Ultra low noise voltage reference

[altaic]

Apologies if this has been discussed before-- I searched and didn't come up anything that answers my specific question, which is: What is the lowest noise voltage reference practically/cheaply available? I'm looking to characterize the noise on a 24-bit ADC that I'm working on.

As I understand it, even batteries have different characteristics due to their chemistries which can introduce or pick up noise. I recall someone mentioning mercury batteries awhile ago, but I don't think noise was discussed. Am I overthinking it, or is there a preferred ultra low noise voltage reference out there that won't break the bank?

[uncle_bob]

Apologies if this has been discussed before-- I searched and didn't come up anything that answers my specific question, which is: What is the lowest noise voltage reference practically/cheaply available? I'm looking to characterize the noise on a 24-bit ADC that I'm working on.

As I understand it, even batteries have different characteristics due to their chemistries which can introduce or pick up noise. I recall someone mentioning mercury batteries awhile ago, but I don't think noise was discussed. Am I overthinking it, or is there a preferred ultra low noise voltage reference out there that won't break the bank?

Hi

Lowest noise and cheapest are pretty much directly at odds with each other. Is $2,000 "cheap" for what you are doing? (it's dirt cheap by some standards).  Is noise well below the 24 bit level required? Is there a stability requirement as well as a noise target? Is this a lab bench measurement or are we building a few thousand copies a week?

Depending on the twists and turns, there are a lot of ways this could go.

Some math:

24 bits is 0.06 ppm. If you are after a lab setup you typically would want around 0.01 ppm of error contribution from your source.

If your ADC has a 0 to 2V input, that would get you to a 0.02 uV error budget. At that level you would need to include thermocouple offsets and other "interesting" things along with noise and drift.

A fairly good reference IC is going to be in the 2 ppm / C range. With a 0.06 budget, you will need to control your drafts and temperature changes to well below 0.1C. That gets into things like oil baths. Is this ok in your application?

Lots of zigs and zags.

This is about as close as you will get with a "cheap chip":

http://www.linear.com/product/LTC6655

It's still a few orders of magnitude away from your requirements.

Bob

[altaic]

Thanks for the response, Bob! This is a bench measurement so I can tweak my design. At this stage, I don't really care about drift and such, only noise in a DC measurement. Re. budget, $2000 would be pushing it currently; later on, assuming this project turns into a marketable product, that wouldn't be a problem.

The ADC in question, BTW, is the LTC2380-24 [http://www.linear.com/product/LTC2380-24]. Quite a nice beastie, which I hope to learn a lot about analog design with.

[uncle_bob]

Thanks for the response, Bob! This is a bench measurement so I can tweak my design. At this stage, I don't really care about drift and such, only noise in a DC measurement. Re. budget, $2000 would be pushing it currently; later on, assuming this project turns into a marketable product, that wouldn't be a problem.

The ADC in question, BTW, is the LTC2380-24 [http://www.linear.com/product/LTC2380-24]. Quite a nice beastie, which I hope to learn a lot about analog design with.

Hi

That's only about a 22 bit part. The last two bits are noise. At a reasonable rate, the last 5 bits are noise (see page 6 in the data sheet).

Bob

[barry14]

Linear Technology recommends their reference device,  LTC6655-5, be used as the reference for this converter.  It has a 2 ppm/C drift and they recommend using a capacitor to lower its noise level. The noise of this reference represents the lowest level that you can achieve. You should look at the specifications for this device.  LT16 bits tests the converter by applying a portion of the reference voltage to the input and then performing FFT's on the output.  The specified SNR, using this method, is about 100 dB (equivalent to about 16 bits).

[uncle_bob]

Linear Technology recommends their reference device,  LTC6655-5, be used as the reference for this converter.  It has a 2 ppm/C drift and they recommend using a capacitor to lower its noise level. The noise of this reference represents the lowest level that you can achieve. You should look at the specifications for this device.  LT16 bits tests the converter by applying a portion of the reference voltage to the input and then performing FFT's on the output.  The specified SNR, using this method, is about 100 dB (equivalent to about 16 bits).

Hi

If you need the link to this part, it's in my post back a few in this thread.

Bob

[altaic]

Thanks guys. I do realize it's ENOB is below 24 bits, and indeed I'm planning on using the LTC6655 voltage reference and the LTC6363 opamp to drive the ADC. I'm also planning on using the averaging filter for increased ENOB, since I don't always require fast sampling.

Once it's fabricated, I would like to characterize the ADC's performance with an ultra-low noise DC source, and experiment with the design to see how noise performance is affected. So, I thought batteries would be a good place to start, since they're cheap and, I imagine, pretty low noise. I'm wondering if anyone has characterized various types of batteries for noise or if there is an appreciable difference between different chemistries.

[retrolefty]

Thanks guys. I do realize it's ENOB is below 24 bits, and indeed I'm planning on using the LTC6655 voltage reference and the LTC6363 opamp to drive the ADC. I'm also planning on using the averaging filter for increased ENOB, since I don't always require fast sampling.

