LM399 based 10 V reference

[blackdog]

Hi Kleinstein,

Thank you for your comments, it's made me think more about it and make some more adjustments. !

1e
The reason that I also made it possible to measure the voltage across the 10K resistor is this, I also want to use it to select electrolytic capacitors on leakage current.

2e
This test box is intended for capacitors from say 1uF.

3e
I am aware of the charge my body has when switching the function.
But by using the isolated bridge of the "MC" Brand and touching the metal of the box with one hand, I can switch " low of charge" for that matter.

4e
My first thoughts were a rotary switch and/or relay's to switch the functions.
But I don't have a good rotary switch that will do, and with relays, my batteries will run out quickly.
It can be done with relays by using bistable versions to save power. but I will have to order those and I also want to build this measuring instrument with the parts that are now available here.

5e
Leaks due to dirt can of course always be dare with the connecting bridge as suggested here.
But measuring instruments of mine, that are already about ten years old still look like new.
I regularly clean the connectors and make sure my hands are clean when operating my instruments.
So i have to be shure this box is clean before i use it!

6e
Regarding the charge of the DMM input on the high impedance setting, that is correct, one of my KeySight 34461A goes on the 10V range in overrange, if you wait long enough.
For those who don't know yet, the source of the problem is the leakage current of the DMM running out of the input and which charges the capacitor and the paracietic capacitors in the DMM.

So short-circuiting the DVM input before you start measuring might be a good plan.
But I'm still thinking about maybe doing it a bit differently, I will show you in the 2nd version schematic.

7e
I measured two of my DVM's in terms of input capacitance, that is for the KeySight 34461A ~300pF and for the Tektronics DM4050 about 210pF, both in the 10V DC range.
The charge in these capacities together with the capacity of the short measuring cables will, as I think about it, give little disturbance to the measuring data if you measure capacitors larger than 1uF.

Kind regards,
Bram

[Kleinstein]

I just did some DA measurements on 2.2 nF low loss caps at shorter time scale (e.g. 100 µs to 50 ms). So I may be thinking about slightly different effects that are less of a problem here, though the smaller cap and shorter time scale require similar insulation resistance.

One could have the circuit with only one switch that is critical - the one to connect the DUT to the circuit. The rest of the circuit could than use less critical switches to select charging, discharging and measuring. The meter could be always connected, to also measure charging voltage. So it would be charge, discharge and "open" for the 2 nd switch. 

In most cases the charge in the DMM input capacitance is not a problem, at least with 1 µF or similar caps. It could be with smaller caps.
So limiting the initial voltage is usually good enough.  In a few seconds the input voltage will not rise very much, the full 15 minutes could be a little long.

Measuring the leakage current of the cap under voltage may requite a very stable reference. Just the DA measurement does not need such a stable voltage (just a 9 V block cold be good enough), as it is not to know the DA to 1% of its value. It is usually more the question if the DA is in the 1%, 0.1 % or maybe 0.01% range.  Looking at different time scales may be more informative than a very accurate scale factor.
DA can be quite temperature dependent.

[maat]

@Blackdog: You might want to put a temperature sensor on that board, close to the capacitor, because leakage current is strongly dependant on temperature.

Judging from your circuit the power consumption will be pretty much negligible, therefore  the temperature will be around room temperature, but if you put those caps into real circuit you will probably some 10 K above ambient. Depending on the capacitor, this might double or quadruple your leakage current. So maybe you want to add some kind of heater to your box.

If you are interested, I do have leakage current values at different temperatures for the WIMA MKS2 10 µF, 50V cap you intended to use. I'll will dig them out.

To get some reasonable values from the WIMAs I had to wait about 8h. Nichicon UKLs needed about a day or two to settle. Kemet T491 about 8h. Regarding the Tantalum capacitors, after eight hours, there was still significant DA, but in comparison to the leakage in absolute numbers it could be considered settled.

I usually measure across a 1 MOhm resistor using a good DMM. Their input current is on the order of a few tens of pA or lower so having them connected in parallel can be neglected when measuring capacitors 1 µF or up.

