connector for LVDS

Have a look at one of the smaller D-type hybrid shells,

The coaxial inserts are essentially SMB - I think we got away with using a standard SMB crimp tool to crimp them - and they are both robust and tidy. The standard D-type pins are a bonus - you could use them for power supplies or extra shield connections.

I don't know of a stocking distributor in the US, but there ought to be at least one. RS Components aren't my favourite distributor. but they are the only source of these parts I know of in Europe.

----------- Bill Sloman, Nijmegen

Those have always seemed a bit too specialized and complex for my taste, surely there's something better suited for this new common problem. One doesn't need 3GHz bandwidth for most LVDS channels. John, what about differential LEMOs? Maybe they aren't so bad?

No, they're beautiful connectors, just expensive.

John

Hello John,

USB plugs come to mind, being cheap and plentiful. Of course that carries the risk of plugging into the wrong system. But you may be able to make sure electrically that nothing will fry if someone does that. Then you'd have four contacts plus shield. The host sides are a bit wide and clunky but the other end is a smaller square. If you'd have the small squares only then nobody could plug into a laptop or PC.

Regards, Joerg

I first ran into this problem back in the late 1980s. Within a rack, the mixed signal DIN41612 connectors worked extremely well, letting us plug in all the coaxial connections at once.

The hybrid D-types use the same coaxial inserts, but they come with back-shells, screw locks and all the rest of the paraphenalia that you need to make a reasonably tidy and secure connection between two free-standing boxes.

They were much cheaper than LEMO connectors, and available off the shelf, in small quantities - LEMO stuff (though quite beautiful) can be hard to get hold of.

At that time we got all our hybrid connectors from the Quadrant Connector company, who looked after Cambridge Instruments very well indeed. They still exist but I don't know anything about what they sell today, or how they sell it.

The temptation to use multiple BNC or SMA/SMB/SMC connectors should be resisted. No matter how carefully you label the cables, some idiot graduate student/technician will plug them in in the wrong order, then give you a hard time about your unreliable design.

The pleasure one gets from pointing out that the cables have been plugged into the wrong sockets is no compensation for having to spend half an hour walking around to the equipment and persuading the offenders to look at what they've done wrong.

The first time we ran into the problem, we threaded all the half-dozen coax cables involved through strategically placed holes in a perspex plate (labelled the "monkey frame" by the technician who did the work) before fitting the coaxial connectors.

We then bound the cables together close to the monkey frame and the connectors so that there wasn't enough slack to let anybody plug the cables into the wrong sockets. It worked fine, but wasn't elegant. Since the machine was minimally improved copy of a Siemens-physicists' development machine, which Cambridge Instruments then sold to Thopmson-CSF, this wasn't the only inelegance. The production version was quite a lot better, and the fully digitised version (which got canned before we could get it into production) was an absolute dream

------- Bill Sloman, Nijmegen

Mini-DIN's are smaller than a BNC twinax. And a lot more readily available than LEMO's.

There is a connector called the "mini twinax" in the computer world but I don't know where to buy the raw plugs/jacks.

Pair-to-pair skew doesn't hurt you when you've only got one pair! Don't knock the easy availability of CAT5 stuff.

Amphenol Fakra's seem to be popping up everywhere these days, and they seem to be exactly dual SMB's. My impression is that this is the GHz range auto-RF connection being pushed these days.

Tim.

Circular DIN connector? XLR audio connector? On the other hand, if a BNC is too big, those might be too big as well. Maybe a mini-DIN (PS/2 keyboard and mouse) connector?

Use one of each on each box and make cables to match - that way it's harder to hook up the cables backwards.

Matt Roberds

John Larkin wrote: > If one had a couple of small boxes and wanted some single-lane LVDS > connections between them, any idea of a suitable connector? Something > like a BNC-type twinax, but smaller. Or a differential LEMO, but > cheaper. Two pins and a shield would be best, so an RJ-type thing > might not be so good. Two SMB or MCX coaxal connectors would work but > seems klunky. >

Amp always had those individual pin type things for coaxial termination- you don't get smaller than that, but they hold on by friction. Then there's Gore, not so expensive when dealing with stock items.

