Small Oven Controler For Voltage Reference

[TiN]

Sorry about useless post, but I love this thread. Keep it going (esp with photos) 

[blackdog]

Hi cellularmitosis,

You are correct! 

Regards,
Bram

[Andreas]

Sorry about useless post, but I love this thread. Keep it going (esp with photos) 

Ok perhaps I can jump in with my voltage reference ageing box:

Build in PWM controler, i will never do that again, that is asking for EMC trouble.

That is why I use 100 Hz as PWM frequency: this will be canceled out by most instruments as it is a multiple of AC mains frequency at my location.

in the background: lower half of the styrofoam box consisting of 2 parts which are overlapping over the full height.

top and bottom are 2 aluminium plates with a heater foil (car mirror heater) each.
On the top side a NTC for (fast) temperature controlling (P-controller).

Additionally a overtemperature switch which switches off when electronic fails.
(I am missing that in the previous ovens: what can happen if the NTC is shorted or open?)

Then the PCB with the voltage references on the left side.
In the middle of the references a further NTC for the slow temperature controlling (I-controller).

In the middle of the PCB the multiplexers (differential) so that one of the references can be measured.

on the right side from bottom to top:
RS-232 interface (with photocouplers = isolated)
Processor for temperature control (12 Bit ADC = 0.025 deg C resolution)
level shifters for the multiplexers and power stages (PWM)

with best regards

Andreas

[David Hess]

ahhhh, now this is all making sense   I ran into a similar regulation problem with my OP07, and quickly realized it was because the output couldn't swing below roughly 1.4V, so I added a reverse-biased zener on the output to solve the problem.

And now I finally realize why there is a strange zener hanging off of the LM723 output, exposed as pin 9...

The 723 voltage output only has a compliance down to 2 volts from the negative supply so the zener is needed to drive pass transistors which have their emitter at the negative supply.  The metal can package only has 10 pins so the zener output is missing.

[cellularmitosis]

Initial estimate of performance: roughly a 20mK change in a 750mK ambient swing, which makes the attenuation about 37.5x (which also includes the error of the Keithley 2015.

Testing the oven in another oven (rather than relying on ambient swing) would help to isolate the Keithley 2015 tempco error.

[cellularmitosis]

Some notes from my quest for compensation:

• A: no compensation
• B: 0.1uF ceramic + 1M (from inverting input to after the output zener)
• C: 0.1uF ceramic + 10M
• D: 1uF ceramic + 10M
• E: 2uF ceramic + 10M
• F: 2uF ceramic + 10M (attached to top of current shunt).  this was less effective than from inverting input to after output zener.
• G: back-to-back 10uF electrolytics + 10M (from inverting input to after the output zener) (caps are positioned negative to negative)
• H: back-to-back 47uF electrolytics + 10M
• I: back-to-back 47uF electrolytics + 10M (from inverting input to before the output zener)
• J: back-to-back 47uF electrolytics + 10M (caps are now positive to positive)
• K: back-to-back 47uF electrolytics + 20M
• L: swapped around the bridge (thermistor now on bottom half of non-inverting input).  forgot to swap the compensation network, which nulled its effect.
• M: corrected the compensation network
• N: changed out the 10k/10k fixed divider for a 100k/100k divider.  yay the oscillations stopped!
• O: 1uF ceramic + 20M.  this still worked!
• P: 1uF ceramic + 10M.  this still worked as well!
• Q: 0.1uF ceramic + 10M.  this caused the oscillations to return (though smaller than before)

(Ignore the component vallues in the schematics below: only their position is indicated)

[cellularmitosis]

The final circuit ended up being nearly the same as a circuit from Charles Wenzel.

http://www.techlib.com/electronics/ovenckts.htm

(I took the idea of trying a 100k/100k divider from him, which seems to enhance the effect of the compensation RC, and was the change which finally stopped the oscillations)

Edit: made a few corrections to the schematic.

[cellularmitosis]

Here is the template (as pdf and svg) I made to cut out EVA "craft" foam pieces to cover a Hammond 1590B enclosure.

I printed this out on paper, cut the paper pieces out with scissors, then used a straight edge and razor blade the cut the foam.

