[Gb-ccb] connectors

[John Ford]

Martin,

This is indeed a tricky business.  The skewclear cables seem like the
best approach, but the crosstalk spec you have is certainly pretty
tight.

First off, we'll connect a big fat ground bus between the receiver and
the backend, which should eliminate nearly all common mode error
between the 2 chassis.

I would, for the sake of sanity, separate out the control signals from
the data lines.  That is, the controls for the phase switches and cal
signals should have their own connector.  This connector is
non-critical.  I suggest a standard DB-25 Spectrum Control 700 series
filtered connector with either twisted pair individual cables, or some
of the spectra-strip twisted pair flat cable.  Use an overall shield
if you like, but I don't think it's necessary.

For the data cables, you will have to decide on an approach, either a
couple of large cables, or a multitude of shielded twisted pairs...
Both have their strong points.

I think that the most workable approach would be to build a PC board
attached to an FCN-260 connector that you've suggested, with the RFI
chip filters (note that a capacitor is not sufficient. You'll need a
ferrite bead at least and I'd use an LC or Pi section filter.  See
http://www.spectrumcontrol.com/pdfs/emi4-8.pdf for example) Then, put
a metal can over the connector and filter assembly.  From this board,
to the A/D board, then, you could either use a card-adge connector and
plug the A/D board diractly into this card, build the A/D's directly
on this card, or use some more of the skewclear.  The skewclear is not
terribly flexible, so I would recommend against doing this.  I would
also use 2 connectors for the data.  Each FCN-260 infiniband connector
can hold either 8 or 24 pairs, so 2 of the 8 pair connectors would
work fine.  It seems like it would be impossible for a mere mortal to
build these inifiniband connectors, so we should stick to commercially
available configurations.  

See my sketch for some ideas.  We think that your best chance for
success and the least cost, least fuss option is to build the A/D
section with the connectors on that board.  The less distance from the
connector to the A/D system the better from a noise perspective.

Obviously, the other idea is to use individual shielded twisted pairs
for the data signals.  This has some drawbacks as well: Bundling up
the cables creates the same problems as putting them inside the
skewclear, you still need to filter them, you still have the problem
of handling them inside the chassis, and now there are lots more
mechanical connections to deal with.  In any event, I would not
attempt to capture all these connectors in a bolt-down plate or set of
plates.  We're smart enough to wire it up right, and the ability to
change the data inputs from one to another comes in handy from time to
time for troubleshooting.  I hate this idea.  It seems it would be a
nightmare inside the chassis and out.

At the end of the day, the connectors used are probably not the
limiting factor in the low-noise, crosstalk, and accuracy departments.
The noise inside the enclosure, the analog circuit designs, and the
digitization will contribute to all of this.  It may be that simply
using standard spectrastrip mass-terminated cables, and using every
other pair as a ground, would give good enough performance.  That
would have to be verified via a test on the bench.  Whatever solution
you choose, it should be tested on the bench.

I think that we could live with whatever you think will give the best
performance.  Keep in mind that the RFI levels are a specification
that cannot be violated.

John


ccb.ps file in separate mail...