When should I use a voltage reference?

[ballsystemlord]

Hello,

As you are probably aware, we can use linear voltage regulators to power our circuits. We can likewise then use a resistive divider to get pretty much any voltage that we want. So, when should a voltage reference be used?

I understand that high-bit ADCs would benefit, as do PSUs (by virtue of trying to regulate their voltage instead of a having a pre-regulated voltage to use). But I was wondering what, if any, rules of thumb there are for deciding if and when to use a voltage reference?

Thanks!

[Konkedout]

Mainly use references when you want voltage to be accurate and low noise.

Having said that I once did a successful production battery charger design where the control circuitry was powered by an original LM317.  This current was in the range of 10-20 mA so not high, and the 12V out of the LM317 was also used as a reference.  The design had a potentiometer to dial in the voltage to about 0.5%.

And.....having said that....The original LM317  from National Semioconductor was a really good chip.  I cannot be sure whether those myriad of different ones available today are as good.

[Smokey]

Noise, disturbance rejection, and stability typically.

If you have other stuff using the same voltage regulator that you are using for your reference, any bounce in that voltage from stuff drawing current will show up in your measurements.  Dedicated voltage reference eliminates that.

[oz2cpu]

when ?
it is quite simple,
what have you got : resolution ?
and what do you want out of it: stability ? Accuracy ? long term short term ? temperature range deviation ?
and a what cost are you willing to put into it to achieve your goals.
you hardly get full performance from 8 bit ADC with a standard regulator and voltage divider resistors
look at the specs, do the math of all parts min-max variations of any parts that affect the reading.

[PGPG]

We can likewise then use a resistive divider to get pretty much any voltage that we want.

Now list the elements using voltage obtained from the divider not loading it (any load, even a voltmeter, changes the voltage).

[MariuszD]

Rules of thumb are made up by people who are too lazy to think. Determining whether you need a stable voltage and how stable it should be is not difficult and requires at most a few seconds of thinking. I don't see the point in looking for shortcuts.

Reference voltage sources are varied, and they do not always produce low noise.

Comparing the reference voltage source with the linear voltage regulator, we have the same band-gap reference inside. But the voltage regulator is less accurate and often heats up because it conducts significant currents to the load. Temperature changes and load changes further worsen the accuracy.


Cheap voltage reference sources like the TL431 are not very accurate. TL431 is very often used simultaneously as a reference voltage and error amplifier in SMPS power supplies. The output voltage of these power supplies is not exceptionally precise.

We can compare it to a Zener diode. Zener diodes <4V had high dynamic resistance and a significant temperature coefficient, and they also required a considerable current for proper operation; these diodes have been completely replaced by band-gap sources. Zener diodes close to 5-6V have a low temperature coefficient. The ultra-stable LTZ1000 reference is based on a subsurface Zener diode. Above 40V, Zener diodes are available, but there are almost no band-gap integrated circuits. If someone wants to make a 150V voltage regulator in a simple way, they will use a Zener diode.

[ballsystemlord]

Rules of thumb are made up by people who are too lazy to think. Determining whether you need a stable voltage and how stable it should be is not difficult and requires at most a few seconds of thinking. I don't see the point in looking for shortcuts.

Umm, I wasn't trying to find a shortcut, rather I understand that not everything EEs learn comes from text books. There is life experience, practical application, etc.. So, I was trying to see if there was more to the subject.

Thanks

[voltsandjolts]

I understand that high-bit ADCs would benefit, as do PSUs (by virtue of trying to regulate their voltage instead of a having a pre-regulated voltage to use). But I was wondering what, if any, rules of thumb there are for deciding if and when to use a voltage reference?

My thought process would be;

1/ Evaluate the requirements for measurement accuracy and stability over the operating temperature range.

2/ Can the measurement be ratiometric, such that no voltage reference is required? For example, measuring a Wheatstone bridge output versus bridge supply is ratiometric and can provide high resolution data (e.g. with 20 bit sigma delta adc) without any voltage reference. Same for PT100; measure the ratio compared to a 100R reference resistor.

3/ Is a voltage rail sufficiently stable and accurate to use as a reference?

4/ Does my mcu or adc have a suitable built-in voltage reference?

5/ Damn, need to spend money on a voltage reference! Evaluate cost/performance of references compared to application requirements.

