Signal Ground VS Chassis Ground - Should they be coupled?

[innkeeper]

I'm assembling a sensitive op amp circuit that will be comparing both balanced and unbalanced inputs for low level measurements.
I am wondering if there should be any coupling between Earth/Chassis ground and the signal ground(PSU ground)?

I am concerned with any kind of external noise influence.

This project will be placed inside a metal case. Also inside this metal case will be an R-core transformer and and EIC connector with its ground attached to the chassis for safety though i am not opposed to moving the PSU external if noise becomes a factor.

My question is should there be any coupling between chassis ground and the power ground of the op amp circuit.

i am of two minds on this..i like that the metal chassis forms a Faraday cage. however, i am concerned that the chassis itself could also introduce noise if the coupling is not handled the right way.

Eventually i can experiment with this once i go from a design to actual implementation, but, id like to get some input on what the traps are here and best practices.

i  know that for certain sensitive measurements the psu is totally replaced by batteries and the entire thing is enclosed in a metal box. I'm not sure if i am exactly at that point but, pretty close.  I am taking care to make the power supply extra clean for example.

I look forward to everyone thoughts on the topic.

[jmelson]

Wow, you've opened Pandora's box!  There are so many things to know before answering.  But, in the most general terms, you don't want the electronics to be floating with respect to the
chassis/shield.  So, the op amp power ground should be tied to the chassis at ONE POINT only.  Instrumentation amps need to have the signals within the common mode range of the
op-amps, so the sensor that is the source of the signal needs to be referenced to that ground in some manner.

Jon

[innkeeper]

i am thinking out loud here... if the signal ground is tied to chassis ground, and i connect it to another mains grounded system, that can case a long loop between grounds that can have voltage difference both in noise and mains. this would make for a noise source... or am i thinking completely wrong?  if this is the case, then id think keeping the signal ground isolated from mains/chassis might make sense?

[JustMeHere]

There's three sets of lectures on learning.oreilly.com that cover this.  You can get a week long trial.  The lectures are by Howard Johnson. He's considered the best signal engineer on the planet.

[teecelly2020]

I use a pair of diodes in parallel, but in opposite polarity and a small cap to break low voltage ground loops in audio equipment.

You can learn more about it here:  https://circuitcellar.com/resources/ee-tips/find-and-eliminate-ground-loops/

Good luck!

[bob91343]

Here's the story.  Signal ground is merely a reference point; the important issue is the current path.  In other words, the signal needs a return path independent of any other currents.

Chassis ground should only be used to prevent the chassis from wandering around, and its main purpose is shielding.  No current should flow in the chassis.

If you don't follow these rules you are setting yourself up for noise, hum, oscillation, and worse.

You can connect signal and chassis grounds together as long as you avoid uncontrolled current flows.

I realize this isn't the simple answer you seek but it's not simple stuff.

[mansaxel]

Here's the story.  Signal ground is merely a reference point; the important issue is the current path.  In other words, the signal needs a return path independent of any other currents.

<snip>

I realize this isn't the simple answer you seek but it's not simple stuff.

(The few things I know in this field are all about analog audio systems, so the following is to be read with that in mind. Other fields of EE might need different treatment. Caveat emptor. )

It is indeed complicated.  Neil Muncy wrote a paper in Journal of the Audio Engineering Society, June 1995, "Noise Susceptibility in Analog
and Digital Signal Processing Systems”, which is a very good description of the problem and pointers towards remedies.  Tony Waldron, Technical Director at Cadac, wrote a series of articles on "Practical interference-free audio systems" that start with how rebar, piping, conduit and electrical grounds are interconnected in theatres and TV studios, and go all the way up to PCB design.

The meat of it all is already mentioned, "one interconnection only" -- per device, that is. The screen of an incoming signal connection shall be well terminated to chassis at the entry point, and only the signal wires shall continue to the PCB. Signal ground on the PCB shall be interconnected with DC PSU 0V and at one place connected to chassis, and power ground.  That is the basic rule, and it can be bent, but at best knowingly so.  Between boxes, no screen shall be interrupted, unless you're building an AM antenna. 

There IS a place for galvanic, i.e. transformer or optical isolation, but it is rarely needed as much as people believe.

[T3sl4co1l]

Yes.  Sometimes.  Maybe.  Depends.

(Congratulations, you've won my laziest post of the day?.. )

Suffice it to say, it's complicated.  You cannot look at one small part of a system in isolation, you must consider the whole system to determine which is best.

