ADR1399 reference

[CalibrationGuy]

Andreas, I wonder what would happen to those references in the ceramic package if you turned them on for a day, then turn them off for a day and then turned them on again for a day for a few months. Do you think the hysteresis drift would lessen?

When I get around to building those boards, you can let me know if there's anything you want me to test (thermal cycling, power cycling, etc.)

TomG.

[Andreas]

Hello,

I think that the hystersis on the LS8 package is happening by the humidity reaction of the EPOXY of the PCB.
Here we have time constants of several (2-7) days. So a short interruption of one day or over the weekend will not have such a large impact.

I have no preferences for your tests. The datasheet should give you enough ideas. (no one has tested dynamic resistance up to now imho).
Your first goal should be to get stable values (over time).
After a few hundred hours the references should be stable enough to begin T.C. measurements.

With best regards

Andreas

[JodelJonny]

Hello,

sorry I never managed to post my KiCAD and gerber files. Here they are.
Feel free to use them as you like. If you improve the design, it would be great if it was made available to everybody as well.

The layout works, I get 10 V and no smoke, apart from that it is very possible that there are errors (wrong footprint for example).

Thanks for all your help and advice 

Since my last post I have build a second one. It has run a few khours but still needs a trim resistor to get close to 10 V.


Best regads JodelJonny

[Kleinstein]

With quite some difference between the hot and cold state there can also be hysteresis from thermal stress. The humidity could add to the stress, though from the tendency the other direction. Solder can creep / deform, especially when hot, but also at room temperature.

For the humidity relaxation the time constant also depends on the temperaure. Expect a much shorter time scale when hot. So maybe 2 days hot and 2 weeks cold.

[Andreas]

Hello,

I finally managed to do the T.C. measurements for PCB#3.
On the first run after ~40 hrs for the 4 LS8 devices I had no luck.
I recognized too late that I had forgotten to activate the temperature compensation of my measurement ADC.
(It was switched off for the ratiometric T.C. measurements of the resistance trimmers).

So I had to repeat the measurement with activated T.C. compensation.
I also had bad luck with the TO52 devices. One of the relays had a bad contact and the voltage variation showed several 10 uV during measurement for certain temperatures. I tried to do some cleaning (current) but had no luck. After exchangeing the relay the galvanic isolation (photocouplers) of the RS232 communication failed. So I had to do another run...

Finally the results:
T.C. on PCB#3 is between -0.008 ppm/K and -0.213 ppm/K near room temperature.
The best is a TO52 device and the worst a LS8 device.
The table shows all measured ADR1399 results and the 2 results of LM399 on PCB#1

with best regards

Andreas

[Andreas]

Hello,

after T.C. measurements I can now supply temperature corrected measurements.
compare to the diagrams from here (which were uncompensated and so more noisy).

https://www.eevblog.com/forum/metrology/adr1399-reference/msg5726673/#msg5726673

with best regards

Andreas

[ITArchitect]

Your graph shows ADR1000 for that square plot.

[Andreas]

Hello,

the ADR1000 (now stable) is only for sanity check of the measurement.

with best regards

Andreas

[Andreas]

Have you looked if the LCC or TO46 package have any difference in 1/f noise?

Ivo asked for noise behaviour between LS8 or TO46 package.
So I did noise measurements. Here PCB#3 (with relay multiplexer).

First the noise floor together with a 2 LiIon (9V) battery on the input of the LNA.
124 nVpp is low enough.


then the 4 LS8 devices and the 4 TO46 devices.

Generally you can say that the noise is very individual (independant of package).
Most devices are over the data sheet spec of 1.44 uVpp.

Even if you take into account that I am doing 100 sec measurements which give a slightly higher noise (statistically more extreme values)
than the 10 second measurement duration which is minimum duration for the 0.1-10 Hz range.
I also use the longer duration and 19 measurements to catch popcorn noise and calculate some average noise values + stray.

Most devices have some visible popcorn noise above 0.5uV (up to 2 uV measured).
Some typical graphs are shown.

with best regards

Andreas

[Andreas]

Hello,

and also for PCB#2 (with MAX4051A multiplexer).

again the same picture: very individual noise.
Mostly above datasheet typical value.
So if you need really a low noise device you should select the voltage references for noise.

Since the PCB´s are measured from a (linear) mains supply. The mains frequency is rather large. (could not be further reduced).
Effectively I have -165 dbV which corresponds to 6 nV eff or 20 nVpp. So the influence is low enough for the measurement.

and finally a overview.
only 3 devices meet ~the datasheet spec. (green)
2 further devices are near that but have already some popcorn noise (yellow)
the other devices either have large popcorn noise or some "high frequency looking" noise. (LS8_08 and TO46_02)

with best regards

Andreas

[iMo]

The T046-4 looks nice, but.. try to run it longer, like 1 day for example, whether it jumps or not.. I have such sample (EDIT: it was an LM399) with lowest noise but it jumps 1-2 times per day (jumps to a different level and stays there).

