Experiments with Vref Ovens

[Andreas]

EDIT: I've added a photo of the daughterboard if anyone is interested. It's double sided and can take 120 x REF102's.

I fear you will have to increase the setpoint temperature when the 120 REF102 are heating (about 2W additional power).
I have ~30 references in my ageing box and temperature without "heater on" is 30 - 40 deg C depending on environment temperature.

with best regards

Andreas.

[essele]

Hi Andreas,

Yes, I think you are right. At the moment, with two layers of insulation and an ambient of around 22.8C, I'm only needing 1.4W to sustain 44C so I don't think I'm realistically going to be able to put more than 60 or so in at the current set point.

The other option would be to reduce the insulation again and use the external temp sensor for better control ... at the moment I'm just using the internal ADC results, I'm trying to get that as good as I can before I introduce additional inputs.

Lee.

[dietert1]

Thanks for interesting link. Several things come to mind.
a) Operate the array of plastic references in a controlled humidity environment.
b) For the physical arrangement there should be a massive heat conductor in touch with each reference that keeps the temperature of all references similar. Otherwise you will have ever changing thermal EMF in the system and that is difficult to fix later. Maybe an aluminum carrier PCB.
c) There should be redundant temperature monitoring in case something fails. When you read that report, the biggest shift happened when the reference had to be repaired. The repair took nine months, strange. I also had that problem with a reference to reproduce the previous temperature precisely.
d) The heat problem can be solved using a TEC to keep the alu box cool and at a constant temperature.
e) Statistics roughly says that averaging 100 references you get a factor 10 better than the base unit. So when building such a beast i would include a MUX. Not only for groups as in the linked report, but for each single device, similar to what Andreas did. Then you can make something like a Maxwell demon to sort things out a little. Depends on what you want to do with your reference.
f) Get access to a national reference lab...

Regards, Dieter

[essele]

Thanks Dieter ... I have a plan for some of these (which may of course be misguided!)

(a) That was the theory behind the oil ... I'm assuming that if everything is immersed in oil then there's very little chance of humidity changes affecting things ... I am completely guessing though, so keen to understand more.
(b) Again I was hoping the oil would help to do this.
(c) I have the capability to add I2C sensors pretty easily, however I'm not sure what failure scenario wouldn't be detectable with the ADC. Any major failure should cut power to the heaters anyway. But I will give this one some more thought.
(d) I have thought about this, but was trying to avoid it.
(e) I have worried about detecting misbehaving references, but I don't think muxing them individually is practical. I'm going to test each one prior to soldering and then just hope. I am also thinking about making it a bit more modular, which will help a little.
(f) Yes, that would be great! Do you know anyone who'd let me into one here in the UK? 

Lee.

[essele]

Validating your oven temp can be a difficult challenge ...

I've been seeing quite some swings on my thermistor measurements (this is just for validation, not part of the control loop) in the order of several tens of mK, at the same time my ADC measurements have typically been within +/- 4mK ... so I've just put in a four wire PT1000 as an additional measurement.

During the same 30 minute period I saw the following results...

This is the PT1000 (measured on a DMM7510) in four wire mode...


This is a 5k thermistor (measured on a 34470A)...


Quite mad ... there's presumably some level of equilibrium going on (with the PT1000, which was new into the device) but the reality is that the temperature should have reduced (because I removed the lid to put the PT1000 in) and then should have been increasing again one the lid was back on.

So I believe the PT1000 at this point, but I'm going to continue watching both of them for a while!

EDIT: Err ... user error. Both devices weren't particularly well fixed and the oil level was slightly low so I think I was getting a bit of a hybrid oil/air temp. Now that they are better fixed they actually match each other reasonably well ... I'll still keen an eye on them for a while.

[dietert1]

Sorry i did not read about the oil before.
Adding the PT1000 is exactly what i meant by redundant temperature measurement. In case one of your sensors fails, you will be able to fix the device without loosing calibration. And it's a good idea to measure the sensors with calibrated instruments, i mean outside the temperature control circuit.
In my failed Geller reference (AD587) i only have one NTC within an analog control loop and i am not sure whether i will ever get the same temperature again. Don't know yet what failed...

