How do I determine parameters in an npn array thermal oven?

[homeless_peep]

Hello folks,
I have a circuit (an exponential current source, for a music synthesizer) that uses two matched transistors, and the whole thing is very sensitive to
1. the temperature difference between the two transistors
2. the absolute temperature of the transistors.
The 3 most used solutions are
1. use a temperature depended resistor to scale the input accordingly and cancel out thermal errors,
2. Blindly throw power to the transistors through a not controlled heater resistor, hoping the temperature will always rise to the same levels for both transistors, and rise at roughly the same level no matter the ambient temperature
3. Design a thermal loop, that keeps the transistors at the exact same temperature no matter what. I've noticed that cheap commercial synths (microbrute and minibrute) use this technique and it works great (been using it for a while).
I've already tried solution 2 with two discrete jellybean transistors and a resistor thermally coupled between them. Most days it will work ok and everything is in tune, but sometimes it just won't (and retuning is really a pain in the butt).

I have two MAT14 npn transistor arrays available and I'd like to go for another prototype, utilising them. They feature 4 matched npn transistors with characteristics favorable to the exponential converter. The classic solution I'm seeing all over the place is something like this:
http://hackabrute.yusynth.net/MINIBRUTE/analog-board/schematics/MiniBrute-04-VCO.pdf
(bottom left, "Temperature Regulator" block)

I understand how the temperature sensing and reference work and the topology used to just generate heat from the transistor. But I'm puzzled when it comes to the OPAMP configuration. It's configured as a comparator, realising what looks to be like on-off control. However, since differences are in the order of milivolts (2.2mV/C° for silicon?), or even less, the output will eventually not be saturated (when differences reach the order of microvolts) and, fingers crossed, the loop will settle somewhere and be stable.
Is that reasoning correct? Would a gain controller perform better?
On to the next question, let's suppose I roll with the above topology. How should I go on deciding the resistor values for my transistors and power supply? I've set a 114uA current on my sensor and calculated the voltage reference for about 50C°. What about the heater transistor? For now, I've set it up such that for maximum (positive) opamp voltage I get the maximum power dissipation(~0.5W), while still staying within the absolute maximum rating (<30mA and 85C°) and for minimum (negative) opamp voltage I got the minimum dissipation (~0.1W).
How does loop stability work in this case?
Any help or resources would be much appreciated.

[David Hess]

Maybe the divider at the output of the TL064 is lowering the gain enough to prevent oscillation.

I agree though that a higher DC gain integrator would be better.  Check out figure 4 of Linear Technology application note 5 where this was done with a CA3096 by overcompensating the 301A operational amplifier.

[Gyro]

The last time I saw a transistor array thermally stabilised in that way was in the Practical Electronics 'Minisonic' project back in 1975. Yes, it was also a synthesizer and it was also used to stabilize the current source!

There are scans of the project on Tim Stinchcombe's site:  http://www.timstinchcombe.co.uk/index.php?pge=pemini

The temperature stabiliser was included as an enhancement in part 4 of the project (the Minisonic, not the Minisonic 2), here's the specific link to the scan pdf:  http://www.timstinchcombe.co.uk/synth/pe_mini/pe_mini_feb_75.pdf

I haven't analysed the circuit, other than noting that it looks pretty straightforward and includes setup instructions. Your request just triggered a very distant memory!

[homeless_peep]

Thanks a lot for that Gyro, it all makes more sense now. The article uses a simple PI stage which is more straightforward than the open loop topologies.
So, the "loop gain" of such an arrangement (heater+sensing circuit) is depended on both the power dissipated from the heater and the voltage gain of the opamp?
I'm thinking about including a couple of different resistors for the heater's emiter, to adjust the "heater gain" roughly, and a trimmer on the opamp feedback, just in case I fire it up and the whole loop goes in oscillator heaven, lol.
How does heat transfer through the chip work in terms of transfer function/control loop?

