HPM7177 ADC from CERN

[SigurdR]

Indeed, the 7177-2 is rather low noise.
Table 6 and 7 in the DS show that noise at 5 SPS is about 0,05uVrms (BW not stated).
"The numbers given are for the bipolar input range with
an external 5 V reference"...what ref used is not stated. LTC6655-5 maybe.

The noise from the bare eval kit without a lower noise reference can be quite high because of the reference noise.
Chances are the sampling rate is still at 10 kSPS. This also means there can be some mains hum included. To get comparable with a DMM one would need to average some 200-5000 readings, so expect the noise to go down by a factor of 10-50 just from averaging - though this does not fully apply to the reference noise, that can have quite some 1/f part.
By itself the AD7177 is rather low noise.

[SigurdR]

Thanks for the reply!

Filtered LTZ1000A is what I will use.
I have only need for 1 SPS or even lower.
My plan is for a DVM for DC.
For AC an audio ADC from ESS is my choice.

The evaluation board comes with an ADR445. It's not too bad, but it's not LTZ1000. And it's well above the ADC noise (even LTZ1000 is).

For long-term measurements you probably can't use 10 kSPS with the evaluation board software. And you can't go slower than 20 Sa/s either. If power line interference is a problem, you can use the built-in 50/60 Hz notch filters.

[Castorp]

I had remembered incorrectly - indeed, the lowest data rate is 5 Sa/s, not 20.

You can calculate the effective bandwidth of the sinc3 or sinc5 filter. They are fairly steep, as the first notch is at the output data rate. The equivalent bandwidth is roughly 1/3 of that, so something like 1.7 Hz at 5 Sa/s.

Filtering the LTZ1000 won't give you much. You can't filter its low-frequency noise efficiently, and the broadband noise wouldn't matter in this case.

[SigurdR]

I have been looking at AD converters trying to find the best alternative for a 8.5 - 9.5 digit DVM. Seems like there is not much development the last 10 years regarding slow, precision ADC:s. I guess there is not much sales for the companies designing them...
Fast ADC:s are plenty.

The AD7177-2 is about 10 years old already.

LT has the LTC2500-32 SAR which might be a candidate.

Is the Ad7177-2 still the best ADC for a DVM ?

[Kleinstein]

Nearly all higher end DVM use seprately build ADCs, not ready made ADC chips. The HPM7177 is the exception in this direction. The next best DMM build around an ADC chip may be the Sigilent SDM3068.

There are mainly 2 difficulties using a SD ADC chip in a high resolution meter:
1) the reference and range of the ADC is usually 5 V or less. This does not combine easy with a 7 V reference. The extra divider for the reference is a point for additional drift.

2) The linearity of most ADC chips is not that great. Low noise does not help that much when the linearity is limited. The LTC2500 and related ones are a relatively new exception in this respect and also the AD7177 with reduced range seems to be relatively good.  The linearity specs are usually for a differential drive with suppresed common mode signal. This needs extra effort for the input stage.
The input stage is at least different from the usual single ended design used in classical DMMs.

Because of the limited linearity and use of a usually not as stable 2.5-5 V reference the ADC chips are normally considered more suitable for the 5.5 to maybe lower end 6 digit meter range.
Noise is only one parameter for a DMM and with modern part it is more like the easiest. The other parametes are linearity and stability against drift (with time and temperature).
The SD ADC are pretty good in the 5.5 digit range. The 6-7 digit range is still mainly multislope or similar ADC build around a FPGA and multiple analog chips. These can work well from a 7 V reference and the native range usually includes 7 V.

In my view the best candidates for high performance from a ADC chip are AD7177, ADS1281 and LTC2378 and related (inlcuding LTC2500-32). New chips come up quite often.
Even getting the nominal INL may not be easy, as the layout and reference filtering / buffer can effect the linearity. INL may be the more important parameter, not noise.
Some of the ADC chips are pretty good when it comes to high speed. This is an area that is difficult with a classical multi-slope ADC.

