Encasing of RoadRunner LM399 10V reference

[jpb]

I didn't want to put this on the LM399 10V reference thread as it is slightly different, and it isn't about the design anyway.

Having bought one of RoadRunner's voltage references:
https://www.eevblog.com/forum/buysellwanted/deeu-simple-lm399-reference-board-improved-second-version/
a while back, yesterday I finally got around to encasing it.

I included a LT3402 board for the 15V and also a L7818CV to feed it 18V and allow more leeway for the input voltage (the LT3402 doesn't like more than 20V and I might want to use a 24V supply - efficiency doesn't matter much because the current drawer is so small).

I have had it running for about a day now and the Voltage reading for my 2015 varies (after warm up) between 9.999991 and 9.999997 and my 2000 tracks it but about 0.000002 or 3 higher. This is well within the 1day spec and I've not had a calibration for years (the 2000 is most recent and that was in February 2014).

I need to set up an RS232 or GPIB connection and plot the results properly but that will have to wait until I've finished my Oscillator measurements (I'm more into time-nuttery than Voltage-nuttery at present).

Photos show the 15V voltage, the 18V voltage (as checks), the casing and the 10V reference measured on various Voltmeters I have.

[jpb]

further photos...

[jpb]

..the end at last...

EDIT - took it to bits to take better photos and also because I felt the foam I'd previously used was squashing things so replaced it with thinner card so there is no risk of stress, then restarted it (the meters were not turned off).

The readings are higher so dropping as it warms up so got a 10.00000 photo on the way!

[guenthert]

Well, at those voltages, there's hardly a safety concern and you probably don't want to plug multimeter probe cables in, so shrouded sockets are not needed.  Typically one would want sockets which offer a screw terminal (some banana sockets offer this as an additional feature) as one might just want to use spade lugs or bare copper wires in an attempt to reduce the error due to thermal EMF (for that the terminals need to be the 'low EMF' kind).

You find some examples in https://www.eevblog.com/forum/metrology/diy-low-emf-cable-and-connectors/.

[jpb]

Well, at those voltages, there's hardly a safety concern and you probably don't want to plug multimeter probe cables in, so shrouded sockets are not needed.  Typically one would want sockets which offer a screw terminal (some banana sockets offer this as an additional feature) as one might just want to use spade lugs or bare copper wires in an attempt to reduce the error due to thermal EMF (for that the terminals need to be the 'low EMF' kind).

You find some examples in https://www.eevblog.com/forum/metrology/diy-low-emf-cable-and-connectors/.

Thanks for the link. I didn't think that EMF issues would affect measurements at 6 1/2 digits (10uV resolution) but the paper referenced in the link suggest that it may be important. As an experiment I reversed the connection (as per the Fluke paper) and the change in magnitude for the 2000 was only one digit but it was 3 digits for the 2015 - see attached photos.

I used the shrouded sockets for a couple of reasons. The main one is I had them in my spares box left over from another project and a lesser reason is that I have a few power supplies with shrouded sockets and my meters have shrouded sockets so I can use standard test cables to connect them.

I would also normally, have the sockets separated by the standard 19.05 mm but it was difficult to do this neatly so I stuck them in the corners. They are also the opposite way around on either side which looks odd but I did this to match the board which also swaps them, though obviously I didn't need to do this.

Given my very rudimentary metal working skills I was amazed that I actually managed to get the sockets reasonably neatly fitted!

[jpb]

I just discovered by reference again whilst tidying my bench so repeated my measurements one year on.
It now seems to have risen by about 100 uV in a year (or else all my meters are reading high by a similar amount).

Attached are the photos. Perhaps I'll make this an annual event, if I remember.

[TiN]

Sadly, those connectors and overall enclosure only making this reference worse. Many new sources of thermal EMF errors. Much better would be remove that cable connect altogether, and add two pure copper wires, terminated on low thermal binding post such as Pomona 3770. Also I would shift reference PCB to one side of the case, and put regulator and posts on other side. This also would allow to put output posts with standard dual-jack spacing, so you could use standard dual banana cables, if you like.