[blackdog]

Hi cellularmitosis,

All the components connected to the inverting opamp input are part of the loop compensation.
Like i told before, be shure to have a tight coupling between the heath source and the sensor.
This will solve a lot of loop problems.


Andreas
Additionally a overtemperature switch which switches off when electronic fails.
In my other oven controllers is an extra sensor included for too high temperatures.

Kind regarts,
Bram

[cellularmitosis]

The prototype oven has an MJE3055 and two thermistors affixed to the bottom side with Arctic Silver thermal epoxy.

However, the oven did not run at 25C like I had expected.

I lifted up the bottom of the oven and pointed a small fan at the thermistors, ensuring they were both at the same temperature and very closely tracked the ambient room temperature.  I then alternated measuring one and then the other thermistor.

It turns out they have a rather large offset, both from each other and from the Si7021 which I used to log the room temperature.  I believe this explains why I didn't settle on 25C as expected.

These are +/-5% parts, which roughly means +/-1C (at 25C).  https://www.digikey.com/product-detail/en/us-sensor-littelfuse-inc/103JG1J/615-1016-ND/1014544

[cellularmitosis]

I hooked up the LM723-based circuit to the same Hammond 1590B + MJE3055, and fiddled around until I came up with this, which seems to be pretty well damped, and is a pleasantly minimalist circuit.

I noticed that there is a tradeoff with the value of the compensation resistor in relation to the set-point divider resistors: as the value of this resistor becomes closer to the value of the divider resistors, the compensation network seems to have a stronger effect.  However, it then also starts to alter the set-point.  Making the compensation resistor much larger than the divider resistors means that it doesn't affect the set-point, but the compensation has less effect.

I tried attaching the compensation network to the top of the current shunt and this seemed to work well (thanks for the suggestion Kleinstein).  Previously, when I had the compensation attached to the output, the voltage could wander around lower and lower during an undershoot, but the current was already zero, because the voltage was below the turn-on threshold of the MJE3055.  This seemed to act like a "wind up" problem, but in the downward direction.  By fixing the compensation network to the shunt, compensation voltage is now directly tied to heater current, which seems to work better.

I also realized that I could split the shunt into two halves, so that I wasn't limited by the 0.65V threshold of the current limiting transistor.  This seemed to amplify the compensation network's effect a bit.

I also realized that with this arrangement, since the set-point (3.5V) is always more positive than the top of the current shunt (or rather, the split current shunt is now chosen to ensure it is always below 3.5V), I can now get away with using a single (polarized) electrolytic in the compensation network, allowing for larger capacitance values than a ceramic (and I can avoid using a back-to-back pair of electrolytics).

I'll test the stability overnight.

[blackdog]

Hi,

It begins to look good 

Now make the capacitor on pin-13 220pF, 10nf is way to much, the openloop gain will be far below 1Hz and you dont want that.
The 220pF is more than enough for HF stability and you wil have a bigger oven gain, that is more temperature stability.
Perhaps a small correction of your loop components is necessary.

Your setup in the bridge  for the negative input is a little on the high side, and yes this helps to keep the loop components smaler.
But keep in mind the bias currents, these are high for a uA723, keep the uA723 in the oven, then you can probably get away with it. 
I would not exceed two resistors of 20K for this bridge section.

I see the connection of the loop components to the emitter of the MJE3055 as no problem at all, it is just a very large emitter follower of the output signal.
Keep op the good work!

Kind regards,
Bram

[blackdog]

Hi,

Andreas,
Some comments about the EMC problem with a PWM controller.

I have tested extensively with a aluminium tube oven with resistance wire wound around it.
I then connected the resistance wire to one of my function generators.
Dit a frequency sweep and measured the field inside the oven, HORROR!

Next step, winding the resistence wire Bifilair on the tube, this was a big improvement.
But even with low frequencies there were big spikes.
This is the oven with the bifilair and also double windings fore parallel (power on quick heating and serial for normal use)
The yellow wire bundle are the three sensors, one for the ovencontroler and the other for measurement and savety (high temp)



So the next step was to remove the scharp edges when using a square wave.
Below is a part of the schematic I use to limit the EMC of a P.ID controler build around a Arduino or Teensy micro contoler.
The 10uF capacitor remove the sharpedges, the P.I.D. frequency is about 1Hz a 3Hz.



