An arbitrary output voltage divider for a precision voltage reference.

[rhb]

Once one has a precision voltage reference one still has the problem of deriving other voltages from it.  The following idea came to me in replying to a new thread on OSHW.

Consider a precision voltage reference with an ADC and a DAC connected to an MCU with switchable  R and C values.   Set  the desired output voltage using the ADC and a timer referenced to a GPSDO.  Use the ADC value to address a lookup table that feeds the DAC.  The curve shape is known analytically.   By switching the R & C values  and measuring the change in time constant against a GPSDO one can correct for all the errors including the ADC and DAC.  Aside from the voltage reference it does not require precision parts. A sparse, minimum summed absolute error  L1 pursuit will handle the divider calibration process very well.  Much better than a traditional least squared error L2 solution. 

The central idea is that precise time is the cheapest precision available and CPU time is essentially free.  I investigated the mathematics of correcting the ADC and RC errors some time ago.  At the moment I'm not quite sure how to do the DAC errors.  Probably a second ADC will do the trick, though it might require another  DAC as well to setup sufficient equations.

The available 24 bit audio ADC and DAC chips should work well for a DC reference.  Does anyone know of prior work on the concept?  24 bits should give 1 ppm or better resolution and accuracy after calibration.

[cellularmitosis]

I didn’t realize the cheap audio DACs could produce DC.

Also, where are you correcting for temperature?

You should check out Linear AN-86, great read!

[ManateeMafia]

The first time I read about an audio DAC being used in test equipment was Brooks Shera's GPSDO.  http://www.rt66.com/~shera/QST_GPS.pdf

[rhb]

Why wouldn't a DAC produce DC?  Most of them are running on a single rail supply.  They only produce AC because they are changing value and there is a blocking capacitor.  In this application they are not changing the output  unless the user has selected a different output voltage or an error needs to be corrected.

This is not a design yet, just a concept.  Precision voltage dividers are difficult and/or  expensive. 

Temperature compensation would have to be handled by additional lookup tables and temperature sensors.   A temperature controlled bath of mineral oil would probably be needed to maintain uniform temperatures.  Temperature compensation doesn't change the fundamental concept.  It just adds some more terms to the equations that have to be solved and adds more lookup tables for correction factors.  It may well be that temperature effects limit the accuracy and 1 ppm is not realistic.

[MisterDiodes]

Audio ADCs/ DAC's are not really precision devices at DC, and are not designed for tight TC or precision offset / gain / INL errors, etc.  That's why we have precision ADCs  & DAC's that are designed for DC & very low freq applications.  Especially if you're after higher accuracy.   

The idea of trying to measure RC charge / discharge curve timing isn't new, and usually where it becomes a real challenge is even with proper ADC and DAC's: your R and C keep drifting and changing over time and temp.  It can certainly be done, but sometimes it can be like measuring a bowl of Jello if you don't have really, really good parts for your "C" especially.

 

[Kleinstein]

Audio ADCs and DACs usually have quite some drift. In addition quite some audio chips include a high pass filter (analog or digital) to remove DC errors - thus no DC out. There are a few chips that allow DC operation, but performance is usually poor.

In addition to audio and precision grades there are also ADCs intended for AC metering purpose, these are still cheap but often have better gain stability than the audio ADCs. However they may not be that good with linearity.

The R + C route is not helping much unless you get perfect parts (R, C, switches and buffer). Normally this is called PWM DAC - it can be an option if done right, but it's not that easy and still has limits.

[cellularmitosis]

Normally this is called PWM DAC - it can be an option if done right, but it's not that easy and still has limits.

rhb, try this google query to follow up on what Kleinstein is talking about: https://www.google.com/search?q=site%3Aeevblog.com+datron+4910+pwm

[rhb]

The R + C route is not helping much unless you get perfect parts (R, C, switches and buffer). Normally this is called PWM DAC - it can be an option if done right, but it's not that easy and still has limits.

I don't see why the parts need to be perfect.  Why is stability not sufficient?

However, I conflated the notion of using the RC curve to quantify the non-linearity with generating the output voltage.  What I was looking for was any way to get around building a precision resistive divider.

