ADR1399 reference

[dietert1]

Those MLCC caps C9, C10, C11, C14 used in the filter are known for microphony.
Easy to make a low pass filter that reduces high frequency noise and adds low frequency or very low frequency noise at the same time. Some say it's better to leave a high quality zener reference as it is and rely on bandwidth limits further down in the signal processing chain. Or use a reference array in order to reduce noise..

Regards, Dieter

[CalibrationGuy]

+1 for the array approach. I believe Datron did that with the LM399. TiN also did that for his 3458a reference board.

TomG.

[KE5FX]

Better to use whatever capacitors you like, and don't expose the PCB to vibration when making sensitive measurements. 

See also the perennial arguments about BNC jacks and microscope boom arms. 

[Andreas]

Hello,

PCB#3 (with relay multiplexer) now passed 11 kHrs of operation time.

see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg6126387/#msg6126387

The LS8 devices are typically drifting still somewhat more than the metal can TO devices.
The TO device KHZ#8 developped 2 downward jumps of ~1 ppm (popcorn noise?) where I do not have any explanation for it.

with best regards

Andreas

[Andreas]

Hello,

PCB#2 (with CMOS multiplexers) now passed 21.4 kHrs total operation.

see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg6126397/#msg6126397

the LS8 devices (also again #7) are drifting slowly upwards
most of the metal can devices are stable except #1 which is drifting downwards with ~-4ppm/year.

with best regards

Andreas

[branadic]

I've finished a quick weekend project. JodelJonny has sent me two of his PCBs https://www.eevblog.com/forum/metrology/adr1399-reference/msg5729949/#msg5729949, that were already populated long time ago. This weekend they made it into a Fischer case together with a modified linear power supply board and frontpanel assembly that contains a 10:1 divider copycat of W/F7000. At the same time there is enough space and connections and juice in the supply available to parallel another two reference boards. 
The more time consuming part is now t.c. compensation and trimming of each reference board.

-branadic-

[Andreas]

Hello,

PCB#3 (with relay multiplexer) now passed 12 kHrs of operation time.

see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg6126387/#msg6126387

The LS8 devices are typically drifting still somewhat more than the metal can TO devices.
No further jump on TO device KHZ#8.
Unfortunately one of my ADC´s (ADC#16) developed a ~1ppm upward jump in readings on day 476
which results in a ~0.3 ppm shift on all averaged ADR references. (averaged readings of 3 ADC´s)

with best regards

Andreas

[Andreas]

Hello,

PCB#2 (with CMOS multiplexers) now passed 22.4 kHrs total operation.

see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg6126397/#msg6126397

the LS8 devices are still drifting slowly upwards
Metal can device #1 now seems to stabilize.
The 0.3 ppm jump resulting from 1 ppm ADC16 drift is visible on day 908.

with best regards

Andreas

[Andreas]

Hello,

PCB#3 (with relay multiplexer) now passed 13 kHrs of operation time.

see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg6203177/#msg6203177

The LS8 devices are typically drifting still somewhat more than the metal can TO devices.
No further jump on TO device KHZ#8.
So no significant change to last kHr.

with best regards

Andreas

[Andreas]

Hello,

PCB#2 (with CMOS multiplexers) now passed 23.4 kHrs total operation.

see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg6203193/#msg6203193

the LS8 devices are still drifting slowly upwards
Metal can device #1 is again drifting so the temporary "stabilisation" was caused by ADC16 drift.
The 0.3 ppm jump resulting from 1 ppm ADC16 drift is visible on day 908.

with best regards

Andreas

[GigaJoe]

I don't solder the LS8 directly on the board
I use AWG32 wire instead, with a small styrofoam encasing. I'm thinking soldering directly on the board is not ideal at all if targeting maximum precision."

[miro123]

I don't solder the LS8 directly on the board
I use AWG32 wire instead, with a small styrofoam encasing. I'm thinking soldering directly on the board is not ideal at all if targeting maximum precision."

I tend to disagree with such solution. AD gives you two possibility
1. Best performance - TO46 package
2. Up to modern design and take advantage of smt production capability. - LS8 package
Dead bug mount LS8 does not provide either of them .

[grx]

Well... the newest references like the ADR1001 (and the TI "copies") are only produced in ceramic packages now, the metal ones are possibly a thing of the past because they are more expensive to manufacture: for example the ADR1000 is very hard to get via normal distributors. I cant get them via mouser, farnell or digikey. The LTZ1000 also seem to have disappeared from these stores in europe.

So I think it is reasonable and interesting to investigate ways to get the best performance out of these ceramic packages too, You never know what happens in the future.

[GigaJoe]

we not speaking of industrial implementation - right ?
We trying to squeeze a maximum from what we have, as no one in a sane mind would use dead bug and syrofoam for SMD industrial mass market ...
So personally I would leave AD or TI advise to industrial thingi.
And I will use a ceramic package, floating in vacuum, inside silver coated sapphire sphere - if I can.