Regards, Dieter

[essele]

So I've been logging data for part of the day, unfortunately (or maybe fortunately) I'm also doing my day job so the door has been open and closed and the room heating and cooling a little bit.

There are three graphs in the attached image:

Top: ktemp -- this is a 4wire PTC connected to an external DMM, it's in the oil very close to the onboard PTC's.
Middle: ambient -- this is the room ambient temperature, very near the oven (I also have humidity and pressure data but it doesn't seem that relevant to this)
Bottom: bbtemp -- the temperature as calculated by the ADC using the baseboard PTC's and controlled by the heaters.

I've been battling with why I'm seeing several 10's of mK variation in my validation temps whereas the onboard ADC seems to be within a few mK. In the attached graphs you can see that ambient swings roughly 2 degrees (from 25.5 to 27.5), at the same time the external validation swings about 50mK (~ 44.66 to 44.71) but the ADC measurements only move by about 5mK (43.999 to 44.004)

The graphs now make me think that the 50mK variations are actually caused by the negative temperature coefficient of the external DMM (a new (to me) DMM7510 in this case), the specs say 0.003 degrees/degree TC (but when when outside of +/-5 from Tcal, which I am outside of in this sauna!) Interestingly it doesn't quote an accuracy, only a "resolution" which is 0.01 degrees ... not sure I understand what to take from that!

So my thinking at this point is that my ADC seems to be the most accurate measurement (especially since it is itself in the oven) and the system is working really quite well ... so therefore I can believe my ADC numbers and go back to focusing on the control algorithm, and perhaps adding some hardware filtering in the next board revision.

Does this theory sound reasonable? Or am I missing something?

Thanks,

Lee.

[dietert1]

50 mK * 0,39 %/K of the PT sensor would correspond to 200 ppm/K. Difficult to believe a DM7510 has such a problem.

I observed something similar with my LTFLU ovens. In the bottom, were the flat cable enters, i have an NTC inside and the TEC outside of the alu box to get a fast response for the TEC controller. With the Arroyo TEC controller i can keep that NTC to some mK.

But then a second temperature measurement from the LTFLU reference chip itself (TC of transistor Ube) yields a temperature offset and temperature variations correlated with ambient. This is feed-through from the other sides of the alu box. I can use that signal for a second control loop and fine-tune the reference chip temperature variations down to better than +/- 0.2 mK typical. That works with a GPIB DAC pulling the temperature measurement of the TEC controller by very small amounts. That second loop is slower than the first one since there is an air gap between the alu box and the reference. At the same time that air gap makes a low pass filter to keep rapid ambient temperature variations away from the reference.

Regards, Dieter

[essele]

Wow ... 0.2mK .. very impressive.

I had originally thought my issue was caused by some kind of feed-through, however it's negative TC is the thing that doesn't make sense to me! The negative TC wasn't obvious until I graphed it against ambient.

I don't think it's an "issue" with the meter because I had (what I think was) exactly the same behaviour with a 34470A.

Perhaps I'll rerun the logging using 4W resistance mode and then post-process the data, that way the spec should be a bit clearer.

It's quite frustrating ... I'm building the oven so I have something at a constant temperature so I can better understand TC in the meters and eliminate it from the reference, but I don't have anything that's at a constant temp so I can check the oven is ok! I'm jealous of the man with two clocks!

[IRFP]

Hello,
this is what I use for my measurements 

https://www.eevblog.com/forum/metrology/manufacture-an-oven-for-measurements-in-a-temperature-controlled-environment/

[enut11]

Hi @IRFP. Interesting design and able to condition much larger projects.

Do you have any temperature/time charts of it working?
enut11

[IRFP]

Hello,
The project is recent, it is in the evaluation and adjustment phase.
Cordially

[enut11]

When you have two ovens each one with its own voltage reference, it's more fun. Then you can measure the voltage difference (to 100 nV in the lowest 3456A range) as a function of the oven temperature while keeping the other temperature constant. You will need to use Low Thermal EMF connections to get that precision. Then you can determine TC of each reference as a function of temperature, find a good operating temperature where the curve is near flat, determine reference noise etc.

Regards, Dieter

Is it also possible to determine Vref noise when two 10v refs are connected opposed and measuring the difference on a 6.5 DVM? If so, what noise component is it?

[MegaVolt]

Is it also possible to determine Vref noise when two 10v refs are connected opposed and measuring the difference on a 6.5 DVM? If so, what noise component is it?

If we assume that the noise is Gaussian and the same for both devices, then the measured noise will be equal to the noise of each multiplied by the root of 2.

[dietert1]

A standard method with voltage references seems to be noise measurement in a 0.1 Hz to 10 Hz frequency range using a large capacitor instead of a second reference, and with a special low noise amplifier. That method yields numbers that can be compared. You can find several threads here on how to do it. Like one paralleling 16x OPA140 OpAmp channels.
What you want in the end is something predictable for months or years, not seconds. And that can't be checked with a capacitor, unless that capacitor is very special, ovenized... Better use a second reference to compare to.
For me, a DVM difference measurement resembles the intended application most. Starting with sampling intervals of 10 seconds and averaging at different time scales. Of course, everybody is tempted to look for a separation between drift and noise. Yet even after months you won't know whether an observed change was permanent or temporary (e.g. seasonal changes).

Regards, Dieter

[enut11]

Is it also possible to determine Vref noise when two 10v refs are connected opposed and measuring the difference on a 6.5 DVM? If so, what noise component is it?

If we assume that the noise is Gaussian and the same for both devices, then the measured noise will be equal to the noise of each multiplied by the root of 2.

Is that the sum of the 2 noise components times sqr root of 2?

ie (RMSnoiseRef#1 + RMSnoiseRef#2) x 1.414

[trobbins]

If any control/power/sensing cables are likely to provide significant thermal conduction through any insulation barrier and to a location that is being temperature controlled, then using the minimum practical gauge wire, and/or adding some cable loop length inside the insulated domain, may be worth checking.  As per standard thermocouple mounting practise.

[MiDi]

Is it also possible to determine Vref noise when two 10v refs are connected opposed and measuring the difference on a 6.5 DVM? If so, what noise component is it?

If we assume that the noise is Gaussian and the same for both devices, then the measured noise will be equal to the noise of each multiplied by the root of 2.

Is that the sum of the 2 noise components times sqr root of 2?

ie (RMSnoiseRef#1 + RMSnoiseRef#2) x 1.414

No, it is RMSnoiseRef#1 = RMSnoiseRef#2 = RMSnoise|| × sqrt(2).
This only applies if both have same noise figure, otherwise it will give false results.

[MegaVolt]

,ie (RMSnoiseRef#1 + RMSnoiseRef#2) x 1.414

Summ = Sqrt((RMSnoiseRef#1)^2+(RMSnoiseRef#2)^2);

if RMSnoiseRef#1=RMSnoiseRef#2

Summ = Sqrt((RMSnoiseRef#1)^2+(RMSnoiseRef#1)^2)=Sqrt(2*(RMSnoiseRef#1)^2)=RMSnoiseRef#1*sqrt(2);

This is one of the typical noise estimation methods of different standards. For example, for crystal oscillators, two generators are multiplied. For voltage sources, noise is observed between two identical ones.

Assuming that the sources are the same, we can assume that the noise that they produce is the same. Therefore, the formula is correct.

[dietert1]

Just finished some noise test with an Arroyo Instruments TecPak 585 and a VREF oven, maybe the post fits here.

Regards, Dieter

[MK]

What often causes problems for measuring and maintaining a very stable temperature is the self heating of the thermistor.
One product I worked on many years ago maintained the temperature of a water bath to 0.002 degrees variation over any week of interest. The glass encapsulated ones are very stable.

