AUTHORS : M. Sirianni, A.R. Martel and G. Hartig.
Our main goal is to determine the correct and most effective method of subtracting the bias level from WFC Build#4 data frames. We will define the overscan regions to use for the offset determination and correlate the physical and virtual overscan levels and their residuals with the data count rates. A similar analysis was performed for build 3 (see WFC#3 Bias Subtraction and Overscan Analysis). The bias stability will be treated in a later report using analyzing more data from the Thermal Vacuum #3.
The bias frames and flat fields were acquired with SMS procedure JGCW32A as part of the photon transfer test for the WFC at gain=1 and offset=3. The internal tungsten lamp T2 provided the illumination and the WFC Build#4 (amps ABCD) recorded the images. The data consist of a sequence of flat field pairs acquired over a wide range of signal levels. A pair of bias frames was acquired at the beginning and at the end of the sequence. During the acquisition of the second bias pair, the T2 lamp was unintentionally left on. The following table lists all the data used in the analysis.
TABLE 1: Images used for the bias analysis.
|ID||t exp (sec)||Exposure Type||Gain/Offset||Shutter Blade||Notes|
The column "Shutter Blade" lists which side of the shutter was in
front of the detector during the readout. This is determined from the
keyword WSHUTPOS in the FITS header, which is populated with the first
engineering snapshot. Its value represents the shutter resolver
position at the beginning of the exposure. The shutter side changes
for all exposures of type INTERNAL or EXTERNAL but not for BIAS or
DARK exposures. Each shutter blade has two open and two closed
positions, rotating a half turn between two consecutive exposures.
The convention for the shutter side designation is :
Side A, WSHUTPOS : 9984 +/- 160 or 42752 +/- 160
Side B, WSHUTPOS : 26384 +/- 160 or 59152 +/- 160
1. DESCRIPTION OF THE BIAS FRAME
In this section we will use as prototype of a bias frame the median of the two bias frames acquired with the lamp off. We will discuss the structure of the bias frame for each quadrant of the WFC focal plane. The default readout configuration on orbit will be a four-amp readout. In this report we will identify each quadrant with the proper amplifier : Amps A and B for Chip 1, C and D for Chip 2. A detailed description of the WFC format is available at Introduction : WFC Format. The dimensions of each quadrant are 2072 columns by 2068 rows. In the following discussion, each of the four quadrants will have the same orientation with the amplifier in the lower left corner. Each quadrant has a 24-pixel wide "leading" physical overscan regions at columns 1-24. Due to the particular readout, no "trailing" physical overscan is present. Each read-out is also accompanied by a virtual overscan of 20 rows at rows 2049-2068.
|Leading physical overscan||1-24||1-2048|
|Active image area||25-2072||1-2048|
1.1 Leading Physical Overscan
In each quadrant the leading physical overscan is the region between columns 1:24 and rows 1:2048. All four amplifiers produce an horizontal ramp in the leading overscan which extends up to 18 columns toward the image area. The amplitude of the ramp is different for each amplifier; it is larger in the A/D quadrants than in the B/C quadrants. The presence of this ramp does not allow an optimal determination of the bias level in the physical overscan. However, the perturbation associated with the ramp fades gradually and disappears approximatively within 15 columns. Therefore, in order to determine the bias value in the physical overscan we first created a median profile of the region (columns 1:24) along rows 1:2048. We then used the median value of the last six columns (19:24) of the resulting 1-dimensional array to set the bias level. The next two figures show the median profile across the leading physical overscan where the signal has been normalized to the bias level. The maximum amplitude of the ramp is ~4 % in the A/D quadrants and ~0.8% in the B/C quadrants. The figures also show that the region used to evaluate the bias level is very flat (better than 0.1%).
|Profile across leading physical overscan|
|FIGURE 1||AMP A/D|
|FIGURE 2||AMP B/C|
Note : All the figures in this report have been created with IDL so the first column has number 0.
