same sequence of observations, but each external exposure was preceded by a preflash of 40 electrons.
So far, the 40 sec, gain 7 F555W exposures and the 1 second F555W exposures with no preflash have been analyzed.
Analysis of the 40 Second Data:
There were 7 gain 7 F555W images were taken, with preflash levels of 0, 5, 10, 20, 40, 80, and 160 electrons. There was only one exposure per preflash level and there were no shifts between exposures. Cosmic rays in the individual frames were identified by comparison to the combined image. A special task was written to produce a CR mask image for a single frame based on comparison to a clean reference frame with a different background level. Stars with cosmic rays in the photometry aperture are not used in the plots. The star aperture used was 1.5 pixel in radius.
The plots show the difference mag minus mag (max preflash) as a function of background (preflash) level. There is a clear indication of a CTE effect in the full-frame data in the following figures: PC WF2 WF3 WF4.
For WF 4, I fainter stars were included, which explains the increased scatter.
For: WF2 WF3 and WF4 we can see the effect of preflash as a function of y position on the chip. The plots suggest that for low row numbers the magnitudes are essentially independent of preflash, while at high row numbers, magnitudes are fainter at low preflash levels. This is the expected behavior for the CTE effect.
These figures show the effect of preflash as a function of brightness: PC WF2 WF3 WF4. Brighter stars are virtually uneffected by preflash, while fainter stars get brighter as the preflash level increases. The magnitude scale was set arbitrarily here, such that a star of mag=25 has 100 DN in a 40 sec exposure.
The relatively smooth change in the magnitude residuals as a function of preflash levels indicates that a short preflash will not do much to cure the CTE ramp effect. However, if the ramp effect is now under 0.05 mag for relatively faint stars (the amplitude of the ramp has not been measured directly from this data set) a preflash of 80 to 160 electrons probably could reduce the ramp effect to under 0.02 mag. This might worth be considering for calibration observations. The idea here is that most calibrations are short exposures and have low background, while most science observations are long exposures and have higher background. Hence it might be useful to calibrate with preflash, and use a ramp correction to derive photometric corrections for the science observations. However, given that the preflash makes essentially no difference for stars with more than 2500 DN, the only calibrations to benefit would be those with relatively few counts in the calibration sources.
It is likely that the wings of the PSF are being eaten away by the CTE effect more than the cores, since a larger fraction of electrons will be trapped at low count levels. If so, then we might expect the wings to be suppressed in exposures with low preflash levels. Thus, for example, exposures with low preflash should show less change between apertures of 3 and 4 pixel radius than exposures with high preflash levels. The effect seen in the following figures is small, but in the sense expected. The figures show the magnitude differences between different aperture sizes as a function of preflash level: PC WF2 WF3 WF4.
Analysis of 1 Second Data (WF4 only):
Difference in measured magnitude from the sum of 21 frames measured with 1 second exosure with no preflash, minus the magnitude measured in a frame with 40 seconds exposure with 160e- preflash, for WF4. Measured values in the 1 second frames are typically 0.1 mag fainter than in the 40-second frame with preflash.
Same as a function of row number. There is a clear trend for stars at high row number to have a large offset -- up to 0.2 mag at the top of the chip. There is only a few percent loss in flux at low row numbers.
Same, but for brighter stars. The magnitude of the effect is perhaps slightly smaller, but still nearly 0.2 mag at the top of the chip.
Same, but for fainter stars. S/N is too poor to tell if there is a difference.
Growth curve for 40 second + 160e- preflash data (open circles) and 1-second no-preflash data (filled circles), for stars near the bottom of the chip. There is no obvious difference.
Same near the top of the chip. Once again there is no clear difference.
-Harry Ferguson (firstname.lastname@example.org) 5/20/96