LM399 Voltage Reference, Fluke 289, Gossen 29S, RM303, AN870

[Trader]

Just sharing an experiment...  There are awesome Voltage References as "DMMCheck Plus" and "PDVS2mini 20bit", and also cheaper options like AD584 and LM399.

I got an LM399 on Aliexpress, I was waiting for a good product and (very important) a Precise Calibration. I tested a very new Fluke 289, Gossen 29S, Richmeters RM303, and Aneng AN870.

None of them exactly matched the LM399, but the difference is between 0.02% and 0.01% (or 1/10,000), an excellent precision!!!

I'm happy that Fluke and Gossen exactly matched ALL the measurements (both have a Calibration certificate and they scored exactly 5.0000 VDC). Also, the chinese DMMs performed Very Well.

I think a factory produces a bunch of LM399 references and each seller is responsible to manually calibrate it. I saw some pics with the Voltages printed, handwritten, or even photocopy. Maybe (maybe) the seller uses a regular DMM and "create a fake decimal value", like 5.000V "un-rounded" to 4.99955V.

Also, even with a stable 15V input and constant 24C temperature, there is a voltage fluctuation of +- 0.1mV.

In summary: get a good quality Voltage Reference, Correctly Calibrated by a Trustable seller, unfortunately, on Aliexpress not only the products and specs can be fake, but also the Services. Now, I need to find somewhere a +6 1/2 digits DMM to check the correct calibration. 

[masterx81]

I've a similar chinese LM399 reference, and neither match my in calibration 34401a 😂😂

[Trader]

I've a similar chinese LM399 reference, and neither match my in calibration 34401a 😂😂

Post 399, a subliminal reference msg? LOL

[Martin72]

Now, I need to find somewhere a +6 1/2 digits DMM to check the correct calibration. 

Why ? 

The values of your calibrated multimeters matching nearly perfect to the values from the reference, written on the board.
However they don´t trim it to 2.5, 5, 7.5, 10V.
They obviously measured only the voltages without trimming anything and write it down.
And you know the accuracy of the giving values.
If you got 5.0000V and measure 5.0000V or you got 4.9777V and measure 4.9777V...What´s the difference inbetween ? 

[TiN]

You could get 5.0000 V known to +/- 100ppm or 4.9777V with uncertainty +/-10 ppm. Obviously second measurement would be more trustworthy. Listing values without uncertainty attached to it is just speculation digits and not defining a measurand.

[Trader]

Now, I need to find somewhere a +6 1/2 digits DMM to check the correct calibration. 

Why ? 

As I said, the farthest value is just 0.02% from the reference, this is awesome, almost perfect.

When I have an opportunity, I'll re-calibrate this Reference, but this is not urgent, just for fun.

[bdunham7]

As I said, the farthest value is just 0.02% from the reference, this is awesome, almost perfect.

When I have an opportunity, I'll re-calibrate this Reference, but this is not urgent, just for fun.

I'd suggest you power it up and let it cook for a few months.  Then report back with the initial drift numbers.

[Trader]

You could get 5.0000 V known to +/- 100ppm or 4.9777V with uncertainty +/-10 ppm. Obviously second measurement would be more trustworthy. Listing values without uncertainty attached to it is just speculation digits and not defining a measurand.

The LM399 has 0.5ppm (maybe the circuit is more than that).

Image 04 shows an 80uV (80 ppm) variation. The variation range is 0.020% and 0.024%.

As I said, it's excellent, is just a test for fun, but because Fluke and Gossen found the same values, I think the reference is not correctly calibrated.

[Kleinstein]

The LM399 may reach the 0.5 ppm/K level.  However the scaling circuit to 2.5 , 5 and 10 V will likely add quite a bit more drift - expect more like 10-25 ppm/K. The 7.5 V output setting is likely the most stable one, as there is little scaling involved - at least if they get the circuit right.
In addition there can be long time drift from aging, both from the reference itself and the resistors for scaling. The long time drift is often the more critical factor than the TC.


The difference to the 2 meter readings that agree casts some doubt on the numbers written on the box.

[masterx81]

I've a similar chinese LM399 reference, and neither match my in calibration 34401a 😂😂

Post 399, a subliminal reference msg? LOL

Next step is at LTZ1000 🤦‍♂️🤣

[Trader]

Next step is at LTZ1000 🤦‍♂️🤣

LTZ1000ACH ($200) WOW...  some Voltages References are so expensive, maybe is better to buy a new BM869s ($212).

[TiN]

0.5 ppm of what? I don't see any evidence to give thinking that reference was not calibrated correctly. Fluke and Gossen have very wide (compared to typical LM399) uncertainty, and their agreement don't mean anything much. They could be both off same way.

