LM399 with external oven

[NaxFM]

I'm trying different approaches for a design i want to build, and I found asking myself: "Does it makes sense to use the LM399 without its heater but relying on an external oven?"

For use as a reference outside of an instrument it really doesn't make any sense to run the internal heater at a scorching 90C, something like a warm but comfortable 40-45C should be sufficient to reduce the output noise, heat the whole board and improve thermal isolation, so i want to try it.
However, before designing and ordering the PCBs, i wanted to ask if you have any experience with this, just to be sure if it's feasible. I'm just starting to work seriously with precision equipment and components, so i have many doubts, and my main doubt for this project is: "What if the LM399 is optimized to run at 90C and a lower but constant temperature makes things worse?"

[Kleinstein]

It is an interesting question. A lower temperature could reduce the long time dirft and one can expect a little less noise, at least for the higher frequency part. For the 1/f noise part and popcorn noise things are not so easy: the higher temperature may well shift things to a higher frequency and this way get less noise at the very low frequency end that really hurts. The 1/f noise curve is only an approximation and the spectrum may actuially get flatter to very low frequencies.
Especially with the popcorn noise that is very common with LM399 more frequent jumps can be a good thing, even though the reference may than look more noisy.

A lower temperature may reduce the power needed, but the extra heater would add to the size and this tends to increase the power needed. So the reduction in power may not be that large.

[branadic]

In this case you can forget about the LMx99 and use LM129/329 instead. With a bit of luck you can find some NOS in the hermetic package somewhere.

-branadic-

[Andreas]

"What if the LM399 is optimized to run at 90C and a lower but constant temperature makes things worse?"

Its most probably not. Otherwise we would not have such a large stray between unheated and heated zener voltage.
The best zero-TC voltage for a LM399 is near 6875 mV.

https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg877716/#msg877716

The other question is: how good can you thermally isolate a package which has much more size than the tiny TO-46 package?
I fear heater power wise you will not gain much.
And keep attention to the temperature sensor(s) of the oven. The LM399 sensor is not placed optimal so that you get output voltage variations depending on orientation of the LM399. (Tilting effect).

with best regards

Andreas

[branadic]

The best zero-TC voltage for a LM399 is near 6875 mV.

I'm still not convienced about that theory, as I remember also the following image of that diagram added with measurements I did on several specimen.

-branadic-

[Kleinstein]

The TC of the LM399 and other compensated zeners does not have to be a fixed number, but the 2nd order TC can become quite relevant. This especially effects those units with a small TC.C(heated). The units with the smallest difference are likely the best at some 60 C( aout the middel of the range), but not decessary good at 30 C or 90 C.  Chances are the units with a small TC at around 90 C would still have a significant change from unheated to heated operation.

Unheated the LM139 would be an option too, but these are long obsolete and only rarely come up used. AFAIR there was a ref. system with multiple LM139 in a custom oven.

With a rather constant supply / current I don't think it would be that important to have the temperature sensor that close to the reference.
With an external oven the temperature control may not be as good. So a good TC at the choosen temperature can help - so for this case those ref with around 6900 mV may indeed be good.

[srb1954]

In this case you can forget about the LMx99 and use LM129/329 instead. With a bit of luck you can find some NOS in the hermetic package somewhere.

-branadic-

This approach is used in at least one commercial product - the Startronics VS4 (aka Guildline 4410) voltage standard. This uses a total of 32 LM329A reference diodes arranged in 4 groups of 8. The 8 diodes in each group are commoned together to reduce noise. All 4 groups, together with their scaling amplifiers, are placed in a common oven running at around 33C and the output of each of the 4 groups is separately available so that cross checking can be done between groups to check for relative drift and errors.

Since the oven is run at only 33C the long term drift of the reference diodes and the associated scaling amplifiers and resistors is possibly reduced compared to running everything at a higher temperatures. The VS4 is designed as a transportable reference requiring that the oven be kept running off batteries while it is being transported for calibration. The lower oven temperature requires less heating power translating into longer battery life.

The disadvantage of running an oven at such a low temperature is that temperature regulation can be lost in very high ambient temperatures. The VS4 is only rated for operation up to 28C ambient to ensure oven temperature regulation is maintained.

[MegaVolt]

Zener diodes SZA263 and LTFLU very often work with external heating.

[dietert1]

Yes, a reference with access to the zener voltage separate from the temperature compensated zener voltage (Ube of compensation transistor/Uf of diode) can provide an on-chip temperature measurement and fine-tuning of TC at any temperature. I posted a schematic in the LTFLU thread. As far as i remember the 7000 reference schematic posted in the LTZ1000 thread demonstrated similar methods.

Running a reference a lower temperature reduces shifts when using it as a portable device, with intermittent power. Using a peltier element that's an easy exercise to make an oven for 25 or 30 °C.

Regards, Dieter

[oz2cpu]

how about you add a new case, round the whole lm399 and its opamps and such,
and this case is now regulated to 40C or so.. now the lm399 and all the parts needed, will have a much easier life
to stay at exactly the same stable temperature

[branadic]

I'm not entirely sure that's really worth all the effort. It's good practice to adjust the zener current to the zero t.c. temperature where you later want to operate the surrounding oven at.
The LMx99 is not a pure temperature compensated zener so you can't adjust the current through the zener and thus you can't adjust the z.t.c. temperature, same goes for LM129/329. Not sure anyone ever tested unheated LMx99 before to find their z.t.c. temperature. Would be interesting to know where that temperature is for a given number of samples. On the other hand I expect a large distribution here, that's why the requirements for the oven stability are rather high, since many of the LMx99 are operating somewhere at the rising slope of the t.c. curve.

