[time-nuts] Opera coordinator has resigned
magnus at rubidium.dyndns.org
Sat Mar 31 21:55:10 UTC 2012
On 03/31/2012 10:29 PM, Jim Lux wrote:
> On 3/31/12 12:24 PM, Achim Vollhardt wrote:
>> Dear John and all,
>> I do work in high energy physics and we use LEMO and other standards.
>> For some years now, I have started to advertise against LEMO (in
>> particular the LEMO 00 size), as it is VERY sensitive to mechanical
>> defects and partial connection (yes, you can ..). We have found very
>> often defective LEMO connections, which could only be detected via time
>> domain reflectometry. Or LEMO interfaces, which changed impedance
>> significantly when rotating them in the fully mated position.. this list
>> could be extended a lot more. And this famous click when mating is
>> inaudible in typical high energy physics electronics barracks.. too much
>> fan noise.
>> In addition, for the very fast signals of modern DAQ systems, LEMO 00 is
>> just not up to speed (literally) anymore. If size is not of ultimate
>> importance, we switched to SMA, or SMC if size matters.
> On bad connectors.. "K" connectors are horrible.. they can intermate
> with SMA, but they don't tolerate the range of SMA mechanical variation,
> so it's possible to damage a K jack by mating an SMA plug with it. Not
> all SMA plugs, but some percentage of them. And it's not obvious that
> the jack has been damaged unless you measure it. You can mate a K or SMA
> (properly) with it, it will work, mostly, but now, there's a tiny gap in
> the center conductor, sometimes, depending on the connector flex, etc.
> We have a dreadful connector at JPL which is a mini twinax (used for
> balanced pairs, like MIL-STD-1553). The key is a stamped metal bump so
> small that it's easy to mismate in the 180 degree reversed position.
> There's a much better version which is hermaphroditic (each side has one
> male and one female pin) and it can only mate one way. But that other
> connector hangs on, and on, because you get the "device A uses connector
> type X", so you build test equipment with mating Connector Type X, then
> the next device B has that connector, so we can re-use the test
> harnesses. The next batch of test equipment is made to be compatible
> with device B, and so it goes. The last people I know (not at JPL) who
> bought those connectors called up the mfr and was quoted a spectacularly
> long lead time (6 months or a year), so they asked who else had bought
> them, maybe they could buy the half dozen they needed. Naturally, JPL
> had bought the last batch, something like 15 or 20 years ago, and of
> course, we had dozens of them sitting in bonded stores.
> I was at an IEEE conference a few years ago on high speed interconnects
> (10 Gbps is slow for those guys), and a bunch of the folks were talking
> about how too much time is being spent on trying to get impedance
> controlled connectors, instead of reliable connectors. Their point was
> that in any real high speed (which would, for time nuts be "high timing
> precision") system, you already have multiple transitions: die to IC
> package pin to PWB trace to connector pin to connector, before you worry
> about the connector's potential mismatch. These other ones are even
> harder to control the impedance on, and trying to do so makes the boards
> hard to route, etc, so why not bite the bullet and just implement a good
> adaptive equalizer in your high speed interface. Then you can use a
> mechanically robust connector that has a positive mating, can't be
> mismated or partially mated, and in fact, the equalizer can tell you if
> the cabling is screwed up.
> From a cost standpoint, it might actually be cheaper. Implementing the
> equalizers in a standard high speed SERDES doesn't cost much more in
> silicon (yes, the IP development costs money, so the selling price is
> higher), and you save a lot in the rest of the system.
We have equalizers as well as pre/post emphesis to shape the eye of the
signal. It's really required. We adjust the sampling point on our bits.
That's standard business today.
More information about the time-nuts