ADR1399 reference

[Kleinstein]

A seprate generate 5 V reference could cause an issue if the reference voltage changes. So far the reports for the ADR1399 are more on the lower half for the voltage.

The graph for the 1 PLC 0 V measurement looks like quite some values are off to both the positive and negative side. So a histogram that does not looks like it would have 3 peaks. The extra noise with 1 PLC compared to 10 PLC points to an issue with the auxiliaray (µC internal) ADC.
So maybe the auxiliary ADC scale is not adjusted with the normal DC cal.
The run-up part allows for 4 cases (2 each for the input and zero reading). A scale factor for the auxiliarx ADC that is off would separate the cases.
One may get an even simpler picture from the 1 PLC non AZ mode for a histogram.

[iMo]

..in my 34401P the 5V_Vcc is 4.902V and the voltage at the +5V_REF divider before the 706 is 4.931V (I have not measured DC at the MCU's +5V_REF as it is jumping there a lot). With the original 399 it would be aprox 4.96V.

In the 34401 the +5V_REF "follows" the x399 voltage..

I have not seen a difference here in the noisiness before/after "adjusting" the 10V range when replaced the x399.

With the constant +5V_REF in case of the 34970 the AUX_ADC readings might be say -XXmV (? how much ?) off when the HP399 is replaced with a 1399 (assuming typically the 399=7V and 1399=6.9V).

So maybe the auxiliary ADC scale is not adjusted with the normal DC cal.

@Alex: did you try to always adjust the 1V range you are using after the x399 replacements?

PS: the good message might be you may fix that potential issue simply with a 100ohm resistor as it is done in the 34401, thus make the AUX_ADC "ratiometric".. 

[Alex Nikitin]

Yes, it appears there is at least some correlation with the reference voltage from my measurements. I did calibrate the DCV after two of the replacements, with no obvious effect on the noise though. Need to look at this again, probably in a little while as I don't have time right now  .

Cheers

Alex

[Kleinstein]

It may need and extra calibration step to adjust the scale factor for the µC internal ADC. There is the extra step for the +10 and -10  V linearity correction. There is a way to read the parameter:
https://www.eevblog.com/forum/testgear/hp-agilent-34401a-hidden-menu/msg4636828/#msg4636828
But I don't know which way to get it adjusted - a kind of factory calibration call.
With the µC ref. set from a separate zener, one could just trim that voltage the analog way, to keep the same ratio to the main reference or to get minimal noise for the 1PLC and faster modes.

[iMo]

It may need and extra calibration step to adjust the scale factor for the µC internal ADC. There is the extra step for the +10 and -10  V linearity correction.

I did the "special -10V" adjustment (the step needed after replacing some ADC parts or messing with cal memory) couple of times, hoping it makes the "34401P anomaly" smaller (SM p.82). It did not help.

It is described in the service manual p.85 for 34970 and it is more complex than with 34401, because of the multiplexer.

[Alex Nikitin]

https://www.eevblog.com/forum/metrology/agilent-34970a-(34401a)-linearity-adjustment/msg6330590/#msg6330590

Cheers

Alex

[Alex Nikitin]

Now I have to measure this lot    . 

Cheers

Alex

[exe]

Where did you get it? And, sorry for derailing the thread a litle bit, is there a cheap source for adr1399/lm399? I couldn't find much( I think the absolute best price I found is lm399 for like 20euro/piece.

[Alex Nikitin]

A fortunate eBay BIN. I was pretty convinced I’m buying a lot of Chinese pulls with welded legs, they certainly looked like it on the photos  , however the price was irresistible, I better not to say it.
Received the lot today, looks completely legitimate, measures too, two date codes, both from 2015 (if my memory is right, in 2015 the LM399H was priced around $10 a piece. My only worry now is these are rejects from someone’s selection process. We shall see, I’m measuring first chips right now.

Cheers

Alex

[iMo]

..do wire 16 or 20 of them in parallel and you will get a much better reference than a single 1399 (except the 0.5A heater)..
Everything else would be wasting the time, imho..
 

[Alex Nikitin]

..do wire 16 or 20 of them in parallel and you will get a much better reference than a single 1399 (except the 0.5A heater)..
Everything else would be wasting the time, imho…

That thought DID cross my mind ! It is not so much 400mA continuous, but 3A start-up current that bothers me. In any case I need to measure all chips and see if there are any really horrible ones.

Cheers

Alex

[iMo]

..you may limit the startup current, from my experience with limiting that current (ie some 30-40mA w/ the 723 w/ "a fold back" I did) it takes much longer the 399 heats-up fully, like several minutes..
So perhaps 4-5 groups, each group powered from a dedicated voltage regulator set to, say, 100-150mA limit..
PS: the 78L15 has 100mA DS current limit (in reality it allows a bit more, like 150mA for some time till overheats)..

[Alex Nikitin]

I am thinking about switching the heaters in groups, that might be more efficient and faster to reach the right temperature.

Cheers

Alex

[bsw_m]

Some ADR1399 for experiments )

[exe]

Meanwhile, I closed the window in the lab. This further improved the noise.

Now, according to AI, with linear temperature compensation, the voltage is within +-1uV about >94% of time. Though there is only one outlier about 6 days ago.

I tried to analyze the outlier, but I can't find any correlation with environmental factors. It does look unusual. Like, I can see what is "normal" noise, and normal drift, that dip doesn't fit it. It also could be that temperature correction has its limits. I think I should start a separate thread for that.

[iMo]

..an occasional jump +/- 1-2uV up or down is normal with 1399, afaik. What outlier do you mean? How do you make the temperature compensation?

[exe]

What outlier do you mean? How do you make the temperature compensation?

