LM399 based 10 V reference

[branadic]

In preparation of some lab intercomparisons I did some measurements on the LM399-PWM-DAC.
First a 1h measurement using an uncalibrated 2182A @ 1NPLC, second a t.c. measurement using the integrated NTC. It turned out the t.c. is -0.275 ppm/K. Further stability measurements and taking some more readings with different calibrated meters are both in progress.

-branadic-

[Andreas]

Hello branadic,

just to get shure:
it is the schematic mentioned here?
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg1955149/#msg1955149

which analog Switch + PWM frequency did you use?
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg2200098/#msg2200098

with best regards

Andreas

[branadic]

Hi Andreas,

it's the version with the modification suggested by Kleinstein, but with the original 8 + 16 bit resolution firmware, 20 MHz crystal and ADG419.

https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg2082544/#msg2082544

-branadic-

[Andreas]

Hello branadic,

so you have the  ~10 kHz PWM frequency with prescaler = 8:1.

with the optimized Code (for 8.14 Bit PWM) from here
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg2189460/#msg2189460
Prescaler 1:1 and XTAL frequency 4.5 .. 5 MHz you could get near zero T.C.

With best regards

Andreas

[branadic]

Thanks Andreas, but I stay with what I have at the moment as I don't have the required components in my drawer.

-branadic-

[oz2cpu]

shodan i dont understand your circuit ?
in your post 1216 you show a schematic with 5K and 10K in the feedback
how can this deliver 10.000V out, when the LM399 is about 7V ?

[oz2cpu]

thanks a lot, this helps : since it is very easy to get 1:2 resistors in the crasy matched type kind of temp match
then it is possible to add a "normal type of ressistor to trim the value to at least 10.0xx
this minor trim will not affect temperature match so much

[free_electron]

TL;DR . Bob pease or jim williams had a circuit where they used like 10 of the lm199's. anyone know where to find that schematic ?

[guenthert]

how can this deliver 10.000V out, when the LM399 is about 7V ?

You can see exact value on legend into post #1227
No one could prove to me "REF should be 10.00000000000V exactly", therefore, I use a voltage around 10.6V.
10.6V Its fine value, because most DMM accept overload to 10% on 10V range, without any problems. 

Probably the main problem is that I'm not a voltnuts, so 10.6V works fine for me. I can calculate ppm drift with 10.6 base w/o problem.

Moreover: Into my last toy, i use direct connection LM399 to 20 bit DAC with OPA189 buffer and mirror... so... as result DAC output is +6.9V...-6.9V, no problem here... works very precise! 

   10.6V is fine as a voltage reference, as is 6.9V.  The reason to chose a 'standard' voltage is so that it can be compared directly (difference method) with other 'standard' voltage sources (using e.g. nanovoltmeters with high resolution, but a limited range or a null-voltmeter with an even smaller range).

   Incidentally, I do however have a 16bit PCI ADC adapter with a 10V range, which doesn't allow for *any* overrange.  Even the 40mV offset on my reference are too much 
So a bit below the nominal voltage would be better in that case.

[Kleinstein]

Directly measuring the difference also helps when looking at the noise. With a modern meter to read the difference and the still relatively noisy LM399 there is no big problem if the difference is in the 1 V range of the meter. The full 7 or 10 V meter can be a bit demanding for the DMM and the reference to compare to is than fixed to the DMM internal one.

There are a few other divider ratios that can be obtained with a reasonable small number of equal resistors. With 6 equal resistors one can get a 10/7 gain (1 :2.333 resistor ratio).
It would anyway be a good idea to still have access to the raw 7 V to at least check the 7 to 10 V stage stability seprately. Long term stability is hard to judge upfront and in the ppm range there are sometimes unexpected small effects even with good parts. With enough time modern DMMs are not that bad in measuring the 7 to 10 V ratio and a resistor ratio has a hard time (expecially not guarantied) to be that stable over a long time.

[oz2cpu]

what i think you are trying to say :
every component you add to the output of your lm399 or other ref
you dont add accurasy or stability, but you add more problems.

[mawyatt]

TL;DR . Bob pease or jim williams had a circuit where they used like 10 of the lm199's. anyone know where to find that schematic ?

Think that was Bob Pease in response to the newly introduced LTZ1000 or whatever it was originally called way back, but can't recall the exact details nor schematic.

Best,

[dietert1]

In this thread (above) i reported a build of a 5x LM399 array, with image and schematic including a PWM gain stage 7 => 10 V. There were some measurements on initial aging and a six months log.
Meanwhile i built a 10x LM399 reference with 14 V output (2 by 5) with a PWM divider 14 V => 10 V and programmable for use as calibrator. Hope to report measurements later.

Regards, Dieter

[free_electron]

In this thread (above) i reported a build of a 5x LM399 array, with image and schematic including a PWM gain stage 7 => 10 V. There were some measurements on initial aging and a six months log.
Meanwhile i built a 10x LM399 reference with 14 V output (2 by 5) with a PWM divider 14 V => 10 V and programmable for use as calibrator. Hope to report measurements later.

Regards, Dieter

why pwm ? use a multiplying DAC ... send output of the dac into summing amplifier.
take 10V ref output , divide by 10 send into dac 1. , divide ref by 100 , send into dac 2 , take ref , dac1 output and dac2 output into summing amplifier.... if you use a 10 bit dac ( 1024 steps)

you don't care what the absolute value of your 10v is. it can be trimmed using the dac.
you take the reference and create a band around it. let's say 9.9 and 10.1 volt using two opamps. use that as the DAC ref+ and ref-. the output of the dac is post-divided 1/10. sum the referenc ewith the post-divided output and you can trim the output very precisely. store trimdata in nvram.

