LM399 based 10 V reference

[TiN]

Unlike my work, you have it up and running, so I would say you are way ahead of my LM399 projects. 
And for prototype testing, your build is nice, tidy and clean. So you have my thumbs up right there. 

You can build two of these, and run them back to back, using 34465 as difference meter between two REFs on 100mV range. That will give you some idea on the noise as well.
Then if surely few friends more local to you with access to calibrated 3458/reference could calibrate your ref and provide you pretty good accuracy for further tests. 

Worthy to put into metal box with low-thermal binding posts and battery, so you can use it as portable test source 

[Galaxyrise]

Since temperature and resistance matching is going to best for the resistors in the same package, perhaps parallel two packages.  For example, for the top half of the ratio, use 1 from package A and 2 from package B.  Then for the bottom half of the ratio, use 7 from package A and 2 from package B.  That produces a 3/7 ratio such that only the ratio within each package needs to be maintained. The combination is 2.2 times the value of a single resistor, so maybe the 5k resistors makes more sense.  But with so many ratios available this way, one could choose something closer to what their LM399 actually needs to minimize trimming.

[JimmyJo]

Thank you for the kind words, TiN. 

Since temperature and resistance matching is going to best for the resistors in the same package, perhaps parallel two packages.  For example, for the top half of the ratio, use 1 from package A and 2 from package B.  Then for the bottom half of the ratio, use 7 from package A and 2 from package B.  That produces a 3/7 ratio such that only the ratio within each package needs to be maintained. The combination is 2.2 times the value of a single resistor, so maybe the 5k resistors makes more sense.  But with so many ratios available this way, one could choose something closer to what their LM399 actually needs to minimize trimming.

That is an interesting arrangement of resistors, but im not sure if it works the way you described.  To simplify the mental math, lets grossly exaggerate the problem.  Lets say the resistance of one of the resistor networks increases by few orders or magnitudes, the overall ratio of the arrangement would then be dominated by the other resistor network.  A quick calculation in excel shows that a 5ppm change in one of the networks will lead to a ~1.2ppm change in my circuit, and ~2.2ppm change in the arrangement you've shown.  Can you confirm my math?

[z01z]

It's possible to do the ratio with one package, as 1.5k/3.5k.

[Svgeesus]

I see six wires going to the LM399 so Kelvin connection. But I'm not sure where exactly those would go. One 0V wire to your output terminal and another to the 0V of the op-amp, I guess. And one 7V wire to the op-amp input and another to the resistor supplying the zener current?

[JimmyJo]

I see six wires going to the LM399 so Kelvin connection. But I'm not sure where exactly those would go. One 0V wire to your output terminal and another to the 0V of the op-amp, I guess. And one 7V wire to the op-amp input and another to the resistor supplying the zener current?

That is my poor attempt at a Kelvin connection, I'm not 100% sure how it should be wired, especially the low side.  You/ve guessed the 7V side correctly, that's exactly the way i wired it.   As for the low side, one goes to power supply ground, the other one goes to output low.  But my 3k/7k return is also tied to the output node, so there is a constant 1mA there.

[enut11]

Hi All
In Reply #533 I purchased four LM399 chips and acquired two more since. The Refs have been 'cooking' now for about 18 months. At irregular intervals I spot tested the Vz and 10v outputs and occasionally reset the 10v to 10.00000v when it wandered too far, presumably due to aging components.
Initial tests were done with my HP3456A but lately on a near-new Keysight 34461A. The object of the exercise is to pick the most stable LM399 PCB for further experiments.

Results pages attached:

[enut11]

Expanded graph sections from the results above. All measurements are spot checks.
The blue line is Vz. Graph shows variation from test-to-test.
The red line is 10v output. Graph shows variation from 10v. I occasionally tweaked the 10v output to bring it back to 10.00000v.
Some of the graphs show off-chart excursions. These are simply missing data from when the Refs were traveling around Australia for the Aussie Cal Club.

