LM399 drift tendency with age (which direction)?

[kj7e]

Apologies if this has been covered 9,475.3 times already, I really did search both google and here.

The question, does the LM399 have a tendency to drift up or down in voltage with age and heat cycles?  Or is it completely sample dependent?

The reason for the question, I have a band new 34465a that tends to read 8-10ppm low, so I'm wondering if it may have been offset calibrated to allow for aging.  I know this is well within spec, but curious non-the-less.

[Edwin G. Pettis]

I have a number of certified drift LM399AH-50s given to me by Bob Pease, they all have similar drift characteristics over 1,000 hours.  The first 336 hours, they all drifted up 20 - 65 PPM, then drifted downward towards the 1,000 hour figures which varied somewhat within the specified 50 PPM drift spec.  One drifted back down to its original starting voltage, three drifted down around -40 to -45 PPM.  They all drifted downward from their initial upward drift and in general they tend to stabilize with a slow downward drift as they get old, often changing under 1 PPM/year.  The records ended at 1,000 hours.

[Andreas]

Hello,

perhaps you have only a version with a buggy firmware:

https://www.eevblog.com/forum/testgear/keysight's-new-34465a-(6-5-digit)-and-34470a-(7-5-digit)-bench-multimeters/msg775998/#msg775998

They should have pre-aged the reference in a way that it will not drift the max stated value during the first year.
The drift consists of 2 drifts: heater setpoint temperature drift together with tempco of the zener. And the zener drift itself.
I pretend: it is not possible to predict the total drift (except when measuring it).

Franks 34465A was spot on when he got his replacement.

with best regards

Andreas

edit:
https://www.eevblog.com/forum/testgear/keysight's-new-34465a-(6-5-digit)-and-34470a-(7-5-digit)-bench-multimeters/msg889215/#msg889215

[Dr. Frank]

The question, does the LM399 have a tendency to drift up or down in voltage with age and heat cycles?  Or is it completely sample dependent?

The reason for the question, I have a band new 34465a that tends to read 8-10ppm low, so I'm wondering if it may have been offset calibrated to allow for aging.  I know this is well within spec, but curious non-the-less.

Virgin LM399 tend to drift randomly, I've salvaged Joe Gellers measurement on 78 EA National LM299, and that does not reveal any special tendency. See picture.
If LTs 399AH behave the same, I can't tell.
The pre-aged and monitored references in Keysights DMMs should be stabilized already, to meet the specified drift rate.

I measured several 34465A, freshly from manufacturing and calibration, and in the 10Vdc range, they all were spot-on (1ppm), after delivery.

You need to use ACAL, of course.

With FW 2.14, the 100 and 1k DCV ranges should also read within a few ppm, but no 10..15 ppm deviation any more, as in earlier versions.
My latest 34465A still is spot-on in the 10VDC range, after 1 1/2 years of use.

So, please check FW version, ACAL, and calibration report and its date.

Frank

[kj7e]

Thank you Andreas, I read the entire Keysight 344(65|70)a thread and the Voltage de-rating topics before my order, its a brand new unit with FW 2.14 and a cal date of 27-Jan-2017.  I recall Franks FW bug findings, the FW version was the first thing I checked.

Perhaps its my 34465a that's correct and my other devices that are slightly off, but that seems fairly unlikely three of four devices will be off exactly the same amount.

I have a 1 month old Siglent 3065x, 1 month old Doug Malone 10v reference, 2 week old Ian Johnston PDVS2 (LM399A version) and now a 3 day old Keysight 34465a.  Both references and the Siglent (of all things) read identical. Its the Keysight 34465a that's the odd ball. Yes, I did have to use the ACAL after the first warm up as it read considerably low by more than 35ppm, this brought it up to where it is now, but still disagrees with my other devices by about 8ppm.

Edit: Ignore the "error", it was caused by me trying to go into the FW menu with the security lock on.  All devices powered on for a min of 12 hours, room temp 73-74 deg F and stable.




I know I'm splitting hairs here and its still well within its spec, just a bit disappointed it does not agree exactly with the other devices.  One thing is for sure, it will be very expensive to try get anymore precise.

