ADR1399 reference

[iMo]

..
..Strange the stability at 55 degrees

That is because there is a TC of the 1399 zener combined with the TC of the trimmer - see below an example, trimmer's TC from -50ppm/C to +50ppm/C. Also your LT3083's TC may contribute as well.
Vrefout in ppm.

[Kleinstein]

I doubt the models for the parts are that accurate to worry about the 2nd order TC.  The main point would be checking the coarse part (e.g. how much is the TC of the trim resistor attenuated) and for the simulation one needs some x-axis (e.g. time or temperature) to get the graphs and more conventional interface.

Beside the 100K/10K LT5400 array, one could also use an array with 8 equal resistors (e.g. TDP or NOMC )  for a 1.5 to 3.5 R ratio. This starts at the nominal 3:7 ratio and would need trim options for up and down.  The ADR1399 is nominal 7.05 V and thus in between the 2 approximation. The shown trim resistor in the LT5400 version only works up to some 7.2 V. A rare but still in spec higher voltage unit would need a different position trim resistors to reach 10 V.

[Conrad Hoffman]

Has anybody come up with a PCB for the 1399 ceramic SMT version that's generally considered to be good?

[Andreas]

Hello,

I think I have now solved my "position dependant drift". see also:
https://www.eevblog.com/forum/metrology/adr1399-reference/msg4683005/#msg4683005

I have put the PCB into a die cast housing (Hammond 1590U)
https://www.reichelt.de/alugehaeuse-1590-u-120-x-120-x-58-6-mm-aluminium-1590u-p221457.html?&nbc=1

The temperature sensor from CPU is now cabled + taped on the bottom of the housing.
The PCB with the 4*ADR1399LS8 + 2*LM399 sits with some cloth (thermal isolation) within the metal housing.
Of course the effective T.C. with respect to the "case" sensor also is different from "environment" and "MUX" temperature sensor.
So I had to determine a new set of T.C. coefficients.

Now the measurements are stable (as I am used from my ADCs) and the T.C. of the references can be compensated.
The temperature drift of the ADR1399 is still much higher than the LM399 samples.
The LM399 shows more stable uncompensated values (except for some popcorn noise on day 4 since Hammond packaging).

The T.C. compensated values are additionally averaged between 3 ADCs to reduce further measurement uncertainities.

The overview shows the new effective linear T.C. values which are minimum a factor 3 higher for the ADR1399LS8 samples than for the LM399s.

with best regards

Andreas

[DavidAzulay5000]

LM399 + OP AMP LT1001 COMBO
PERFECT 10V Reference 10.000.00
v

[gamalot]

The simplest ADR1399 comes, only two 1uF capacitors and a 1K ohm resistor, measured with 34465A, it seems not bad! 
 

[Kleinstein]

The 34465 compares the ADR1399 to it's internal LM399.

[gamalot]

The 34465 compares the ADR1399 to it's internal LM399.

I also did a test with my 6581T (1000 samples, 20 NPLC, 14uV Span, 2uV Std dev), these results were almost identical to the 34465A.

[iMo]

Evaluating an ADR1399 (or an LM399 and better refs) with a 6.5digit voltmeter is difficult. You have to know your meter well at the 10V range - you have to know its TC to compensate for its internal temperature and/or ambient, you have to log an 8-10 hours long measurement (at 100NPLC) and evaluate several thousands of samples at minimum (analyze the graphs, compensate for internal/ambient temperature, remove the outliers, provide filtering/averaging/smoothing). And you get a single number (I do it this way). That all with an assumption (not valid one, btw.) the long therm drift of your meter is zero.

The LM399 inside those 6.5meters does a random walk typically within 5-10uV at 10V range (therefore the minimal resolution with those meters is 10uV at 10V) and there are sporadic random jumps 4-5uV large (the popcorn) on top of that walk as well. And your DUT does the same too, basically..

[dietert1]

As long as the ADR1399KHZ isn't available, that minimum setup with the SMD part as "dead bug" is a good solution. It reduces stress on the package and the heating power. With that minimum setup one can make two or more of them. Once you observe more than one, it gets a little more conclusive.
And the differences can be measured to 100 nV with a good 6.5 meter. Except one needs to avoid systematic errors affecting all of them in a similar way.

