Low-cost, high-precision
By Allan Linton-Smith
Many digital thermometers have readouts with a 0.1°C
resolution but rarely are they accurate to within ±0.1°C.
Despite their claims, some can be several degrees out,
giving a false sense of accuracy. This simple, low-cost
thermometer checker will tell you just how accurate your
thermometer is. In some cases, you may even be able to
adjust the thermometer to be more accurate.
T
here are many reasons why you might need
an accurate thermometer. Checking to see if someone
(especially a child) has a fever is an everyday use
case. This requires pretty good accuracy, as the difference
between a normal-but-elevated temperature (as can happen
when someone has been exercising or crying for example)
and a fever is just fractions of a degree.
Or maybe you’re a keen chef, and you want to use processes like tempering chocolate, where you need to heat the
chocolate to a temperature within a fairly small window,
eg, 31-33°C.
A 1°C error could mean that you think you’re in the
window, but you aren’t, and the batch could be ruined.
Whatever the reason for using it, if you have a thermometer that will read out to within 0.1°C, you want to know
if it’s at least ‘in the ballpark’ before you trust its display
fully. This simple device allows you to do that.
In some industries such as food manufacture, storage and
distribution, temperatures are critical. This is especially
true when food poisoning is a potential problem. So in
these cases, it is essential to check that your thermometers
are accurate. A device like this is therefore invaluable.
This design is based on the LM35CAZ IC, a temperature
sensor that has been available for some time now. But
it has really come down in price lately. If managed
correctly, it can be expected to give readings
within ±0.2°C at 25°C.
It works over a –40°C to +110°C range,
but its accuracy is not as good when
reading temperatures further away
from room temperature.
It’s worth building this yourself because other devices with
precise temperature readings, eg,
28
±0.1°C, are not commonly available and are very expensive. For example, the Fluke 9142 and 9143 are excellent
calibrating instruments with a display accuracy of ±0.2°C
over their full range, but we recently spotted a used one
for sale for over £3,000! And new? Don’t even ask!
Some say that glass thermometers are very accurate.
Usually, their accuracy is accepted as ±0.5 divisions,
which typically translates to ±1°F or ±0.5°C, but they are
becoming quite rare.
And they are still susceptible to reading errors, some of
which are described in the side panel.
When designing this device, we found that there are a
few temperature sensor ICs that are even more accurate,
such as the LMT70, but we decided against using this (for
now) for a few reasons.
One is that it only comes in a tiny
SMD package (0.94 × 0.94mm)
which is hard to work with.
Another is that its output voltage is non-linear and requires
a lookup table or polynomial
curve-fitting to convert to a temperature reading.
You can buy them pre-soldered to a
module, but these test boards cost more
than £30, which is not worth it for only slightly
better accuracy.
To give you an idea of how hard it is to measure temperature precisely with a digital sensor, here is a passage
from the LMT70 data sheet:
‘Although the LMT70 package has a protective backside coating that reduces the amount of light exposure on
the die, unless it is fully shielded, ambient light will still
reach the active region of the device from the side of the
Practical Electronics | November | 2021
Reproduced by arrangement with
SILICON CHIP magazine 2021.
www.siliconchip.com.au
The three DMMs are reading the outputs of the LM35s but we have also inserted the probes of five cheap digital
thermometers and two lab-grade glass thermometers (see opposite) into the device. The cheap thermometers have a
0.5°C spread, quite a bit larger than the 0.2°C difference between the LM35s.
package. Depending on the amount of light exposure in a
given application, an increase in temperature error should
be expected.’
‘In circuit board tests under ambient light conditions, a
typical increase in error may not be observed and is dependent on the angle that the light approaches the package. The
LMT70 is most sensitive to IR radiation. Best practice should
include end-product packaging that provides shielding
from possible light sources during operation.’
Circuit details
The LM35CAZ is a precision integrated-circuit temperature
sensor with an output voltage linearly proportional to the
temperature in degrees Celsius.
