Is it considered a good practice to use the chassis as a return path to a DC Power Supply?

I have been given a rack mounted device that switches 480 volt power on/off remotely. The remote feature is handle via a Raspberry Pi which is powered via 120v to 5v power supply (http://www.mini-box.com/OpenUPS).

I was surprised to find that the return pin on the 5v power supply is tied directly to the chassis and the GND on the Pi is also tied to ground and they are using the chassis as the conductor for the return path. There is only a positive 5v wire going between the power supply and the Pi. Is this considered good practice?


I would think not.

For one, if either device was removed from the chassis (by the unwary) for fault diagnosis then the operation would fail and damage may result.


simulate this circuit – Schematic created using CircuitLab

Figure 1. (a) What you've got. (b) Standard wiring practice with ground at one point. (c) Ground at two points and 0 V line connection.

Grounding at both devices as in (c) risks ground loops which are problematic in low-voltage signal circuits. In your case with both devices on the same chassis it might not be a problem.

| improve this answer | |
  • 1
    \$\begingroup\$ Grounding at two points as shown does NOT create a ground loop. To create a ground loop you would need a Gnd wire between the two units as well. It is however never good policy to used a frame or sheet metal cover to carry current. It typically has bolted connections which need to be tightened and the metal is a poor conductor compared to wire (which is why you get loops when you have multiple connection point). It is good design to carry a ground wire back to a common point on the frame/chassis, typically near the AC mains entry point. \$\endgroup\$ – Jack Creasey Jul 25 '18 at 3:03
  • \$\begingroup\$ Thanks, @Jack. I had made that point in some of comments on other answers and then muddied my answer with the addition of Figure 1a. Un-muddied with addition of 1c. \$\endgroup\$ – Transistor Jul 25 '18 at 6:10

It's generally not a good idea to rely on the chassis as the return conductor.

Magnetically coupled interference happens because a stray magnetic field is coupled into a magnetic loop in your circuit. The smaller the cross-sectional area of your magnetic loops, the smaller the interference.

Consider twisted pair ethernet. This has little susceptibility to interference because the twists mean the area between the conductors is small.

If you use the chassis as your return path, your currents are travelling around a loop that is probably several centimetres across, and so gives a wide area to receive interference. With a conventional two-conductor power cable, the loop is only the distance between the conductors - less than a millimetre.

Additionally, you may not have a good low resistance connection depending on factors like corrosion.

What about cars? Cars use the chassis as ground. But they're full of high current components (lights, pumps, etc) that aren't susceptible to such interference - when connecting to a device that is sensitive, like a radio or an ECU, there will be a ground wire on the connector.

| improve this answer | |
  • \$\begingroup\$ Where would the loop form if there was no 0 V connection between the devices and only a chassis connection as explained in the question? \$\endgroup\$ – Transistor Jul 24 '18 at 21:59
  • \$\begingroup\$ The loop is: 5V of PSU -> cable -> Pi -> chassis connection -> chassis -> chassis connection -> GND of PSU. That's why it's large (and potentially high resistance if one of those connections isn't good). \$\endgroup\$ – user1908704 Jul 24 '18 at 22:01
  • \$\begingroup\$ This is the first time I hear about "Magnetically coupled interference". Even in your description any hypothetical current induced in 1 loop of power circuit will go into power lines where it will have no effect whatsoever. Unless you are talking about NEMP, in which case it does not matter anymore. \$\endgroup\$ – Maple Jul 24 '18 at 22:28
  • \$\begingroup\$ The power supply is DC. Coupled interference is at kHz (from SMPSUs) or MHz (radio transmissions/leakage). Even if it's coupled into the power input, the voltage regulator only has so much capability to block RF. However it's more of a problem when interference is coupled into data lines rather than power lines - for example other devices connected to the Pi. It depends whether the ground reference for those is the Pi or the PSU end - if they're referenced from the PSU end then it'll be superimposed on the Pi data signals. \$\endgroup\$ – user1908704 Jul 24 '18 at 22:37

Nope. It was common practice in old tube amplifiers but it is not a good technique by today's standards. And such a device would be rejected by any cert required (UL or what have you).

There should be one point in the unit (usually a PCB but it could be a barrier strip or similar) where all grounds converge. One of those grounds should be from the Pi (presumably neatly with it's 5V in which would then go on to the supply). Then run a separate thick wire from the convergence point to the chassis with a proper bolt that has a toothed washer. But there should be no return currents in that wire or through the chassis. It is simply an earth safety connection.

| improve this answer | |
  • \$\begingroup\$ In this case it's more likely to be noise related rather than safety related. The Pi will work quite happily with an ungrounded, isolated USB PSU. What UL standards would it be foul of? \$\endgroup\$ – Transistor Jul 25 '18 at 6:13
  • \$\begingroup\$ It is not likely that you can leave the Pi floating. Not unless everything it's connected to does not need a ground connection. If it's a bunch of relays, then yes it will work. But anything connected to it that needs a ground reference will require that the Pi is also connected to ground. \$\endgroup\$ – squarewav Jul 25 '18 at 15:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.