IC Temperature Sensor Provides Thermocouple Cold-Junction Compensation

National Semiconductor
IC Temperature Sensor Application Note 225
Michael X. Maida
Provides Thermocouple April 1979

Cold-Junction
Compensation
Introduction acteristic differs from that of an ideal thermocouple and has
nothing to do with nonlinearity. Tolerance is typically ± 3⁄4% of
Due to their low cost and ease of use, thermocouples are still reading for J, K, and T types or ± 1⁄2% for S and R types, so
a popular means for making temperature measurements up that a measurement of 1000˚C may be off by as much as
to several thousand degrees centigrade. A thermocouple is 7.5˚C. Special wire is available with half this error guaran-
made by joining wires of two different metals as shown in teed.
Figure 1. The output voltage is approximately proportional to
Additional error can be introduced by the compensation
the temperature difference between the measuring junction
circuitry. For perfect compensation, the compensation circuit
and the reference junction. This constant of proportionality is
must match the output of an ice-point-referenced thermo-
known as the Seebeck coefficient and ranges from 5 µV/˚C
couple at ambient. It is difficult to match the thermocouple’s
to 50 µV/˚C for commonly used thermocouples.
nonlinear voltage vs. temperature characteristic with a linear
absolute temperature sensor, so a “best fit” linear approxi-
mation must be made. In Figure 2 this nonlinearity is plotted
as a function of temperature for several thermocouple types.
The K type is the most linear, while the S type is one of the
least linear. When using an absolute temperature sensor for
cold-junction compensation, compensation error is a func-
tion of both thermocouple nonlinearity and also the variation
in ambient temperature, since the straight-line approximation
to the thermocouple characteristic is more valid for small
00747101
deviations.
VOUT . ∞(TM − TREF)

FIGURE 1. Thermocouple

Because a thermocouple is sensitive to a temperature differ-

ence, the temperature at the reference junction must be
known in order to make a temperature measurement. One
way to do this is to keep the reference junction in an ice bath.
This has the advantage of zero output voltage at 0˚C, mak-
ing thermocouple tables usable. A more convenient ap-
proach, known as cold-junction compensation, is to add a
compensating voltage to the thermocouple output so that the
reference junction appears to be at 0˚C independent of the
actual temperature. If this voltage is made proportional to
temperature with the same constant of proportionality as the
thermocouple, changes in ambient temperature will have no
effect on output voltage. 00747102
An IC temperature sensor such as the LM135/LM235/
LM335, which has a very linear voltage vs. temperature FIGURE 2. Thermocouple Nonlinearity
characteristic, is a natural choice to use in this compensation
circuit. The LM135 operates by sensing the difference of Of course, increased error results if, due to component
base-emitter voltage of two transistors running at different inaccuracies, the compensation circuit does not produce the
current levels and acts like a zener diode with a breakdown ideal output. The LM335 is very linear with respect to abso-
voltage proportional to absolute temperature at 10 mV/˚K. lute temperature and introduces little error. However, the
Furthermore, because the LM135 extrapolates to zero out- complete circuit must contain resistors and a voltage refer-
put at 0˚K, the temperature coefficient of the compensation ence in order to obtain the proper offset and scaling. Initial
circuit can be adjusted at room temperature without requiring tolerances can be trimmed out, but the temperature coeffi-
any temperature cycling. cient of these external components is usually the limiting
factor (unless this drift is measured and trimmed out).
Sources of Error
There will be several sources of error involved when mea- Circuit Description
suring temperature with thermocouples. The most basic of A single-supply circuit is shown in Figure 3. R3 and R4 divide
these is the tolerance of the thermocouple itself, due to down the 10 mV/˚K output of the LM335 to match the See-
AN-225

varying composition of the wire material. Note that this tol- beck coefficient of the thermocouple. The LM329B and its
erance states how much the voltage vs. temperature char-

