[time-nuts] temperature sensor

[Neville Michie]

Hi,
as a bit of a temperature nut, here are some observations.

Diodes work as temperature sensors, but better is the trans-diode,
a bipolar transistor with collector connected to base. Sensitivity
about 2.2 mV/K, I did not use them much in spite of their linearity
and low cost.

PT100 is only useful with a four terminal measurement with an HP3468A instrument
or equivalent measuring 4 terminal ohms. It is stable to one milli-Kelvin and I use it for 
measuring the temperature of thermostatic water baths when calibrating thermometers.
The sensor has self heating, and is poor in placement on small components, with 
high thermal resistance to measured item and high heat conduction along leads.
Without the HP measuring instrument the PT100 relies on the resistors it is compared with.
Although the PT100 specs are good, resistors that can match those specs are very expensive,
so to set up a measuring bridge requires a lot of expensive technology in resistors 
and stable amplifiers. If you wont go that far you are better off with thermistors.

Thermocouples only compare temperatures, so cold reference is a problem 
as are cold junction compensators. With only 40 microvolts/Kelvin they need good 
to very good amplifiers. They are small and are not compromised by using 
microscopically fine wire. They can measure tiny items with very little heat 
transmitted along the wires. they have no self heating.I use them in a Pile 
of six elements to measure Relative Humidity to a precision of 0.1% RH in 
an appropriate portable instrument. One logging instrument I have built 
uses three thermocouples with ICL7650 amplifiers and a AD590 common cold junction
sensor. A four channel logger produces data that is reduced by spreadsheet
to three temperatures. Copper /constantin is a good pair, sensitive (40uV/K), reasonably linear
and the copper solves many switching and amplifier circuit problems.

Thermistors are useful. They are available in tiny packages, sealed in glass.
With high impedance their self heating can be negligibly small. As has been 
pointed out they comply well to a general calibration curve.
If Paralleled with an equal value resistor, or connected in an equal 
value resistance bridge, they are linear over about 20C range.
They are very sensitive (4%/K), and so need no special amplifiers. They prove quite stable.

Calibration is done with a known PT100 sensor and a HP3468A in a thermostatic bath
(Colora or Haake). Tested devices a placed in a copper tube immersed 10cm (4 inches)
at least. 

Icepoint cell, in a Dewar flask (Thermos) with shaved ice gives a temperature at zero C
to 0.01C. Test thermometer inserted 15 cm.

Hypsometer, a boiling point cell has special splash and radiation shields, gives 
measurements stable to 1mK, but after all the corrections needed to a Fortin barometer
for temperature etc, relies on knowing local gravity to some accuracy. If the local 
geoid is assumed with usual corrections the ice-point is still more reliable as an absolute standard.

Conduction along leads is a major consideration, even with fine (0.002" ) copper leads,
a typical installation needs 20mm immersion. It is a logarithmic function, the immersion error,
I allow about 50mm to be sure with thermocouple sensors into a 3mm hole drilled into aluminium plate 
edge wise to measure the plate temperature. You only have to do a few experiments to find 
that this is necessary.

cheers,
Neville Michie