I have already made a prototype of a triangle core VCO I designed for a modular synthesizer system (it is also my final project for an ECE master). Tuning instability is enhanced by the fact that my design called for a PNP source, but guess what! There are no low cost PNP quad+ arrays on the market to make a stabilised current source, at least none that i could find. That already limits my options a lot so I just heated the damn thing and hoped for the best. It's not that bad actually, but certainly not ok for a polyphonic system. Good enough for graduating though, lol.
And don't even get me started on waveshaping.

[Gyro]

You're welcome. 

In terms of loop / thermal gain, I'd try not to over-think it (difficult, I know, when it's a masters project). You have very little to go on however - all you have in terms of package spec is thermal resistance junction-case and case-ambient, (which is of course implementation specific). I think you can do it by just adjusting the gain of the feedback loop and allowing for some damping of the response. You would be best to breadboard (solder, not pluggin) this section before commiting to PCB. It would be quick to do and would give you a much better feel of any issues. You can easilly see if it goes into oscillation by monitoring a Vbe and also applying loading to the other transistors to see how well the circuit responds and tracks. Control loop theory is fine when you have all the necessary figures. Some things, you have to do the best you can and then  try it out.

[homeless_peep]

I have already experimented with simple thermal loops on to-92 packages on the breadboard. I used a heating resistor thermally coupled to an lm335 and fooled around loop parameters like gain. It was all very "random" in terms of values (just went with whatever felt right), but it was very easy to trim it to work properly. Measuring with a multimeter and a thermocouple, I could get it up to 70C° and keep it there fairly easily. I also observed the temperature error (output of the difference amplifier), it always settled to some final value when the loop was stable.
I will follow the article you posted, adjusted for the MAT14 (more current gain, lower Vbe values). If all else fails, i will disable the thermal control and bombard the sucker with heat, until it stays in tune, lol.

[Gyro]

If all else fails, i will disable the thermal control and bombard the sucker with heat, until it stays in tune, lol.

Yay, the pragmatic approach. There's hope for you yet! 

I suspect that you will have much better success with the much closer coupled monolithic transistor array. You should be able to derive some useful parameters for yourself. eg:

- Absolute die temperature... externally heat the package to known temperatures while monitoring the Vbe of one transistor.
- Thermal losses... Apply a fixed dissipation to one transistor and monitor change of your monitor Vbe.
- Thermal time constant... Apply/remove the dissipation, as above, and observe the rate of monitor Vbe change.

[David Hess]

So, the "loop gain" of such an arrangement (heater+sensing circuit) is depended on both the power dissipated from the heater and the voltage gain of the opamp?

I'm thinking about including a couple of different resistors for the heater's emiter, to adjust the "heater gain" roughly, and a trimmer on the opamp feedback, just in case I fire it up and the whole loop goes in oscillator heaven, lol.

Exactly, and the emitter resistors will also make the "heater gain" more linear and predictable.

How does heat transfer through the chip work in terms of transfer function/control loop?

The time constant for a monolithic transistor array is on the order of milliseconds or faster.  (1) If you monitor the temperature at different points on the array using the different transistors, then you can *see* the waves of heat propagate across the part.  Jim Williams mentions this and transistor selection for heater, sensor, and active device in the application note I referenced.

There are no low cost PNP quad+ arrays on the market to make a stabilized current source, at least none that i could find. That already limits my options a lot so I just heated the damn thing and hoped for the best.

The lack of inexpensive monolithic transistor arrays and variety is a real problem in these applications.  THAT Corporation makes some which are suitable including PNP  parts.  If you collect enough part numbers, then you might be able to find some NOS (new old stock) or pulls for a low cost.  While not ideal, attaching matched discrete transistors to a temperature controlled heat sink can work and is how it was done in the distant past.

(1) The propagation of heat across the semiconductor die is one of the primary limitations in operational amplifier setting time.  Thermal feedback from the output transistors to the input transistors also limits open loop gain.  "Precision" operational amplifiers take steps to minimize these problems by using symmetrical layouts to cancel the effects of temperature waves.  Minimizing the output loading with an external buffer can help.

