LM399 based 10 V reference

[branadic]

Thanks Andreas, we hopefully get the pictures of our small experiment within the next days
The crystal is not thermally isolated with foam directly on its surface, but the complete board is running inside this blue styrofoam box. I just opened it for this pictures.

[babysitter]

Today I teached our apprentice how to make PCBs and grabbed his product... and assembled a LM399 Reference ("Portable calibrator" from Datasheet, LM317T preregulator, available OP727R as Amplifier, 0.1% chicken food THT resistors, some C and L to guesstimated-improve EMC and interesting parts hidden in pink foam) during lunch break.

More later...

BR
Hendrik

[branadic]

More later...

What is coming next?

[babysitter]

This. Sorry, its in german. Description of my lunch break reference.

http://dg3hda.primeintrag.org/doku.php?id=lm399

Greetings
Hendrik

[branadic]

I thought there would be something special, cause you worte:

and interesting parts hidden in pink foam)

[babysitter]

Isn't it a nice package ?

Well, there is nothing special about it, as the application circuit doesn't call for anything special, right ?
No wire resistors wound by virgins at full moon...

Greetings
Babysitter

[babysitter]

I received the thermal videos from Branadic, also some background story, and will post it to youtube and will publish a dedicated webpage about his efforts soon.

[branadic]

Thanks to Hendrik for setting the videos and the publication online

http://dg3hda.primeintrag.org/doku.php?id=lm399_thermographie

[Andreas]

Mhm,

on temperature compensated zeners (1N829A) there exists a "zero tempco current".
Has anyone tested if the LM399 also has some kind of zero tempco current
(near the heater temperature) in the valid range of 0.5 .. 10mA for the zener?
Can the 0.5 - 1 ppm/K be further optimized by this current?

With best regards

Andreas

[branadic]

Has anyone tested if the LM399 also has some kind of zero tempco current
(near the heater temperature) in the valid range of 0.5 .. 10mA for the zener?

I didn't, but an interesting question you ask.

[mwilson]

I really dislike buying schmartboards and soldering TSSOP/SOIC packages...

Sorry to not answer your actual question, but it's worth pointing out that SOIC and TSSOP are very different beasts. I'd really recommend you become comfortable hand-soldering SOIC packages. No Schmartboard required... you can buy very inexpensive SOIC adapter boards (for example, I've used many of these: http://www.dipmicro.com/store/PCB-SOIC-SSOP16. AdaFruit has some as well). There are so many interesting parts that simply aren't available in DIP packages, but if you're lucky are available in SOIC instead of a finer-pitch package. Make sure you use flux and good (which includes being small diameter) solder and you can pretty much just solder SOIC packages one pin at a time.

TSSOP is twice as small (or worse) as SOIC, but with flux and good technique is still pretty doable without needing to buy expensive/bulky Schmartboards. A lot of the adapter boards have SOIC on one side and SSOP on the other side, so you only need to stock one kind of board for a given package pin count and you're covered. It's really not that bad and keeps a whole world of interesting parts open to you.

[Andreas]

Hello,

I did some further tests regarding the LM399 stability with my LM399 #3 reference after I have removed the socket and soldered the LM399 directly with long leads on the bread board.

Again the results might also be valid for the LTZ1000 reference. So I was not shure wether I should post it here or on the LTZ1000 thread. But since the thermal pictures from LM399 are in this thread it fits better here.
The question from the LTZ1000 thread is still how to achieve the best stability.

I found that on LM399 there is a voltage deviation depending on orientation in room (tilting of LM399). On my LM399 references this is around 3-4 ppm. On a Keithley 2000 I have found even larger deviations of readings depending on flat on table or placed on left or right or top or bottom or front or back side.

The LTZ1000 has a better internal construction. But also some tilting effects can be seen. (see discussion of Dr. Frank and me in September 2010 on the volt-nuts forum).

------

First experiment is LM399 #3 with long leads mounted into a folded box to keep direct air currents away from LM399. (see picture IMG1458w)

Each of the 6 possible orientations (0-5) has been tested with >=15 minutes settling time and 5 minutes measurement time. The 5 minute values are averaged. Measurement setup consists of a 2:1 precision voltage divider and a LTC2400 based ADC. My measurement standard deviation is around 0.25ppm giving up to 2uV  uncertainity with a confidence level of 2 sigma.

The result is on 20140102_LM399_3_boxed.PNG.
Difference between orientation 1 (= pin 1 marker LM399 on bottom) and 3 (= pin 1 marker on top) is 18.7uV or 5.4 ppm as maximum.

The same test with thermal isolation of LM399 #3 by a cotton pad on the solder side and a cotton pad on top of the LM399 fixed with a expanded plastic sleeve (see IMG1452w).

The result is on 20140101_LM399_3_cotton.PNG
The maximum difference is again between orientation 1 and 3 but this time only 4.7uV or 1.4ppm as maximum.

So the thermal isolation seems to minimize the tilting effect on the LM399.

to be continued ...