We then investigate the vertical profile of the leading physical overscan. We created the median profile of the last six columns of the overscan, the same columns we used above for the offset determination and which are not contaminated by the ramp perturbation. As in the previous case we normalized the resulting profile to the offset level. For all four amplifiers, the vertical profile is extremely flat (within 0.2%, see Figs 3-4).
|Profile along cols 19:24 of the leading physical overscan|
|FIGURE 3||AMP A/D|
|FIGURE 4||AMP B/C|
In order to verify if the bias level determined from the leading physical overscan is representative of the bias level in the image area, we compared the signal in the central region of each quadrant [500:1500,500:1500] with the bias level in the physical overscan. In three cases, amps A,B and C, the image area has a bias level higher than the the physical overscan. The offset varies for each amplifier and can be as large as 3.5 DN. The causes of this offset are unknown. A detailed analysis will be performed with a larger number of bias frames taken during the July 2001 Thermal-Vacuum campaign at GSFC (a preliminary analysis shows that such an offset is fairly constant within 0.2-0.3 DN). Figs 5-6 show an averaged profile which includes the leading physical overscan and 500 columns in the image area.
|Profile across cols 0:500|
|FIGURE 5||AMP A/D|
|FIGURE 6||AMP B/C|
1.2 Virtual Overscan
The virtual overscan consists of 20 rows over-clocked after the readout of row 2048. For each amplifier the virtual overscan is very uniform and it represent a natural extension of the image area. In order to determine the bias value of the virtual overscan we first created a median profile of the overscan (rows 2049:2068) along columns 1:2048. We then used the median value of the last five rows (16:20) of the resulting 20 pixels long array to set the bias level. There is no offset between the bias determined from the virtual overscan and the image area. The next four figures show the median profiles across and along the virtual overscan. As in the previous plots the signal has been normalized to the bias level.
|Profile across virtual overscan|
|FIGURE 6||AMP A/D|
|FIGURE 7||AMP B/C|
|Profile along virtual overscan|
|FIGURE 8||AMP A/D|
|FIGURE 9||AMP B/C|
1.3 Virtual Overscan or Physical Overscan ?
The physical or virtual overscan can be used to determine the offset to apply to each image before correction of the bias pattern. Which of the two regions gives the better representation of the bias offset in the image area ? The first possible check is to compare the median bias level in the three different regions :
The differences are small but the physical overscan seems to be systematically lower than the image area and the virtual overscan slightly higher than the image area and hence higher than the physical overscan. These offsets can be due to a pattern in the image area which makes the bias array non-uniform. To verify this hypothesis, we created a mean profile of the columns and rows of the bias frame and compared the residuals after the subtraction of the offset level from the physical overscan and the virtual overscan. The following table contains the images for each quadrant and the comparison of the residuals along the rows (X-DIR) and along the column (Y-DIR) for each amplifier. All the images have the same orientation, with the amplifier in the lower left corner, the physical overscan in the left side and the virtual overscan at the top.
A close look at the images will show a small gradient in the horizontal direction for all four amplifiers. There are no gradients in the vertical direction. In particular, amps A and C have a region (between 300 and 500 columns wide) next to the leading physical overscan with higher signal than the remaining regions. These ramps are more visible in the plots in the central column of the previous table. These plots also show that using the bias level either from the virtual or the physical overscan results in very small residuals, less than 0.1% and 0.25% respectively. The plots in the last column show the residuals in the vertical direction, and indicate that the bias level is very well represented by the value from the virtual overscan. From this analysis alone, we could conclude that the bias offset from the virtual overscan seems to be a better representation of the bias level of the image area and should be taken as the region of reference for the offset calculation. However, a previous report (WFC#3 Bias Subtraction and Overscan Analysis) showed that the virtual overscan in internal flat field frames is not as flat as in the bias frame. The first rows of the virtual overscan can be affected by parallel CTE effects or light contamination. In the following sections, we will show that this problem is also present in WFC#4 and that the virtual overscan is therefore not suitable for the offset determination.
2. OVERSCAN LIGHT CONTAMINATION
In the following sections, we will compare the physical and virtual overscans in a bias frame in low and high signal level flat fields. In particular, in this section we will compare the features in the following frames :
We want to test whether the exposure to a light source changes the structure of the physical and virtual overscan, possibly from light leaks, CTE effects, or other anomalies.
