Pulsing to circumvent noise?

[eromlignod]

Gentlemen:

I have an application where I have a large number of lines that
transmit square waves across about a six foot distance. It is an
extremely noisy environment: electromechanical relays, other signals
in close proximity on a PCB, etc. All I need from each square wave is
to determine its period by accurately measuring the time between two
successive low-to-high transitions. I have tried shielding the hell
out of each individual line and all sorts of other measures, but I
still get enough residual noise, especially at transitions, to cause
problems.

Rather than transmitting the whole wave, would it be possible to only
transmit a short pulse (say 1.0 us) for each low-to-high transition at
the source? Since the signal spends most of its time grounded, do you
think these short pulses would be able to escape the noise? If I can
get reasonably clean pulses, I can still get an accurate period
measurement.

Thanks for any help.

Don

 

[Rich Webb]

On Mon, 13 Jun 2011 07:29:38 -0700 (PDT), eromlignod
<eromlignod@aol.com> wrote:

Gentlemen:

I have an application where I have a large number of lines that
transmit square waves across about a six foot distance. It is an
extremely noisy environment: electromechanical relays, other signals
in close proximity on a PCB, etc. All I need from each square wave is
to determine its period by accurately measuring the time between two
successive low-to-high transitions. I have tried shielding the hell
out of each individual line and all sorts of other measures, but I
still get enough residual noise, especially at transitions, to cause
problems.

Rather than transmitting the whole wave, would it be possible to only
transmit a short pulse (say 1.0 us) for each low-to-high transition at
the source? Since the signal spends most of its time grounded, do you
think these short pulses would be able to escape the noise? If I can
get reasonably clean pulses, I can still get an accurate period
measurement.

What does your receiving side look like? If it's noise at the rising
edge that's giving you problems, have you tried using a comparator with
a lot of hysterisis? Possibly also have the comparator fire a
retriggerable one-shot for a belt and suspenders approach.

--
Rich Webb Norfolk, VA

 

[Sjouke Burry]

eromlignod wrote:

Gentlemen:

I have an application where I have a large number of lines that
transmit square waves across about a six foot distance. It is an
extremely noisy environment: electromechanical relays, other signals
in close proximity on a PCB, etc. All I need from each square wave is
to determine its period by accurately measuring the time between two
successive low-to-high transitions. I have tried shielding the hell
out of each individual line and all sorts of other measures, but I
still get enough residual noise, especially at transitions, to cause
problems.

Rather than transmitting the whole wave, would it be possible to only
transmit a short pulse (say 1.0 us) for each low-to-high transition at
the source? Since the signal spends most of its time grounded, do you
think these short pulses would be able to escape the noise? If I can
get reasonably clean pulses, I can still get an accurate period
measurement.

Thanks for any help.

Don

A simple rc network on each line, followed by a Schmitt trigger would do
nicely.

 

[Globemaker]

On Jun 13, 10:29 am, eromlignod <eromlig...@aol.com> wrote:

Gentlemen:

I have an application where I have a large number of lines that
transmit square waves across about a six foot distance.  It is an
extremely noisy environment: electromechanical relays, other signals
in close proximity on a PCB, etc.  All I need from each square wave is
to determine its period by accurately measuring the time between two
successive low-to-high transitions.  I have tried shielding the hell
out of each individual line and all sorts of other measures, but I
still get enough residual noise, especially at transitions, to cause
problems.

Rather than transmitting the whole wave, would it be possible to only
transmit a short pulse (say 1.0 us) for each low-to-high transition at
the source? Since the signal spends most of its time grounded, do you
think these short pulses would be able to escape the noise? If I can
get reasonably clean pulses, I can still get an accurate period
measurement.

Thanks for any help.

Don
1 yes.

2 no.

tv remote controls sometimes use an optical burst of band limited
signal near 50khz, with a matched filter at the receiving end. A
similar plan could help your signaling to reject all except that
unique signal.

