Q: Measuring voltage with isolation

[SharpEars]

The basic idea is this:

I have multiple supplies whose outputs are 24V DC. I want to accurately measure their voltages using the 16-bit ADCs of a microcontroller, but without the microcontroller in any way interacting with the supply voltages themselves. In other words, I need to keep the microcontroller completely isolated from the supplies.

The only way I have found to do this is to use a device such as:

http://www.avagotech.com/products/optocouplers/industrial-plastic/isolation-amplifiers-modulators/isolation-amplifiers/acpl-c870

It is basically a 1:1 op-amp with isolation, but these things are expensive and cost almost $8 a piece (and are single channel, not to mention require separate power for each side of the amplifier (i.e., input and output)).

Without isolation, the task is easy: You divide the voltage you are measuring with a resistor divider to fit ADC input range on your microcontroller and you're good to go.

But, with a requirement for isolation between the voltage source being measured and the microcontroller, the problem seems much more difficult and expensive.

Is there a cheaper/better way to use a single microcontroller to measure voltages at multiple power sources while maintaining isolation between the microcontroller and those sources (and the sources themselves, of course)?

[Kleinstein]

Isolation amplifiers are expensive parts. There are a few that also include an isolated DC/DC converter to power the input side. You still have to make sure the isolation amplifier is good enough and does not cause to much errors.

Sometimes it can be simpler to have an separated ADC to do the measurement and transfer the digital signal with an opto-coupler or digital isolator.

There is a way to charge a capacitor and than use a special kind of relay (that ensures there is no cross connection) to switch both ends from the input side to the output side.

[oldway]

If you don't need to measure, but only know if yours 24V power supplies are ok (for exemple, output voltage between 24.5 and 23.5V), you can have a windows comparator in each power supply (power good signal) and transmit the power good signal to your microcontroler by an optocoupler.
Power ok must be 1 for power supply ok, so you can use the 24V to supply your comparator.

[nctnico]

Last time I designed something similar I used an I2C isolator from Analog devices, a (safety rated) dc-dc converter and an ADC but this was for measuring something connected to mains.

[Ian.M]

You cant get complete isolation + a reasonably accurate 16 bit reading across the isolation barrier without getting fancy.  You'll never get 16 bit accuracy with a commodity analog isolator so you'll *ALWAYS* need either a boutique precision isolator priced to make you weep, or a 16 bit ADC on the isolated side + a digital isolator.  Both require power so if you cant draw it from the isolated side, you'll need an isolated DC-DC converter, and as you want 16 bit accuracy, it will almost certainly need external filtering and regulation.

OTOH if the isolation requirement is simply so the supplies can be stacked in any order, a high input impedance, high voltage tolerant potential divider based differential solution may be possible.  You will run into problems with noise and gain matching so 16 bit accuracy would be extremely difficult.

[ebclr]

AMC1304x High-Precision,
Reinforced Isolated Delta-Sigma Modulators with LDO




http://www.ti.com/lit/ds/symlink/amc1304l05.pdf

[Jeroen3]

You either isolate the input signal. Or you lift the entire mcu/adc to the input voltage.

[Alex Nikitin]

1) How accurate do you want to measure the supplies?

2) How fast do you need this measurements to be?

If the answers are - with ~0.1% accuracy or less and not fast at all - i.e. once in a second, then there is a simpler and cheaper option with a capacitive analogue transfer, using relays.

Cheers

Alex

[rs20]

AMC1304x High-Precision,
Reinforced Isolated Delta-Sigma Modulators with LDO




http://www.ti.com/lit/ds/symlink/amc1304l05.pdf

+1, I am using this part in one of my design.
This chip also has analog output version (which is essentially the same chip with internal oscillator and low pass filter).

Are you kidding, that's not very appropriate to the OP's question at all -- it mandates a microcontroller with a DSFM, or some other solution to convert the bitstream to a useful reading. Any of the countless digital isolators out there, coupled with any of the countless suitable ADCs out there (as already suggested multiple times in this thread), seems like a far more viable solution.

[Ian.M]

Its expensive - $9.38 one-off from digikey and around $4 in bulk so unless the O.P. is going to production and can get it at reel pricing it isn't that attractive.  However its an all in one solution and for low enough sample rates could be multiplexed into either a single DSFM module or ANDed with its clock and a 65536 clock gate pulse and counted in a 16 bit counter on less capable MCUs right down to 8 bit PICs.

[danadak]

There is a technique where almost anything can be linearized, for example
the Light vs I of an ordinary LED, or an opto coupler transfer function. The
basic idea attached.


Regards, Dana.

[Ian.M]

LED output drops with aging and use.  Even with a reference optocoupler to compare against, long term relative drift will result in a need for relatively frequent recalibration, and calibration in general and especially doing temperature sweeps is $EXPENSIVE$.   A pure analog isolation solution is unlikely to be viable for more than 8 bit accuracy even with calibration.

