LM399 based 10 V reference

[branadic]

Correct but as you can see the temperature changes by 6K, humidity by 45%rH and pressure by 28hPa. Even though the experiement is running in a temperature controlled lab it's enough change to make a raw estimation of what is goning on.
I agree, a better way would have been a climate chamber, but currently all of our chambers are in use. This way you would perform two measurements, one with varied temperature and constant humidity, a second one with constant temperature and varied humidity. I don't know of any commercial climate chamber that allows for controlling pressure on the DUT. We have a calibrated pressure controller, but it's to much work to set up a system to control ambient pressure for a DUT inside a climate chamber.

-branadic-

[Andreas]

I performed the compensation for temperature, afterwards for humidity and at least for ambient pressure. Attached are the single steps as a series of diagrams.

The compensation is resulting in:

Mean: 10.001790V
Std: 2.58µV

Mhm
What does "Mean" mean here.
Is it interpolated to 0 deg C and 0% rH and 0% pressure or what?
from the diagrams the value should be more near 10.00177xV.

Thats why I normalize all my interpolations to 25 deg C. (for components).
For instruments it should be more 23 deg C.

It gets not clear to me:  you did the correlation one after the other
or all coefficients together
or incremental (with correction of the previous coefficients)?

On the first view the dew point temperature seems to be better than rH for humidity.
But since the time constants for humidity are in the 3-7 day range a correlation with a pt1 filtered dew point temperature would be also interesting.

with best regards

Andreas

[branadic]

Is it interpolated to 0 deg C and 0% rH and 0% pressure or what?

Exactly, even though it's only an academic value I can correct the values to "any condition" I want.
However, for me my measurements are showing, that each precision voltage reference should contain sensors for temperature, humidity and pressure! I have evaluated different sensors and will use SHT25 and MS5611 for my next designs. Also I got an own feeling of what a case and the electrical connections through the case should look like. I think that is a very important point, if you want to do precision stuff you must run all the way through the dessert and do all the mistakes on your own to develop a gut instinct.

It gets not clear to me:  you did the correlation one after the other or all coefficients together or incremental (with correction of the previous coefficients)?

1. I first compensated for the most significant variable, temperature.
2. After this I compensated the temperature compensated values for humidity (the second significant variable) and
3. at least I compensated the temperature and humidity compensated values for ambient pressure.

With your words: the incremental way.

On the first view the dew point temperature seems to be better than rH for humidity.

Dew point is a calculated value including temperature and humidity using Magnus approximate formula, I wouldn't use that to compensate for humidity.

But since the time constants for humidity are in the 3-7 day range a correlation with a pt1 filtered dew point temperature would be also interesting.

Still I don't understand what you mean by "pt1 filtered". Maybe you can give me an example?

[Kleinstein]

.....

But since the time constants for humidity are in the 3-7 day range a correlation with a pt1 filtered dew point temperature would be also interesting.

Still I don't understand what you mean by "pt1 filtered". Maybe you can give me an example?

The effect of humidity is usually not instant. It is not air humidity but more like the humidity contend in the board and parts that influences the circuit. This is delayed against the air humidity. A pt1 filter (= simple first order low pass filter) could be used as a first estimate. However there is still the time constant (e.g. 5 days) that is an additional parameter. Using a filtered humidity is a little like using an intentional very slow sensor.

[Andreas]

Still I don't understand what you mean by "pt1 filtered". Maybe you can give me an example?

Hello,

Example: you want a time constant for the low pass filter of 3 days = 259200 seconds.]
Your measurement samples are 100 NPLC so each after 4 seconds (with zero adjustment active).

You calculate the slope K of the PT1 filter between the measurement points.
e.g.  K = 4 s/ 259200 s
You calculate the filtered values from the measurement values M[sample]
Init: PT1[0] = M[0]
PT1[1] = PT1[0] + (M[1] - PT1[0]) * K
...
PT1[n+1] = PT1[n] + (M[n+1] - PT1[n]) * K
for all samples.

you repeat the same for 7 days time constant so that you have 2 pt1 filtered rows with different time constants.
then you calculate the correlation coefficients (together).

From the ratio of the 2 coefficients you can check wether you are nearer to 3 or to 7 days.
Repeat the same for 2 new estimations of time constants until you get no improvement in standard deviation for the corrected values.

