7.xV to 10.00000V AutoCAL circuit

[e61_phil]

Hi,

I'm afraid there is quite a lot disscusion about 7V to 10.00000V in the LM399 and LTZ1000 Threads, but my "idea" is related to both references.

It seems to be quite hard to get a relieable and stable (long-term and tempco) amplifier to generate 10.00000V out of the ~7V from a LM399 or LTZ1000. VHP100 are quite expensive, hard to get and not adjustable. So even if you get these things you can't adjust the voltage to 10.00000V without extra effort.

I was thinking about something cheaper and much easier to get (in germany most things are available at Reichelt Elektronik).

What if one uses a LTC2400 ADC as "reference divider"? The tempco is quite small (0.02ppm/K) and I could imagine it is also long-term stable. My approach is to create an amplifier which is able to generate about 10V out of the 7V and fine tune this amplifier with a DAC. A rather cheap (or two) DAC is sufficient, because it sits in a closed loop and will only tune the last digits.

I would use a LTC1043 to divide the 10V output down to 5V. This 5V will be the digital/LTC2400 GND. The 10V output of the amplifier will be the reference voltage of the LTC2400 (which seems to be 5V for the LTC2400). Now, I'm able to measure the about 2V (7V-5V Ref) with the LTC2400 while the output voltage is tuned to 10.00000V by the DAC.

After this calibration it is only neccessary to tune the DAC after some drift back to the ADC value measured at the calibration. Therefore, even the INL error of the LTC2400 doesn't matter.

Furthermore, it is possible to start this AutoCAL procedure, tune the DAC and switch off the microcontroller, so there will be no swichting digital circuits until using the reference.

What do you think about it? Is it a old story or doesn't it work?

Philipp

[uncle_bob]

What if one uses a LTC2400 ADC as "reference divider"? The tempco is quite small (0.02ppm/K) and I could imagine it is also long-term stable. My approach is to create an amplifier which is able to generate about 10V out of the 7V and fine tune this amplifier with a DAC. A rather cheap (or two) DAC is sufficient, because it sits in a closed loop and will only tune the last digits.

I would use a LTC1043 to divide the 10V output down to 5V. This 5V will be the digital/LTC2400 GND. The 10V output of the amplifier will be the reference voltage of the LTC2400 (which seems to be 5V for the LTC2400). Now, I'm able to measure the about 2V (7V-5V Ref) with the LTC2400 while the output voltage is tuned to 10.00000V by the DAC.

After this calibration it is only neccessary to tune the DAC after some drift back to the ADC value measured at the calibration. Therefore, even the INL error of the LTC2400 doesn't matter.

Furthermore, it is possible to start this AutoCAL procedure, tune the DAC and switch off the microcontroller, so there will be no swichting digital circuits until using the reference.

What do you think about it? Is it a old story or doesn't it work?

Philipp

Hi

Floating the 2400 in "mid air" means that all your digital stuff will be sitting on your 5V "2400 ground" rail. You need a pretty good amp to do the buffering on the output of the 1043 and then you are right back "into a mess". Amps don't like to drive cap loads. Rails want to be highly bypassed to be quiet. You now have multiple loops and compensation around the amp.

Since the 2400 is a cap sampler, it pulls bursts of charge off the source. That's worked out in the app notes from the 2400 side of things. It's still a very "spiky" current flow. To keep your ref from seeing these ... more op amps.

The "big guys" decided long ago to simply take whatever the reference put out and use that as a standard. Checking a 8.39205228002894 V calibration reference is not any worse than a 10V ref with modern gear.


Bob

[Vgkid]

If you want some old time circuits look at the Systron Donner m106, or the newer m107.

[e61_phil]

Floating the 2400 in "mid air" means that all your digital stuff will be sitting on your 5V "2400 ground" rail. You need a pretty good amp to do the buffering on the output of the 1043 and then you are right back "into a mess". Amps don't like to drive cap loads. Rails want to be highly bypassed to be quiet. You now have multiple loops and compensation around the amp.