[Andreas]

Hello,

next setup is with the PCB from branadic which was used for the thermal measurements:
For mechanical fixture and air current shielding I put the pcb on one styrofoam sheet and used a second with cutout to fix the pcb within a plastic box. (IMG1465w).


A piece of additional foam (which was used on the thermal measurements) was put on the top of the 2 LM399s. (IMG1464w).

The 2 LM399s are with shortened legs. I call the one with the closed PCB the CH6 and the other with the slotted PCB CH7. The orientations are the same as with my LM399 #3.

0 = upright
1 = pin 2+3 upside (pin 1 marker bottom)
2 = pin 1+2 upside
3 = pin 1+4 upside (pin1 marker top)
4 = pin 3+4 upside
5 = upside down

Unfortunately now I have to measure 2 devices in each orientation. The measurement duration is longer than I can manage within one day when doing my other ageing measurements. Interruptions within measurement are shown as a short dip to orientation zero within the diagram.

The result for shortened legs without slots is shown in 20140104_LM399_CH6_foam.PNG

The maximum difference is about 4.4uV = 1.3ppm which is comparable to the LM399 #3 value with good isolation. Although this is no prove that long and short legs make no difference with good thermal isolation  (since not measured with the same device) it seems to be a indicator for that.
I wish I had equipment with better stability.

The result for shortened legs with slots is shown in 20140104_LM399_CH7_foam.PNG

Here the maximum difference is about 10.1 uV = 2.9 ppm which is significant worse than the CH6 value.
This is what I had expected from the thermo scan pictures since the temperature gradients on the slotted area are much larger than on the non slotted area.

further measurements will follow ...

[Andreas]

Hello,

3rd setup of branadic PCB with better thermal shielding similar to LM399 #3:

See picture IMG1467w and IMG1468w.
Cotton pads are on solder side fixed with tape.
And on component side over the LM399s fixed with a foam hose + tape.
All together put into the plastic box from previous setup.

Results:
20140105_LM399_CH6_cotton.PNG
the non slotted board shows 4.1uV maximum difference which is around 1.2 ppm.
So similar to the previous setup.

20140105_LM399_CH7_cotton.PNG
the slotted board shows 3.6 uV maximum difference which is around 1.1 ppm.
Around a factor 3 better than the previous setup.

But stop:
there are shifts in measurement between the two days which I cannot explain by measurement uncertainity.

Mhm, was there enought pre-ageing of the 2 LM399´s?

As branadic stated to me he had only done his thermal tests on the 2 LM399's. A pre-ageing was not necessary for this.
As can be seen in 20140105_LM399_CH_6_7_age.PNG the ageing is about 2.25 uV per day for CH6 and around 3.24uV per day for CH7.

So to say: shi(f)t happens.
I will have to do some correction calculations or repeat the measurements after some run-in phase.
In the mean time I can do some noise comparisons. (e.g. 0.1 - 10 Hz noise).

But anyway:
One result is already clear:
- nothing can replace a good thermal shielding.

There seems to be a second result:
- A slotted board is bad in case there is no good thermal shielding.
- From the first view: if I calculate out the ageing shift it seems to be that with proper thermal shielding the slotted board seems to have a slight advantage.

So the most important question for now is:
- what is the best material with the lowest K-value and at least fair electrically isolation (>> 10 MegOhm) which can be used for a hobbyist to shield a LTZ1000 or a LM399?

With best regards

Andreas

[Andreas]

Hello,

I have done the correction calculation for a linear ageing (linear regression coefficient).

The correction is applied on the averaged measurement values and not for the original 1 minute values.

The spread values with the foam do not really change:
CH6 foam corrected:   4.4 uV = 1.3 ppm (unchanged)
CH7 foam corrected: 10.4 uV = 3.0 ppm (was 10.1 uV)

The spread values for the "cotton" isolation seems even to be better for the slotted PCB:
CH6 cotton corrected:  4.6uV = 1.4 ppm (was 4.1 uV)
CH7 cotton corrected:  2.4uV = 0.7 ppm (was 3.6 uV)

remember that standard deviation for my setup is around 0.25 ppm or 1 uV.

With best regards

Andreas

[Mickle T.]

Thank you, Andreas! It's very interesting result.
I tried to reproduce it with my old triple LM399H 10 V reference connected to the HP34401A DMM. Reference is on-line 24/7 about six months, LM399H without thermal isolation, PCB without slots, HP34401A preheated. Maximum deviation (upright->upside down->upright) is about 0.5 ppm.

[Andreas]

Hello Mickle,

your experiment orientations are that what I call orientation "0" and "5" where I also have only little differences (up to around 0.5ppm) in my setups.

The pin 1 marker directs to the right side on the photo.
So the critical directions would be the putting them on the left and on the right (connector) sides.
I guess that in this case there would be also about 3-4 ppm difference.

When looking at photos from older HP34401a or Keithley 2000 units the critical directions should be putting them on the left and on the right side. The pin 1 marker shows either to the left or the right side in this case.

With best regards

Andreas

[Mickle T.]