I found a  good article on building a 1 ppm voltage reference using a 10 V reference and an AD5791  in the  Analog Dialogue 44-04, April (2010),  "The 20-Bit DAC Is the Easiest Part of a 1-ppm-Accurate  Precision Voltage Source" By Maurice Egan.  It looks as if the BOM would run around $100 which seems perfectly acceptable for such a device.  Does anyone know of board Gerber files for that design?

The TI DAC1282 has a DC mode and is about 1/2 the price of the AD part, but the AD part looks more appropriate for  the task.

[chuckb]

You can purchase an already build development PCB at Digikey ($150). The chip is already on the PCB, you add a Power supply and an external voltage reference. There are also provisions on the PCB to add an LTZ1000.

EVAL-AD5791SDZ Guide
https://www.digikey.com/products/en/development-boards-kits-programmers/evaluation-boards-digital-to-analog-converters-dacs/793?k=ad5791

[cellularmitosis]

Watch out — there are two grades to that part, A and B grade.  The B grade is the one which looks like it actually does 1ppm, and is something like $130 $107 per chip at Digikey.   Combine that with an LTZ and you’d have a nicely upgraded version of Ian Johnson’s LM399+AD5781 project!

[thermistor-guy]

The R + C route is not helping much unless you get perfect parts (R, C, switches and buffer). Normally this is called PWM DAC - it can be an option if done right, but it's not that easy and still has limits.

I don't see why the parts need to be perfect.  Why is stability not sufficient?

However, I conflated the notion of using the RC curve to quantify the non-linearity with generating the output voltage.  What I was looking for was any way to get around building a precision resistive divider.
...

A Cutkosky-type resistive divider seems like a very handy piece of gear to build, even if only to verify your RC-based prototype.
https://www.mijpn.com/app/download/12847871390/CIM2017.pdf

[rhb]

You can purchase an already build development PCB at Digikey ($150). The chip is already on the PCB, you add a Power supply and an external voltage reference. There are also provisions on the PCB to add an LTZ1000.

EVAL-AD5791SDZ Guide
https://www.digikey.com/products/en/development-boards-kits-programmers/evaluation-boards-digital-to-analog-converters-dacs/793?k=ad5791

Cool!  Thanks.  I wasn't able to find the LTZ1000 reference board for sale, but I'll look further.

[cellularmitosis]

He means the eval board already has unpopulated footprints where you can solder in your own LTZ components.

[rhb]

That's what I assumed, but the datasheet I looked at showed a daughterboard marked "EV-LTZ1000-REFZ".   At any rate, very satisfactory for my needs and wants.  So after I bring up my VNWA, Bodnar GPSDO, Zybo, MicroZed, LimeSDR and BeagleBoard X15 I'll get one.   Right now I am in "missing part" purgatory.

Edit: I went back and looked at the datasheet from Digikey and the LTZ1000 is on a daughterboard with no place for it on the AD5791 eval board.  i don't know if the unpopulated daughter board is part of the AD5791 eval board or a separate item. i've not yet fund a source for the LTZ1000 board.

[hwj-d]

Watch out — there are two grades to that part, A and B grade.  The B grade is the one which looks like it actually does 1ppm, and is something like $130 $107 per chip at Digikey.   Combine that with an LTZ and you’d have a nicely upgraded version of Ian Johnson’s LM399+AD5781 project!

Same idea. Have already laying around this eval board from Digikey waiting to play with it. On my board is the AD5791BRUZ (B grade). At this time I bought the board, it was only $20 more than the chip itself. The LTZ1000 layout on this board isn't exactly trustworthy, don't use it.

I have the impression, maybe, don't really know it at the moment, the AD5781 is also the better, sufficient and cheaper one.

[MisterDiodes]

The older AD5791 demo board comes with the layout for the LTZ as a side panel  - but look carefully at the layout - depending on which version you get (and if it's like the one that showed up at our lab) - the LTZ might have to be mounted on the the board bottom side.  I know some of those are OK, you'll have to trace it out when you get it...don't just solder parts on before you trace out the LTZ circuit.

Newer versions you need to order (or build) the newer LTZ Vref board, or talk to the sales rep.   I think there is a boxed demo unit available with the demo LTZ Vref included, or at least that's what our rep was talking about. Somebody here might have the part number.

