Square root circuit

R9 just softens the steep zener slope and defines the gain better.

OK, but R9 shouldn't get very hot.

I should probably do something less clever, like a straight thermostat.

I don't really understand thermostats, but they always seem to work.

The guys who designed the original assembly, the science types, apparently had a lot of thermal transport delay from the heaters to the sensor, with a linear loop, and it's not stable. Thermostats seem to be always stable... I think. But a better thermal path would sure help, which I can do by soldering the ground end of the thermistor to the ground end of Thing.

Den onsdag den 1. juli 2015 kl. 20.52.49 UTC+2 skrev John Larkin:

how about one of numerous protected fets? current limited and temperature shutdown

tiny zener from drain to gate and you have a power zener

-Lasse

"Oh Oh!".. vigorously waving his hand... Put the sensor right next to one of the heaters.. ala Phil H. (and Bill S.)

Re thermistor thermal coupling: or use a diode connected transistor as temp sensor and stick it's collector to ground.

George H.

For a factor e in collector current, the base has to go up 2kT/q. Say you have it running, with some programming current I_p getting you a collector current I_C, and you want it to go to (1+epsilon) I_C. This will require (1+delta) times as much programming current.

delta V_B = 2kT/e *epsilon due to the two transistors in series.

delta V_B = kT/e * delta

The two have to be equal, so delta = 2 epsilon. This is a differential, so when you integrate, I_C goes as sqrt(I_p).

There are all sorts of galloping offsets and drift, but inside a loop that has enough current drive available, it looks OK.

Cheers

Phil Hobbs

Depends on the situation. I really needed a simple analogue loop with nearly constant bandwidth, because I have this trick I'm trying out to make faster loops, and it's going near some sensitive transducers.

Cheers

Phil Hobbs

Here's the .asc of one version.

Cheers

Phil Hobbs

---------- Version 4 SHEET 1 880 680 WIRE 16 -144 -80 -144 WIRE 64 -144 16 -144 WIRE 432 -144 64 -144 WIRE -80 -128 -80 -144 WIRE 16 -112 16 -144 WIRE 192 -80 64 -80 WIRE 64 -64 64 -80 WIRE 192 -64 192 -80 WIRE -80 -32 -80 -64 WIRE 16 16 16 -32 WIRE 16 16 -16 16 WIRE 64 16 64 0 WIRE 64 16 16 16 WIRE 192 32 192 16 WIRE 432 32 432 -144 WIRE 64 48 64 16 WIRE -80 96 -80 64 WIRE 432 144 432 112 WIRE 64 176 64 128 WIRE -80 240 -80 176 WIRE 64 288 64 256 WIRE -80 352 -80 320 FLAG 432 144 0 FLAG -80 352 0 FLAG 192 32 0 FLAG 64 288 0 SYMBOL voltage 432 16 R0 WINDOW 123 0 0 Left 2 WINDOW 39 0 0 Left 2 SYMATTR InstName V1 SYMATTR Value 16 SYMBOL res -96 224 R0 SYMATTR InstName R1 SYMATTR Value 10 SYMBOL res -96 80 R0 SYMATTR InstName R2 SYMATTR Value 32 SYMBOL diode -96 -128 R0 SYMATTR InstName D1 SYMATTR Value ES1D SYMBOL pnp -16 64 R180 SYMATTR InstName Q1 SYMATTR Value D45H11 SYMBOL diode 48 -144 R0 SYMATTR InstName D2 SYMATTR Value 1N914 SYMBOL diode 48 -64 R0 SYMATTR InstName D3 SYMATTR Value 1N914 SYMBOL res 176 -80 R0 SYMATTR InstName R3 SYMATTR Value 1kR SYMBOL res 48 32 R0 SYMATTR InstName R4 SYMATTR Value 3k SYMBOL voltage 64 160 R0 WINDOW 123 0 0 Left 2 WINDOW 39 0 0 Left 2 SYMATTR InstName V2 SYMATTR Value PULSE(16 0 1m 10m 10m 5m 30m) SYMBOL res 0 -128 R0 SYMATTR InstName R5 SYMATTR Value 22k TEXT 280 392 Left 2 !.tran 12m

So what is wrong with the obvious answer? Isn't there a good reason why it is obvious? I guess if you can do it easily and simply with analog circuits that is good. After all, every MCU has to be not only designed, but incurs recurring costs for programming it. No matter how small the package, how cheap the cost every MCU is a small problem of it's own accord because it contains software.

