Author Topic: Differential pair transmission vs. single ended  (Read 1777 times)

0 Members and 1 Guest are viewing this topic.

Offline MomchiloTopic starter

  • Regular Contributor
  • *
  • Posts: 122
  • Country: de
Differential pair transmission vs. single ended
« on: July 21, 2019, 07:47:42 am »
Hi,
I have one control board and 3 additional boards that send and receive data from the control board over a distance of ~1m (3ft). All signals are single ended.
I have to design the boards, but I'm not sure about the best connection method.
Each of the 3 additional boards receive following signals from the control board:
  • 10x PWM @12kHz
  • CLK @16MHz
  • 4x digital signal lines @low frequency (turn on the boards etc.)
  • +15VDC @50mA
  • -15VDC @50mA
  • +5VDC  @200mA
  • Ground
  • Analog Ground
And in the other direction, every additional board send following signals to the control board:
  • 10x 1bit stream @16MHz
  • 5x analog signals -10V to 10V @2mA

Now I'm unsure if I send them as single ended or use differential line drivers and receivers. What would you do?

That would be at least 34 single ended lines with one ground and one analog ground connection. Where would you put the ground lines, between each signal? Then we're talking about ~70 lines per board. And could I use a differential pair cable or should it be straight ribbon cable? Or do know a better solution? 2 cables per board would be also okay, but not ideal.

If I use differential drivers and receivers for the PWM, CLK, analog and 1 bit stream signals, we're talking about 52 signal lines (26*2) + 4 slow digital lines + power and ground per board.
3 68 pin D-Sub cables with twisted pairs should work in this scenario or not? In my opinion this is the best alternative. But also the most expensive one, because I need 15 differential line drivers and 15 differential line receivers. 3 drivers and 12 receivers (each IC 4 channels) on the control board and on every additional board 3 receivers and 4 drivers, if I convert every single ended signal to
a differential signal pair.
So is it necessary to use differential signals? Maybe just for particular signals to reduce the number of additional ICs?
Thanks for every advice and your help :)

Best regards
Momchilo
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 17427
  • Country: us
  • DavidH
Re: Differential pair transmission vs. single ended
« Reply #1 on: July 21, 2019, 11:21:05 am »
Two wires is not only associated with differential signaling.  What matters is the ground return for the fast signals.  With differential signalling, another driven signal becomes the differential ground return but any fast signal needs a ground return whether differential or not.  So you are already looking at pairs of wires for your fastest signals.

This also applies to low frequency analog signals when "ground" is at different potentials at two locations.  Here the second wire serves to sense the remote ground.
 
The following users thanked this post: Momchilo

Offline JagV12

  • Contributor
  • Posts: 25
  • Country: fr
Re: Differential pair transmission vs. single ended
« Reply #2 on: July 21, 2019, 11:58:54 am »
It also very much depends on the electromagnetic environment. I personally would make prototypes for various solutions and see how far I can go with each...
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Differential pair transmission vs. single ended
« Reply #3 on: July 21, 2019, 01:47:19 pm »
Will this ever be applied to greater length runs, or is this a one-off circumstance?

Is this connectorized and user-facing (needs to withstand hairless gorillas), or can a plain ribbon cable suffice?

What bandwidth are the analog signals?  Slow like the PWM?

Offhand I would be looking at, say, RS-422 for the clock, since it takes priority (I assume it's synchronizing everything else), at least modest filtering for the PWM and slow digital signals (assume they're being generated by a nice sharp MCU, or some kind of CMOS output pin, but that you don't need the sharp edges transmitted as well), some kind of shielded or differential cable for the bit streams, and the analog signals can run in plain multiconductor cable.

The ideal cable would be a foil-shielded twisted pair (differential clock), foil-shielded multi (single ended bit streams), and loose or foil-shielded multi (slow digital, analog), all collected inside a single insulation jacket, and braid too if desired.  All the grounds tie together (braid, foil drains), to whatever enclosure or primary system ground is.  Digital ground probably, too.  Analog ground can come along as a DC reference, assuming you don't need much analog bandwidth (otherwise if you do, consider a separate braid on that bundle, or even individual coaxes).

Making do with off-the-shelf multiconductor cable, may be possible, but crosstalk may get too bad, mainly clock getting into everything it is accidentally nearby.