[blackdog]

Hi, maat,

All information is welcome, I think also for other forum users.
There is not much information available about capacitors except some generalities.

Currently I am testing a 100uF MKS4 from WIMA with a 10V reference module and a 10K resistor,
after 8 minutes of charging the difference between the reference output and the voltage across the capacitor is 10uV.
The current that was flowing at that time was 1nA and it is still dropping.

Of course this is not 6x the WIMA MKS2 10 µF, 50V capacitor, but I will do a test with that tonight.

So all information is welcome and I will also do a test with the capacitors in the 42C oven to see if the leak is significantly higher than at 25C.

Thanks!

And have all a nice last day of the year, here still 3 hours to go before the end of the year!

Bram

[blackdog]

Hi,

I had a little time before the lady of the house called me. 

Here some measurement data done on the philips PM6303 measuring bridge and the 100uF WIMA at 10V leakage current test setup.

This is the offset of the meter when I short-circuit the meter (10K resistor).
If the MC short-circuit bridge applies, you have to wait about 5 minutes until Mr. Seebeck has left. 
When the display is stable on the meter, I use the "Null" function on the meter to adjust this offset away.

.
This is the value after using the "Null" function.
2uV across the resistor means a current of 200pA, there is about 50pA bias current from the meter.
I forgot to see if the current is off the meter right now or running out of the DVM.
Take these measurements with a grain of salt, they are difficult to perform reliably, just see that it indicates a direction.


These are the six WIMA 10uF 50V capacitors in parallel, total capacity 58.31uF.
This is already difficult to do a good 4 wire measurement because of the low impedance.


With well connected sense wires, the meter shows a relatively high Q.


This is the "D" of the capacitor which is also 1/Q.


And this is the impedance of the six capacitors at the measuring frequency of 1KHz.


Kind regards,
Bram

[Kleinstein]

If the current is still drop, it may very well be part of the DA and not real leakage. DA is not just happening as charge recovery, but also as delayed charging of the capacitor. With the usual 0.5% DA number for polyester caps one can about expect 0.5% of the charge and thus some 5 µC to go to the delayed part. The number is only approximate, as it is for a given time scale. If the effect is slower, there could be even more and the fast part is already done.
Still 5 µC are good for 1 nA over 5000 s, so this could be most of the current seen.
1 nA and 10 K source resistance and thus some 10 mV may still be noticeable for the LM399 reference. 200 pA may be below the level where it is a real problem.

With high DA like in MKS caps it is really difficult to measure leakage, as it is difficult to separate it from DA effects. One could see the difference comparing the charging and charge recovery part. The DA part in principle comes back as charge recovery, the leakage does not. However it is a slow process and the difference may not be visible over short time (e.g. less than hours).

[blackdog]

Hi,

You all have a nice 2020 

Yesterday I connected the 6x 10uF 50V WIMA to get an impression of the leakage current and DZA of these capacitors.
The large WIMA 100uF 63V MKS4 had become almost immeasurable, so low was the voltage across the 10K series resistor.
It was close to the drift, input leakage and noise of the reference and the measuring instrument used.

This 100V 63V WIMA MKS-4 capacitor could well be used in my LM399 reference circuit,
the leakage is very low and the DA can be used for a reference that is in a oven because of the heating time to get stable.
But it is much too big for my oven.

Great, let's go back to the 6x 10:00 WIMA in the 50V version...
It's always good to put your components to an extra test.

I have already used these capacitors in several filters and had no problems with that.
But with the filter application of this LM399 reference design, many more properties of the capacitor are important.
Of course I looked in the datasheet of this type of WIMA MKS2 and hoped that the insulation value would be better than mentioned in the datasheet.
And I think this is also the case but I use six capacitors in parallel, that doesn't really help.
And when I think of what forum user "maat" explained, that the leakage current of this type of capacitor is quite temperature sensitive, it is better not to use this type of capacitor in this project.

