HPM7177 ADC from CERN

[Castorp]

AFAK some of the AZ OPs also have a little bit of 1/f noise. So it is not only the AD7177 and the resistors as possible sources. I would expect the foil resistors to be much better than -40dB noise index. I would be more surprised if the resistor contribution is significant at all.

Having more noise from the ADC and scaling means the noise contribution from the LTZ1000 reference is a little smaller. The overall noise has not changed, only a slight change in where the noise comes from. The reference still is the largest contribution.

For the really critical parts it would be more useful to have 2 completely separate units in parallel and not use one ADC with 2 x LTZ1000 for the reference. Looks like this is the plan anyway.

If it's 1/f noise (voltage/current) from the AZ op amps, I wouldn't expect it to scale like this with voltage. It would be the same at zero and elsewhere, right?

I also thought the NI should be much better. If LTC5400 has -55 dB and that's thin film, then Vishay are almost certainly under-specifying. However, now I have some doubts again.

You're right, it's still mostly the LTZ1000. That extra bit from the amps/resistors/ADC is less than 30% at 10 V.

And yes - the plan is to have 2 completely separate units. It's always like that. In normal operation the controller uses the average of the two, so we're always sqrt(2) better in terms of noise. This fact is not reflected in the specs, we keep it as a safety margin.

[Kleinstein]

The OP at the input would add noise independent of the voltage. The OP at the reference would add noise to the reference, and depending on where the 5 V is actually taken to the test input this may give voltage dependent noise.

For the resistor noise, i can imagine the measurement gets tricky much below -50 dB. At some point that can be just temperature variations that cause variation in the resistor ratios. Here the LT5400 can have a slight advantage because of very good thermal coupling and matching. in the extremes, combining resistors with positive and negative TC is not as good as one material with a low TC everywhere.

It is not such a surprise that the ADC itself has some extra 1/f noise also for the reference path. The internal switches also have some resistance and these can have excess noise (I expect the MOSFET ON resistance to have a rather poor noise index). I don't know the details of the switched capacitor SD converters, but in the continuous time version and multi-slope ADC with a high input voltage there is more uneven weight to the resistors if the voltage is large and thus more excess noise from the resistors gets visible. This is what I see in my version. A can imagine a similar effect with the switched capacitor version too.

[Castorp]

Well, at least I'm sure that these in situ measurements are not affected by temperature fluctuations, turbulent air flow or thermal EMFs. At the moment I really can't pinpoint the origin. I'll dig deeper if I get more time, even if it's just for academic reasons.

[Castorp]

I have a question - is anyone aware of a good and reliable reference on excess noise in bulk metal foil resistors?

The Vishay datasheets claim <-40 dB, here they claim -42 dB:
https://www.ieee.li/pdf/viewgraphs/ultra_high_precision_resistors.pdf

Frank Seifert's LIGO report is a great reference, but it doesn't have much on this type. There's a lot of speculation and hand waving around this whole thing. Not to mention audiophile statements like "this type of resistor sounds better than that type", as if we're comparing tomatoes on the market.

I can try to measure it myself, but at the moment I just can't free up enough time and the needed equipment for it.

On the other hand, I came up with a test that I can do with the HPM7177. By feeding a common-mode signal from the internal multiplexer I can exclude the ADC as a source of additional 1/f noise  -it would still measure near-zero differential voltage. At the same time, the four 20 K elements will be biased with with up to 6.4 V (for CM input of +10V). Three measurements at CM = 5V, 7.1 V and 10 V will be enough to fit the model. The problem is that the HPM7177 setup is also busy, running another long-term stability test

[MegaVolt]

is anyone aware of a good and reliable reference on excess noise in bulk metal foil resistors?.

[Castorp]

Thanks, I'm well aware of this reference. There's just one measurement of a 100 Ohm S102K resistor that shows well below -60 dB, but I'm not sure how representative that is for higher values (10 K and higher).

[Castorp]

OK, now I have the answer.

Today I had to stop the long-term test, so I did the measurement at VCM = 10 V. So far, with 10 FFT averages, it looks pretty much the same as with VCM = 0 V. So the extra noise comes from the ADC, and not from the resistors. If they contribute something, it's really negligible.

The voltages across the 20 K elements (R1-R3-R5-R7) and 3.5 K ones (R2-R4-R6-R8) in the R array has the following dependence on VCM (input):



As you can see, due to VOCM=2.5 V, all 8 elements see zero voltage when VCM (input) = 2.5 V. At VCM=+10V they see 3 times higher voltages than at 0V.