On the plot there are two curves: raw voltage, and voltage with applied ~1.9 uV/C correction that I called "temperature compensation".

By outlier I meant that dip that I marked on plot attached.

[dietert1]

It's easy to get a 1 uV deviation from thermal EMF, even when using low thermal EMF connectors and copper wires. In the same sense: There are temperature effects proportional to the speed of temperature change.
Trying to work at < 1 ppm requires a lot of patience, really.
Our ADR1399 is a double oven. Maybe one day in the future i will replace its ADR1399 evaluation module by a 6x ADR1399 array - parts are there. I once made a 6x LM399 array with good results.

Regards, Dieter

[Andreas]

Hello,

PCB#3 (with relay multiplexer) now passed 17 kHrs of operation time.

see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg6302962/#msg6302962

The LS8 devices have low drift except LS8#12 which developped a -1ppm step.
The metal can devices are rather stable with a slight negative slope except KHZ#8 which again developped a -1ppm step.

with best regards

Andreas

[Andreas]

Hello,

PCB#2 (with CMOS multiplexers) now passed 27.4 kHrs total operation.
see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg6302982/#msg6302982

the LS8 devices are very stable on continous operation

Nothing new for the metal can devices stable behaviour except KHZ#1 which drifts -3..-4 ppm/year

with best regards

Andreas

[exe]

Here is my report after 4k hours on my single adr1399 that I have. It was testing my patience to wait that long   

There was a permanent jump +2uV from when I put a mains cable next to the reference  Not sure what to conclude from this. Like, keep your references shielded 100% of time? Or, adr1399 is not good for 1uV accuracy?

So, if not disturbed, with temperature correction the whole system is stable to +-2uV 99% of time, at least for last 43 days. But that jump... Anyway, I conclude that my adr1399 and my dmm7510 are fine not drifting significantly. Good-enough for me.

[CalibrationGuy]

Here is my report after 4k hours on my single adr1399 that I have. It was testing my patience to wait that long   

There was a permanent jump +2uV from when I put a mains cable next to the reference  Not sure what to conclude from this. Like, keep your references shielded 100% of time? Or, adr1399 is not good for 1uV accuracy?

So, if not disturbed, with temperature correction the whole system is stable to +-2uV 99% of time, at least for last 43 days. But that jump... Anyway, I conclude that my adr1399 and my dmm7510 are fine not drifting significantly. Good-enough for me.

When you get to ppm and sub-ppm measurements, you need something like a well aged 732 with calibration history as a reference source. How are you compensating for drifts in your measurement setup? If the thing you're measuring, and your measurement setup both drift in the same direction, you will not notice an absolute drift.

I'm just curious as to how confident you are of the measurements you're taking, and your estimation of error sources.

TomG.

[Andreas]

There was a permanent jump +2uV from when I put a mains cable next to the reference  Not sure what to conclude from this. Like, keep your references shielded 100% of time? Or, adr1399 is not good for 1uV accuracy?

Hello,

shielding alone is not sufficient.
You need also some filtering in EMI related environment. (Supply and Output lines!!!)
Otherwise you get demodulation effects (offset voltages) from diode junctions.

Is the reference buffered?

with best regards
Andreas

[exe]

You need also some filtering in EMI related environment. (Supply and Output lines!!!)
Otherwise you get demodulation effects (offset voltages) from diode junctions.

Is the reference buffered?

I agree about EMI. The reference is not buffered. Perhaps, adding a cmos opamp such as opa189 as output buffer would make it even less sensitive to EMI?

If the thing you're measuring, and your measurement setup both drift in the same direction, you will not notice an absolute drift.

I'm just curious as to how confident you are of the measurements you're taking, and your estimation of error sources.

TL;DR: based on data I believe my instruments are stable, and measurements are repeatable, but I don't know the absolute voltage.

Are saying that there is a reasonable probability that a 10+ years old dmm and a 1 y.o. reference have been drifting perfectly in sync for 4000+ hours? I'm not buying that.

I'm confident that my tools are well in spec, yes. I'm not saying my measurements are accurate, or that there is no hysteresis. I'm saying, based on the data, that my measurements they are reproduceable: voltage doesn't drift much with time, predictably changes with temperature, reconnecting leads didn't change measurements, cycling power on reference didn't change voltage (although sample is low, so I don't count on that).

I also have 1month of ageing data (after 1 month of pre-ageing) of my adr1399 measured by another DMM7510 (I asked the seller to do measurements before shipping it to me), it was stable, with noise 4 uV p-p (30 measurements total, once a day, 1uV resolution).

Laying wires differently may shift voltage in my setup. At least it was the case before shielding. My wires are also cheap. So, EMI and thermal EMF are the weakest spots in my setup, and I don't have any calibration certs. Therefore, I can't accurately measure absolute value.

I can't estimate error contribution of individual components, all I can say is that the tempco of this setup is ~+1.5 uV/C, validate roughly over 20-35C room temperature.  The noise you can see on the plots. I also predict that my dmm is 6-10uV off (at ~6.9V) from the dmm that was used to calibrate adr1399 because we measure the same value, but my DMM7510 has the fan replaced, causing internal temperature to rise by 4-5C.

Since I don't want to continue the current measurement (4k hours is enough for me), I can experiment a little bit. Like, power-cycling instruments, leave them unpowered for a while, change shielding, change polarity, etc. Readers suggestions are welcome.

[Andreas]

Hello,

even then I recommend the buffer cirquit shown here and a common mode choke:
https://www.eevblog.com/forum/metrology/emi-measurements-of-a-volt-nut/msg2684100/#msg2684100
(measured with an LM399)

with best regards

Andreas