But then what?
Back to my old filter which I also used in other voltage references including my main LAB 10V reference which is also in an 42C oven.
That is a bootstrap electrolitic capacitor filter, placed here in the latest version of the schematic.



So now we are going to perform the next test with 330uF capacitors.
There are already a few types here, which I have already bought for low leakage applications in this kind of filters.

So now we are going to perform the next test with 330uF capacitors.
There are already a few types here, which I have already bought for low leakage applications in this kind of filters.

For the first test, I attached two 470uF 35V Nichion and two 470 Nichion 16V to my 10V reference PCB in a "real" professional way! 
This test is with new capacitors, no formatting in place.


And here I am formatting 13 capacitors at 15.5V.
I'll leave this on for about 36 hours.


I still use my series HP and Harrison power supplies for general testing in the LAB.
Here I use the Harrison 6823A for the capacitor test, on this power supply I also mounted a BNC so it's easier to connect it to one of my function generators.


The test with these capacitors on the 10V reference is now on for 0.5 hours and the leakage current is already lower than with the 6x WIMA 10uF 50V capacitors!
And that is with completely new capacitors without having formatted them on 15V first.
I now measure 18uV over the 10K resistor and the 6x 10uF was 45uV after 12 hours.
Looking good for the first test...

Kind regards,
Bram

PS
I forgot to add the Spice simulation of the low pass filter.
Does anyone know how I can get the graph in LTSpice so that there are more lines so that it is easy to read?
I don't use spice much so I don't have much experience with it yet.

[iMo]

Your R6/7 (or R2 in your sim) should be something like 100k-200k, imho..

Does anyone know how I can get the graph in LTSpice so that there are more lines so that it is easy to read?

What do you mean by "more lines"?

[blackdog]

Hi imo,

R5 and R6 are fo charging the capacitor C3 to almost the full reverence voltage, soo over C2 wil than be only a verry low voltage, therefore also very low leakage current.

Kind regarts,
Bram

[iMo]

R5 and R6 are fo charging the capacitor C3 to almost the full reverence voltage, soo over C2 wil than be only a verry low voltage, therefore also very low leakage current.

But the bottom R value impacts the corner freq of the low pass.

[Kleinstein]

One can use parameter sweeps to get more curves in one plot. With a command like
.step param par LIST 100 1K 10 K
The parameter par  (use as {par} for values)  gets assigned the 3 values and one gets a curve each. However it is sometimes tricky to see whick curve is which.

With the bootstrapped electrolytic cap, one may reduce the leakage problem, but could still get DA effects. So in the simulations it may be worth to include some DA equivalent circuit: e.g. a capacitor of some 10% the size and series resistor in the high MOhms range. Settling in electrolytic caps may take several days.

[blackdog]

imo,

I know that the turn-over point and the steepness influences it is of course always a consideration of what you want and what comes out for your application.
Most of the applications I've seen from others was a resistance ratio of about 10x.
Thanks for showing me the step parameter in LTSpice!

Kleinstein
This picture made by forum user RR shows what I mean, all small steps are also visible in the frequency and amplitude scale.


Thank you both for the help.

Bram

[iMo]

Plot settings -> Grid tick

[blackdog]

imo,

YES!



Thanks
Bram

[dietert1]

As imo already commented: For a meaningful application the lower capacitor in this circuit should be much bigger. Unless its RC time constant is larger than that of the filter itself, the circuit won't work as intended. Can't you keep your 60 uF MKS cap for the upper part?

Regards, Dieter

[blackdog]

Hi Dieter1 and 3roomlab, :-)

My time for this project is up now, I'll probably have time again coming weekend.

I already played in LTSpice with a larger value of the lower capacitor, the last version the lower capacitor was 1500uF and the top capacitor was 470uF.
That looks good in LTSpice, but I still have to think about whether or not other effects will play a role.