I am zeroing in on Ref #5 as the best of this lot.
enut11

[VK5RC]

Hi Enut11,
Did you use the same cables etc during testing? Any Temp data ?
Wth my 34461, I found if I used the Maths function, after 1hr warm up, connected everything, waited a while ie 2-3mins and then the display was often quite stable and would show down to uV levels.
My favourite cable is Pomona's twin banana to twin banana shield low EMF cable - mainly as I can't accidentally short out the reference - don't ask!!!!!!!!
Have you checked for any interference - I had a grounding issue - and switching some lights on/off would blow my reading by 30-100s of uV. Once I solved this - putting everything in a metal box - connecting electrically everything - it all settled down a lot.
Kind regards
Rob

[enut11]

Hi @VK5RC
I did use the same cables for the latter 12 months of testing as I made up a special 3-way set to measure 10v and Vz. I also simultaneously compared Vz on the HP3456A and the Keysight 34461A.

Before I make my final choice on the best one I will do a more thorough comparison.

At the moment I am working on Max's version of the insulated Ref oven and getting good results - around 0.1C max variation at 34C over 24hrs. Playing around with different insulators. It all takes time.
enut11

[Wolfgang]

IMHO the LM317 is a lot noisier than even more classic regulators like the LM723 with a PNP pass transistor.

DK7JB has created a paper (unfortunately in German) comparing a lot of voltage regulators for output noise (just look at the graphs). The LM723 won hands down.
https://www.bartelsos.de/dk7jb.php/rauschen-von-spannungsreglern?download=115

Powering a LM399 should be done by a supply with minimum noise.
Some hints can be found in:

https://electronicprojectsforfun.wordpress.com/power-supplies/a-collection-of-proper-design-practices-using-the-lm723-ic-regulator/

[David Hess]

IMHO the LM317 is a lot noisier than even more classic regulators like the LM723 with a PNP pass transistor.

DK7JB has created a paper (unfortunately in German) comparing a lot of voltage regulators for output noise (just look at the graphs). The LM723 won hands down.
https://www.bartelsos.de/dk7jb.php/rauschen-von-spannungsreglern?download=115

All of the integrated regulators which use a bandgap reference are pretty poor compared to the zener reference on the 723.  They not only have have higher reference noise but must multiply it to get their output voltage.

I would like to see how using a 317 as the pass element of a 723 would perform.  Enclosing the noisy 317 inside the 723's feedback loop and low noise reference should work pretty well.  The advantage is integrated protection of the pass element.

[Wolfgang]

Hi Dave,

I have never seen this arrangement, but I could give it a try when I'm back at the bench. While protection is cool, I am afraid that the 317 produces more noise than the LM723 is able to kill (the error amp gain is only 33dB).

We will see 

[Gyro]

IMHO the LM317 is a lot noisier than even more classic regulators like the LM723 with a PNP pass transistor.

DK7JB has created a paper (unfortunately in German) comparing a lot of voltage regulators for output noise (just look at the graphs). The LM723 won hands down.
https://www.bartelsos.de/dk7jb.php/rauschen-von-spannungsreglern?download=115

All of the integrated regulators which use a bandgap reference are pretty poor compared to the zener reference on the 723.  They not only have have higher reference noise but must multiply it to get their output voltage.

I would like to see how using a 317 as the pass element of a 723 would perform.  Enclosing the noisy 317 inside the 723's feedback loop and low noise reference should work pretty well.  The advantage is integrated protection of the pass element.

If you want a protected pass element, then the LM395 would be a better choice. No bandgap reference at all.

[Wolfgang]

Hi,

the LM395 might not be a bad choice, but I missed any noise specs for it. My guess is that is way better than a 317, for a few €s more.
Anyway, also worth a try. 

Some Horowitz and Hill precision supplies as well as Linear Tech App Notes used a separate preregulator using a 317, but outside of the main regulators loop and with extra noise filtering before. Adding current limit to the precision regulating loop adds noise.

[David Hess]

I have never seen this arrangement, but I could give it a try when I'm back at the bench. While protection is cool, I am afraid that the 317 produces more noise than the LM723 is able to kill (the error amp gain is only 33dB).

I have used it many times but with an external reference bootstrapped off of the output and a precision operational amplifier as shown in the example below.  I just never tried it with a 723.

If you want a protected pass element, then the LM395 would be a better choice. No bandgap reference at all.

I agree but they are way more expensive.

Adding current limit to the precision regulating loop adds noise.

