I build and service machines that include unfortunate noise sources (arcing relays and highly inductive motors) as well as a variety of low voltage electronics. Several of my co-workers and I have been mystified as to why what seems like very heavy duty grounding along with careful adherence to star grounding methods does so little to protect our low voltage circuits from noise related problems, like glitches on a character display every time a motor starts.

I think I may finally be starting to understand at least one of the issues in play here, but wanted to stop by for a sanity check.

We thought our grounding should be solid because it seems to go to all the right places, following all the right paths, and with very, very low resistance. But it occurred to me recently that impedance is a function of frequency and that our noise bursts include high frequencies (behind my ability to accurately measure with my beat up oscilloscope, but appears to be at least 1MHz.)

Is it possible that, although our grounding is already severe overkill for DC returns and for safety concerns regarding 60Hz AC, it presents a much higher impedance to this high frequency noise?

If so, where do I start in trying to improve things? We've got #8 copper wire connecting the frame ground to the building ground, #12 copper wire bonding the only two frame components that could otherwise become electrically isolated from one another, and less than 8" of #16 copper connecting the DC common for our low voltage board to the star grounding point. The frame itself is maybe 3' x 2' x 1' and fabricated from stainless steel, most of it nearly 1/8"thick. All of these components show incredibly low resistance with an ohm meter, but I only recently thought of the varying impedance for different noise frequencies.

 

The skin effect can significantly increase the impedance of even a large wire at high frequencies.
One way to reduce this effect is to use large wire braid (typical example shown below) as ground wire, since the many individual small wires reduce the skin effect.
A flat copper strap also has a lower impedance.

 

Is it possible that, although our grounding is already severe overkill for DC returns and for safety concerns regarding 60Hz AC, it presents a much higher impedance to this high frequency noise?

Absolutely. It's not only possible, but inevitable.

If so, where do I start in trying to improve things?

There's a bunch of things you can do, some of which may be applicable and some which may not be:

• Control ground system topology, such as by use of a single-point ground, to prevent the ground return current from one subsystem (especially one that is switching high currents with fast rise/fall times) from flowing through the grounding system of other subsystems. You're already doing this.
• Minimize ground system impedance, such as by using braided/flat ground straps as crutschow suggested.
• Minimize inductive coupling from one subsystem/component to another by minimizing the area enclosed by signal paths and their ground returns, for both inputs and outputs, such as by use of twisted pair to connect relay coils, panel indicators, etc. to their control points on a circuit board, and twisted pair (or even coaxial cable) to connect sensors to their circuit board input points.
• Use decoupling to keep fast current transients out of the ground system as much as possible. Feed DC power to each subsystem through a series RF choke followed by a bulk storage capacitor (size dependent on circumstance) to local subsystem ground; this will force sudden current transients to be drawn from the local bulk storage capacitor rather than from the main DC power supply. (In effect, this performs the same function at the circuit board level that is performed at the component level by putting 100 nF decoupling caps across the power pins of logic ICs.)
• Minimize capacitive coupling of switched high-voltage signals (such as from relay coil drive points) into sensitive circuits through use of electrostatic shielding. Shield all sensitive circuit sections.
• De-sensitize board-level logic inputs, by slowing them down so they operate no faster than functional requirements demand. Also, avoid taking high-impedance logic inputs off-board; shunt them with a low impedance (either a resistor or a resistor and capacitor in series) to make them less sensitive to capacitively-coupled interference.
• Isolate transient-generating subsystems from sensitive, susceptible subsystems by galvanic isolation (optocouplers, etc.).
• Use separate AC mains feeds, each through its own power line filter (a Corcom filter, for example), for the transient-generating subsystems and the sensitive, susceptible subsystems.
• Choose switch-mode power supplies very carefully; some can spray huge amounts of electromagnetic "hash" all over the place, while others are more quiet.

That's all I can think of for now, with only two cups of coffee down; if I think of anything else after I've fully "spun up" I'll post it.

