9.4.2 The direct test

A direct way of checking the linearity of the CCD is to plot level pr. unit time versus level, since if we have $N \propto I \cdot t$, and I is constant, then N/t is constant.

The level used before (e.g. in Figure ) was computed using imstat with a suitable upper cut to avoid cosmic-ray-events. To get more precise level, fitsky was used. This task fits a Gaussian to the histogram of the sky values and determines the mean level from the fit. The center area was defined as (x,y) = (500,200), and the right hand side area as (x,y) = (995,300), both with an inner radius (annulus) of 2.5 pixels and an outer radius (dannulus) of 40.0 pixels.

Since we have a number of images with the same exposure time (10 sec, see Table ), we can check whether the intensity really was constant by plotting the level versus the (clock) time. This is shown in Fig. . It is seen that the illumination level was not constant.

 

The points shown in Figure  were fitted with a 5 piece (JB) / 2 piece (GR) cubic spline using gfit1d, as also shown in the figure.

All the N/t values for the 68 images were now divided by the fit to the time variation. Now N/t vs. t could be plotted, as shown in Figure .

 

As Figure  shows, the CCD is in fact linear up to the very highest levels.

The rms standard deviation of all the data with levels > 100 ADU is 1.1%. The rms for the first sequence (JB) and the second sequence (GR) separately, is 1.4% and 0.7%, respectively. The reason for the difference is probably the more ill-behaved lamp variation for the first sequence, as seen in Figure . We conclude, that the CCD is linear within 0.7%.