Hey, do anybody have experience with damn accurate thermometry

Thread Starter

Janis59

Joined Aug 21, 2017
1,350
I am willing to make a best choice between three possible thermometry candidates measuring most smallest ever possible temperature DIFFERENCE between multiple points in the rather harsh GHz range EM-noise field :

Must to be dT=0,0007 C what is 16 bits
Must to be digital reading, thus spi or better i2c but both are digestable

Candidate 1=LM95172EWG with ultra-fast reading (=good but not essential); with damn bad metrological linearity 1C (very bad) and noted that "serial interface means never before been noise immunity" (very extra +++ important for my case but I dont believe no a word in this sensless motivation - all the candidates have identical serial interfaces); SPI interface (not best but okay); the tablet have none thermal contact pad (wonder if it is useful for measuring at all)

Candidate 2=ADT7320 with slow measurement 1 sec (rather OK or at least not tragic); metrological accuracy 0,2 C (brilliant); SPI interface (not brilliant but ok); nothing said about noise immunity (great unknown); thermal pad exist

Candidate 3=ADT7410 with slow as the previous, metrological accuracy 0,5 C; i2c interface (wonderful, however it isnt the most important thing); thermal pad exist. ADT 7420 the same but metrology 0,25 C and price 6x up.

Candidate 4=STS31DIS or better STS35DIS with unique metrology 0,1 C and drift 0,03 C/per annum, sensitivity the same 16 bits, SPI interface but relaitively unknown producer (Sensirion) first hear such.

Candidate 5=Max31725 with 50 msec and metrology 0,5 C and i2c

Candidate 6=G-Nico TSYS-01 (metrology 0,1 C) 16 bit - have experience with it cheap end brothers - just wonderful however the soldering of TDFN/QFN is real nightmare); exists 24-bit version (just a sweatheart candy) TSYS01-1 -GNICO-023D however I deny to understand what means resolution 0,01C at 24 bit reading. Something there smells by fishing. Elementary calc show the 24 bits means the 5 micro Celsius.

So - what is the best candidate according Your experience?
And, maybe there are some more candidates unknown to me?
 
Last edited:

jpanhalt

Joined Jan 18, 2008
11,088
I am using the MAX31856 amplifier and linearizer: https://datasheets.maximintegrated.com/en/ds/MAX31856.pdf

In continuous mode, it can complete a read every 100 ms. Resolution is 0.0078 °C (19-bit). You can select continuous or one shot mode. I use one shot, which is a little slower due to the filtering for 50 Hz or 60 Hz, which is user selectable. It also has built in cold-junction, but can use an external one too. I am quite happy with it. Be sure to monitor the "Fault " line.

Maxim makes several similar devices with different numbers of channels, analog and digital.
 

Analog Ground

Joined Apr 24, 2019
456
I think you are looking in the wrong place for ultimate temperature measurement. I have measured to 0.001 C. It takes a discrete sensor like a platinum RTD and some very low noise electronics measuring fractions of a microvolt. For example, a 100 ohm RTD with 1 milliamp of excitation gives 385 nanovolts per 0.001 C . The Maxim part looks pretty good but the sub-milli-kelvin world you are looking for is pretty specialized. That said, my experience is about a decade old. Maybe there is something new.
 
Last edited:

jpanhalt

Joined Jan 18, 2008
11,088
I agree on the RTD . Did it once for characterizing a reaction. However, I assumed the apparent discrepancy between this,
Must to be dT=0,0007 C what is 16 bits
and the six products being considered was to be interpreted as resolution, not accuracy.
 

Thread Starter

Janis59

Joined Aug 21, 2017
1,350
1) I must measure ONLY temperature DIFFERENCE not the temperature itself. Example - one sensor shows 20,00001 C and another shows 20,00003 C (means about 293 Kelvin), then I must log into logbook the 0,00002 C. So I am not interested to know anything about linearity or metrological warranty but only the sensitivity of younger bit. The whole measurement range is maximum ever imaginable maximorum minus 20 to plus 40 Celsius and maximum T difference never may be larger than one centigrade.

2)However EMI immunity is rather bad thing there, something so bad as maximum 200 V/m of 1 GHz up to 300 GHz may happen (or not) and it ought not to paralyse the sensor. And more than sure some 1...10 V/m there will be all the time.

3)Therefore all kind of thermoelectric couple, platin resistor, thermopile sensor or all other kind of low-voltage devices ought be condemned.... however, may be, my noise never will be slower as some 400 MHz thus by proper filtering immunity may be solvable.... but no, I must have a as much as possible cheap and damn easy replicable system not a system where installed 4 thermometers each for 500$. The aim of project is factory replicable product.
 

jpanhalt

Joined Jan 18, 2008
11,088
1) I must measure ONLY temperature DIFFERENCE not the temperature itself. Example - one sensor shows 20,00001 C and another shows 20,00003 C (means about 293 Kelvin), then I must log into logbook the 0,00002 C. So I am not interested to know anything about linearity or metrological warranty but only the sensitivity of younger bit. The whole measurement range is maximum ever imaginable maximorum minus 20 to plus 40 Celsius and maximum T difference never may be larger than one centigrade.
Now you have added another zero...

1) What stumps me is that you consider the MAX31725 a candidate:
1584735411098.png

when you must have two orders of magnitude better. Presumably that applies to the other candidates too.

2) Have you considered what thermal noise will be at that frequency and 293°K?
 

joeyd999

Joined Jun 6, 2011
4,477
That's one interpretation.

Another is that your sensor has a lot of short-term noise (thermal?) and drift. Was THAT your validation?
Interpret it as you will. If you read the thread, I specified the total noise in both peak-to-peak and RMS. But the resolution, as displayed on the graphs, was 10µ°C. As far as accuracy, I don't believe I was better than 0.25°C -- but that was not important for that test.

Measuring small signals is hard. Measuring small temperature changes accurately and and with high resolution and absolute accuracy is even harder.

The point of that design was to provide about a 140dB dynamic range for the property I was trying to measure. This entailed not just one, but two very high resolution temperature sensors to compensate the 2 "property" sensors.

The design was a resounding success. That's the only interpretation I need.
 

jpanhalt

Joined Jan 18, 2008
11,088
When I validated any analytical system, I always used some independent measure. Showing change or even a measure of randomness means nothing. A "sink test" can be extremely reproducible. Did you compare to any standard such as a quartz thermometer? (An example: https://ocw.mit.edu/courses/mechani...ents-13-998-spring-2004/readings/seawater.pdf)

It's not a question of whether you did or didn't do it, but a question of whether you can show you did it. So far as we know, you just had a constant temperature bath that wasn't working properly.

Oh, by the way, Wikipedia was not very helpful. Search on "microdegree" : https://en.wikipedia.org/wiki/Microdegree

I am well past that stage in life.
 

cmartinez

Joined Jan 17, 2007
7,420
No catch. My customers pay for the details. How much you willing to spend?

Here's one you haven't heard: my grandfather invented binary. I know many will scoff, but this is bona fide true.
Your grandfather was Gottfried Leibniz? ... you must come from a family where extreme longevity is common...

And as far as spending goes... I'd like a refund for the time I spent in the previous two posts and this one... ;)
 
Top