LTZ1000 digitally controlled oven temeprature

[miro123]

Dear Voltnuts.
I’m new in this forum and initialy posted message in wrong section –
https://www.eevblog.com/forum/testgear/ltz1000-digitally-controlled-oven-temeprature/

I have seen many LTZ1000 voltage references boards.
My three questions,
 1. why nobody do digital control of LTZ1000 oven temperature?
 2. How big is the part of LTZ1000 1/f noise caused by bad temperature control? Does is worth  to make it beter?
 2. Is it an crazy idea to do digital control with such vintage component.? see the notes below
  - modern SD ADC  can sence LTZ Ube with almost unlimited resolution, accuracy and low noise
   - other adc channels can be used to sence data for my model - e.g. several PT100 anemometers etc.
  - Output DAC does not need to be ultra linear. - combined .dual 16..18bit DACs must be enough
  - wide BW noise from digital parts can easily be mitigated
  - DAC ADC Vref can be tricky - for first prototype i can use one temperature controlled KX board.
  - later on with two voltage references I can make even beter digital control.
  - Digital control is area of my interest - once I find the open loop model of my voltage reference board I can buld control loop - I'm prepared that model wil be multidimansional and time variant.

Critics, suggestion or advices are welcome.
I have zero experience with LTZ1000
Best Regards,
Miro

[guenthert]

    The reason why it hasn't been done (afaik) is likely that the reference design uses analog control.  Digital control wasn't the obvious choice when the LTZ was developed (and that's the aspect in which its 'vintage' could be relevant).

    As others have already pointed out, the reference design performs quite well, so it's not quite obvious how a digital control can improve on this, particularly due to small distance / low delay between heater and sensor.  The Zener with the moderate TC (50ppm/K, iirc) is clearly the most critical piece to be temperature controlled, but others, especially the current setting and voltage dividing resistors would benefit too (or allow for cheaper components).  I'd rather think a controlled oven for the whole assembly would be the lower hanging fruit (where digital control would be the obvious choice due to the much longer control loop).

[TiN]

1. Nobody "publicly". It has been done before by industry.
2. 1/f noise mostly electrical/physical issue, not thermal. Difference would be rather marginal.
3a. SD ADC with unlimited accuracy? Do tell, I'd like to have one.
3b. Your ADC in best case as good as your reference is.
3c/d. It is not easy nor cheap to mitigate wide bandwidth noise.

Most reasons why hobby players don't bother with digital control/compensation for LTZ1000-based reference is elegance of analog ways. Adding digital complexity over something like LTZ also introduce all the digital issues, noise, charge coupling, PCB layout impacts, additional filtering, more points to generate parasitic thermal EMFs (extra heat from DAC/ADCs), bigger power consumption (bye bye battery operation). And then spending hundreds hours troubleshooting and revising design to overcome heroically all these challenges, and for what end result?

That being said, don't listen too much, and go ahead on trying it. I'm sure it will be fun to learn and something useful can be discovered in the end result 

[dietert1]

Recently there was a thread on a smart Mosfet driver (TI DRV8320). It has an SPI interface and lets you configure dead time, gate drive currents, current limit etc etc. At TI they seem to be convinced this will be the future. In the past we inserted a resistor or a diode to get it done.
In the case of a voltage reference it would make sense if you had some algorithms or an AI setup to automate "optimization" of a reference. You would need to insert knowledge from eevblog metrology into that AI system. And i guess one would need to make a unit with multiple references, because somehow the AI would need to decide which one or which configuration works better. As far as i understand optimization of voltage references is something extremely rare, due to the long observation periods. A plausible improvement may turn out to be an error only after some years.

Regards, Dieter

[Kleinstein]

The problem with the temperature regulation is not with the control loop, but in measuring the difference from the actual temperature to the set point.
The problem only shifts from the divider to the ADC and with modern SD ADCs the divider to divider down the reference from some 7 V  to some 2.5-4 V for the ADC. The demand for the ADC would be quite high. The digitial control would not save much, as the actual control is not the weak point. The standard circuit even gets away with the aquare law heater response and thus power dependent loop gain.

As the LTZ1000 ref. is often used in combination with some high resolution DMM / ADC , it may be still feasible to occasionally read the voltage of the set point divider. If drift is detected there one could still do a numerical correction of the results.  The analog solution is well good enough for the shorter ( < weaks) time frame - the weak point is long term drift of the resistors, that may be limiting. At least the long term drift is not known upfront and hard to test.

[dietert1]

The LTZ1000 reference has two transistors and in its basic datasheet circuit they run at the same collector current. One way to discover thermostat problems may be looking at both Ube voltages and comparing them with some precision. The recommended application circuits are simple and convenient, but also a bit sloppy. I can't remember a quantitative characterization of typical LTZ1000 thermostat errors.