Regards, Dieter

[iMo]

Btw. - the DCV Ratio (34401a) always returns 7 significant digits after the decimal point over the serial, could be somehow useful when having a 10V ref and the DUT is an LM399 (7V). I have not used it this way yet. Do not forget to wire the LO and the LO together 

[Andreas]

Hmm,

I have here 7.5 life digits with 34401A ratio measurement.
(Measuring the Ratio of a ADR1000A between 10V and 6.6V output)
Ratio is in mV/V VOut/VRef.
The pitfall is only  1 measurement all 12 seconds with 100 NPLC.

21:57:08;  1499.99030;               1
21:57:20;  1499.99020;               2
21:57:32;  1499.99040;               3
21:57:44;  1499.99030;               4
21:57:56;  1499.99010;               5
21:58:08;  1499.99030;               6
21:58:20;  1499.99060;               7
21:58:32;  1499.99060;               8
21:58:44;  1499.99040;               9
21:58:56;  1499.99020;              10
21:59:08;  1499.99040;              11
21:59:20;  1499.99020;              12
21:59:32;  1499.99040;              13
21:59:44;  1499.99020;              14
21:59:56;  1499.99030;              15
22:00:08;  1499.99030;              16
22:00:20;  1499.99040;              17
22:00:32;  1499.99040;              18
22:00:44;  1499.99050;              19
22:00:56;  1499.99030;              20
22:01:08;  1499.99020;              21
22:01:20;  1499.99040;              22
22:01:33;  1499.99050;              23
22:01:45;  1499.99030;              24
22:01:57;  1499.99040;              25
22:02:09;  1499.99060;              26
22:02:21;  1499.99000;              27
22:02:33;  1499.99030;              28
22:02:45;  1499.99020;              29

The Ratio measurement is rather stable (shown by Allan Deviation mV/V of 1 minute averages)

with best regards

Andreas

[iMo]

Yep, as I wrote above there is always 7 sig digits after the decimal point. There is 9.12345670e-01 or 1.12345670e0 I get.
It works such it makes 100NPLC with V1 and then 100NPLC with V2 and calculates the ratio. Thus the internal LM399 is still involved, afaik..

[Kleinstein]

With many modern DMMs the ratio measurement is just measuring the the 2 external voltages relative to the internal reference and than calculating the ratio.
This may not be that bad with a reasonable fast and low noise ADC. Many of the old Keithley 19x meters used a 3 reading cycle with a reading of the reference voltage on there normal AZ mode. So using the ratio mode with a reading of the external reference instead would at least not slow down things much or add extra noise.

Requiring 12 seconds for a 100 PLC ratio measurement on the 34401 looks a bit slow, as 4 conversions (high and low side for the input and reference) should be enough and thus 4 x 2 seconds.
I don't expect the ratio mode to be high priority for the 34401.

Only very few older meter (e.g. Solartron 7071/7081) do an a true direct ratio by using one of the voltages for the ADC reference.

The often much more practical solution is to looks at the difference of 2 external references, if they are comparable in voltage (e.g. less than 20 % difference).

[JodelJonny]

Thanks for the feedback. I tried to implement your suggestions.

-The temp sensor is moved to the resistor network. I'd like to stick to the digital one because it makes it easier to display and save the measurements with minimal external stuff needed. If it interferes with the reference it could only be powered on for brief periods between measurements.

-I switched to the NOMC resistor network with pads for external trim resistors.

-I tried my best at star grounding. Just for the heater it felt wrong to wire it all the way to the star point and back. I don't see any problems wiring it directly to the input. 

-I will try the 7815 on the board. If it causes thermal problems it is easy to short it out and use an external regulator. I want the reverse polarity protection on the pcb in any case because between the power supply and the pcb is the most likely place to mix up the polarity.
Is there a better regulator than the 7815 for this application?

If no one spots any big problems anymore I'm going to order the parts and the pcb and give it a try. This is just supposed to be the first step to get a feel for things.

[Conrad Hoffman]

I'd probably use an LM317. I think it's a better regulator and it gives you the option of trying different voltages. Reg on board worries me, but give it a try.

[Andreas]

-I tried my best at star grounding.
-I will try the 7815 on the board.

You obviously have not understood the principle: remove high courrents (heater, zener supply) from critical wiring (zener voltage sensing).
Doubling the lines (seen on the zener ground) has no effect (the voltage drop from the 3 mA Zener supply will be seen on the output connector).

