SRS sells (we have one) a 10 MHz brick oscillator that replaces an unavailable HP thing. It's an SC cut rock, so the jitter is very low, like 1 ps RMS at 1 second delay, thousands of times better than a cheap oscillator.
Their thermal loop design is amusing. It apparently oscillated for reasons they didn't understand, so they killed the gain and added a feedforward sensor from ambient.
Hey, I have this...
Frequency counters used to be a big deal.
-- John Larkin Highland Technology, Inc jlarkin at highlandtechnology dot com http://www.highlandtechnology.com Precision electronic instrumentation Picosecond-resolution Digital Delay and Pulse generators Custom laser drivers and controllers Photonics and fiberoptic TTL data links VME thermocouple, LVDT, synchro acquisition and simulation
than,
urethane.
euld
ch
s. A
AC-
ood
at
But they don't modulate the temperature-indicating signal in the first place, which rather takes away the magic in this application.
On the other hand, I've never seen anybody complain about "conductivity fluctuations" as a source of 1/f noise in micro-degree temperature controllers - knowing about Johnson noise at all is already fairly up-market for that literature
hBut would have 4mW of dissipation in the sensor, which would be 4 millidegrees of self-heating with a 1C/Watt thermal resistance to ambient. You'd need to keep the excitation voltage tolerably stable if you wanted to control to micro-degrees with such a sensor.
Sloman, A.W. "On microdegree thermostats", Journal of Physics E: Scientific Instruments, 11, 967-968 (1978) discusses this briefly.
It reduces the thermocouple worries, until you start dissipating milliwatts in the sensor.
kBut it does make an outer layer of foam a whole lot more effective.
It's one way to win.
A re-invention of a solution that got published in Journal of Scientific Instruments sometime in the 1970's. A tetrahedron with just four faces would be slightly less extravagant, if rather poorer at space-filling. A dodecahedron with l2 faces could be admirably symmetrical, if even more extravagant.
-- Bill Sloman, Nijmegen
That's where the platinum sensor wins over thermistor. With the right sensor you can get close thermal contact and a short time-constant between the sensor and the volume whose temperature is being controlled. That allows you a higher bandwidth in the control loop and a smaller offset for a give rate of temperature change.
-- Bill Sloman, Nijmegen
The outer-oven/inner-oven design gets rid of that, because the outer oven (even if its regulation is crude) doesn't pass any long duration of ramp in temperature.
P-I-D is simple enough, but tuning it gets time-consuming. Aren't there any buy-off-the-shelf ovens? For an ovenized oscillator, of course, you can just buy one (Bliley is one supplier).
To keep temperature uniformity inside the oven, a design with multiple points of control has some advantages, too; I've seen PTC thermistors advertised for self-regulating heaters, has anyone tried to dot a bevy of those onto a metal shell, to make a uniform ovenized layer?
than,
urethane.
Why? With the amp inside the controlled zone, where's the error coming from?
The thermal resistance to the copper could be quite a bit lower than that, I think. I have some thin film RTDs that are about 3mm x 10mm x
0.6 mm, so their thermal resistance should be around 0.4ish. They're cheap enough that you could wire 10 PT100s in series instead, and get 0.04. (I used a whole bunch of thermistors in series-parallel, because they're 10x more sensitive and 100x cheaper. I wrapped the flex around the foam, and then used an anodized aluminum thermal shield over that, held in contact with the metallization of the thermistors by the foam.As long as that stays still, I don't care about it. It's inside the box, so dissipation-induced gradients should be constant.
I didn't claim to have invented temperature control, Bill, or that any of this was particularly novel--it's just a counterexample to the microkelvins-requires-AC claim. Multiple heating zones were used in buildings since probably 2000 BC, so the extension of that to smaller objects isn't that hard to imagine. We were just discussing how best to do it in instruments. In my case, the geometry was dictated by having to fit down a 2-inch cased drill hole, so I wasn't going to be doing any tetrahedrons. It was six patches on a piece of flex circuit, arranged three round the circumference times two axially. The ends could have lots of extra insulation, so I just extended the shield.
The inner shield was a solid piece of copper with just one heating zone, but lots of heaters and thermistors. Since it was in nearly isothermal surroundings, any gradients caused by slightly uneven heating were pretty well constant.
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
Even when it's tuned optimally (for some definition of optimal) it's not going to be perfect. Industrial controllers self-tune the PID parameters, typically a bit on the conservative side IME.
Humm.. have not seen any.
10MHz Rubidium-vapor clock modules are pretty cheap surplus, not sure how expensive they are new. They're supposed to be stable to something like 2E-11 per day. 200 years ago you could have been burned at the stake for owning a clock that stable.SMT OCXOs are 2-3 orders of magnitude worse.
My gizmo was small enough to be supported entirely by the insulation, and used meandering-conductor flex cable, thermally grounded to the outer layer. The thermal conductivity of 4 mils x 1/4 oz Cu, 10 cm or so long, isn't very much. (Part of the reason for the series-parallel resistors for the heaters was so I could use 36V drive at very low current.)
