LM399 based 10 V reference

[David Hess]

From the stability and simplicity it is hard to beat the resistor from the 10 V output. Even when build with separate drive and sense it is still the preferred solution.

It does have the disadvantage of adding a second order error from positive feedback because the bias current increases with output voltage, but I assume this error is insignificant.

Ratiometric sensors, including strain gauge bridges and RTDs, may add this error deliberately to improve linearity, in some cases by a lot.  I think bandgap references also use this error or something like it for bow correction over temperature.

[Kleinstein]

The reference is stable and thus not really much of 2nd order error. There is a tiny bit of amplification of effect like the TC, but this is minute. With some 3 K from 10 V to provide 1 mA and 0.5 ohm typical impedance for the LM399 this would be some 1.0002 fold gain on the TC or similar drifts. So nothing to even notice. Even for the 1N82x zeners with there 20 times high impedance and 8 x higher current the positive feddback is still a non issue. What changes is that current source and TC of the resistor can start to matter.

With the LM399 with the rather low impedance the current source is not critical. So other solutions will also work, though more complicated and still most of them less accurate (even if it does no matter). With a roughly 14000 fold (7 V / 1 mA / 0.5 ohm) attenuation of drift the qualtiy of the resistor is not that critical (just don't use low grade carbon).

[mawyatt]

From the stability and simplicity it is hard to beat the resistor from the 10 V output. Even when build with separate drive and sense it is still the preferred solution.

It does have the disadvantage of adding a second order error from positive feedback because the bias current increases with output voltage, but I assume this error is insignificant.

Ratiometric sensors, including strain gauge bridges and RTDs, may add this error deliberately to improve linearity, in some cases by a lot.  I think bandgap references also use this error or something like it for bow correction over temperature.

A similar means (+ feedback) was employed with the Caprio Quad Cell based differential amplifier to improve linearity.

Best

[senso]

Hi everyone!
Long time since I last had free time to play with electronics, and I just discovered a couple days ago 4 LM399 that I ordered from Aliexpress some 5-7 years ago.
Do they look fake?
They all have burn marks, seeming to have been extracted from some equipment. They all have gold plated leads, all have some solder in the end.

I would like to just make a basic PCB to learn how to use KiCAD since I no longer have Altium and Kicad 8 seems pretty nifty.
This would be just to say that I have a voltage reference and to leave them aging, maybe some day I will get a decent multimeter.

I read some 10 pages back on this topic, plus a couple of the initial ones, is there any "recommended" design, or just go with the basic from the datasheet, since I'm not hunting for 10.0000V..

Best regards.

[Kleinstein]

The chips look legit as reclaimed parts. The LM399 are normally quite robust, but there could still be a damaged one.

The standard circuit is with amplification to some 10 V and providing the current for the zener from there. So a bit like the "portable calibrator" circuit in the data-sheet, but without the trimmer. Depending on the amplifier used one may not need the 200 K for the start up.
I could make sense to also have an extra output for buffered 7 V.

[Overspeed]

Hi everyone!
Long time since I last had free time to play with electronics, and I just discovered a couple days ago 4 LM399 that I ordered from Aliexpress some 5-7 years ago.
Do they look fake?
They all have burn marks, seeming to have been extracted from some equipment. They all have gold plated leads, all have some solder in the end.

I would like to just make a basic PCB to learn how to use KiCAD since I no longer have Altium and Kicad 8 seems pretty nifty.
This would be just to say that I have a voltage reference and to leave them aging, maybe some day I will get a decent multimeter.

I read some 10 pages back on this topic, plus a couple of the initial ones, is there any "recommended" design, or just go with the basic from the datasheet, since I'm not hunting for 10.0000V..

Best regards.

Hello

On Github : https://github.com/circuitvalley/LM399_Voltage_reference

Not tested but with all data to be able to check including gerbers

Regards
OS

[ivo]

I would like to just make a basic PCB to learn how to use KiCAD since I no longer have Altium and Kicad 8 seems pretty nifty.
This would be just to say that I have a voltage reference and to leave them aging, maybe some day I will get a decent multimeter.

Best regards.

If you want simple, you can follow along with IMSAI Guy's design-

[iMo]

..there are definitely better designs here than the IMSAI's one..

[Overspeed]

..there are definitely better designs here than the IMSAI's one..

Hello

Could be interesting to create a post with the ( one ) best solution circuit ? or even link the Gerber files or PCB design tips as ground plane , thermal control ....

Regards
OS

[Korchahin]

I heard about a good option using 1n829a. I will be grateful for such a scheme and comments on the ADR1399 + 1N829 combination.

[Kleinstein]

A combination of ADR1399 and 1N829 makes relatively little sense. The 1N829 are more an outdated old thing.