2.1 Leading Physical Overscan
As a first step we checked if the ramp in the leading physical overscan is similar in the three frames. We created the median profile of the physical overscan and normalized the result to the value of the bias from the last 6 columns, in the same way we did for the bias frame in section 1.1. Figs 10-13 show the leading ramp in the three images we used in this test. In the case of the A and D amplifiers, the three profiles are very similar, with almost no variation between the bias frame and the low signal level flat field and with just a small increase in the signal of the first 8 columns in the high level flat field. For the other two amplifiers (B and C), the differences between the high level and low level flat fields are enhanced, but still within 0.5%, and extend up to 10 columns from the edge of the chip.
|Profile across leading physical overscan|
|FIGURE 10||AMP A|
|FIGURE 11||AMP B|
|FIGURE 12||AMP C|
|FIGURE 13||AMP D|
The second step consisted of verifying whether the bias value from the physical overscan changes with the illumination signal in the central region [500:1500,500:1500] of the quadrant itself. For this test, we used the entire dataset (Table 1) and therefore we have two measurements at each signal level for each amplifier. In FIGURE 14, we indicated with a different symbol each amplifier and used a filled or empty symbol to distinguish the two series. For example, at each signal level, the first image with amp A is AMP-A1 (filled box) and the second one is AMP-A2 (empty box). The figure shows a small but clear linear dependence of the bias level in the physical overscan from the illumination level. For all four amplifiers, the dependence is very similar, with a slope of 0.000250 (+/- 0.000075). This trend is related to the illumination level; we can exclude any thermal or temporal cause since the results for the four bias frames (two taken at the beginning and two near the end of the sequence) overlap.
2.2 Virtual Overscan
We performed a similar analysis for the virtual overscan. We reproduced the median profile of the virtual overscan for the two flat fields, and normalized the result with the bias value from the last five pixels. Figs 15-18 show the comparison for the four amplifiers. The profile of the virtual overscan along the column direction is not uniform in a flat field image. A small ramp is present in the first five rows of the virtual overscan. The amount of signal in the brightest, first row depends on the signal level in the image area and it is larger (~0.8% of the bias level) in the high signal flat field than in the low signal flat (~0.2% of the bias level). However, in both cases the upper region of the virtual overscan is quite uniform.
|Profile across virtual overscan|
|FIGURE 15||AMP A|
|FIGURE 16||AMP B|
|FIGURE 17||AMP C|
|FIGURE 18||AMP D|
As for the physical overscan, we verified how the bias level from the virtual overscan changes with the signal in the central region [500:1500,500:1500] of the image itself. Again, we used the entire dataset of Table 1, resulting in two measurements at each signal level for each amplifier. In FIGURE 19 we indicated with a different symbol each amplifier and used a filled or empty symbol to distinguish the two series. This figure highlights several important aspects that we will investigate in the following sections :
We finally compared the normalized profile across the columns of the virtual overscan for the three frames used as reference. Figs 20-23 show that the virtual overscan region in the case of the flat field is not as flat as in the case of a bias frame.
|Profile along virtual overscan|
|FIGURE 20||AMP A|
|FIGURE 21||AMP B|
|FIGURE 22||AMP C|
|FIGURE 23||AMP D|
3. SHUTTER LIGHT LEAK
In this section we will discuss two features of Fig. 19 : the difference between the bias level of the bias frame with the lamp on and off, and the differences in bias level in each pair of flat field with the same exposure time. From Fig. 19, it is clear that when the lamp was on, the bias level was higher than in a normal bias frame taken when the lamp was off. Is an offset in the bias level the only effect ? We can compare the profile of the physical and virtual overscans in the two cases, after proper normalization to the bias level :
|FIGURE 24||AMP A|
|FIGURE 25||AMP B|
|FIGURE 26||AMP C|
|FIGURE 27||AMP D|
|Profile across virtual overscan|
|FIGURE 28||AMP A|
|FIGURE 29||AMP B|
|FIGURE 30||AMP C|
|FIGURE 31||AMP D|
Figs 24-31 show no significant differences in the structure of the physical overscans or across the virtual overscans. However, if we compare the profile along the virtual overscan (Figs 32-35) the difference between the two cases are of the order of ~2%.
|Profile along virtual overscan|
|FIGURE 32||AMP A|
|FIGURE 33||AMP B|
|FIGURE 34||AMP C|
|FIGURE 35||AMP D|
The difference in the virtual overscan are just a symptom of the problem. During a bias exposure the shutter does not open. The fact that we see some structure along the virtual overscan means that there is a light source (and therefore a light leak). This problem appears clear when we compare the aspect of the full image in the two cases :
This light leak around the shutter explains the difference in the bias level between the two cases. However, when a bias frame is taken, the shutter blade does not rotate and the two frames in the same pair have the same bias value (Fig. 19). In the case of flat field pairs, the lamp stays on during the readout and the shutter changes its position for each exposure. The two parallel lines in Fig. 19 tell us that the amount of the light leak depends on which shutter blade covers the chip during the readout. This leak occurs only when the lamp is on during the CCD readout.