[homeless_peep]

Thank you both for your suggestions. I had contacted THAT corporation for samples, but they didn't answer my e-mail. Analog devices MAT14 did arrive, so I was designing around that. But, surprise surprise, the people from THAT not only did send me samples (got them yesterday), but they were awesome enough to include 4(!) from each array on the 300 series, all in DIP packages. MAT14 has better logarithmic characteristics, but THAT320 has PNPs, eliminating the need for a current mirror.
So I got to the breadboard today and experimented a bit. This was easy after all.
I used transistors 1 and 3 (top left, bottom right), reasoning that it's better to have the heater and sensor equidistant from the expo transistors, plus, not have immediate thermal contact between the two.
When operating in open loop mode (like the microbrute schematic I posted) there is indeed (like we predicted) an AC voltage on the output of the opamp, at a frequency of about 80Hz. The duty cycle will vary depending on the temperature set and the heater gain, but the frequency is fairly consistent (always between 70 and 80 hz). Perhaps there is a way to calculate the time constant of the die from that frequency value. I'm not sure if this causes low frequency thermal oscillation, but certainly having about 20mA of current pulsing through the heater might cause noticable interference.
Next, i tried a 120nF cap across the opamp output and negative input. The output is now completely smooth, the loop settles in just a couple of seconds and seems to stay put. Touching or blowing the package doesn't change the sensor voltage at all, so i assume the regulation is good enough for my purpose. I tried lower values too, at 33nF the loop behaviour was similar, but i could notice some ripple at the output.
Finally, i replaced the capacitor with an 1M resistor. I'm not sure I can see a difference from the capacitor version (didn't bother to compare settling times, they seemed to be in the same order of magnitude). When changing the temperature setting, the sensor voltage follows really fast (looks like 'real time"). I'm not sure how I can verify that there is no low frequency oscillation. I think i can barely make out something on the (digital) oscilloscope, but it's hidden deep into the noise floor. Any suggestions on that?
On the voltage reference, the eurorack synthesizer standard calls for +- 12V supply rails, but I expect small variations from one user to the next. I think it'll be worth it to use some kind of stable, low thermal drift reference voltage. Would an onboard 7809 be adequate for that? Should I go into something more sophisticated?


We don't have the equipment to do extensive thermal tests and there are hardly any use to my application. Each unit is always hand trimmed and tuned before use, so what's important here is that the temperature is always the same, not what it's absolute value will be.
To calibrate the temperature setting I measured the sensor Vbe at room temperature (~630mV) and set my trimmer to 580mV. The die should be at about 50C° (~25C° above ambient). Since I don't really care about absolute measurement accuracy and linearity, I guess that's good enough.
Observing Vbe through time is very hard due to the scope noise. I tried averaging, but I had no luck. There is a specific lower boundary for the power dissipation below which the energy lost to ambient is greater than what the transistor dissipates, so the loop doesn't work. I couldn't verify this (I wouldn't want my transistors to fry, yet), but I assume that increasing the gain of the loop (either by more collector current or more error amplifier gain) brings the whole system closer to oscillation conditions, while reducing the settling time. If I open the thermal loop and attempt to make it's bode diagram just by feeding inputs at the + of the opamp and measuring outputs at the sensor voltage would i arrive to anything useful?

[Kleinstein]

Having just the 1 M resistor in feedback to reduce the gain of the thermal loop, still keeps a rather high gain for the OP. It may not be the very best choice to have the heater referenced to the +12 V and the sensor relative to the GND side. Though this would be mainly a problem with a capacitor in feedback. This might give some extra ripple if the supply voltage is not absolute stable. A LM7809 type regulator would be well good enough. With the resistor in feedback the regulator is mainly proportional type (with rather high gain). With the capacitor it would be an I type regulator. The more normal type of compensation would be a capacitor and resistor in series, thus a PI type regulator.

The usually adjustment for a P-Type regulator would be to increase the gain until it just oscillates and than reduce the gain by something like a factor of 2 to 4. A similar adjustment is possible with the PI or even a PID regulator. It is called the Ziegler–Nichols method. So I don't think an extra open loop measurement is really needed.

Usually such a regulator is either stable or it is oscillating with large amplitude - so a small amplitude oscillation so that one would not see is nothing to worry about.

