Ovenizing vs. compensation

On a sunny day (Wed, 03 Jun 2015 15:10:07 -0400) it happened Spehro Pefhany wrote in :

Power consumption

Predictability and linearity of the tempcos. And maybe thermal hysteresis.

Compensation can get pretty good if you are willing to thermally cycle and make a cal table for every unit. That's a nuisance. Ovenizing doesn't need that.

Some parts, like manganin shunts and quartz crystals, can have a parabolic TC, namely a flat/sweet spot at some known temperature. In that case, ovenizing exploits that curve and works much better than compensation. Really good OCXOs tweak the temperature to park exactly on the flat spot.

Ovenizing an entire circuit lets you use (and not think too hard about) lots of cheap parts.

We've done it both ways. Here are some ovenized circuits:

Three have a manganin shunt and a DAC and stuff, PPM stable. One is an XO. All run at about the sweet spot.

I did a design for

a gezillion years ago (literally), that tabulated compensation capacitance versus temperature for a crystal oscillator and wrote it into ROM... which data then controlled a voltage-variable capacitance, thus getting TCXO performance without the power needed by an oven. ...Jim Thompson

The "classic" TCXOs use a couple of thermistors and trimpots or selected resistors to fake a polynomial temp-voltage curve into a varicap. AT-cut crystals are parabolic, so a linear compensation is not ideal.

I think that most all TCXOs are now digital, with a rom like you describe.

Some TCXOs don't use a varicap, they *skip cycles* to tweak the frequency. That's fine if you're making a wall clock, not so good if you're running serious logic.

Willing to do that, the trade-off in this particular case is that mass is expensive (which hurts both approaches) and time/parts cost is cheap. Power consumption is a consideration, but a watt or so isn't a deal killer.

You point is well taken in general that if you have a good oven you're done, no fuss, no muss. In one compensated device we do two- dimensional curve fitting and interpolation which is a total PITA to calibrate.

You can even put most of the controller in the oven.

Thanks!

something.

gs.JPG

The oven is probably slower in responding to changes. Either outside change in T or internal change in power consumption.

George H.

Well, duh, that makes it more stable!

One was for controlling magnet currents in a particle accelerator, for Jlabs. Two were the core controllers for pulsed-gradient drivers for NMR systems. I only get to keep the ratty dirty kluged engineering units; the production ones are prettier.

Spehro wants low mass, so the obvious fix, massive amounts of aluminum, are out.

There's something to do with gradients and such like too.. both will work better with a bit of insulation and some thermal mass in the isothermal-ish region.

There's a bunch of commercial stuff (pressure transducers, accelerometers come to mind) that are first-order internally compensated reasonably well under *static* conditions, but all heck breaks out under dynamic conditions. And usually that information is missing or obfuscated.

An early low-noise amplifier (uA725, from Fairchild) had very impressive modeling (probably SPICE) results, but on testing had much poorer performance. When the engineers figured out the problem, they put on a demonstration, with a jitter/noise display, of one of the new chips. Lots of noise and jittery offset. Dunk the circuit into a tank of oil. Noise goes way down. Audience applauds.

As in all things, it depends on a bunch of factors. How much space & power you have available vs. how precise it needs to be and how much time you have to mess with it.

Big, definitely works -- ovenize.

small, guaranteed employment for someone -- compensate.

Having said that, I had a small contract with an Electronics Manufacturer who Shall Not Be Named* on the possibility of improving the performance of an ovenized product over temperature, because even an oven isn't perfect. If you really need to go there, you can get into all these absurdities with nested ovens and heat spreaders and whatnot.

A former coworker of mine worked on cell phone test instruments, and with at least one of them they went down Compensation Road. They found that if you really, really need to compensate out everything, then you not only have to compensate at every temperature, but you have to compensate for temperature gradients as things heat and cool. Different locations on the board(s) heat up and cool down at different rates, and have components with their own thermal sensitivities, so life gets -- interesting.

• It's in the contract. I'd have to give up my first born, which actually tempts me to go ahead and violate that provision, but they'd want to sue me for bux, too, so I guess I'll live by it.

I should say, I've been thinking mostly about the oven-model lately.