Oddly enough there may be another good reason to add a current limit via a shunt between the pass element and output filtering capacitor.  A small amount of AC feedback can be taken from before the current shunt to stabilize the feedback loop with low ESR output capacitors.  Some integrated regulators use this trick but they have the advantage of sampling the AC feedback from the pass transistor structure itself without adding any series resistance.

Effectively this adds a controlled series resistance to the output capacitor for stability purposes while maintaining a low ESR for filtering purposes.

[Andreas]

Powering a LM399 should be done by a supply with minimum noise.

Why that? in which application. (the LM399 dampens the supply noise at least by 3K / 1 Ohm).

And why do you look at the wideband noise and not the 1/f noise which is more important for integrating voltage measurement instruments?

For a HP34401A a LM317/LM337 seems to be enough.

with best regards

Andreas

[Wolfgang]

As with all regulating circuitry, PSRR deteriorates with frequency. I may be conservative, but I try to kill noise or ripple in a signal or PSU chain as early as possible. This is not a solution for the problem of 1/f noise that is *generated* in (pre) regulator stages, I agree.

[GigaJoe]

that it ....


top 10V, then 1V then 0.1V ;  Right side variable output 0-20V, adjusted precision up to 1uV
it based on 2 LM399, that i grilled a while ago; self made 1:10 dividers. and other cheapo thingi ..  TC I measured +0.7 ppm\C (spent 2 weeks to do adjustment, finally give-up , but can be done better) on 10V, and not measurable on 1V.

variable based on 2DW232 and some 1:10 divider if need some millivolts output ... 

reference and variable has 2 independent transformers. so I had fun idea,  I set Variable output to compensate 1Vout,   and after 12 ours it was really stable , but after a day pass, it state -20uV - I guess due to env temperature up , as I didn't do anything for a thermal compensation on  variable output.

u use my K2001 to adjust as it suppose to,  consider a performance report on K2001 i did a year ago, soo will see around december how it will change ... if so ...

now I will think about current source now, a bit more challenge due to 1A output needs.

[TiN]

BNC connectors are not good for precision microvolt stuff .

[GigaJoe]

but HP 3245A has it .... at least connect disconnect on 0.1V i had repeatable result .... anyway ....will see ...

[branadic]

While I was waiting for the parts for LTZ1000 references I had the time to draw circuit and board file for the "LM399 PWM DAC reference" based on the "ION" MickleT. presented over here as a representitive of the russian forum (radiokot.ru).
As far as I understood the original design came from Sergei and is available here:
The original design comes from adver, but sources can be found on Sergei's google drive.
You can find the source code for two versions:
- 8bit + 14bit fractional
- 8bit + 16bit fractional
and lot's of pictures.

I made some minor changes and put some additional parts into the circuit:
    - LT1763 voltage regulator for voltage reference (12V) as well as microcontroller (5V)
    - additional resistor divider feedback (for testing purpose at first glance)
    - output buffer (LTC2057)
    - NTC temperature sensor on one of the ADC pins, to add a temperature compensation scheme
    - optical transmitter (TORX1952 or SFH752V) for galvanically insulated data transfer of temperature data and pwm setting to develope temperature compensation algorithm

The board does fit, what a miracle, into a Hammond 1455C801 case.
The boards have been ordered at Elecrow and will arrive soon. In case someone wants to duplicate it and contribute to this reference, attached are the design files (Gerber data). Eagle files on request.

-branadic-

[DC1MC]

I'm chewing my fingernails, it's actually the first PWM implementation that I know off.
What range it's planned to have ?

 Cheers,
 DC1MC

[branadic]

Actually the PWM is only to boost the zener output voltage to 10V. I'm still waiting for the boards to arrive. Customs took quite long as the boards arrived in Frankfurt on 22th of november, so 10 days ago.

-branadic-

[DC1MC]

Actually the PWM is only to boost the zener output voltage to 10V. I'm still waiting for the boards to arrive. Customs took quite long as the boards arrived in Frankfurt on 22th of november, so 10 days ago.

-branadic-

Ah, so it's not planned to be adjusted in a wide range, but just like a finely controlled amplifier. Still very interesting.

 Cheers,
 DC1MC