I am working on an industrial control board + node network system. This system needs to be very robust, both for signal integrity as well as protection from transients.

The basic setup is shown here:

enter image description here

Since the Ethernet connection going to the server is fully isolated, I don't think there really is much to discuss there. Going from the control board to the IO board is a CAN bus and balanced audio line. These will either be on a separate cable, or on the same Cat6e cable.

All cables will be shielded, with the shield directly terminated to the enclosure/chassis. I have left the PCB ground and chassis unconnected in the image, as I would like suggestions on that as well. The plan would be to dedicate a single PCB layer for chassis ground. (Of course, if chassis gnd = digital gnd, then there will be multiple layers for them.)

For the following discussion, I will only show the CAN bus setup, as the audio will be fully isolated thanks to the transformers. It will also not require a common ground wire to equalize the grounds.

Option 1:

enter image description here

No isolation. Neither the CAN nor audio lines will have any isolation. To ensure that the common mode voltage stays within a safe range, CAN_GND = CONTROL_BOARD_DIG_GND = IO_BOARD_DIG_GND.

Option 2:

enter image description here

Full isolation on one side (let’s say the control board side for discussions sake.)

Option 3:

enter image description here

Full isolation on both sides.

For the isolation, assume that for the CAN, some sort of powered, digital isolator is used. For the audio, as stated above, a 1:1 transformer will be used.

My assumptions/thoughts:

  1. Option 2/3 is an improvement over option 1, as it eliminates most ground loops. (There will still be a potential loop from chassis 1 -> shield -> chassis 2 -> earth -> chassis 1. Either capacitively from the chassis to earth, or if both metal boxes are mounted to a metal connection that is earth grounded.)
  2. Option 3 may be an improvement over 2 because any surges/transients that hit the cable will be limited to the CAN transceiver, since both sides are isolated.
  3. If there was a 8kV strike on the cable, option one would potentially show that voltage on the digital ground for both boards. Unless it was fully clamped and protected, this would destroy pretty much everything on both boards. Option 2 would show the voltage on the IO board digital ground only. Option 3 should be fully protected. Note: The CAN transceivers should survive in option 3, since the relative voltage between tx/rx and isolated transceiver ground would stay low, since all conductor would (or should) show the same 8kV.
  4. Whichever route I choose, I will still fully protect lines with TVS diodes.


  1. Are my assumptions correct for the most part?
  2. If they are, then I will likely go with option 3. Should the chassis on both boards stay isolated from the digital ground?
  3. If I go with option 3, should I use a safety cap to bridge the isolation gap? How to use isolation to improve ESD, EFT and surge immunity in industrial systems seems to think that I should.
  4. Any other notes/thoughts that I am wrong or missing here?
  • \$\begingroup\$ Look what this chip does on page 18. I have used them for isolated 485 and page 18 describes how you add a safety cap although it's for noise reduction really: analog.com/media/en/technical-documentation/data-sheets/… \$\endgroup\$
    – Andy aka
    Jun 25, 2022 at 15:05
  • \$\begingroup\$ I think you have too many questions in one question. It would take a book about subject of grounding in a system to answer them, but what the whole system even is? For example, likely a server PC has chassis earthed and likely Ethernet connector metal shield is connected to chassis, and a shielded CAT cable connects shield to connector shell. So your device metal chassis will likely be earthed too. So depending on why you want galvanic isolation and where is a mystery, what problem there is that isolation solves? And usually shield is for shielding, not equalizing potential whics needs a wire. \$\endgroup\$
    – Justme
    Jun 25, 2022 at 15:35
  • \$\begingroup\$ Transformer isolated balanced audio doesn’t require a separate ground line with it. \$\endgroup\$
    – Bryan
    Jun 25, 2022 at 16:10
  • \$\begingroup\$ @Andyaka Thanks for that. Interesting how they recommend just using an "embedded" cap instead. Makes a lot of sense. \$\endgroup\$ Jun 25, 2022 at 18:26
  • \$\begingroup\$ @Bryan Whoops, you are indeed correct. I will make some edits to the post. Much appreciated. I do think that now makes option 3 a no-brainer, since both the audio and CAN will be of a similar setup. \$\endgroup\$ Jun 25, 2022 at 18:27

1 Answer 1


Any other notes/thoughts that I am wrong or missing here?

Answer: All your design Specs.

Rule #1

  1. All designs must have "SPECS" or a design specification before choosing an implementation.

  2. All specs must be validated (DVT), documented, and revised/guaranteed for customers with process margins, including yours. This is part of a contract that you choose to warrant.