There is nothing wrong with a 7815/LM317 if the cirquit is mains powered.
With battery supply I usually prefer a LT1763.

with best regards

Andreas

[iMo]

I would also fill the circle on the bottom pcb side beneath the 399 with copper.
Also - what is the pitch of the 399 pins (between the adjacent pins)?

[Kleinstein]

The optional trim resistors R4 and R5 would have no effect as they are shorted. The trim part with 4 resistors are 4 different ends still looks a bit odd. I would more prefer 2 positions to get a trim up or down. Suitable are the resistors that are in series with another and thus have a lower weight to start with and than 2 or 3 optional resistors there.

The star ground idea is still way off. 
The heater, regulator ground and supply to the OP-amp can be treated separate and no real need to use a star ground for these.
One should use the star ground part mainly for the critical parts: the neg output terminal, the zener neg side, the low end of the resistor array and one link to the normal ground.
All 3 points have or can have some current flow. The use of a start ground thus does not fully compensate for voltage drop on the ground lines. This would need exrta current compensation from a negative supply, which may be a bit overkill. Just using relatively short and not too thin traces should be good enough. The main point is to keep the heater current isolated from the critical part.

[JodelJonny]

The doubled up ground trace was left after thinking about using the SMD package with separate sense pins. I forgot to delete it. It does nothing in this case but as long as the enclosed are is near zero it shouldn't hurt either.
 
The pitch between the 399 Pins is 1,796 mm.

I'll think again about the ground configuration but I honestly don't see the problem right now.

R4 and R5 are in series with the circuit so they have jumper links to short them out between the pins of the resistor network that could be easily cut if I'd like to use those resistors instead. I should have made that more obvious in the circuit diagram.

[dietert1]

C9 could be a 1nF film capacitor.
Also i would try to avoid vias in the reference ground connection and at the negative opamp input. I'm not shure how much thermal EMF vias can contribute and whether it is possible to "untwist" the design.
The thermal ring can be ground and under the sensor signal lines there can be ground, too. Same applies to the pad under the resistor array.

Regards, Dieter

[iMo]

An example of a star grounding for a typical LM399 circuit (an example only).. @Kleinstein - plz do advice us directly in the schematics..

[Andreas]

I honestly don't see the problem right now.

Just one example:

Your connection between output pin and LM1399 has about 25 mOhms (assuming 1 mm width and 35um Copper with around 50 mm length).
It carries the reference current of 3 mA.
This gives a increase of 75 uV of the zener voltage (or 10-11 ppm increase).
If all would be constant that would not  be a problem.
Now the T.C. of the copper trace with nearly 0.4%/K increases the zener voltage T.C. by 40 ppb/K.

An example of a star grounding for a typical LM399 circuit

The connection between output and LM399 has to be as short as possible in this case (ok for LM399 its a factor 3 less critical).

with best regards

Andreas

[Kleinstein]

The ground is indeed not that bad as it looks at first sight. Going to both sides makes it a bit hard to follow though.
The neg output terminal is used as the star ground point. This is a coice for a very low output impedance, but external wires woud add anyway.

The link from the neg output to the resistors should be thicker - that resistance adds to the divider and may add a tiny bit to the TC. If really needed one could still add parallel copper to correct.
The link from the neg side of the reference could be made a bit shorter and maybe use 2 or 3 vias. I am not that afaid of thermal EMF of the vias, more the resistance of small vias.
The 2 parallel lines are confusing, but could be used for some TC tweek if really needed, as the restances adds a little to the ref. voltage. So they are not that bad.

There is no need to also have the neg supply of the OP-amp from the star ground, but as the supply current is pretty stable it is also not that bad. Ideally the ground here would not go through the star ground for the critical levels.

One should avoid isolated copper islands and better connect those areas made for thermal reasons to ground or a simlar potential.

Cutting a trace directly at the SO16 chip can be a bit tricky. I would prefer trim by just adding resistance in parallel and not opening a link. There are 2 suitable position for a trim up and down.

[JodelJonny]

Thanks. Good to know, that my thoughts were not totally off after all.

I'll work on improving the "readability" of schematic and layout.

One of the thoughts was to be able to support up to 10 mA output current without major problems. That's way more than needed, I know. Therefore using the output terminal as the star point was chosen.