It isn't really a second-order problem, because thermal diffusion takes over at some point, turning it into an exponential decay. In foam, diffusion is very slow, so the loop bandwidth doesn't have to be that large to keep the offset error small. So the slope of any ramp you're going to be putting through the inner insulation won't be very large. And in any case, some tiny amount of feedforward can reduce it a lot.
The device was designed to sit on a modestly temperature-controlled platform attached to a thermoacoustic fridge, to give us a 25 +- 10 C ambient, even down-hole. With very wide swings, you'd have to worry a bit more, of course, but another layer of temperature control would be a lot cheaper than a kilogram of nicely machined copper.
Temperature control is hard, and I sure wouldn't claim to have solved it entirely. This design is a good one for the purpose, though, and didn't require AC excitation. (I've got nothing particulary against AC, except for the circuit complexity and AC magnetic fields.)
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
Hi Spehro,
Actually there is a "thermal degaussing" technique used in the highest accuracy zener references. Don't see why it would not work for OXCO as well as a current source.
E.g. US 5369245.
Think timescale is minutes-hours rather than seconds there, but might be an idea.
-- John Devereux
n
ter than,
han urethane.
ner
d the
r would
at -
which
grees. A
to AC-
las good
s.
n that
de a
d wThink about it. The point about AC excitation of a resistance bridge is that the output is an AC signal with a well defined frequency, and what ought to be a tolerably well-defined phase. Anything else on the output is noise and can be filtered out.
Even the best chopper amplifier working on the output of a DC excited bridge doesn't have that advantage.
I don't really have to specify where the noise is going to come from
- if you are looking at uV signals, noise comes from all over. Even the local AM radio station can contribute if you aren't totally paranoid and very careful.
so
with
so
What "copper" do you have in mind?
And who makes your sensors? I want their contact details. Nothing that I was ever able to buy was anything like that good, but if you could get somebody to deposit a thin film of platinum on an extensive chunk of alumina substrate you might be able to do that well - as I fantasised about back in 1978.
Units? 0.4C/Watt?
Alumina has a thermal conductivity of 30W/m/K. Your substrate would have a conductivity 1.5W/K - 0.67K/Watt - from one side to the other, but the other side isn't going to be in direct contact with ambient.
This is what Farnell sells as a 0.39C/W heat sink.
It's rather larger than 3mm x10mm.
of
f sShould be? Nature is rarely that cooperative.
-- Bill Sloman, Nijmegen
n
.g.
that
My claim was that microkelvins
might well require AC excitation. Thermistors are ten times more sensitive, and you can get them with rather higher resistances, both of which help. Priel in 1978 got +/-3.5uK with a DC-excited thermistor.
To quote what I wrote in 1996 "Dratler(1974) described a system also using a DC-excited Wheatstone bridge, but dissipating 60u=16W in a Yellow Springs Instruments thermistor, for which he claimed =06+/-5u=16K stability. In his system all the components of the Wheatstone bridge and the bridge amplifier were mounted within the controlled temperature space, which presumably made his system less susceptible to adventitious thermocouple voltages."
And IIRR it too went down a bore-hole.
Dratler J 1974 Rev. Sci. Instrum. 45 pages 1435=9644
It's a long time since I've looked at the paper, but I don't think that he needed six separate zone thermostats to get that.
-- Bill Sloman, Nijmegen
Too true. The development phase of the thermostat design I published in 1996 relied on a self-tuning PID controller.
When we got hold of a copy of
Ziegler J and Nichols N 1942 Trans. ASME 64 759
and tuned it ourselves, it settled a lot faster.
-- Bill Sloman, Nijmegen
snip
they also sell 0.07C/W thermal pads
I think you are comparing pears and apples
-Lasse
than,
urethane.
the
would
which
microdegrees. A
AC-
good
that
Not so. An AC measurement has an inherent 2x bandwidth disadvantage compared with DC. In lots of cases, the 1/f noise and other junk more than makes up for this, so that AC measurements are a win. (My Ph.D. thesis was about heterodyne interferometry, so you know I'm a fan.) The point under discussion is, I take it, whether temperature control is still one of those cases.
True of all sensitive measurements. Probably not such a big worry 5000 feet down a drillhole, inside three concentric metal enclosures, but in some other cases, sure.
The metal thermal shield. Copper or aluminum, usually.
Not sure--I've had them in the drawer since about 1995. Probably distributed by Omega.
The point is that it's easy to make it very much lower than either of our single-sensor numbers, by using many cheap thermistors with semi-decent thermal contact to the copper-or-aluminum shield. My scheme uses a mechanical preload to push them into a fairly thin, soft, anodized aluminum shield with a quilted pattern on it, but there are lots of other methods. For instance, you can indium-solder to ordinary anodized aluminum, so the resistor/thermistor array could be soldered to the flex and then indium-soldered to the inside of the shield. I don't pretend to know enough to say which method is the absolute best, but there are lots of good ones.
Nature is very cooperative, obeying all sorts of mathematical laws. All she asks is that you do the right calculations, which in this case aren't that hard.
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
I was talking about the thermal resistance from the temperature sensing resistor to ambient, and Phil Hobbs responded talking about the thermal resistance from one side of his alumina substrate to the other - which happens to be one of the thermal resistances which add up - in series - on the way from the sensing resistance to the "ambient" which is supposed to be being controlled.
Comparing the thermal resistance from one side of an alumina substrate to the other to the total resistance from sensor to ambient really is comparing apples and pears, so I sent him up with an even more ridiculous comparison.
I'm glad that you noticed that it was ridiculous. I'd be even happier if you'd realised what I was ridiculing.
-- Bill Sloman, Nijmegen
pads
Ridicule and depression seem to encompass your skill set lately. Nice paring.
-- John Larkin Highland Technology, Inc jlarkin at highlandtechnology dot com http://www.highlandtechnology.com Precision electronic instrumentation Picosecond-resolution Digital Delay and Pulse generators Custom laser drivers and controllers Photonics and fiberoptic TTL data links VME thermocouple, LVDT, synchro acquisition and simulation
ark>
better than,
er than urethane.
e inner
uch
round the
hour would
t that -
OCXO which
rodegrees. A
have to AC-
well
be as good
ovens.
even that
inside a
would
are
ea few
rst
ming
Unless you go single side-band, which is practicable, if rarely practical.
Most cases.
Micro-kelvin temperature control with a platinum resistance temperature sensor - a rather specific example of temperature control.
A rather specific example of temperature control, rather different from the OCXO the John Larkin was talking about. I can see applications where you might want to stick one of them down a bore- hole, but it's not a high volume market.
and so
eRTD with
in, so
fine
if
So your temperature sensor - controlling the very specific environment whose temperature you want to control at the micro-kelvin level is bonded to a "metal thermal shield". If that's the temperature you want to control, what is it that that "metal shield" is actually shielding?
The actual sensor isn't bonded to any shield, it bonded to the lump of metal or whatever whose temperature you want to control.
If the sensor itself is injecting a couple of milliwatts of heat flux into that object, you've got a significant thermal gradient between the point whose temperature you are measuring and thus controlling, and the point you want to control.
oft,
So what was the object whose temperature you needed to control?
It obviously wasn't "fairly thin, soft anodised aluminium with a quilted pattern on it", because it's properties aren't going to matter to anybody.
he
sort of
e--if
lates
e lBut the way you find out that they are the right calculations is by building the thing and seeing what it does in reality. That rarely works first time around.
Mathematical modelling is a process of simplifying reality down to something simple enough to be tractable while retaining enough reality to make the results of your simulations useful. Frequent reality checks are obligatory.
-- Bill Sloman, Nijmegen
;)
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
t t
Hi Phil, I'm enjoying the conversation... particularly because I've got more temp control in my immediate future. Do you mind waxing some about diode sensors, (and diode connected transistors.) I was thinking about doing a pulsed heat capacity measurement, and what could I use for the heater, and thought about a transistor (which I've done before). I then remembered your idea (or the idea I heard from you first) of a dual transistor, one side for T sensing and the other as the heater. Now I'd really like a dual transistor in a 'power pack'. (which I don't think exists) And then I got to thinking that maybe I could use one 'power pack' transistor to do both jobs. (I'm not sure how to do the switching.... maybe JL's telecom relay?) If I pulse a transistor and then measure it's temperature I'm thinking I'll see a whole lot of thermal time constants. (which is a good thing.)
George H.
t -
an
at
k ,...
Sadly, it doesn't seem to pare down the nonsense as effectively as I'd like. Maybe it would work better if I were manic.
-- Bill Sloman, Nijmegen
Good temperature sensing needs three things: stability, sensitivity, and speed. Thermocouples are fast, but not too good otherwise. RTDs are stable and reasonably fast, but not very sensitive. IC sensors are sensitive but unstable and very very very slow.
I haven't used diodes in really high-stability applications, but one place they work really really well is in temperature controlling diode lasers. The monitor PD is brazed right on to the same metal header as the laser, so its response is good out to several hertz, which is really hard to achieve any other way. You do have to kind of stand on one leg to use it, of course, because you have to reject the photocurrent contribution somehow. (There are several methods for doing that, e.g. turning off the laser briefly or dumping in a pulse of forward current and servoing on delta V_F with the photocurrent kept constant by an auxiliary feedback loop.)
Some diode packages are pretty good for this, I expect, but you want a small diode with solid leads (like a 1N4148) rather than something with wire bonds. Those cylindrical SMD ones might work pretty well.
The notion of indium-soldering the assembled flex circuit directly to an anodized aluminum heat shield is a big win, I think.
Semiconductors have a fair amount of stress sensitivity, so I'm not sure how good diodes are in the presence of board stress and so on. Probably not as good as RTDs. Their sensitivity is similar to an RTD, and their resistance can be adjusted by varying the forward current. An ideal diode has half the noise of a resistor at a given temperature, so you can get lower noise at the same sensitivity, at the price of more dissipation.
All of which is an extended way of saying "Diodes have a lot in common with RTDs, and are much cheaper, but I don't know how stable they are in real life."
Cheers
Phil Hobbs
-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.