I am designing a dumb "isolation box" that is to sit between a computer and medical data acquisition equipment. The purpose of this box is to inject power and to provide communications and digital signal isolation. The overall cable length that the digital signals (short pulses >10µs) will travel is 10m or less, but edge integrity is critical.

On the isolated side of the box I only have access to 48V DC, and would like to avoid adding another over-specified switching power supply just to power the digital interface (I already need one for the non-isolated side). I would like to get away with just "passive" optocouplers and a minimum of additional circuitry on the data acquisition equipment side, but I am concerned with proper termination of the 100Ω differential pair.

The receiving side seems simple enough as it would be a relatively high-current signal driving the optocoupler LED (and a balancing diode) that can use a 100Ω resistor on each end of the line.


simulate this circuit – Schematic created using CircuitLab

The transmitting side seems more problematic as any solution I can think of would imply a reduced signal level that would then require a level shifter or comparator on the other side (such as this one) which would then require a few more components to function and to add proper ESD protection.


simulate this circuit

Am I over-complicating this? Should I just byte the bullet and add the wasteful power supply and a couple of RS-422 transceivers at each end?

  • \$\begingroup\$ Welcome to EE.SE. First of all use a ultra fast digital opto-coupler like a 6N137, rated to 10mbps. R1 plus R2 in series should not exceed 350 ohms or the maximum clock rate will drop off below 10mbps. Use unshielded twisted pair like Ethernet does, with an opto-coupler at the receiving end only. That is a partial answer at best. \$\endgroup\$
    – user105652
    Oct 28, 2018 at 0:49
  • \$\begingroup\$ @Sparky256 The isolation barrier has to remain inside the box, I cannot have partially isolated lines traveling alongside isolated ones. So my problem is really in the transmitting side. In principle I could use a second optocoupler to terminate the transmitting side, but that is just an inefficient level shifter. \$\endgroup\$ Oct 28, 2018 at 1:11
  • 1
    \$\begingroup\$ Well, to start with, the pairs of wires you show are not in any way differential or balanced. If you really care about signal integrity, you're going to want to balance both the impedance and the voltage/current in each pair. How many pairs are in your cable? Can't you devote a pair to sending some power down to the left side of your diagrams? \$\endgroup\$
    – Dave Tweed
    Oct 28, 2018 at 2:26
  • \$\begingroup\$ @DaveTweed The signal of interest is differential, even if the common-mode is all over the place. But you bring up a good point, I had not considered the effect these unbalanced pairs will have on the other pairs in the cable. That might settle the issue. But no, I don't have any more available pairs. \$\endgroup\$ Oct 28, 2018 at 2:36
  • \$\begingroup\$ your signals aren't balanced, can you use coaxial cable instead? \$\endgroup\$ Oct 28, 2018 at 3:54

2 Answers 2


Opto Isolators are non-ideal in many ways, including speed and difficulties with charging the driven cable capacitance.

You would be better to consider the new generation of digital isolators such as those made by TI.
For example if you are isolating at the send (Tx) side, the TI ISOW7821 is a two channel isolator that can be run from a 3.3 - 5V supply. It's good for 100Mbps so should be adequate for your 10us pulse/timing requirement. If you can't find a 3.3-5V supply then you'd have to generate this.

enter image description here

This application shows a CAN Bus being drivern, but the isolator could be used to drive your signal cable with almost any configuration you want.

  • \$\begingroup\$ Those are an alternative, but they (1) require power on both sides (which is what I am trying to avoid), (2) have additional EMI considerations, and (3) most of them don’t meet medical qualifications. I used a TI part with internal isolated supply in a related application before, but we had to scrap that revision because of excessive EMI from the internal power supply. I am considering Linear Devices magnetic isolation part, but its power supply running at 180MHz makes me nervous. \$\endgroup\$ Oct 28, 2018 at 21:41
  • \$\begingroup\$ I believe that that's the particular TI part we used and I referred to in the previous comment as causing EMI problems. They claim to be seeking medical (60601-1) certification, but their own specs only specify 1000Vrms isolation. I am considering the ADuM6201 from Analog Devices (analog.com/media/en/technical-documentation/data-sheets/…) which has the proper certifications (including a 5000Vrms 1min isolation and a >8mm creepage path). \$\endgroup\$ Oct 28, 2018 at 23:07

Just for completeness. I decided against this. I just went with standard RS-422 Rx/Tx differential pairs driving a medical-certified AduM2286 magnetic isolation IC and a couple of (over-dimensioned) supplies. The reasons for this are:

  1. I did not want the wandering common mode of these signals to interfere with my main communication channels, which travel along these cables and are unshielded.
  2. The application relies on a floating ground and is very sensitive to changes in this ground. Having a wandering common mode that can couple to the environment and move this ground around would be a problem. The relatively large currents and voltages required could have become an issue.
  3. Optocouplers degrade relatively fast. We have had issues with these failing after just a few of years in service, even with some apparently conservative design choices.
  4. Although there are all-in-one RS422 magnetic isolation ICs with integrated power supply from the same company, these run the power supply in excess of 150MHz. It is quite cumbersome to get the PCB design to pass EMC regulations (particularly without a metal case).

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.