The intrapixel non uniformity [4,5] was not measured because we believe that its impact is marginal. It has been shown that when a pixel is illuminated by a spot smaller than the pixel area and moving in the x or y direction, the flux measured by the illuminated pixel can vary by several tens of %, but the flux summed over the nine adjacent pixels remains constant to within about 1% and this variation cannot be distinguished from the pixel to pixel non-uniformity. This is true only for thinned back-side illuminated CCDs. This diffusion process is negligible for our photometric application, the total flux being measured in a large aperture (300 pixels for AS channel and 80 pixels for PF channel).
The CCD was illuminated with the integrating sphere. 30 images are summed to reduce readout and photon noise. The measurement accuracy on the relative quantum efficiency is estimated to be less than 0.07%. To reduce the contribution of non uniform illumination, the local PRNU is computed in windows of 32x32 pixels. Figure 1 shows the evolution of the local PRNU with wavelength measured with a bandwidth of 10 nm.
For all CCDs the PRNU is lower than 0.6% except for wavelengths greater than 800 nm because of fringing. When the bandwidth is larger, e.g. with LED (Light-Emitting Diode) images, the fringing disappears entirely as can be seen Fig. 1 with measurments made with LEDS. The led at 880 nm has bandwidth of 80 nm.
We note on flat field images (on all CCDs) the presence of rows with lower or greater value than the mean value by a few percent. These rows are # 512, 1024 and 1536, and correspond to junctions between masks of 512 rows used to manufacture CCDs. At these junctions pixels can have a different size which implies a different collecting surface. The quantum efficiency does not differ strongly on these rows, but the difference of collecting surface changes the collecting efficiency.
4.3 Gain of CCD
The CCD gain (GCCD) is deduced from the camera (g) gain measured on the photon transfer curve. The slope of the curve variance versus the mean is equal to 1/g . This is true only if the quantum efficiency is perfectly uniform. To avoid the PRNU contribution to the variance, the variance was computed on the difference of 2 images with the same mean illumination level. In this case the slope of the relationship nadu = f(cradu) is equal to 2/g.
The CCD gain mean value for all CCD is 4.28 ^V.e-"1 with a dispersion of 0.12 ^V.e--1 and the difference between left and right output is lower than 5% for all CCD. The precision on the gain measurement is 1%. The gain is measured for different temperatures to determine relative sensitivity to CCD temperature aG expressed in ppm. K-1. The mean value for all CCDs is -900 ppm.K-1 with a dispersion of 100 ppm.K-1. The measurement precision is about 15%.
Was this article helpful?