© 2002 National Semiconductor Corporation AN007471 www.national.com

AN-225
Circuit Description (Continued) With a room temperature adjustment, drift error will be only
± 1⁄2˚C at 70˚C and ± 1⁄4˚C at 0˚C. Thermocouple nonlinearity
associated voltage divider provide a voltage to buck out the results in additional compensation error. The chromel/alumel
0˚C output of the LM335. To calibrate, adjust R1 so that (type K) thermocouple is the most linear. With this type, a
V1 = < 5˚C T, where < 5˚C is the Seebeck coefficient compensation accuracy of ± 3⁄4˚C can be obtained over a
(Note 1) and T is the ambient temperature in degrees Kelvin. 0˚C–70˚C range. Performance with an iron-constantan ther-
Then, adjust R2 so that V1−V2 is equal to the thermocouple mocouple is almost as good. To keep the error small for the
output voltage at the known ambient temperature. less linear S and T type thermocouples, the ambient tem-
To achieve maximum performance from this circuit the resis- perature must be kept within a more limited range, such as
tors must be carefully chosen. R3 through R6 should be 15˚C to 50˚C. Of course, more accurate compensation over
precision wirewounds, Vishay bulk metal or precision metal a narrower temperature range can be obtained with any
film types with a 1% tolerance and a temperature coefficient thermocouple type by the proper adjustment of voltage TC
of ± 5 ppm/˚C or better. In addition to having a low TCR, and offset.
these resistors exhibit low thermal emf when the leads are at Standard metal-film resistors cost substantially less than
different temperatures, ranging from 3 µV/˚C for the TRW precision types and may be substituted with a reduction in
MAR to only 0.3 µV/˚C for the Vishay types. This is espe- accuracy or temperature range. Using 50 ppm/˚C resistors,
cially important when using S or R type thermocouples that the circuit can achieve 1⁄2˚C error over a 10˚C range. Switch-
output only 6 µV/˚C. R7 should have a temperature coeffi- ing to 25 ppm resistors will halve this error. Tin oxide resis-
cient of ± 25 ppm/˚C or better and a 1% tolerance. Note that tors should be avoided since they generate a thermal emf of
the potentiometers are placed where their absolute resis- 20 µV for 1˚C temperature difference in lead temperature as
tance is not important so that their TCR is not critical. How- opposed to 2 µV/˚C for nichrome or 4.3 µV/˚C for cermet
ever, the trim pots should be of a stable cermet type. While types. Resistor networks exhibit good tracking, with
multi-turn pots are usually considered to have the best reso- 50 ppm/˚C obtainable for thick film and 5 ppm/˚C for thin film.
lution, many modern single-turn pots are just as easy to set In order to obtain the large resistor ratios needed, one can
to within ± 0.1% of the desired value as the multi-turn pots. use series and parallel connections of resistors on one or
Also single-turn pots usually have superior stability of set- more substrates.
ting, versus shock or vibration. Thus, good single-turn cer- Note 1: See Appendix A for calculation of Seebeck coefficient.
met pots (such as Allen Bradley type E, Weston series 840,
or CTS series 360) should be considered as good candi-
dates for high-resolution trim applications, competing with
the more obvious (but slightly more expensive) multi-turn
trim pots such as Allen Bradley type RT or MT, Weston type
850, or similar.

www.national.com 2
 AN-225
Circuit Description (Continued)

00747103

Thermocouple Seebeck R4 R6 Thermocouple Seebeck R4 R6

Type Coefficient (Ω) (Ω) Type Coefficient (Ω) (Ω)
(µV/˚C) (µV/˚C)
J 52.3 1050 385 S 6.4 128 46.3
T 42.8 856 315
K 40.8 816 300

*R3 thru R6 are 1%, 5 ppm/˚C. (10 ppm/˚C tracking.)

†
R7 is 1%, 25 ppm/˚C.

FIGURE 3. Thermocouple Cold-Junction Compensation Using Single Power Supply

A circuit for use with grounded thermocouples is shown in follows: first short out the LM329B, the LM335 and the
Figure 4. If dual supplies are available, this circuit is prefer- thermocouple. Measure the output voltage (equal to the
able to that of Figure 3 since it achieves similar performance input offset voltage times the voltage gain). Then apply a 50
with fewer low TC resistors. To trim, short out the LM329B mV input voltage and adjust the GAIN ADJUST pot until the
and adjust R5 so that Vo = < 5˚C T, where < 5˚C is the output voltage is 12.25V above the previously measured
Seebeck coefficient of the thermocouple and T is the abso- value. Next, short out the thermocouple again and remove
lute temperature. Remove the short and adjust R4 so that Vo the short across the LM335. Adjust the TC ADJUST pot so
equals the thermocouple output voltage at ambient. A good that the output voltage equals 10 mV/˚K times the absolute
grounding system is essential here, for any ground differen- temperature. Finally, remove the short across the LM329B
tial will appear in series with the thermocouple output. and adjust the ZERO ADJUST pot so that the output voltage
An electronic thermometer with a 10 mV/˚C output from 0˚C equals 10 mV/˚C times the ambient temperature in ˚C.
to 1300˚C is seen in Figure 5. The trimming procedure is as

3 www.national.com
AN-225
Circuit Description (Continued)

00747104

Thermocouple R1 Seebeck Thermocouple R1 Seebeck

Type (Ω) Coefficient Type (Ω) Coefficient
(µV/˚C) (µV/˚C)
J 377 52.3 S 45.8 6.4
T 308 42.8
K 293 40.8

*R2 and R3 are 1%, 10 ppm/˚C. (20 ppm/˚C tracking.)