 

[George Herold]

On Jun 13, 10:29 am, eromlignod <eromlig...@aol.com> wrote:

Gentlemen:

I have an application where I have a large number of lines that
transmit square waves across about a six foot distance.  It is an
extremely noisy environment: electromechanical relays, other signals
in close proximity on a PCB, etc.  All I need from each square wave is
to determine its period by accurately measuring the time between two
successive low-to-high transitions.  I have tried shielding the hell
out of each individual line and all sorts of other measures, but I
still get enough residual noise, especially at transitions, to cause
problems.

Rather than transmitting the whole wave, would it be possible to only
transmit a short pulse (say 1.0 us) for each low-to-high transition at
the source? Since the signal spends most of its time grounded, do you
think these short pulses would be able to escape the noise? If I can
get reasonably clean pulses, I can still get an accurate period
measurement.

Thanks for any help.

Don

A few more details might help.
Is 1 us timing accuracy good enough?

What's the noise look like on the signal?

Why is there more noise at the transistions? Do you have ringing?
Maybe cross-talk between the numerous signal lines?

How fast is the rise time of the square wave?

Can you ground the cable shield at both ends?

Is the cable shield have braided copper, aluminum foil, or both?

George H.

 

[Rich Grise]

eromlignod wrote:

I have an application where I have a large number of lines that
transmit square waves across about a six foot distance. It is an
extremely noisy environment: electromechanical relays, other signals
in close proximity on a PCB, etc. All I need from each square wave is
to determine its period by accurately measuring the time between two
successive low-to-high transitions. I have tried shielding the hell
out of each individual line and all sorts of other measures, but I
still get enough residual noise, especially at transitions, to cause
problems.

Rather than transmitting the whole wave, would it be possible to only
transmit a short pulse (say 1.0 us) for each low-to-high transition at
the source? Since the signal spends most of its time grounded, do you
think these short pulses would be able to escape the noise? If I can
get reasonably clean pulses, I can still get an accurate period
measurement.

Typically, in industrial settings, which are horrendously noisy, they

use a 4-20 mA current loop with a twisted pair, and even shielded
twisted pair isn't out of the question.

Of course, that's assuming that it's not too late to redesign your device.

Good Luck!
Rich

 

[Jamie]

eromlignod wrote:

Gentlemen:

I have an application where I have a large number of lines that
transmit square waves across about a six foot distance. It is an
extremely noisy environment: electromechanical relays, other signals
in close proximity on a PCB, etc. All I need from each square wave is
to determine its period by accurately measuring the time between two
successive low-to-high transitions. I have tried shielding the hell
out of each individual line and all sorts of other measures, but I
still get enough residual noise, especially at transitions, to cause
problems.

Rather than transmitting the whole wave, would it be possible to only
transmit a short pulse (say 1.0 us) for each low-to-high transition at
the source? Since the signal spends most of its time grounded, do you
think these short pulses would be able to escape the noise? If I can
get reasonably clean pulses, I can still get an accurate period
measurement.

Thanks for any help.

Don
Use a twisted balanced pair of wires.. a differential TX and RX circuit

or a current loop which will regulate the current to maintain it on the
other end.
If you are using shielded wire, connect the shield to ground only on
one end.

If you want to keep with your current method, then lower the R value
of the load on each end to the lowest R and gain the most load the
transmitter can handle and use hysteresis type logic on the RX end, like
a Schmitt Gate,inverter or comparator with a hysteresis window in it.

Jamie

 

[whit3rd]

On Monday, June 13, 2011 7:29:38 AM UTC-7, eromlignod wrote:

Gentlemen:

I have an application where I have a large number of lines that
transmit square waves across about a six foot distance. It is an
extremely noisy environment...

Rather than transmitting the whole wave, would it be possible to only
transmit a short pulse (say 1.0 us) for each low-to-high transition at
the source?

Yes, that'd be possible. You could capacitor-couple and the
square wave would become alternating positive and negative
pulses, then (at the receive end) a Schmitt trigger (NE555 actually
is a pretty good one for this application) will regenerate the
square wave. This kind of operation (bandwidth limiting the
sent signal, then using a receiver that's insensitive in a wide
noise-margin range) is common.

The 'short pulse' low-to-high is no more important than the
alternate high-to-low pulse, in this scheme. The '555 needs
both a SET and a RESET event in each cycle.

 

[Rich Webb]

On Mon, 13 Jun 2011 18:15:19 -0700 (PDT), eromlignod
<eromlignod@aol.com> wrote:

Thanks for the answers.

The output is from a 339 comparator with a 3K pull-up resistor. It's
a quad comparator and each circuit only uses three of them, so there
is one that is wasted. Is there any way to take advantage of the
extra comparator?

How are you using the comparator? One really must have it set up with
positive feedback to achieve enough hysterisis. If it's just wired with
a fixed reference versus your input, yes it will go nuts at transitions.

--
Rich Webb Norfolk, VA

 

[eromlignod]

Thanks for the answers.

The output is from a 339 comparator with a 3K pull-up resistor. It's
a quad comparator and each circuit only uses three of them, so there
is one that is wasted. Is there any way to take advantage of the
extra comparator?

Don

 

[eromlignod]

I meant the circuit output before the long lines. The comparator is
generating the square wave. When the circuit is operated alone, it
produces a beautiful, crisp square wave. When I add the other
circuits I get crosstalk between signals and noise from the coils.

Don

 

[George Herold]

On Jun 13, 6:07 pm, Rich Grise <ri...@example.net.invalid> wrote:

eromlignod wrote:
I have an application where I have a large number of lines that
transmit square waves across about a six foot distance.  It is an
extremely noisy environment: electromechanical relays, other signals
in close proximity on a PCB, etc.  All I need from each square wave is
to determine its period by accurately measuring the time between two
successive low-to-high transitions.  I have tried shielding the hell
out of each individual line and all sorts of other measures, but I
still get enough residual noise, especially at transitions, to cause
problems.

Rather than transmitting the whole wave, would it be possible to only
transmit a short pulse (say 1.0 us) for each low-to-high transition at
the source? Since the signal spends most of its time grounded, do you
think these short pulses would be able to escape the noise? If I can
get reasonably clean pulses, I can still get an accurate period
measurement.

Typically, in industrial settings, which are horrendously noisy, they
use a 4-20 mA current loop with a twisted pair, and even shielded
twisted pair isn't out of the question.

Of course, that's assuming that it's not too late to redesign your device..

Good Luck!
Rich- Hide quoted text -

- Show quoted text -

I've never done any industrail stuff. So 4-20 mA means I get to hang
a resistor on the end (impedance of twisited pair is ... 100 ohms?)
and measure 400mV as 'zero' and 2 Volts as 'one'?

George H.

 

[George Herold]

On Jun 13, 9:38 pm, eromlignod <eromlig...@aol.com> wrote:

I meant the circuit output before the long lines. The comparator is
generating the square wave.  When the circuit is operated alone, it
produces a beautiful, crisp square wave.  When I add the other
circuits I get crosstalk between signals and noise from the coils.

Don

I haven't dealt with much cross talk. The brute force approach, is
separate shields for each line. I'd try and keep everything grounded
if it's possible. (both ends!)

Sometimes it's good to slow the transitions down, less dV/dt. A bit
of C on the output of a comparator can slow it down..... but what's
the C of 6 feet of twisted pair?

Oh, have you looked at your power supply when everythings on?

George H.

 

[Rich Grise]

George Herold wrote:

On Jun 13, 6:07 pm, Rich Grise <ri...@example.net.invalid> wrote:
eromlignod wrote:
I have an application where I have a large number of lines that
transmit square waves across about a six foot distance.  It is an
extremely noisy environment: electromechanical relays, other signals
in close proximity on a PCB, etc.  All I need from each square wave is
to determine its period by accurately measuring the time between two
successive low-to-high transitions.  I have tried shielding the hell
out of each individual line and all sorts of other measures, but I
still get enough residual noise, especially at transitions, to cause
problems.

Rather than transmitting the whole wave, would it be possible to only
transmit a short pulse (say 1.0 us) for each low-to-high transition at
the source? Since the signal spends most of its time grounded, do you
think these short pulses would be able to escape the noise? If I can
get reasonably clean pulses, I can still get an accurate period
measurement.