[Alex Nikitin]

A pure analog isolation solution is unlikely to be viable for more than 8 bit accuracy even with calibration.

It depends. As I've said earlier, use a sample and hold circuit with a P/P capacitor and a DPDT relay per each 24V supply, and you can get 16bit + accuracy though at a pretty slow sampling speed. Good for slow but accurate measurements, not good for a continuous fast real-time monitoring.

Cheers

Alex

[Ian.M]

And then you are fighting contact wear and must use relatively large and clunky relays to get enough isolation + a guarantee that if one pole's contacts have welded, the other pole cannot switch. . . .

[Alex Nikitin]

And then you are fighting contact wear and must use relatively large and clunky relays to get enough isolation + a guarantee that if one pole's contacts have welded, the other pole cannot switch. . . .

We don't know how much isolation the topic starter needs. A quality small signal telecom relay is good for over 50 000 000 operation and can provide over 1kV isolation . The operating current during switching is essentially zero, so there is a very little chance of contacts welding.

Cheers

Alex

[SharpEars]

AMC1304x High-Precision,
Reinforced Isolated Delta-Sigma Modulators with LDO




http://www.ti.com/lit/ds/symlink/amc1304l05.pdf

Nice! I like this answer the best so far and am looking into it. I can't find it's analog output counterpart part number, though, as mentioned by blueskull in a reply in this thread:

...
This chip also has analog output version (which is essentially the same chip with internal oscillator and low pass filter).

[Alex Nikitin]

AMC1304x High-Precision,
Reinforced Isolated Delta-Sigma Modulators with LDO




http://www.ti.com/lit/ds/symlink/amc1304l05.pdf

Nice! I like this answer the best so far and am looking into it. I can't find it's analog output counterpart part number, though, as mentioned by blueskull in a reply in this thread:

...
This chip also has analog output version (which is essentially the same chip with internal oscillator and low pass filter).

AMC1301, AMC1100, AMC1200

Cheers

Alex

[SharpEars]

This chip also has analog output version (which is essentially the same chip with internal oscillator and low pass filter).

AMC1301, AMC1100, AMC1200

Cheers

Alex

Thanks, I am looking at those now! The ISO124 also looks interesting, but is several times more expensive than the above.

[Kleinstein]

I would consider an floating ADC like an MCP3421 (cheap 18 Bit sigma delta ADC with internal reference) and an I2C isolator chip. It is affordable and rather low power.

[max_torque]

do you need to measure 24vdc with 16b resolution (<< note resolution not accuracy!!!) or 0 to 24vdc with 16b accuracy?

If the former, ie you only care about a lower voltage range of measurement, then there are various VtoF single IC's that can be set up to send a frequency modulated square wave across an optocoupler for a low cost.

For example the Linear Tech "TimerBlox" devices: http://www.linear.com/products/timerblox

 Alternatively, if you are willing to reduce the resolution requirement, you could use any number of low cost micro's, floating on the power supply side ground with their internal ADC's to do exactly the same thing

[danadak]

LED output drops with aging and use.  Even with a reference optocoupler to compare against, long term relative drift will result in a need for relatively frequent recalibration, and calibration in general and especially doing temperature sweeps is $EXPENSIVE$.   A pure analog isolation solution is unlikely to be viable for more than 8 bit accuracy even with calibration.

Thought and a question.

Using a calibrated current the LED can be examined periodically for light
output. So question is does the behavior of LED light reduction over time
have a known relationship ? If so still hope that a DUT test during manufacture
could solve.

Regards, Dana.

PS : Looking online I see various curves that are not uniform.....

[Kleinstein]

There are several mechanisms that contribute to decrease of LED output efficiency. There are some predictions, but every unit may be a little different. Anyway the better analog Ok (e.g. IL300) use a single LED and use feedback from a second photo-diode. So aging of the LED is not critical, it is only the ratio in sensitivity of the two detectors and maybe the way the light is split.  There is also the option to use a transformer for transfer: so have chopper - transformer and synchronous detector, for higher precision again with an extra secondary with detector for feedback.  If done right this can be very accurate (e.g. ppm level), but also very expensive.

[Doctorandus_P]

My vote is for measuring the voltage locally and then use a digital isolator because it's the least troublesome.

20 years ago a common approach was to build a voltage to pwm ( or frequency) converter and put that signal through the isolation barrier.
Hmm... Put the isolation somewhere halfway a discrete dual slope converter?
Mwa, use I2C / SPI adc and some opto's and be done with it.

[rs20]

There is also the option to use a transformer for transfer: so have chopper - transformer and synchronous detector, for higher precision again with an extra secondary with detector for feedback.  If done right this can be very accurate (e.g. ppm level), but also very expensive.

I can't wrap my head around why one would choose this option over 20+ bit ADC, digital isolator, DAC. (Yes, I know I'm conflating resolution with accuracy here, but still.)