†
R1 and R6 are 1%, 50 ppm/˚C.
FIGURE 4. Cold-Junction Compensation for Grounded Thermocouple

www.national.com 4
 AN-225
Circuit Description (Continued)

00747105
All fixed resistors ± 1%, 25 ppm/˚C unless otherwise indicated.
A1 should be a low drift type such as LM308A or LH0044C. See text.

FIGURE 5. Centigrade Thermometer with Cold-Junction Compensation

The error over a 0˚C to 1300˚C range due to thermocouple Construction Hints
nonlinearity is only 2.5% maximum. Table 1 shows the error
due to thermocouple nonlinearity as a function of tempera- The LM335 must be held isothermal with the thermocouple
ture. This error is under 1˚C for 0˚C to 300˚C but is as high reference junction for proper compensation. Either of the
as 17˚C over the entire range. This may be corrected with a techniques of Figure 6a or Figure 6b may be used.
nonlinear shaping network. If the output is digitized, correc- Hermetic ICs use Kovar leads which output 35 µV/˚C refer-
tion factors can be stored in a ROM and added in via enced to copper. In the circuit of Figure 5, the low level
hardware or software. thermocouple output is connected directly to the op amp
The major cause of temperature drift will be the input offset input. To avoid this from causing a problem, both input leads
voltage drift of the op amp. The LM308A has a specified of the op amp must be maintained at the same temperature.
maximum offset voltage drift of 5 µV/˚C which will result in a This is easily achieved by terminating both leads to identi-
1˚C error for every 8˚C change in ambient. Substitution of an cally sized copper pads and keeping them away from ther-
LH0044C with its 1 µV/˚C maximum offset voltage drift will mal gradients caused by components that generate signifi-
reduce this error to 1˚C per 40˚C. If desired, this temperature cant heat.
drift can be trimmed out with only one temperature cycle by
following the procedure detailed in Appendix B.

5 www.national.com
AN-225
Construction Hints (Continued)

00747106
a

00747107
*Has no effect on measurement.
b

FIGURE 6. Methods for Sensing Temperature of Reference Junction

TABLE 1. Nonlinearity Error of Thermometer Using Type K Thermocouple (Scale Factor 25.47˚C/µV)
˚C Error (˚C) ˚C Error (˚C)
10 −0.3 200 −0.1
20 −0.4 210 −0.2
30 −0.4 220 −0.4
40 −0.4 240 −0.6
50 −0.3 260 −0.5
60 −0.2 280 −0.4
70 0 300 −0.1
80 0.2 350 1.2
90 0.4 400 2.8
100 0.6 500 7.1
110 0.8 600 11.8
120 0.9 700 15.7
130 0.9 800 17.6
140 0.9 900 17.1
150 0.8 1000 14.0
160 0.7 1100 8.3
170 0.5 1200 −0.3
180 0.3 1300 −13
190 0.1

www.national.com 6
 AN-225
Construction Hints (Continued) the thermocouple function with a straight line and determine
< 5˚C from the line’s slope for the temperature range of
Before trimming, all components should be stabilized. A interest. On a graph, the error of the line approximation is
24-hour bake at 85˚C is usually sufficient. Care should be easily visible as the vertical distance between the line and
taken when trimming to maintain the temperature of the the nonlinear function. Thermocouple nonlinearity is not so
LM335 constant, as body heat nearby can introduce signifi- gross, so that a numerical error calculation is better than the
cant errors. One should either keep the circuit in moving air graphical approach.
or house it in a box, leaving holes for the trimpots. Most thermocouple functions have positive curvature, so
that a linear approximation with minimum mean-square error
Conclusion will intersect the function at two points. As a first cut, one can
Two circuits using the LM335 for thermocouple cold-junction pick these points at the 1⁄3 and 2⁄3 points across the ambient
compensation have been described. With a single room temperature range. Then calculate the difference between
temperature calibration, these circuits are accurate to ± 3⁄4˚C the linear approximation and the thermocouple. (Note 2) This
over a 0˚C to 70˚C temperature range using J or K type error will usually then be a maximum at the midpoint and
thermocouples. In addition, a thermocouple amplifier using endpoints of the temperature range. If the error becomes too
an LM335 for cold-junction compensation has been de- large at either temperature extreme, one can modify the
scribed for which worst case error can be as low as 1˚C per slope or the intercept of the line. Once the linear approxima-
40˚C change in ambient. tion is found that minimizes error over the temperature
range, use its slope as the Seebeck coefficient value when
designing a cold-junction compensator.
Appendix A An example of this procedure for a type S thermocouple is
Determination of Seebeck shown in Table 2. Note that picking the two intercepts (zero
Coefficient error points) close together results in less error over a nar-
rower temperature range.
Because of the nonlinear relation of output voltage vs. tem- Note 2: A collection of thermocouple tables useful for this purpose is found
perature for a thermocouple, there is no unique value of its in the Omega Temperature Measurement Handbook published by Omega
Seebeck coefficient < 5˚C. Instead, one must approximate Engineering, Stamford, Connecticut.