Since the 2400 is a cap sampler, it pulls bursts of charge off the source. That's worked out in the app notes from the 2400 side of things. It's still a very "spiky" current flow. To keep your ref from seeing these ... more op amps.

I thought about a buffer with an auto-zero opamp around to stabilize the 5V rail (LT1010 togehther with LTC2057 for example). Perhaps, the LTC2400 isn't the right ADC for this.

I think the nice thing which an approach like this is you can get rid of any switching noise during measurements. This ist not possible with PWM, so you have to put much more effort into this to get it very quiet. I can't believe that this will be less effort than maintaining a rough 5V for the ADC. But, perhaps I am on the wrong track.

[montemcguire]

The real trick with PWM is getting the minimum 0.1ppm adjustment resolution you need without a many-GHz PWM master clock.  I figured out how to do it after some time-- [can't say much more about it though, sorry]...

One way to extend PWM resolution is to use a high resolution PWM duration register and add high quality random dither to that 'real' duration value before truncating it to the actual low resolution duration value. Since we're concerned with essentially DC, the small but random variation of the dithered PWM edges is just a HF component that will disappear when lowpassed. But, the edge timing dither will allow a much higher effective resolution of the PWM.

One requirement is to generate some high quality random numbers with an average of zero. I'm doing this in an NXP MCU by using their 12 bit ADC to sample the Johnson noise of some high value resistors, amplified by a few chopper stabilized op amps (OPA333). The combination of NiCr resistors and the noisy chopper amp produces very close to perfectly random noise with no 1/f component. The bottom 8 bits are extracted from the 12 bit ADC and stuffed into a RAM circular buffer, to be clocked out as needed when each new PWM edge timing is generated. While this circuit and ADC generates low bandwidth random numbers (~40KB/s), if the circular RAM buffer is large enough, there should be no problems with periodicity, even though the buffer will be cycled through a few times as the random data is slowly being updated.

There are still some gotchas here - the DC error of the chopper amps (and the ADC) will pollute the random noise signal, slightly altering its average value, but some coupling caps can minimize that for all but the last amp in the noise generator chain. Further, this dither is scaled down somewhat before it is added to the high res register, reducing this DC error. Finally, when the bottom 8 bits are taken from the 12 bit ADC values, this randomizes the random signal's DC error - it is effectively modulated by the value of the 4 bits that are thrown away from each sample. As long as the analog noise is large enough to make those top 4 ADC bits move, but small enough to prevent the ADC from clipping with any regularity, any DC error (causing a non-zero µ in the random data) should disappear.

Not sure if this scheme is what you came up with (and there's no need to confirm or deny!) but I'm using this method to randomize and extend the resolution of a PWM application and it seems to work reasonably well at low cost with a modern MCU.

[Andreas]

Hello,

the LTC2400 is not bad.
The self calibration and thus the low drift are the advantages.
Also a low power consumption helps to keep switching noise away from the analog part.

On the other side this ADC is relative noisy (10 uVpp) so you will have to average many
measurements (around 1 minute) to get a stable result.

For the DAC you should think twice. A cheap 8 Bit DAC will not be able to compensate for
a 20 ppm temperature + around 20-50 ppm ageing drift with a 1 uV resolution.
And perhaps you even want to compensate the T.C. of the voltage divider by small steps of the DAC.

PWM is not the solution of all problems.
Ok you have no resistor T.C.
But the other challenges are charge injection, temperature dependant switching times, and temperature + voltage dependant switch impedance to name a few.

With best regards

Andreas

[e61_phil]

Hello,

the LTC2400 is not bad.
The self calibration and thus the low drift are the advantages.
Also a low power consumption helps to keep switching noise away from the analog part.

On the other side this ADC is relative noisy (10 uVpp) so you will have to average many
measurements (around 1 minute) to get a stable result.

For the DAC you should think twice. A cheap 8 Bit DAC will not be able to compensate for
a 20 ppm temperature + around 20-50 ppm ageing drift with a 1 uV resolution.
And perhaps you even want to compensate the T.C. of the voltage divider by small steps of the DAC.