The main problem with the AD5791 is the relatively high noise on the output at DC - look carefully at the datasheet and notice how the noise units change for 0.1~10Hz.  Suddenly you go from 7.5nV/rtHz to 1.1uVp-p...  That will tend to happen on IC scale resistors.  This may or may not be important to you, just pay attention.

As CellularM pointed out - You'll  want the more expensive B version if you're attempting to go into more serious lower PPM territory with 1.5LSB INL.  The Demo boards used to come with A, at least ours did, sounds like hwj-d got a good deal for $20 over cost of B chip!.   

[TiN]

Somebody here might have the part number.

Ref module for AD board is EV-LTZ1000-REFZ, available on Analog for $500.

[rhb]

Somebody here might have the part number.

Ref module for AD board is EV-LTZ1000-REFZ, available on Analog for $500.

Ouch!  Mouser has it for *only*  $450.  I clearly need to learn more about the voltage reference as I had not expected that a $50 part would cost 10x on an eval board.  Especially one intended to plug into a $150 board.

This thread  is what got me started thinking about this:

https://www.eevblog.com/forum/oshw/lets-discuss-how-to-make-oshw-more-successful-(inspiration-by-foss-design-cycle)/msg1450960/#msg1450960

Calibration gear seems to me a natural for what the OP would like to do.

I'm in the process of going from a pile of old gear mostly  in need of repair to a functional lab with new gear.  A 10 ppm reference would suffice for my needs.  I'm trying to get to basic measurements with some level of self  calibration *before* I'm too old to use it.  Buying parts is driving me nuts.  Finding even basic hardware such as #4-40 1/4" fillister head machine screws at sensible prices and quantities is an ordeal.  The web is full of people trying to sell a $3/100 screw for $0.75 each.

I have agitated a bit for some way to summarize topics whether commercial test gear or DIY.  It gets very daunting trying to find the important parts of threads of 50+ pages.  Better search within the forum would help, but would still be difficult.  A major thread wiki or a summary thread with links to the important posts in a thread would be a huge help for someone like me trying to come up to speed on the LTZ1000.

[cellularmitosis]

You make a great point rhb, a beginners guide to these circuits and to bootstrapping your own in-house levels of precision would be great.  I’ve even thought about starting a YouTube channel which would just be devoted to summarizing the LTZ1000 thread.  So far I’ve only gotten as far as buying a lapel mic

TiN made an index of many of the important threads in this section of the forum, which is definitely helpful.  Conrad Hoffman’s mini metrology lab articles are also a good starting point.  In fact, perhaps that’s where effort for creating a guide might go — into making an expanded sequel to his articles.

Have you considered starting with an LM399 circuit before the LTZ?  It is much easier to get started with, and the 7v output is stable enough that you can use it to do a pretty good job of comparing two 6.5-digit meters to each other, to see if their cal matches.  I can send you a small lm399 board for free, and you can populate it for about $15 in parts.

[rhb]

I'm not a big video fan.  I can read much faster than people can talk and looking up a piece of information in a video is very time consuming.  A video lecture in combination with a web page would be really nice.  A video intro for the novice and the web page for someone who just wants to look something up. 

At a certain level I'm not sure that "beginner" is a meaningful term.  At PhD level, not having done something before is more or less routine.  You're supposed to be doing things that you haven't done.  The whole point of that level of education is to be able to teach yourself.  I have a 5000+ volume personal technical library in which I swim from topic to topic as whim or necessity dictates. I gave up trying to remember everything long ago.  I just try to remember when  I need to pull out some books.

I've looked at some of Hoffman's articles, but need to go back over the series.  I looked at the index, but the threads themselves are *rather* long.

An LM399 would be more than satisfactory for now.  I don't have even one 6.5 digit meter.  I'll PM to arrange shipping, etc

[AG7CK]

The AD5791 has been discussed till the cows come home on this forum.

These threads / posts are easily found by searching " site:https://www.eevblog.com/forum/ ad5791 "

The video below shows what kind of "precision / linearity" that can be expected for a slapped-together AD5791 + LTZ1000 + HP 34970A DMM.



https://youtu.be/qifVn6V9ECI