I’m building some automation equipment for my shop that will be used in extreme noise environments (plasma cutting and TIG welding nearby, for example) and I’m trying to come up with the best signaling for connecting modules that is still reasonably cost-effective. The modules will live in a couple of rack cabinets, and have dual-shielded 8P8C modular cabling connecting units within each rack and between cabinets to carry both moderate-frequency (200kHz-4MHz) and near-DC signals.

My initial direction was to use AM26C31/32 RS-422 ICs on these links, for cheap and easy differential communications.

The ICs are rated for +/-7V common mode, and the power supplies are isolated. Therefore, the racks require some mechanism to limit ground potential differences. The RS-422/485 standard recommends terminating the cable shields at both ends through a few hundred ohms to tie the systems together, but if the building ground is continuous there is now a room-sized loop and it's connected to ground through a low impedance. Some of that might be mitigated by inductors on the shield-ground tie but I’m not sure to what extent, and high-frequency high-energy noise could radiate through the stub.

Second, in the case that the line is undriven, such as when a cable is unplugged, the received voltage has an indeterminate state. Therefore, they need fail-safe biasing, which potentially connects uneven power supplies. RS-422 Schematic I am now considering using a single-ended driver for one wire of each twisted pair, the other wire being a dedicated return tied to ground at the driver but not receiver. Receiving would be done by optocoupler, resulting in full isolation. This has the advantage of ignoring the GPD, removing the need to tie the cabinets strongly to ground and so while the ground loop through the building still exists, it should be better isolated from the ground plane. The other advantage is that logic zero in this system is <5mA or so, and the undriven state of the line is 0mA for a decent noise margin without biasing. Sadly, high-speed optos are expensive. CAN Schematic It’s still technically differential, but the swing is half that of the RS-422 drivers and I’m not sure how much unbalanced drive will affect noise emission at the wavefront vs the opposing magnetic fields of balanced drive. Four pairs share each cable so that noise would be of concern. CAN uses unbalanced drive in voltage mode and is generally considered pretty robust, but it’s not clear how much of that is its fault-tolerant protocol and lower data rates at short range vs RS-485.

I will have to use at least one of each, since off-the-shelf stepper drives use optoisolator inputs, and there is a required RS-485 bus to serve, but RS-485 transceivers are available with features to mitigate the need for biasing networks (though not GPD). So, given the possibility of a 200A welding arc going off a few feet away, what is the most noise-immune thing I can do? Can I solve the problems with RS-422 without going to expensive fully isolated receivers? Can anyone shoot some new holes in my non-EMC-engineer-approved plan? Am I overthinking it and both systems are good enough?

Thanks for any input anyone can provide.

  • \$\begingroup\$ using the opto (or a galvanic isolator chip) to eliminate ground differences might go a long way to a reliable signal. Shielded 422 / 485 should be okay even in a moderately noisy environment \$\endgroup\$
    – Pete W
    Jan 5, 2021 at 5:40
  • \$\begingroup\$ also maybe AC couple the shield to GND on one of the ends??? \$\endgroup\$
    – Pete W
    Jan 5, 2021 at 5:47
  • \$\begingroup\$ Generally CAN is more rugged and modern because it has a much better data link layer than dinosaur UART. Though if you are stuck with Modbus then I guess you have no other option. Galvanic isolation is a good idea in either case. \$\endgroup\$
    – Lundin
    Jan 5, 2021 at 7:33
  • \$\begingroup\$ You can mix and match PHYs and data link layers, like use UART with CAN PHY. \$\endgroup\$
    – Justme
    Jan 5, 2021 at 8:01

3 Answers 3


For best possible immunity, I would recommend current loop transmission. It can be shown with some calculation that current loop transmission is superior to voltage transmission by orders of magnitude.

The reason is that an induced error current of e.g. 1mA does not have much influence on a 10mA signal in a current loop, whereas in a voltage loop with e.g. 10kOhms receiver input impedance, this error current would lead to 10V error voltage.

Of course you would still need twisting and shielding of the cables.


Why do you need rs485? Using fiber optic ethernet is reasonably cheap and easy these days. This solves your problem in an instant. As for the stepper drives, shielded cable routed away from the plasma/welder cables should be adequate. I suspect you may be overthinking the problem. For the most part you need to concern yourself with earthing, routing and shielding.