This evening I also tuned the amplifier stage in LTSpice in terms of AC behavior.
I'm already satisfied with that, but I still have to do some checking of the DC settings, if the output from my the last schematic is able to deliver the 10mA I always want to beavailable.
I haven't shown this schematic here yet, that will come at the end of the week, afther I've done some real noise tests with the Audio Precision measuring set.

Thank you guys for al the input. 

Kind regards,
Bram

PS
The 2x 470uF capacitors now only give 5.5uV over the 10 K resistor, so that goes very well for these unformatted parts.
Probably this is a mix of the DA and the formatting.

[iMo]

Here is a simple simulation with pink noise (1/f) from a file.
Download pink.wav (various sources) and assign the file as the voltage value of a voltage source (wavefile="C:\Pink.wav" chan=0).
I've added a popcorn too..
PS: the green and red signals look weird, the amplitude is around 1uV, needs a better setting somewhere.

[blackdog]

Hi,

A small update about the capacitors under test (2x 470uF) de voltage over the 10k resistor dropt to <2uV, that is less than 1ppm deviation on the 10V output voltage.

Later today I will use the DVM "Null" again to see if it has been changed to get a more accurate measurement.

Kind regards,
Bram

[maat]

Here are the numbers promised, but first the setup. I used a lead acid battery (12.84V) in series with the capacitor and a 2.2 MOhm (2.192 MOhm measured, 1%, 50 ppm/K, Vishay MRS25000C2204FRP00). I *continuously* measured the voltage across the resistor using an HP3458A (~1.5 pA input bias current).

Ambient temperature in the lab is stable to within 0.5 K, DMM temperature to within 1 K, DUT 0.1 K over the course of a day.

Temperature range: 25 °C to 45 °C in 5 K steps. I am currently limited to 45 °C, because of the sensor setup. I might change that in the near future. I could add temperatures down to  15 °C and, if I replaced the sensor, up to 65 °C if there is interest. The temperature of the DUT is usually a little 'off', because the sensor used for the set-point is different from the one attached to the DUT and there is a thermal gradient inside the box.

WIMA MKS2B051001N00JSSD, 10 µF, 50V, 5%, Sample size: 5

Temperarture of DUT in °C	Leakage @12.84V in nA	Insulation Resistance in GOhm
44.7	4.12	3.11
39.5	2.45	5.22
34.3	1.49	8.59
29.1	0.89	14.38
24.0	0.55	23.27

And when I think of what forum user "maat" explained, that the leakage current of this type of capacitor is quite temperature sensitive, it is better not to use this type of capacitor in this project.

I did by no means imply that these capacitors are bad in any way. They are fantastic and I would say WIMA is one the best brands you can find. They are a few orders of magnitude better than their datasheet says (which is likely due to the timescales involved in measuring significant sample sizes during production).

The problem is, that you are 'magnifying' the issue by a factor of 10k. Using that much gain, will make anything look big in comparison to microvolts.

So lets do a quick back-of-the-envelope calculation:

• Oven temperature 42 °C
• Estimated board temperature 50 °C close to the references
• Estimated change of leakage current @7V: 0.3 nA/K per capacitor

This will yield you a net tempco of 2.63 ppm/K using 6 capacitors (likely a bit better)!

Regarding the use of a bootstrapped electrolytic cap. This is not a good idea either, because not only does the leakage current change significantly more, but also the  capacitance. The latter will severely diminish the effect of the bootstrapping and get you horrible 1/f noise. I did try that at one point before using a 249/2490 Ohm + 330 µF Nichicon UKL bootstrapped setup and it already got me all kinds of weird drift.

I don't know your application, but in my opinion I would keep the the rc filter somewhere in the lower Hz (only use one capacitor for example), because any DMM you are going to use to compare your ref will low pass filter the result anyway (NPLC=100), so you only need to push the noise down until the DMM filter does the rest. If you want to improve the tempco beyond a single MKS2 and you have the space, use a polypropylene cap (as suggested by 3roomlab). I can recommend the MKP4 series. They come in 10 µF at a space (and price) premium .