There is very limited need for even 8 digit resolution. Not many reference are good for this level and for a voltage measurement one has the noise of the reference and the intrinsic noise if the signal source. Unless one is measuring a high end reference the DUT may very well swamp the noise.

[Castorp]

Well, HPM7177 (and the other digitizers we build and maintain) was never meant to be a general-purpose DVM. It has features specifically tailored to our needs, while completely lacking others. I thought that was clear from the start, but I keep getting questions that suggest otherwise 

High-resolution ADCs and high-end voltage references don't automatically click together. It's true that high-resolution Sigma-Delta and SAR ADCs progress slower than other ADC types, but they progress nevertheless. Check out AD7768. I think in the coming years we may see more parts with lower INL and equally good noise performance. I doubt there would ever be one that takes 7 V reference though.

[SigurdR]

I hope I did not say that the HPM7177 would be a general purpose DVM. If so, that was not my intention. It is a grand piece of work for your specific needs!

CERN has always been a remarkable place just to be interested in what goes on there. I am a physisist originally but ended up in electronics and management. I still follow what goes on in particle physics.

Thanks for the tip about the AD7768 - i will study it.

Well, HPM7177 (and the other digitizers we build and maintain) was never meant to be a general-purpose DVM. It has features specifically tailored to our needs, while completely lacking others. I thought that was clear from the start, but I keep getting questions that suggest otherwise 

High-resolution ADCs and high-end voltage references don't automatically click together. It's true that high-resolution Sigma-Delta and SAR ADCs progress slower than other ADC types, but they progress nevertheless. Check out AD7768. I think in the coming years we may see more parts with lower INL and equally good noise performance. I doubt there would ever be one that takes 7 V reference though.

[Castorp]

@SigurdR, I wasn't referring to your questions or the discussion going on in this thread. There are plenty of people who do their homework and ask clever questions, either here or through other channels. And then there are others who ask "Why do you make a 8.5-digit DVM and you don't even put a display on it?" 

[SigurdR]

Thank you, Kleinstein, for your informative reply!

This makes me wonder how many people, or man/woman hours, are put into a project like designing for manufacturing, an 8.5 digit DVM...

Nearly all higher end DVM use seprately build ADCs, not ready made ADC chips. The HPM7177 is the exception in this direction. The next best DMM build around an ADC chip may be the Sigilent SDM3068.

There are mainly 2 difficulties using a SD ADC chip in a high resolution meter:
1) the reference and range of the ADC is usually 5 V or less. This does not combine easy with a 7 V reference. The extra divider for the reference is a point for additional drift.

2) The linearity of most ADC chips is not that great. Low noise does not help that much when the linearity is limited. The LTC2500 and related ones are a relatively new exception in this respect and also the AD7177 with reduced range seems to be relatively good.  The linearity specs are usually for a differential drive with suppresed common mode signal. This needs extra effort for the input stage.
The input stage is at least different from the usual single ended design used in classical DMMs.

Because of the limited linearity and use of a usually not as stable 2.5-5 V reference the ADC chips are normally considered more suitable for the 5.5 to maybe lower end 6 digit meter range.
Noise is only one parameter for a DMM and with modern part it is more like the easiest. The other parametes are linearity and stability against drift (with time and temperature).
The SD ADC are pretty good in the 5.5 digit range. The 6-7 digit range is still mainly multislope or similar ADC build around a FPGA and multiple analog chips. These can work well from a 7 V reference and the native range usually includes 7 V.

In my view the best candidates for high performance from a ADC chip are AD7177, ADS1281 and LTC2378 and related (inlcuding LTC2500-32). New chips come up quite often.
Even getting the nominal INL may not be easy, as the layout and reference filtering / buffer can effect the linearity. INL may be the more important parameter, not noise.
Some of the ADC chips are pretty good when it comes to high speed. This is an area that is difficult with a classical multi-slope ADC.