Its like you trying to calibrate precision 10-digit frequency standard with handheld watch, counting seconds between readings.. If you like to play with metrology I can recommend getting PDVS2mini, its far more capable and useful thanks to adjustable output and Ian's traceable calibration. It also have better output terminals instead of bad SMA (for DC reference).

[bdunham7]

Has anyone popped the can on one of these to see what is actually in there?  Or how about a photo of the backside of the PCB board, which I notice none of the sellers post?

[MikeLud]

Has anyone popped the can on one of these to see what is actually in there?  Or how about a photo of the backside of the PCB board, which I notice none of the sellers post?

[Trader]

I don't see any evidence to give thinking that reference was not calibrated correctly.

I also don't see any evidence to give thinking that reference was calibrated correctly.

2 calibrated DMMs matched the measurements, someday I'll check the LM399 VR with a +6 1/2 digits DMM.

[Trader]

Has anyone popped the can on one of these to see what is actually in there?  Or how about a photo of the backside of the PCB board, which I notice none of the sellers post?

[bdunham7]

Well, OK, looks legit (except maybe the soldering).  It must be quite a chore for them to bin all the resistors to get that close a match on the voltages.  And the tempco of the LM399 would seem insignificant compared to the divider resistors.  They can't possibly afford low-tempco precision parts in that application.

[TiN]

I also don't see any evidence to give thinking that reference was calibrated correctly.

I didn't say it was It's just my point was that you don't have equipment to properly verify LM399-based reference. There is no such thing as "correctly calibrated", because every device/reference/DMM is incorrect by design/operation (short of quantum standards, but that's another story). Question that matters - by how much is it incorrect?

If you'd buy two of these modules - you could use Gossen 29S to test one against the other using lowest mV range of the DMM.

Difference between two references is often little easier to measure to good accuracy than absolute large scale value. It's valid method even on commercial lab level metrology, using nanovoltmeters to test one zener reference vs other, getting better results than even 8.5-digit meter

[masterx81]

My china lm399 reference after 1.30h of warmup measure 7.38621v on a in-cal 34401 while on the reference there is wrote 7.38599. 220uv of error.
While the 9.84664v measure 9.84713. The 2.54773 measure 2.46511.

[mawyatt]

It appears these PCB references are biasing the LM399 Zener section with a 7.5K ohm resistor from the input +15 Volts which passes thru a series protection diode which also feeds the heater circuitry. If this is correct it seems like the heater current draw will affect the zener current as will temperature due to the diode drop temperature dependancy and the heater current temperature variation. Aren't most uses "bootstrapping" the LM399 from the 10.00 volt op-amp output to reduce the effects of input voltage and other factors?

After a few months continuous power, the couple LM399's we have from DigiKey read a stable ~7.04 volts with a new KS34465A and a couple 34401As. Earlier had an LM399 from eBay and it was not stable and read under 7 volts, sold as new but had solder on the leads

Best & Happy Holidays,

[Kleinstein]

The circuit (as far as one can judge from the pictures) looks indeed odd, not like a proper reference circuit.
The output is directly from the OP -  so excessive capacitance can make it oscillate.

The normal way is to take the current for the reference from a bootstrapped reference level of some 10 V or so.  This would however need a second OP for the lower voltage settings.

[Trader]

My china lm399 reference after 1.30h of warmup [...] 2.54773 measure 2.46511.

WOW, this is a Huge difference.

[Martin72]

Why using such suspicious Voltage References ? For calibrating Multimeters ?
I let my multimeter calibrate/adjusting external for appx 79 bucks/year, knowing they use a much, much, much more expensive reference for doing it.

[Trader]

Why using such suspicious Voltage References ? For calibrating Multimeters ?
I let my multimeter calibrate/adjusting external for appx 79 bucks/year, knowing they use a much, much, much more expensive reference for doing it.

The Chinese product (LM399) and service (calibration) aren't trustable, but I believe in excellent quality and cost/benefit if using the DMMCheck Plus.

[Trader]

I figure out this chinese LM399 Voltage Reference is MUCH Better as a CURRENT Reference.

If you keep the 10V setup (jumps J1-J2, J3-J4, J5-J6) and probe the jumps J7-J12 you have a CONSTANT 1925.8uA or 1.953mA (the difference maybe is the burden voltage) !!!

I logged it for +20 minutes (uA, and mA). When I started, decreased 0.1 for <1s, but keep stable the same current for mA and uA all the time, never changing, no ripple (on my DMM).

Bellow 2,000 counts is an excellent value for testing all DMMs with the highest resolution.  Different jump combinations also gave other interesting lower stable values.

A cheap and not too bad stable and precise V, mA, uA reference. Awesome.