The LTFLU (aka SZA263) reference zener diode circuit

[branadic]

With the command "SETEXTENDEDRES 3 \n" you can turn on the reserve digits and the instrument will deliver 1000 times more resolution.

Thanks, that is awesome! Not sure where you got this information from, but in the Arroyo Computer Interfacing Manual for 5305 there is no such command listed? Are there more hidden commands?

-branadic-

[dietert1]

That information came from P. Corr at Arroyo Instruments, but i remember it's also in one of the documents on their website. This was introduced with one of the firmware updates. I have been using it with a TecPak 585, but as far as i understand it should work on the 5305 as well. and also on the 5235. I posted a little report on my TecPak oven experiments here:

https://www.eevblog.com/forum/metrology/experiments-with-vref-ovens/?action=dlattach;attach=1084858

Regards, Dieter

[branadic]

Thanks dieter1, unfortunately I wasn't able to find any hint on the Arroyo Instruments website or in the documents over there, even the firmware file doesn't include any hint. I already had firmware 1.44 installed on my 5305, but didn't know about that option. However, having 10µK resolution is crazy and amazing. 
As shown in my former post, the temperature of my oven setup at 25°C, some Kelvin above room temperature, shows low noise (<<1mK peak-peak) and good stability over hours, measurement still ongoing.
Next step, get one of the meters free to find the ztc point and measure stability of LTFLU at it. Since Autotune was able to find PID parameters, it should be possible to transfer them to an analog circuit for a fixed temperature, right?

-branadic-

[Kleinstein]

Depending on how the Arroyo regular is programmed it may not be so easy to transfer to an analog circuit. There are 2 possibly difficult points:
1) the time constants may be quite long, not very comfortable with an analog circuit, though likely still possible.
2) As a high end regulator the Arroyo may use linearization for the TEC to get improved regulation - this can be a little tricky in the analog domain. It may not be needed when using the TEC only at relatively low current and thus in the more linear range.

[branadic]

The PID parameters found are as followed:

P=0.4760217, I=0.0012075, D=4.1800661

Not sure about any linearization, I guess only Arroyo Instruments e.g.  P. Corr can answer that question. I have limited output voltage and output current to what I think an analog oven controller would look like, thus 12V and 130mA for the BPR10J101 resistor on a 12V supply, controller set to heat only.

-branadic-

[Kleinstein]

The PID parameters looks a little strange. From the ratio of the P and I part would get a time constant there of some 400 seconds (or whatever internal time units are used). This is relatively slow, but reasonable for the thermal system.

The ration of D and P part on the other side suggest a time constant of some 9 seconds for the differential term, which looks reasonable. There is still the possibility the instrument uses an internal time unit and not seconds.
The overall gain factor would probably be adjusted anyway for the analog version and the main part to take from the digital regulator would be the time constants.

The relatively large time constants could be pushing the limits for an analog regulator (especially the simple 1 OP circuit) with rather large capacitors / resistors (e.g. 10 µF and 40 M to get the time constant of the integral part).
A resistive heater is inherently nonlinear. So chances are the analog regulator would ideally also have at least an approximate linearization.  The minimal solution may be a relatively close set upper limit for the heater. Getting to high in the heater power would increase the regulator gain and thus can cause instability. At lower heat the gain would be reduced, which only makes the regulation slow.

[branadic]

Sample rate is more like 31.25Hz, time between two samples about 32ms.

-branadic-

[dietert1]

In my little report under "Reference Oven Experiments" (link above) i had the PID numbers as determined by Autotune for my setup:

TEC:PID? => 4.7796493,0.0258901,180.5900726

They are much bigger, about 10 times for the P term, about 20 times for the I term. The D term normally does not matter and i often set it to zero by hand. I would not trust Autotune with a nonlinear heater setup. The best method is using a TEC mounted on a large aluminum heat sink. P. Corr wrote to me that the regulator calculates at 10 Hz. The power stage is a constant current source and the proportional term is in units of A/K, so P=1 will output 1 A at 1 K temperature error between set temperature and measured temperature.

Regards, Dieter

[branadic]

dieter1

I have a BPR10J101 (10W, 100R) resistor as a heater attached to my LTFLU on an aluminum board and no TEC installed, which explains the different parameters.

-branadic-

[branadic]

Waiting for components in one project results in progress in another project.
I've updated my report on LTFLU, but it's still not finished.

