RTD linearization

On a sunny day (Sun, 13 Dec 2009 11:19:49 -0800) it happened John Larkin wrote in :

would be better.

signals

Then why not use a simple chip sensor? i2c or any other protocol exists. My inside thermostat is the temp_pic:

You can see the system working in real time here now:

I have written the soft so it is in 1 C steps... The outside sensor is a LM135 or something like that, almost connected like your cicuit, but only 3 meters wire. It will be freezing here tonight perhaps. Nice and warm inside, all temps ara logged too, and can be controlled from anywhere in the world.

How do you cancel the offset error from the opamp? I'd go for an 12bit A/D converter with internal mux. In my circuit the voltage across the RTD varies 2.8mV per degree Celcius. With a 12 bit A/D referenced at

Why not a variation on Jim Williams' idea...

I've recently implemented such a variation on a custom chip.

Advantages...

AC Gain avoids DC offset accumulation

It's trivial to implement DC restoration (aka pick your own baseline) ...Jim Thompson

There's also a clever way to use a third current to compensate for the base resistance of Q2 and lead wire resistance. Hard to filter the snot out of it if you're switching currents around though, and diode-connected transistors are reputed to rectify.

Best regards, Spehro Pefhany

On the chip I (naturally) used multiple devices switched in and out rather than changing the currents. ...Jim Thompson

Well, it's your cabin :-)

I've seen this sort of stuff go poof despite rather large caps. It was sometimes hard to convince the original designer (who has to ECO this) that the double-diode plus one more resistor would be good but when the new field failures numbers come in during the year they become believers.

Yes, they are nice.

delta-Vbe temp sensors are cool.

There are at least two amusing errors in that Williams article-- o as drawn the current ratio is 2:1, not 10:1 o the chopper amp switch only ever feeds the chopper 0v(!)

But, we know what he meant...

-- Grins, James Arthur

About -2.47k depending on what range you want.

Cheers

Phil Hobbs

Gee, no one wants to play?

Another error: the LTC1150 feedback resistor should be 1 meg, not 1 ohm. (the stage needs a gain of about 504 to scale 198uV/K to 10 volts @ 100c.)

Does anyone see the advantage to Q1? Two resistors in series to +15v, switch across the lower resistor, would be simpler, reduces the influence of Vbe on i(c), and minimizes the switch's on resistance.

As it is, on the "100uA" setting, Q1 forces significant base current into the 330 k base network impedance, causing a beta and temperature dependence. That isn't obviously useful.

-- Cheers, James Arthur

Ignore it. The offset is 100 uV typ, and the signal is 5 millivolts per degree C. And I'm digitizing a 0.1% 1K dummy resistor as a reference anyhow.

I'd go for an 12bit

I'm using a purchased data acquisition board to talk serial to my PC

which has a 10 bit ADC. It does have 8 analog inputs, but I need my own mux so that I can add the gain after muxing.

In my circuit the voltage across the

I'd only get about 1 degree C resolution without the amp, and how much fun would that be?

I'll probably have enough noise that software lowpass filtering will be a good thing to do. Temperatures don't change all that fast anyhow.

I just finished the signal-conditioning/control board layout. Two layers, thru-hole parts, a time warp for sure.

John

With those base resistor values on Q1, it's not a very good 2:1.

Or two equal resistors in parallel, one switched and one not.

Oh. Right. I said that!

Jim W does some weird stuff sometimes. I think he mostly fiddles until things sort of work. His stuuff tends to be "component rich" and the only way he ever compensates control loops is by adding big caps.

John

I got -2750 for -40 to 40C.

Go back to the beach! Have one of those mango drinks with an umbrella in the glass. I think that keeps the rum from getting diluted in case it rains.

John

Those of us in the know actually run the equations and sort it all out. Ms prissy Pants needs staff ;-) ...Jim Thompson

e)

No, that's worse! It'd have to be 49.9k and 499k to get the ~10uA-to- ~100uA Williams wants. (And that doesn't yield an accurate 10:1 either, obviously.)

But @ i(c)=3D100uA, a 200-gain PNP will dump half a uA into 330k, causing a ~170 mV error. On a ~4.4V emitter bias that's a 4% error.

It doesn't matter much--as long as the current ratio is stable, the circuit's initial trim will work across all future sensors. But variations in the PNP's beta would affect the initial calibration, plus, beta changes with temp.

That compromises one of the circuit's appeals--that it can be accurate /without/ calibration.

-- Cheers, James Arthur

I'm just heading down there now. Occidental Grand Papagayo, Liberia, Guanacaste, Costa Rica. Nice place. Cheap, too, thanks to #1 daughter being in the travel business.

(The 2.47k number is for a much wider range---50 to 250 iirc. It's from my thermal chapter on the web site.)

Cheers

Phil Hobbs

ine)

Hi James, I think delta-Vbe temp sensors are cool too! (But I'd rather call it a Ebers-Moll temperature sensor. The shorting of the base and collector is crucial. If you just fed current into the base the I-V slope is not as close to ideal.) I found that the collector resistance of Q2 also causes an error in this circuit. (Sphero mentioned this already.) I determined a collector resistance of about

George H.

If I buy 1K thinfilm RTDs and 1K 0.1% thinfilm resistors, I can hit a fraction of a degree C absolute accuracy with no calibrations whatsoever. The signal levels are millivolts per degree C, not microvolts, and emi hazards are esentially nil. The ADC only needs to be linear.

The ceramic slab and 1206 platinum RTDs that we've tested are much better than their guaranteed specs. I can't imagine how they manage to trim them that well.

John

eline)

Yes, diode-connected NPN Q2 is the sensor. You can also use a straight diode there, but the diode-connected transistor behaves better.

But I was questioning the PNP current-source (Q1) that feeds it. Q1's i(b) produces a significant error voltage in its 500k-1meg bias network, thus on its base node, which also injects a temperature dependence into the switched current ratio (it makes the high/low ratio increase with temp.).

I was too lazy to evaluate all the higher-order effects of that. Maybe Williams is trying to compensate something, but first-order, it looks counter-productive.

Replacing that Q1-based current source with a couple switched resistors avoids that problem, is more accurate, and it's simpler.

Yes, and reduces wiring resistance error too.

-- Cheers, James Arthur

Wow, looks like it's turning into a major science project now. Pretty soon you can say with confidence that the temperature in Truckee is exactly 22.74 degrees Fahrenheit :-)