-branadic-

[miro123]

I'm trying different approaches for a design i want to build, and I found asking myself: "Does it makes sense to use the LM399 without its heater but relying on an external oven?"

Why onchip ovenized references are industrial proven standard?
There are many reasons for it, One of them is that inside temperature is tightly controlled.
It is fundamental of control theory.  Having actuator as close as possible to the feedback sensor guaranties great phase margin and higher BW of control loop.
Maybe my explanation is too abstract. I''ll give you a real life example - Go to the shower with  and try to archive  comfortable water temperature by turning hot and cold water taps - It is difficult isn't it? Now remove the long pipe and control the water temperature directly from taps output - It is much easier isn't it?
That is how  the most convesional temeprature control systems work. They are 100% feedback control based
Modern digital control oven use advanced techniques like feed forward control based on model or observer or prediction blocks

My point is that you cannot achieve the tight temperature control of internal LM399/LTZ100/ADR1000 heater

[dietert1]

One disadvantage of the on-chip oven is that there will be other critical parts without oven, i mean unless one can use the 7.xxx V as an ADC reference in a multimeter. For a 10 V metrology reference the external oven is a good solution as it can keep the reference gain stage and output buffer at constant temperature as well. And it can be a Peltier "oven" running at lower temperatures. There are reference chips with on-chip temperature measurement that do support tight temperature control. I posted a schematic in the LTFLU thread.

Regards, Dieter

[Kleinstein]

As the power consumption of the reference chips is essentially constant, one can still get pretty good temperature control with an external oven. As long as the insider of the oven is not changing, if would be OK to just regulate the outside.

The chip internal regulator is in may cases limited to simple proportional control and is thus somewhat limited in accuracy. With the tight couling it can still work reasonably good. An external oven can use PI or even PID control to compensate for much of the disadvantage from having the sensor and heater more separated. So it is not that clear that a chip internal temperature control is actually more accurate. A sensor directly on the chip would defenitely help.

[miro123]

@Klreinstein & Diether - can you give some numbers of 24h stability of yours solution.
I hardly stay that with simple SISO, PID control and DIY sensors and heater cannot achieve decent stability in 24h++ range.

Back in old day Fluke 732 achive this with their MISO analog control - https://www.eevblog.com/forum/metrology/fluke-732a-temperature-control/
It is needless to say that Fluke carefully selected insulation mechanics and accurately placed multiple sensors
Today such solution is much easer to achieve and fine tune using digital control.

[Kleinstein]

I don't have 24 h data (only shorter time) for the stability for my analog temperature regulator. Given that it is only a LM358 (inside the oven) with a diode, I would expect something in the 0.05 K range. For the long term the plastic case of the transistor used as a sensor may be an issue. For me this is good enough with a relatively well adjusted TC voltage reference (JFET based) and the temperature regulation mainly to fight the square law part.

Edit: I found a stability test for a bit over 2 hours:  This is the sensor also used for the regulation and may this be misleading. Chances are much is LM358 noise. Some 5-10 µV_pp noise over 1 hour sounds plausible.

[dietert1]

As you wrote thermal stability is a physics problem, if one does everything right with the digital controller and with the temperature sensors. For example i used the aluminum oven for heat distribution and the heat transfer to the pc board inside and its components constitutes a low pass filter. The on-board (actually on-chip) sensor is used for integration and the Peltier element has its own glass NTC to achieve rapid response against ambient temperature changes entering through the Peltier element.
I used a fairly large heatsink on the other side of the Peltier element and the whole setup is inside an outer case made of plastic. The heatsink is protected from ambient air draft.
I don't have access to the logs right now, but 24h stability is better than 1 mK (RMS). In the winter there used to be a 10 mK excursion when the heaters in the lab turn on at 7:00 in the morning. After that ambient air tempature rises from 16 °C to 20 °C within 45 minutes.

Regards, Dieter

[MDO]

LM399 are interesting parts and i have quite a few, i was thinking to adjust the zener current and try to find the zero t.c temperature for a couple of unheated samples. My plan is to use my old calibration equipment to get the zero t.c zener current.

The plan.
1: Use my Tek Know TC400 dry well and ice bath to set the temperature.
2: Using my process calibrator i can generate different zener currents for the references to find the optimal value.
3: Read/plot zener current, zener output voltage and temperature using the process calibrator and a Datron 1071 for precision voltage measurements.

The equipment specifications.

TEK know TC-400
TEMPERATURE RANGE: ~Ambient to 400°C
ACCURACY ±0.3°C
STABILITY ±0.03°C
RESOLUTION 0.01°

Process calibrator specifications: Voltage range 60V with 0.01mV resolution, Current generation range 0…25 mA uncertainty about 1.1-3uA

The Datron 1071 is 7.5 digits and quite stable over time so it should do alright in these tests.

Soon when i get some vacation time of work i will start working on this

[Kleinstein]

The LM399 reference part has more than just a zener. The actual zener current is regulated internally and unless going too low the TC does not change much with changing the current to the reference part.

AFAIK the datron 1071 has a LM399 or similar reference. So it would be a comparison of two comparable references only.

[iMo]

Yep, the LM129/399 zener's current is set to 250uA internally, and stabilized, afaik. I doubt you could force the zener's current more than, say, +/- 20uA from that value.
Have a look at the 399's schematics in its datasheet.
The other suitable part for experimentation will be the ADR1001, when it gets finally available. You may set the zener's temperature with an external resistor, as well as the zener's current..

[alm]

AFAIK the datron 1071 has a LM399 or similar reference. So it would be a comparison of two comparable references only.

I'm pretty sure it has one or two selected 1N82x temperature compensated zeners. I'm not aware of a Datron instrument using LM399, just temperature compensated zeners or LTZ1000.

So it might be a bit better than the LM399 in noise and drift.