Seems to me that using two separate coax'es for the two sides of the LVDS is asking for trouble. If one has a different prop delay from the other, then, after a transition, there will be a short time when both signals are at the same level, which sure sounds nasty from a jitter standpoint. If I went for two connectors, we could use SMB or MCX types, both cheap and easy to get.

Actually, I guess a twisted-pair has a similar but less severe problem: classicly there's an even mode and an odd mode, but it seems to me there are actually three prop modes: line1 to ground, line2 to ground, and differential. And the first two modes could have different prop delays. So you could get silly looking waveforms at the receiver.

John

The rise and fall times for LVDS are 0.5nsec, which is four inches (100mm) of propagation delay in normal coax. You shouldn't have much trouble matching your coax cable lengths to better than that - even half an inch (12.7mm) difference in length would be pretty obvious after assembly.

-------- Bill Sloman, Nijmegen

Hello John,

How about the low-tech low cost path, red and white RCA? Then you could use audio cables from Walmart and buy color coded jacks for your modules almost anywhere and any style :-)

Regards, Joerg

Hello John,

True. I guess nowadays they are all machine made and molded but the coaxes can come from two separate spools during the process. I doubt the prop delays will be much different though because that's all in the materials.

SMB cheap? Wow. Last time I heard that was from the mil spec world.

Twisted pair is surprisingly good in group delay flatness. It does come shielded but it's not very flexible and looks ugly once somebody made a sharp bend.

Regards, Joerg

Speeds up to 240 MBit/s can be handed via cheap plain old-fashioned SUB-D connectors and STP, IF you take some precautions:

- chose the pinning so that the LVDS pairs are on two adjacent pins

- all pins surrounding a pair must be connected to ground

- use a shielded twisted pair cable with matched differential and single ended impedances

- keep the number of connectors in the line as low as possible

- once built, verify the LVDS signals for integrity / eye pattern.

Of course good HF connector is different.

If the cables were long enough for propagation delay differences to become ostensibly significant (aproaching 0.5nsec), the degradation in the pulse edge speed would probably be much more important - coax isn't perfectly non-dispsersive.

SMA is the mil spec minature coax connector, though SMB doesn't seem to be much cheaper.

----------- Bill Sloman, Nijmegen

I know you said you wanted small, but CAT 5/5e is damn cheap and very suitable for differential use -- it might be perfect if you can afford the larger RJ-45 jacks.

We're paying $2.31 for the basic pcb-mount right-angle SMB. I hate SMA's... you spend half your life screwing/unscrewing them.

MCXs are very nice, and there are tons of cheap cable assemblies on ebay.

John

I've used SCSI2 connectors and commercial cables for differential PECL over short distances, but differential prop delays (one wire of the pair relative to the other) added jitter, on the order of 30-40 ps rms over a meter or two of cable.

John

Yeah, it's sounding like a plain ole D9 would be the best pick for two or three shielded lvds signals. At least they're easy for everybody to get and wire to.

John

I don't have hands-on experience with LVDS, but wouldn't that need a controlled transmission-line impedance for the differential signal? Others have proposed guiding the lines through two separate MCX/RCA/whatever connectors here in s.e.d.- wouldn't that introduce a significant loop and thus lumped inductance to the signal path? Hmmm, I suppose if the termination is w.r.t the common ground (rather than between the diff wires) and the drive is strictly symmetrical the loop won't pose a problem.

The reason I'm interested is that I need to feed a number of twisted pairs through some sort of a multi-pin connector. So far I've been using the 1.27mm pitch double-row headers, using a pair of pins for the twisted pair and grounding the neighbouring pin pairs. I'm sort-of happy with this, even though opening the twist and feeding through the header introduces ~10nH worth of lumped inductance, but the 1GHz cutoff on the 60-ohm differential line due to this (and more importantly the phase shift present at lower frequencies) is tolerable.

That header is not suitable for John because there is no shell. I would be glad to hear of a shelled connector for multiple twisted pairs, and I suppose John could use those also. To make the matter not-too-easy I'd prefer the connector to be vacuum tight. One possibility would be to use Stycast to make a leaky connector vacuum tight. Any suggestions? (Microminiature D ? SCSI2 ? Others?)

Regards, Mikko