Meanwhile I've designed and build an analog oven controller for the reference, but than found that noise is not satisfying. Turned out that noise of the network used is worse than expected for such application, an experience also other members made meanwhile. I've made a new board revision with some minor changes, such as using TOMC network instead of NOMCA and spend a second NTC. Thus, almost all of the former steps had to be repeated.
Things improved, but the project goes on.

-branadic-

[Kleinstein]

A TEC is nice if one wants a temperature rather close to room temperature. For just a stable 40-50 C it is easier to use just a heater. 
With TEC element one usually has quite a fast path from the outside via the peltier element and strong couling to the outside - this makes the oven overall relatively fast to react to external disturbance. Ideally one could us the other side of the TEC element for feed forward correction, but this is hard to implement analog.

I am currently also in looking at on reference ofen. The analog circuit looks similar, though using the output transistor as heater. With the transistor as heater the power is linear and not much power is lost.  A similar circuit is used in the HP Model 10811A/B crystal oven  (with some extra complications of a 2 nd heater and overly complicated current limit).

For a fixed setpoint I see no need to have the extra filtering for the set point voltage, so no real need for the 2nd OP in the regulator circuit. This takes out some of the OPs noise. For the shown circuit I would prefer the LT1013/LT1006 over the AD822. With a shifted DC level (other resistor values) one would not need a single supply capable OP and could use the more modern OPA202 too.

For the final version, the noise looks not so bad. Remember that one gets the combined noise of the reference and the meter, including it's LTZ1000 ref. It would need a comparison to a 2nd ref with known noise to get a definitive value.
With the oven temperature adjusted so close to the zero TC point, I would not expect much noise from temperature fluctualtions. The TOMC resistor are lower noise than NOMCA, but may still contribute a little to the noise.

[branadic]

Hello Kleinstein,

I fear you misunderstood. I used a heater resistor BPR10J101,  but a TEC controller (Arroyo 5305), that allows heating only mode for the temperature sweeps I've performed. With it, I've then started to develope a discrete analog oven controller.
AD822 was used for the analog oven controller as I want to supply the whole assembly (oven and reference) with a single 12V rail only. Some inspiration for the oven controller came from F732B, though TL062 was replaced by AD822.

By now, not all of the required measurements are done, but F7000 is waiting for exactly the measurement you describe. This is just an update on my current progress to help others contributing to voltage references based on LTFLU.

-branadic-

[Andreas]

With the transistor as heater the power is linear and not much power is lost. 

Hello,

that is true but the heater area is punctual and very small.
So you need extra effort for a heat spreader compared to a heater foil.
Otherwise you have large temperature gradients of several deg C within some 10s of mm.
If I would use transistors I would use as many transistors as can be placed on the backside of the PCB.

with best regards

Andreas

[branadic]

Agree Andreas, that is why I've chosen something in between heater foil with thermal gradients and transistor with its local hot spot and came to the solution thickfilm resistor on ceramic substrate.



-branadic-

[Kleinstein]

A agree that a transistor is a much more local heat source, but it depends on the power level and thermal design how bad this is.

With nonlienar (e,g, square law) heater, one could consider to linearize the loop at least a little, e.g. with a divider + diode in the feedback path of the regulator. So one could have something like a divide by 2-3 at low power, going over to some divide by 1.2 at higher power.  This could improve the regulation at the lower power end, where the resistive heater is less effective.
Without linearization the loop gain would depend on the power level, with less than ideal gain at low power.

[branadic]

The TOMC resistor are lower noise than NOMCA, but may still contribute a little to the noise.

At least a factor of 10 difference, something that is hidden in the datasheet, as they are both spec'ed with <-30db. Maybe it's the limit of what Vishay can measure with equipment they have?

Maybe some words to the construction technic used. The design approach here is somewhat similar to the voltage reference of F5700 DAC module, though a ceramic board is used there, which has better heat conductivity compared to aluminum. But they use a ceramic spacer between ceramic thickfilm heater and ceramic reference board, most likely for "act as a heat spreader" purpose.
Not sure what's the power consumption for their oven assembly - assembly only covered with some plastic shield and power consumption doesn't play any rule in a calibrator at all - but with my reference board and resistance heater inside the small styrofoam box - I should post a picture of it at some point - and everything thermally stabilized, the oven draws about 0.9W.
This is larger compared to an LTZ reference - no doubts about that - but probably smaller compared to what Fluke needs for either the F732B oven assembly, which is physical large - or the mentioned F5700 reference module including the oven.