And yes.. making this oven is difficult 
The wires are not allowed to touch each other.



Testing the EMC in the oven.



Testing with a square wave.



And this is how it looks after removing the fast edges.


For more info look at this link and use google translate...
https://www.circuitsonline.net/forum/view/133966

Kind regards,
Bram

[cellularmitosis]

Looks like the LM723 oven's attenuation / performance is about 50x (0.5C ambient change results in roughly 10mK change).

Bram, I'll try a smaller cap on pin 13, thanks.

[Andreas]

Some comments about the EMC problem with a PWM controller.

Hello Bram,

of course you are right.
Bifilar winding is essential. (Is also done on the heater foils).

And I also forgot to mention that I usually do not use "normal" FETs when it comes to "power".
I usually use "automotive fully protected" FETs.  (like VNP5N07 or BTS432).
They are optimized for low emissions.
And further they limit the current in case of short cirquits.

with best regards

Andreas

[IconicPCB]

Now that the temperature is monitored and controlled by the LM723...

Has anyone checked performance of the REF voltage of temperature stabilised buried zener inside the LM723?

[Wolfgang]

Hi,

I built a self-heated LM723 reference. It can be seen here:

https://electronicprojectsforfun.wordpress.com/silly-circuits/silly-circuits-a-heated-lm723-reference/

Drift is a millivolt per a few days (out of ca. 7.15V), but this could be a burn-in effect.

[IconicPCB]

Wolfgang,

I saw Your article prior to posting above comment.
I was hoping to see other confirmation of Your experience as a sanity check.

[Wolfgang]

A word of warning regarding high-capacity (X7R, Z5U, ...) ceramic capacitors:

- Their capacitance decreases drastically with applied voltage
- Their tempco is evil
- they are microphonic - See Dave Jones teardown of a Rohde scope where tapping at the touchscreen created voltage jumps - caused by ceramic Cs !
- They age
- In my opinion, the should never be designed into a control loop - Use high quality foil when possible.

Just ranting - much luck experimenting !

[blackdog]

Hi,

+1 for Wolfgang 

Kind regards,
Bram

[Wolfgang]

Hi,

I am continueing the long term tests, but unfortunately I have not seen others doing the same or something similar yet.
Maybe you try yourself ?

[IconicPCB]

My highest resolution instrument is an advantech 16 bit data logger which is not good enough to get me to the resolution and stability  of measurement.

[Wolfgang]

Yeah, you are right. When it comes to checking references, the checking instrument must be at least an order of magnitude more stable than the expected deviations.
Just to be on the safe side, I teamed a RIGOL 6 1/2 digit M300 datalogger with a Keysight 34401A multimeter. At a measured value of around 10V, they disagree for ca. 300uV maximum. When you have a look at my webpage, there is a multimeter comparison at the bottom. 6 1/2 digits at this range means about 300uV accurracy. A 16bit *resolution* (thats not enough, we need accuracy, too) is about 5 3/4 Digits, but again absolute values will be a lot less, so no option.

If you really want more, then you have to go for an 8 1/2 digit multimeter unit, bringing uncertainty down to +/-26uV (Keysight 3485A, 4ppm). But then the whole setup must be improved: Temp control for all instruments, low EMF cables, .... and you need some extra 10k€.

If you can afford all that, then you can also invest in a LTZ1000A ... My LM723 reference creeps around in the mV range, so I would say that 300uV of tolerance is legit.
We are far beyond what the LM723 datasheet was promising, so its fine for me. My reference was meant for radio amateurs with budget constraints.

[iMo]

Or, you may try with a "differential method". Using your 16bit datalogger as the "zero indicator" only. You would need a voltage source more stable as the 723''s ref. On that "stable" source wire a 10turn 10k Bourns (or something like that), from the wiper to your datalogger input, the datalogger's gnd lead to the 723 ref output. Set zero with the Bourns at the beginning.

In past I was using a REF01 as the voltage source for the Bourns.

This method may eliminate the 7.10V offset problem. Your logger will show the actual difference between the wiper and the 723. In case the REF01 and Bourns will be put in a temperature stable box, and you will cool/heat the 723 ref only, you can easily get its tempco.
Just an idea..

[IconicPCB]

Yes, the data logger can be configured as a differential instrument.