11

It seems that some users on another Stack have reported GFCI nuisance trips caused by mains leakage through twisted-pair Ethernet cables connected between computers on different branch circuits, or more specifically, between a computer with a Class I, chassis-mounted, supply conforming to IEC 60950 connected to a grounded receptacle with UL 943 Class A GFCI protection, and a switch that is a Class III appliance with a Class II power supply, connected to a grounded, but unprotected, receptacle on a different branch circuit.

While, conceptually speaking, the idea that there could be a leakage path through the data cable makes some sense, and I have seen Ethernet reference circuits that have termination RC networks from the port-side center-tap terminals in the magnetics to chassis ground as well as a 1nF capacitor between chassis and signal grounds, it seems to me that it would be very poor engineering for this leakage path to allow the mains leakage current to rise to a magnitude exceeding the IEC 60950 standards.

What is the magnitude of this Ethernet-connection-induced leakage current rise, what factors in the design of the equipment involved control this rise, and can someone describe to me the precise leakage loop involved?

ThreePhaseEel
  • 8,858
  • 4
  • 26
  • 41
  • 2
    I think the user on DIY is full of it. if the power supply in the PC is isolated, there shouldn't be any leakage that will trip a GFCI. Maybe he routes his ethernet cables by coiling them around the power cables? – The Photon Apr 03 '19 at 03:00
  • 1
    @ThePhoton -- unfortunately, I'm not in a situation to test it (don't have the network setup or the sensitive leakage clampmeter needed for that) or else I would put this theory to the test! If anyone wishes to experiment with this, though, I'd love to hear about it! – ThreePhaseEel Apr 03 '19 at 03:05
  • I've got a router, computer and laser printer networked together on a GFCI receptacle. Networked back to FIOS box and to another computer both not on this GFCI, and the GFCI hasn't tripped ever (~ 13 years). On the other hand, all the GFCI stuff is also going through a UPS (laser is on the surge-protection-only part) so that may mask any potential problem, though until ~ 6 years ago it wasn't on a UPS. But (as noted in DIY), I've never seen this problem anywhere and I have quite a few customers - I would think I'd come across the problem occasionally if was at all common. – manassehkatz-Moving 2 Codidact Apr 03 '19 at 03:10
  • 4
    One caveat to my above comment: It's entirely possible (in fact, it's certain) there are some shitty power supplies out there being used in PCs. – The Photon Apr 03 '19 at 03:33
  • Related: [Does an ATX power supply have any isolated outputs?](https://electronics.stackexchange.com/q/140854/6334). – The Photon Apr 03 '19 at 03:45

3 Answers3

4

The ethernet connection leakage current should be negligable, with UTP. Every port has a array of transformers for high frequency, the leakage at 50 Hz common mode will be very low. enter image description here

However, if shielded cable is used, S-UTP or CAT7 cables, there will also be made a chassis connection between the two devices.
Then the power supply leakage enters the equation, and those may leak several milliamperes.

The simple act of reversing the plug might remove the nuisance tripping of the GFCI/RCD.

*(image source)

Jeroen3
  • 21,976
  • 36
  • 73
2

So I'm the user on DIY.

I had some original experience at work where we coudn't get the new portable generator to power more than one computer even though the old one did. We eventually bisected it to the GFCI outlet in the new generator.

Later on, I had to track down why my AFCI breaker kept tripping. The electrician I called tested the AFCI breaker by bridging a resistor between power and ground. That tripped it. He said that AFCI breakers work by detecting ground faults. I originally said he was nuts, but it turns out it was true.

I got a copy of a circuit for RJ45 magnetics. The critical point is RXN and TXN are tied together by a pair of identical resistors R6 and R7 and capacitor C15 which is 10nF ties that bridge line to ground. In steady state, C15 would indeed conduct no current; however when sending a packet, the impedance of C15 1 / jωC = 1 / j(2·10⁹)(10·10⁻⁹) = 1/j20. This gives the resulting current flow of I = V/R = 3.3/2/49.9² + 1/20²)¹ᐟ² = .033 amps.

And that's just that one capacitor. I haven't been able to locate the indicator LEDs yet. I've noticed that the connection indicator LED on quite a few computers will light even when the board is unpowered but not when unplugged. Conclusion: that LED is tied between the Ethernet cable on one side and the ground on the other, and that ground is often the neutral wire rather than the house ground (two wire devices ...).

Now the electrician was in fact telling the truth. Old series AFCI breakers would trip at something like .1 amps of ground loop by specification. The gigabit switch I was using at the time was a two wire device (no dedicated ground) so all of that current had to go into the neutral wire. New AFCI breakers have since been fixed to work by other means than ground fault detection and replacing the AFCI breaker was the solution.