[grx]

Hello all,
Unsurprisingly I am also trying to build a adr1399/lm399 reference.
One important difference between these is the RC network that must be added in parallel with the "zener" of the adr1399.

ADR1399 very much wants his 1 uF in series with 5 ohms.
But the LM399 is usually connected to a simple 100nF and no resistor.

I have read all the topics of course but they're big and I would have to reread everything to get an answer to my current question.

Would anything bad happen if I used the compensation network of an ADR1399 with an old LM399 ?

Like critical problems, stability, oscillations, etc.

This is still an experimental board, not a final project.

Thanks!

[Kleinstein]

The datasheet Z_out curve for the LM399 stops at 100 kHz. From the looks I would expect the LM399 to be OK with the 5 ohms and 1 µF. I expect the RC element more lowering the overall impedance than causing issues. In most cases the difference would not be noticed.

[Andreas]

Hello,

PCB#3 (with relay multiplexer) now passed 14 kHrs of operation time.

see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg6237399/#msg6237399

I tried a one week off time to check influence of humidity.

The LS8 devices are typically affected more (0.5-0.7 ppm)
than the metal can devices (within measurement noise).

with best regards

Andreas

[Andreas]

Hello,

PCB#2 (with CMOS multiplexers) now passed 24.4 kHrs total operation.

see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg6237407/#msg6237407

I also had a one week off time to check for humidity sensitivity.
Although humidity is currently low (< 40 % rH)

again the LS8 devices are affected 0.5-0.7 ppm.
Except LS8#6 which is again less affected.

The metal can devices show less influence.

with best regards

Andreas

[GigaJoe]

i finish my array end of october ... i stops drifting around april .. in may , i use one of my 2001 for 2 weeks run it basically flat with 2 microvolts up and down as i believe it device itself ...    as i mention early it 3 x 1399 LCC ,  connected by awg32,  LCC has option of kelvin connection that i use , buffers and a final buffer.   All circuit in the metal box, not really air sealed but close, and 7cm thick styrofoam box where it self heating to around +40C , resultant variation to ambient temperature around 6:1 , let say +6 outside +1 inside. it increasing to +15 microvolts during warmup.  The power supply actually 2 one for references, one for heater as a negative.  ( somehow it think negative are another 2 cents .. )   1399 out of shelf not selected ...  some errors i fix in the output circuit ...  Ah .. total consumption for all 3 heaters 27 mA at 11.33V   basically 100mW per LCC.  I assume stable temperature environment and small dissipation also help in stability as heater less fluctuating ... and Zener i think ( not remember exactly) around 3-4 ma. I decided to feed all 3 as derivative from summing, not each itself.

so i'm thinking based on my stuff 6 month and it stop drifting ... as well i didn't noticed in off\on shift ..

[Jacques]

[..] PCB#3 (with relay multiplexer) now passed 14 kHrs of operation time. [..]

Just wanted to say thank you for continuing to share your measurements, Andreas.

As Miro mentioned earlier in the thread, it seems clear that the TO-46 package delivers better performance overall than the LCC package. Some people suggest using a flex PCB with the LCC package, but it looks to me that in small volume production scenarios the added cost of a flex or hybrid rigid/flex PCB would be higher than having a manual solder step for the TO-46 package.

I do wonder how the ADR1399KHZ holds up to the LM399AH in terms of long-term stability. Do you also have some insight into this?

The LM399 references in some of the lab equipment that we built for our own use appear to hold their values very well (way better than the datasheet would suggest). I don't have many opportunities to open these devices up and measure the references, however, as the equipment is perpetually in use.

[gamalot]

I just received two chip of LT5400‑8 resistor array, together with the four ADR1399KHZ I already have (around 6.92 V average), I can build another 10 V reference now. 

[CalibrationGuy]

I assume that is a simplified version of the unit you want to build. Don't forget the output transistor, input voltage regulation, capacitors (Andreas and Kleinstein recommended, etc.).

TomG.

[gamalot]

I assume that is a simplified version of the unit you want to build. Don't forget the output transistor, input voltage regulation, capacitors (Andreas and Kleinstein recommended, etc.).

TomG.

Yes, this is just a simplified diagram to show how the LT5400‑8 is used in the feedback loop. 

[Andreas]

I do wonder how the ADR1399KHZ holds up to the LM399AH in terms of long-term stability. Do you also have some insight into this?

Hello,
I have only a limited sample size.
Additionally my LM399 (non-A) are not fresh from the factory but most of them I bought when the NS-parts became obsolete.

The devices behave very individually.

Some stabilize after 2 weeks, others need 5-10 kHrs until they are in the range 1-2 ppm/year.
I think that is the reason why DMM manufacturers pre-age and select the devices (drift + popcorn noise) before they put it into a 6.5+ digit DMM.

with best regards

Andreas