Another product I worked on oscillated when I inherited it, but dropping the bridge supply rails cured that as the self heating issue went away.

[dietert1]

Thanks for the proposal to use a bridge circuit to improve this measurement.

Meanwhile more experiments have shown that the HP 3456A used in the above tests has a TC of 0.54 ppm/K in its +/- 12 KOhm range. This is the TC of its reference resistor. This was determined by replacing the thermistor by SFernice foil resistors (10K2 + 1K3) inside the oven and waiting for some days of initial drift. When measuring the oven temperature using that HP 3456A with a 10K thermistor, the DVMs own TC contributes a temperature dependent error of 12,4 uK oven/K ambient (factor 11500 Ohm/500 Ohm more). When using the DVM for an outer control loop as described in my little report this effect remains hidden, but must not be forgotten. It results in much larger fluctuations than the 1.6 uK rms noise limit observed.

Next i will try to determine the own TC of the Arroyo TecPak 585. Will probably put it into an incubator and use another TecPak to run the oven.

Regards, Dieter

[dietert1]

Meanwhile there is an observation of an Arroyo TecPak 585, regarding the resistance measurement by that unit. It turned out to be worse than expected. The measurement of a very stable ovenized 11K5 resistor exhibited drift of about 140 ppm within some weeks. Would be interesting to find out which one is the reference resistor inside the TecPak and maybe replace it by something better.

There has been more info on reference ovens in the LTFLU thread, e.g. here: https://www.eevblog.com/forum/metrology/the-ltflu-(aka-sza263)-reference-zener-diode-circuit/msg2637174/#msg2637174 and recently calculations of oven power requirements: https://www.eevblog.com/forum/metrology/the-ltflu-(aka-sza263)-reference-zener-diode-circuit/msg3429710/#msg3429710.

Regards, Dieter

[iMo]

..
There has been more info on reference ovens in the LTFLU thread, e.g. here: https://www.eevblog.com/forum/metrology/the-ltflu-(aka-sza263)-reference-zener-diode-circuit/msg2637174/#msg2637174 and recently calculations of oven power requirements: https://www.eevblog.com/forum/metrology/the-ltflu-(aka-sza263)-reference-zener-diode-circuit/msg3429710/#msg3429710.

Regards, Dieter

Added the xls file there for the heat transfer estimation calculation..

[dietert1]

Yesterday i tested the experimental TEC oven driver shown in the image.
It has a Nucleo STM32L432 module, an ADS1256 ADC module, a linear power supply for the output stage with 4x IRF540 and a 0.235 Ohm shunt. The SIC473 buck converter adapts the supply voltage according to TEC output current, so the lower mosfets of the bridge stay cool and work in linear mode. The design TEC current is +/- 2 A.
The output stage is controlled by a DG4053 MUX for heat/cool switching and an OPA140. Hidden below the ADC module is another MUX for voltage reversal of the thermistor and its reference resistor (blue, 10K2, 5 ppm/K). Also there is a DAC8551 for TEC current control.
The thermometer with voltage reversal gives one measurement per second at a noise level of 10 uK (standard deviation). Noise of the resistance measurement is 0.4 ppm.
So i connected a little oven and implemented configurable PID control. Gains were adjusted by hand. They are pretty high and the first diagram shows the damped oscillation (47 s period). At steady state the effectve gain is less due to thermometer noise and TEC current digitization noise. The diagrams show temperature and TEC current logs at two different time scales. Oven temperature in °C in red and TEC current in A in blue. The oven and its heatsink are protected from air movement. During the last hour the oven temperature log exhibits a 11 uK standard deviation.
The thermometer circuit has unused channels, so one can also measure ambient or TEC heatsink temperature and/or use a second thermistor mounted on the oven lid.
Last image shows a similar four channel TEC output stage that i made for a LTFLU reference array. It has an OPA4140 and a DAC8554. There is only one buck converter as all four ovens run at similar temperatures.

Regards, Dieter