[RoGeorge]

As you are probably aware, we can use linear voltage regulators to power our circuits. We can likewise then use a resistive divider to get pretty much any voltage that we want. So, when should a voltage reference be used?

First thing to say, any voltage regulator has a voltage reference inside.  So, if you have a linear regulator with a resistor divider, you are already using the voltage reference from the linear regulator.  Modern IC linear regulators have pretty good Vref inside them, often a buried Zener or a bandgap reference.

All good so far, except not all Vref are the same.  Some are better than others.  The things that matters most is voltage accuracy (how far they are from the specified voltage), temperature drift, long term stability and noise.  These parameters are given in the datasheet of the voltage regulators, or in the datasheet of voltage references.

Usually, a voltage reference IC will have better performance than a linear IC regulator.  If that difference in performance is needed or not in a circuit, that's up to you to decide.

For example, if you have a 16bit ADC, and you use a 7805 regulator (~2% error) as a voltage reference, those 2% will mean about 1300 counts of error out of the 65536 full scale.  If a +/-1300 accuracy error is acceptable for your ADC, that depends of the destination of the circuit.  It's up to you to decide if a 7805 is good enough, or if the application needs a better Vref.

Selecting the Optimum Voltage Reference
https://www.analog.com/en/resources/technical-articles/selecting-the-optimum-voltage-reference.html

Voltage reference selection basics
https://www.ti.com/lit/wp/slpy003a/slpy003a.pdf

[xavier60]

I wanted something better than an LM7805 to power a PIC as I'm using the 5V supply for the ADC reference.
I chose a TL431 characterized for a wide temperature range, https://au.mouser.com/ProductDetail/595-TL431BQLPM

[David Hess]

I wanted something better than an LM7805 to power a PIC as I was using the 5V supply as the ADC reference.
I chose a TL431 with a wide temperature range, https://au.mouser.com/ProductDetail/595-TL431BQLPM

If you want a real improvement, use the circuit below, modified with modern parts, to deliver an accurate and low noise supply directly to the power and ground pins.  Replace the regulator with a 2.5 volt regulator, or a bipolar output stage if you like, (1) and the JFET with a PNP bipolar transistor.  Add a 2.2 or 2.7 kilohm resistor between the regulator output and adjustment pin to force 1 milliamp through the transistor.  The reference could be a 2.5 volt bandgap so the resistor divider now uses equal value resistors for a gain of 2 and 5 volt output.  The operational amplifier is not critical because reference error and noise dominate, but low drift and high open loop gain help.  I have used this circuit before to provide 10.000000 volts at up to 1 amp with excellent results; it is essentially a power reference, and illustrates the major difference between regulators and references.

With an integrated regulator, thermal feedback from the power output stage ruins the performance of the reference.  If these are separated, as shown below, reference performance can be achieved at high power levels.  The best precision operational amplifiers are low power devices for the same reason, and should be lightly loaded for best performance.  Here the JFET or PNP transistor reduces the loading on the operational amplifier although an integrated regulator with low adjustment pin current can be driven directly.

(1) I like using an integrated regulator for the output stage because it provides built in fault protection.  The relatively high noise and drift of the integrated regulator is reduced and dominated by the excess gain of the operational amplifier, so can be ignored.

[CatalinaWOW]

My thought process goes like this.  Why do I care about voltage?  Three obvious reasons, I am measuring things, something I am powering has specific requirements or there is a safety requirement.  From there it goes into the deeper thinking stuff.  How well do I need to know.  What environments.  How long.  Price limits.  Design time limits.  The latter is often the driver.  References are cheap and often it isn't worth the time to come up with the "optimum" design.

[negativ3]

It can help greatly with ADC stability when the ADCREF pin is fed separately from the supply rails.
I found a factor of ten improvement when testing a certain dev board recently which had the 3V3 rail tied to ADCREF, and when swapped to a decent reference, gains in stability were notable.

[TimFox]

A simple answer is that ICs sold as voltage references should not be used to supply large currents to the load, but can be used inside a larger stabilizer circuit where the heat is kept away from the reference.

