A. Mean Bias Properties


The SBIG target market is the amateur astronomy community. One consequence of this is that the SBIG Camera electronics do not provide an overscan region upon readout. An overscan region is produced when the CCD output amplifier is instructed to read out some additional number of columns, n, over and above the nominal numbers of columns, N, of the CCD. For a CCD with a geometry of P rows by N columns, the resulting image after readout is P X (N + n) in size. The additional P X n pixels at the end contain information on the DC-offset and Read Noise only. This is true no matter what type of an image has been recorded. Thus an overscan region for (say) an image of a bright nebula contains information on the DC-offset of that image. With such information, one can correct for the DC-offset term on an image-by-image basis.

Overscan-upon-readout is the standard procedure for all professional astronomical CCD cameras, and has been so since the late 1980s. SBIG's failure to comply with this standard means that anyone wishing to do quantitative work with our system must be very careful about the detector temperature while obtaining both data and calibration frames. The reason all professional CCD systems generate overscan regions is that this allows the observer to explicitly separate the DC-offset and the 2-D pattern term. The overscan region of any image (not just bias frames!) provides a measure of the DC-offset appropriate for that frame. Thus, if one has data with overscan regions, one can subtract the DC-offset from all the frames individually. One can then measure the 2-D pattern term from a series of "overscan-corrected" bias frames, and median them together to generate a much higher signal-to-noise estimate of the 2-D pattern than can be obtained from an individual bias frame. As detailed below, the lack of an overscan region for our system introduces a systematic uncertainty of ~0.5 - 1% in all images obtained with our system.

When using the instrument, one sets a target detector temperature and the system tries to reach that temperature. The thermoelectric cooler cannot cool the detector more than about 30 degrees below the ambient temperature. Also, even when the system reaches its target temperature, the temperature stability is not exceptional. It will flutter up and down by 1 degree or so. The temperature recorded (to 13 places!) in the image header is the temperature at read-out. For bias frames it is probably accurate to 3 places. For longer exposures, even when the temperature control is behaving well, the temperature given in the image header is not to be trusted to better than 1 degree.

Figure 1 shows the measured average bias level as a function of temperature. There are six sets of data shown. Five were obtained by PBE on the dates indicated on the figure. All of the data from these sessions was obtained unbinned. The fifth data-set is a collection from various nights obtained by former MnSU undergraduate Andy Monson. Three essential points should be derived from this plot.

  • Figure 1: Mean Bias level as a function of detector temperature for bias frames obtained on five separate dates from mid-summer through mid-winter.

    First, there is no evidence of any night-to-night systematic offsets in the bias level to within the scatter in the data. Thus one can reliably expect to understand what the bias level of a given exposure should be (to within the intrinsic scatter) if one knows the detector temperature at the time of the exposure.

    Second, the bias level is a very strong function of temperature at detector temperatures T > -10 degrees C. This means that good quantitative work will be very difficult, if not impossible to do unless T < -10 degrees C.

    Third, even at T < -10 degrees C, there is a scatter of ~1 ADU at fixed temperature. If one takes a simple mean of the data in Figure 1 with T < -10 degrees C, one finds

    This is shown by the dotted line in Figure 1. There does appear to be a slope in the data, even at T < -10 degrees C. An unweighted one-sided linear least-squares fit to the data, using T as the independent parameter, yields the following:

    This is shown by the solid line in Figure 1. The least-squares fit is formally a bit better than a straight mean, but one should pay attention to the fit dispersion. Without an overscan region, the fit dispersion determines the limit of precision we can obtain when doing bias-correction.

    One final test that illustrates this last point is shown in Figure 2. The figure shows the mean bias level for a set of 13 bias frames all taken with approximately the same system temperature (T ~ -35 degrees C) over the course of about an hour and a half. These data were obtained as part of a set of tests to study the dark current of our system (see Section 3, below). The following points emerge from the figure:

  • Figure 2: Mean Bias level at a constant temperature as a function of time.

    There is no evidence of any temporal shift in the bias at a constant temperature. Thus one can reasonably expect the bias to be stable if the detector temperature is kept stable.

    There is evidence for a "jitter" in the bias level of 1 ADU from exposure to exposure. This reinforces the point above that a given exposure will have a systematic uncertainty in the bias level of on the order of ~1 ADU.


    Procede to Next Section

    Return to Main Bias Page

    Return to Outline


    Updated: 2005 September 20 [pbe]