logic levels over long cable runs

We have a customer who wants us to generate some fast pulse-type timing signals and run them around a big machine. The longest run will be around 150 feet, and he wants sub 1 ns resolution on pulse timings and widths, and below 100 ps RMS jitter.

So we measured some CAT6A and coax cables.

ftp://jjlarkin.lmi.net/DDG_cables.zip

Note the characteristic "drool" step responses one gets from a lossy transmission line. As expected, the 10/90 rise time goes as the square of the cable length. Good coax is much better than CAT6.

None of this looks promising. Coax is faster, but is still slow and will have ground loop problems. We want DC coupling, so transformers are out.

Anybody who gets 10Gbit ethernet through CAT6 must be doing some very serious adaptive equalizing. We could equalize CAT6, but that would be a nasty mess.

Looks like we should go fiber. We'll probably need to run 16 signals

150 feet, so we'll need 16 laser/pin diode sets and 16 fibers, or maybe look into the MPO or MTP fiber ribbon cable things, if we can find the driver/receiver things for the ends.

Another nuisance is that most tosa/rosa gadgets assume telecom type apps, with AC coupled laser drivers and agc/AC coupled receivers, which don't work with baseband timing pulses.

John

Reply to
John Larkin
Loading thread data ...

We have a customer who wants us to generate some fast pulse-type timing signals and run them around a big machine. The longest run will be around 150 feet, and he wants sub 1 ns resolution on pulse timings and widths, and below 100 ps RMS jitter.

So we measured some CAT6A and coax cables.

ftp://jjlarkin.lmi.net/DDG_cables.zip

Note the characteristic "drool" step responses one gets from a lossy transmission line. As expected, the 10/90 rise time goes as the square of the cable length. Good coax is much better than CAT6.

None of this looks promising. Coax is faster, but is still slow and will have ground loop problems. We want DC coupling, so transformers are out.

Anybody who gets 10Gbit ethernet through CAT6 must be doing some very serious adaptive equalizing. We could equalize CAT6, but that would be a nasty mess.

Looks like we should go fiber. We'll probably need to run 16 signals

150 feet, so we'll need 16 laser/pin diode sets and 16 fibers, or maybe look into the MPO or MTP fiber ribbon cable things, if we can find the driver/receiver things for the ends.

Another nuisance is that most tosa/rosa gadgets assume telecom type apps, with AC coupled laser drivers and agc/AC coupled receivers, which don't work with baseband timing pulses.

John

---------------------

I couldn't make out what termination resistor you used for these tests.

mike

Reply to
m II

It's often done with a combination of preemphasis and decision-feedback equalization. In the fibre world, you can get 4x and 12x transceivers at 850 nm multimode, but they're not especially cheap. They do have pretty nice EMI resistance compared with wire, especially in factory situations.

The big problem with the dribble-up step response is inter-symbol interference. If the duty cycle is low enough that the transient is all finished by the time the next pulse arrives, you might be able to set the threshold lower down on the step and do well enough for your purposes.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
 Click to see the full signature
Reply to
Phil Hobbs

We drove one end from a Tek 11801 sampling scope TDR head, 30 ps steps, and terminated the other end in another sampling head. The scope pics are what comes out the far end of the cable. The termination for coax was 50 ohms, and the termination for the twisted pairs was 2x50 = 100 ohms. We did measure the CAT6A differential impedance and it was dead on, 99.6 ohms.

The SD24 sampling head will do single-ended or differential TDR steps.

John

Reply to
John Larkin

Since we're sending various-width pulses at unpredictable duty cycles, ISI shows up, to us, as pulse width errors. Our expected pulse widths straddle the range where these cable responses will make trouble. Since the customer is picky about timings, we'd have to carefully equalize each installation as a function of cable length, probably manually, times 16 runs. Sounds like not-fun.

John

Reply to
John Larkin

One other possibility in the low-duty-cycle case would be to hit it with a 15 volt pulse at the TX end and use a series clipper or something like that at the RX end to square it up without blowing the logic input.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
 Click to see the full signature
Reply to
Phil Hobbs

Hmmm, we could send the rising and falling edges on separate pairs, and do a little logic at the receiver. That would take twice as many pairs.

John

Reply to
John Larkin

The brute force method would be to have repeaters every 50 feet, or possibly to send bipolar pulse pairs, where the drool would pretty well cancel out--you could send a +- pair to start a pulse and a -+ one to end one, for instance. Trigger on the edge in the middle.

Even with fibre, you have to worry about signals rattling round inside the fibres causing problems at the 1% level.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
 Click to see the full signature
Reply to
Phil Hobbs

Are these equal distance timing pulses ? If not, how much in advance is the pulse time known ?

Could you use some GPS style PRN sequence and at each receiver have an identical PRN generator, which is synchronized with the incoming PRN sequence (early/late detection). This would average out most of the jitter caused by ISI in the band limited transfer medium.

This would also allow AC (transformer) connection.