[GigaJoe]

I made a circuit with lm399 a brand new one , 10V output , but output growing steady +10 µV per day, the thing is, it steady going up for about a month, should i consider this particular LM faulty or wait a few weeks ....

[The Soulman]

Have all influencing factors been stable for that month? (temperature, humidity, supply voltage,....)
If so, also log the LM399 output voltage for a while (aprox. 6,9 Volt) if that is stable than its probably
the gain resistors that are drifting.
More info on your circuit, components used and measurement set-up would help.

[Andreas]

I made a circuit with lm399 a brand new one , 10V output , but output growing steady +10 µV per day, the thing is, it steady going up for about a month, should i consider this particular LM faulty or wait a few weeks ....

Only the 10V output or also the 7V raw voltage?
perhaps also your DMM is drifting.
I have also some of my ADCs drifting since it is getting warmer and thus more humidity (absolute) in the air.

with best regards

Andreas

[Muxr]

Could be the probes. Try making your own shorter probe with some twisted pair. I've found noticeable difference in readings between different probes I've used, twisted pair really helped me eliminate a lot of the interference I was having.

[GigaJoe]

well, not ...

that was 2 days ago, i did adjustment , yesterday was 10.00000  , today +1; tomorrow will +2 .... and so on .... that going on constantly for 3 weeks already.

I'm wonder will it stop...
And yes it LM; the LM output also rising exact as output on opamp, no differ it I change opamp, it just shifting Vout due offset.

probes in fact made of, CAT5 shielded cable,  so it already twisted and shielded ...
100 and 10 plc  -same numbers, I assume not much noise in overall.

temperature in basement pretty steady, +5 Celsius  give around +2-3 ppm; but that unusual (when outside over +30C), it always +20

The fun thing, adjustment pot is set to '0' i have no room to adjust it (initially I calculate 200 mcV room)

[Andreas]

Hello,

Ok 1 ppm/day drift is rather much.
But on the other side 21 days are much less than the first 1000 hours.

When I look through my measurements then LM399#CH7 needed nearly 300 days to stabilize.

Of course there may be other reasons.
- A bad socket for the reference.
  (on one LM399 I recognized that the heater was not connected -> how is the power consumption?).
- Bad stabilisation for the heater voltage.

Up to now I have only one continouosly drifting LM399.
But this one was not a "new" but a "used" one with unknown history.

added some other LM399s. (all with sockets so results may be worse than without sockets).
at day 80 (~2 kHrs) I pulled and reconnected the LM399 to the sockets and re-started drift measurement.

So its always good to build at least 4 devices to select the best from.

with best regards

Andreas

[ebclr]

Do have any sense to make a system with N pieces of LM 399 on averaging configuration to have better long term stability?

[Andreas]

Hello,

perhaps.
but you have to sort out the "stinkers" first.
So pre-ageing+noise measurement is a must if you do not want only the "datasheet values".

with best regards

Andreas

[GigaJoe]

ok .will see in a month ... July 4 ( if aliens not show up )

the construction I made is 2 independent circuit one is LM+ curr resistor , all soldering on air (no board) and wrapped in thick styrofoam, then independent wires for heater and LM ; going to the opamp board which is again in styrofoam, that 2 bricks wrapped in the aluminum foil, grounded to common ground point, transformer - bridge caps, LM+15 volt - output to heater; then LM +12V caps - power for opamp , opamp + transistor comm. emmiter, to unload opamp, is feeder of LM and output .....

I bought 10 , one blow up , second drifting, thinking just hook all that left under some voltage and left for couple month .... problem to trace changes need extremely precise voltage output to feed zener, due 1:1000 feed voltage sensitivity, ( in my practice) , I mean 1 millivolt of feed voltage, translated in 1 microvolt on zener , due to current change.   A reason why i combine resistor and zener together , heat stabilizing resistor a bit. ( Again this statement may not true , I may messing something in my head)

but but ... i just realize .... i can be current resistor that drifting .....

[hwj-d]

@kj7e Hello,

why not measure 24h drift from one piece to others. Set statistics on, set it to Trend, show All and set vertical manual Center and Span, and let it run from the beginning with 10 and 100 plc , to be clearer, whats happen there in your Lab.