Regards, Dieter

[Kleinstein]

I would consider 3 possible types of thermostat malfaunction: the heater stuck off, stuck on (possibly bruning the heater if there is no extra protection / limit) and finally some kind of oscillation. There may be some large signal instability that can be difficult to spot. The control loop is nonlinear (more gain at higher power) and once the excursion is large it may go to a permanent oscillation even though it is stable in the small signal linear theory.

[dietert1]

When i wrote error i did not mean failure but deviation in °C. You can find lots of opinions about whether board slots help etc. and also the opinion that reference noise is not caused by thermostat operation and - no numbers. Many LTZ1000 references use plastic covers over the IC, so the oven does have its limits.
I remember someone wrote that a basic LTZ1000 reference has a typical TC of 50 ppm/K. What is a typical oven temperature change with an ambient temperature change of lets say 5 °C?

Regards, Dieter

[miro123]

1. Nobody "publicly". It has been done before by industry.
2. 1/f noise mostly electrical/physical issue, not thermal. Difference would be rather marginal.

Thanks everybody for feedback. I have some questions. I also try to clarify my cryptic post.
1. Clear
2. Do you have some numbers. How big is the oven thermal error by analog design. e.g you KX design

3a. SD ADC with unlimited accuracy? Do tell, I'd like to have one.
3b. Your ADC in best case as good as your reference is.
3c/d. It is not easy nor cheap to mitigate wide bandwidth noise.

Modern SD are designed to interface with sensors - not to make voltmeters. Sensing Ube is perfect match for them. Lets put some numbers
   - Voltage range - 0...600mV
    - glueless interface
    - accuracy 10ppm = 6uV.
Assuming Ube has -2mV/C . 6uV= dT accuracy =0.003C=3mC.
Are my calculation correct. If yes, is  0.003C accuracy enough.
   

Most reasons why hobby players don't bother with digital control/compensation for LTZ1000-based reference is elegance of analog ways. Adding digital complexity over something like LTZ also introduce all the digital issues, noise, charge coupling, PCB layout impacts, additional filtering, more points to generate parasitic thermal EMFs (extra heat from DAC/ADCs), bigger power consumption (bye bye battery operation). And then spending hundreds hours troubleshooting and revising design to overcome heroically all these challenges, and for what end result?

That being said, don't listen too much, and go ahead on trying it. I'm sure it will be fun to learn and something useful can be discovered in the end result 

Clear. I realize the enormous job done from industry leaders and comunity. The easy way is to copy exisiting design. Most likely I'll copy , adapted version of your design.  WHy
 1. At best I can achieve the same performnce as yours. - makes no sense to goes further
 2. I can acieve that with 1/10 of efforts. - still a lot of work for shingle person. I'll explain why digital aproach can be achieved 10 times faster.

   - easy to quantify influence of external factors. Air flow, te,perature change rate, humidity, air preassure etc.
   - Well defined theory, no magics and mysteries
           - straightforward approach from  transfer function to control loop design
           - known theory of multi variable control
    - fast iteration loops - change software vs redesign board
    - easy debug - easy to quantify the difference - software diff vs  components/PCB/ Soldering and other fragile parameters

BR
Miro

[miro123]

The LTZ1000 reference has two transistors and in its basic datasheet circuit they run at the same collector current. One way to discover thermostat problems may be looking at both Ube voltages and comparing them with some precision. The recommended application circuits are simple and convenient, but also a bit sloppy. I can't remember a quantitative characterization of typical LTZ1000 thermostat errors.

Regards, Dieter

This is one of advantages of digital design.
In general - monitoring of all known disturbances and finding the correlation between disturbances and open loop output voltage /Vz/ - Control thery call it tranfer function.
Creating from multiple transfer functions an multi varible optimal controller is science. No mystery or magics.

[miro123]

The problem with the temperature regulation is not with the control loop, but in measuring the difference from the actual temperature to the set point.
The problem only shifts from the divider to the ADC and with modern SD ADCs the divider to divider down the reference from some 7 V  to some 2.5-4 V for the ADC. The demand for the ADC would be quite high. The digitial control would not save much, as the actual control is not the weak point. The standard circuit even gets away with the aquare law heater response and thus power dependent loop gain.

As the LTZ1000 ref. is often used in combination with some high resolution DMM / ADC , it may be still feasible to occasionally read the voltage of the set point divider. If drift is detected there one could still do a numerical correction of the results.  The analog solution is well good enough for the shorter ( < weaks) time frame - the weak point is long term drift of the resistors, that may be limiting. At least the long term drift is not known upfront and hard to test.