And here's what the noise spectra look like. For completeness, the spectrum with internal ADC short is added.
(Edit: replaced the 10V CM plot line (red) with one averaged over longer time)



There's no need to take further measurements to confirm it. Those Vishay resistors definitely have noise index of -50 dB or lower.

[TexasRanger]

Are there any 1/f data for LT2500-32?

Cheers, Gerhard

Here you go, the dB scale is however a bit off 

Getting this Plot with an actual differential fullscale input signal seems quite impossible to me.
Compared to the 2380-24 the 2500-32 has better DNL.

(Sorry for Gravedigging)
EDIT: There is a typo in the plot, Ref Level is 3.5Vrms/10Vpp

[oPossum]

Marco Reps built one...

[quarks]

Marco Reps built one...

very nice

[branadic]

Really nice to see Marcos attempts to this Cern project. Fortunately he shared his measurements on github, so I couldn't resist to use his INL data to check my linearization approach.
Would be nice to see, if this improves his measurements.

-branadic-

[guenthert]

Really nice to see Marcos attempts to this Cern project. Fortunately he shared his measurements on github, so I couldn't resist to use his INL data to check my linearization approach.
Would be nice to see, if this improves his measurements.

-branadic-

     Well, can the INL error assumed to be constant in time?  Isn't it rather the result of all the little errors which do shift a bit over time?

     Further, Marcos cheerfully neglected the INL of the source and assumed it to be perfectly linear.  The Fluke 5700 certainly has impressive specifications, but how well does that specific unit still perform?  Not sure, how strict one should or realistically can be here (what's the objective?), but J. Williams et. al. in AN86 went through great length in determining INL using a (self-calibrating) KVD and multiple 3458A.  I'd think pen-ultimately one would use hamon dividers for a few test points, if one doesn't have access to a Josephson junction ...

[branadic]

Hopefully everyone understood, that this approach is linearization to the source only, not to absolute linearity as most of us don't have access to PJVS or at least 3458A with linearity verified to PJVS. So I used what data are available. Please take the results with a grain of salt.
Once Marco showed up at PTB and got results on their PJVS and thus proper INL measurements we can adjust the coefficients or the error correction function I presented

-branadic-

[View[+]Finder]

"Marco Reps built one" is a bit of an understatement: he did it with "a sugar-powered pick and place machine," in his kitchen, with a chemical-assisted reflow crafted from food-service line parts. Freaking amazing! And it works.

branadic, thanks for posting your paper on error correction in INL measurement using fitted regression models. I need to conduct similar measurements on my KS34465 DMM with a transplanted reference from the 7.5 digit version. Also, I would like to experiment with the AD7177 Evaluation Kit as a less expensive way to 8.5 digits. Did you use the actual CERN-design boards in your analysis? Do you have any tips, photos of your 7177 build that you could share?

[Kleinstein]

The fluke 5700 definitely has some INL limitations around zero. The negative side is genrated by switching relays and these can add thermal EMF offsets. Also noise / drift during the offset cal can add to the jump at zero.

Another possible contribution to the simpe INL test with the 5700 can be reference drift, At the µV level the low frequency noise of the LTZ1000 reference in the meter and LTFLU in the 5700 can add a bit to the difference. So Ideally one would repeat the sweep of the test points a few times.

There are 2 spot tests for the INL, that can be done without special instruments:
1) the so called turn over test to check if a votlage is read the same with reversed input therminals. If done at different voltages, this can test the even powers of the INL curve. Due to the internal construnction of the ADC this tends to be quite good.

2) with a crude, but stable 1:1 divider (possibly with buffer) one could check the linearit at half the scale, by comparing the sum of the two halves to the direct sum. 

[Castorp]

@branadic, I'm also very much interested in this approach. Even if it's just for academic reasons (for now).

I can tell that the INL curve is stable in time. At least over one year, at least to the uncertainty limit of my measurement (multiple 3458As).

There are some newer integrated ADCs that are even more promising in terms of native INL. Check out the plots in the datasheet of AD7768 for example. Well, I already explained why INL is not the most important spec for us, so for now I won't be building any HPM7768s 

[chuckb]

@branadic, I'm also very much interested in this approach. Even if it's just for academic reasons (for now).

I can tell that the INL curve is stable in time. At least over one year, at least to the uncertainty limit of my measurement (multiple 3458As).

