Codidact Meta
Codidact Meta
The Great Outdoors
The Great Outdoors
Photography & Video
Photography & Video
Scientific Speculation
Scientific Speculation
Electrical Engineering
Electrical Engineering
Languages & Linguistics
Languages & Linguistics
Software Development
Software Development
Code Golf
Code Golf

Mark all as read

Should I connect signal reference and chassis ground in a battery powered application?


Hello. I have a grounding problem on my acquisition system. I have read many articles, app notes and whatever but still could not find a convincing solution for the topic.

The acquisition system is made of an anodized aluminium box with a PCB inside and two external load cells. More details:

  • each load cell:
    • is made up with strain gauges in a Wheatstone bridge arrangement
    • is connected to the board through a 2.5 meters cable with a metallic cable gland
    • has exposed steel parts, which are directly connected to the cable shield, with the other end of the shield connected to PCB screws
  • the acquisition system:
    • is battery powered (NiMH)
    • has a USB input for battery charging
    • has a WiFi connection with an external antenna
    • is IP65
    • passed RED certification
  • the box is anodized (insulated), but there are masked sections for external connections (USB connector, power button, mounting screws, cable glands, panel antenna connector) and for the PCB screws inside
  • the WiFi antenna connector carries the signal reference on its cold pole
  • the antenna cold pole is mounted on the box in a masked section, thus creating a direct connection between signal reference and metal box (chassis ground); the connection is only made in this point
  • the acquisition system is battery powered and hence it has no earth ground reference; but it is used in a railway application, and when used the exposed steel parts of the load cells are in direct contact with rails (which are grounded)

After the RED certification phase we shipped a few systems to a lab in order to certificate system measurements, but two systems came back to our lab with some damaged ICs. I suspect an ESD event, as the ADC and LDO converter connected to the load cells were damaged and are not protected against ESD.

Now I am redesigning the system in order to solve this issue. I will definitely add ESD protections and connect the shield of the load cell cables to the cable gland (so as to have a 360 degree coverage). But my main issue is: is it ok to connect signal reference and chassis ground? Should I leave that connection, or it would be better to unmask the panel antenna connection thus separating the two grounds?

As I said, I found many articles on the topic, but none of them mentioned the battery powered case. So I am still a little confused about the possible consequences of both choices.

Thanks in advance for your help.

Edit: The following diagram shows my acquisition system (some external devices such as power button are not shown). I tried to make it easier to read by using different colors for different signals:

  • black for system reference (PCB ground)
  • light gray for chassis ground
  • brown for earth ground (on the rail)
  • red for positive power lines
  • other colors for signals

The panel antenna connector is intrinsecally connected to PCB ground and it is also connected to the chassis ground; at the moment this is the only connection between the two. The traces on the PCB carrying the cable shields to the screws (shown in the diagram) are not connected to PCB ground on the PCB itself.

@Olin Lathrop: from your answer I think I should leave the "mask" on the aluminium box in the panel antenna connector section, so as to use it as the single point connection between system ground and chassis ground. But this leads me to a new implication: what happens when the two load cells are placed on the rail (see diagram) and the chassis ground is bound to be at earth potential? Could it cause some voltage shift I should be aware of?

Edit 2: @Andy aka: the guys at the measurement lab were not able to tell exactly what happened. What I know is they put all systems on a metal table: this creates a second connection between metallic parts (chassis ground) but I think there would be no issue, as this does not affect PCB ground. Even small noise currents circulating on the chassis ground (ground loop created by the two connections) should not be an issue as there is still only one connection point between PCB ground and chassis ground. Moreover, 4 system were used and only two of them got damaged. The load cells are made up with strain gauges and a flexible PCB, so the only insulation between the signals on the Wheatstone bridge and the metal parts on the rail are the strain gauge carrier (a 40 um thick insulating material) and the flexible PCB material (50 um thick). There was no lightning, and an ESD event is my opinion based upon the fact that the LDO and ADCs (not protected against ESD) got destroyed. My first goal is understanding whether or not I should keep PCB ground and chassis ground connected in the point they are now, with a technical explanation about it (and I think Olin's is very clear). The second goal is to understand what can I do to make my system more robust against external events (such as ESD), as the issue happened with PCB and chassis grounds connected.