It requires no external calibration or trimming. It is low
in cost, can operate on a wide variety of single supply
voltages and has low self-heating.
There’s little to the circuit besides three of these devices,
and a battery to power them, as shown in Fig.1. IC1-IC3
can run from a wide supply range of 4-20V, so they are very
well suited to be powered from a 9V battery.
Fig.1: the
circuit
couldn’t be
much simpler;
it’s just the
three LM35s
with a shared
100nF bypass
capacitor,
power switch
S1 and a 9V
battery for
power.
V+
IC1
LM35CAZ OUT
GND
LM35 CAZ
METER+
METER–
CON1
GND
V+
OUT
V+
IC2 OUT
LM35CAZ
GND
METER+
100nF
METER–
CON2
ON/OFF
S1
V+
IC3 OUT
LM35CAZ
GND
SC
METER+
METER–
CON3
2020
Practical Electronics | November | 2021
BAT1
9V
The output of each device can be measured by a multimeter connected across one of CON1-CON3, set to its 1V
range or thereabouts (ideally, with 1mV resolution). IC1IC3 have a nominal 0V output at 0°C, rising by 10mV/°C.
So, for example, in the photo above showing a 155mV
reading on the multimeter display means that the temperature is 15.5±0.2°C.
Note that the LM35CA is only guaranteed to be within
±0.5°C at 25°C, but in reality, a typical sample of the device
is within ±0.2°C from around –25°C to 50°C.
The reason for using three different devices is threefold.
First, it increases your confidence that you have an accurate
reading when they are all giving similar results. Second,
it also lets you get an idea of which sensors read a little
Parts list – Thermometer Calibrator
1 diecast aluminium box, approx. 115 x 90 x 55m
[eg, Jaycar Cat HB5042]
3 LM35CAZ temperature sensors [eg Mouser LM35CAZ/NOPB,
Digi-key LM35CAZ/NOPB-ND, RS Cat 5335878]
3 voltmeters [eg, Jaycar Cat QM1500]
3 red banana plug to banana plug leads
3 black banana plug to banana plug leads
3 black chassis-mounting banana sockets
3 red chassis-mounting banana sockets
1 chassis-mounting 9V battery holder
1 9V battery clip with flying leads
1 9V battery (alkaline recommended)]
1 100nF ceramic, MKT or greencap capacitor
1 SPST toggle switch
1 small piece of protoboard
1 3mm ID solder lug
1 M3 x 10mm machine screw and nut
1 adhesive TO-3P or TO-247 insulating washer
1 small tube adhesive heatsink compound [eg Jaycar NM2014]
Various lengths of ribbon cable or hookup wire
29
Accurate temperature measurement is not easy
Making precise temperature readings (say
to within ±0.1°C) is difficult. Devices to do
this are not commonly available and are
very expensive.
For example, if your backyard weather
thermometer is showing 40°C, it could
actually be 38°C or 42°C. It could even
be much higher or lower than this if your
thermometer is poorly sited (eg, near an air
conditioner or road) or in a poorly designed
enclosure or bad position, which allows its
reading to be affected by direct sunlight.
Assuming your sensor is linear, you can
calibrate it using a stirred ice bath (to determine its reading at 0°C) and vigorously
boiling pure water (100°C), both at sea
level. But unless you do this correctly, your
readings could still be out considerably.
For example, at around 300m elevation,
the boiling point of water is about 98.9°C.
Normal day-to-day atmospheric pressure
variations can have a small effect on the
boiling point, too.
Any salt in the water or ice can have a
dramatic effect on both the boiling and freezing points. According to the CRC Handbook
of Chemistry and Physics, 2.92% sodium
chloride in solution reduces the melting point
of ice by 0.19°C and increases the boiling
point by 0.05°C.
Calibrating Theremometers, a practical
thermometer calibration method from NFSMI
can be downloaded from the June 2021 page
of the PE website.