 

some of our emi/rfi problems have had to be fixed with going to fiber instead of twisted pair for serial cables. one set of machines was so bad, that there was 1 micro second 600 amp pulses on everything, including grounds.

 

Wow, thank you so much. Great info, and lots for me to read up on. Most of that list I have at least a vague understanding of (or at least feel like I know how to search for more details,) but I'm not quite following this one:

Also, avoid taking high-impedance logic inputs off-board; shunt them with a low impedance (either a resistor or a resistor and capacitor in series) to make them less sensitive to capacitively-coupled interference.

Any chance you could describe that another way or give me an example? Are you saying that the main board (with the best DC common/ground connections) can have high impedance inputs, but that other boards and subsystems should be designed to have low impedance inputs (presumably because of the limitations to how good the grounding can ever be off the main board?)

 

1 micro second 600 amp pulses on everything, including grounds.

Yikes!!! I haven't tried measuring the current from our noise pickup, but that sounds crazy to me!

However, as in your situation, I'm definitely finding noise between grounding points. In one case I simply put the probes at the ends of a #16 wire that's only 8" long, connected to the star grounding point on one end, figuring there couldn't possibly be noise there... but it was still there, plain as day. I have so much to learn!

 

Are you saying that the main board (with the best DC common/ground connections) can have high impedance inputs, but that other boards and subsystems should be designed to have low impedance inputs (presumably because of the limitations to how good the grounding can ever be off the main board?)

No, I'm saying that avoiding sending high-impedance signals off-board (regardless of which board we're talking about) will minimize the chances for capacitively-coupled interference. Even something as simple as a low-value pullup (or pulldown, whichever is appropriate) resistor on an input can sometimes help.

Any chance you could describe that another way or give me an example?

Sure.

• Buffering a board logic input with a relatively slow common-emitter transistor inverter stage;
• Bringing an input in to a 74HC14 Schmitt trigger inverter, preceded by an RC filter with a time constant of several microseconds;
• For REALLY high noise immunity (provided speed is not important), do board-to-board interface through RS232 transceiver chips (like MAX232);
• Or, where speed is important, use differential signalling such as with RS422 or RS485 interface chips.

In short: minimize the opportunities for stray capacitively-coupled interference.

 

a number 16 wire isnt a good ground for noise. you need more surface area for lower impedance. it wuld work for a safety ground, but would have a high inductance. sometimes paralelling several grounds of differing lengths works better, spoiling resonances in the ground connection.

 

No, I'm saying that avoiding sending high-impedance signals off-board (regardless of which board we're talking about) will minimize the chances for capacitively-coupled interference. Even something as simple as a low-value pullup (or pulldown, whichever is appropriate) resistor on an input can sometimes help.


Sure.

• Buffering a board logic input with a relatively slow common-emitter transistor inverter stage;
• Bringing an input in to a 74HC14 Schmitt trigger inverter, preceded by an RC filter with a time constant of several microseconds;
• For REALLY high noise immunity (provided speed is not important), do board-to-board interface through RS232 transceiver chips (like MAX232);
• Or, where speed is important, use differential signalling such as with RS422 or RS485 interface chips.

In short: minimize the opportunities for stray capacitively-coupled interference.

I think I'm with you now. Thanks!

 

One prime reason for displays acting up, is the absence R/C snubbers across coils and solenoids etc, diode in the case of DC types.
http://www.automation.siemens.com/doconweb/pdf/840C_1101_E/emv_r.pdf?p=1
http://www.ormec.com/Portals/0/file.../ApplicationNotes/Shielding_and_Grounding.pdf
Max.

 

One prime reason for displays acting up, is the absence R/C snubbers across coils and solenoids etc, diode in the case of DC types.
http://www.automation.siemens.com/doconweb/pdf/840C_1101_E/emv_r.pdf?p=1
http://www.ormec.com/Portals/0/file.../ApplicationNotes/Shielding_and_Grounding.pdf
Max.

Yeah, we used to see a LOT more noise, then you guys educated me on snubbers. We added snubbers to all new machines starting several months ago and now we can only find noise related problems in very specific, fairly unlikely circumstances.