[Kleinstein]

Usually a DMM will do a lot filtering. The main part to filter is something like the 2.5 Hz range, as a DMM may use 10 PLC sampling the signal and than 10 PLC sampling zero. This would make the DMM slightly sensitive to noise around 2.5 Hz that can be avoided. Ideally the DMM would include such a filter, but only a few meters do, and the filter in the Datron1281 does not look like low enough to suppress 2.5 Hz (if they really use 10 PLC).

Instead of spending a lot on filtering, there would be the alternative of using a 3rd or maybe even 4 th LM399. Of cause the return of more references goes down, so does the gain from a little more filtering.

For multiple references it may be interesting to also get access to the individual voltages and differences, to see if some of them are drifting more than others.

[blackdog]

Hi, maat,

Thank you for the information about the WIMA MKS-2 capacitors, it is appreciated.
I am the last one to say that these are bad capacitors, I think they have made a nice product as far as the 10uF 50V capacitors are concerned, I used them many times.

But as always, use the components that fit your application.
The 10uF 50V MKS-2 is just not the right capacitor for this application.

Bootstrap filter
My 10V LAB reference which contains 4x an obsolete LT1021, in an oven of around 42C, drifts extremely little in the last four years.
According to my calibrated 3458A it is still within 2PPM.

The 3458A I have, drift according to the Keysight reports, not more than 1PPM per year.
This LT1021 10V LAB reference also contains a bootstrap filter, but I do not experience any strange behavior.
Maybe that's because I use select capacitors...

I tried to buy this capacitor from Farnell: 330 µF Nichicon UKL, but they don't have this type in stock.
Then I have to wait or see if I have a list for another electronics wholesaler, this because otherwise the order costs will be too high.

At this moment, the test of de new 470uF 16V capacitor is goin wel and it looks like its hit the bottum at 2uV over de 10K resistor.
The other 470uF capacitor are stil forming @ 15,5V, late this evening or tomorrow I wil place in series with every capacitor a 10K resistor to test de Leakage current.

I'm also happy with all the input I'm getting here.
Yesterday I even put the opamp circuit in LTSpice and tuned it and tuned the filter responce.
Without examples and help on this forum, I would never have done this.

I am especially someone from the "Old School" do real measurements because Spice often fools you.
But I've known for a long time that Spice can also take a lot of work out of your hands.
But I always do real measurements, most of what I do with Spice is calculate filters and that almost always goes well.

Sorry for the crooked sentences, I suffer from dyslexia and English is not my native language, that's a lose, lose situation 

Kind regards,
Bram

[iMo]

I think the KSFs here are a) to make the oven with say <0.15Kp-p stability, b) to make <1ppm/K "ratio TC" 7->10V divider, and c) to find popcorn free LM399s (difficult). A lot of challenges, indeed 

[Kleinstein]

I don't think there is much hope to find LM399 really free of popcorn noise. It is more about finding some where is popcorn noise is not too high.

The  7 to 10 V divider can be the real challenge: the TC matching may the easier part. The really hard part is to have the ratio long time stable.
Especially with an oven one may not even need good TC matching.
For a possible check one could also consider having the option to get the raw 7 V out, not just the final 10 V.

[Andreas]

Hello,

fortunately most of the LM399s cannot read the papers (nor data sheets).
Usually the ageing is only the first few weeks and turns to a "random walk" afterwards.
This first ageing is most probably equalizing the tensions of the cement between chip and housing.

attached a ageing curve LM399#2 and #3 are pre-aged devices.
LM399#CH6 and #CH7 are from day zero for the first 2000 days.
As reference 2 LTZ devices (unbuffered) with some "accidents"
(see the sudden shifts of the short cirquits).

All measured against a LTC2400 with temperature compensated AD586.

with best regards

Andreas

[iMo]

@Andreas: my conspiracy theory would be AD/LT is selecting somehow (before a burn-in) the LM399s based on popcorn or 1/f noise level. They sell parts with "LM399" markings to the general public, and the parts marked with "weird numbers" to the selected customers. Guess who gets what..