Even if your calibration method is flawless,
you also need to know that the sensor response is perfectly linear to have confidence
in readings between the two extremes.
IC-based temperature sensors like the
LM35 do suffer from some level of non-
higher or lower than the others. And third, it also lets you
check that the case is at an even temperature before making your readings.
In the photo overleaf, with all three giving readings
within 0.2°C of each other, note how the cheap digital
thermometers with their probes inserted into the same
metal case, and presumably reading the same temperature,
are all reading high (by about 0.5-1°C) and also have a
considerably greater spread than the LM35CA devices.
You must use the LM35CA version for accuracy, as the
LM35/LM35A/LM35C/LM35D cannot achieve the same
accuracy. (The ‘Z’ suffix indicates a TO-92 package).
Note though that the LM35CA is limited to measuring in
the range of –40°C to +110°C, while the less accurate LM35
and LM35A versions can measure from –55°C to +150°C.
The three multimeters we’ve used here are low-cost
devices that you can get for a few dollars from Jaycar, and
we’ve found that they are very accurate. They have a voltage
linearity, even though they are designed
to be as linear as possible.
And it isn’t just electronic sensors that
suffer from accuracy problems, either. As
one meteorologist pointed out, even the
meniscus (bulge in the top of a column
of liquid in a tube) in a mercury or alcohol
thermometer can lead to significant inaccuracies in the readings. He also mentions:
‘… mercury freezes at -38.8°C. It
becomes increasingly less malleable as it
approaches that temperature and makes
low temperatures with mercury thermometers of no value. The 18th century
observers of the Hudson’s Bay Company
using thermometers provided by the Royal
Society were unaware of the problem ...’
Because of problems like this, interpreting historical air and sea temperature data
is quite tricky!
accuracy rating of ±0.5%, which equates to an additional
error of just ±0.1°C in the temperature readings.
To demonstrate the accuracy of the LM35CAs, we also
have two laboratory-grade analogue thermometers measuring the same temperature. As shown in the separate
photo (frst page), they are both reading just under 16°C,
just slightly higher than the figures shown on the DMMs.
Do not buy cheap LM35 sensors online if you are expecting accuracy, or even for them to function. We also
purchased several LM35Ds cheaply on the internet to
compare, but NONE of them worked at all! So it is essential to obtain them from a reputable supplier (eg, the ones
mentioned in the parts list).
Construction
We recommend that you build this into a diecast aluminium
box. This will not only provide some shielding, it allows
you to check glass thermometers and to help maintain
While it might seem
like overkill, placing
this project in a
diecast case has
several benefits – it’s
shielded, of course,
and the thick
aluminium provides
some thermal
inertia. Placing the
LM35CAZs inside the
box also means they
will be less affected
by external variants.
Of course, a smaller
diecast case could be
used, providing the
various components
will fit.
30
Practical Electronics | November | 2021
Thermocouple
LM35CAZ
+200 to +1750°C -40 to +150°C
±0.5 to ±5°C
±0.2°C at 25°C
Variable
0.2°C/year
Non-linear
Linear
Self-powered
4-20V DC
0.1-10s
2-15s
Susceptible
Susceptible
High
Moderate
2.5
TEMPERATURE ERROR (°C)
Thermistor
RTD
Range
-100 to +325°C -200 to +650°C
Accuracy
±0.05 to ±1.5°C ±0.1 to ±1°C
Stability <at> 100°C
0.2°C/year
0.05°C/year
Linearity
Exponential
Fairly linear
Power
Small current
Small current
Response
0.1-10s
1-50s
Interference
Rarely
Rarely
Cost
Low to moderate
High
2.0
LM35D
1.5
LM35C
1.0
LM35CA
0.5
TYPICAL
0.0
±0.5
LM35CA
±1.0
LM35C
±1.5
±2.0
Table 1 shows the typical parameters of various temperature sensors, while the
graphs at right show the errors in the different iterations of the LM35.