[homeless_peep]

Kleinstein, could you elaborate on the supply referencing?
I tried changing the supply voltage by +-0.5V and the sensor voltage remained the same. I referenced the heater to both rails so i could get more power dissipation for a given collector current. Would it improve overall robustness to use the LM7809 for all voltages except for the opamp (heater, sensor and reference).
P and I controllers were very easy to stabilise, and since there is no low amplitude oscillation to watch out for, if it looks good it has to be. Would there be some considerable gain in making a PI version? What is most important here is repeatability of the final temperature and long term stability while in use. I don't really care about how fast it settles, in the synthesizer world warm up times up to 20 minutes are not unheard of.

[Kleinstein]

For the hand drawn circuit, the heater current is controlled by the base current and this the OPs output voltage relative to the + 12 V. The OPs input side is more like referenced to the GND side. It depends on the size of the base resistor on how much supply ripple would come through to the output side. I don't think it will be so much trouble. If at all it would be the faster ripple that could be a problem - all the slower part will be compensated by the control loop. Also supply ripply would compete with the rather high control voltage of the OP, not the 2 mV/K at the sensor.

For the supply, it would be mainly the divider with the 10 K and 100 K where the supply should be stable. So it might be worth to use something like a TL431 as a 2,5 V reference there. The other supply could stay at the rough regulated 12 V.

The purpose of a PI regulator would be an even more accurate control, not a faster response. It looks like there is already a really high gain and accurate control in the proportional mode. So changing from I control to PI control would be like going from maybe 100 mK temperature stability to 1 mK temperature stability. So nothing to worry about in a synthesizer.
If you would really go for highest precision, one could use the sensing transistor already as a first amplifier: so supply an adjusted voltage to the base. This could reduce the influence of OPs offset by something like a factor of 100-500.

This way the transistor might even already have enough gain to get away without using an OP at all. I attached a corresponding circuit. The compensation part C1 R5 would be optional and may not be needed here.

[homeless_peep]

If you would really go for highest precision, one could use the sensing transistor already as a first amplifier: so supply an adjusted voltage to the base. This could reduce the influence of OPs offset by something like a factor of 100-500.

I kinda lost you here.
I'll give that circuit a try tomorrow. If anything, i love how elegant it is.

[floobydust]

Just recently saw this approach to analog synthesizer VCO fail.  You have to have the whole thing in an isolated chamber/oven. NOT ON A PCB with everything else.
The problem is the heat loss from the matched-transistors+heater to the PCB, and if ambient temperature is up there you cannot cool anything. A lot of thermal design.

I would say do not re-invent the wheel- you are better off to use the Curtis Electromusic CEM3340 VCO, it has returned to production since the 1980's. It's a proven stable VCO for analog synths.

[David Hess]

I would put a resistor in series with the emitter of Q3 to better control its gain with a voltage drop of at least one Vbe at the operating current.

Then for adjusting the loop response, I would use the DSO to measure the output of the operational amplifier and perturb the loop with the square wave output from a function generator.  The settling time of the temperature control loop is not too important but if too slow, settling time of the translinear function may be affected.

[homeless_peep]

Just recently saw this approach to analog synthesizer VCO fail.  You have to have the whole thing in an isolated chamber/oven. NOT ON A PCB with everything else.
The problem is the heat loss from the matched-transistors+heater to the PCB, and if ambient temperature is up there you cannot cool anything. A lot of thermal design.

I would say do not re-invent the wheel- you are better off to use the Curtis Music CEM3340 VCO, it has returned to production since the 1980's. It's a proven stable VCO for analog synths.

I can't really do that, since my subject is "Design and implementation of a VCO for the eurorack synthesizer standard"  Plus, this has turned into a real obsession, making a viable triangle core design that doesn't use obsolete parts and has more than 5 octaves of rock solid tracking. Still, it can't be that dangerous of a solution, considering most new synth boxes use a technique along those lines rather than go for the CEM.
You raise a fine point though, the thermal conduction on a pcb full of metal traces is probably increased compared to the breadboard. However, I'm working with a DIP package and, because of it's cost, i will surely use IC bases. The chip will not be in contact with the pcb, and the insides of a eurorack are a bit warm anyway, so I doubt it will have a problem reaching it's goal temperature. The reference temperature could be increased by 5-10 degrees to allow for warmer settings.