[Kleinstein]

DACs are not that stable on the long run.  Many internaly use resistors for the R2R chain and are thus not more stable than the resistors.
Trimming with resistors is possible and this may be combined with a DAC for the very fine part. Combining multiple DACs is often not helping very much, as often (the PWM DAC is one of the few exceptions) the stabilty of the DAC for the coarse part is not much more stable than it's resolution.

The PWM DAC can be very long term stable. There is some drift (e.g. from charge injection that might change over time), but usually not much.

[dietert1]

Those projects require some patience to try various revisions. For example i first used an integrated MOS switch to implement the PWM. Then i found that the delays of those switches (50 to 100 nsec typ.) have temperature dependence, a detail largely unspecified in any datasheet, yet important to get the best precision. So the calibrator (second revision) got a discrete implementation resembling those Fluke calibrators, with switching delays of less than 5 nsec. Then temperature dependence and aging effects mentioned by Kleinstein shall also be 10x less than with the integrated switchess.
I think a good test of a PWM DAC is its native precision, i mean without any calibration. If one can get the output within some ppm of the predicted value, it can be made sub-ppm by calibration. If using 10 or 20 MHz as clock, a one second integrated DC with sub ppm resolution is possible. My calibrator test setup uses a MSP430F2616 running at 16 MHz. So one may even use a temperature measuement to numerically compensate residual TC.

Regards, Dieter

[David Hess]

DACs are not that stable on the long run.  Many internaly use resistors for the R2R chain and are thus not more stable than the resistors.

Analog Devices has a 20-bit DAC which is suppose to be stable to 22 bits.

[oz2cpu]

i dont understand what is the idea about pwm a voltage ratio,
since the voltage is critical in the sub ppm range, the timing demands and later filtering, to make it DC again,
is super critical..

the fundamental : pwm ratio, of voltages, sounds like a good idea, since the ratio is a digital value,
but we want to have a stable, noise free DC

[dietert1]

The PWM circuit can be built very stable. This was the result of Fluke research many years ago. I remember a list of download links in another thread in this forum with Fluke papers describing how and why this is a good method.
Here the discussion may be a bit off-topic.
The basic idea is like this: ppm precision in a 1 second DC integral requires timing correct to 1 usec or lets say 100 nsec, which is easy to do. Many parts we are using nowadays work at bandwidths of 500 MHz, so that would be less than 1 nsec. When you start with circuit design, it will be a similar adventure as building a high resolution DVM. You can read one of the threads about building DVMs, e.g. Kleinsteins.
You will find that you shouldn't use the built-in clock generator of the MCU, but a separate clock generator.
You will find that you shouldn't use a 74HC74 or a FPGA for synchronization, but you should rather select a 74lcv74 or the like.
Then the MOS switches. The discrete solution i found uses a dual n-channel mosfet with pair specs. The Rdson difference that shifts output voltage can be as little a 0.1 Ohm. When the filter impedance is 50 KOhm or so, this is an error of 2 ppm at most.
The AD DAC 5791 mentioned above is a good part and it will beat many of those calibrators available at ebay. Yet it won't replace a Fluke 5700 calibrator. I'd guess to build a PWM DAC of that quality one needs a 1 GHz digital scope.

Regards, Dieter

[guenthert]

The PWM circuit can be built very stable. This was the result of Fluke research many years ago.

Minor correction, Valhalla beat Fluke to it and their 2720GS was preceded by research and patent of a French fellow in the 60s: https://adret-electronique.fr/appareils_adret_metrologie/104a/104a.html

[iMo]

At the end of the day you have to create the DC out of the PWM train. Do we have data on how the low pass filter components stability affects the created DC?

[Kleinstein]

The PWM to DC filter can indeed effect the stability. If usually uses a relatively large resistor in the 50-200 K range and this makes it somewhat sensitive to changes in the bias / leakage current.
For the level of accuracy of the LM399 this is usually OK with a reasonable moder AZ OP like LTC2057. The version with gain adds some additional complications. The possibly more tricky part than the filter / buffer after the filter can be the driver to provide a low impedance reference signal to the PWM switches. Here the settling performance to the current spikes can matter.

Using PWM as a long term stable divider is not necessary the easiest way to go, but definitely an option worth considering.  The brute force way with very stable reistors. Another, a bit more exotic way is to use a precision transformer, somewhat similar to the ACAL part in the Datron 1281. This can also be quite stable on the long term, though the resolution is a bit limited to integer turn numbers.

Resistors tend to drift over time. The PWM ratio an a transformer turn ration can be very stable over time, but it takes extra effort and includes some extra interference even for a short time scale. For a short time scale the resistors are the obvious winner but the longer the time, the more attractive the alternatives get.

[dietert1]

Making a precision PWM source is much easier than making a precision DVM. Once time division has been implemented with good quality, the filter is very easy in comparison to the integrator. The caps are selected for low leakage, dielectric absorption is no issue as they run at constant voltage. I found caps rated at 160V would perform very well at 10 V or so. In fact i even implemented kind of a guard scheme to run the caps at near zero voltage. 1 pA leakage on 50 KOhm gives 50 nV.
Also i found on the web a ripple cancelation circuit that simplifies the filter and that reduces the ripple of the PWM DAC input as well. It's in the schematic i posted above.

Regards, Dieter

[iMo]

@dietert1: do you plan to sell the final version of your PCB (inside the EU, for example).. ?

[dietert1]

To be honest i looked into this after reading the 2019 reports of Andreas and branadic on their experiments with PWM gain stages (see above). Meanwhile they seem to have better results, too. As far as i remember they initiated kind of a calibration circle in May, using a PWM.

Regards, Dieter