Tim

[innkeeper]

@JustMeHere - is oriley safari worth the 399 a year?

@teecelly2020 - that is a very interesting approach to connecting chassis ground. The more i think about it the more i like the concept, though, not sure how it will behave int he wild.  What i might do is take my pcb and add some pads to connect components from the chassis ground points to the signal gorund. this way i can try this and other ideas. or not use it.

@bob91343 - sounds like the way to control current flow in the chassis would be to only allow 1 connection point to the chassis?

@mansaxel - in my case, i will have a device under test, with a single ended signal going into both the device under test and my device and then the single ended output of the device under test also going into  my device. then yet another output from my device to other equipment for measurement purposes, all sharing common grounds.  I could break the two grounds to the DUT by potentially making the input to my device differential and detecting the difference from the incoming ground and the signal line and therefor not grounding the returning line to the device however the measurement equipment wold still share the same ground as the input to the device under test. sorry if thatch a bit convoluted.

[bob91343]

That is correct, only one point on chassis should connect.  Otherwise there will be currents flowing in the metal with resultant voltage drops which can interfere in the signal path.

If you have a PCB with a signal ground, it should be one point only.  And the input signal from the jack, mounted on chassis, should be insulated and its ground carried to the PCB signal ground (NOT touching chassis!).  Never have two signal grounds!  Never rely on the chassis to have zero impedance!

A while back a friend had a signal mixer to use in an audio system.  It had some hum and he sent it to the factory for some standard hum-removing modification which didn't do the job.  I got the unit and discovered the cause of the hum.  If you guessed chassis current, you get the prize.  I had to redo the whole circuit philosophy to fix it.  Later he came over and showed me how he used it.  I looked at him and said, no hum?  He did a double take and said you are right, no hum!  How did you do that?

[EEEnthusiast]

The circuit ground need not be connected to the chassis ground. For e.g. in case of USB powered devices, the USB cable has both a shield and a GND wire. The best practice is to have the GND (black wire) connected to the circuit ground. The shield wire (metal braid) gets connected to the chassis. It is also a good practice to add a small capacitor (few 100pF) from the circuit GND to the Shield for the high frequency grounding. This helps to minimize EMI.

As a general rule, the chassis should be at the earth potential to avoid any safety hazards. The circuit ground can be at any potential w.r.t. chassis.

[mbless]

If you have a PCB with a signal ground, it should be one point only.  And the input signal from the jack, mounted on chassis, should be insulated and its ground carried to the PCB signal ground (NOT touching chassis!).

Can you clarify the "insulated" part of your statement? From what I've read elsewhere, I'm surmising that a shielded cable has the shield connected to the chassis (typically through the connector), and the signal and signal ground go directly to the PCB. Or are you talking about shielding the signal in the interior of the chassis from the connector to the PCB?

[T3sl4co1l]

A lot of nonsense happens when designers don't know about RF versus galvanic or audio grounding.  Most of the solutions seen for one type, are horrid when considered by the other, and vice versa.

Single point star grounding, case in point.  What an electromagnetic disaster.  There is, however, an RF version which is general, covering both cases; the trouble is, few audio designers know it.

The system is the biggest problem in audio.  You can't fix a badly wired system even with the best possible amplifier.  Like I said before, you can't analyze a component in a vacuum, it's the system that matters!

In RF, you don't much care about ground loop currents, because they are rejected out-of-band, and isolation is easy (some coupling capacitors or a little isolation transformer).  Differential signaling may also be used.  You can make up for some sins this way.

In audio, the fault arises because ground currents are allowed to flow in signal conductors.  Namely, signal cables are one pair and that's it: signal and "ground".  Except it's not really "ground", or it shouldn't be: it's signal reference, negative.  It's forced to serve both purposes, and the system is fragile as a consequence.

Oscilloscope probes suffer the same fate: they are both wideband, unbalanced, and sensitive to common mode (= ground loop) interference.

Commercial signaling systems avoid this by separating the signal reference and ground, and often using differential.  Examples:
Shielded XLR, a direct analog example, differential
USB, a digital example, differential with poor CM range making it highly vulnerable, if not for its shield
ADSL, a hybrid example, differential, symbols are transmitted as phased AC with up to 32768 states

ADSL must be isolated?  And Ethernet is transformer isolated (it uses differential signaling and relatively few states, e.g. PAM-3 for 100BASE-T), with the bonus that their signals are high frequency, so avoid low frequency noise problems doubly so.

XLR is sometimes transformer isolated, but this should only be needed in the absolute worst case, where ground loop voltage is so high that the receiver's CM range is exceeded (probably around 15V in most cases?).

