64 
Fig. 5. SNR of a CCD-element at a 
saturation load of500,000 electrons and 
rms-noise of235 electrons. 
kpixcl 
Fig. 6. Effect of flat field illumination in 
the focal plane. 
One can clearly see the effect of the shading of the 
lens (at the edges the light intensity falls off to about 
40%) and of the influence of the PRNU. The 
differences of sensitivity of CCDs are usually 
indicated in the datasheets as PRNU values in 
percent of the values of the videocurrent for the 
range far below saturation (mostly at 50% of U mt ). 
We will adhere to this definition here as well. Thus 
in the linear range of the CCD the fixed pattern 
noise of the pixel sensitivity can be expressed 
directly as a signal dependent noise, which is 
converted into a time dependent noise during 
transfer of the charge. 
PRNU PRNU 
o,.. = //.. = cr; 
100% 
100% 
(3) 
Depending on the PRNU of the CCD the signal to 
noise ratio results as: 
SNR = 
(4) 
1 + 
PRNU 
100% 
cr. 
Figure 7 shows the highest attainable SNR at full 
exploitation close to saturation (400,000 to 500,000 
signal electrons) in relation to PRNU, based on the 
aforementioned parameters. Up to a PRNU value of 
0.02% the SNR is determined exclusively by the 
photon noise of the signal, the rms noise of the CCD 
and the noise of the analogue channel. At 0.1% the 
PRNU influence becomes dominant. In the 
engineering model of LH Systems’ new airborne 
digital sensor, described in sections 6 and 7 below, 
the PRNU correction is done pixel-wise. 
Fig. 7. SNR in relation to PRNU assuming 
a thermal and electronic noise of S e [ = 235 
rms-electrons and a signal electron counts 
of500 000 e~. 
Normally light fall-off of the lens system 
(approximately 30%) is corrected simultaneously 
with the PRNU correction. For the estimation or the 
PRNU correction, this fall-off has not been taken 
into account because it does not contribute directly 
to an increase of the SNR. It only contributes 
indirectly through the adaptation of the signal to the 
analogue channel. The correction of light fall-off of 
the lens does not restrict the SNR significantly. 
The efficiency of the correction can be seen in 
Figure 8, which shows imagery of the Reichstag, 
Berlin, taken with the engineering model of the LH 
Systems airborne digital sensor on 23 April 1999. 
The flying height was 3 km and the ground sample 
distance is 0.25 m. In the radiometrically zoomed- 
out image parts no noise can be seen.