±2.5
±75
±25
25
75
125
175
TEMPERATURE (°C)
a uniform and stable temperature, without any thermal
gradients. The sensors have very little self-heating, but it
is still present; the large thermal mass of the case helps to
mitigate this.
The LM35s also detect temperature variations through
their pigtails. If these are exposed to small amounts of heat
variations, such as human breath or wind, it can disturb
the measurements and give false readings. By placing the
ICs inside a metal box, we can eliminate these errors.
Solder the three LM35s to a small piece of protoboard,
veroboard or similar. Join their V+ and GND leads together,
and solder the 100nF capacitor across these rails. Also connect pairs of wires to the GND and OUT terminals of each
device, plus one pair of wires between the V+ and GND rails.
Ideally, the pairs of wires should be figure-8 cable (eg,
stripped from ribbon cable). If you are using individual
wires, it’s best to twist them together so that any interference is mostly cancelled out between the two conductors.
Now glue the three TO-92 plastic packages to the inside
of the diecast box using thermally conductive adhesive. We
used Jaycar NM2014 adhesive thermal paste.
Drill holes in the case for the power on/off switch and
9V battery holder, plus holes for the three pairs of banana
sockets in the lid. Also drill a 3mm hole for the chassis
grounding screw, near the battery holder, and one or two
extra holes in the lid for analogue thermometer calibration, if desired.
Deburr all the holes and mount these parts. Then solder
the pairs of wires from the LM35 GND and OUT terminals
to the banana sockets, with the OUT terminals going to
the red sockets.
The remaining pair of wires then goes to the switch (V+)
and case (GND). Solder the other switch terminal to the red
lead from the 9V battery, so that V+ is connected to the battery when the switch is in the on position (usually down).
Join the remaining GND wire to the black wire of the 9V
battery to the solder lug and attach it to the inside of the case
using an M3 machine screw and nut (not shown below).
Stick the insulating washer on the inside of the case directly below the analogue thermometer insertion holes in
V+
Using it
Avoid using this device in a windy environment or one
with rapidly changing temperatures, such as near a window
that’s exposed to full sun where clouds may pass by. Ideally,
it should be used indoors with still air in an environment
with a stable temperature.
Switch it on and allow everything to stabilise for around
20 minutes before using it for best results.
Your best bet since MAPLIN
Chock-a-Block with Stock
Visit: www.cricklewoodelectronics.com
O r phone our f riendly kn owledgeable staf f on 0 2 0 8 4 5 2 0 1 6 1
Components • Audio • Video • Connectors • Cables
Arduino • Test Equipment etc, etc
K
IC1 OUT
LM35
METER+
METER–
A
D1
1N4 148
18k
SC
K
Fig.2: by adding three
components to each
LM35, you can measure
temperatures below 0°C.
the lid. This will provide the thermometers with a bit of a
‘cushion’ so that they do not break when inserted.
Now connect the battery clip to the battery, slot it into
its holder and switch on the power. Use a red and black
pair of banana plug leads to connect one of the DMMs to
one of the pairs of binding posts, and check that you get a
reading that’s fairly close to ambient temperature.
For example, if it’s around 25°C where you are, you
should get a reading around 250mV. Verify that all the
outputs are similar values.
1N4148
A
V+
GND
C009
Figure 9. Accuracy vs Temperature (Ensured)
2020
A
D2
1N4 148
LM35
K
GND
V+
OUT
Practical Electronics | November | 2021
V is it o u r S h o p , C a ll o r B u y o n lin e a t:
w w w .c r ic k le w o o d e le c tr o n ic s .c o m
0 2 0 8 4 5 2 0 1 6 1
V is it o u r s h o p a t:
4 0 -4 2 C r ic k le w o o d B r o a d w a y
L o n d o n N W 2 3 E T
31