The trick for "pro-sumer" audio is to use RCA jacks differentially.  The shield must be RF grounded with capacitors.  The shield can additionally be semi-grounded to the circuit or chassis through a resistance, but this should be nonlinear for protection only -- i.e., a large resistor to bleed down static, in parallel with a TVS to shunt fault currents from cross-wired circuits.  The differential voltage is then sensed with a diffamp, and you don't have to worry about ground loop currents on board at all.  Mind to include an RF filter there, to clean up remaining trash, especially if you're using slower or older or bipolar circuits that love to rectify RF.

Such a method won't solve CM currents already on the cables, or between other equipment in the system, of course.

Tim

[bob91343]

To mbless, yes the shield of the input cable and the low side of the input jack should be insulated from chassis.  They need to go to the signal return in the amplifier.

It's good to ground to chassis at some point.  But only one point.

Some systems use a shield ground and a signal ground at the input connector.  Whatever it takes to eliminate the so very common ground loops.

It's easy to get sloppy and just ground stuff.  Sometimes it comes out okay but there are too many times when one gets oscillation, power supply hum, and other undesirable effects.

A shield should shield from fields.  Don't ask it to carry current unless you are very careful how you do it.

Trace the current path of the signal and make sure it only goes where you want it to.

[mansaxel]

A lot of nonsense happens when designers don't know about RF versus galvanic or audio grounding.  Most of the solutions seen for one type, are horrid when considered by the other, and vice versa.

What was discovered, there in the middle 90s, was that one had to learn both; mostly because digital logic at then-modern CPU speeds acts more like RF than DC... And excepting luddites, most every audio has device a computer inside.

As you write, star grounds are effective in the audio passband, but not further up. Their impedance at higher frequencies is simply too large.  The conclusion by Muncy and Waldron among others was that one had to build a broadband grounding scheme.  They ended up with a few things:

• As noted above, mixing interference with signal defiles the signal
• The practice of interrupting screens that sometimes, at random, gives useful results in audio is an efficient antenna design paradigm at HF.
• The job of a screen is to siphon interference current off of the signal path.
• Therefore, the screen needs to be as complete as possible, in order to be able to carry interference currents.
• The screen needs to be not only continous in a DC ohms sense, but also for RF, which means that things like barrier strips were very problematic, since they can't be screened and present a high impedance bump at RF.
• Since we can't mix signal and screen, balanced audio is a requirement for any connection that will travel any length, especially between powered devices.

These conclusions led to a few recommendations and products:

• A device design model that takes only signal positive and negative onto the PCB, leaving screen and 0V and phantom power ground etc to travel together with interference through the chassis instead, meeting DC 0V in the PSU. Not in the preamplifier where there's a lot of eager, non-discriminating, gain.
• New RF-shielded XLR connectors and XLR chassis connectors that short Pin 1 to ground as broad-band-ly as possible
• A grounding paradigm that did away with all those crazy ideas of putting racks on wood plinths to isolate them, instead opting for as much grounding as possible in the audio system design, cross-bonding with much everything, frequently.

This works. But, as also has been said upthread, the system needs to be analyzed as a whole.

[T3sl4co1l]

Regarding single-end shields: it's very old advice indeed, old enough that I wonder if it's nothing but a holdover from the tube days.

Namely, in vacuum tube circuits, circuit impedances are quite high (kohms in RF circuits, 100s kohms in audio), and being higher than Zo/2 (Zo = impedance of free space), magnetic fields don't matter very much.  Or to put it another way, signal levels are high enough that magnetic field induction is almost negligible.  Electric induction (i.e. capacitive coupling) instead is the major offender.  So just putting a screen around everything sensitive, and grounding one end to sink the electric charge to ground, is adequate shielding.

Circuits were also low bandwidth and gain, unlikely to oscillate from random transmission line stubs.  Except when they weren't (peppy RF tubes are more than happy to!), but grid stoppers were standard practice as well.

Tim

[David Hess]

i am thinking out loud here... if the signal ground is tied to chassis ground, and i connect it to another mains grounded system, that can case a long loop between grounds that can have voltage difference both in noise and mains. this would make for a noise source... or am i thinking completely wrong?  if this is the case, then id think keeping the signal ground isolated from mains/chassis might make sense?

That is a real problem with audio as discussed above when RCA connections are used but there are ways to fix it.  The usual solutions are to either use differential signaling which rejects the difference in the grounds/commons, or to sense the remote ground/common potential with a difference amplifier so it can be subtracted.