-branadic-

[Kleinstein]

The < -30 dBi noise level seems to be a common level, like the limit for a simple test system.  They also give the same number for the ORN resistors (only 4 resistors), with a measured noise level that is much lower.

10 x better than NOMCA may still not be good enough to ensure low noise also at very low frequencies.  I did see quite some noise with NOMCA (50 K and 10 K) already at 25 Hz.

For the in plane conductivity the aluminium core board should be higher than most ceramics (I doubt they go the trouble with saphire or BeO). The advantage of ceramics is lower thermal expansion, but this should not matter so much in a stable temperature oven.

Some 0.9 W sound reasonable for a module of that size - more insulation could probably reduce it a little more.  This is still quite a bit higher than the power consumption without the heater - so the minimal temperarure would not be too high. 

[branadic]

Some 0.9 W sound reasonable for a module of that size - more insulation could probably reduce it a little more.  This is still quite a bit higher than the power consumption without the heater - so the minimal temperarure would not be too high.

Would be interesting to hear your suggestion on how to improve insulation.

Hear is how it's currently build:
I have a 1cm thick-walled styrofoam box, with some cotton wool on the bottom, the board with the heater resistor attached to its back is laying flat on it. Then more cotton wool covers the top and fills the inner volume of the styrofoam box until a 1cm thick-walled styrofoam lid closes the box. The inside of the insulating box is slightly bigger than the aluminum reference board (20x40cm²) with its heater and has only enough room for bending the wires towards the lid, where they feed through.

Indeed some Aerogel like material could be used, but that is nothing commonly available in your next hardware store. Even a box made out of a bottle with pu foam could be used (already have one), but needs some kind of form, to get it in shape. So no easy task either.
Am I'm missing a much more simple but effective solution, with material that is easy to get? Don't forget, all hardware stores are closed due to lockdown.

-branadic-

[Kleinstein]

1 cm styrofoam looks already quite good isolation - not that much more to improve. There maybe slightly better isolating gardes, but likely not much difference. A little more thickness can help. A layer of aluminium foil at the outside (and maybe parts of the inside) may help. With the not so good grades of foam quite a bit of the heat goes through as IR radiation. The main point to improbe would be too have the board not too large - not so easy to change and likely not worth a change.  The wires may also be important  for the heat loss.

So maybe the 0.9 W are not that much to improve on.  A small mains transformer often has more no load loss so for the overall power it would not make too much difference.

I don't think the PU foam from the sprax can is an especially good isulator, chances are it would be more one of the not so good grades. Better wait for the hardware stores to open and than get some foam made for isolation.

[iMo]

I put my 1.2W ocxo into an 1cm thick box made of of expanded polystyren (a white 1cm thick plate made of white balls ) and it is not enough..
The thermal conductivity of that material is 0.03-0.04 W/(K.m) so you may somehow calculate the effects..

[dietert1]

P = const * a / d * dT. With a 10 x 4 x 4 cm box a = 200 cm², d = 1 cm => a / d = 2 m times dT = 20 K times const = 0.03 W/mK gives 1.2 W.
Double thickness gives a little more than half of the power.

Regards, Dieter

[iMo]

Out of curiosity I've put the size of my ocxo into dietert1's calculation.
Pretty good guess

[branadic]

As the batteries for the reference were running out of power and I have to charge them, I took some pictures and measured the actual inner size of my box: 5.5cm x 3.6cm x 3.6cm with a thickness of the walls of ~1.5cm.
I also have to correct myself, the power drawn for the oven is in average 0.484W, not 0.9W as stated earlier. I also took the chance and wrapped some self-adhesive aluminum tape around the outside of the styrofoam box, will see if that makes any difference.

-branadic-

[dietert1]

By the way, the same formula applies for copper, except this time const = 384 W / (m K). So if your wires have a total cross section a = 1 mm² of copper, then d = 1 cm gives a/d = 1E-4 m and with dT = 20 you get P= 0.768 W. If we assume half of the power gets lost in the styrofoam and the other half in the wiring, then this means that your wires are colder than expected inside and hotter than expected outside of the oven (d = 5 cm or so). Otherwise power loss would add up to something like 1 W or more.

Regards, Dieter

[Sprock]

Hello

just want to show my LTFLU. Its case in case. Outside a Zeissler Feltron Alu case
whitch was extended bei an Alubar 12x165x50 mm just to get the right high
for  the 4x isolation a´26 mm. So the LTFLU is in Rose ABS case apr. 76x120x55 mm.
Data are a bit slurry.