GFCI outlets are documented to trip at .004 amps. Guess what happens when you run Ethernet cables between devices on different circuits where one of them doesn't have a ground wire. And I'm pretty sure from the bisection that most of these cheaper power supplies were tying the motherboard ground to the neutral wire not the ground wire despite the ground wire being available.

Joshua
  • 287
  • 1
  • 6
  • So HIGH SLEW RATE Ethernet signals are, lacking OTHER return paths, using the PowerLine as RETURN? – analogsystemsrf Apr 03 '19 at 03:43
  • @analogsystemsrf: Yup! Was expecting it as soon as I learned how carrier sense worked. Come to think of it, that's probably why that line is tied to neutral rather than ground in the PCs. If it were tied to ground it would have a hard time returning to a two-wire device. – Joshua Apr 03 '19 at 03:46
  • 1
    When you say "that ground is often the neutral wire", if that's true then your PC vendor hasn't provided a properly isolated power supply and is violating safety regulations in any first-world country. – The Photon Apr 03 '19 at 03:50
  • 1
    The ground on this schematic would be the negative power rail for the entire circuit. Every part of the computer will send current to the same ground that's on this schematic. In electronics, ground usually just means zero volts, not "real ground". – user253751 Apr 03 '19 at 03:53
  • (Even if it was real ground, you didn't account for the fact that RXN and RXP should change with opposite polarity to each other, so the voltage on the capacitor shouldn't change) – user253751 Apr 03 '19 at 03:55
  • @ThePhoton: That "properly isolated power supply" (and many are not) is still only providing DC isolation. But you really want a GFCI to trip if there's significant AC leakage, because you're trying to prevent not only burns but also fibrillation. – Ben Voigt Apr 03 '19 at 03:56
  • @immibis: They both vary between 0 and 3.3 volts above ground. – Joshua Apr 03 '19 at 03:57
  • 1
    @Joshua One goes down to 0V at the same time the other goes up to 3.3V and vice versa - that's how differential signals work. The capacitor voltage will be around 1.65V at any time. – user253751 Apr 03 '19 at 03:58
  • @immibis: There's a 2 in the calculation dividing the 3.3V that accounts for that. – Joshua Apr 03 '19 at 04:05
  • 1
    @BenVoigt, There will be a capacitive coupling across the isolation barrier. This should only pass current signals at well above the mains frequency. Meanwhile the GFCI response time is measured in milliseconds. – The Photon Apr 03 '19 at 04:33
  • @ThePhoton: I was thinking less about capacitance and more about the (intentional) mutual inductance that exists in both Ethernet magnetics and AC step-down transformers. The former, as well as DC-DC switching power supplies, I agree are designed to couple at higher frequencies. The AC step-down transformer is definitely designed to couple 50-60Hz, though, which is slow enough to trip the GFCI (unavoidably) – Ben Voigt Apr 03 '19 at 04:41
  • 1
    @BenVoigt, but the mutual inductance can only produce a balanced current on the other side, equal current flowing in one terminal and out the other, which doesn't trip a GFCI. – The Photon Apr 03 '19 at 04:47
  • 1
    The RJ45 socket is with magnetics. It means that motherboard signals are isolated from the cable signals with transformers. Even the transformers are driven differentially. There cannot be a DC path via the magnetics unless they are faulty. Only DC path would be from RJ45 connector chassis, which is connected to PC chassis, which in turn is connected to PE on mains plug within PC power supply. And only AC path is via 1nF capacitor from RJ45 side of the transformer to chassis ground. – Justme Apr 03 '19 at 05:50
  • @Joshua The current through a capacitor is proportional to the rate of change of voltage. If the voltage does not change there is no current. – user253751 Apr 03 '19 at 20:43
  • @immibis: And what do you think the rate of change of voltage is in gigabit ethernet? – Joshua Apr 03 '19 at 20:47
  • @Joshua On that capacitor, zero, because the signal is differential, like I already said. (But also it's irrelevant because current through that capacitor is not a ground fault) – user253751 Apr 03 '19 at 20:51
  • @immibis: My ethernet cables are more than a nanosecond long. – Joshua Apr 03 '19 at 20:55
  • @Joshua What's your point? On that capacitor, zero, because the signal is differential, like I already said. (But also it's irrelevant because current through that capacitor is not a ground fault) – user253751 Apr 03 '19 at 21:02
0

Tripping a GFCI usually occurs when there is a missmatch between the current going in and the current going out. While there could exist mutual inductance between cables, I don't think it would generate enough current because properly designed Ethernet ports have megaohms of impedance at DC. I'll find some impedance graphs for the chokes tomorrow, but if I remember right there is high attenuation for lower frequencies through chokes, and highly unlikely to pass much current 60Hz through DC.

If the cable was improperly built there could be a pathway there.

manassehkatz-Moving 2 Codidact
  • 2,522
  • 11
  • 19
Voltage Spike
  • 75,799
  • 36
  • 80
  • 208