LM399 based 10 V reference

[dietert1]

Looking at this image i noticed a wiring error. The second PWM DAC (ripple cancellation) got wired to flip flop pin 10 instead of pin 8. Pin 8 remained unconnected. In effect ripple cancellation was implemented but remained inactive. Some days ago i corrected that mistake. PWM output shifted by -2 +/- 0.5 ppm and noise is lower now. Right now the HP 3456A used for logging delivered more than 18000 consecutive readings with no other results than 10.00000 and 10.00001 V. Need to improve logging for better resolution.

Some months ago i got a batch of 10x LM399AH and made two more 5x LM399 modules like the one shown above. When i started fitting them into an enclosure i got the idea to change the circuit into 5x 14 V instead of 10x 7 V. Then the PWM DAC divides the 14 V average down to 10 V. The DAC will no longer be in a control loop, but just feed an output buffer. And the reference will have programmable output voltage proportional to PWM ratio. I already got the ten heaters and the 5x 14 V pairs working. Power consumption of the ten heaters is about 1.8 W but there will be more insulation. This concept can be very cost efficient, since a TEC controller isn't needed.

Regards, Dieter

[Kleinstein]

Starting from some 14 V (2 reference in series) and go down with a PWM divider makes some sense, especially with a variable PWM ratio, to also get voltages different from 10 V.  A similar configuration is used with some of the fluke calibrators.
Driving the 14 V reference may need an auxiliary voltage a little higher than 15 V, so likely more than 15 V supply needed.

With multiple LM399 in parallel it may be worth having a way to measure the individual reference too, to identify more drifty units and also the higher / lower noise ones. The noise can be quite different between units and a single noisy unit would effect the average quite a bit. Ideally a more noisy unit would get a larger resistor to contribut (R ~ En²) to get the lowest noise for the average.

[dietert1]

Yes, the supply is completely different now. The heaters are running from 19 V. The same voltage is used to drive the dual n-channel mosfet for the DAC output stage. And there is a 1.2x amplifier to derive 17 V from the 14 V reference voltage. This amplifier isn't very critical, deviations get attenuated by about a factor 3000 (2 * 0,5 Ohm / 3 KOhm).
Then there is a buffer between reference voltage and PWM output stage. So, together with output buffer of the PWM filter this time there are two opamps involved with their errors. The output buffer may need bootstrapping for perfect linearity, don't know yet.

Regards, Dieter

[TiN]

Was the ref powered all the time with no interruptions? Drift can't be estimated easily with just few points for not permanently on zeners.

[dietert1]

Yes, after there was some interest in this PWM reference i kept it logging continuously. Will show results later.

Regards, Dieter

[TiN]

Tested LM399 out of Keithley 2001 for operation at sub-zero temperatures. Perhaps this is interesting for someone other than me. 

Device is SL40057 IC-257 (custom Keithley P/N for selected LM399) date code 30 week 1991. Heater part of LM399 was not connected to anything, and "zener" part biased with +24V from Keysight E36312A and 11.1kOhm BMF resistor (static at ambient temperature). Output (kelvin-connected) from zener circuit was monitored by trusty Keithley 2002 DMM.

Run took about 1 hour, and temperature of GM cooler cold-tip with attached LM399 (without plastic casing) was reduced from ambient +27°C to -262°C.



Reference stopped working at around -185°C or so, with current drawn from PSU reduced down to 770uA (vs 1.45mA nominal) and output voltage going over 8V, all the way until 15.4V. 

Crude logging equipment setup:



I guess there is too much going on inside LM399 to operate this much far out of specified 0°C 
Next specimen is obvious LTZ1000...

[dietert1]

The curves don't show a sudden failure. Can you say to which degree this test was destructive for the poor LM399?

[saturnin]

LM399's output voltage is out of specs at room temperature already (the datasheet says its max reference voltage is 7.3 V). Has it been tortured in another experiment?

I also miss what the purpose of the whole exercise is. It is well known fact that silicon devices generally do not work well at deep cryogenic temperatures (BJTs are worse than MOSes in this respect). In presented case extremely low temperatures were reached (~ 10 K). No surprise LM399 ceased to work at some point...

[ramon]

Just for testing, I guess.

His eyes looked around the table to find which creature can be sacrificed for science. And he found that LM399. Testing a resistor noise could had been more interesting .. or boring (who knows!). But just for testing the setup I think that it doesn't matter which device you choose.

The final goal might be to find the ultimate VOLT reference.

[FransW]

A reminder:

Absolute measurements do NOT exist.

Frans

[razvan784]

Absolute measurements do NOT exist.

Number of people inside a building.
Number of molecules inside a vessel.
The imbalance in electric charge between the plates of a capacitor, expressed as a number of electrons.
Should I go on?

[donlisms]

Are counting and measuring the same thing?

[IanJ]

Jings!......thats a chart you don't see very often!.......I can feel the cold from here!

Ian.

[MegaVolt]

This is the first version schematic.

What program allows you to draw such beautiful circuit?

[snik]

It could be sPlan.

https://www.electronic-software-shop.com/lng/en/elektronik-software/splan-70.html?type=N&opt=%7B11%7D22&language=en

[voltsandjolts]

Yes, could well be sPlan.
But inkscape can also do similar nice diagrams. Once you have a few basic parts/blocks, it's surprisingly quick.

How about a +200C run for the LM399?