@Olin Lathrop: "rail" is a single railroad track, sorry for not clarifying this point. The system is used for weighing trains. The sensors have a 2.5 meters cable connecting them to the box. Isolation is one of my options for redesign, but if possibile I would avoid. I already have a microcontroller, and I used isolation in other designs, so this is not a problem, but it would be a major circuit modification, requiring time and efforts. Still, of course, I will go for it if it is my best option.

Thank you all


Edit 3: @Olin Lathrop, I owe you some clarifications (and maybe a few more questions):

  • the system cannot be used while the USB charger is connected, hence the "battery powered isolation" is always present
  • the system is portable and therefore the sensors are positioned on the track only for the time necessary to take the measurements, then they are taken away
  • the data wires from the sensors are treated as differential, and are filtered as soon as they enter the circuit, just before the ADC inputs
  • the sensors' 0V lines are not directly connected to the cable shields but they are connected to the PCB ground
  • the sensors' cable shields are attached to the sensors' metal parts, and the metal parts are connected to the rail when in use; there could not be a direct connection between the cable shields and the rail as the sensors have to be taken away when not in use
  • I cannot change the sensors' chassis; I could change the way cable shields are connected to them, but from waht I understand it does not seem to be a good choice
  • about ground loops: I took many voltage measurements between different chassis ground points and found no significant voltage (a few uV at most, maybe due to multimeter uncertainty).

Again on 0V lines: they are not directly connected to the cable shields, but the sensors' 0V lines are connected to PCB ground and cable shields are connected to chassis ground. Being PCB and chassis grounds connected in one point, 0V lines get connected to cable shields. Is it ok or should I change something?

And: the "single point connection" between PCB ground and chassis ground is on the panel antenna connector, as the antenna cable - connecting the inside connector to the external antenna - carries the PCB ground and is attached to the box. So basically I have two possible choices for the one point connection between PCB and chassis grounds:

  1. keep the single point connection on the antenna; for me this is the quickest and easiest way
  2. do not connect the panel antenna connector to the box and make the single point connection on a PCB screw; this way I see a possible fault in the fact that PCB ground will be accessible both from the antenna connector (isolated from chassis ground) and from chassis ground (connected to PCB ground through the screw), thus giving the way to possible ground loops.

What do you think about that?

Again, thanks for your help.

Edit 4: @Olin Lathrop: From your last sentence I assume that choice 1 above is correct and the load cells' grounding is ok as well. Thanks a lot

Why does this post require moderator attention?
You might want to add some details to your flag.
Why should this post be closed?


Show a diagram. I got lost in some of the hand waving. Olin Lathrop‭ 6 months ago

What testing was conducted that destroyed the failed devices? Are you looking for reasons behind these failures? Can you be specific about what advice you are seeking? What insulation have the load cell bridges got from the grounded rail? What specific tests were conducted? Was it ESD or indirect lightning as well? Andy aka‭ 6 months ago

Can you address my questions above as a comment because I don't want to have to trawl through the whole question trying to work out things. What I can say is this: without knowing what specific testing destroyed the devices is not going to help..... Andy aka‭ 6 months ago

..... I spent a whole year on one job trying to find an answer to this specific type of problem where a device had allegedly passed an impulse test but there were no records kept. In the end I concluded that someone had added TVS diodes to a design after testing had been done, didn't do their homework and got it wrong. The moral being this: find out what caused the destruction of the device specifically or it's a fool's errand. Andy aka‭ 6 months ago

@Andy aka: Apparently the failed and good devices were submitted to the very same tests. The guy at the (external) lab could not be more specific, and I will not have any other reply on that side. I agree with you, with no clear idea on the specific tests that damaged my devices I cannot be sure I can address the real issue. That's why I asked for "generic" help about grounding. I will now redesign the board according to what I learnt. And, by the way, I think we'll search for another lab... stinf‭ 6 months ago

1 answer


It's not totally clear from just the verbal description, but it seems the real question is when/if signal and chassis ground should be connected.