With best regards

Andreas

[branadic]

I stopped measuring my LM399 for the moment. Here are the latest results. It seems that dew point temperature represents best the change in output voltage of my LM399.
I think I will now perform some hardware modification on the reference, such as a die-cast case or similar to minimize humidity influence. Also thermal isolation will be improved to reduce ambient temperature influence. Both are resulting in dew point temperature without any lag as it seems.

-branadic-

[Andreas]

Interesting result.

For me it seems that there is a 1 ppm drift over time additionally to the dew point sensitivity.
It also would be interesting if the result is due to the scaling resistors or the LM399 itself.

With best regards

Andreas

[branadic]

For me it seems that there is a 1 ppm drift over time additionally to the dew point sensitivity.

Yes, you are right Andreas, there is drift in the order of 1ppm.

It also would be interesting if the result is due to the scaling resistors or the LM399 itself.

Not a measurement I will perform yet.

-branadic-

[2N3055]

Hello everybody!

I was wondering, dew point influence without time delay, could that also be thermal influence?

Thermal conductivity of 0% saturated air as opposed to 100% saturated air (at dew point) is 28% more (0.032 W/m*K at 0% vs 0.025 W/m*K at 100% humidity) at 90°C (temp of LM399). At room temp, there is practically no difference, but as temperature rises, there is difference between humid and non humid air. This can create thermal conductivity gradients between hotter and colder parts of circuit, and emphasize temperature gradients... Cool parts are being cooled better than hot parts when it's humid...

Usually not a problem but at ppm levels maybe it is...

Just an idea...

Regards,

Sinisa

[Kleinstein]

For the thermal conductivity it is the absolute humidity or dew point that is important. So 90 % RH at 90°C is far from what one could get from room air. Under normal conditions the partial pressure of water is more like in the 10-20 mbar range, thus something like 1% abs water contend this one would expect a water contribution in the 0.5% range. So it would be a limited effect on the thermal properties.

[2N3055]

For the thermal conductivity it is the absolute humidity or dew point that is important. So 90 % RH at 90°C is far from what one could get from room air. Under normal conditions the partial pressure of water is more like in the 10-20 mbar range, thus something like 1% abs water contend this one would expect a water contribution in the 0.5% range. So it would be a limited effect on the thermal properties.

I should have known you guys already thought about it... 

Regards,

Sinisa

[Spikee]

Would anybody here be interested in buying pre-aged LM399AH's ? As in with a minimum of 1k on time in ambient environment or using the accelerating aging method ?

[pelule]

Interested, send a PM
/PeLuLe

[branadic]

Almost done with modifications on my LM399. Main modification was that the reference was put into a Hammon aluminium die cast, filled with cotton batting for thermal insulation  and some selfmade true copper feedthrough. Therefor some installation copper wire (1.5mm²) was glued (Roth Thermokitt) into a brass part with external thread, so that I can screw the selfmade feedthrough into the Hammond die cast. Further I can now measure reference voltage as well as amplified reference voltage. Hoplefully I can start new measurements within the next days.

Will add some pictures soon.

-branadic-

[branadic]

Now here is a picture of my selfmade copper-feedthrough. As already mentioned the mainbody is made of brass which was coated with chemical tin. The copper wire is glued with Thermokit inside the main body.
My LM399 reference is sitting inside its Hammond aluminium die cast with the copper-feedthrough installed to supply power for the reference and crystal heater on top of the gain setting resistors and to measure the reference voltage as well as the amplified reference voltage. The reference is now again running and monitored by 3458A. Reference voltage is very stable. It currently seems like ambient influence is now fully cancelled out, but with the additional thermal isolation inside the Hammond case I can observe a drift presumably of the gain setting resistors 6.95V --> 10V. Maybe I have to remove some of the wadding to allow some heat dissipation? Some further pictures will follow soon.

-branadic-

[Gyro]

You could save yourself a fair amount of effort on the feedthroughs. Hollow bore glass feedthroughs are readily available on ebay for running copper wire through.

https://www.ebay.co.uk/sch/i.html?_odkw=glass+feedthrough&_osacat=12576&_from=R40&_trksid=p2045573.m570.l1313.TR0.TRC0.H0.Xglass+hermetic+feedthrough.TRS0&_nkw=glass+hermetic+feedthrough&_sacat=12576

Of course, if using a hammond encloure you'd need to solder then into drilled out a brass screws, or glue them. Some of them are very small.

