WFC#3 Bias Subtraction and Overscan Analysis

AUTHORS : A.R. Martel, G. Hartig, M. Sirianni, J. McCann.

PURPOSE :

Our main goal is to determine the most effective method of subtracting a bias frame from WFC data frames for the June calibration campaign at BATC, correct for any residuals, and correlate the physical and virtual overscan levels and their residuals with different parameters such as the data count rates, exposure times, and filter wavelength.

DATA :

The bias frames and flat fields were acquired with SMS procedures JGCW27A-E as part of the dust filter search. The internal tungsten lamp T2 provided the illumination and the WFC Build#3 (amps BC) recorded the images. For each SMS sequence of broad-, medium-, and narrow- band filters (JGCW27A-B), a bias frame was first acquired, followed by pairs of flat fields through each filter. The first flat-field of a given pair is obtained at the nominal filter position and the second image with a filter offset of three steps. The F892N and polarizing filters were acquired in the same fashion (JGCW27E), but at the wheel position appropriate for these small filters. Their flats were read out as 1/4-field subarrays. For the ramp filters (JGCW27C-E), six consecutive images were acquired at three wheel positions with offsets to completely map the filter.

METHOD :

We simply subtracted the bias frame from the individual flat fields. In Table 1 and Table 2, residual mean counts per pixel in the physical and virtual overscans are compiled (cols 7-12) for the WFC Chips 1 and 2. The small filters, read out as subarrays, are not included. The residual mean (and its standard deviation) of the leading physical overscan was calculated between columns 10-20 to avoid any contamination from the leading edge ramp and from columns 4132-4142 for the trailing overscan to minimize the CTE contamination near the data region (see Figs 6 and 12). The rows of the two physical overscans were also restricted within the data regions (rows 5-2040). The mean of the residual virtual overscan was evaluated around its peak (cols 1000-3000). All means in the overscans were evaluated after clipping pixel values that deviate by more than 2.7 sigmas from the mean (cosmic ray hits and hot pixels). The minimum (MIN) and maximum (MAX) pixel values (cols 13-14) in the data regions were evaluated from the histogram of each image. The median (MED) (col. 15) was calculated with the msstat task in IRAF after constructing pixel masks for each chip over the bad pixels/columns. In the tables, only the MIN and MAX of the first image (at the nominal position) of each pair are given since these are essentially the same for both images. For the ramp filters, only every other image was measured.

After subtraction of the bias frame, the final bias-corrected image was obtained by removing the mean of the residual of the leading physical overscan (col. 7). This assumes that to first-order, the residual is constant over the leading physical overscan, which is generally true for the broad-band filters (see Figs 16-17, 19-20, for example) but not necessarily so for the ramp filters (Figs 27-28).

RESULTS :

1. Overscan Structures

A straightforward subtraction of the bias frame does not remove the parallel and serial CTE effects in the virtual and trailing overscans - considerable structure remains. On the other hand, the leading edge ramp in the physical overscan is greatly reduced. This is apparent in the following figures extracted from the F555W frame (CSIJ00167185419_1.fits) after bias subtraction (CSIJ00167184222_1.fits).

A net residual remains in the physical overscans after bias subtraction. For the F555W image shown below, it is approximately 5.5 DN/pixel in the leading physical overscan and 6.2 DN/pixel in the trailing overscan. This difference between the physical overscan levels can probably be entirely attributed to the CTE effects in the trailing overscan which artificially inflate its counts.

• Fig. 1 : Chip 1, Across Virtual Overscan (cols 760-1515, rows 2030-2068)
• Fig. 2 : Chip 1, Across Virtual Overscan : Blow-up (cols 760-1515, rows 2049-2068)
• Fig. 3 : Chip 1, Across Leading Physical Overscan (cols 1-30, rows 50-2000)
• Fig. 4 : Chip 1, Across Leading Physical Overscan : Blow-up (cols 1-24, rows 50-2000)
• Fig. 5 : Chip 1, Across Trailing Physical Overscan (cols 4111-4144, rows 50-2000)
• Fig. 6 : Chip 1, Across Trailing Physical Overscan : Blow-up (cols 4121-4144, rows 50-2000)

• Fig. 7 : Chip 2, Across Virtual Overscan (cols 760-1515, rows 2030-2068)
• Fig. 8 : Chip 2, Across Virtual Overscan : Blow-up (cols 760-1515, rows 2049-2068)
• Fig. 9 : Chip 2, Across Leading Physical Overscan (cols 1-30, rows 50-2000)
• Fig. 10 : Chip 2, Across Leading Physical Overscan : Blow-up (cols 1-24, rows 50-2000)
• Fig. 11 : Chip 2, Across Trailing Physical Overscan (cols 4111-4144, rows 50-2000)
• Fig. 12 : Chip 2, Across Trailing Physical Overscan : Blow-up (cols 4121-4144, rows 50-2000)