[CatalinaWOW]

It can help greatly with ADC stability when the ADCREF pin is fed separately from the supply rails.
I found a factor of ten improvement when testing a certain dev board recently which had the 3V3 rail tied to ADCREF, and when swapped to a decent reference, gains in stability were notable.

True.  But only meaningful if you needed more stability.  There are other ways like a simple RC filter on the 3V3 rail.  Less improvement and further limits range, but might be the right answer in some cases.

This is just saying what several people have said in a different way.  The answer to the OPs question lies in his application.  Not in the experience of others.  The experience of others benefits when someone has discovered a clever outside the box solution to some problem.

[negativ3]

It can help greatly with ADC stability when the ADCREF pin is fed separately from the supply rails.
I found a factor of ten improvement when testing a certain dev board recently which had the 3V3 rail tied to ADCREF, and when swapped to a decent reference, gains in stability were notable.

True.  But only meaningful if you needed more stability.  There are other ways like a simple RC filter on the 3V3 rail.  Less improvement and further limits range, but might be the right answer in some cases.

This is just saying what several people have said in a different way.  The answer to the OPs question lies in his application.  Not in the experience of others.  The experience of others benefits when someone has discovered a clever outside the box solution to some problem.

I answered the question and gave a real world example.
What exactly is your point?

[xavier60]

I wanted something better than an LM7805 to power a PIC as I'm using the 5V supply for the ADC reference.
I chose a TL431 with a wide temperature range, https://au.mouser.com/ProductDetail/595-TL431BQLPM

The PCBs containing the regulator design in Reply #9 arrived, so was able to test for the first time.
I used 8.2K for the divider resistors as the value was already being used on the board.
It came up at 4.997V. I don't expect all subsequent builds to land this close.
Theoretically, the regulated voltage should be 2x 2.495= 4.990V.
The few microamps drawn by the R pin adds a few extra millivolts across the top divider resistor.

[CatalinaWOW]

It can help greatly with ADC stability when the ADCREF pin is fed separately from the supply rails.
I found a factor of ten improvement when testing a certain dev board recently which had the 3V3 rail tied to ADCREF, and when swapped to a decent reference, gains in stability were notable.

True.  But only meaningful if you needed more stability.  There are other ways like a simple RC filter on the 3V3 rail.  Less improvement and further limits range, but might be the right answer in some cases.

This is just saying what several people have said in a different way.  The answer to the OPs question lies in his application.  Not in the experience of others.  The experience of others benefits when someone has discovered a clever outside the box solution to some problem.

I answered the question and gave a real world example.
What exactly is your point?

I guess it is this.  If you know enough to be designing things you know you use a voltage reference to get a voltage with some degree of accuracy.  A zener with a series resistor is a voltage reference.  The 3.3 volt supply is a voltage reference.  One which many understand is not a very good reference, but not very good is a loose term.  It might actually be better than a zener.   There are threads dozens of pages long in the metrology section on how good various voltage references are, and how to make them better. 

So saying a voltage reference in your A/D circuit improved performance doesn't really give any information on whether to use a voltage reference.  There is an implication that in your application you weren't happy with the performance using the 3.3 volt rail.  And an accompanying implication that it was better with the voltage reference.  So not really any help on whether or not to use a voltage reference.  There is an enormous design space for A/D converters.  There are applications which effectively work with 4 bits of performance.  And others that need far better.   At what point did linearity errors in your 8 bit converter become larger than the reference noise errors?  Could you have gotten satisfactory results by synchronizing your samples with the dominant source of ripple on the 3.3V supply?   Was the noise on the 3.3 volt supply random and white so you could have eliminated it with a software filter giving a smaller BOM and physical footprint?  Those are the general type of questions that lead to reasons why or why not to use a voltage reference in a given application.    I will also admit that there are applications where you may just put one in because it might help.   You don't care about space, or unit cost or anything.  Something like an audiophool product where you can spin stories about head room, and performance beyond anything else.  But even there you need to borrow from the medical profession and convince yourself you are doing no harm.  Usually that isn't too hard, but sometimes polishing the apple a bit can bite you.

[negativ3]

It can help greatly with ADC stability when the ADCREF pin is fed separately from the supply rails.
I found a factor of ten improvement when testing a certain dev board recently which had the 3V3 rail tied to ADCREF, and when swapped to a decent reference, gains in stability were notable.