Andreas, you are quite right!
I repeated all the measurements with Datron 1071 in 7.5-digits mode (4 hours preheating). The difference (in critical directions) was 2.1 ppm. Orientation "0" and "5" gives 0.3 ppm.

[Andreas]

Hello,

as promised now the noise measurements of LM399 in different configurations.

I use a amplifier   x 10000 with a 0.1 Hz .. 10 Hz limitation.
3rd order high pass (+ 10 sec measurement time) and 4th order low pass.
Readout is done on the oscilloscope.
In this case the 5mV / DIV on the scope corresponds to  0.5uV / DIV at the input.
Of cause all measurement are done in a metal cookie box with batteries.

I did 18 measurements in each configuration in upright orientation (see pictures above).
The plastic box + the foam sheets were used to hold the pcb in orientation.
Result is average peak-peak noise and the standard deviation of the 18 measurements.

CH6: cotton (both sides) + foam hose thermal isolation (non slotted board).
average: 2.89 uVpp   stddev: 0.32 uVpp

CH7: cotton (both sides) + foam hose thermal isolation (slotted board).
average: 2.45 uVpp   stddev: 0.27 uVpp

CH6: foam isolation on top (non slotted board).
average: 3.03 uVpp   stddev: 0.29 uVpp

CH7: foam isolation on top  (slotted board).
average: 2.46 uVpp   stddev: 0.21 uVpp

In this case I see no significant change between
good (cotton) and poor (foam) thermal isolation.

CH7 has slightly better values.
But this is most probably due to the individual noise level of the LM399.
(Has nothing to do with the slotted board).

For me it seems that measuring 0.1 .. 10 Hz noise is no good
way to judge the influence of a mounting method
on voltage reference stability.

With best regards

Andreas

[Mickle T.]

Noise measurements of 6 LM399H in simple one resistor configuration. Used EM Electronics nanovolt-amplifier with LF filtering (0.14-10 Hz).
LM399h No1:  6.6 uV p-p;
LM399h No4:  4.5 uV p-p;
LM399h No6:  5.3 uV p-p;
LM399h No7:  5.6 uV p-p;
LM399h No9:  6.8 uV p-p;
LM399h No10: 6.3 uV p-p.

Noise measurement of triple LM399H with constant current supply and simple filtering (see pictures above):
Pass No1: 0.88 uV p-p;
Pass No2: 0.87 uV p-p.

[Andreas]

Noise measurement of triple LM399H with constant current supply and simple filtering (see pictures above):
Pass No1: 0.88 uV p-p;
Pass No2: 0.87 uV p-p.

Hello,

That's a very good result: what is the band with of the filter?

Generally I have also large differences on my LM399 references:

The theoretical value on buried zeners with around 1mA is around 0.6ppm (p-p).
So for a 7V reference the theoretical value should be 4.2uVpp.

LM399#1 (LM334 temperature compensated current source) av: 3.85 uVpp s: 0.36 uVpp
LM399#2 (LM334 temperature compensated current source) av: 6.55 uVpp s: 0.35 uVpp
LM399#3 (LT1013 bootstrap current source) av: 3.56 uVpp s: 0.42uVpp

CH6 + CH7 are simple resistor configurations. (But with low noise LT1763 voltage regulator).
But since the dynamic resistance of the LM399 is around 0.5 Ohms
the difference current source / resistor has only very low influence on the 0.1 - 10 Hz noise.

With best regards

Andreas

[iTist]

Hello,

i am following both Voltage Reference threads the LTZ 1000 and the LM399 one.
I am new to the reference tobic only build a LT1027 based one but i want to do the next level step :-)
i think its the LM399. Inspired by the forum i have  a question about the trimming of the 10 Volt output Voltage.
Is it possible to trim the voltage with an DAC?
I have attached an simpified Schematic.


Thanks in advance.

Greetz
Oliver

[Andreas]

Hello Oliver,

yes this is possible.
At least the FLUKE 7000  voltage standard uses this.
Look for the brochure 10777-eng-01-a.pdf on the Fluke page.

You should use the LM399 output as reference voltage for the DAC.
And of course your feedback resistors should have a very low tempco so that the LM399 TC is not compromized.

With best regards

Andreas

[Mickle T.]

Hello Oliver,
There are many russian voltage standards uses this too.

[iTist]

Hi,

thanks for the answers. I made a more detailed schematic (see attachment). Do i need a separate reference output for the DAC (REF which is a bufferd output of the LM3999) or could i use the 10V output. I thought i build a Voltage divider(switched Capacitor with an LTC1043) to feed the DAC.

The 9k and the 20K resistors need to be low TC ones. The 100 ohms and the 1k Resistor aren't so critical is this assumption right?
Sorry for my silly questions but I am just a hobbyist who is interested in Reference Voltages.
Is it better to use one 14 or 16 bit DAC or two (cheaper) 8 bit DACs. For the 8 bit DACs i wouldn't need a µC. I can trim it with simple DIP switches.

Thank you for your comments.

Greetz

Oliver