If the circuit is not intended to be truly isolated, then signal and chassis ground need to be connected at some point. The trick is to think about all the ground currents, and make sure that things like power return currents or ground offset currents don't flow along the signal ground wire.

Generally, this means the signal and chassis grounds should be connected in exactly one place. That lets the signal follow whatever common mode voltage is on the chassis ground, but does not allow chassis ground currents to flow thru the signal ground wire.

The best place for this single connection is usually right where the signal ground wire enters the ground-connected box. Any other ground connections between the circuit and the chassis must then be avoided.

Sometimes it is more convenient to let the signal wires feed thru a small hole in the chassis to connection points on the PCB. In that case, connect the PCB ground to the chassis ground close to where the signal ground is connected to the PCB. I sometimes do this with a mounting screw hole that is also a PCB pad, with the holes for all the other mounting screws insulated.

Think of the chassis as shunting unwanted ground currents around the circuit. You have to assume there are currents flowing between any two points of the chassis. By connecting the circuit to the chassis in exactly one spot, these chassis currents can't flow thru the circuit. If you connect the circuit to the chassis at two points, then the currents that chassis was meant to shunt might flow thru your circuit, causing ground offsets.

what happens when the two load cells are placed on the rail (see diagram) and the chassis ground is bound to be at earth potential?

That depends on what this undefined "rail" thing is, and then possibly whether the sensors are electrically connected to it.

If the sensors are not close to the box, then you should be using differential signals.

If they are really far, then you should consider isolating the immediate analog input circuitry, with data shipped to the rest of the circuit digitally over opto-couplers or something. That would require a small isolated power supply, enough to power the analog circuitry and microcontroller to manage the conversions and ship out the digital values. I've done that a few times using a small POE transformer in flyback mode.

"rail" is a single railroad track

I re-read your whole question, and things make more sense now. You did mention railroad track before, but I had forgotten that by the time the diagram was added.

In general, you have too many ground connections, and therefore need to consider ground loops carefully.

One advantage you have is that your circuit is isolated due to being battery powered. However, that no longer applies when you connect the USB charging input to something that is ground-referenced. If normal operation is always without a USB connection, then you get to consider this isolation as part of the design.

I would tie all the "external" grounds together everywhere, but make sure it never carries any signal. These external grounds would be the rail, the two cable shields, and the box your circuit is in. This becomes a shield, connected to earth by the rail. This should be connected to your circuit ground in exactly one place. I'd probably make one of the mounting screws carry the ground connection.

The two data wires from each sensor carry a single differential signal, and should be received that way by your analog front end. Each would have a little filtering immediately where they connect to the PCB, then go into an instrumentation amp. The point of the filtering is to squash frequencies, like radio pickup, that the diff amp can't handle.

The tricky question is what to do with the two 0 V lines coming from the sensors. This depends on whether the sensors are electrically connected to the rail.

Ideally the sensor is not electrically connected to the rail, in which case you don't really have a "too many grounds" problem. If the senors are connected to their external mounting plates, maybe you can put an insulating pad between the mounting plates and the rail. Or, get a different model of sensor that is insulated. Note that the cable shields should still be connected to the rail.

If the 0 V lines and the cable shields are unavoidably connected at the sensors, then things get more tricky. You need to tie the 0 V lines to your analog circuit ground because that carries the return current of the 5 V lines. In this case, I would not add a deliberate connection between your circuit ground and the chassis. You basically already have two such connections, one at each sensor.


From your latest description, it sounds like you have done all the reasonable things. I don't know what else I can say from here.

Why does this post require moderator attention?
You might want to add some details to your flag.


Sign up to answer this question »