[Andreas]

Interesting,
how high are the jumps and how much samples did you test?
with best regards
Andreas

[iMo]

  My brain fart..  It was LM399.. The jumps were around 5uV..
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg5684641/#msg5684641
My best 1399 is currently in burn-in, no such infrequent jumps so far..

[CalibrationGuy]

Andreas, how many data points are you taking per day for those eight voltage references on your board? (4 LCC + 4 TO46) I assume you take more samples for noise measurements. How many samples/sec, or min, etc. do you take for the noise measurements?

TomG.

[Andreas]

Hello,

one single measurement has 100 sec and 5 MSamples in total ( so 50 kSps / sec)
I am doing 19 measurements (the whole usable buffer of the oscilloscope) in one series.
So I can also do averaging on the FFT outputs. (not inputs).
The 19 measurements are then evaluated for average 1/f noise and AC rms values.

Usually I am doing more than 1 series on one sample to screen for popcorn noise.
So actually one sample needs more time than the 1900 secs for one series.

with best regards

Andreas

[CalibrationGuy]

That's a lot of data. How about the drift data. How often do you sample the 8 references on your aging boards?

TomG.

[Andreas]

Hello,

at the moment I am starting one (automated) measurement per day.
Each of the 8 references is averaged over 1 minute measurement time.
The first 1000 hours I measured twice per day.
During T.C. + noise measurements there is a gap in the ageing drift.

with best regards

Andreas

[Andreas]

Hello,


PCB#3 (with relay multiplexer) now passed 2 kHrs of operation time.
The gap in the measurements is due to T.C. measurements and noise measurements where I had to do a different setup.

Some of the references show a shift after the temperature cyclings.
And all of the LS8 devices changed direction of ageing.

The PCB#3-Devices are shown with drift over linear time (days) and in a 2nd diagram as SQRT(kHrs) as x-axis.
Y-Axis is temperature compensated according to the linear T.C. from T.C. measurements.

see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg5748407/#msg5748407

with best regards

Andreas

[branadic]

Attached is an update of my ADR1399-PWM-DAC reference. The reference was initially measured by Solatron 7081 and Prema 6048 white. At some point the Prema 6048 left and made its way to its owner. Then at some point the 7081 stopped working (at about hour 4700), so I had to take it from the setup and repair it. Fortunately, recapping fixed the problem.
Luckily, another Prema 6048 white showed up at MM2024, stayed with me until I finally became the new owner and this meter was used to measure from hour 4900 on.

Overall, this reference is constantly drifting upwards with no obvious sign of stabilizing.  So continuous observation is still needed.

-branadic-

[Kleinstein]

The drift is still pretty small. The Solatron meter seemed to have been a bit temperature sensitve. Most of the "noise" there seems to be a thermal effect at the meter. With the downward temperature trend for quite some time it is a bit hard to tell if we really see the reference drifting or maybe more a temperature effect at the meter.

[Andreas]

Hello,


PCB#3 (with relay multiplexer) recently passed 3 kHrs of operation time.
The gap in the measurements is due to T.C. measurements and noise measurements where I had to do a different setup.
Y-Axis is temperature compensated according to the linear T.C. from T.C. measurements.

2 of the TO-46 devices showed a jump during the last days. Generally they now seem to stabilize.
This can also be seen when drawing the standard deviation (SDev) over the last 42 measurements (1000 hours).
It is trending towards the 0.1 ppm of the ADR1000 reference.

The direction change of the LS8 devices remains. Here the standard deviation is more or less constant.

see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg5771905/#msg5771905

with best regards

Andreas

[Andreas]

Hello,


PCB#3 (with relay multiplexer) now passed 4 kHrs of operation time.

Most of the devices are stabilizing. (except one of each group (TO-46 and LS8))
Again standard deviation is shown for all devices.

with best regards

Andreas

[Andreas]

Hello,

PCB#3 (with relay multiplexer) now passed 5 kHrs of operation time.

Most LS8 devices still have a drift of ~2 ppm / year
resulting in ~0.2 ppm/kHr std dev.

Most of the TO-46 devices are further stabilizing
std dev here is between 0.1-0.2 ppm/khr (where 0.1 ppm/kHr is my measurement limit shown by ADR1000 comparison)

with best regards

Andreas

[Andreas]

Hello,

PCB#3 (with relay multiplexer) now passed 6 kHrs of operation time.

On day 220 I opened the die cast housing of the PCB to inspect the wiring of the RS232 interface.
with removing and installing again the thermal isolation cloth.

This had mayor influence on the LS8 devices. So also the standard deviation is rising.
The TO-46 devices are not affected by day 220. Although one device has begun to drift again on day 200.

with best regards

Andreas

[iMo]

I wonder how this would work in practice (provided we have quite good 1u and 3u3 foil caps handy)..

PS: with 50G (an example) of parallel parasitic resistance at the foil capacitors the voltage drop from 399 to the non-inverting input of the OPA145 will be 130uV.. not good..