Re: Connectors and cables for high-speed digital signals

Tim Nustad x7-4361 (nustad@sol.med.ge.com)
Fri, 26 Jul 96 10:22:59 CDT

Per Torstein Roeine had a recent question on cabling solutions
for board to board high speed digital interconnects. I thought
I would pass along some experiences we have had. In both of the
cases discussed, costs requirements prevented the use of a typical
RF coaxial connectors.

AMP Z Pack
===========
We have used the Amp Z Pack connector system for one application
here that required high speed board to board interconnect. It
may give you some ideas.

If the flexibility required of your "cable" is not extreme, it is
possible make your cable system out of the Z pack connectors mounted
to a small flexible circuit board. The flex circuit can be built
as a 50 ohm strip line or microstrip. In a simple picture:

________________|--| |X|--------------|X| |--|_________________
^ ^
circuit board ^ ^ ^ circuit board
| | |
Zpack | | flex circuit board
|
Zpack

We built a flex circuit board with 10 layers (5 signal layers, 5 ground
layers, with the outer layers being ground). The flex circuits where
about 6cm wide, and 1m long. There where 64 signals in each flex circuit
"cable" The cables where able to bend around an 8cm radius.

The flex circuit can be costly, but the assembly of the cable is simple and
cheap as you do not have to terminate multiple wires, you simply solder
the Zpack connectors to the flex board. The completed cable was nearly the
same cost as a traditional cable made of miniature coax bundles.

Amplimite 050
=============
If you need the flexibility of a traditional cable, we have also made
successful use of the Amplimite 050 series shielded connectors. In this
case we have cable bundles of 16 to 32 miniature 50 ohm coax cables
terminated to the Amplimite connectors. Each coax center and ground
have their own pin pair in the connector.

This Amplimite system does not maintain a 50 ohm impedence through the
connector. We have found that the small discontinuity the connector
presents is very tolerable in our application up to 200Mhz. I would encourage
you to experiment with some of the well shielded "non RF" connectors such
as this. They can have quite good results, if the cabling and termination
impedences are well behaved.

Regards,

Tim Nustad
-------------------------------------------------------------------
Lead Program Integrator
GE Medical Systems, Ultrasound
phone: (414) 647-4361
fax: (414) 647-4117
email: nustad@sol.med.ge.com

> From perr@ifi.uio.no Fri Jul 26 07:53:35 1996
> Mime-Version: 1.0
> To: si-list@silab.Eng.Sun.COM
> Subject: Connectors and cables for high-speed digital signals
> From: perr@ifi.uio.no (Per Torstein =?iso-8859-1?Q?R=F8ine?= )
> Date: 26 Jul 1996 14:42:19 +0200
> Lines: 54
> X-Mailer: Gnus v5.3/Emacs 19.31
> Content-Length: 2747
>
> I am currently designing a prototype for evaluating several
> possibilities for high-speed low-swing digital interconnects. This
> includes differential signals in parallel and a new method using
> 3-wire pseudo-differential signals (patent pending). On the PCB, 50
> Ohm microstrips or striplines will be used, with on-chip termination.
> Frequency range is 150-500MHz (0.3-1Gb/s), distance less than 3m.
>
> I need to connect the prototype boards together, and to an
> oscilloscope. In order to do comparisions, I would like the cables and
> connectors to have similar properties. As I have little experience in
> this field, I need your comments and answers to some questions.
>
> One option is to use coax cables and SMB connectors (or similar). This
> will be a bit clumsy and expensive, as 32 connectors are needed per
> board. I think skew may be a problem. What kind of skew can be
> expected? (for connectors only, and per meter of cable). What kind of
> delay differences can I expect between pre-made cables from the same
> vendor/batch? I also need addresses of companies that can supply such
> cables, preferrably in Europe.
>
> For the differential signals, twin BNC connectors can be used. This
> cuts down the number of connectors to 22 per board. I guess that skew
> between the two signals of each differential pair is better controlled
> this way, but delay differences between cables is still a problem. Are
> there any reason that it should be worse/better than for coax cables?
> Where can I buy pre-made twin BNC cables? Are there any other similar
> concepts that I can use? I think the best would be a cable with 2-3
> 100/50 Ohm shielded twisted pairs.
>
> For the three-wire signals, I would use triple BNC connectors if they
> existed. Instead, I think I have to treat them as three separate 50
> Ohm single-ended signals, and use three coax cables per signal. Other
> suggestions?
>
> Another option is to use some kind of high-speed connectors and ribbon
> cables, like Amphenol Champ or something. Twisted pair ribbon cables
> is probably good for differential signals, but what about the other
> ones? Do you know of any microstrip or stripline connector systems
> (like AMP Z-PACK) that can be used with cables? I have seen references
> to Skewclear from Amphenol and something called Molex-Tensolite
> milli-Z, but I have problems finding any more information. Again,
> addresses to producers and distributors will be helpful.
>
> What is _your_ favourite board-to-board interconnection system for
> things like ECL, PECL or LVDS?
>
>
> Thanks,
>
> --
> Per Torstein Roeine email: perr@ifi.uio.no
> University of Oslo phone: +47 22 85 24 52
> Dept. of Informatics, Microelectronics Group fax: +47 22 85 24 01
> Box 1080 Blindern, N-0316 OSLO, NORWAY
>