I want to use an op amp with a transdiode to linearize the output of a thermistor.

Thread Starter

studiousscholar

Joined Jul 15, 2008
13
I have a project that will have a number of thermistors. It will require measurement of temperatures higher than the range digital sensors can accommodate. Rtds and thermocouples would be expensive. I would like suggestions as to whether use the thermistor as the input resister with the transdiode in the feedback loop of the op amp, or to use the transdiode as the input with the thermsitor in the feedback loop. In either case I plan on using an analog multiplexer to switch in different thermistors.
Since matched pair transistors and logamps are no longer readily available at a reasonable prices, I have a different idea to use a circuit similar to how a chopper op amp functions to cancel out temperature variations in the transdiode. The way I figure it, there's no better match for a transistor than itself anyway. However that will be the subject of a different thread.
 

Thread Starter

studiousscholar

Joined Jul 15, 2008
13
Since an NTC thermistor has a exponential response to temperature, inputting it in a log amplifier will make the response linear. A transdiode is a bjt connected as a diode. Using a transdiode in the feedback loop of an opamp will make it an log opamp. Conversely using a transdiode as the input would make the opamp an antilog amplifier, then using the thermistor in the feedback loop would make the output linear.
 

Thread Starter

studiousscholar

Joined Jul 15, 2008
13
Thermocouples seem to have a reasonably linear performance, and can be assembled with a spotwelder and appropriate wire material.
https://en.m.wikipedia.org/wiki/Thermocouple
Select one that meets your range of temperature.
There is a slew of problems with making your own thermocouples.
1)You have to have cold junction compensation.
2)The multiplexer resistance would swamp the signal
3)The response from thermocouples isn't linear over a wide enough range.
 

crutschow

Joined Mar 14, 2008
23,542
From the curve below, the thermistor resistance is only somewhat logarithmic with temperature.
It would be a straight line if the device were completely logarithmic.
So a logarithmic correction from the transdiode will still give significant linearity error.
1572769857974.png
How accurate do you need to measure the temperature?
What are you doing with the measurement value.?

What about linearizing the device with a correction factor using a microprocessor?
 

Thread Starter

studiousscholar

Joined Jul 15, 2008
13
From the curve below, the thermistor resistance is only somewhat logarithmic with temperature.
It would be a straight line if the device were completely logarithmic.
So a logarithmic correction from the transdiode will still give significant linearity error.
View attachment 190499
How accurate do you need to measure the temperature?
What are you doing with the measurement value.?

What about linearizing the device with a correction factor using a microprocessor?
 

MrChips

Joined Oct 2, 2009
19,425
What about linearizing the device with a correction factor using a microprocessor?
That would be my choice. Furthermore, I do not use ADC to measure thermistor resistance. I use a simple R and C in parallel connected to two GPIO pins of an MCU. No additional circuitry required.
 

Thread Starter

studiousscholar

Joined Jul 15, 2008
13
I need some of them to be accurate to about 2 degrees Fahrenheit, some of them can be quite a bit less accurate, but that is misleading because I intend to use some of them outside of their rated temperature range. I can do this because NTC thermistors still work above their rated range but this is usually not used because of their nonlinearity, which is especially bad at their high end. The method I am trying will make their response linear over the entire range, up to when parts of the thermistor melt. (By the way, I don't plan on getting them THAT hot.)

I am going to be using them to monitor and control several different flow streams of intake air, feed air, and fuel feed for a foundry furnace. I will be measuring ambient air temp., starting temp. for fuel (used motor oil), using a servo to mix part of the exhaust gas to mix with ambient air to preheat the fuel, a pid controller for this servo, and a pid controller for the fuel pump to keep the oil just below its flash point when it comes out of the preheater.

I am going to be using a microcontroller, but I am going to be using different thermistors for different uses. The ones for ambient temps will be different than the ones for the feeds. I only need to get one of these signal conditioners working for all the different thermistors. If I use a microcontroller for this I will have to calibrate at least two different types of thermistor. The upper range extreme will be about 350 degrees C for the preheat air for the fuel heater.
 

MrChips

Joined Oct 2, 2009
19,425
I can linearize any thermistor over any temperature range to better than 1°F (0.5°C).
List all the situations that you require:

Min temp
Max temp
Make and model of thermistor
 

Thread Starter

studiousscholar

Joined Jul 15, 2008
13
I can linearize any thermistor over any temperature range to better than 1°F (0.5°C).
List all the situations that you require:

Min temp
Max temp
Make and model of thermistor
Any thermistor as opposed to ALL thermistors. I my case eight thermistors over a range of 25c to 350c. Nice hammer you got there. I need T25, a #2 Phillips, a 7/16ths, and 10mm. Thanks for playing. Read the original post. I DON'T NEED A HAMMER.
 

Thread Starter

studiousscholar

Joined Jul 15, 2008
13
I have a project that will have a number of thermistors. It will require measurement of temperatures higher than the range digital sensors can accommodate. Rtds and thermocouples would be expensive. I would like suggestions as to whether use the thermistor as the input resister with the transdiode in the feedback loop of the op amp, or to use the transdiode as the input with the thermsitor in the feedback loop. In either case I plan on using an analog multiplexer to switch in different thermistors.
Since matched pair transistors and logamps are no longer readily available at a reasonable prices, I have a different idea to use a circuit similar to how a chopper op amp functions to cancel out temperature variations in the transdiode. The way I figure it, there's no better match for a transistor than itself anyway. However that will be the subject of a different thread.

I'm reposting this because I apparently did not make my request for help specific enough. I have used Arduinos with voltage divider connected thermistors before. I have used thermocouples with Arduinos before. These are not the solutions to my problem. I need help with nonlinear op amps and nonlinear sensors.
 
Last edited:

nsaspook

Joined Aug 27, 2009
6,506
Most of us are saying that your proposed analog linearize solution will be most likely be inferior to a modern 12-bit ADC/Digital sampling solution.
 

MrChips

Joined Oct 2, 2009
19,425
What we are saying is you do not need any fancy circuits or voltage dividers.
Measure the voltages directly with your ADC.
I prefer not to measure voltages. I measure the resistance directly using a frequency measuring technique, i.e. direct-to-digital ADC.
 

Thread Starter

studiousscholar

Joined Jul 15, 2008
13
I have a project that will have a number of thermistors. It will require measurement of temperatures higher than the range digital sensors can accommodate. Rtds and thermocouples would be expensive. I would like suggestions as to whether use the thermistor as the input resister with the transdiode in the feedback loop of the op amp, or to use the transdiode as the input with the thermsitor in the feedback loop. In either case I plan on using an analog multiplexer to switch in different thermistors.
Since matched pair transistors and logamps are no longer readily available at a reasonable prices, I have a different idea to use a circuit similar to how a chopper op amp functions to cancel out temperature variations in the transdiode. The way I figure it, there's no better match for a transistor than itself anyway. However that will be the subject of a different thread.
Most of us are saying that your proposed analog linearize solution will be most likely be inferior to a modern 12-bit ADC/Digital sampling solution.
What I'm trying to do is a one time, high effort solution as opposed to eight (or more) moderately difficult ones.
 

Thread Starter

studiousscholar

Joined Jul 15, 2008
13
Unfortunatey this has become somewhat of a lost art. This type of circuit at one time was extremely common. The SR71 and the NASA rockets of the sixties were designed with analog computers. A log amp was as common, or more so, as an op amp adder. There is some old fart in his eighties out there that could breadboard this up in a minute.
 
Top