Typically, in industrial settings, which are horrendously noisy, they
use a 4-20 mA current loop with a twisted pair, and even shielded
twisted pair isn't out of the question.

Of course, that's assuming that it's not too late to redesign your
device.

I've never done any industrail stuff. So 4-20 mA means I get to hang
a resistor on the end (impedance of twisited pair is ... 100 ohms?)
and measure 400mV as 'zero' and 2 Volts as 'one'?

Yes, and the impedance or even resistance of the wires doesn't make any

difference because the 4-20 mA supply has voltage compliance, i.e., it
will develop however many volts as it takes to drive the current.

It has the added benefit, or maybe just a bonus - if there's 0 current,
that indicates "fault."

Some of these might be useful:
http://www.google.com/search?hl=en&source=hp&q=4-20+ma+loop

Cheers!
Rich

 

[Rich Webb]

On Mon, 13 Jun 2011 18:38:39 -0700 (PDT), eromlignod
<eromlignod@aol.com> wrote:

I meant the circuit output before the long lines. The comparator is
generating the square wave. When the circuit is operated alone, it
produces a beautiful, crisp square wave. When I add the other
circuits I get crosstalk between signals and noise from the coils.

Are you using the 339s as your line drivers? If so, not a good plan.
They're not really designed to drive long lines in a noisy environment.

Try shifting your interconnects to using differential signaling over
twisted pairs. Line drivers/receivers labeled for RS-422/485 are easy to
find (and typically multiple-sourced) and would probably help a lot.
Common mode interference (a transient that affects both signals) will,
for the most part, disappear.

Common CAT5 cable would probably be sufficient for the twisted pair. If
you're in an extreme environment, shielded twisted pair may be needed,
but I'll bet that just moving to the differential drivers/receivers
clears up most, if not all, of the problems.

--
Rich Webb Norfolk, VA

 

[krw@att.bizzzzzzzzzzzz]

On Tue, 14 Jun 2011 08:04:20 -0400, Rich Webb <bbew.ar@mapson.nozirev.ten>
wrote:

On Mon, 13 Jun 2011 18:38:39 -0700 (PDT), eromlignod
eromlignod@aol.com> wrote:

I meant the circuit output before the long lines. The comparator is
generating the square wave. When the circuit is operated alone, it
produces a beautiful, crisp square wave. When I add the other
circuits I get crosstalk between signals and noise from the coils.

Are you using the 339s as your line drivers? If so, not a good plan.
They're not really designed to drive long lines in a noisy environment.

Shouldn't be *too* bad if you drive a current loop and detect with another
'339. They are a little weak, though.

Try shifting your interconnects to using differential signaling over
twisted pairs. Line drivers/receivers labeled for RS-422/485 are easy to
find (and typically multiple-sourced) and would probably help a lot.
Common mode interference (a transient that affects both signals) will,
for the most part, disappear.

They aren't cheap and suck a lot of power. A '339 is less than $.10. An RS422
drivers and receivers go closer to a buck. '339s come in four-packs. Use two
(half). ;-) I agree, though. RS-422 is the way to go if the budget allows.

Common CAT5 cable would probably be sufficient for the twisted pair. If
you're in an extreme environment, shielded twisted pair may be needed,
but I'll bet that just moving to the differential drivers/receivers
clears up most, if not all, of the problems.

CAT-5 will almost always be good enough. We use it for runs specified up to
1500' (with RS-422 from DC to 1Mb) and have never had a problem reported.

 

[Rich Webb]

On Tue, 14 Jun 2011 09:10:34 -0500, "krw@att.bizzzzzzzzzzzz"
<krw@att.bizzzzzzzzzzzz> wrote:

On Tue, 14 Jun 2011 08:04:20 -0400, Rich Webb <bbew.ar@mapson.nozirev.ten
wrote:

On Mon, 13 Jun 2011 18:38:39 -0700 (PDT), eromlignod
eromlignod@aol.com> wrote:

I meant the circuit output before the long lines. The comparator is
generating the square wave. When the circuit is operated alone, it
produces a beautiful, crisp square wave. When I add the other
circuits I get crosstalk between signals and noise from the coils.

