The Leitz PMM 654 is a 3-axis coordinate measuring machine. Movement along the horizontal x- and y-axes is done by separate air-bearing granite blocks. The optical system is focussed on the plate using the vertical z-axis movement, which is mounted on the y-axis (note that there are two definitions of the coordinate systems in use: see below).
A photographic plate or film is mounted in a holder on the granite x-axis. Both the CCD imaging system above the plate and the illumination system under the plate are mounted on a "bridge" which wraps around the plate but which is not physically connected to it: the bridge is suspended from the y-axis. A plate is scanned in a series of lanes, by moving the x-axis and thus moving the plate through the optical path of the imaging system. The field-of-view of the imaging system is about 1.5 cm: at the end of each x-axis movement, the x-axis is returned to its start position, the y-axis is stepped by an amount which defines the lane width (at the present time this is 1.28 cm) and the next lane is scanned.
The coordinate system is defined as follows:-
| y | | \/ <---------- x SuperCOSMOS /\ | x | | -----------> y LeitzThe coordinate system convention assumed by the SuperCOSMOS analysis software has the {x,y} origin at top right: note that the intrinsic coordinate system used by the Leitz control computer differs from this, and hence the SuperCOSMOS control software makes a conversion between coordinate systems: these coordinate systems have become known as the "Leitz" and "COSMOS" coordinate systems.
During the scanning of a lane, the table is simply driven at constant speed
along the Leitz x-axis. In practice, of course, there are small variations in
speed which, if not taken into account, would result in positional errors in
the output, since if the CCD camera were read out at uniform time intervals,
the table speed variations would result in non-uniform spacing of pixels in
the final image. This is allowed for by linking the readout of the CCD to the
table position. The table's position is measured by Moire fringe gratings:
those gratings produce two sinusoidal signals whose wavelength is 20 micron,
the two signals being out of phase by 5 micron. A bistable flipflop is
switched every time one of those signals crosses zero (i.e. every 5 micron):
further electronics converts one edge of the resulting step function into a
pulse: pulses are thus delivered every 10 micron movement. Those pulses are
then used to trigger readout of the CCD, and the CCD 
-axis coordinate
system is directly tied to the Leitz table coordinate system.
The positional accuracy of this method can be assessed by measuring the amount
of light collected by the imaging system at each readout. Variations of
percent in collected light are observed. This places an upper limit on
the positional uncertainty: if those variations in collected light were
entirely due to positional errors in the output from the gratings, with the
table being driven at constant speed, then the largest positional error would
be 
percent, or 0.2 micron. In fact, some tuning of the grating
output was required in order to achieve this, as the zero-crossings of the
output were not initially set up to be equidistant. The zero-level was varied
until the difference between successive readouts was minimised. This was
measured by a Fourier timeseries analysis of the CCD output (see below): the
Fourier amplitude at the Nyquist frequency, corresponding to 10 micron
sampling, was minimised.
In fact, the Fourier analysis showed that much of the variation in collected light is due to variations in table speed due to mechanical imperfections in the drive system. On the power spectrum of the fluctuations clear mechanical resonances can be seen. These were minimised as far as possible by changing the drive motor from the original system.