The above also applies to differences between grounds/commons within a chassis as well.  With precision circuits, it may apply on a single printed circuit board or ground plane like when you have 16+ bit converters.

[EEEnthusiast]

i am thinking out loud here... if the signal ground is tied to chassis ground, and i connect it to another mains grounded system, that can case a long loop between grounds that can have voltage difference both in noise and mains. this would make for a noise source... or am i thinking completely wrong?  if this is the case, then id think keeping the signal ground isolated from mains/chassis might make sense?

That is a real problem with audio as discussed above when RCA connections are used but there are ways to fix it.  The usual solutions are to either use differential signaling which rejects the difference in the grounds/commons, or to sense the remote ground/common potential with a difference amplifier so it can be subtracted.

That the above also applies to differences between grounds/commons within a chassis as well.  With precision circuits, it may apply on a single printed circuit board or ground plane like when you have 16+ bit converters.

Yes, it is always a good idea to keep the shield and signal return (circuit ground) as separate points. I think the circuit reference should never be called as Ground as it creates confusion. It all started off with the main wiring connection where one of the conductors was physically connected to the earth (ground) for safety reasons. For battery powered devices, there is no connection to the physical ground but there is always a shield to keep away EMI. For systems where there is some connection to the earth, for e.g. PC, Audio Amps, or most of the line powered devices, it is always recommended to keep the circuit reference separate from the shield (Ground). The shield can carry noise current which if connected to the circuit reference, may show up as SNR degradation. By keeping them separate, you split the noise paths and the signal return paths.

Take another practical example of a system which is enclosed in a metal case. The power supply to the system is completely floating (isolated using transformer). The circuit reference is not connected to the metal case. However there is a shield on the PCB which is connected to the circuit reference. And the mains Ground (also called earth in many countries, and usually a Green Wire in the mains supply) is connected to the metal case (chassis). Should the shield on the PCB be connected to the metal case of the system to make a single point ground? The simple answer is NO. The internal shield on the PCB is used to keep the EMI away by containing the fields within the shield. These carry mostly high frequency currents that are in low amplitude. The metal case offers protection for the user and in case of any breakdown, the metal case is held at the earth potential by the earth (ground) connection. By shorting the metal chassis to the RF shields on the PCB, we could induce some currents on the chassis which was never meant to flow to the earth. These RF currents could cause radiation as the earth wires are long and are very good radiators.

Keep the circuit reference away from the shield (ground) whenever possible. In some cases where you use a connector like BNC or SMA where the Ground is the same as the shield, you do not have a choice but the short them. But these are mostly RF equipment where the circuit reference and the ground are coupled due to capacitive coupling. Solving the ground issues in RF designs is tricky as this is the area of black magic design, where the observations in the lab are seldom written on paper.. They remain in the minds of the engineers...

[EEEnthusiast]

I would recommend the new book from Henry Ott (Electromagnetic compatibility engineering)
It is a bible for EMI engineers...
https://cds.cern.ch/record/1188673/files/9780470189306_TOC.pdf

[David Hess]

Analog Devices has several old application notes covering this subject:

https://www.analog.com/media/en/technical-documentation/application-notes/AN-202.pdf
https://www.analog.com/media/en/technical-documentation/application-notes/6001142869552014948960492698455131755584673020828AN_345.pdf
https://www.analog.com/media/en/technical-documentation/application-notes/41727248AN_347.pdf

[bob91343]

One approach I use is to draw on the diagram the complete loop of signal current.  For example, from collector to coupling capacitor to output jack to load, through load, back to low side output jack, back to emitter, through transistor and finally back where we started, collector.

Any 'ground' should not be in the circuit, although it's okay to connect some point to chassis (say, output low side) for shielding purposes but not to pass any signal.   And once you pick a ground point, use it for all grounds.

That won't make it perfect due to cross currents in the actual ground connection but it will make it possible to avoid major loops.  Sometimes when connecting several wires to a chassis ground point it will be advantageous to disconnect them and try reconnecting in a different order.  Only a fraction of an Ohm can make a difference.

Don't rely on rivets and screws to stay connected; they loosen and corrode.  Not only do they lose some continuity but they also become diodes and rectify signals with resulting intermodulation.  A large bolt and toothed lockwasher and solder lug is a good thing if tight enough.  A copper strip is good; bolt it to the chassis after soldering all grounds as close to one another as possible.