  3. Choose from Industry standards before re-inventing the wheel, unless you know of specific exceptions. Then compare different experts' advice/ IC datasheets and follow the common guides to achieve this.

  4. Under the general guideline of EMI, ESD is a subset with well-defined standards such as the human body model, HBM of the fingertip (100 pF), and the Cart model. There are also EM ingress standards for pulse, continuous (CW), and modulated carrier which may be coupled by L or C physical properties of all conductors and insulators in the network as well as a choice of isolation and ground (=0V). Crosstalk is a parameter of choice for suppression levels. Often these are symmetrical with some emission spectrums by a shared filter.

enter image description here ref


enter image description here


  1. Are my assumptions correct for the most part?
  2. If they are, then I will likely go with option 3. Should the chassis on both boards stay isolated from the digital ground?
  3. If I go with option 3, should I use a safety cap to bridge the isolation gap? How to use isolation to improve ESD, EFT and surge immunity in industrial systems seems to think that I should.
  4. Any other notes/thoughts that I am wrong or missing here?


  1. yes
  2. Optional C=1nF RF cap like CAT5 terminations. or PE bonded to suppress SMPS floating CM noise. Verify with tests.
  3. if protection diodes are faster than the arc surge Yes, if not maybe. It depends on layout and surge current paths, crosstalk or ground shift and slew rate in kV/ns. Model using 10 nH/cm on wires.
  4. See 1st answer.

If ESD is rated for an impulse <10 us @ 8kV and the device is rated for 4kV for the 60s and you want to improve choose a tougher spec like 8kV for 1s, then one must examine all sources of Breakdown voltage using a current limited low C tester, examine slew rate limiting filters to attenuate 8kV impulses with impedance matching and current limiting tradeoffs to clamp. however it is easier to limit current from crosstalk if floating but CM rejection of RF depends on cable quality for imbalance, kinks, and dB of isolation at RF spectrum from impulses is an additional task.

For example, a BIL-rated insulator for 200kV 1,10us impulse can easily break down at 50kV continuous AC or DC due to air contamination with %RH and surface creepage. This stress is unlikely for CANBUS environments but the ratios tell you about ionization delays and how greater impulses can be withstood. E.g. 8kV 100 pF ESD.

  • \$\begingroup\$ Erm, I suppose thanks for that? Didn't really help much, or any. When the customer doesn't require meeting a specific standard for their application, finding one for the sake of finding one doesn't really solve that much. Typically they want it to be resilient "enough" to be reliable. At least that's how most contract design work, well works. Also, even with a specific standard to design towards, most of these questions would still be valid, as the standard won't lay it out completely for something like this. \$\endgroup\$ Jun 25, 2022 at 18:31
  • \$\begingroup\$ When your customer doesn't know any better , you should and make a guestimate and choose a std. So follow the rules in this datasheet or choose another, If you cannot , then I cannot help, nor anyone. I used to write datasheet specs from customer's fuzzy light descriptions and even "Paper Napkin" specs from Lucent. For DIY it doesn't matter and then produce something I verified to same, even with dry ice and a heater in a picnic box if you don't have a Tenney Chamber \$\endgroup\$ Jun 25, 2022 at 19:13
  • \$\begingroup\$ OK, for discussions sake, I choose to follow the datasheet you shared. Now, I still need to answer my above questions... Lol, that's what I was getting at, having a standard really doesn't help things for what I need. \$\endgroup\$ Jun 25, 2022 at 19:45
  • \$\begingroup\$ I once had a Power One 180W that failed miserably for Hipot after the process control was slack moving from SD to Mex. They tested with DC floating. I tested with DC PE bonded. This is more stressful on the highest impedance insulator on the AC side and caused breakdown arcs and destructive failures. I later added a 1M current limiting R to my HiPot test and disqualified them until all issues fixed. Their QA Mgr was in disbelief. We used 1k units /mo . end customer was AVAYA , my design. \$\endgroup\$ Jun 25, 2022 at 20:24
  • \$\begingroup\$ Thanks for your edits, very helpful! With that in mind, and seeing ti.com/lit/an/slla486a/slla486a.pdf?ts=1656184482024& (Number 7) and analog.com/media/en/technical-documentation/application-notes/… (Page 3.SYSTEM ISOLATION OVERVIEW), I think the plan may be to use option 3, but tie the isolated ground on the control board to PE, so all will be referenced to that single ground point. Still fuzzy on what the chassis would be, but at least making progress! \$\endgroup\$ Jun 25, 2022 at 21:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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