[dietert1]

As far as i remember some Keithley meters like 2000, 2700 and 3700 use that method, i mean continuous recalibration of a low noise reference by an ovenized reference like the ADR1399. It probably requires digital processing to get a crossover frequency lower than 0.1 Hz.
By the way, our double ovenized ADR1399 with an analog low pass filter built into the 7 -> 10 V gain stage turned out to be a nice solution. It was shown in the ADR1399 thread before https://www.eevblog.com/forum/metrology/adr1399-reference/msg4559587/#msg4559587 and provides peak to peak stability of about 0.2 ppm over 24 hours. If you want something better, it's probably a LTZ1000 or an array of ADR1399. I mean one LTZ1000 buys four ADR1399 and they are easy to use.

Regards, Dieter

[Kleinstein]

The Keithley 2001, 2282 and likely 2010 are using the 2nd reference zener for the ADC. Definitely not the 2000 and 2700. The idea is getting a cross over in the frequency with the old style zener having slightly better higher frequency noise and the LM399 the better long time stability. The way the meters are build with relatively frequent checks of the ADC gain, there is little extra effort in combining the 2 references. It still adds complications for the ohms part.

However with the ADR1399 it is no longer clear that the 1N829 would really be lower noise - probably have to select with quite some effort. For he relatively small improvement it would be quite some effort. For a 10 V reference it would be extra effort to combine the 2 reference to cover different frequency ranges.

[iMo]

The major issue with the 399 and 1399 designs (good ones or bad ones) when targeting the 10V output are the two 7->10V resistors availability/quality.
Frankly, the easiest solution is the ADR1001. It combines "similar" performance of the 1399 and adds up the on-chip ovenized/trimmed resistors as well (all in ceramic). It has got some minor quirks and it might require a proven history of a metrology grade measurements, sure, but for generating 2.5/5/7/10V output there is hardly a "cheaper*" and a simpler solution available..
*ADR1001 is pretty pricey, tbh, but..

With a "traditional" 399/1399 designs the "best" solutions would be those with 2/4/N refs in parallel, with the zener's cathode resistor biased off the 10V output (and buffered by an npn/jfet transistor), with those 2 critical 7->10V resistors having "ratio TC" better than the TC of those 399/1399s used (resistors difficult to get, people use "single chip statistical resistor arrays" sometimes). Another option is to ovenize the entire stuff to mitigate the resistor's bad ratio TC somehow. Have a look into this thread there are several such designs elaborated again and again..

..there are definitely better designs here than the IMSAI's one..

Hello

Could be interesting to create a post with the ( one ) best solution circuit ? or even link the Gerber files or PCB design tips as ground plane , thermal control ....

Regards
OS

PS: below a typical design I would consider "good". Sure, there are nuances we may discuss, as usual..

Notes:
1) All caps in the "signal path" -> foil.
2) Input voltage >19V, ie. 5*18650.
3) 7815 off the Vref pcb (7815 changes its temperature with the input voltage variations).
4) Most critical for the TC and the long term stability is the resistor's array and the trimming resistor (with a trimmer adjusting you would need a special wiring).
5) PCB design - star config(s) with grounds (low current and high current tracks and grounds, the schematics does not show the nuances), special handling of the area around the 399 pins (thermal gradients removal), cutouts (mechanical stress avoidance),
etc. etc..
6) Resistor array - that one in the example has "ratio TC" <=1ppm/C, the trimming resistor (say 10ppm/C) makes it even worse. That is 2-5x more than the TC of the 399/1399. The vintage OP07 (or use a newer one there) may have less influence on the TC than those 7->10V upscaling resistors.. Another point is the noise of the resistors (there are better arrays people say).

[iMo]

.. here is a sim showing a possible trimming with a pot-trimmer (assuming its TC=0) and 470k (TC=20ppm/C for example) wired to pot's wiper (aprox 10x value of the lower resistor in the divider).
The X axis is the wiper's position (0..1), the temperature sweeping from 20C to 30C is in color..
In that temperature delta (10C) the 10V output voltage changes by aprox 3ppm, that is aprox 0.3ppm/C in total.
As you may see the trimming to the exact 10V is quite difficult exercise even with a 10turn pot..

[dietert1]

For the trimming pot a better solution is to connect the wiper to the opamp negative input in order to make the wiper current as small as possible. The wiper has some extra and quite unstable contact resistance that will later add voltage variations depending on wiper current.
Once i got from ebay hermetic foil resistors of 350 and 900 Ohm that make a 0.72 low noise divider, near perfect for a LTZ1000 with the notorious 20 Ohm TC compensation resistor. As i bought two of each i wired them as a bridge, connected a 10 V DC supply and looked at the bridge imbalance with a  sensitive meter. Of course there was a difference but it would remain constant to better than a uV for days, even under ambient temperature changes of 1 or 2 °C. Those resistors were quite expensive but i think with a little patience one can arrive at a good DIY solution.
Other possible paths mentioned before are using NOMCT resistor arrays or a 14 V reference array with a PWM divider. The PWM divider has the advantage that calibration including TC compensation can be performed in firmware.

Regards, Dieter

[iMo]

The wiper is connected (via the 470k) to the opamp's negative input, afaik.
The current through the 470k is 6.5uA.
That is a huge current, indeed, considering the wiper's resistance fluctuations in, say, +/- XX ohm..