If ribbon is acceptable, I would try to alternate signal and ground as much as possible (note that power supplies also serve as ground, particularly if bypass capacitors are nearby on both ends), still use a diff pair for clock, and wrap the whole thing in a foil shield, grounded at both ends.  This could even be rolled up and jacketed, but it's still not terribly robust compared to a proper cable.


If this needs to run over greater lengths as well, I would seriously consider adding an FPGA with clock divider (8MHz will be more comfortable for RS-422 transceivers), clock multiplier (up to a few 100s MHz), and serdes (to combine all the digital signals into one or a few serial streams).  ADC/DACs can be applied to the analog levels as well.  The PWM could also be moved to a local generator, that's set through a command sequence (ala I2C or SPI -- preferably, derived from this serial stream), thus reducing its bandwidth.  Then a USB cable would be a quite adequate medium, or if the serdes can be configured for DC-balanced coding, Ethernet hardware and cabling could be used.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 
The following users thanked this post: Momchilo

Offline MomchiloTopic starter

  • Regular Contributor
  • *
  • Posts: 122
  • Country: de
Re: Differential pair transmission vs. single ended
« Reply #4 on: July 21, 2019, 03:32:13 pm »
Thanks a lot for your answers :)

Two wires is not only associated with differential signaling.  What matters is the ground return for the fast signals.  With differential signalling, another driven signal becomes the differential ground return but any fast signal needs a ground return whether differential or not.  So you are already looking at pairs of wires for your fastest signals.

This also applies to low frequency analog signals when "ground" is at different potentials at two locations.  Here the second wire serves to sense the remote ground.
So basically twisted pair and each pair is one signal and one ground with respect to the signal?

Will this ever be applied to greater length runs, or is this a one-off circumstance?
It's a prototype for a project in the university. So one off.

Is this connectorized and user-facing (needs to withstand hairless gorillas), or can a plain ribbon cable suffice?
Ribbon cable would be enough. I prefer D-Sub cable, because it's more rugged, screwed in and is shielded, but it's not required.

What bandwidth are the analog signals?  Slow like the PWM?
The bandwidth is around 200kHz.

Offhand I would be looking at, say, RS-422 for the clock, since it takes priority (I assume it's synchronizing everything else), at least modest filtering for the PWM and slow digital signals (assume they're being generated by a nice sharp MCU, or some kind of CMOS output pin, but that you don't need the sharp edges transmitted as well), some kind of shielded or differential cable for the bit streams, and the analog signals can run in plain multiconductor cable.
The clock synchronizes 10 ADCs (Sigma-Delta modulators). And these return the results as 1bit streams to a FPGA. The clock is only for synchronizing the sample rate of the ADCs.

The ideal cable would be a foil-shielded twisted pair (differential clock), foil-shielded multi (single ended bit streams), and loose or foil-shielded multi (slow digital, analog), all collected inside a single insulation jacket, and braid too if desired.  All the grounds tie together (braid, foil drains), to whatever enclosure or primary system ground is.  Digital ground probably, too.  Analog ground can come along as a DC reference, assuming you don't need much analog bandwidth (otherwise if you do, consider a separate braid on that bundle, or even individual coaxes).

Making do with off-the-shelf multiconductor cable, may be possible, but crosstalk may get too bad, mainly clock getting into everything it is accidentally nearby.

If ribbon is acceptable, I would try to alternate signal and ground as much as possible (note that power supplies also serve as ground, particularly if bypass capacitors are nearby on both ends), still use a diff pair for clock, and wrap the whole thing in a foil shield, grounded at both ends.  This could even be rolled up and jacketed, but it's still not terribly robust compared to a proper cable.
I thought maybe a 50 Pin twisted pair D-Sub cable like this one?
But of course a 50 or 60 pin ribbon cable is also a option. But they will be moved a lot. I don't think I could build a shielding for them that will last for some time. And shielded ribbon cables are very expensive.

In the attachment are two solutions, do you think one of these could work? Maybe a 68 pin shielded twisted pair D-Sub cable could be an option too, but it's more expensive than the 50 pin one.

And how would you filter the signals on the PCBs? I guess a low pass filter with a cap to ground and a resistor in line with the signals. But what should be the cutoff frequencies, 200kHz and 16MHz or other values are more practical?
And for which signals I should use a filter? Because we're talking about 3x10PWMs + 3x10bits + 3x5analog signals =75 filters + CLK.
Sorry for the many questions, but I never designed something like this and I have no idea what will work and what not.
« Last Edit: July 21, 2019, 03:34:06 pm by Momchilo »
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Differential pair transmission vs. single ended
« Reply #5 on: July 21, 2019, 08:29:00 pm »
Yeah, those pinouts should work pretty well.  You would of course want to get the CLK/!CLK into the pairs-version somehow. Or in another cable I suppose, if that's not so irritating.