PWM is not the solution of all problems.
Ok you have no resistor T.C.
But the other challenges are charge injection, temperature dependant switching times, and temperature + voltage dependant switch impedance to name a few.

With best regards

Andreas

I don't want to use 8bit DACs. I thought about a MCP4922 (2x 12bit  3ppm/K gain error) and use one for coarse and one for fine adjustment.

[uncle_bob]

Hello,

the LTC2400 is not bad.
The self calibration and thus the low drift are the advantages.
Also a low power consumption helps to keep switching noise away from the analog part.

On the other side this ADC is relative noisy (10 uVpp) so you will have to average many
measurements (around 1 minute) to get a stable result.

For the DAC you should think twice. A cheap 8 Bit DAC will not be able to compensate for
a 20 ppm temperature + around 20-50 ppm ageing drift with a 1 uV resolution.
And perhaps you even want to compensate the T.C. of the voltage divider by small steps of the DAC.

PWM is not the solution of all problems.
Ok you have no resistor T.C.
But the other challenges are charge injection, temperature dependant switching times, and temperature + voltage dependant switch impedance to name a few.

With best regards

Andreas

I don't want to use 8bit DACs. I thought about a MCP4922 (2x 12bit  3ppm/K gain error) and use one for coarse and one for fine adjustment.

Hi

As mentioned above, the "big guys" pretty much have gone over to PWM for this sort of thing. Once you start tossing op amps into the mix, some of the PWM stuff becomes less of an issue. Because it's a "well known technique" there are a lot of schematics and app notes to give you a trail of bread crumbs to follow.

A somewhat crude example:

Start off with "sort of 20V" derived from who knows where. Divide it by two (your 10 V) and by 4 (your 5V) and maybe by some other integers in the 2 to 16 range. Yes this sounds a bit stupid, but there are some advantages:

1) You can run things at reasonable speed (no 1 Hz lowpass filters).
2) You can have consistent switching in all your outputs (except for the obvious odd / even issue).
3) You can get useful voltages in most ranges without a lot of crazy effort.
4) Your lowpass filters are not very crazy to do.

So where is this going you wonder? (Yes, there is a very large  ex-bottle of wine somewhere around here .. hmm ....).

Part of your problem with PWM is working out what the TC of "this" compared to the TC of "that". Building lots of PWM's is dirt cheap. Feed them into a bank of 2400's and go to it. Average for tens of minutes if you need to. Feed back the results to a display. Let that tell you what you have on each of your outputs.

Ok so back to the top (remember the 20V starting point?).

Play with your same servo dac stuff to get "whatever" to produce 20V. Put that (after PWM) into one side of a 2400 (or multiple 2400's). Run the other side of the 2400 off of your magic super reference (after PWM). Your comparison is now good at the "who knows what" fraction of a ppm level. The real answer is going to depend on a bunch of averaging issues and firmware and those pesky op amps.

Bob

[e61_phil]

A PWM might be the way to go if you want a stable output for a very long time. But it isn't easy to build. My approch will get you only a new calibration from time to time for the amplifier. I think cheap precision resistors are good enough to create a stable divider for a few hours of measuring time. After this few hours you rerun this AutoCAL procedure. It is not meant to keep track on the tempco. It's only a cheap way to recalibrate the amplifier every time before you use it.

[uncle_bob]

A PWM might be the way to go if you want a stable output for a very long time. But it isn't easy to build. My approch will get you only a new calibration from time to time for the amplifier. I think cheap precision resistors are good enough to create a stable divider for a few hours of measuring time. After this few hours you rerun this AutoCAL procedure. It is not meant to keep track on the tempco. It's only a cheap way to recalibrate the amplifier every time before you use it.

Hi

I guess a lot depends on what you consider cheap and what you are after in terms of accurate. A fixed width PWM is dead simple to build.

Bob

[e61_phil]

Hi

I guess a lot depends on what you consider cheap and what you are after in terms of accurate. A fixed width PWM is dead simple to build.