Some cold gas flow, a thermal shield with heater, around an internal sample. And how much mass can you spend on the shield? How big do you need?

At which point I always fold back to the needed time constant.

George H.

Some TCXOs have gradient problems. The TC is great steady-state, but any changes - even air currents - wiggle some bits here and there. The temperature sensor senses one point in the assembly. Putting a cover over a TCXO improves the phase noise... reduces transient forcings.

So your substrate should be low thermal resistance, and as massive as you can stand, and buffered from the outside world.

I don't know what material has the best specific heat to mass ratio, or the best thermal conductivity to mass ratio. I should know that. Probably diamond.

One Outfit That Shall Not Be Named (initials are SRS) makes an ovenized SC-cut XO. The thermal design is awful. The thermal path from the heater mosfet to the sensor is a lot of thin metal, so the thermal delay is slow and diffusive. The resulting stable controller is low gain, proportional only, horrible. So they added another sensor outside to sense ambient and added a scaled feed-forward correction to make up for the wimpy slope of the main loop.

Works pretty well, actually.

I think I know some of the SRS design guys, at least the pres. John W. and other's I talk to at trade shows. Except for trivial stuff, I've got no complaint with their gear.

Getting thermal stuff right for some is hard, my boss likes big thermal masses.. stability. I want something small and quick that I can shake around.

Are there any good thermal books?

Oh I've got a good thermal control loop question. I wanted to make a circuit that did P, PI and P plus negative feed back around the plant. But doing the last one means that I have to move the gain stage.

George H.

Their stuff works pretty well, but the user interfaces tend to be very clumsy. Some of their newer boxes may actually have spinner knobs.

I have a few, but they aren't very good. One is all in crazy American units, like BTU/(hr-ft-degF).

Here's some stuff I've collected:

Got a sketch?

Diamond has anomalously low heat capacity, though its conductivity is so much higher than others that it should be no contest.

Nothing really has higher specific heat, on its own. There's a theorem for that, I think. Phonons and crystals and such.

Vaporization inside a sealed pressure chamber would be much more effective; it would also have the benefit of better heat conduction even than diamond, I think (i.e., heat pipe action). Possibly not as fast for very rapid changes?

Tim

A couple of alternatives are possible and may work better than a massive substrate.

On option would be to scatter thermistors and heaters all over the outer walls of the device which could be made from PCBs - these could be SMT thermistors and transistors or resistors. With an individual PID loop for each thermistor, gradients could be made small without adding much mass. Many PID controllers for the many thermistors and heaters could fit in a small microcontroller.

Another option would be a fairly thin and light aluminium can surrounded by light insulation, then surrounded by another thin, light aluminium can. Due to the intervening layer of insulation, the total mass of aluminium is much less than an equivalent-performance single can. It is a bit like if you want to minimise the voltage drops between the emitters of the two transistors in a current mirror, make that track separate from the ground track to your big electric motors and battery chargers. Just interconnect it at one point. Wenzel has a paper about that, see figure 7 in this:

If one is not dissipating much power in the circuit that is to be ovenised, then a vacuum flask works nicely.

Maybe consider doing both - then the oven doesn't need to be as good, and the compensation doesn't need to be as good.

Obviously if you can, then eliminate or reduce the thing with the bad tempco that you are trying to fix, as this is better than compensation.

Objective criteria for comparing approaches...

Maximum ambient temperature that you can deal with - the oven setpoint has to be higher than the highest expected ambient temperature, especially if the circuit itself has significant power consumption.

Power consumption

Mass (as you mentioned)

Size

Cost

Shock and vibration susceptibility

Component count & reliability as a whole

Any reduction in reliability of the circuitry due to the added oven thing forcing the junction temperatures of the circuitry (and any passives) to be higher

Circuit misbehaviour (drift etc.) as a function of ambient temperature

Circuit misbehaviour as a function of gradient in ambient temperature or heat flux or radiant heat flux.

Circuit misbehaviour for a given turbulent airflow of constant temperature air (if the circuit itself has significant self-heating that is affected by drafts)

Circuit misbehaviour due to a given turbulent flow of different temperature air that is not well mixed