Are you using the 339s as your line drivers? If so, not a good plan.
They're not really designed to drive long lines in a noisy environment.

Shouldn't be *too* bad if you drive a current loop and detect with another
'339. They are a little weak, though.

Try shifting your interconnects to using differential signaling over
twisted pairs. Line drivers/receivers labeled for RS-422/485 are easy to
find (and typically multiple-sourced) and would probably help a lot.
Common mode interference (a transient that affects both signals) will,
for the most part, disappear.

They aren't cheap and suck a lot of power. A '339 is less than $.10. An RS422
drivers and receivers go closer to a buck. '339s come in four-packs. Use two
(half). ;-) I agree, though. RS-422 is the way to go if the budget allows.

I was going to ask if you were channeling Joerg when I realized he would
have opted for a couple of one penny discretes and some fractional-cent
passives. ;-)

--
Rich Webb Norfolk, VA

 

[krw@att.bizzzzzzzzzzzz]

On Tue, 14 Jun 2011 11:53:53 -0400, Rich Webb <bbew.ar@mapson.nozirev.ten>
wrote:

On Tue, 14 Jun 2011 09:10:34 -0500, "krw@att.bizzzzzzzzzzzz"
krw@att.bizzzzzzzzzzzz> wrote:

On Tue, 14 Jun 2011 08:04:20 -0400, Rich Webb <bbew.ar@mapson.nozirev.ten
wrote:

On Mon, 13 Jun 2011 18:38:39 -0700 (PDT), eromlignod
eromlignod@aol.com> wrote:

I meant the circuit output before the long lines. The comparator is
generating the square wave. When the circuit is operated alone, it
produces a beautiful, crisp square wave. When I add the other
circuits I get crosstalk between signals and noise from the coils.

Are you using the 339s as your line drivers? If so, not a good plan.
They're not really designed to drive long lines in a noisy environment.

Shouldn't be *too* bad if you drive a current loop and detect with another
'339. They are a little weak, though.

Try shifting your interconnects to using differential signaling over
twisted pairs. Line drivers/receivers labeled for RS-422/485 are easy to
find (and typically multiple-sourced) and would probably help a lot.
Common mode interference (a transient that affects both signals) will,
for the most part, disappear.

They aren't cheap and suck a lot of power. A '339 is less than $.10. An RS422
drivers and receivers go closer to a buck. '339s come in four-packs. Use two
(half). ;-) I agree, though. RS-422 is the way to go if the budget allows.

I was going to ask if you were channeling Joerg when I realized he would
have opted for a couple of one penny discretes and some fractional-cent
passives. ;-)

I'm learning. ;-)

 

[whit3rd]

On Monday, June 13, 2011 6:15:19 PM UTC-7, eromlignod wrote:

The output is from a 339 comparator with a 3K pull-up resistor.

To drive a line with that comparator, your best bet is to hit
110 ohms output impedance, with a 220 ohm pullup to +5 and
a 330 ohm pulldown to GND. That makes the drive a good
match to CAT-5 wiring (or most twisted pair wiring) while
maintaining TTL signal levels. To receive the signal,
an old MC1489A is suitable, or any logic gate with hysteresis.
Best, though, is a differential receiver with hysteresis and
good common-mode rejection. Like RS-422 serial receivers.

Twisted pair wiring and shielded-pair wiring and differential drive
all reduce pickup from nearby machinery.

 

[Rich Webb]

On Thu, 16 Jun 2011 06:45:45 -0700 (PDT), eromlignod
<eromlignod@aol.com> wrote:

What do you guys think about fiber optic lines?

Probably overkill. If you're adding new drivers/receivers, go with a
422-style and twisted-pair setup first. *If* that doesn't work, consider
fiber but be aware that it comes with its own issues WRT
connectorization, termination, bend radius, etc.

--
Rich Webb Norfolk, VA