ADR1001 - Ovenized Voltage Reference System

[iMo]

I expected ADR1000 with resistors integrated on-die, some cheap R2R dual opamp from Analog and a bunch of wires jumping back and forth.
Maybe the package is too small for that. Such design would also need a few MLCCs.

That hybrid solution would be a pretty expensive device these days, I am afraid. They targeted (end 2022) $60, what is the price of an LTZ/ADR1000..

[Kleinstein]

Not having large capacitors is a good point. The standard LTZ1000 circuit uses quite some capacitors, especially in the temperature regulation. Even when integrated on the chip the termal system is relatively slow and ideally would like rather long time constants in the regulation. Getting this with the limited on chip capacitors is tricky. One the other side the heater is using transistors instead of a resistor. This makes the heater much more linear and thus simplifies the regulation.  The LM399 also gets thermal regulation without large capacitors.

For the voltage loop not having a normal OP-amp can even simplify the compensation, as the LTZ1000 system has a gain of some 200 from the transistor that make the compensation with a normal OP-amp tricky. Here the ADC1001 could likely use a simple dominant pole compensation and no large capacitor needed.

For the extra output OP-amps the circuit looks large. In part this could be from splitting it between the temperature regulated part in between the heater to get low drift and the output stage that ideally has some distance to the reference and distribuited heat at the outside.  One should still avoid drawing much current from the OP-amp as the extra heat on the die could disturb the reference. So the position of the extra gain setting resistors at the op-amp is odd. I would really have preferred the resistors connected to the inverting input instead.

[iMo]

An update..
Note:
34401A, 100NPLC, 10M input
7.1.23 powered off for 10minutes
24.1.23 powered off for 20minutes
18.2.23 powered off for 16 hours
5.3.23 powered off for 20minutes
17.3.23 - the first "metro-LAB_A measurement" Vref = 9.999.895 Volt, Fluke 8588A, powered off for 3 hours

[Andreas]

Hmm,

-1.7 ppm in 2.5 kHrs looks better than my ADR1000 with -2ppm/srqrt(kHr)

with best regards

Andreas

[iMo]

Hmm,

-1.7 ppm in 2.5 kHrs looks better than my ADR1000 with -2ppm/srqrt(kHr)

with best regards

Andreas

Sorry Andreas, that is what I see here.. Perhaps the ovenized on-chip resistors make the difference? Anybody else is measuring the ADR1001???

[sahko123]

nobody else will be able to get one for a while so guess not

[iMo]

An update..
Note:
34401A, 100NPLC, 10M input
7.1.23 powered off for 10minutes
24.1.23 powered off for 20minutes
18.2.23 powered off for 16 hours
5.3.23 powered off for 20minutes
17.3.23 - the first "metro-LAB_A measurement", powered off for 3h
18.4.23 - the first "metro-LAB_B measurement", powered off for 1h

Another great day - my first LAB_B measurement (thanks!), this time with a HP3458A and a trusty Fluke 8508A. This time my box spent almost a day in the lab B prior to the measurement, the measurement with thr Fluke 8508A has been provided together with a quick check with a hot 732C (0.6uV uncertainty as per last report) and it proved to be spot on within some +/- half of a uV with my test leads Below some shots while playing with the meters, it has been a great fun
PS: fixed the part numbers

[Dr. Frank]

Hello,
I recommend for the next time when you have access to this lab, to use statistics and voltage reversal to get proper measurements, i.e. including Standard Deviation and e.m.f. removal.
See my example here:
https://www.eevblog.com/forum/metrology/lowest-drift-lowest-noise-voltage-reference/msg4792004/#msg4792004

On my 3458A, I always use its built-in statistics function with 10 .. 16 times NPLC 100 measurements for each polarity, i.e. about 2x 30 secs. 30 secs give the lowest StD of about 2x10^-8, judging from Allan Deviation statistics, and as can be seen in my protocol (e.g. 80nV/7.176V ~ 0.011 ppm StD).

The difference of these averaged measurements divided by 2 gives the mean value of the reference, w/o e.m.f.
Their sum divided by 2 gives the e.m.f. value, so you can estimate, how good your electrical connections are.

Frank   

[iMo]

@Frank - thanks, I'll do next time. I tried the reversal with both (therefore my shots show negative numbers). For precise measurements as you do one needs some time I had not.. My box spent 19 hours there, not me , unfortunately . The goal with the first LAB measurements had been to get some basic numbers and refresh the contacts. Next LAB A/B measurements in summer (this year hopefully) and I will try harder with the built in math too..