For the trimming pot a better solution is to connect the wiper to the opamp negative input in order to make the wiper current as small as possible. The wiper has some extra and quite unstable contact resistance that will later add voltage variations depending on wiper current..

PS: with this below the current through the wiper is the same..

PPS: AI on wiper resistance

| Parameter                                                                    | Typical Range for Small 10T Trimmers | | -----------------------------------------------------------------| ------------------------------------------ | | Wiper contact resistance                                                 | 10 mΩ to 100 mΩ                             | | Wiper resistance fluctuation[while moving the wiper?]       | ±1% to ±3% of total resistance         | | CRV (Contact Resistance Variation) [??]                           | Up to 50 Ω or 3%                              |

[Korchahin]

Hi guys!
I was wondering if it is possible to use a resistive divider of the current stabilizer in the circuit for 399/1399 to connect to the ADC as Vref? I sketched out a circuit in which I wanted to show what I am talking about. I am interested in the point where XMM1 is connected.

[Simmed]

Hi guys!
I was wondering if it is possible to use a resistive divider of the current stabilizer in the circuit for 399/1399 to connect to the ADC as Vref? I sketched out a circuit in which I wanted to show what I am talking about. I am interested in the point where XMM1 is connected.

XMM1 --> opamp buffer --> ADC

[iMo]

Just curious I made a sim of the wiper's TC contribution in my above schematics (this time with a jfet at the output).
In order to see some temperature dependency I had to increase the wiper's resistance to extreme 1000ohm and I used 500ppm/C.
The sim below (X axis is the wiper's position 0-1).
The voltage change is 3.3uV from 15C to 40C, thus 130nV/C.
The current through the wiper is aprox 6.2uA, the question is whether it is too much or too low..
In reality the wiper's resistance may fluctuate due to its oxidation and/or mechanical stresses as well.

PS: the R3=470k value is set such the 10V at the output is aprox in the "middle of the trimmer's path". So R3 may change based on the zener voltage and the exact fb divider ratio..
The R3 with 25ppm/C (example only) contributes with 3.2uV/C.

PPS: the trimming with the trimmer to 10.0V in this scenario would be rather difficult, as you may see from the data on X axis (the wiper's position).. You may pad the trimmer with resistors, for example, but the resistor's ratio TC will contribute then (we assume the "ratio TC" of the trimmer itself is 0).

[iMo]

An example of an LM399#2 which is the one of "the lowest 0.1-10Hz noise" one I have here, but doing this.. The jumps are exactly 4uV.
A day long measurement - 34401A in the "ratio mode", raw data, the ADR1001#1 as the reference, LM399#2 is the DUT.
With 100NPLC a single measurement takes aprox 8secs. The LM399#2 wired with a cathode 100ppm/C resistor only, the snubber is there as well, 3.4mA current (the TC is not important in this case, blue is the ambient temperature), battery powered.

[David Hess]

An example of an LM399#2 which is the one of "the lowest 0.1-10Hz noise" one I have here, but doing this. The jumps are exactly 4uV.
A day long measurement ...

Looks like classic popcorn noise, which seems to be more common in the lowest noise references.  It is difficult to test for in production because it can take days or weeks or longer to manifest, and some customers pay extra for burned in references to avoid it at all cost.  It used to be a big problem in operational amplifiers and analog ICs in general, but parts have gotten better.

Is LM399#2 a new part?

Shown below is an example of popcorn noise that I found in the analog channel switch of a Tektronix 2230 DSO made in 1986 that took more than a month to track down.

[iMo]

Yep, LM399H, date code 1836.. Some time back I put it inside my 34401A with a potential for replacing the original HP399 one (with higher noise), but I saw one shot per day. So I repeated the measurement again outside the meter (with 3.4mA current) and I've got 7 shots per day. Just charging the batteries and I will repeat the same with my lowest noise 1399..

BTW - with ratio measurement - would not be better to switch the AZ off? Or do it with 10NPLC? I mean - the shorter the period between the two measurements (extREF<->HP399 and DUT<->HP399) the less deviation the internal HP399 may cause?

[Kleinstein]

I have seen similar 4 µV range jumps in all 4 of my LM399 reference. While the size of the steps is similar the rate for the jumps varies. Some faster (like 1 jump per minute) to rather slow like about 1 jump per hour.

The HP34401 would likely still need the AZ mode to reduce it's own 1/f noise from the amplifier and ADC.
As the 34401 is relatively noisy and has a LM399 itself one could get less noise by looking a the difference between the LM399 the other low noise ref.

[iMo]

A straight difference is 3V, that will not help much. When dividing 10V ref down to 7V you get null meter method. But getting down from 10V ref to 7V for null meter comparison is not an easy task either..
PS: I built such a divider for null meter myself, perhaps I will try with it. I can go below 100uV "easily" with it, but the precise divider's value is not known (so perhaps good for relative measurements only)..
Below a single 12hours long test I ever made with the divider, where the ADuC845 (quite noisy btw) has been used as the "null meter"..