If the timing is not at regular intervals, but known some time in advance, in the idle state run some preamble (101010..) then switch to the PRN and the actual timing is at the end of the PRN sequence. Running a 1023 bit sequence at 1 GHz would require that the actual timing is known 1 us in advance (corresponding to 200 m propagation delay).

Anyway, you need some propagation delay compensation for different receivers.

Using separate non-identical lasers for each fiber will add timing uncertainity to the pulse. Using a single laser and passive fiber splitter will avoid this problem,

What about using two lasers and two fibers, driven in a complementary way. On the receiver side, use an analog comparator to detect when one rising signal exceeds the falling signal and hence compensate for the amplitude variations.Using two lasers at different wavelengths in this complementary fashion, would work with a single fiber/station.

Reply to
upsidedown

Maybe so, but that sounds tricky, getting sub-ns accurate results from links with tens of ns risetimes. My customer has a few gigabucks at stake and is on record as being opposed to tricky.

Hmmm, maybe so. We'd better allow for angle-polished fiber in case the reflections become a problem.

John

Reply to
John Larkin

If you want differential drive/receive (which is best for long lines), you should also be looking at twinax, with differential ECL logic (like 100k series). Cat6 is intended for low crosstalk, but is otherwise not well optimized for speed and straggle (it only works for gigabit at short distances, with multinanosecond timing errors allowed).

Reply to
whit3rd

If you shape the pre-emphasis properly and use a comparator as the receiver there's no need for a clipper. ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
 Click to see the full signature
Reply to
Jim Thompson

The resolution requirement would suggest a 2 GHz sine carrier driving a PLL at each receiver.

Omitting one cycle of the carrier at the start of the pulse and two cycles at the end of the pulse could be used to gate the PLL at the next transition.

Make 16 equally long cables and coil the excessive length.

If this is not accurate enough due to aging or temperature variations, then you would have to use a return path to calculate the round trip time.

Reply to
upsidedown

MC10EL89 is a nice part. Fast, and almost 2 volts of swing at each output pin.

Cat6 is intended for low crosstalk, but is

Even high-quality teflon coax has a risetime of 24 ns at 150 feet, and I'd imagine twinax would be worse. There just may be no reasonable (and safe) way to pipe sub-ns-edge-precision pulses over copper at this distance.

John

Reply to
John Larkin

"John Larkin" wrote in message news: snipped-for-privacy@4ax.com...

I think the ground loops will play havoc equally as well.

Cheers

Reply to
Martin Riddle

If you need cost effective fibre parts, you might look at fibre channel disk array transceivers, which plug in to the disk arrays and also the fc hubs. I think the current spec is 4Gb/s, but possibly 8. Also of course, 10Gb/s networking which uses fibre as well as copper.

Not an easy task, and no cheap solutions, whichever way you do it...

Regards,

Chris

Reply to
ChrisQ

Take a look at the thing called "SNAP12". However, it's only 12 lanes.

Here's the so-called MSA (multi-source agreement) doc:

formatting link

We use the Avago version and pay about $900 for the transmitter/receiver pair, and we buy the 10Gbps/lane version in order to run 100G ethernet. I don't know how much the fiber cables are.

Avago p/ns:

AFBR-776BxxxZ AFBR-786BxxxZ

formatting link

Bob

--
== All google group posts are automatically deleted due to spam ==
Reply to
BobW

...

10G over copper is a PAM-16 system and it needs THP coding (pre-emphasis based on estimated channel on the receive side), deterministic noise cancellation (fext, next and echo) and heavy-duty FEC (LDPC). Even 1000BaseT is PAM-5. Twisted pair at the length you're looking at is not a good medium to transmit pulses. Fiber is a much better bet.
--
Muzaffer Kal

DSPIA INC.
 Click to see the full signature
Reply to
Muzaffer Kal

Those are interesting. Both the TX and RX are DC coupled, which is sort of unusual. One problem we have with a lot of parts is that they are AC coupled, often with receive AGC, so don't work with baseband pulses.

Thanks; we'll research these some more.

John

Reply to
John Larkin

Sure, but doing the preemph right requires active tweaking in general, whereas hitting it hard and clipping it off effectively gives you the

2%-20% rise time, which is much much faster than the 10%-90% rise. (The pulse shape is trying to be proportional to sqrt(t).)

IBM uses something like 8-sample FIRs to put a big fat ring on the leading TX edge, with DFE at the receiver, and active tweaking of both. They also use 'elastic buffers', i.e. adjustable RX clock phase per pin, to make sure they don't violate setup and hold time requirements.

John's production run will probably be too small to justify that much engineering, and it sounds like a pretty high-end application, so lasers are probably a good solution, assuming they have somebody that knows the ropes.

A short tone burst, like 1 cycle, won't show significant ISI because it produces the second derivative of the step response, so long slow ramps go away. Triggering on the zero crossing in the middle of the burst should be very stable. Whether I'd guarantee it to be stable to 100 ps, I don't know. In the lab, sure, but in a factory or a big laser facility (like the NIF), the EMI environment is pretty awful.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
 Click to see the full signature
Reply to
Phil Hobbs

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.