1. I think that temperature dynamic control is limited in analog design. They solved redusing the problems
      - reduce thermal mass using - PCB slots small traces in inner layers
      - keep PCB temperature distribution constant in time == hard coded thermal EMF in all junctions.
      - using foam or beter isolated LTZ1000A helps to control temperature with limited analog control
2. I agree that stady state of digital control will be equal /at best/or worse that analog
My questions are
 -  how good is dynamic control of temperature in current analog design. Any numbers are welcome?
 - is there any correlation between temperature error and low frequency noise of LTZ1000 reference circuit?
3. sensing critical points in LTZ circuit without control the output seems great idea. Thanks. Do you know other critical pints in LTZ circuit that needs monitor. I think 7V /10V ration is critical too


BR
Miro

[miro123]

Insipered by Kleinstein Idea - I would like to change the subject from digital control to digital monitor.
Monitoring of all critical electric and physicals disturbances will be enough. The aim is to achieve better short term (>1week) stability.

[Kleinstein]

Getting 3 mK temperature stability would be marginal. The LTZ1000 withput heater has a TC of some 50 ppm/K and 3 mK would thus be some 0.15 ppm of error. The would be about 1  µV or about the peak to peak noise for 0.1-10 Hz.
However the weakest point with an extra ADC would likely be there refrence to that ADC. Which would likely be a divider (e.g. to 1/3) from the 7 V LTZ ref.  The divider from 7  to some 2 V for the ADC reference would be similar demanding as the  1/15 divider for the temperature set point, which is the weakest point in analog control.

The analog control loop is not perfect, but it still pretty good and fast settling as the thermal part is quite fast. The P to I cross over for the analog regulator is typically at some 100 ms. External thermal disturbance is usually slow, if the reference is in a case and not flapping around in the breeze. So the dynamic part of the temperature control is not a problem. The DC loop gain is exceptional high: some 1 million from the OP, some 100 from the transistor and around 0.01 for heater and the 2 mV/K response for the BE junction.  The remaining temperature error is from the thermal setup in slightly different response to external heat and the internal heater. The quality of the control is thus not limited by the control loop, but more by the position of the heater / sensor.

The logic way to improve on the temperature control would be a thicker case to smooth faster disturbance and in the extremes a 2nd oven layer around the whole circuit including the resistors. For such a larger oven digital control may be convenient, as the much longer time constants in the 10s to 100s of seconds are a bit difficult to handle analog.

I would not expect much correlation between the reference noise and the temperature variations, especially not under relatively stable conditions. The problem with the temperaure is more slow drift, e.g from aging of the 1/15 divider or the sensor itself.
Monitoring the voltage for the set point may be a real option, if one has a good ADC in the system anyway.

[magic]

It seems you could superimpose 1V RMS at various frequencies on top of the heater voltage and see how much difference it makes in reference output. Then measure the actual noise on heater supply in normal operation and see if it's significant.

The recommended circuit is decent enough that it uses the temp sense transistor (Q2) as the input stage of the temperature regulation loop. So Q2 should be about the only significant noise contributor. You can't escape Q2 noise by wiring Q2 in unity gain (B-C shorted) and using external ADC to measure it, because the same noise will still be there.

If you think the recommended Q2 current is suboptimal for noise, tweak its collector resistor.

[dietert1]

You can simulate the dynamic response of the LTZ1000 oven on short time scales. I can tell you that i observed 50 Hz oven oscillation when running the control loop without the lowpass filter caps, so the effective delay time between heater and sensor is about 10 msec. And the response will depend on the low pass filter actually implemented. Certainly the proposed control circuits avoid critical response, but implement some stability margin.
The correct way to find the low frequency oven gain (determined by chip geometry) is to put the DUT into a temperature chamber and impose external temperature change. The test needs special attention to avoid confusion of oven residual temperature change (observable) and thermal EMF (disturbance) as far as possible. One way to achieve that is running the test at an ambient temperature near the oven temperature, with small heating power of LTZ1000.

Regards, Dieter

[Kleinstein]

The heater has a square law effect and is thus nonlinear. The linarized loop gain for the temperature loop thus depends on the current heater level. The higher the power, the higher the loop gain. For this reason the control loop in the standard circuit needs to be quite conservative, not to get unstable when at higher power and thus higher loop gain. At a very low power level the loop gain goes down, but likely still high enough.
So while one could linarize the the loop and add more gain for the low power range, this would probably not make a real difference.

The standard current for T2 is at a little under 100 µA, which is perfectly fine noise wise. Even less current would be OK - the problem is more that higher value resistors tend to be less stable.