• Fig. 15 : Chip 1, Along Virtual Overscan (cols 1-4144, rows 2051-2066)
• Fig. 16 : Chip 1, Along Leading Physical Overscan (cols 10-20, rows 1-2068)
• Fig. 17 : Chip 1, Along Trailing Physical Overscan (cols 4132-4142, rows 1-2068)

• Fig. 18 : Chip 2, Along Virtual Overscan (cols 1-4144, rows 2051-2066)
• Fig. 19 : Chip 2, Along Leading Physical Overscan (cols 10-20, rows 1-2068)
• Fig. 20 : Chip 2, Along Trailing Physical Overscan (cols 4132-4142, rows 1-2068

As seen above, the physical overscans are relatively flat and show little structure along their lengths (Figs 16-17, 19-20). This is generally true for the broad-band filters. On the other hand, for the ramp filters, the physical overscans sometimes exhibit some structure, also seen in the "raw" flat field and so can not be an artifact of the bias subtraction. Curiously, the mean level of the physical overscans of the ramp filters often lies *below* the corresponding level of the bias frame acquired immediately before, resulting in negative residuals. This is especially true for Chip 1. As an example, we show below the overscan structure of the bias frame of Chip 1 (entry 16122) acquired in SMS JGCW27E and of Chip 1 of the FR459M image (entry 16123) obtained immediately after, before subtracting the bias frame. Clearly, the leading physical overscan (Fig. 21) of the bias frame is higher than that of the FR459M frame (Fig. 25) by approximately 6.8 DN. Moreover, both physical overscans of the FR459M frame (Figs 27-28) show a smooth slope, with a variation of 3 DN between rows 1 and 2068 (bottom to top of the chip).

The reason for the negative offset between the physical overscans of the bias frame and the ramp filter frames and the smooth structure along the overscans of the ramp frames is unknown. This behavior will need to be monitored in the future for the flight build detectors. In particular, we can try to correlate the overscan residuals with the illumination pattern across the field. Does a bright ramp near one edge give rise to unusual residuals on that side ?

• Fig. 21 : Chip 1, Bias, Across Leading Physical Overscan (cols 7-24, rows 50-2000)
• Fig. 22 : Chip 1, Bias, Across Trailing Physical Overscan (cols 4121-4143, rows 50-2000)
• Fig. 23 : Chip 1, Bias, Along Leading Physical Overscan (cols 10-20, rows 1-2068)
• Fig. 24 : Chip 1, Bias, Along Trailing Physical Overscan (cols 4132-4142, rows 1-2068)

• Fig. 25 : Chip 1, FR459M, Across Leading Physical Overscan (cols 7-24, rows 50-2000)
• Fig. 26 : Chip 1, FR459M, Across Trailing Physical Overscan (cols 4121-4143, rows 50-2000)
• Fig. 27 : Chip 1, FR459M, Along Leading Physical Overscan (cols 10-20, rows 1-2068)
• Fig. 28 : Chip 1, FR459M, Along Trailing Physical Overscan (cols 4132-4142, rows 1-2068)

2. Residuals and Correlations

In this section, we attempt to correlate the quantities tabulated in Table 1 and Table 2 with the median count rates in the flat field regions, the exposure times, and the central filter wavelength for the broad-, narrow-, and medium- band filters (JGCW27A-B). The ramp filters are not treated because of their complex illumination pattern and the peculiar behavior of their overscans.

a. Median Count Rates

In Figs 29-31 (Chip 1) and Figs 32-34 (Chip 2), we plot the residual count rate in the leading, trailing, and virtual overscans after bias subtraction as a function of the median count rate calculated in the flat field data region. The images of each flat field pair, the first at the nominal filter position and the other slightly offset from it, are treated separately in these plots, hence the obvious 'pairing' of the data points. A strong linear correlation between all three overscans and the data count rate is obvious from these plots, in particular for Chip 2. For the virtual overscans (Figs 31 and 34), there is an apparent step between the broad-band filters, and the medium- and narrow-band filters, but the correlation generally holds. Thus, we conclude that a high data count rate leads to a high net overscan residual.

The most straightforward interpretation of this direct correlation is that scattered light falls on the pixels of the serial shift registers. At these short exposure times, the contribution from dark current is insignificant. Sirianni finds that for Build#3, the dark rate is roughly 4 e-/pix/hour at -80 Celcius, and approximately twice this value for the temperature of -77 Celcius of our data. This then corresponds to corresponds to a rate of 0.0022 DN/pix/sec or 0.28 DN/pix for our longest exposure (127 sec) and gain=1. This is well below the 1-5 DN/sec residual count rates that we observe in the overscans.

• Fig. 29 : Chip 1, Leading Physical Overscan Rate vs Median Count Rate
• Fig. 30 : Chip 1, Trailing Physical Overscan Rate vs Median Count Rate
• Fig. 31 : Chip 1, Virtual Overscan Rate vs Median Count Rate

• Fig. 32 : Chip 2, Leading Physical Overscan Rate vs Median Count Rate
• Fig. 33 : Chip 2, Trailing Physical Overscan Rate vs Median Count Rate
• Fig. 34 : Chip 2, Virtual Overscan Rate vs Median Count Rate

Similarly, it is apparent from Figs 35-36 (Chip 1) and Figs 37-38 (Chip 2) that the differences between the trailing-leading and virtual-leading overscans also follow a linear relation. If these offsets are due to parallel and serial CTE effects, then such a correlation is expected, since a larger data count rate will lead to a larger number of trapped charges.