Bob

A fixed PWM isn't tuneable to 10.00000V and a tuneable PWM with >20bit with at least 120dB filtering isn't that simple. Or I don't know how

[fmaimon]

Try using this: https://www.eevblog.com/forum/projects/general-purpose-power-supply-design-7488/msg105509/#msg105509

I've tested it (some results further down the thread above), and it works very well.

[e61_phil]

Try using this: https://www.eevblog.com/forum/projects/general-purpose-power-supply-design-7488/msg105509/#msg105509

I've tested it (some results further down the thread above), and it works very well.

I think it is next to impossible to achieve 120dB of filtering with some kind of PWM dithering. The dithering will produce low frequency stuff which is next to impossible to filter with a DC accuracy needed for these reference circuit.

The way I would go is something with combined PWMs like this 32bit PWM DAC from the EDN magazine or something like this. But that is all much more effort than tuning with an additional DAC which only correct for the last digits. Furthermore, for the DAC output no filtering is necessary.

[uncle_bob]

Hi

I guess a lot depends on what you consider cheap and what you are after in terms of accurate. A fixed width PWM is dead simple to build.

Bob

A fixed PWM isn't tuneable to 10.00000V and a tuneable PWM with >20bit with at least 120dB filtering isn't that simple. Or I don't know how

Hi

Ok, going back to my original post on how to do this:

You have a synthesized 20V it has no PWM on it's output so no 120 db questions.

You have a steering circuit for the 20V with a D/A just like you originally proposed, so no 120 db questions on that either.

So your output signal at 20V has no issues.

====

Now, you divide the 20 to 10, that's with a 2:1 PWM that likely is at a nice high frequency. You are doing "many samples" (as in minutes) on your ADC. The bandpass of the filter does *not* have to be very large.

Pick a frequency, let's call it 100 MHz to have something to work with. You want 120 db? One ideal pole will get it for you at 100 Hz. Two poles at 100 KHz. Both are a long way away from "many minutes". Neither one is a very complex filter. If you need more poles, there is *lots* of room for improvement. If you really are concerned, it's a fixed frequency, put in a tuned trap.

Yes indeed your grounding and board layout will have to be up to snuff to achieve good isolation. If you are after 120 db isolation from the "stuff" on the AC line, they will have to be pretty good for that as well. The audio crap on your AC is a *lot* closer to your intended output passband than *anything* coming from a PWM. Cascading high frequency filters is pretty trivial compared to isolating the 10, 15 and 20 (etc) line crud.

Again, the guys who make the high dollar calibrators seem to have no problems doing it this way. It's generally easier to copy what is known already to work than to go off and find a whole bunch of issues doing it some other way.

Bob

[e61_phil]

You have a synthesized 20V it has no PWM on it's output so no 120 db questions.

You have a steering circuit for the 20V with a D/A just like you originally proposed, so no 120 db questions on that either.

So your output signal at 20V has no issues.

Perhaps, I missed your point, but isn't that exactly what I want to do?

[Alex Nikitin]

What is wrong with the "standard " autocal approach? All we need is to keep the ratio between 7V and 10V constant and correct. A hi-res delta-sigma ADC, like LTC2400, is a good device to measure that ratio and it is fairly easy to (mostly) remove other influences but the ADC noise and long-term gain drift. Measure the 7V and 10V with the same 2:1 divider and the same reference voltage (for simplicity, 10V output divided by 2 by the second divider, as the LT5400 has two pairs of resistors) and, as long as the drift and noise of that divider and the reference is low enough for the duration of both sequential measurements, the result is accurate. The rest is pretty straightforward and a 12 bit multiplying DAC should be enough to correct the main 7 to 10V amplifier gain drift with temperature. And the ADC can sit on the ground rail, no need to float it.

Cheers

Alex

[uncle_bob]

You have a synthesized 20V it has no PWM on it's output so no 120 db questions.

You have a steering circuit for the 20V with a D/A just like you originally proposed, so no 120 db questions on that either.

So your output signal at 20V has no issues.

Perhaps, I missed your point, but isn't that exactly what I want to do?

Hi

In your approach, the ADC floats between 5 and 10V. In a more conventional approach, there is one ground and everything is referenced to it. In the conventional approach, PWM's are easy to build high accuracy dividers to get it all worked out.