Once it's fabricated, I would like to characterize the ADC's performance with an ultra-low noise DC source, and experiment with the design to see how noise performance is affected. So, I thought batteries would be a good place to start, since they're cheap and, I imagine, pretty low noise. I'm wondering if anyone has characterized various types of batteries for noise or if there is an appreciable difference between different chemistries.

 That is a very good idea/project. With so many volt-nuts around here I'm surprised it hasn't happened already.

[ChunkyPastaSauce]

So, I thought batteries would be a good place to start, since they're cheap and, I imagine, pretty low noise. I'm wondering if anyone has characterized various types of batteries for noise or if there is an appreciable difference between different chemistries.

http://tf.nist.gov/general/pdf/1133.pdf "Measurement of voltage noise in chemical batteries"

[altaic]

http://tf.nist.gov/general/pdf/1133.pdf "Measurement of voltage noise in chemical batteries"

Yes, thank you! Surprising that Ni-Cd performed best.

[mojoe]

Very interesting paper. I wonder if NiMH would perform as well as NiCad. I don't know the internal resistance of NiMH.

Am I correct that the paper stated with NiCad, the noise did not increase while float charging?

[Fungus]

I'd have thought adding a decent capacitor across the output would do more to reduce noise than buying a more expensive reference.

[Kleinstein]

NiMH cells tend to have a higher internal resistance. Also the hydrite has lots of material internal stress and could thus cause noise. So I won't consider NiMH worth a try. Some cells are sensitive to mechanical stress from the outside, so there might be some very low frequency noise / signals.

If a LT6655 is not good enough, one could use the average of 2 or 4 of these to get lower noise. This also applies to other reference chips. So if temperature dirft is not critical, even a larger quantity of LM329 might be a way to go.

[ChunkyPastaSauce]

Also resistance isn't dependent only on chemistry - structure matters. As in the paper, simply selecting larger current capacity cells from the same chemistry significantly reduces noise due to lower cell resistance.

Temperature will also effect noise level, but not sure of the relationship... typically battery resistance increases with falling temperature, while many noise processes decrease with [falling] temp... not sure which wins.

[suicidaleggroll]

I'd have thought adding a decent capacitor across the output would do more to reduce noise than buying a more expensive reference.

Really high frequency noise, sure, but what about noise in the 1-100 Hz region?  A cap won't do anything for that, you'd need an RC combination, and to get the cutoff frequency that low you'd end up needing a series resistance so high it would make the output pretty much useless for driving an ADC reference line.

[Fungus]

I'd have thought adding a decent capacitor across the output would do more to reduce noise than buying a more expensive reference.

Really high frequency noise, sure, but what about noise in the 1-100 Hz region?

I think the voltage reference itself is supposed to deal with that (by adjusting its output).

[suicidaleggroll]

I'd have thought adding a decent capacitor across the output would do more to reduce noise than buying a more expensive reference.

Really high frequency noise, sure, but what about noise in the 1-100 Hz region?

I think the voltage reference itself is supposed to deal with that (by adjusting its output).

It is, and the better the reference, the better the job that it will do.

[splin]

Ni-cd batteries are almost certainly the way to go - relatively low cost and much lower noise than any solid state reference but you have to be careful of their temperature dependence when making low frequency measurements below 1 Hz. See reply #27 here:

https://www.eevblog.com/forum/projects/low-frequency-very-low-level-dc-biased-noise-measurements/25/

[splin]

Ni-cd batteries are almost certainly the way to go - relatively low cost and much lower noise than any solid state reference but you have to be careful of their temperature dependence when making low frequency measurements below 1 Hz. See reply #27 here:

https://www.eevblog.com/forum/projects/low-frequency-very-low-level-dc-biased-noise-measurements/25/

On reflection, the temperature sensitivity of the AA cells that I measured was very likely due to the change in thermal EMF at the negative given that the steel battery case is directly connected to the negative but not the positive terminal. Thus any change in the environment temperature will get reflected at the negative connection very quickly through the thermal conductivity of the battery sleeve, with the positive taking much longer to equalize. A well insulated box and lots of time to stabilize should be OK.

[uncle_bob]

http://tf.nist.gov/general/pdf/1133.pdf "Measurement of voltage noise in chemical batteries"

Yes, thank you! Surprising that Ni-Cd performed best.

Hi

No available battery will be stable enough / quiet enough to do what you are trying to do.

If you have a 6655 putting 5V into the ADC, use another one as the "calibration" device. Run it at 2.5V or whatever you decide to check things at. The ADC does not care which input the noise comes in on. With identical sources, the "real" result will be 0.707 times whatever you read.

If you want to step the voltage off as you test, you are going to need a pretty good bridge. That gets you right back up into the > $1K range on the surplus market.