True.  But only meaningful if you needed more stability.  There are other ways like a simple RC filter on the 3V3 rail.  Less improvement and further limits range, but might be the right answer in some cases.

This is just saying what several people have said in a different way.  The answer to the OPs question lies in his application.  Not in the experience of others.  The experience of others benefits when someone has discovered a clever outside the box solution to some problem.

I answered the question and gave a real world example.
What exactly is your point?

I guess it is this.  If you know enough to be designing things you know you use a voltage reference to get a voltage with some degree of accuracy.  A zener with a series resistor is a voltage reference.  The 3.3 volt supply is a voltage reference.  One which many understand is not a very good reference, but not very good is a loose term.  It might actually be better than a zener.   There are threads dozens of pages long in the metrology section on how good various voltage references are, and how to make them better. 

So saying a voltage reference in your A/D circuit improved performance doesn't really give any information on whether to use a voltage reference.  There is an implication that in your application you weren't happy with the performance using the 3.3 volt rail.  And an accompanying implication that it was better with the voltage reference.  So not really any help on whether or not to use a voltage reference.  There is an enormous design space for A/D converters.  There are applications which effectively work with 4 bits of performance.  And others that need far better.   At what point did linearity errors in your 8 bit converter become larger than the reference noise errors?  Could you have gotten satisfactory results by synchronizing your samples with the dominant source of ripple on the 3.3V supply?   Was the noise on the 3.3 volt supply random and white so you could have eliminated it with a software filter giving a smaller BOM and physical footprint?  Those are the general type of questions that lead to reasons why or why not to use a voltage reference in a given application.    I will also admit that there are applications where you may just put one in because it might help.   You don't care about space, or unit cost or anything.  Something like an audiophool product where you can spin stories about head room, and performance beyond anything else.  But even there you need to borrow from the medical profession and convince yourself you are doing no harm.  Usually that isn't too hard, but sometimes polishing the apple a bit can bite you.

Fair point, but my example was pretty relevant.
On an RP2350 dev board with a custom AFE circuit, the final op-amp stage looks rock solid on the scope. But the ADC readings over serial were jumping ±20 counts. Swapped ADCREF from the 3.3V rail to a cheap ZTL431 and it dropped straight to ±2 counts, matching the AFE output.
Wasn’t some fancy reference, just a basic one that cleaned up the noisy 3.3V rail. For a lot of real projects chasing decent 10-12 bit performance, this is a quick win, not audiotomphoolery.
Application dependent for sure, but this kind of before/after is why people share experiences.
For my own RP2350 build I’m using a better-than-stock LDO, but I’m still leaving the footprint for a reference just in case, easy to compare as another step.

[DrijSkij]

Rules of thumb are made up by people who are too lazy to think. Determining whether you need a stable voltage and how stable it should be is not difficult and requires at most a few seconds of thinking. I don't see the point in looking for shortcuts.

Umm, I wasn't trying to find a shortcut, rather I understand that not everything EEs learn comes from text books. There is life experience, practical application, etc.. So, I was trying to see if there was more to the subject.

Thanks

I was sensing a bit of arrogance too.  Yep -  I said it.

[MariuszD]

Rules of thumb are made up by people who are too lazy to think. Determining whether you need a stable voltage and how stable it should be is not difficult and requires at most a few seconds of thinking. I don't see the point in looking for shortcuts.

Umm, I wasn't trying to find a shortcut, rather I understand that not everything EEs learn comes from text books. There is life experience, practical application, etc.. So, I was trying to see if there was more to the subject.

Thanks

I was sensing a bit of arrogance too.  Yep -  I said it.

I do not intend to show off. I just wanted to convey the harsh realities of design. Reality doesn't care whether you have good knowledge or not. If you don't know the answer to a question, you will be forced to design the circuits from scratch. In hindsight, once you know what you did wrong, the answer seems trivially simple. I'm writing this because I've been thru it myself. When I asked similar questions as the topic author, I didn't know what to pay attention to in order to do it right. I also thot there was some magic in it that isn't in the books, but really, you just need to use the knowledge that's on the first page.

The difficulty certainly lies in determining what you don't know and writing a good specification. When we don't know what we expect from the system, it can't be designed correctly. I learned this from the failures of myself and people with many times more experience than mine.