[Kleinstein]

The setup is not very good with respect to thermal EMF, but for a 100 µV = 10 ppm  change this is likely not the relevant part.
There is some drift in the reference, though supposedly little when a LM399 is cold. Still if not operated much before the starting point one is still in the steeper initial phase. The other point is drift of the resistors for the 7 to 10 V step.

[jpb]

Thank you for the advice.
I think as Kleinstein says, the thermal EMF effects are smaller than the general drift. I did do some experiments last year which I reported in this thread:
https://www.eevblog.com/forum/metrology/why-dont-thermoelectric-emfs-cancel/
and concluded that thermal emf effects were less than the accuracy of my meters.

When I first turn everything on there is quite a lot of drift and I wait for it to settle (the voltage drifts down a few parts in the least significant digit).

I'm not really a voltage nut (more a time nut) so these experiments are not to the precision that they should be. Perhaps I should leave the reference powered up for a long time (like days or weeks rather than a few hours at a time)?

To start becoming a Voltage Nut I'd need to get a 7.5 digit meter.

The ironic thing is that I suspect my meters have much less drift than my reference!

[Kleinstein]

The LM399 references are supposed to mainly drift while they are one, because they run quite hot. So for an initial burn in it can be good to leave it on for some time, like a week or month. This generally slows down the later drift. Later it may be even good to only power it up as needed e.g. for a few hours at a time.

However the largest drift for the reference circuit is likely with the resistors for the 7 to 10 V scaling.  I can't see the resistors on the picture and this kind of indicates they may not be very high quality. The circuit around the trimmer can also be done in a poor way and cause drift - the trimmers itself are not very stable. So the range for the trimmer should be limited.

The meters should be more stable than the shown reference, maybe comparably to the LM399 alone.

[Andreas]

Hello,

I would do a run in phase of at least 200 days with a LM399 reference (see attachment run in phase of two LM399s on 7V output).

But in this case the trimming scheme is not according to precision rules (see old National Semiconductors LM399 data sheet)

with best regards

Andreas

[jpb]

Following Andreas's post, I'm leaving it on for 200 days (will try to - assuming no power cuts etc).

After 6 days (or is it seven?) it has settled down to nearer 10V (as defined by my not-calibrated-for-many-years bench meters )

Here is today's update (my meters are normally around 2 figures in the last place different after warming up - I cheated a bit by taking the photo while they were just 1 different) :

[Kleinstein]

For the burn in the reference does not have to be on all time. For the LM399 it is more like the total on time, not matter if 24/7 or only 1 h a day.

[kawal]

 I made a few of these based on road runners files and have it sitting on the bench for 1.5 years now.  Waiting to stabilize . I get drift on 10s of uV position so not happy just yet.

[jpb]

I'm leaving mine on continuously but I'm not sure I want to tie up one of my power supplies for 1.5 years!
Currently it is still drifting down
As I only have 6.5 digit meters 10uV is the ls digit.
Here is today's reading:

[Andreas]

Hello,

below 10V you can get another digit (1uV resolution) on the display by the math function of the K2000:

https://www.eevblog.com/forum/metrology/getting-one-more-digit-from-a-6-5-digit-meter-without-using-gpib/msg1300529/#msg1300529

with best regards

Andreas

[jpb]

Hello,

below 10V you can get another digit (1uV resolution) on the display by the math function of the K2000:

https://www.eevblog.com/forum/metrology/getting-one-more-digit-from-a-6-5-digit-meter-without-using-gpib/msg1300529/#msg1300529

with best regards

Andreas

Thanks for that tip - here is the latest readings down to a microVolt - its fun to do but I suspect the uncertainty is around 15 microVolts or more.

[jpb]

Here is an update to give another (crude reading).
Posted in response to a question on this thread:
https://www.eevblog.com/forum/buysellwanted/deeu-simple-lm399-reference-board-improved-second-version/msg3487034/#msg3487034
but I wanted to keep all the measurements in the same place.