: d 1 
ns a (1) 
The measurement is desraded particularly by vibrations of 
the plane about its cross axis. The error caused by the lack 
of certainty whether row II actually scans the very strip 
previously registered by row I can be reduced by optical smearing. 
PROCESSING OF THE OBTAINED DATA 
The comparison of the brishtness profiles of row I and row II 
mentioned above means a determination of the degree of correla- 
tion between the two rows. Row II has recognized the bright- 
ness profile of row I if the correlation measure adopts an 
extremum. the time available (At = 50 ms) is too short to. 
allow the correlation function to be determined. As a sub- 
stitute, the mean of the absolute differences (MAD) of the 
pixel values is formed. This MAD function has a minimum where- 
ever the correlation function has a maximum /1/. Thus, the 
search for maximum correlation will, hereafter, always be a 
search for the minimum of the MAD function. As the formation 
of the mean of absolute differences always extends over the 
same number of pixel pairs, division by the number u of pixel 
pairs may be spared. Then, the formula for computing the MAD 
funetion is 
u 
m( u) => Fa $ or /a) - b(i)| ja = =] ees +] (2) 
i=0 
where ,u designates the shift of row II relative to row I. 
The quéntities a and b are the digilbed brishtnesses of the 
pixels of row I and II, respectively. Frequently the diffe- 
rence of thefrow means (à - b) is subtracted within the absolute 
value siens. This measure would be of importance especially if 
the rows were exposed for different times or if different 
drifts of the equal shares might occur in case of different 
temperatur coefficients due to temperature variation. To 
Solve these problems, exposure times are determined anew prior 
to every osure cycle and maint_ained through the respective 
cycle for both rows. The temperatur problem is further reduced 
by mounting both CCD rows on a ceramic base. 
In the device here described, the rows have been extended at 
either and by |j|=7 pixels, so that altogether 15 shifts are 
examined. The pixels used for row extension have to be so 
selected that they do not in any case exhibit a correlation 
with row II; they only serve the purpose of always having 
u pairs of values in order to save the division. Fig. 3 shows 
the resulting representation of funetions for five exposures 
of row II. The graph also indicates that, with an assumed 
shift of three pixels, the correlation maximum, i.e. the MAD 
minimum, is found after four correlations. Thus, the v/h ratio 
is n=4 and the drift angle = 3. It should be added that in 
our instrument a shift by seven pixels corresponds to the 
130