  • \$\begingroup\$ Another point against using fiber optic Ethernet is that also standard Ethernet over CAT cabling uses transformers for isolation. So it would just require a way to ensure the two device grounds are not connected via the CAT cable shield to form a ground loop. \$\endgroup\$
    – Justme
    Jan 5, 2021 at 7:27
  • \$\begingroup\$ There's many off the shelf solutions for ethernet to modbus RS485 and industrial i/o. If you bring Functional Safety into the mix, then expect to pay a few more dollars. In your diagram with the shield tied to local gnd at each end, if you're worried about gnd currents, then use a capacitor from the shield to gnd to shunt high frequency currents. I'm not sure what use the 1M resistor and 33uH inductor - 1M isn't going to pass much current so what does the inductor do? \$\endgroup\$
    – Kartman
    Jan 5, 2021 at 9:58
  • \$\begingroup\$ Shunting high frequencies to ground is exactly what I am trying to avoid. There's up to 200A of broadband-emitting plasma arc nearby, that energy needs to stay as far away from reference planes as possible. The only reason they're tied together at all is to limit static potential between earth and ground. There's no point trying anyhow; the grounds are galvanically isolated by the power supply transformers. Only the shield earth forms a loop. The inductors on the second schematic are vestigial from editing the first. \$\endgroup\$
    – Ben H
    Jan 5, 2021 at 11:20
  • 1
    \$\begingroup\$ What do you want the shield to do? The general idea is that it forms the preferential path for the interfering signal rather than it going via your data signals. The transformers will give you galvanic isolation but will still couple AC due to interwinding capacitance \$\endgroup\$
    – Kartman
    Jan 5, 2021 at 11:58
  • \$\begingroup\$ That is a severe oversimplification of the mechanism of shielding. If it were that simple, countless installations with shields terminated at only one end to break ground loops wouldn't have any effect. You're also still missing that the point is to keep the noise OFF the PCBs and power lines. \$\endgroup\$
    – Ben H
    Jan 5, 2021 at 18:27

Background Anecdotal experience

I faced this exact problem in a large factory in’77 using biphase 100kbd over a wide deviation FM carrier on a high quality coax over a 30m demo test with SCADA robotics Eddy current scanning custom design during the commissioning test with all the stake holders present. A disaster due to an arc welder in the adjacent attached bldg cause a data glitch. The little “Robbie the robot” was walking inverted on a heat exchange tube array and just when it was suspended on one probe leg , ready to rotate and insert & lock the other leg when the Rx received an impossible command to was unlock the only one that was locked by pneumatic collets and fell on the floor of the mock-up bldg. Fortunately no damage but it had to be manually inserted to complete the walk sequence and Eddy Current probe scan of each adjacent tubes out of 2000. Moral of the story is The coax differential ground signal induced by the plasma arc was enough to penetrate the strong carrier signal and alter enough bits to change the command. I used 3 redundant solutions which were designed that week including Hamming codes, controller illegal state masking and more robust filtering.

The plasma arc is broadband and very high E-H field but it is possible to reduce the risk of errors by estimating the signal energy per bit and noise energy per bit and using a spectral matched filter and carrier level to ensure a SNR >20 by design of CNR and SNR conversion. Coax and STP wire alike all have imperfections measured by CMRR and Spectral Transfer Impedance but is hard to come by so it must be tested.

It may require redundancy or reducing the bit rate or improved CM chokes or tighter filtering with group delay implications with bit jitter.

To start , you must define all the variables for signal spectrum, your BER vs SNR integrity requirements, the bandwidth required and then research what types of redundancy, processing power, shielding ( not all shields are the same), double shields, as well as feedthru on power conducted noise reduction testing from AC to DC.

Even this may not be enough, but it’s a good start to solving the problem.

In another factory, a large flashing Xenon safety strobe would block all cell phone communication while HV tests were running as the repetitive arc pulse is a great unintentional radiator.

In another test, Train traction motors 300 meters away would distort many mid and cable TV signals down a back alley due to corroded coax earth grounds for CATV distribution. The noise ingress was significant.

I have dozens of these EMI problem experiences and know that these problems and solutions exist for Products. Another was a large investment firm in the tallest building downtown would get occasional data errors on mainframe disk drives when the International radar pulse coincided during some data transfers , fixed by better signal grounds and shielding.

Good luck. Ask a better question when you get some spectral test results with various cables terminated with arc welding.


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.