# Temperature Measurement, Part 2

In my first blog in this series, I discussed the generalities of the different temperature measurement techniques. In this blog I will deal with converting the temperature into electronic signals and start with purely analog conversion. All the major analog semiconductor manufacturers produce standalone devices — Analog Devices, Linear Technology, Maxim Integrated, and Texas Instruments amongst them. Just search for both the data sheets and application notes on their websites.

Some of these designs even can detect open-circuit on the sensors and linearize the output. Three design ideas demonstrates how linearization can be achieved for an RTD (IC generates second order polynomial), with an improvement using a spreadsheet (Design an RTD interface with a spreadsheet), and an improvement (A better approach to designing an RTD interface with a spreadsheet) proposed by yours truly.

Figure 1. This is extracted from a Burr-Brown data sheet of an obsolete device (and much lamented, on my part). It is the device used in Figure 1 in my last blog. We can still learn how to interface to an RTD by analyzing the operation.

Using current sources allows a modification of the Wheatstone bridge approach and is useful with 3-wire RTDs. The XTR103 in Figure 1 has two current sources (0.8mA) which are at a low value to prevent self-heating of the RTD. I have also seen the current pulsed to reduce the self-heating. Rz is mounted on the PCB and is equal to the value of the RTD at its minimum temperature of the desired range. One current source drives the RTD and the second drives Rz.

The voltage across the RTD increases with temperature, while the voltage across Rz remains constant, so with the use of a differential amplifier across both resistors you can measure only the change in voltage from its minimum range. The currents join on the return path of the RTD and pass through Rcm. This shifts the common mode of the voltage difference up into the operational range of the difference amplifier.

You can see from the notes on the figure that the line resistance is cancelled out. It is simple enough to replicate parts of this for your design by creating your own pair of current sources, as you can see in Figure 2. Also of interest is the fact that the circuit in Figure 2 has the ability to detect if any one of the RTD wires is broken. Although you may not use this exact approach in your design, the possibility of cable failure is something you need to consider.

Figure 2: Two independent current sources to drive an RTD and zeroing resistor (R11). The difference was fed to an instrumentation amplifier, which is built into the PSoC1.

Let us move our focus to the conversion of the measured RTD signal to digital format.

Modern A/D devices cater for devices like RTDs and include current sources. Since temperature changes very slowly, the Delta-Sigma ADC is ideal for the application, and today there are some with very high resolution. Some also allow for differential inputs, but because of the high resolution it is possible to use a different technique if there is no differential input option. Simply use an ADC system with two input channels, and the micro can numerically subtract one channel from another to give you the difference.

With the high resolution it is still possible to have meaningful results. Let’s assume you have a 20bit ADC with a span of 2.5V — the LSB represents 2.4µV which is 2.4mΩ at 1mA current through the RTD. With a DIN 43760 RTD the sensitivity is about 0.385Ω/°C, so theoretically you could resolve to milli-degrees. There is yet another approach if you have sufficient resolution and inputs. You could configure a precision resistor in series with the RTD and measure the voltage across it to give the actual current flowing. We will see more on these ADC systems in a while.

A thermocouple produces a very small signal and will benefit from a noise-immune amplifier as well as quite a bit of gain, since the signal is so small. As always, protection against surges is good practice. It will also need (as previously mentioned) cold junction compensation.

Here is a suggestion from Linear: Ultraprecise Instrumentation Amplifier Makes Robust Thermocouple Interface. And Analog Devices has a ton of information here: Thermocouple Interface.

In years gone by you would condition the signal based on thermocouple type and temperature range so that the signal would be within the common mode of the input and with the maximum possible output and then feed the signal to an ADC.

Figure 3: This device is a universal thermocouple to 4-20mA signal conditioner. It was based around the 8051 microcomputer and was configured by a PC for the type and temperature range by downloading necessary lookup tables to an EEPROM. The thermocouple input and micro were galvanically isolated from the 4-20mA output.

In many applications the temperature range and thermocouple type can be pre-determined. My employer sells signal conditioning modules, but the Holy Grail is to have a universal input that can cater to all types and ranges so that we only need to make one product and differentiate by software. So the philosophy has changed to use the high-resolution Delta Sigma ADCs as well. Let’s assume the range of the ADC is 1.2 volts and a resolution of 24 bits, so the least significant bit is equal to 72 nV. Further let’s assume that you only want a 250°C range which is about 10 mV for a J-type thermocouple.

That range would be represented by a count of 10mV/72nV=139000 which is about a 17 bit resolution. Now, as promised, we will still get to these ADC systems, but I think it would be better to point out here that the current source on the ADC system (as mentioned in the RTD discussion) can be used to periodically test the integrity of the thermocouple connection, although obviously not at the same time as taking a temperature measurement!

On a practical note, it is possible to use lookup tables or formulae to figure out the temperature, but most published tables take the temperature and give you the voltage when, in fact, you are measuring the voltage and want the temperature. A great resource is the National Institute of Standards and Technology (NIST) Thermocouple Database: NIST ITS-90 Thermocouple Database. I still have its program to produce the inverse tables, but that doesn’t seem to be available any more. However it should be no problem to generate any table using a spreadsheet and the different coefficients, either normal or inverted.

Shifting our focus once again, this time to thermistors. It seems to me that because the thermistor is so simple and cheap, most designers opt for a simple resistor divider technique, but I could be wrong. A single current source driving the thermistor also seems to have some traction in the industry. Again, the current through it can cause self-heating, and, given the sensitivity of the thermistor, this heating could be more problematic than in applications using the RTD.