As i see, you have your temp measurement already there. Take away all unnecessary switching power supplies, also from constant voltage drill your probes and switch off all led and fluorescent tubes, (edit:) Wlan, Bluetooth and Cellphones.

After that, you get something like this on my (terrible) pictures, where i play around with my shielded (not optimized) REF5020. What you see, of course, is only the drift behavior from each one of your combination. Don't know, this is possible with this Siglent too.

I mean, your Keysight tell you the truth, if it is ok. That all others show the same, over time, incl. the last digit, don't know, that could be right.

[kj7e]

@kj7e Hello,

why not measure 24h drift from one peace to others. Set statistics on, set it to Trent, show All and set vertical manual Center and Span, and let it run from the beginning with 10 and 100 plc , to be clearer, whats happen there in your Lab.

As i see, you have your temp measurement already there. Take away all unnecessary switching power supplies, also from constant voltage drill your probes and switch off all led and fluorescent tubes.

After that, you get something like this on my (terrible) pictures, where i play around with my shielded (not optimized) REF5020. What you see, of course, is only the drift behavior from each one of your combination. Don't know, this is possible with this Siglent too.

I mean, your Keysight tell you the truth, if it is ok. That all others show the same, over time, incl. the last digit, don't know, that could be right.

I sent one of my 10v references back to Doug Malone with my measurements for him to verify.  Once it returns Ill run some more tests and have a better idea of whats going on.

I'm also building an LTZ1000A reference, just waiting on the VPG resistors.  Once the LTZ1000 ref is built and burned in Ill look at sending it out for measurements.

[IanJ]

Hi all,

Must admit, when I first started dev of the PDVS1 I noticed more than anticipated tempco/humidity related drift. Back then I was using a brand new 34465a only and as it turned out it was the 34465a that was drifting as I built a temperature chamber and ran the PDVS1 inside it and varied the temp both in terms of varying the internal temp and holding the internal temp/varying the w/shop temp and proved it had to be the 34465a. Then, when I got my 3458a I confirmed it (I run it in a much more controlled temp room).
But, as I never use that meter these days for final calibration/testing I never got around to looking fully into it........it's a must do sometime.

Ian.

[Dr. Frank]

Hi all,

Must admit, when I first started dev of the PDVS1 I noticed more than anticipated tempco/humidity related drift. Back then I was using a brand new 34465a only and as it turned out it was the 34465a that was drifting as I built a temperature chamber and ran the PDVS1 inside it and varied the temp both in terms of varying the internal temp and holding the internal temp/varying the w/shop temp and proved it had to be the 34465a. Then, when I got my 3458a I confirmed it (I run it in a much more controlled temp room).
But, as I never use that meter these days for final calibration/testing I never got around to looking fully into it........it's a must do sometime.

Ian.

Ian,

did you use the ACAL function of the 34465A?

This function is not as capable as the 3458As, but should bring back 10V reading to about 1ppm, when temperature changes by about +/- 5°C (my experience / estimation).

Under which conditions did you identify the 34465A to drift, i.e. how did you determine the environmental temperatures?

Frank

[kj7e]

Hi all,

Must admit, when I first started dev of the PDVS1 I noticed more than anticipated tempco/humidity related drift. Back then I was using a brand new 34465a only and as it turned out it was the 34465a that was drifting as I built a temperature chamber and ran the PDVS1 inside it and varied the temp both in terms of varying the internal temp and holding the internal temp/varying the w/shop temp and proved it had to be the 34465a. Then, when I got my 3458a I confirmed it (I run it in a much more controlled temp room).
But, as I never use that meter these days for final calibration/testing I never got around to looking fully into it........it's a must do sometime.

Ian.

I think you may have meant your 34461a, almost the same as the 34465a but the 61 lacks the ACAL tempco correction.

[Alex Nikitin]

Has anyone observed this phenomena on his 34465A/34470A unit as well ?