During the readout of scientific frames, the internal lamp will be off, as for reference bias and dark frames. The lamp will be on when internal flat fields are acquired (for flat field correction and CTE monitoring, for example) and during the readout of these frames an extra amount of light will be collected in each pixel. This contaminating light will depend on the filter inserted in the optical path and, as we saw in the previous section, on the shutter position. All the frames listed in Table 1 were acquired with the filter F625W. This fact gives us the opportunity to study in detail how significant the light leak could be even after the full frame bias subtraction. In fact, Fig. 14 shows that the light leak does not affect the physical overscan which is the selected region where we measure the bias offset. The light leak will appear instead as a bias pattern which needs to be removed with a full frame subtraction of a bias acquired with the lamp on. Using flat fields with two different signal levels, we can study the residuals of the subtraction of the bias with the light contamination produced by both the shutter blades. In only one of the two cases, when the bias and the flat field have the same shutter blade side, the light pattern on the bias will match the pattern in the flat field. In the following table, we first compare for each amplifier and for each case the profile along the virtual overscan of the flat field frame of the bias frame with the lamp on and the bias frame with the lamp off. We then compare the residuals in the virtual overscan region after the subtraction of the full frame lamp-off and lamp-on bias.
|AMP||Low Signal f.f.||High Signal f.f|
From this analysis, we conclude that only the subtraction of the full bias frame acquired with the lamp on and the same shutter position removes all the structure due to the bias pattern and light leak. By subtracting the bias with the lamp on, but without matching the shutter blade, there are some residuals (up to 10 DN). The worst case, of course, is a bias lamp-off subtraction which does not remove the contribution of the light leak. Again it should be stressed that the amount of the light leak depends on the filter selection. The tungsten lamp has a relatively red spectrum and therefore we expect an higher contamination in red broad band filters than in blue or narrow band filters. The filter F625W is a broad band, high efficiency filter and therefore this analysis could roughly represent a worst case scenario. How important is it to subtract the best bias from the flat field ? Since the light leak occurs during the readout, the amount of extra light is fixed for a given readout pattern. Therefore, the higher the signal level, the lower the residuals in percentage. The following plots show the minimum and maximum residuals as a function of the signal level in the image area.
|AMP A||AMP B||AMP C||AMP D|
In normal conditions, reference flat field frames will have more than 10000 DN. From the previous plots, we derive that at such signal levels the light leak can be neglected. But it needs to be taken into account for CTE measurement purposes, when flat field frames with signal level as low as 100 e- are acquired.
APPENDIX 1 : CALACS Bias Determination
CALACS is the calibration pipeline of STScI, consisting of a package of IRAF scripts specifically written for ACS data reduction. The 'doBlev' routine subtracts the bias computed from the overscan. The specification of the overscan regions to be used for this correction will be contained in the reference table OSCNTAB. The bias level is subtracted from each pixel after a linear fit of the median of values from the overscan regions designed in the reference table. In normal conditions, the routine 'doBlev' would use both the leading physical overscan and the virtual overscan. However, due to the offset between the physical and virtual overscans discussed above, we decided to run some tests. In the first case, we used both overscan regions and ran 'doBlev' with its default parameters. In the second case, we modified the OSCNTAB table to force 'doBlev' to use only the leading physical overscan (columns VY1 and VY2 set to O). In the following images and plots, we compare the original bias frame, used as input for 'doBlev', and the results for the two different overscan subtraction cases described above. The images show the 4kx2k array of each chip with a scale of +/- 8%. The row profiles in the third column show the median profile of all the 2048 rows before and after the bias level subtraction, using the two different methods. Finally, the last column shows the mean column profile obtained by averaging cols 1000-2000 for the first amplifier (black line) and cols 3000-4000 for the second amplifier (red line).
|Frame||Image||Row Profile||Column Profile|
|INPUT||chip1 chip2||chip1 chip2||chip1 chip2|
|Physical and Virtual ov.||chip1 chip2||chip1 chip2||chip1 chip2|
|only Physical ov.||chip1 chip2||chip1 chip2||chip1 chip2|
Comparison of the residual signals in the image area after the bias level subtraction shows that 'doBlev' performs a correct subtraction only when the bias level is calculated using the region selected in the leading physical overscan.