Bob

[altaic]

Ni-cd batteries are almost certainly the way to go - relatively low cost and much lower noise than any solid state reference but you have to be careful of their temperature dependence when making low frequency measurements below 1 Hz. See reply #27 here:

https://www.eevblog.com/forum/projects/low-frequency-very-low-level-dc-biased-noise-measurements/25/

On reflection, the temperature sensitivity of the AA cells that I measured was very likely due to the change in thermal EMF at the negative given that the steel battery case is directly connected to the negative but not the positive terminal. Thus any change in the environment temperature will get reflected at the negative connection very quickly through the thermal conductivity of the battery sleeve, with the positive taking much longer to equalize. A well insulated box and lots of time to stabilize should be OK.

Fascinating thread; thanks for the pointer.

Hi

No available battery will be stable enough / quiet enough to do what you are trying to do.

If you have a 6655 putting 5V into the ADC, use another one as the "calibration" device. Run it at 2.5V or whatever you decide to check things at. The ADC does not care which input the noise comes in on. With identical sources, the "real" result will be 0.707 times whatever you read.

If you want to step the voltage off as you test, you are going to need a pretty good bridge. That gets you right back up into the > $1K range on the surplus market.

Bob

Ah, interesting. So using two 6655s to double the noise, and take the RMS (thus sqrt(2)/2)?

[uncle_bob]

Ni-cd batteries are almost certainly the way to go - relatively low cost and much lower noise than any solid state reference but you have to be careful of their temperature dependence when making low frequency measurements below 1 Hz. See reply #27 here:

https://www.eevblog.com/forum/projects/low-frequency-very-low-level-dc-biased-noise-measurements/25/

On reflection, the temperature sensitivity of the AA cells that I measured was very likely due to the change in thermal EMF at the negative given that the steel battery case is directly connected to the negative but not the positive terminal. Thus any change in the environment temperature will get reflected at the negative connection very quickly through the thermal conductivity of the battery sleeve, with the positive taking much longer to equalize. A well insulated box and lots of time to stabilize should be OK.

Fascinating thread; thanks for the pointer.

Hi

No available battery will be stable enough / quiet enough to do what you are trying to do.

If you have a 6655 putting 5V into the ADC, use another one as the "calibration" device. Run it at 2.5V or whatever you decide to check things at. The ADC does not care which input the noise comes in on. With identical sources, the "real" result will be 0.707 times whatever you read.

If you want to step the voltage off as you test, you are going to need a pretty good bridge. That gets you right back up into the > $1K range on the surplus market.

Bob

Ah, interesting. So using two 6655s to double the noise, and take the RMS (thus sqrt(2)/2)?

Hi

If you get to the point that they add as power, yes you get the 0.707. If you believe they add as voltage then it is 1/2.

If you have a "noiseless" bridge, and run it off the *same* 6655 as runs the ADC, you would not get either correction. In that case, what you read is what you get.

Bob

[splin]

http://tf.nist.gov/general/pdf/1133.pdf "Measurement of voltage noise in chemical batteries"

Yes, thank you! Surprising that Ni-Cd performed best.

Hi

No available battery will be stable enough / quiet enough to do what you are trying to do.

How on earth did you come to that conclusion? The NIST noise measurements of an AA Ni-Cd @1Hz were more than 2 orders of magnitude quieter than an LT6655 when normalised to 1.2V. The Ni-Cd noise measurements at 1 Hz must include all the drift, stability, thermal EMF etc. contributions.

Below 1 Hz the 6655 exhibits rising, 1/f, noise but the NIST measurements stop at 1 Hz unfortunately.  I don't have much doubt that 10 second measurements for 0.1 to 10 Hz noise characterisation would be significantly worse than at 1 Hz - especially if larger C or D cells are used with their greater thermal mass and significantly lower noise.

Perhaps you have some evidence to the contrary?

[uncle_bob]

http://tf.nist.gov/general/pdf/1133.pdf "Measurement of voltage noise in chemical batteries"

Yes, thank you! Surprising that Ni-Cd performed best.

Hi

No available battery will be stable enough / quiet enough to do what you are trying to do.

How on earth did you come to that conclusion? The NIST noise measurements of an AA Ni-Cd @1Hz were more than 2 orders of magnitude quieter than an LT6655 when normalised to 1.2V. The Ni-Cd noise measurements at 1 Hz must include all the drift, stability, thermal EMF etc. contributions.

Below 1 Hz the 6655 exhibits rising, 1/f, noise but the NIST measurements stop at 1 Hz unfortunately.  I don't have much doubt that 10 second measurements for 0.1 to 10 Hz noise characterisation would be significantly worse than at 1 Hz - especially if larger C or D cells are used with their greater thermal mass and significantly lower noise.

Perhaps you have some evidence to the contrary?

Hi

Having done the same testing they did in a real world temperature environment ....

Bob

[alanambrose]

I wonder if anyone has a sense of where LiPo's would fall on the NIST analysis? One theory seems to be that noise is a ~function of source resistance for batteries so LiPo should be up there?

A.