[iMo]

Out of curiosity I wrote a short python prog which reads my .csv files, extracts T_amb and raw 34401A voltage, and takes last 5000 samples (NPLC100) of the longer over-night measurements only, thus the measurements are evaluated upon samples acquired during the very morning while the meter has been warmed-up for at least 6-8 hours. Below the results for comparison, together with the LAB A/B values. I will know more after the second LAB A/B measurements..
PS: the original graph I've been posting here contains some shorter measurements (not good) which made the January pretty hairy. Also I do there adjusting for gain/offset/TC_dmm, but the difference is rather small - taking in account the unknown uncertainty of my measurements 
Edit: removed the TC graph (not related)

[iMo]

I think I have to rethink my power supply in the box. Currently I've been using AC w/ the transformer - I chose a small EI core trafo with split bobbins for primary and secondary, no shielding. It seems to me the capacitive leakage is still there (even I see no diff here with my 34401A - w/ or w/o an additional isolation transformer made of two b2b tranformers). While watching MarcoR'es video on "the extreme isolation" it is clear it affects the measurements and the meters may react differently on a common mode leakage. Most probably I would add a connector on the box for an external Li_ion (5S) bypassing the trafo during the measurements (the trafo will work during aging only).
PS: the video YT/watch?v=9JinSfCKuNQ
PPS: how to add a link to an YT video such it will not start to play here??

[dietert1]

In my ADR1399 setup the transformer is external, as well as recitifier, capacitors and linear regulation pass transistor, as its heat production is affected by mains voltage variations. Opamp and resistors of the supply remained inside the box. Everything in the box runs from a constant 20 V supply voltage derived from the 10 V precision output.
I think the ADR1399 works better with 20 V for the heaters.
I also want to add as backup a lithium ion pack that i bought at the local discounter (meant for handheld drill/screw driver).

Regards, Dieter

[iMo]

I modded the box a little bit - the front panel post for the internal Vcc measurement is now external battery input (via a diode). I made an overnight measurement with 20V ext battery and the temperature inside the sandwich dropped by aprox 1.5C (against the AC powering where the transformer generates some heat). That T_board drop may result in aprox +2.2uV increase at the same ambient. In my box the PSU part (in the back of the box) is in "open air" (there are the vent slots in the top/bottom walls), the reference sandwich is isolated with foam (in the front of the box).

[iMo]

An update:
In order to create more confusion - here are both graphs - with "my TC compensated 34401A" and with "the raw 34401A" data.

[iMo]

An update: with the higher ambient the measurements get more hairy
PS: I've been using +0.5ppm/C for the TC_DMM compensation, the latest "statistics" shows it should be 0.47ppm/C, what is rather small difference considering the behavior of the 399 inside the DMM (the typical random walk of the 399 is within 1ppm during an overnight measurement) 

[Dr. Frank]

iMo,
I think your measurements using the 3458A will be done quickly. Have fun!

Here's a recent stability measurement on the ADR1000, and its Allan Deviation.
That indicates, that a sampling interval of about 20..40sec will give best StD performance for the 3458A.
I.e. NPLC 100, 10..16 samples is sufficient

I will have a closer look on the 3458A stability directly after an ACAL..
Thanks again for the hint to the latest 3458A manual from March 2023, page 77, that 15min. of waiting time after an ACAL DCV should be obeyed.

Frank

[iMo]

Hi Frank!
Thanks for the hints on the 3458A measurements, unfortunately the 3458A is in a permanent action during working hours this summer, thus I've made measurements with the Fluke 8508A only.. 3458A in October/November this year most probably. There is still LAB1 with the Fluke 8588A and with a planned measurement in August this year.
Anyhow, I incorporated your hints into below measurements, at least I hope so 

BTW - I've got an info from ADI the ADR1001 will most probably be released in FEB 2024..

####

A new LAB measurement today - this time in LAB2 and with Fluke 8508A.

The previous measurement with the same 8508A in the LAB2 (April 18th):

Vref =  9.9998697
Vref = -9.9998745  at T_amb=21.1C, T_board=35.49C, AC powered.

This time aprox 5.5uV higher at T_amb=23.0C, T_board=35.52C, battery powered.