[iMo]

After I collected and burn-in a bunch of 399s I am thinking to design a small 10x10cm large board for 4xLM399, LT5400 + OP07/2xNPN.
Here is the schematics - it is a classic design, what would be great is a feedback from experts here
a) on resistor's values (and types, TC)
b) capacitor's values and types, and
c) the layout - here the rectangles depict the star points, every wire will be a separate track on the pcb.
The power supply will be off this board.

PS: on the first glance the 100nF at the 399 zener should be modified as the ADR1399 has got some specific reqs for the stability..

[Kleinstein]

C14 and L2 don't combine that well. I would tend to leave out C14. For the supply I would consider a common mode choke, if really externally. Some Protection diodes would also be nice.

I am not a big fan of the foild caps directly parallel to the LM399. This may make the life for the internal stabilzer more tricky. Chances are some 10 Ohms in series each would be a good idea. This is still relatively little compared to the higher frequency output impedance. It kind of sets a realistic traget for the regulator circuit.

Chances are C1-C4 can be smaller, like 1µF local and than the common C18. 10 µF MLCC tend to be low voltage and show much capacitance drop at some 15 V.

C9 makes the regulation rather slow, allowing for a lot of capacitance at the output. I think some 1-10 nF should be enough.

P.S.:  with the relatively large C9 the supply hum suppression would be not so great. So maybe split R9 and add some local filtering there.

[iMo]

There are 2 common mode chokes (small FT toroids) now.
Not sure where to ground the new C14 (R9 split)..

PS: Version 3 - Added 78(L)15 for analog only..

What about R1, R2, R3, R4 values??

[dietert1]

Your circuit looks perfect.
In the last LM399 array i made i used an opamp to derive the power supply from the avaraged reference voltage as well, so it's also stable to 10 or 20 uV. Of course a 7815 with low TC serves as well.
The output circuit stability/transient response can be optimized with a pulsed load.

Regards, Dieter

[Andreas]

There are 2 common mode chokes (small FT toroids) now.

PS: Version 3 - Added 78(L)15 for analog only..

What about R1, R2, R3, R4 values??

Mhm,

10 uH  is rather small for a common mode choke. I would take at least 50 uH.
Typically you need a PI-Filter arrangement so at the output a capacitor is missing.

What are the 10R resistors doing in series with the 100nF?
The 100 nF are useless in this case because the LM399 has 0.5 - 1 Ohms dynamic resistance.
The 100 nF are not functional (not reducing noise) but are a EMI-Filter.
https://www.eevblog.com/forum/metrology/emi-measurements-of-a-volt-nut/msg2684100/#msg2684100

There is also a T.C. dependant on heater voltage at the LM399. So why only stabilizing the analog part?

The R1-R4 values have to limit the cross currents well below 1 mA. With typical +/-100mV.
So the 560 R look about right to me.

with best regards

Andreas

[dietert1]

I am using 274R resistors that we had for averaging, as the LM399s i got agree to within +/- 15 mV. As the average is near the middle, max currents are 15 mV/274 Ohm = 0,055 mA or so. The idea of the 100 nF caps close to each LM399 is to protect the zener regulator from external EMI. The averaging resistors can run from the caps, leaving the LM399 output pins with very short traces (< 5 mm). One can also try and place the averaging resistors close to the LM399s and omit the additional filters 10R+100n.

Regards, Dieter

[iMo]

The 100nF ser 10ohm - EricM indicated the new ADR1399 is more sensitive to capacitive loads than the LM399..
The closest components around the LM399 will be aprox 20mm off its pads (there will be the "ring")..
Needs to be investigated (EMI vs. stability aspects)..
The heaters "power" voltage will be stabilized too (say 18-19V). It will be a separate PCB.
All components TH except opamp and LT5400 (provided it fits the PCB)..
PI filter - the output cap could be added (ie 1nF cer).
Btw. - I do not like ceramics in the signal path..
I've increased the spread of the 399's voltages - you may see the currents through the averaging resistors.

The cathode's and averaging resistors will be 25ppm/K one (?).
Below a sim of TC (ppm) with those resistors between +10 to +25ppm/K.. (refs and opamp and LT5400 are 0ppm/K here).
The copper traces will add up some TC too (could be simulated as well).

[Kleinstein]

The LT5400 is a bit difficult to solder (especially the thermal pad, that is not absolutely needed). A ORN type network in an SO8 case would be a possible alternative, though with not as tight specs.

The averaging resistors can also be larger. The added noise is still small compared to the LM399 noise, and the filtering action with the 1 µF cap would get a little more effective.

For EMI prodection it may have to also include some filtering (e.g. Ferrite beeds or sereis resistors) with the individual reference ouputs.

Even with some series resistance the capacitors would still be daming high frequency interference.  In the relevant frequency range where the capacitor has any effect the output impedance of the LM399 is no longer at the 0.5-1 Ohms level valid for low frequency. A 100 kHz it is more like 10-50 ohms and quite inductive (some 10 µH). So a pure capacitance can cause a resonance and some series resistance helps to dampen the resonance.

[dietert1]

Apropos "ring": One problem to think about with LM399 arrays is board temperature. The larger the array, the closer the board temperature rises to oven temperature, possibly causing one of the ovens to shut off. Since that risk depends on ambient temperature, one should watch oven currents. It's a pitty one cannot adjust each oven temperature.

Regards, Dieter