Bob

[Alex Nikitin]

You have a synthesized 20V it has no PWM on it's output so no 120 db questions.

You have a steering circuit for the 20V with a D/A just like you originally proposed, so no 120 db questions on that either.

So your output signal at 20V has no issues.

Perhaps, I missed your point, but isn't that exactly what I want to do?

Hi

In your approach, the ADC floats between 5 and 10V. In a more conventional approach, there is one ground and everything is referenced to it. In the conventional approach, PWM's are easy to build high accuracy dividers to get it all worked out.

Bob

In the conventional auto-calibration approach, if your ADC is stable enough at ratio measurements, you don't need high accuracy dividers. You just need dividers with a good short-term stability and decent WW resistors should be good enough for that  .

Cheers

Alex

[uncle_bob]

You have a synthesized 20V it has no PWM on it's output so no 120 db questions.

You have a steering circuit for the 20V with a D/A just like you originally proposed, so no 120 db questions on that either.

So your output signal at 20V has no issues.

Perhaps, I missed your point, but isn't that exactly what I want to do?

Hi

In your approach, the ADC floats between 5 and 10V. In a more conventional approach, there is one ground and everything is referenced to it. In the conventional approach, PWM's are easy to build high accuracy dividers to get it all worked out.

Bob

In the conventional auto-calibration approach, if your ADC is stable enough at ratio measurements, you don't need high accuracy dividers. You just need dividers with a good short-term stability and decent WW resistors should be good enough for that  .

Cheers

Alex

Hi

If the OP had an ADC with a 10V input range, this all would be even simpler still.

Bob

[Alex Nikitin]

Hi

If the OP had an ADC with a 10V input range, this all would be even simpler still.

Bob

Sure, however there isn't one available and suitable for this application that I know of  . Two WW resistor dividers and a CMOS switch is not too complicated.

Cheers

Alex

[e61_phil]

Hi

In your approach, the ADC floats between 5 and 10V. In a more conventional approach, there is one ground and everything is referenced to it. In the conventional approach, PWM's are easy to build high accuracy dividers to get it all worked out.

Bob

Hi,

I think I'm still missing your point. In my approach there is no need for any PWM. The only stable divider by 2 is realized with a LTC1043 (and even this divider has to stable only. It is not necessary to divide by 2.00000). It is also possible to use two dividers (both in one LTC1043) as mentioned by Alex Nikitin and divide all voltages down and measure all against one common GND (one will lose ~1bit of resolution by doing this, but it shouldn't matter).
I'm still thinking to use a floating GND for the LTC2400 is much more easy to handle in PCB layout and buffering than handling with tenth of MHz PWM frequency. Furthermore, there is no fixed PWM needed.

I'm only interested in how "the guys who make the high dollar calibrators" create their stable 20.00000V which is used by the PWM to create all other voltages. Is this done with DACs only (and precision resistors) or is there any PWM involved, too?

Philipp

[Alex Nikitin]

The only stable divider by 2 is realized with a LTC1043 (and even this divider has to stable only. It is not necessary to divide by 2.00000). It is also possible to use two dividers (both in one LTC1043) as mentioned by Alex Nikitin and divide all voltages down and measure all against one common GND (one will lose ~1bit of resolution by doing this, but it shouldn't matter).

Hi Philipp,

In your configuration the divider quality (tempco and long-term drift) is still important. If you only measure the voltage ratio using the same divider for both voltages, only a very short term divider stability is required, so it is one uncertainty less. Even LT5400 has some drift    .

Cheers

Alex

[e61_phil]

Which divider do you mean? I don't want to use LT5400.

[Alex Nikitin]

Which divider do you mean? I don't want to use LT5400.

You can use just two plain metal film 5-10ppm/C resistors in this case, as long as the divider ratio is stable for a very short time required for every autocal cycle.

Cheers

Alex

[e61_phil]

I think I will need a stable divider by 2 and the stable LTC2400 for long term stability. Otherwise you will need some switching between different divider, I think.