There is very limited need for even 8 digit resolution. Not many reference are good for this level and for a voltage measurement one has the noise of the reference and the intrinsic noise if the signal source. Unless one is measuring a high end reference the DUT may very well swamp the noise.

[SigurdR]

After having read about lowering noise from the LTC6655 and LTC6655LN
https://www.analog.com/en/analog-dialogue/articles/why-does-voltage-reference-noise-matter.html
and here
https://www.analog.com/en/technical-articles/reference-filter-increases-32-bit-adc-snr-by-6db.html
for use with the LTC2508-32 to increase its SNR,
I wonder if that method would also work for the LTZ1000, or what is the main issue to why it is difficult to lower its noise efficiently?
(my being a 20-20kHz low noise audio person)


Kind regards,
Sigurd

Filtering the LTZ1000 won't give you much. You can't filter its low-frequency noise efficiently, and the broadband noise wouldn't matter in this case.

[Kleinstein]

For most electronic designs one would not start from zero.  A moderm higher resolution DVM is kind of modular with parts that are relatively independent: display and interface part, power supply, ADC, reference,  input protection and input switching and amplification.

A first design on paper can be rather fast, the tricky parts comes with refining the actual performance, so that it works like planed or at least close too. How much is needed here is hard to predict: one can be lucky and tricky parts work from the start. However there is also a chance to find nasty surprises, like unexpected ringing, resonances or interactions.  A difficulty is testing when at the cutting edge.
Finding and fixing such quirks is what can take a lot of time, but it is hard to predict. If lucky, it may be a few weeks and it works, but there is also a chance for failure, so that after 2 years one may have no useful product and just knowlege on how it does not work. One still gains knowledge and the experience of the team can make a big difference. Pushing the envelope at the cutting edge is a difficult thing and not very predictable: it sometimes works nice and than new road blocks come up. A the high end failure is always an option.

I spend quite some time for my vesion of a high resolution (now 8 digit range - though the original target was more like cheap 6 digits) MS ADC - the initial version was ready surprisingly fast (~ 20 hours for a version on a bread board). Then there were months (and many hours spend) with no real succes in identifying quirks and improving things that are irrelevant. The closer it gets to perfection the longer it takes to solve the small remaining weaknesses. The ADC part is likely the most tricky one - others like the supply and switching is quite a bit easier. It is still the less intersting part for a hobby project.  For a hobby it is not so much about getting the result, but also have the challange - so there is no strict target performance to be met, but also hard to find a good enough.

AFAIK it took several years to just redesign the boards for the HP3458 to replace obsolete parts.  This was a simpler thing with a known working solution.
The software part can also take quite some time. This is more predictable - though a similar problem with bug fixing applies there too.

For the reference filtering, there can be some improvement be gained from filtering the higher frequencies where it is easy. There can be kind of an analog to aliasing also at the reference side. This would also apply to the LTZ1000.  The SNR tests with the LTC2508 were more at higher frequencies and the LTC6655 has a relatively low 1/f cross over. For a DVM the more nasty noise is the really low frequency part and there filtering is not practical.
In my ADC design I also have some similar reference filtering for a LM399 ref. It helps for the more white noise  - the main effect is from the 25 Hz and  5-100 kHz range.  It still does not help with low frequency noise that is the limiting part of the reference.

[View[+]Finder]

Time for a little update on the "fan-boy, making-do" version of the CERN precision current fluctuation measurement device.

The attached photo shows the AD7177 EVAL board with a piggy-back AD8475 EVALUATION BOARD attenuator. It functions to buffer and reduce the input voltage to a level acceptable to the AD7177, i.e. from ~10VDC to less than 4VDC. The AD8475 is powered by 3.3V from the black micro-controller board.

Having spent most of the last month testing the AD7177 in a variety of configurations, I am happy to report that my investment in the experiment was well worth the time and money. The software (FREE) from AD is both easy to use and sufficiently sophisticated to explore most of the issues raised in this thread. More data will be posted  as it becomes available . . .