[branadic]

I tried something similar with ceramic feedthroughs before without success, because the inner metallization broke while soldering. Maybe because of the high temperature gradients and the involved stress. So this solution is fine for me and with only small effort as I needed the brass body with the outer thread anyway. The rest is straight forward, just a thick copper wire and the Thermokit. As I'm an engineer for devoloping new sensors based on assembling technology this solution perfectly fits my profession
This is not a hermetically solution, but close enough to increase time constant for humidity to a maximum amount.

-branadic-

[ManateeMafia]

I bought some from that seller. They look ok but I haven't tried soldering them yet to see if they can handle the heat.

[d-smes]

You can also buy off-the-shelf feedthrough EMI filters with silver-plated copper lead.  NOT hermetic, but epoxy sealed.  For example: http://www.mouser.com/Tusonix-CTS/Passive-Components/EMI-Filters-EMI-Suppression/EMI-Feedthrough-Filters/_/N-bw7oz?P=1z0zkur

[branadic]

As mentioned my reference is running in it's new case with my selfmade feedthrough. I first monitored the reference output to see the behaviour to ambient changes. Nothing visible.
Afterwards I monitored the amplified output (~267h) and was observing a signficant drift.
I than switched back to the reference output and monitored its ouput till today (~95h). As you can see, the reference voltage is pretty stable and no influence of ambient is visible.
However, the amplified 10V output is drifting. Thus, the constantly heated resistors (crystal heater on top of 5ppm/K SMD resistors) seem to drift.
I will go on measuring the reference output another few days. I will then try to remove some of the batting inside the case to see, if this gives any improvement in amplified output. Maybe I have to allow for some heat dissipation to decrease the drift of the resistors. If this still doesn't give any improvement the only way I see is to quit the SMD resistors + crystal heater approach and go back to classical leaded resistors.

Attached are some picture of the measurement setup and the measurement results.

-branadic-

[Kleinstein]

SMD resistors are sensitive to board stress and thus indirectly may react to humidity. The main effect of a partially sealed case is slowing down changes in humidity. So it can take a long time to get a stable value.

[branadic]

SMD resistors are sensitive to board stress and thus indirectly may react to humidity. The main effect of a partially sealed case is slowing down changes in humidity. So it can take a long time to get a stable value.

I don't think that this is the case with my setup. I'm currently running the reference with crystal heater turned of. After some recovery time (a few days) with opposite drift direction the 10V output seem to have stabalized around 10.001802 with noise in the order of <±5µV at first sight. Updated diagrams following soon.

-branadic-

[branadic]

As announced attached is a diagramm how the readings have recovered with crystal heater (on top of the ±5ppm SMD resistors) turned of.
Not sure if I will leave it as it is or if I will change the resistors to wired ones. I found that RS Components do have resistors from Alpha Serie MCY for a decent price and ordered some 10k and 20k that I now have on my desk. I had in mind to parallel the 10k with some ±25ppm/K resistors  to trim the output to 10V. Maybe I will wait some additional time before I decide this step.

-branadic-

[branadic]

Did it, replaced the gain setting resistors by Alpha MCY ±2.5ppm/K (10k || selected 82k ±25ppm : 20k). Instead of the crystal heater I placed some 33k NTC to the connections. So the story goes on.
Will use this 10V-reference for the AD5791 board that is coming within the next days, so that I can make some automated linearity measurements on my Prema 5017 together with the 3458A, but that's another story of another thread.

-branadic-

[branadic]

As reference is measured with 3458A I found a source for some uncommon Vishay foil resistor values with 5ppm/K on ebay. So I ordered some 82k2 to replace the 82k, 25ppm/K metal film resistor, that is paralleled to 10k and that is gaining the reference voltage with another 20k to 10V. I hope to get a little closer to 10V and get it a little more stable. Actual diagrams below.
This modification is the last one I will perform on this LM399 before I connect it to AD5791 that is now on my desk. 

-branadic-