... timing¹

The BRE method is more affected by randomn variations in the exposure time but has the important advantage of not being affected by any systematic offset. However, if the software for controlling a particular CCD and shutter does not allow, or makes it particularly inconvenient to obtain, multiple exposures without reading out, then a simple bracketed method may be adequate, e.g., 3s, 6s, 3s, 9s, 3s, etc. (see also Gilliland et al. 1993). The exposures should be long enough so that shutter timing is not affected by a constant offset or other inaccuracy (which may vary with position on the CCD). With a base exposure of 3s, the exposure time must be accurate to within about 6ms to obtain 0.2% accuracy in the non-linearity curve. Increasing the base exposure time (with reduced illumination) will relax this criteria but increase noise associated with fluctuations in the intensity of the lamp.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... gain²

This `relative gain' is proportional to ADU/electrons (inverse to the normal definition of gain).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... expected-counts.³

The initial normalisation was made using a linear fit between 0 and 30000 counts. After the combined quadratic fit, the data were renormalised to a relative gain of 1 at zero counts. This is not a completely unbiased way of combining the data, i.e., there could be systematic errors, but it is adequate for the demonstration in this paper.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.