Thermocouple IC's

Those are medium horrible.

One common problem with t/c circuits, and most t/c instruments, is rotten cold-junction compensation. Errors like 0.1 or even 0.2 degC/degC are common.

We use a platinum RTD and a 24-bit ADC to acquire the ref temp, and convert the measurement to ohms, then do the math to properly convert that to temperature. Then we go through a reverse thermocouple lookup to conveer that to microvolt offset, for the compensation.

Most people apply a diode drop or a bandgap temperature sensor directly into the thermocouple voltage. Errors pile up fast.

Of course, the cold-junction region has to be isothermal. Most people do that badly, too.

Why not use an RTD?

Temperature is hard to measure! That's one reason to be skeptical of certain things.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Hmm, well medium horrible could be worse. (I already ordered a few of each.)

Ouch. that is a lot of work! It would be easier for me to make a second reference junction, stick it on a little TEC (or maybe just a heater?) and hold it's temperature at 0 C or 25 C or 50 C (or something) But I don't want to do that.

Yeah I've seen your layout and such for tc's.. very nice, but I want to be cruder.... with a medium horrible part I can do medium horrible layout and not make it too much worse. :^)

It's an existing product. Thin type T tc for space and magnetic field reasons. (The thin piece of constantan is not very magnetic.) We use these omega controllers I don't like them, and I'm always getting calls for replacements.. Like yesterday, and that just motivates me to try and do it better myself.. with fewer buttons that stop working, or need cleaning.

Oh dear, let's not make too much fuss about that! I understand your skepticism... I've pointed you to the physics prof in Berkeley who was also a skeptic.

George H.

My embedded systems programmer does the work! He seems to enjoy it. But all that has to be done to make an aerospace-grade t/c acquisition or simulation box.

Our customers sometimes want to remotely locate the reference junction, so we allow them to acquire external 4-wire RTDs as refs too.

Here's our latest t/c gadget, which uses all those tricks. I let my new hire, the girl from Mexico, design this one. Everything worked fine first try, except we had to add one capacitor to fix the erratic bahavior of that (expletive deleted) AD7793 ADC.

It looks like a lot of the cold fusion confusion was based on bad calorimetry.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

And most of it wasted;; your temperature range, and the accuracy of thermocouple alloys, only results in 8 bits of useful signal. Ten bits, if you calibrate a batch of thermocouple wire.

If your 'cold junction' is at room temperature, just expand the thermocouple voltage-correction-for-cold function out in series:

deltaV = A0 + A1 * (T - 20) = A2 *(T-20) **2 + ...

and if you stay with T near 20 C (like, plus/minus 5 degrees), it might be that you only need add a constant, and a proportional-to-temperaure term. The sloppy kind of 'use a diode' compensation is maybe good enough.

This might help:

Oh. We used 24.

Most thermocouples seem to be a lot better than the specs.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

I don't really care about accuracy. (I must be able to set one point so it's right,

500mV = 50C or so.) Thermal stability is what I care about. I'll do my own PI (d?) loop, but I want a good input signal. An indoor environment so the local temperature won't change much.

George H.

But, since thermocouples measure the difference between two junctions, your cold-junction compensation has to be correct over the permissible range of cold-junction temperatures, regardless of any constriction on the hot-junction temperature range of interest.

It does mean, you might want to expand the thermocouple correction (voltage) using dV/d(Tcold) with Thot = 50, and d^2V/d(Tcold)^2 with Thot = 50...

Then the reference junction will be your biggest source of variability. And maybe some gain-set resistors.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Just to clarify, the the net thermocouple voltage does not depend on delta-t, it depends on the delta-v at the two temperatures. Since thermocouple potentials are nonlinear on temperature, those are not the same. Most t/c ref junction compensation assumes they are the same.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Thus, the interesting case of using a type T thermocouple to stabilize a 50C oven: if you can warm a reference block to 50 C (as reported by any thermostat design you like) and make it your cold junction, the error signal is simply the thermocouple EMF, no temperature compensation reqired.