[View[+]Finder]

Another test of AD7177 setup . . .

[View[+]Finder]

Some more with different settings . . .

[View[+]Finder]

One problem I have encountered is that my AD7177 setup (using what I have determined to be a 'precision' voltage source) places a burden on the source on the order of several millivolts. This might not be a problem for CERN in their application, however it poses a problem for anyone who might want to use it as a precision meter on a reference voltage like the HP3459 A9 board for example. As @Kleinstein noted, there is no easy path to precision measurement.

[Kleinstein]

Discussing the AD7177 evalutation board goes a bit off topic for this thread. If there is more to come in this dirction it may be better to start an own thread for this.

The higher speed very low noise SD ADCs seem to be quite demanding on the drivers, both for the inputs and also for the reference. It is not only the simple DC load current, but the more tricky part seems to be the higher frequency (e.g. 10 MHz)  part with fast transients. So the ADC chip is not that easy to use. A 7 V ref adds a little (but not not that much) to this.

If one does not have a very low noise reference, a first point to look is at the noise with a zero input voltage. Also the external reference used to generate the 3.x V test voltage would add to the overall noise.

[Castorp]

One problem I have encountered is that my AD7177 setup (using what I have determined to be a 'precision' voltage source) places a burden on the source on the order of several millivolts. This might not be a problem for CERN in their application, however it poses a problem for anyone who might want to use it as a precision meter on a reference voltage like the HP3459 A9 board for example. As @Kleinstein noted, there is no easy path to precision measurement.

I suspect in this case the culprit is the AD8475. It's a reasonably good FDA with integrated resistors but it doesn't have buffered inputs.

Unless you want to squeeze the lowest LF noise and INL from the AD7177-2, it makes perfect sense to use its built-in buffers. They make life much easier when it comes to driving the inputs and Vref pins.

[View[+]Finder]

Discussing the AD7177 evalutation board goes a bit off topic for this thread. If there is more to come in this direction it may be better to start an own thread for this.

The title of the thread is "HPM7177 ADC from CERN" so, considering the EEVBLOG forum as a place for interested amateurs of various levels of skill to exchange information, and noting MarcoReps success to DYI the HPM7177, I offered up a lower-cost alternative for other readers to try out. From my experience thus far, it has been an excellent way to learn about ADC and the difficulties of precision measurement.

My objective is to participate (as a mere observer) in the profound effort and accomplishment of CERN scientists and engineers and with the luminance project in particular in mind. I believe my postings are consistent with the public engagement mission evident in the CERN videos on YouTube that I have seen and the cost of my bit of kit is within the budget of, well anyone.

What I envisage as my future posts remain within the scope of the CERN HPM7177: the use of an FPGA (or initially a Raspberry Pi) to control the ADC--in principle the same as the HPM7177--and a low-budget alternative to the LabVIEW PXI rack that CERN will use for data collection. This could be a science project for kids . . . of any age.

[Castorp]

On a similar note, people have been asking me whether it's possible to build a cheaper, lower-grade version of HPM7177. I believe it's possible to reduce the BOM cost by a few hundred euros and keep a similar level of performance. It wouldn't be the same of course, but I think it's possible to have the same noise, linearity and perhaps 2-3 times worse temperature drift (which still means <=0.1 ppm/degree C) with certain thin-film resistor arrays. This approach would open up another possible path - to use a higher temperature set point and replace the Peltier element with a heater.

I think replacing the LTZ1000 with another voltage reference wouldn't make sense. Then you may as well replace the ADC, and also get rid of the complex frontend and some other parts. At CERN we already have a solution like that with LTC2378-20, LT1236, THS4531-based driver and SMN resistor networks. The price is an order of magnitude lower, and so is the performance. It's a tried and proven "Swiss army knife" solution, with hundreds of units used throughout the entire accelerator complex (not just LHC).