As you can tell, although they are far from rare I have never used one. I have heard reports, though, that there are tremendous variations among manufacturers for similar parts, so your design had better cater to the possibility of variations during the product’s life. Any technique used to energize an RTD can be used on a thermistor as well, and I won’t spend any more time discussing them, to avoid duplication.

How time flies! We have looked at some techniques for the three traditional methods of measuring a temperature and we still haven’t got to semiconductor sensors. Next time, I promise. Even though this is a blog in process, comments are invited.

## 13 comments on “Temperature Measurement, Part 2”

1. etnapowers
April 7, 2014

“The voltage across the RTD increases with temperature, while the voltage across Rz remains constant, so with the use of a differential amplifier across both resistors you can measure only the change in voltage from its minimum range”

The resolution of the voltage meter is important to measure correctly the change in voltage. This implies a proper amplification of the differential voltage by mean of a proper gain of the differential amplifier.

2. antedeluvian
April 8, 2014

I just came across this reference- Maxim's Thermal Management Handbook

3. RedDerek
April 9, 2014

antedeluvian, For that universal product, I wonder if you can grab values at two conditions, freezing and boiling, you could then have software determine the type and parameters for the temperature device being used.

I believe all parts may give different values at both extremes. I have not really looked into this thought.

4. RedDerek
April 9, 2014

Found this web site that discusses the drawbacks to NIST's coefficients.

http://www.mosaic-industries.com/embedded-systems/microcontroller-projects/temperature-measurement/thermocouple/calibration-table

Also be faster for a microcontroller to execute the equations.

5. antedeluvian
April 9, 2014

Derek

For that universal product, I wonder if you can grab values at two conditions, freezing and boiling, you could then have software determine the type and parameters for the temperature device being used .

I am not sure I follow your argument. In our industry, the customer normally knows what sensor he/she wants and the input temperature range. Often the type is decided by the voltage range or because there is a tradition for that kind of sensor withoput any rationale as to why it was originally chosen.

For argument let's say the rangfe is from -50degC to 350degC. Standard outputs are 0-10V or 4-20mA. Let's choose 4-20mA. That is at the minimum input temperature range of -50degC there would be an output of 4mA and at an input temperature corresponding to 350degC the output would be 20mA. Any temperature between the two extremes would scale linearly in the 16mA output range.

Normally calibration is done with a calibrator (which I believe I mention in part 4) which simulates the sensor and a particular temperature, so that there is a traceabilty and uniformity to the calibration. We had one custiomer who tried to calibrate by sticking his sensor into a mixture of freezing water and then boiling water. He did not get very good results.

6. RedDerek
April 9, 2014

I was thinking more in line with a customer having a thermocouple, not knowing which. Then purchasing a box that will read a thermocouple temperature. But the box needs some base-line temperature input that it could work with – boiling water for example; but boiling temperature can change depending on altitude and barometer values. The box would self-identify the type and then be able to report the temperature afterwards of what ever is measured.

The above is looking at a thermocouple. But to use an RTD or other device there would have to be more intelligence to end up pushing current through to measure the resistance in some fashion.

7. antedeluvian
April 9, 2014

Derek

Found this web site that discusses the drawbacks to NIST's coefficients

Interesting- thanks. I must say it seems a bold step to challenge NIST when your instruments are supposed to trace back to them.

8. antedeluvian
April 9, 2014

derek

I was thinking more in line with a customer having a thermocouple, not knowing which

It's an idea, but I am not sure if it warrants a product. The wire insulation colours normally define which type of thermocouple. Also there must be 8 or 9 (if not more) thermocouple types. I wander how dissimilar they each are at 0 and 100.

9. Sachin
April 12, 2014

There are many gadgets that are used in temperature measurement and they are produced with different companies. It is true that what should be considered here is that the design, however, there might be any difference, it must cater for variation in the product`s lifespan. It is not that obvious that technique used to energize RTD can be used to power thermistor, consider the current flow in it first.

10. Sachin
April 16, 2014

There are so many factors that go into the actual design of a thermocouple that it is highly unlikely that any two thermocouples would give the same readings if they were both calibrated by measuring their extremes. I know of 7 different types of thermocouples and I have to side with antediluvian in saying that at some point (even if you started by measuring any extreme to establish a baseline temperature) you will still need to use a calibrator.

11. SunitaT
April 22, 2014

@SachinEE, I believe there is some significance in the point that you raise about the components being from different manufacturers. But, and please do correct me if am wrong, the temperature measurement and the calibration are carried out on the final product that has been made after putting together all the components from different sources. So if it is to be tested as a single, complete unit then the individual differences between the components should not really be of any significance at that point.

12. amrutah
April 22, 2014

@SunitaT0,

“the temperature measurement and the calibration are carried out on the final product”

This may or may not be true.  The different components of a product come from different vendors.  The different vendors have a specification to meet and so they have their own trimming, calibration, offset adjustments made. Once all the discrete components are put into a product, I think it becomes difficult to calibrate a particular device or part of the component.

All that can be done is a firmware calibration to improve the accuracy.

Others,

Any thoughts??

13. Sachin
April 29, 2014

It is indeed true that there are many temperature measurement gadgets available in the market. I think when it comes to choosing a temperature measuring gadget, design doesn't matter much but what will make prefer one design to the other is input temperature range with respect voltage rage and not the other way round.

“All that can be done is a firmware calibration to improve the accuracy”

@Amruta, I agree with your statement, not every vendor will meet perfectly create a component that will perfectly match with another vendor's product. There must be an error.

This site uses Akismet to reduce spam. Learn how your comment data is processed.