Below the results - the measurements were performed in a sequence from "A" to "O".
The green 10xFast measurements - the math has been performed inside the Fluke.

The shot as usual for demo purposes, during the measurements the myADR box was placed about 50cm off the meter.

[Dr. Frank]

Hello iMo,
many thanks for your measurements.
The Fluke 8508A should have very similar noise figures compared to the 3458A.
I don't quite understand, which NPLC number you have used, or what means "SLOW" and "FAST", so to compare your measurement with the NPLC 100 on the 3458A. A NPLC 100 measurement takes 2sec sampling and 2 sec Auto Zero, so about 4.1 sec in total.
20sec each measurement on the FLUKE 8508A, does that indicate NPLC 500 or even NPLC 1000?

The StD values seem to be much too high, I always achieve around StD = 120nV or less on the ADR1000.
On the usual LTZ1000 references, I see slightly higher noise.
Therefore, I suggest that you do an absolute noise measurement.

Btw: I have to build an ADR1001 board as well, as I got one sample here.

Frank

[iMo]

Yep, the 0.6-0.7uV STDEV is in range I get with my 34401A measurements at 100NPLC and over 100 samples (in quiet periods).
There are some results with around 0.3uV - that is definitely off my 34401A capabilities, indeed.
It could be the opamp (OPA189) in my wiring is causing that? Cables and the air drafts caused by the air-condition?

Has to be measured by the experienced voltnuts here who are having those nice LNAs

The NPLC number of the Fluke5808A in "slow" and "fast" settings - hmm, I have to investigate

Slow - ~20secs per sample
Fast - ~6secs per sample - I did it with NRDGS=10, performing 10 samples in series, then it printed out the Mean and STDEV, thus the "measurement" was 60secs in total.

[Dr. Frank]

Yep, the 0.6-0.7uV STDEV is in range I get with my 34401A measurements at 100NPLC and over 100 samples (in quiet periods).
There are some results with around 0.3uV - that is definitely off my 34401A capabilities, indeed.
It could be the opamp (OPA189) in my wiring is causing that? Cables and the air drafts caused by the air-condition?

...

Slow - ~20secs per sample
Fast - ~6secs per sample - I did it with NRDGS=10, performing 10 samples in series, then it printed out the Mean and STDEV, thus the "measurement" was 60secs in total.

So the FAST measurement is comparable to the NPLC 100 measurements with the 3458A.
You'll always see a convolution of noise from the ADR1001, the OpAmp and the 8508A noise.
Your OpAmp will presumably not contribute that much, as well not the FLUKE 8508A.
I think that the ADR1001 itself is more on par with the LM399, than with the ADR1000.. don't know why.

I'll have to build my sample asap, to find out.
Frank

[iMo]

Fluke 8508A in DCV:
Normal mode (aka "slow") at 8.5 digits resolution is 16x64PLC   (25secs)
Fast mode at 8.5 digits resolution is 4x64PLC  (6secs)

[iMo]

Below you may see the schematics of the output buffer as is used today - V1.
The critical parts I see are the R1=10k, and the missing opamp's feedback blocking (like 1k+1n).

I plan to make changes in it - see the new V2.
Not sure whether it could help with the noise, however.
Anyhow - I will change the R1 to 1k, add the feedback RC (R5, C7), add the R6 and remove the tantalum 1.5u.
Also what opamp to use?
Any hints would be welcomed..

PS: the TVS is P4KE12A (next time perhaps P4KE15A, to be not so close to the 10V)..

PPS: also thinking to lower the oven temperature, to something like 45degC..

[iMo]

The sim shows something like +0.35C oven temperature per kiloOhm on the lower end (RTEMP wired to the 6V6). It looks like no special low TC resistors are needed there..
Mind it is a model only.

[Andreas]

PS: the TVS is P4KE12A (next time perhaps P4KE15A, to be not so close to the 10V)..

Hello,

I would go for a stand off voltage of 11 V (so either a P4KE13A or a PTVS11VP1UP)
see here:
https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg2372532/#msg2372532

TVS voltage should clamp @20-50 mA below supply voltage of the OP-AMP in case of a accident with back to back operation with a calibrator.
Hopefully R5 limits the current in this case.

with best regards

Andreas

[iMo]

The question is whether the TVS diode is not starting to fire (thus generating noise) when so close with the stand-off voltage to the 10V.
A good topic for a measurement when you have the LNA etc handy..