It is present on the 34465A here, quite small, just under 1ppm, but nevertheless noticeable.On the meter here it appears to be a scaling error, i.e. on +10V it drifts 5-7uV down after ACAL, for example from 10.000018V to 10.000011V over 2-3min on the trend display, however on -10V it drifts up about the same amount, i.e. from -10.000018V to -10.000011V.
Cheers

Alex

P.S. - here is a sample of that drift. The voltage source (LTZ1000, not boxed) is measured simultaneously by the K34465A and HP3458A, the HP meter at NPLC 10 is triggered from the Keysight which logs at 1sec intervals. The drift is present as the change in the difference between two measured values, so it is about 0.7ppm. The scale is 0.5ppm/div. The measurement was started immediately after ACAL on the Keysight meter has finished.

[hwj-d]

Thanks Alex Nikitin,
but, wow, that's pretty good for a 6.5 digit 1.5k$ meter against the >8k$ reference 8.5 digit meter. The HP3458A is drifting too about 0.4ppm over this time, and the 34465a has only +0.5ppm difference.
What more do we want from this lm399 keysight meter?

[Kleinstein]

The drift shown for the 3458 is the combined effect of the reference measured and the meter. It is expected that about half (or maybe more because of missing case and aging) of that drift is the external source, not the 3458.

It might be interesting to see if that effect after the ACAL on the 34465 is repeatable. Given that there are numbers for the positive and negative readings, this seems to be the case. So I won't expect it to be a random drift of the reference.  A possible source for this effect could be something like dielectric relaxation or some software correction used for that. It might be that a similar effect occurs just after switching the range (e.g. go from 100 V range to 10 V range).

[hwj-d]

The drift shown for the 3458 is the combined effect of the reference measured and the meter.

Yes, that's true. But the measurement shows and are in the same phys. situation. So, the relation from the devices to each other should be correct. Correct me, if i'm wrong? 

[IanJ]

Hi all,

Must admit, when I first started dev of the PDVS1 I noticed more than anticipated tempco/humidity related drift. Back then I was using a brand new 34465a only and as it turned out it was the 34465a that was drifting as I built a temperature chamber and ran the PDVS1 inside it and varied the temp both in terms of varying the internal temp and holding the internal temp/varying the w/shop temp and proved it had to be the 34465a. Then, when I got my 3458a I confirmed it (I run it in a much more controlled temp room).
But, as I never use that meter these days for final calibration/testing I never got around to looking fully into it........it's a must do sometime.

Ian.

I think you may have meant your 34461a, almost the same as the 34465a but the 61 lacks the ACAL tempco correction.

Err, yeah I meant my 34461a.............thanks for keeping me right. I'll blame burning the candle at both ends.....

[Alex Nikitin]

The drift shown for the 3458 is the combined effect of the reference measured and the meter. It is expected that about half (or maybe more because of missing case and aging) of that drift is the external source, not the 3458.

It might be interesting to see if that effect after the ACAL on the 34465 is repeatable. Given that there are numbers for the positive and negative readings, this seems to be the case. So I won't expect it to be a random drift of the reference.  A possible source for this effect could be something like dielectric relaxation or some software correction used for that. It might be that a similar effect occurs just after switching the range (e.g. go from 100 V range to 10 V range).

1) Most of the drift on the HP3458A graph is that of the reference itself. It is not very stable and uses cheap resistors. It has about 0.5ppm/C tempco and I've used it mostly for a low noise here. Unfortunately, at the time I've done that measurement, there was an open door in the lab and a bit of draught, so the voltage did drift. The HP3458A just back from repair and full calibration at Keysight UK, as it did develop a short-term instability. Right now it is quite stable on voltage measurements, CAL?72 changes between two ACAL ALL in about a week less than 0.1ppm.

2) Yes, it is not a random error. I've noticed that drift after ACAL on the K34465A some while ago, however as I know about it and it is reasonably benign (<1ppm), I can live with it. I did not try to actually measure it till today though, as normally I would just wait 3-5min after ACAL for the meter to settle. It behaves in a similar fashion on the resistance ranges too (noticed it on 10K measurements - will try to run a similar measurement next time I'll bring the Fluke 742A-10K to the lab).

Cheers

Alex