Improving the absolute X measurement
Below you read about the ideas I had in the past few years and then go
to the preleminary system description for ATLAS.
Or you can go there directly: the ATLAS
RASNIK system.
Pixel synchronious camera's
Using the same test set-up, we connected a pixel synchronious camera to
the frame grabber. Thus the systematic error in the X position and X size
can be eliminated by grabbing the edges of the picture. Also the analog
video data is digitised with the clock coming from the camera. This eliminates
the jitter, normally introduced by the PLL in the framegrabber. The results,
after analyzing the data of several 'pictures', show that the X direction
is now more accurate then the Y direction. This is caused by the bigger
amount of edges in the X direction in the picture grabbed.
Normally, pixel synchronious cameras (and grabbers) are expensive.
So this might be an option when not too many channels are needed. You get
excellent results with little effort.
New chips on the market
More and more new chips become available that might be used to build camera's
that can be read out synchroniously or even random accessable. This is
pretty new (at least the cheap ones) and these chips employ photo-diodes,
rather then the CCD structures as used in cheap observation camera's. The
read-out structure makes the photo-diodes more suitable to be integrated
in the process. Drawback is that the fill-factor (how much of the area
is sensitive to light) is relatively low. This, and the process itself,
makes them less sensative. Also the sensitivity over the surface used to
be not very homogenious.
We tested such a chip (FUGA15 from IMEC) and came to the same conclusion.
However, new chips using the CMOS process enter the market constantly.
Using a mechanical reference
An other possibility is to use reference at the location of the camera,
that is 'seen' at the same time as the projected image of the mask. In
principle this reference should be reconstructed with the same accuracy
as the projected image. This means that half of the digitized data is used
as a reference. In practice, this reference will be more stable then the
projected image and can therefor be much smaller then half the sensor's
size.
In the test set-up two wires, of 30 microns thickness, were stretched
on top of the CCD sensor window. These wires become the reference, rather
then some X offset in the framegrabber settings. The distance between the
wires determine the scalefactor in X direction, rather then the PLL synthesized
clock of the framegrabber. The data, taken with this setup, have not been
analized.
You can have a look at one of the grabbed pictures,
however. Drawback of this system is that it is not easy to separate the
reference from the projected image.
Using a professional frame-grabber
The more expensive frame-grabbers allow the user to change the start position
of digitization and its frequency. This way the intrinsic black pixels
at the begin and end of the video line can be grabbed. Thus the mechanical
edges of the CCD sensor act as the reference for the picture projected.
The accuracy of the position reconstruction of these edges is limited
however. This is described on this page.
Special software, analysing the picture contents near these edges, may
improve this accuracy a bit.
I had an other idea (not tested). The video signal is activily split
near the frame grabber. One branch is used to synchronize the grabber.
The other branch is bandwidth limited to app. 2 MHz. This hardly influences
the signal generated by the projection of the mask. If the time constant
is known, the picture can be reconstructed to its original location. The
fast edges however, are stretched with a time constant of 80 ns. Thus,
3 or 4 sample points become available for the reconstruction of the begin
and endpoints of each video line. Thus the edges of the intrinsic black
pixels can be reconstructed very accurately. This seemed to be a promising
and cheap method.
Develop a special frame grabber
It may be interesting to develop a dedicated frame grabber for this special
purpose. One can use simple cameras, which share a bit more expensive grabber.
There are some vague ideas about how this grabber should work. The edges,
introduced by the intrinsic black pixels, could by measured by means of
a TDC. The TDC values can be used to correct the grabbed picture. The clocks
of the TDC and ADC should be coupled closely.
Going a bit further, is to switch to a TDC completely. The video signal
is differentiated twice. The first differentiation gives the direction
of the edge. When the second differentiation goes through zero, it points
to the middle of the edge of the video signal. Because of the different
polarities, two TDC inputs are needed. There will be differences between
the comparators and TDC channels. But since there are app. as many black/white
as white/black transitions on a reconstructed vertical line, this will
be of little influence. The same problem is seen in the classical system,
where the white areas are bigger then expected in the reconstruction. This
is most likely caused by the gamma correction, which is standard for cheap
observation cameras.
Simulation results
The circuit described above has been simulated, using SPICE. You can have
a look at the results.
The begin and end simulations include the differentiated and double
differentiated video signals V(dif) and V(ddif). The falling edge of ff2
points to the middle of the rising edge of the video signal. The falling
edge of ff1 points to the middle of the falling edge of the video signal.
You can see that the first falling edge is missed by the detector because
of the strong signal of the first rising edge. The results have not been
analysed in detail but the principle seems to work.
Drawback of this scheme is that one cannot look at the actual picture.
This is "lost" after the conversion, which may make debugging more difficult.
One cannot even recognize dust on the sensor.
We have now choosen to use the VV5430
CMOS sensor from Vision (VVL) for the
ATLAS RASNIK system. From here on the story continues on the RASNIK
for ATLAS pages.
Or go to the old CCD_Rasnik
home page.
You can mail 
suggestions for (or questions about) this page to: Henk
Groenstege.
March 1999