Diff pairs being used unbalanced (one side grounded) isn't ideal, but the crosstalk should be okay as long as the bandwidth is kept below ~100MHz (being the 1/4 wave resonance of the 1m cable).

Filtering doesn't need to be fancy, and can be done in many ways.  It can be done digitally for example, though it's not ideal because you've got 100% quantization noise added -- a digital input pin is essentially a comparator, a 1-bit ADC.  Debouncing and properly phased clocking (sampling) are ways to implement this.

Transmitters, definitely should be impedance controlled in some manner; typically, CMOS output pins are a little too strong for ribbon or multiconductor cable, so a small series resistance should be added (source termination), typically 33 to 100 ohms.  Alternately, a bus driver can be used (the driver is even stronger, but also made to handle continuous heavy loads, not just intermittent switching loads), and termination placed at the destination end (typically a resistor divider with a Thevenin resistance matching the line).  Of course, the latter consumes a lot more power, which is annoying.

Analog transmitters should also be terminated, to avoid oscillation.  The cable is going to be <100pF, which shouldn't be a problem for many sources.  Just in case, you can add some resistance (100 ohms should be good enough) to help dampen it.

That addresses basic signal quality, giving high bandwidth without peaking.  To actually reduce bandwidth, we can increase the source resistance further, or modify it with reactances.

Ferrite beads are an easy one.  Pick one with an impedance (usually as measured at 100MHz) a few times the line impedance.  This isn't bad to begin with (it'll "take the edge off" better than a larger series-termination resistor will).  If we pair it with a shunt capacitor to make a not-quite LC lowpass, we can take off even more.

Ferrite beads have a diffusion characteristic, i.e., in the absorption band, Z ~ sqrt(freq) (and Arg(Z) ~ 45 degrees, i.e., equal parts R and X_L).  Working it out with a capacitor, we have a cutoff frequency around:
Fc = (Z_FB * C / sqrt(F_FB))^(-2/3)
Example: for a ferrite bead of 330 ohms at 100MHz (Z_FB and F_FB) and C = 470pF, Fc = 16MHz and Zc = 21Ω.  That's kind of a low Z, but with the driver impedance in series with it, or if we want to add more resistance, it'll roll off quite gently and not overshoot.  For this I think the minimum would be 100pF and somewhere between that and 470pF should give good filtering without screwing up signal levels too much.

ESD diodes are nice as well, maybe not strictly required with a shielded cable (and with female pins in the jack, less likely to be zapped by stray fingers) but nice to have.

For so many signals, you'll want to use arrays.  Quads seem to be the most economical.  Everything is available: resistors, capacitors, ferrite beads and ESD diodes, in chip form.

That'll serve nicely for the bit streams; the digital signals can be similar, or larger resistors and capacitors (1k, 1nF?).  Probably wouldn't bother with LC filtering.  (RC filters are also quite soft rolloff, and combined with the increased risetime, some digital filtering may still be desirable.)

Analog signals can be active filtered if you like, but that's probably not necessary.  Maybe reserve some space to patch in better filters, in the extreme case that it's necessary?  Idunno.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 
The following users thanked this post: Momchilo

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 17427
  • Country: us
  • DavidH
Re: Differential pair transmission vs. single ended
« Reply #6 on: July 22, 2019, 03:39:49 am »
Two wires is not only associated with differential signaling.  What matters is the ground return for the fast signals.  With differential signalling, another driven signal becomes the differential ground return but any fast signal needs a ground return whether differential or not.  So you are already looking at pairs of wires for your fastest signals.

This also applies to low frequency analog signals when "ground" is at different potentials at two locations.  Here the second wire serves to sense the remote ground.

So basically twisted pair and each pair is one signal and one ground with respect to the signal?

Exactly, this works very well when true differential signalling is not required.  Often a ribbon cable will either have a ground adjacent to every signal or a ground between every signal.  Twisted pairs will ground one wire of each pair.

The rules for transmission lines still need to be observed no matter what configuration is used.  For logic without differential signaling, the simplest way is parallel termination at the receiver.
 
The following users thanked this post: Momchilo


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf