But it does allow you to operate at a frequency where 1/f noise sources aren't a problem. The only way to reduce Johnson noise is to reduce the temperature, which really isn't an option in this application.
Sometimes. Platinum resistance sensor work up to 550C - to quote somebody's claim from earlier in the thread. Chopper amplifiers stop working at appreciably lower temperatures.
Do list a few. But at least it isn't sensitive to thermocouple voltages and other 1/f noise sources.
This another example of the consequences of careless quoting.
"But, Bill's point remains. The use of AC excitation buys your way out of several traps. If Johnson noise is higher than the temperature signal, the AC method works. The best chopamp doesn't. If self-heating of the sensor matters, RTD sensors will always give more accurate measurements with small excitation signals. AC doesn't lose accuracy. Chopamp loses accuracy."
Originally "Chopamp loses accuracy" was the last sentence in a paragraph, and forms part of the conclusion to a coherent argument,. I've retrieved the entire paragraph, which had been ripped apart in the subsequent discussions.
John Larkin won't agree - his self-image is at issue - but he should at least have had enough sense to find out where the statement came from, and what it had originally been intended to mean.
Good stuff, thanks. They're using a DC bridge with the control electronics in the outer control volume, to reduce the thermal forcing. Inside the inner one would be better, but with the temperature adjusted to the crystal's turning point, it doesn't matter so much. Variable gradients inside the inner box would probably be more of an issue.
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've done both AC and DC.. and DC is easier, electronics-wise :^)
Phil can you elucidate the 3dB loss in the AC measurement? Is this just the rms value of the sine wave? So square wave excitation would reduce that advantage? Or something more subtle? And if you were to do a lockin measurement, don=92t you get a 3dB improvement because you get rid of the noise in the other quadrature? Maybe even square wave modulation and then switched gain (+1/-1) phase detection.
George H.
(still planning to use DC excitation for the next temperature thing I have to do.)
Sure, if there is any relevant low-frequency noise, but the whole point is that in this case there isn't. I've often done AC measurements for just that reason, but this isn't one of those cases, not any more. (I even bought myself a lock-in amplifier just this year. I'm a fan of AC methods when they're necessary.)
Read the OPA378 datasheet, and think about the effect of putting all those time-varying thermocouple offsets inside the inner oven. They go away, to any accuracy relevant to the problem (i.e. if your thermomechanical design permits no gradients due to external forcing, they go away completely, and if not, they go away to the accuracy of the rest of the system and therefore don't limit its performance.
Moving the goal posts is usually an excellent alternative to doing some more complicated measurement.
My other worry about transformer bridges is that they're almost impossible to test to the required accuracy. I'm not necessarily against putting all your eggs in one basket like that, but you do need to be able to watch that basket carefully.
A bead with a crack in it could easily raise the imbalance from 0.1 ppm to 10 ppm, which could destroy the measurement, but you'd never find out until it flunked final test, or failed in the field.
accuracy.
Constant gradients are no problem. Why would they be? It's just the uncontrolled time variations that you want to get rid of. You can very easily make them constant down to ~ 10 nW if you need to (which you don't).
And, once again, in the flatband a DC measurement has a 3 dB SNR advantage for the same sensor dissipation.
An offset is not a noise source. If you care about it, it can be trimmed out by moving the setpoint very slightly. In instruments, one very rarely cares about the absolute temperature--it's the stability that matters.
Some things like diode lasers may need to work at some specific temperature, but you don't know what that is until you find the right setpoint, so you don't need to know the actual number in kelvins to any great accuracy. You just search for the setpoint that gives you the behaviour you want (optical frequency, stability, interferometer operating point, what have you). Then you just want it to sit still.
voltage
Can you give a numerical estimate as to how large that will be, using, say, a 5-ppm voltage reference, 2 mW of sensor dissipation, and an OPA378 driving a high impedance load?
The dissipation of the chopamp can be controlled to the same tens-of-nanowatts level by putting a second integrator in the loop, outside the inner box, in front of the power amp. (You need to manage the frequency compensation of the resulting second order loop so it doesn't oscillate, but that's not too hard.) Doing that will keep the output voltage of the chop amp pretty well constant, so any dependence of I_bias on V_out won't have a chance to cause problems. This takes one extra capacitor, no more. (It's trivial in software if the loop is digitally controlled.)
Gradients will also > respond to atmospheric pressure or gas composition. That DC error is NOT > a reliable constant.
It's true that there may be effects of atmospheric density and composition. However, those affect all the gradients inside the controlled volume and not just the ones from sensor dissipation. The biggest effects of that would almost certainly be changing the thermal resistance of the insulation, and thermal forcing from gas exchange. If those are big enough to matter, the AC method is in trouble too, by the same amount, and for the same reason--it's a thermomechanical problem.
Sealing the box would be the usual approach to that, as was done in the HP 10811A in the teardown--they soldered the box shut and used hermetic connectors.
As I said, it's thermomechanical design that's the hard part, not the control loop.
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
hobbs at electrooptical dot net
http://electrooptical.net
If you have a baseband signal with a bandwidth of B, and use it to amplitude-modulate a carrier at frequency f, the resulting signal will extend from f-B to f+B, which is twice the bandwidth. (*)
As Bill points out, all of the information is present in either sideband alone, so it's possible to use SSB detection to get rid of the 3 dB penalty. However, the circuitry required has a strong tendency to be both complicated and drift-prone, so it isn't a big win in this case, especially since there are lots of other things you can do to get back the 3 dB if you really need to, e.g. using a bridge with two sensors diagonally opposite, or just jacking up the excitation voltage.
Cheers
Phil Hobbs
(*) I talk about this in gory detail in my book, and I often get asked if there isn't some intermediate case, say if you increased the carrier frequency smoothly from 0, wouldn't the bandwidth go smoothly from B to 2B?
The bandwidth would do just that, but if you allow the baseband signal to get folded back on itself, information gets destroyed and the measurement doesn't work. Another way of seeing that is that for real-valued signals, you actually have two carriers at +-f, so if f <
2B, the USB of the (-f) lobe lands on top of the LSB of the (+f) lobe. To preserve information, you have to jump from f=0 to f > 2B, so AC is always 3 dB worse than DC (except with SSB).
--
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
hobbs at electrooptical dot net
http://electrooptical.net
What particular case would this be? Whatever it is that you stuffed down a bore-hole, or John Larkins OCXO? Or have you found yet another new set of goal-posts?
What prompted all of this was the proposition that they can become necessary - or at least advantageous - if you want to use a Pt resistance thermometer to get micro-kelvin temperature sensitivity.
They never go away completely, but they can certainly be usefully reduced in a lot of systems. Whether they can be reduced enough in any particular system is always an interesting question.
If you can get away with it. Mother Nature punishes misapprehensions.
Standards labs love them, and they are the last people to take stuff on faith. Testing isn't "almost impossible", it's just non-trivial.
What bead? Standards labs have a preference for mu-metal toroids - I'd go for a ferrite core.
In either case a break in the magnetic path is going to make a big difference to the inductance of any one of the coils involved, and you could see it with the simplest inductance meter.
You can perhaps make the dissipation constant. It's harder to be sure that you can keep the thermal gradients constant and oriented the same way.
Only if you can't be bothered to make the system only look at one side-band.
An offset is not - per se - a noise source, but it always has the capacity to become one. In practice no offset is completely stable.
Don't forget to figure in the acoustic noise genenerated by your bank manager beating down your door.
Pretty well constant ...
That doesn't actually follow.
It can be. Particularly if you leave yourself vulnerable to thermocouple voltages by measuring at DC, rather than setting up an AC excited bridge.
On Monday, September 17, 2012 7:08:23 AM UTC-7, John Larkin wrote \[about ovenized crystal oscillators]
I thought the AT (shear mode) quartz crystals were not preferred for reference oscillators, because they are sensitive to minor material losses from the large polished surfaces. There used to be another crystal cut, GC, which had some kind of internal-oscillation mode that didn't stress or accelerate the surfaces, that had better aging characteristics.
SC cut (stress compensated) is best, but they are really expensive. I think they are still shear mode.
--
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
How odd. I was under the impression that they were standard today, now that the patents have run out. I guess catalog cut sheets and such may reveal the current situation.
Traditional analog SSB uses a "sequence asymmetric polyphase filter" to get a more or less constant 90 degree phase shift over a couple of decades of frequency. "The Art of Electronics" spells out the gory detail, or you can go to
M.J. Gingell ?Single Sideband Modulation using Sequence Asymmetric Polyphase Networks? Electrical Communication 48 21-25 (1973). See also
formatting link
This isn't going to be all that practical for the fractional-Hz bandwidths you end up with in temperature control, and you are much more likely to be doing most of your processing in the digital domain, where complexity isn't much of a problem and drifts are non-existent.
There's also the point that SSB radio is all about rejecting strong signals in adjacent channels, whereas in this application the "adjacent channel" has precisely the same strength as the channel you want to detect. In this application, 90% rejection is going to be almost as good as 99% rejection, when it would be a total disaster in a radio receiver.
OK I think I get it. The AC bandwidth has to be bigger than the DC bandwidth and (all other things being equal) you get more noise AC- wise because of that.
Say these 'gains' on the edges of things, reminds me of the Pseudo- derivative feedback of Phelans. (Really negative feedback around the plant (heater).) Besides taming the overshoot of a PI controller to a step change in the set point, the real 'win' in Phelans idea seems to be a more efficient use of the plant. It keeps the whole loop running in the linear mode over a bigger error range. Rather than a step change in the plant output, followed by an RC decay, the plant turns on more slowly, but for a longer time. Keeping the total energy the same. (I still have to try out these ideas in practice.)
George H.
B?
OK I'll go look it over... can you quote me the chapter and verse? (ie Hobbs 7:16) (I hope the slight profanity does not offend.)
The other thing I should have pointed out about SSB is that the filter you'd need (either Hilbert or bandpass) has to cut off infinitely sharply at f0, and so will have infinite group delay.
However, I do have a correction to make.
I'm so used to the usual optical AM case that I wasn't really thinking about the phase sensitive detection step, which of course gets you the 3 dB back for free--half the noise is in the quadrature phase, and so goes away when you mix and filter. So the 3 dB is a red herring--AC and DC are theoretically equivalent when you're using a 1-phase lock-in technique.
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
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.