There are some newer integrated ADCs that are even more promising in terms of native INL. Check out the plots in the datasheet of AD7768 for example. Well, I already explained why INL is not the most important spec for us, so for now I won't be building any HPM7768s 

Do you know how well the INL tracks between assemblies? Could you make some INL improvement with hard coding a correction curve?

[doktor pyta]

It would be interesting to know how much of this INL is added in analog front end circuit.
This would require 3458a measuring -10V ...+10V steps from the calibrator and then measuring the same steps at AFE output.

Often in high end devices front end is being bootstrapped to minimize nonlinearity.

[Castorp]

Only the positive half of the INL is consistent between units. So yes, it's possible to linearize at least 0 to +10 V using some average curve. It would bring it from 0.6 to let's say 0.2 ppm.

As for the contribution of the frontend - I have strong reasons to believe it's significantly smaller than the ADC nonlinearity. But I admit I haven't tried to separate it.

My indirect evidence comes from many tests of the ADC in different modes - for instance by swapping the internal signals using the built-in mux. Just by swapping IN+ and IN- inside the ADC you get a different curve. Also if you use a quartz oscillator rather than the internal 16 MHz or an external CMOS oscillator. And most importantly - INL seems to be very sensitive to the RC networks on the inputs and the Vref pin. What you see in the latest schematics is the result of empirical optimization. Just to remind you - all internal buffers are disabled. That gives a little advantage in noise and big advantage in INL.

[branadic]

Updated the former document with a less saticfying solution for HPM1, which could still lead to good results. It has to be tested though.

@branadic, I'm also very much interested in this approach. Even if it's just for academic reasons (for now).

Due to a still pending paper I can't show any details at the moment, but the solution only.

-branadic-

[MK]

Hi Branadic,

Which document has been updated?

[Castorp]

Branadic, it's really not urgent. But it does look very promising. It would be nice to demonstrate it once I get an absolute INL measurement against 10 V PJAS.

I also have some interesting stuff to publish, pending for now. Hopefully later this year.

[coppercone2]

haha, I got a sample AD7177 (or maybe it was the other 32 bit ADC) back in like 2015 or 2016 and I was imagining making a pickup using it but I stopped because I could not find an application. It's good to see this thing being used for something, the applications were always quite nebulous. I thought something maybe with seismometers but I gave up

What exactly is this thing going to measure anyway?

Specifically I am interested in what transducer its connected to and how the extra resolution benefits research. I really wanted to do something with my old samples, maybe the explanation will inspire me. I just got a nanovolt amplifier and a transformer for it because I got a good deal, but the only thing that came to mind was ELF receiver, depending on the CMRR of the thing, but I am even at a loss for applications despite the fact that it is in a easy to use box. Current amplifiers are a little easier to stomach because you can do light with them.

Do you have maybe ideas for me for applications since you have a good feel for what these machines are useful for , is there anything to do with these low voltage levels and st abilities other then making benchtops jiggle less ? Not that stability is not interesting in its own right but I am not ultra fascinated by it like other things so this part remains at a low priority.

[Kleinstein]

A seismometer may be a possible use for a high resolution ADC. However this usually would not need that much of gain stability.

AFAIK the main target for the HPM7177 is measuring the current to some beam magnets at the accerator, so the beam can be stable at the same position and also fixed focussing. More stable magents should can allow better beam quality and possibly less loss.

The articles show DC current transformers as transducers. In this context I am wondering a little why they need a 10 V range: the shunts (on the secondary) should likely also work with a slightly lower voltage so that one might get away without the divider.

Chance are the meter could also be used in some other places, e.g. to replace a HP3458. It should be slightly lower noise and at the full 10 kHz speed could even be more linear. Especially at the higher speeds (using the lower input resistor) the linearity of the 3458 may not be that great.

I aggree that there are not that many uses for so much resolution.

Chip manufacturers could make use of such a meter to test high resolution DACs - maybe a little faster than the current setup. They may allready use a similar setup.

[coppercone2]

oh its for the focusing/bending coils, so thats kinda like a really really fine tuning CRO

seismometer is like the only cool application I ever got from people about it and it was only proposed like 'maybe those big guys do that'. I can maybe fit one magnet from a small accelerator in my garage, maybe.

not that I usually care about applications but the nanovolts are so much extra cost and work that its almost unappealing to make/work on.

i wish there was a application that can sweeten the deal.