I have to provide an isolated voltage that spans between 0V and 3V. Accuracy and space are more important than cost.

The only solution I see (without go to digital) is to use something like ISO122/124. The drawbacks are:

  • they require a dual supply, so I need two DCDCs to provide rails for both sides
  • space: the package is quite big and also the two DCDCs are bulky

I've already two isolated 5V voltage on both sides, and I would like to use them.

I found many cheap isolation amplifier but with only few hundreds mV of input range (they are for shunt resistor). Another way is to scale down and then up my signal to fit these constraints but I'm afraid about the accuracy when I'm close to 0V.


I add some more information about my scenario:

  • the signal is pretty close to DC, say 1 Hz bandwidth
  • on the second side there is a 16-bit ADC, the accuracy needed is about 1-2 LSB

To summarize the question, you want to get a analog signal across isolation. The bandwidth is 1 Hz, and you need about 1 part in 32k accuracy.

This needs to be done digitally. Analog methods good to 15 bits will be hard to find, and expensive if you do.

This low bandwidth is well within the range of delta-sigma A/D converters. Those can easily do 20 bits, with many available better than that. Measuring the input signal and converting it to digital is therefore no problem.

The digital signal can be transmitted across isolation easily. In this case, a single opto-coupler would do. You don't even need a fast one. Let's say you use UART-style serial data at 300 baud, and you transmit 3 bytes per sample. That allows transmitting 10 samples/second, which is enough to carry 1 Hz bandwidth quite comfortably.

Since you have 5 V available on the receiving side, it's no problem to capture the output of the opto-coupler, feed it into a UART, process the bytes from the UART, and drive a D/A accordingly. It would be hard to find a microcontroller that couldn't do this rather easily. All you need is one with a built-in UART, which is most of them. There are plenty of 16 bit D/As out there, so that's no problem either.

You might even look at filtered PWM out of the microcontroller instead of a separate D/A. To get the equivalent of 16 bits out, you need 65535 PWM slices in a PWM period. For 10 samples/s, you need a minimum PWM clock of 655,350 Hz. Finding a micro that can do 1 MHz PWM is trivial, but you need to look more carefully for one that can do 16 bit duty cycle. Generally, you'll get that with a "16 bit" micro. Pretty much any PIC 24 or dsPIC can do this task.

Keep in mind that one limit on output accuracy is how accurate the voltage reference is that the D/A (in whatever form) uses. 16 bits is one part in 65536, which is 0.0015%. That is actually the toughest requirement of this whole problem.


You should probably post more info about your scenario, to give us more hooks into your issues. You don't even tell us what you need to amplify, so it seems like you need an analog isolator, not an isolation amp, per se. You haven't mentioned specs for accuracy, either.

Isolators that are rail to rail might be problematic. You have a 3V dynamic range, so you should be able to come up with a 0V-5V scheme with enough headroom on either side. You might consider something like an HCNR200, but you might need to put a small offset on your input and subtract it later on the isolated side. There are some creative ways that you can use that chip -- but all of them require at least one op amp on either side as support circuitry.

If you're trying to just sample across an isolation barrier, you can probably just handle the offset after sampling.

Another option might be to sample on one side, and send the signal with a digital isolator.

Hard to be more specific without more info on your scenario.

  • \$\begingroup\$ Thanks. I've update the question with some more information. Feel free to ask if I'm still missing something. On the other side of the barrier there is a resistor divider and an ADC but I have to provide the actual signal (1:1) because I have no control on the target board. \$\endgroup\$ – Mark May 12 '17 at 14:42
  • \$\begingroup\$ @Mark -- What are your requirements for value? Is this a one-off, or is this for production? If it's a one-off, it would be tempting to pop an ADC on the isolated side, I2C or SPI, and isolate it with the appropriate digital isolator. \$\endgroup\$ – Scott Seidman May 12 '17 at 14:47
  • \$\begingroup\$ The volumes are extremely small: about 10 pcs / year.Yes this is the way I was (cryptycally) talking when I said "without go in digital". But to provide the correct analog signal I need also a DAC (and MCU that I already have on the first side). \$\endgroup\$ – Mark May 12 '17 at 14:50
  • \$\begingroup\$ +1 for linearized opto \$\endgroup\$ – Reinderien Oct 3 '18 at 13:37

Consider using a linearized optocoupler - this is specifically what they're for. Since you're short on space you'll want one of the SMD packages like:

Within those families there are a few different options, but the concept is identical: you need op-amps on both sides for linear output. There's a great walkthrough better than I can be bothered to write here.


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.