George, in the back row?

But then I'd be closing the loop on the heater temperature, not the Thing temperature.

That would sure have a short thermal time constant. A thermistor would be more accurate, in that I could predict the setpoint temperature right off the thermistor data sheet, and pretty much depend on it.

An LM45 would work, with its ground pin soldered to the Thing. But it needs a power supply.

Zetex males a cool nickel RTD in a SOT23 package. The top lead is the thermal contact, isolated from the resistor.

And non-recurring costs for programming it! Ten years from now, my soldering iron will still warm up in 30 seconds, but I can't be sure that I can still find/run some compiler.

Sometimes a few opamps and resistors is just the thing.

Yes, I mentioned the costs of designing the MCU. But the idea that you would not be able to *redesign* the MCU in ten years is a bit silly. Ask Joerg what his target lifespan for small MCU devices is and what he achieves. I expect the latter is significantly more than 10 years. Ever hear of the 8051?

you can't be sure you can find what ever magic analog part you need either

buying a preprogrammed mcu is just like buying any other single sourced part

it is, and sometimes an mcu will replace so much stuff you could stick a spare on the pcb and still make is smaller and cheaper

-Lasse

-Lasse

An mcu that replaces two diodes and three resistors is silly.

Resistors and diodes will be around a lot longer than some MCU. Replacing X cpu with Y cpu will be a project.

Repulsive little beast. The 8051, not you.

Diodes, resistors, caps, an opamp in a SOT23 package. Should be around for a while. You can still buy 741 opamps.

We buy tons of programmed LPC1754s from Arrow. We use 12 per PC board. Each one does some serious algorithms and isolated SPI, so that makes sense. Using one in a simple, low-volume temperature controller probably doesn't. George can't PWM, so that leans towards opamps too.

I'd guess we would spend at least a man-week developing, testing, documenting, and releasing a uP based temperature controller design. It still needs a PCB layout and all that stuff, same as an analog design. I could design the analog version in an hour or two.

Sometimes.

The 8051 is kind of brain dead, even though they used it in a lot of stuff. It was enormously popular as the brains behind a bunch of analog polysynths in the mid 1980s, for example.

But it's kind of brain dead and hard to write good code for, kind of like low power x86 anything (Intel Atom, etc)

This fellow has gotten a lot of design ideas out of BJTs.

Using PWM makes the problem go away. Power is proportional to on-time.

The people in Germany who used to convert their cars to run on (un-transesterified) straight vegetable oil used a bunch of regulators with their outputs shorted, as a thermostatic heater for some part of the engine.

If you find a regulator without hysteresis on the thermal shutdown temperature, it could be quite a nice heater. Unfortunately the thermal shutdown temperature is normally higher than the maximum specified junction temperature, so the part is not specified to be reliable when it is permanently in thermal limiting, and the chip designers probably did the electromigration current density calculations at a lower temperature so it is likely to fail eventually.

If one of the chip manufacturers would do a variant of the 3 terminal regulator that has a somewhat lower and well specified shutdown temperature, with no hysteresis and guaranteed reliability as a heater, it would be a nice solution for small crystal ovens etc. I guess that then we would lose the main advantage of the regulator as thermostatic heater which is that it is dirt cheap.

Actually the LTC4059 IS specified to run constantly in thermal regulation mode, at 115 deg C. Maybe that is what I was asking for.

Chris

Your resistors and diodes are not the full solution. You need to add at least one opamp and likely more. A same size and cost MCU can handle everything other than the temperature measurement (and some can even do that too) to driving the load switch.

I get that adding something that needs to be programmed adds a type of complication, but it can also simplify the BOM and design issues considerably.

I would have agreed with you about the 8051 being a really crappy MCU. But Joerg has been able to point out to me the quite significant advantages of its insanely long life, including that you can still find versions that are footprint compatible - drop in replacements. So it is even second sourced if you want. If you need longevity, the 8051 is the way to go, doing better than many analog parts with no end in sight.

If you really want to design with materials that will be around for a long time, perhaps you should take up woodworking. ;)