TABLE 2. Linear Approximations to Type S Thermocouple

Approximation #1 Approximation #2
Centigrade Type S Zero Error Zero Error
Temperature Thermocouple at 25˚C and 60˚C at 30˚C and 50˚C
Output (µV) Linear Error Linear Error
Approx. µV ˚C Approx. µV ˚C
0˚ 0 −17 −17 −2.7˚ −16 −16 −2.8˚
5˚ 27 15 −12 −1.9˚ 16 −11 −1.7˚
10˚ 55 46 −9 −1.4˚ 47 −8 −1.3˚
15˚ 84 78 −6 −0.9˚ 78 −6 −0.9˚
20˚ 113 110 −3 −0.5˚ 110 −3 −0.5˚
25˚ 142 142 0 0 142 −1 −0.2˚
30˚ 173 174 1 0.2˚ 173 0 0
35˚ 203 206 3 0.5˚ 204 1 0.2˚
40˚ 235 238 3 0.5˚ 236 1 0.2˚
45˚ 266 270 4 0.6˚ 268 2 0.3˚
50˚ 299 301 2 0.3˚ 299 0 0
55˚ 331 333 2 0.3˚ 330 −1 −0.2˚
60˚ 365 365 0 0 362 −3 −0.5˚
65˚ 398 397 −1 −0.2˚ 394 −4 −0.6˚
70˚ 432 429 −3 −0.5˚ 425 −7 −1.1˚
< 5˚C = 6.4 µV/˚C < 5˚C = 6.3 µV/˚C
0.6˚C error for 0.3˚C error for
20˚C < T < 70˚C 25˚C < T < 50˚C

Note 3: Error is the difference between linear approximation and actual thermocouple output in µV. To convert error to ˚C, divide by Seebeck coefficient.

Appendix B Technique for

Trimming Out Offset Drift
Short out the thermocouple input and measure the circuit
output voltage at 25˚C and at 70˚C. Calculate the output
voltage temperature coefficient, β as shown.

7 www.national.com
IC Temperature Sensor Provides Thermocouple Cold-Junction Compensation
Appendix B Technique for LM329B) and op amp offset drift. Performance will be limited
only by TC nonlinearities and measurement accuracy.
Trimming Out Offset Drift (Continued)

Next, short out the LM329B and adjust the TC ADJ pot so References
that VOUT = (20 mV/˚K − β) x 298˚K at 25˚C. Now remove the R. C. Dobkin, “Low Drift Amplifiers,” National Semiconductor
short across the LM329B and adjust the ZERO ADJUST pot LB-22, June 1973.
so that VOUT = 246 mV at 25˚C (246 times the 25˚C output
Carl T. Nelson, “Super Matched Bipolar Transistor Pair Sets
of an ice-point-referenced thermocouple).
New Standards for Drift and Noise,” National Semiconductor
This procedure compensates for all sources of drift, includ- AN-222, February 1979.
ing resistor TC, reference drift ( ± 20 ppm/˚C maximum for the

LIFE SUPPORT POLICY

NATIONAL’S PRODUCTS ARE NOT AUTHORIZED FOR USE AS CRITICAL COMPONENTS IN LIFE SUPPORT
DEVICES OR SYSTEMS WITHOUT THE EXPRESS WRITTEN APPROVAL OF THE PRESIDENT AND GENERAL
COUNSEL OF NATIONAL SEMICONDUCTOR CORPORATION. As used herein:
1. Life support devices or systems are devices or 2. A critical component is any component of a life
systems which, (a) are intended for surgical implant support device or system whose failure to perform
into the body, or (b) support or sustain life, and can be reasonably expected to cause the failure of
whose failure to perform when properly used in the life support device or system, or to affect its
accordance with instructions for use provided in the safety or effectiveness.
labeling, can be reasonably expected to result in a
significant injury to the user.
National Semiconductor National Semiconductor National Semiconductor National Semiconductor
Corporation Europe Asia Pacific Customer Japan Ltd.
AN-225

Americas Fax: +49 (0) 180-530 85 86 Response Group Tel: 81-3-5639-7560

Email: support@nsc.com Email: europe.support@nsc.com Tel: 65-2544466 Fax: 81-3-5639-7507
Deutsch Tel: +49 (0) 69 9508 6208 Fax: 65-2504466
English Tel: +44 (0) 870 24 0 2171 Email: ap.support@nsc.com
www.national.com Français Tel: +33 (0) 1 41 91 8790

National does not assume any responsibility for use of any circuitry described, no circuit patent licenses are implied and National reserves the right at any time without notice to change said circuitry and specifications.