resetting a filter

True but a U part is what you want there I think.

fields?

Yikes! That's right.

Open-drain drive would solve any most TC variation, as I posted a couple of days ago in...

Subject: Re: resetting a filter Date: Thu, 09 Aug 2012 17:21:54 -0700 Message-ID: ...Jim Thompson

Agreed, depending on how you define "much". It will leak at RF frequencies, which admittedly is not a consideration for the delay line application under discussion. The problem is balance. Each wire has to have the same capacitance to ground, or it will radiate. In the distant past, I build a capacitive bridge that measure this balance. The distance between the windings and any nearby part or PCB has to be equal for both wires in the bifilar pair. That can be done with a fairly large PCB, or inside a carefully constructed can, but not scramble wound on a coil form or twisted into a toroid pretzel. To reduce balance issues and reduce coupling between turns, delay lines are usually solenoid wound on rods, rather than around toroids, or multi-layer forms, which would be more compact. With a toroid, there's some added capacitance between windings where they touch adjacent turns, which causes coupling between turns, which causes more leakage.

Another example is CAT5 cable. If you peel back the insulation, you'll notice that the twist rate on each or the four pairs is different. That's mostly to minimize mutual coupling between pairs (NEXT/FEXT) mostly through minimizing the points of contact. CAT6 does it better by adding a plastic separator between pairs.

Ours was a labyrinth of spring tensioned lever arms with pulleys at the ends. I vaguely recall three or four short tension arms per supply reel. At low tension, the supply reel would bounce, causing wire breakage. At high tension, small wires would break. The zone of acceptable tension was small and was affected by many things.

Most of the problems were due to lubrication issues caused by using inferior oil for lubing the tension arms bearings and supply reel friction mechanism. I got fed up late one night, tore the tensioners apart, gave them a solvent bath, and relubed everything with overpriced clock oil (no evaporation, no gum, and temperature stable). After adjusting, everything ran perfectly. The problem was that the bearings would get warm as the machine ran, changing the viscosity of the oil. That had a small effect on the tension, which was enough to break fine wires.

CAT6F/UTP still varies the twists, which adds skew. CAT6U/FTP shields each pair, so they don't have to vary the twist, but the attenuation is worse. If it ain't something, it's something else.

No shit. It is part of the design spec/rule.

Did you think they were going to stop "still varying the twists"?

It is one way to tell when your lines (patch cables) are wired wrong.

Try using a different set of pairs on the connectors. Create the same patch from a continuity POV, and find out it fails because you have placed the wrong pairs in the wrong locations.

The spec itself utilizes this to determine a properly wired link.

Also, it isn't about "UTP", idiot. It is the ENTIRE spec, so it doesn't matter if the wire is shielded or not. If it is CAT6 it gets varied twist rate pairs incorporated into it.

AlwaysWrong.

We've opened up, and measured the prop delay of, a bunch of CAT6A cables. The ones with individual shielded pairs that we checked had identical (nearly identical?) twists and much lower prop delay skew (and more HF loss) than the ones with unshielded pairs. We're seeing skews like 40 ns/100m for UTP, more like 10 for STP and FTP. But the risetimes are better for UTP.

Varying the twists, unfortunately, makes the prop speed different on the different pairs. That matters to our current application.

Yep. However, not all individually shielded cables have a constant twist rate. This indoor CAT6 cable does (enlarge the photo):

while this outdoor version does not (enlarge the photo):

Hopefully, the skew is within spec.

Also, some cables with variable twist rates have different doping in the insulation to produce different dielectric constants. This also changes the velocity factor (speed of propagation as a percentage of speed-o-light) which somewhat compensates for the differences in wire length caused by different twist rates. I don't know if such cables are actually available. Some details:

Some amount of twist difference may be for mechanical reasons.

Delay skew in high school physics class? I missed that somehow.

Clearly ethernet gear has to compensate for the skew, by fifoing lanes or something. At GbE rates, the differences between pairs will run into many tens of bits.

Ever heard of packetization? Framing headers?

Sheesh. Wake up, people.

When it gets up into the 10GbE stuff, they start synching entire system elements up with distributed pulses. I think we call 'em "reference signals".

Word salad. If you actually understand any of this, explain how GbE corrects for cable skew, and how much it can actually tolerate.

Too interested in your electronics projects to pay attention in class? Tou seem to have had the same probblem at university too.

Unless they've changed the basic idea since I got immersed in local area networks in the early 1980's, Ethernet is a serial packet-based broadcast system that doesn't care about skew between the various serial channels sending out and returning the various packets.

At the bottom level, the packets are broadcast without any guarantee of timing - collision avoidance can impose an arbitrary delay on when the packet gets broadcast - and the error detection mechanism just discards corrupted packets, and relies on a higher level protocol to asked for the discarded packet to be retransmitted.

-- Bill Sloman, Nijmegen

Sounds like a lot of bull to me and completely irrelevant to the discussion.

Jamie

He picks up buzzwords while he's carrying stuff around for the engineers.

GbE uses more than one pair in each direction, at the symbol rate they obviously need to be aligned

that is at a totally different level

-Lasse

rote:

taff

r 42

ore

t.

e,

at

elds?

You really shouldn't drive twisted pair single ended.

ECL line drivers/receivers always have complementary outputs and differential receivers. Using them - as PECL - in a basically TTL system with negative ground, makes the system less reliable, because any chunk of grounded metal (like a scope probe ground lead) and destroy an ECL output in no time.

Helwett-Packard admitted the 90% of the returns for repair on their laser interferometer electronics were caused by just that problem. Dabbing blobs of potting compound onto the exposed connection can help.

LVDS parts weren't quite as fast when I last looked - which was long time ago - but they are designed to work with regular logic levels.

-- Bill Sloman, Nijmegen

Just put a crossover cable in halfway along the run. ;)

Cheers

Phil Hobbs

I got A's in high school chemistry and physics. Went to the State science fair (got first place in physics) and the National in Baltimore (didn't win, but it was fun. Hung around with Amory Lovins!) I also won the Navy Science Cruiser Award (the prize was a week in the US Navy, in Charleston) and a trip to Bell Labs, where I had lunch with Walter Brattain.

But my physics teacher never mentioned Ethernet cable skew.

Got A's there, too. Easy.

The ancient stuff, from your day, used coax. No twisted pair skew. I remember the coax as being fairly flakey about termination and stuff. One bad tap or a short stub would bring the whole network down.

That's really not a bad idea! I'd need two crossovers, three segments, to preserve the pairing (Ethernet mostly doesn't care any more, but I do).

We could go to a skew-insensitive data protocol, I guess. What we're doing now, per customer requirements, is kinda nuts. A data packet is a burst clock, plus two RZ data lanes.