[Kleinstein]

The heat loss for a dip chip one a board can be quite high. The thermal resistance is usually in the 70 K/W range. So with a 200 mW heater power this is only a 14 K temperature rise.  This would limit the temperature range that can be compensated to about this value.  So a thermal resistance would be really needed. For a larger range also some extra heating power (e.g. an external Zener in series to heater transistor) might help.

[homeless_peep]

My calculations say that I operate at roughly 20-30mA (about the maximum allowed) and bias the heater with both rails, going up to 0.5-0.6W. That's 42 degrees above ambient using that 70K/W figure...

[David Hess]

I looked up a few transistor arrays and saw values up to 160C/W and a lot of variance between different types.  In a production application I would use an external heater to get more operating range but for a prototype or something which will run in a controlled environment, this will not be needed.

[homeless_peep]

I looked up a few transistor arrays and saw values up to 160C/W

isn't that a good thing in our case?

[David Hess]

I looked up a few transistor arrays and saw values up to 160C/W

isn't that a good thing in our case?

Sure and the higher the better but as Kleinstein eludes to, even the ones with high thermal resistance are inconveniently low.  Real ICs that rely on this use a thermal insulating die attachment to raise their thermal resistance into the 300 to 400 C/W range.

The part I originally saw this done with was the LM389 audio amplifier which has 3 uncommitted NPN transistors.  It has a 70C/W thermal resistance however the output stage of the amplifier can handle up to 0.75 watts so its low thermal resistance is not as much of a problem.

[homeless_peep]

Would using electrical tape do any good in that respect?
Choosing a temperature of 50-55C wouldn't be ok for ambient temperatures of up to 45C? How different a performance should I expect when moving from the breadboard then? Would it be worth my time to solder this circuit on a protoboard just to see how it behaves?

so many questions 

[David Hess]

Most of the heat conduction is through the lead frame and pins into the printed circuit board so insulation will have limited effectiveness.  As long as the die temperature is held a few degrees above the maximum ambient temperature surrounding the IC, the thermal regulation should work great.

[floobydust]

  I found it's not about having the NPN array at a constant temperature- the other VCO parts (op-amps, capacitor etc.) have to be in the oven too. Then you will find system power supply voltage (i.e LM7815) drifts quite a bit which was not expected. Very difficult to design. Where you measure the matched-pair temperature verses where you heat the matched-pair; the difference has to be quite small, and heat (leakage) to the PCB absolutely low as possible.

One way to beat this is autocal- on power up get a high and low note and set an EEPOT to tune things. As the synth warms up, do it again. If you have an MCU involved, it can also be done in the background.

Elektor Formant music synthesizer I don't think it used an oven but others added one.

[homeless_peep]

The lm78xx regulators have very good long term stability and fairly low temperature drift. With the 7812 it will not be exactly 12V out of the box,but very close.
The important thing to understand here, is what kind of detuning you need to compensate and get rid of, and what kind is expected and can simply be tuned out by the user during performance.
Broadly speaking, the V/oct setting is a real bitch to trim (by ear) and is what is immediately affected by the converter temperature. Slow drifts in my oscillator core produce a uniform frequency drift. A musician can simply turn the tune knob slightly and come back in tune. Maybe in the technical world it's a taboo, but any musical instrument is expected to drift and you will need to retune, mid performance.
To counter that, I am incorporating 9V regulators in the places stuff is referenced to one of the supplies. Apart from the sensor circuit we are discussing here, there is a schmitt trigger based on an LM13700 that needs to be stabilised. Putting everything in an oven is the "mystical" solution- instead of a good design you just brute force it (not very elegant and usually very expensive). I have taken extra care to use as little ICs as possibles in the VCO core, and those used are operating in a way that is fairly independent from the supply voltage. I'm surprised none of you guys didn't talk about resistor and capacitor drift yet, which is much more important than a TL082 working at 30 instead of 25 C.