It's a lot better to seperate the cold junction from the amplifier.

Use an analog temperature sensor and a good op-amp, and calculate a few resistors.. is that too hard?

--sp

--
Best regards,  
Spehro Pefhany 
Amazon link for AoE 3rd Edition:            http://tinyurl.com/ntrpwu8 
Microchip link for 2015 Masters in Phoenix: http://tinyurl.com/l7g2k48

Grin... Well I would have to look up all the voltages I need. (This way I just pay LT to do my math. :^) And then sweat all uV's of DC offset and drift... The "best" low offset opamp in my "stable" is the opa277,

20 uV. (I use the cheaper 227PA version) Looks about the same as the LTK001. Or do I want a chopper opamp... I've got zero experience with those.

Re: separate CJC, I think that is what these two LT IC's do. Do you agree with JL that the biggest error/ variability will be in the CJC? I always get stuck thinking about typical versus worse case values.

George H.

The delta-sigma ADCs come with 20 or 24 bit resolution, sort of free, so why not? They are nice low-noise integrating ADCs with PPM linearity, and many have mux's and good internal references, fairly cheap.

Integrating for 100 milliseconds notches out both 50 and 60 Hz pickup, and a lot of amplifier noise too.

In real life, even ordinary thermocouples seem to be a lot better than the guaranteed limits. I want my instruments (measurement or simulation) to be a lot better than the RTDs and tcs, so we don't dominate my customers' system error budgets.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

You can roll the CJC room temperature calibration and initial zero offset (and any stable measurement delta between CJC and terminals) into a single point trim. With a 20uV Vos amplifier you're within

and whatever tracking error occurs.

Getting the CJC sensor isothermal to the cold junction connections is usually worse than the sensor accuracy (the slope, I'm talking about here for the latter errors) considerations.

If there's any significant power dissipation in your setup you may find the CJC sensor is running several degrees C different than it should anyway, and worse if the sensor is some stupid signal conditioning chip that has its own self heating and can't be optimally placed.

Chopper amps are cheap these days and work fine- they're noisy but probably not enough to bother you after low pass filtering. Thermocouples are nice low source impedance sensors.

--sp

--
Best regards,  
Spehro Pefhany 
Amazon link for AoE 3rd Edition:            http://tinyurl.com/ntrpwu8 
Microchip link for 2015 Masters in Phoenix: http://tinyurl.com/l7g2k48

You should use a chopper opamp for thermopiles (and thermocouples, if you don't want built-in cold junction compensation). I like the ADA4528.

e.g.:

or this one which only works for positive inputs:

For the first circuit I tried to measure the input-referred offset. It was really too small to measure reliably using my setup but I recall seeing something around 300nV. I suspect that I was having more error from the second stage (OP-07) than the first! Also, if this is in a damp and dirty location then you don't need much leakage between pins 3 and 4 of the OP-07 to give more error than I was seeing.

I don't like the very low supply voltage capability of the ADA4528, but there is a newer one (ADA4522-2 ?) which can handle higher supply voltages. I haven't tried it yet. I can't stand the new ADI website at all, so I only ever see their new parts if they are linked from another website.

Chris

We use ADA4638 in a thermocouple simulator. +-15 rails, max offset 4 uV; we cal out the offset. The lower-voltage parts, like AD8628, have lower offsets.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

CJC?

OK Thanks Spehro, I'll think about how to do the connections. Inside the electronics box there are these big fat TC wires running from the input jack to the measurement input terminal block. I guess with one being a big piece of copper.... Anyway I'll have to think.

George H.

Thanks Chris, I was digging through a pile of parts today. (Things other people had ordered.) And I found an LTC1050 which I put to one side....

George H.

We make a box that has t/c terminations and a reference RTD. The barrier on the left forces the wires to snake around and get a little more isothermal to the PC board.

Luckily, thermocouple wire is usually not a good heat conductor, but

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics