[time-nuts] Temperature sensors and quartz crystals (was: HP5061B Versus HP5071 Cesium Line Frequencies)

Attila Kinali attila at kinali.ch
Sun Jun 4 19:59:23 EDT 2017


This discussion is kind of getting heated.
Let's put some facts in, to steer it away from
opinion based discussion.

On Sun, 4 Jun 2017 08:44:33 -0700
"Donald E. Pauly" <trojancowboy at gmail.com> wrote:

> I stand by my remark that thermistors have been obsolete for over 40
> years.  The only exception that I know of is cesium beam tubes that
> must withstand a 350° C bakeout.  Thermistors are unstable and
> manufactured with a witches brew straight out of MacBeth.  Their
> output voltages are tiny and are they inconvenient to use at different
> temperatures.

If you really mean thermistors, and not, as Bob suggested thermocouples,
then I have to disagree. The most stable temperature sensors are
platinum wire sensors. The standards class PRT's are the gold standard
when it comes to temperature measurement, for a quite wide range
(-260°C to +960°C) and are considered very stable. They offer (absolute)
accuracies in the order of 10mK in the temperature range below 400°C.
Even industrial grade PRT sensors give you an absolute accuracy better
than 0.1K up to 200-300°C. The "cheap" PT100 are more of the order of 1-10°C
accuracy... all numbers just using a two-point calibration.

For more information on this see [1] chapter 6 and [2] for industrial sensors.

NTC sensors have a higher variablity of their parameters in production
and are usually specified in % of temperature relative to their reference
point, which is usually 25°C. Typical values are 0.1% to 5%. Additionally
there is a deviation from the reference point, specified in °C, which
is usually in the order of 0.1°C to 1°C.

The NTC sensors are less accurate than PT sensors, but offer the advantage
of higher resistance (thus lower self-heating), higher slope (thus better
precision). Biggest disadvantage is their non-linear curve. Their price
is also a fraction of PT sensors and due to that you can have them in
many different forms, from the 0201 SMD resistor, to a large stainless
steal pipe that goes into a chemical tank. NTCs are the workhorse in
todays temperature measurement and control designs. 

The next category are band-gap sensors like the AD590. Their biggest
advantage is that their 0 point is fix at 0K (and very accurately so).
Ie they can be used with single point calibration and achieve 1°C accuracy
this way. Their biggest drawback their large thermal mass and large
insulating case, because they are basically an standard, analog IC.
Ie their main use is in devices where there is a lot of convection and
slow temperature change. Due to their simple and and quite linear
characteristics, they are often used in purely analog temperature
control circuits, or where a linearization is not feasible.
But only if price isn't an issue (they cost 10-1000 times as
much as an PTC). Their biggest disadvantage, beside their slow
thermal raction time, is their large noise uncorrelated to the
supply voltage, and thus cannot be compensated by ratiometric measurement.
They are also more suceptible to mechanical stress than NTC's and PT's,
due to their construction. Similar to voltage references (which they
actually are), their aging is quite substantial and cannot be neglected
in precision application.
With a 3 point calibration, better than 0.5°C accuracy can be achieved
(modulo aging) within their operating temperature range, which is
rather limited, compared to the other sensor types.

I don't know enough about thermocouples to say much about them, beside
that they are cumbersome to work with (e.g. the cold contact) and
produce a low voltage (several µV) output with quite high impedance,
which makes the analog electronics difficult to design as well.

With todays electronics, the easiest sensors to work with are NTC and
PT100/PT1000 as most high resolution delta-sigma ADCs have direct support
for 3 and/or 4 wire measurement of those, including compensation for
reference voltage/current variation. Using a uC as control element
also opens up the possibility to linearize the curve of NTCs without
loss of accuracy. Usually measurement precision, with a state-of-the-art
circuit, is limited by noise coupling into the leads of the sensor
and noise in and around the ADC. (see [3-5])

> Where did you get the idea to use a 1 k load for an AD590?  

Jim was refering to a circuit _he_ used in a satellite. Not to your circuit.

> The room temperature coefficient of an AT crystal is -cd 100 ppb per
> reference cut angle in minutes.  (-600 ppb/C° for standard crystal)
> The practical limit in a crystal designed for room temperature is
> about 0.1' cut accuracy or ±10 ppb/C°.  If you have access to an
> atomic standard, you can use feed forward to get ±1 ppb/C°.  If the
> temperature can be held to ±0.001° C, this is ±1 part per trillion.
> This kind of accuracy has never been heard of.

It has been heard of. The 8607 was spec'ed to <2e-10 p-p deviation
over temperature range (-30°C to 60°C). Also, to hold the temperature
stable to 0.001K in a room temperature environment (let's say 10K variation),
you need a thermal gain of >10k. That's quite a bit and needs considerable
design effort. Most OCXO design's I am aware of are in the order of 100
(the DIL14 designs) to a few 1000 for single ovens, to a few 10k for
double ovens. The only exception is the E1938 which achieves >1M.
But that design is not for the faint hearted. I don't remember seeing
any number, but i would guess the 8607 has a thermal gain in the
order of 100k to 1M as well, considering it being a double oven in
a dewar flask.

Also, what do you mean by atomic standard and feed forward?
If you have an atomic standard you don't need to temperature
stabilize your quartz. You can just simply use a PLL to lock
it to your reference and achieve higher stability than any oven

>  Feed forward also
> allows you to incorporate the components of the oscillator into the
> thermal behavior.  It does no good to have a perfect crystal if the
> oscillator components drift.

Beyond tau=100s, the temperature and moisture sensitivity of the
electronics, combined with the aging of the electronics and the
crystal will be the limit of stability. Of course, this is under
the assumption that you achieved a thermal noise limited design
and thus the 1/f^a noise of the oscillator is negligible in the
time range considered.

			Attila Kinali

[1] "Traceable Temperatures - An Introduction to Temperature Measurement
and Calibration", 2nd edition, by Nicholas and White, 2001

[2] "Thin-film platinum resistance thermometer for use at low temperatures
and in high magnetic fields", Haruyama, Yoshizaki, 1986

[3] "Completely Integrated 4-Wire RTD Measurement System Using a Low Power,
Precision, 24-Bit, Sigma-Delta ADC", Analog Circuit Note CN-0381

[4] "Completely Integrated 3-Wire RTD Measurement System Using a Low Power,
Precision, 24-Bit, Sigma-Delta ADC", Analog Circuit Note CN-0383

[5] "2- 3- 4- Wire RDT (Pt100 to PT1000)Temperature Measurement"
Ti Presentation

You know, the very powerful and the very stupid have one thing in common.
They don't alters their views to fit the facts, they alter the facts to
fit the views, which can be uncomfortable if you happen to be one of the
facts that needs altering.  -- The Doctor

More information about the time-nuts mailing list