[niner_007]

After having read about lowering noise from the LTC6655 and LTC6655LN
https://www.analog.com/en/analog-dialogue/articles/why-does-voltage-reference-noise-matter.html
and here
https://www.analog.com/en/technical-articles/reference-filter-increases-32-bit-adc-snr-by-6db.html
for use with the LTC2508-32 to increase its SNR,
I wonder if that method would also work for the LTZ1000, or what is the main issue to why it is difficult to lower its noise efficiently?
(my being a 20-20kHz low noise audio person)


Kind regards,
Sigurd

Filtering the LTZ1000 won't give you much. You can't filter its low-frequency noise efficiently, and the broadband noise wouldn't matter in this case.

If you want the reduce the noise of an LTZ1000, you can parallel multiple LTZ1000 as described in the data sheet. That’s the only way you can possibly reduce 1/f noise that I’m aware of. Diminished returns though. You might have better luck with a carefully selected zener diode in a heated oven. If low noise and not long term stability is what you need, an LTC6655 in a thermally stabilized chamber could solve your temperature stability and noise.

[branadic]

If low noise and not long term stability is what you need, an LTC6655 in a thermally stabilized chamber could solve your temperature stability and noise.

That is a misconception and here is why: Datasheet states "0.25ppmP-P (0.1Hz to 10Hz) 625nVP-P for the LTC6655-2.5"
If you multiply that noise figure by 2.88 (7.2V/2.5V) you end up with 1.8µVpp @ 7.2V, which is larger than the 1.2μVpp noise for the LTZ @5mA and typ. 7.2V.

-branadic-

[branadic]

If you refer to the typ. 0.12ppmpp for the LN version, this value is also misleading if you read it out of context, as the LN version comes in a MSOP plastic package. So you might have lower pink or 1/f noise, but have to deal with mechanical strain and humidity influence and suddenly all your low noise advantage is gone forever.

-branadic-

[Kleinstein]

A single LTC6655 is higher noise than a LTZ1000. However with the LTC6655 it is very feasable to combine 2 references (e.g. 2 x 2.5 V in series) and than the 1/f noise gets comparable to the LTZ1000.  The LTC6655 may even get away without an extra reference buffer, as it has quite some capacitive drive capability. The LTZ1000+buffer circuit can include some filtering for the higher frequency range. Not sure how much the AD7177 reacts to this higher frequency noise - it may or may not be a factor.
0.1-10 Hz is the standard frequency range used in datasheets, because this is about the practical limit for a measurement with capacitive coupling. Real world may need performance at even lower frequencies, that may be worse with the LTC6655 because of thermal effects.

[SigurdR]

CERN must be a gold mine for us electronic engineers! I am realy glad that CERN is making the 8,5 digit DVM open to the public!
Is the LTC2378-20, LT1236, THS4531-based driver and SMN resistor networks also open to the public?

I think replacing the LTZ1000 with another voltage reference wouldn't make sense. Then you may as well replace the ADC, and also get rid of the complex frontend and some other parts. At CERN we already have a solution like that with LTC2378-20, LT1236, THS4531-based driver and SMN resistor networks. The price is an order of magnitude lower, and so is the performance. It's a tried and proven "Swiss army knife" solution, with hundreds of units used throughout the entire accelerator complex (not just LHC).

[Castorp]

Is the LTC2378-20, LT1236, THS4531-based driver and SMN resistor networks also open to the public?

Unfortunately it isn't. But you can see a block diagram here:
https://kt.cern/sites/knowledgetransfer.web.cern.ch/files/technology/magnet-power-supplies/tech-brief/magnet-power-supplies-posterpdfpdf.pdf

And a photo of the newer version on slide 4 here:
https://indico.cern.ch/event/976008/contributions/4117627/attachments/2160132/3644408/HL-LHC_ADC_beev.pdf

The SMNs are on the bottom side of the board under IC4, IC8, etc. The voltage reference is on the bottom right corner, on the island with L-cutout, surrounded by resistor networks used as heaters. There's a small plastic enclosure that goes over this island.