• Fig. 35 : Chip 1, Trailing-Leading Physical Overscan Rates vs Median Count Rate
• Fig. 36 : Chip 1, Virtual-Leading Physical Overscan Rates vs Median Count Rate

• Fig. 37 : Chip 2, Trailing-Leading Physical Overscan Rates vs Median Count Rate
• Fig. 38 : Chip 2, Virtual-Leading Physical Overscan Rates vs Median Count Rate

b. Exposure Time

The above results can also be visualized by correlating the overscan residuals and their differences against the exposure times. Since the data count rate and the exposure time are inversely related, as seen in Figs 39-40 below, we expect a strong correlation. In Figs 41-43 (Chip 1) and Figs 44-46 (Chip 2), the residual overscans are clearly correlated with the exposure times. Similarly, in Figs 47-48 (Chip 1) and Figs 49-50 (Chip 2), the differences between the overscan residual rates also show a strong correlation. These results then reinforce the correlations found in the previous section.

• Fig. 39 : Chip 1, Median Count Rate vs Exposure Time
• Fig. 40 : Chip 2, Median Count Rate vs Exposure Time

• Fig. 41 : Chip 1, Leading Physical Overscan Rate vs Exposure Time
• Fig. 42 : Chip 1, Trailing Physical Overscan Rate vs Exposure Time
• Fig. 43 : Chip 1, Virtual Overscan Rate vs Exposure Time

• Fig. 44 : Chip 2, Leading Physical Overscan Rate vs Exposure Time
• Fig. 45 : Chip 2, Trailing Physical Overscan Rate vs Exposure Time
• Fig. 46 : Chip 2, Virtual Overscan Rate vs Exposure Time

• Fig. 47 : Chip 1, Trailing-Leading Physical Overscan Rates vs Exposure Time
• Fig. 48 : Chip 1, Virtual-Leading Physical Overscan Rates vs Exposure Time

• Fig. 49 : Chip 2, Trailing-Leading Physical Overscan Rates vs Exposure Time
• Fig. 50 : Chip 2, Virtual-Leading Physical Overscan Rates vs Exposure Time

c. Central Filter Wavelength

In the figures below, the overscan residual rates and their differences are plotted as a function of the central filter wavelength for both chips. There are no apparent trends or correlations.

• Fig. 51 : Chip 1, Leading Physical Overscan Rate vs Central Wavelength
• Fig. 52 : Chip 1, Trailing Physical Overscan Rate vs Central Wavelength
• Fig. 53 : Chip 1, Virtual Overscan Rate vs Central Wavelength

• Fig. 54 : Chip 2, Leading Physical Overscan Rate vs Central Wavelength
• Fig. 55 : Chip 2, Trailing Physical Overscan Rate vs Central Wavelength
• Fig. 56 : Chip 2, Virtual Overscan Rate vs Central Wavelength

• Fig. 57 : Chip 1, Trailing-Leading Physical Overscan Rates vs Central Wavelength
• Fig. 58 : Chip 1, Virtual-Leading Physical Overscan Rates vs Central Wavelength

• Fig. 59 : Chip 2, Trailing-Leading Physical Overscan Rates vs Central Wavelength
• Fig. 60 : Chip 2, Virtual-Leading Physical Overscan Rates vs Central Wavelength

CEI SPECIFICATIONS :

None.

CONCLUSION :

• After subtraction of the bias frame, the leading physical overscan is relatively flat while the trailing physical overscan and the virtual overscans remain contaminated by serial and parallel CTE effects. The leading physical overscan is therefore more representative of the 'true' bias level and we therefore remove its mean residual from the bias-subtracted flat fields. The trailing and virtual overscans are ignored.
• There is a strong linear correlation between the residual physical and virtual overscan count rates (after bias subtraction) and the data region count rates in the flat field. The higher the count rate in the data region, the higher the overscan rates. The simplest interpretation is that scattered light contaminates the serial registers. The dark current rate is too low to account for the observed levels. This hypothesis can be tested by repeating the measurements (or variation thereof) with an *external* light source, such as RAS/HOMS, which should produce a different scattered light pattern, and possibly different behavior in the overscan regions.
• No wavelength dependence is observed in the overscan residual rates and their differences.
• The negative overscan residuals in the ramp flat fields and the smooth structure along their overscans remains a mystery. Future tests might include a correlation of the residuals and overscan structures with the brightest part of the filter on the chip, similar to the above analysis of the broad-, medium-, and narrow-band filters.

REFERENCES :

• Description of the WFC Build#3 detector (M. Sirianni).
• Description of the WFC filters (Z. Tsvetanov)