I'm trying to galvanically isolate a device which is controlled by ethernet (1000base t) via computer. I've been looking and apparently the ethernet standard provides some isolation, but after I connect the cable to my otherwise isolated device to my computer, my multimeter reads short between its ground and earth ground. So clearly it's not isolating enough.

I'm wondering how I should proceed; according to wikipedia, gigabit ethernet is 8-bits, so a digital isolator won't work. Can anyone suggest a kind of chip that would work? They don't seem to make 8 bit digital isolators. In any case, it seems the ethernet signal can go below ground, so I'm not sure opto couplers would be a (direct) option, although I suppose I could bias them before and remove the bias after with a cap? It also seems like they are not fast enough for gigabit ethernet...

  • \$\begingroup\$ What magnetics part did you use? How did you connect the pins and the housing shield to your circuit? \$\endgroup\$ – The Photon May 29 '15 at 16:01
  • \$\begingroup\$ These are stock ethernet ports, one on a computer, the other on the redpitaya (the microcnotroller I want to isolate). I didn't use anything special, just a regulat cat6 cable. \$\endgroup\$ – Paul L May 29 '15 at 16:43
  • \$\begingroup\$ OK, now I read the question more closely. A short between the cable ground and earth ground is expected for EMC/ESD reasons. It does not mean that the cable shield is connected to the PCB ground. It does not mean you don't have isolation between the cable and circuits inside the box. You need to open up the box and measure the resistance between the cable shield and a circuit ground on the pcb (whichever pcb you are worried about having isolated). \$\endgroup\$ – The Photon May 29 '15 at 16:49

Buy a pair of gigabit fiber media converters and a few meters of fiber. That will give you more than enough isolation.


Ethernet does provide 'transient' isolation (to around 1.5kV but don't quote me on that) as distinct from functional isolation (a permanent ac or dc bias, which is a different spec), via a transformer which may be a discrete transformer on the PCB, or integrated into the RJ45 connector ("integrated magnetics"). This provides a floating differential pair driven/received by the transformer at each end - 2 of them for 10/100BaseT, or all 4 pairs (of Cat5/6) for 1000BaseT. Unless this is a PoE scenario, I would expect open-circuit between the "ground" of the main circuitry at each end. If you don't, I think there's something wrong somewhere.

A separate issue is whether the "ground" of the circuitry of your PC & mains-earth are also connected, which varies - desktop PCs usually are, laptops usually aren't. This has nothing to do with the 'galvanic isolation' you're seeking between PC & peripheral.

GigE isn't "8 bits", it uses all 8 wires of Cat5/6 cabling, 4 (floating) differential pairs. True 'Ethernet Isolators' can provide true 'functional isolation' between each end, for (rare) situations where the "ground" at either end has a permanent ac or dc differential.

Forget about digital isolators or optocouplers, this isn't your problem or your path forward :)

  • 1
    \$\begingroup\$ If it is an STP cable, won't there be a connection through the shield? Shielded cables are becoming more popular these days, to support 10G ethernet. \$\endgroup\$ – Oleksandr R. May 29 '15 at 11:20
  • \$\begingroup\$ The pairs are not float, they are typically DC (non-PoE) or AC (PoE) coupled between each other and AC coupled (thru HV cap) to the chassis, for examples see here and here. \$\endgroup\$ – asndre Jun 14 '15 at 16:18

Well, this was a lot simpler than I thought. Thanks The Photon for the comments giving me the idea; it turns out the cat6 cable I was using is metal shielded. The metal shield is connected to ground at the computer end, and to the ground on my microcontroller. Using a cable with no metal shield breaks the connection.

  • \$\begingroup\$ There is actually another user (mainly on Physics.SE) named Phonon. \$\endgroup\$ – The Photon May 29 '15 at 16:57
  • \$\begingroup\$ oops! corrected. \$\endgroup\$ – Paul L May 29 '15 at 17:20
  • \$\begingroup\$ There is usually a capacitor as well, in the nF range, but rated at a goodly high voltage (like 600V). \$\endgroup\$ – Spehro Pefhany May 29 '15 at 17:44

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.