Precision heater chips (fixing temperature instability)?

Given that you're doing a one-off, and that if you're halfway decent at closing a control loop the solution you propose will work with a known amount of fuss, and if you use an off the shelf chip the chances are even that it'll work like a charm or never work at all -- I'd stick with what you have.

Maybe use a little PIC for a controller if you need really long time constants and you get tired of fighting with your capacitors, but that only if you like the idea of closing a loop with a microprocessor.

^^^^^^^^^^^ How many are we talking about here? Can you manage to stuff all of them into a single temp controlled enclosure?

Hard to tell, from the JPG, but it looks like a VCO with a triangle output? Is that so?

Transfer function desired?

...Jim Thompson

Just two oscillators, each with its own CA3046, some three inches apart on the PCB -- so I'd need two controlled heater circuits. No problem at all to build (I estimate it at an hour or so), but I just wondered if there were ready-made solutions available.

Richard Rasker

...

Yup, it produces a triangle or sawtooth.

Not really necessary, I just wondered a) what caused the temperature instability, and b) why the built-in temperature compensation (one of the '3046 transistor used as a heater) doesn't work properly -- but I'm slowly beginning to understand both things. Especially the failing compensation is something to take yet another look at, because fixing that would mean I no longer need to build my own heater circuit. Anyway, it's funny how you can get the "feel" of a circuit simply by looking at it and thinking about it long enough ...

Thanks for the response anyway :-)

Richard Rasker

Indeed! I've always maintained I "think" and "speak" circuits ;-)

That current diversion AGC doesn't do away with kT/q temperature terms.

The reason I was asking is that there are trivial ways to amplitude control without adding temperature sensitivity.

...Jim Thompson

Have you thought about replacing the caps in the OSC section, with more stable types ?

Also, you can place a diode in series with the circuit to bias it when the diode heats up, place the diode near the hear source and have it in the circuit to help balance it.

...

Did that at my very first attempt, with less than stellar results. But as soon as I fix the instability with the transistor array, I'm sure that the new caps (2% tolerance, as opposed to the 5% types previously used) will pay off after all. And oh, I also replaced all 270-degree single-turn trimmer potentiometers with multiturn precision types.

Hmm, I think I'll try precision temperature control of the transistor array first, before introducing yet more components. But thanks for the suggestion anyway.

Richard Rasker

You might try something like the following before going to the trouble of building an oven. Try to isolate the component(s) that are most sensitive to temperature change. Get a hot air soldering tool and direct a warm stream of air at each component in turn until you find one that really affects the tuning (oscillator frequency). Replace that component and see how the instrument performs. Alternatively, get a can of freeze spray and direct a very quick blast of cold (1-second blasts) at components until you find one that seriously affects the tuning. Replace.. etc. I've found that semiconductors and resistors are usual culprits.

Cheers

Erm, this exact procedure is what led me to the CA3046 being the prime suspect (and convicted and jailed already) in this case. But thanks for your contribution anyway.

Richard Rasker

You could fabricate a heatsink for the devices. Use a big chunk of metal with lots of thermal mass so the frequency drift isn't as noticeable over short term.

Also, there are probably MOOG forums where this problem is discussed.

Until the 3046 configuration is changed, the OP will always have a TC to contend with.

The OP should "do the math" to understand the issue, rather than hand-waving.

...Jim Thompson

That won't help him if the ambient temp is causing the frequency shift.

Right. Although this is a bit outside my field, I'd model the entire oscillator to see what else might be temperature dependent. It might turn out that more components, possibly the entire oscillator need to be put inside a temp controlled environment.

temperature

is

^^^^^^^^^^^

them

on

to

were

Another potential issue may be that, the capacitors in the oscillators have high tempcos. They may not be fun to temperature stabilize.

Well, in theory you're right. But this isn't my synth, and the guy who owns it doesn't want me to put dozens of hours into it, as that would be a bit too costly. So I resorted to "hand waving" after all -- or more precisely: waving a can of freeze spray to cool down individual components, while monitoring each oscillator's frequency. It turns out that the transistor arrays are by far the most heat-sensitive components. So now I have those heated to a toasty

Richard Rasker

Yep. That solves it because... that 3046 current diversion scheme is fundamentally flawed... for amusement, analyze that transfer function using the transistor non-linear model.

Also the triangle wave oscillator is about as hare-brained as they come ;-)

...Jim Thompson

Use a monolithic quad e.g. a MAT04. The transistors are laid out in a highly symmetric square array, so you can use Q1 & Q3 as a diff pair, Q2 as a temperature sensor, and Q4 as a heater. Works great. The LM3046/86 has 6 transistors and isn't quite as symmetric, but the on-chip ovenizing trick works pretty well.

Cheers

Phil Hobbs