Shielded wire can be a trap.  Use the shield as a shield and not for signal return, meaning 'ground' it at one end only.  If you need a return wire, get two-conductor shielded wire and use one internal wire for signal return.  Back in the day, we learned a lot when trying to connect phonograph cartridges to amplifiers.  Getting the hum out took a lot of thought.

[T3sl4co1l]

Shielded wire can be a trap.  Use the shield as a shield and not for signal return, meaning 'ground' it at one end only.  If you need a return wire, get two-conductor shielded wire and use one internal wire for signal return.  Back in the day, we learned a lot when trying to connect phonograph cartridges to amplifiers.  Getting the hum out took a lot of thought.

Again, broken shield means introducing 100.0% of interference picked up by said shield, into the signals inside.

If you're doing a differential pair, maybe that's okay.  In which case you don't even need the shield in the first place.  15V CMR is enough to handle a lot of ambient noise sources.  Won't stop clicks and pops from ESD and such.  Maybe those are brief enough they don't matter.  Maybe you prefer the sound of random clicks, because it's like a phonograph...

But you better hope your CMRR is good, including at radio frequencies.  And if it's not, I hope you like the local broadcast stations...

Tim

[mansaxel]

Shielded wire can be a trap.  Use the shield as a shield and not for signal return, meaning 'ground' it at one end only.  If you need a return wire, get two-conductor shielded wire and use one internal wire for signal return.  Back in the day, we learned a lot when trying to connect phonograph cartridges to amplifiers.  Getting the hum out took a lot of thought.

As T3sl4co1l writes, that is not always the good idea you describe it as. I'd limit that method to perhaps inside a device, where you have an external good RF screen and need to go from one place to another by cable. Outside, entering the system design domain, the "lifting" of shield connections was tried, erratically and with very mixed results, by those in the trade, aided by stupid and sometimes dangerous methods and measures offered by habit and box designers, like ground lift switches on devices and ungrounded outlets.. 

Once you understand that the interference is an induced voltage that wants to become current, the solution is obvious. Let the current flow, but in a managed way, in places where it won't harm your signal. 

As a benefit, you'll get less harm to people too:  I once worked at a theatre, where the stage managers position has got an intercom wallbox just beside the iron fire curtain. I happened to stand there and touched the intercom box while leaned against the iron curtain, and was zapped with 117VAC, not much current, but it hurt. I had my suspicions, as to why, and after the show I opened the wall outlet where the mixing desk power supply was connected and replaced the ground wire. The Y2 capacitors in the PSU will act as voltage dividers and put 1/2 * mains voltage on the ground pin if it's left floating. The entire audio system shield / 0V / "ground" (including human-touching parts like microphone pop screens etc; the mixer provides a ground connection that is continous all the way out to the vocalists mouth.) had 117VAC 50Hz on it, because the mixing desk of course was connected to the chassis of its PSU. With a suitably beefy ground connection, the problem disappears, and the Y capacitors can work again.

[T3sl4co1l]

The entire audio system shield / 0V / "ground" (including human-touching parts like microphone pop screens etc; the mixer provides a ground connection that is continous all the way out to the vocalists mouth.) had 117VAC 50Hz on it, because the mixing desk of course was connected to the chassis of its PSU. With a suitably beefy ground connection, the problem disappears, and the Y capacitors can work again.

Yikes!  Did you notice that everything felt weird when you lightly run your finger over it?  It's not just weird--there's an electroresponsive reaction (I forget if it's the sense of touch, or the skin itself somehow moving?) to AC electric fields!

Tim

[mansaxel]

Yikes!  Did you notice that everything felt weird when you lightly run your finger over it?  It's not just weird--there's an electroresponsive reaction (I forget if it's the sense of touch, or the skin itself somehow moving?) to AC electric fields!

Oh, yes. My laptop has exactly the same problem. Portable macs, with their metal shell and wall warts that can take 2-prong plugs or 3-prong cables (at least in Europe) will need to be fitted with a 3-prong cable/plug to be safe and tolerable.  If not, they will present a very special tingling sensation when you type on them. Your palms typically rest on the metal case and will conduct if you're better grounded than the floating computer. An ESD carpet is low resistance enough to set this off, and leaning towards a metal cabinet will literally  

I did some testing with this setup:



The laptop is plugged into the mains with a 2-prong plug, and measuring from the ground lug in the outlet to the shell connection of the headphone jack (which is the same as the case of the computer) yields this nice measurement: 



Altering the setup by fitting a grounded mains cable, but otherwise similar, produces a different result, roughly 3 orders of magnitude lower: