At some point, using a crystal isn’t a good idea. If you want an extremely accurate frequency standard, at a reasonable price, look here: http://www.accubeat.com/prod_Rb.asp
Oh, and these days everyone has access to exceptionally accurate timebases using GPS…
What if you need a high-stability frequency standard?
Realistically, for what?
Yes, GPS isn’t going to work, but if you really need something like that, you are better off using a device like I mentioned above. I’ve never encountered a application that required frequency stability better than a crystal could provide. If you are doing research or making some sort of reference, then you should have the budget to buy a certified primary (or at least NIST-traceable) device.
What I’m working on here is that the standard way of making the oscillator hit the frequency you want by compromising other things wastes what the oscillator is best at, to get it to do something microprocessors are way better at anyway.
I’ve never heard of a quartz oscillator made to be stable without trading some of that away to get the frequency closer to the nominal value. Yet, often, the choice of the nominal frequency is arbitrary anyway. Since people do put a lot of effort into making oscillators stable, even in oscillators without Rb or other atomic physics packages attached. It just seems obvious to me that you should make the oscillator stable at the cost of making the frequency precise, if you’re spending money to make it stable, and have processing power involved anyway.
Well, I’m not an expert in Quartz crystal oscillators, but I have done some work in error budgets for other applications (pressure sensing). To make an extremely precise (not extremely accurate) oscillator, I would think that one would have to understand all of the aging and drift errors very well. Some of these errors are reasonably well understood (temperature effects, for example), but others can’t be controlled as well (aging due to mechanical stress relief). If I were going to design such a device, I would probably shoot for as many different oscillators as i could reasonably afford, and run them in parallel (which of course, is the way the original time standards were done). Still, there may be unpredictable errors that can’t be compensated for. One that comes to mind is microphonics, which might be significant once all the other errors are accounted for.
You might be interested in this paper on modern TCXO design. I ran across it while doing a google search. They use an ASIC to do a fairly sophisticated temperature compensation function on an AT-cut crystal.
Not really. The idea of a crystal-controlled motor is that you can shoot an entire magazine without losing synch. The 60hZ cycle is a constant speed governing the motor and shutter. Without it, films cannot be shot in synch. In theory ( though not always in practice ), this means you can shoot an 11 minute take without going out of synch at all. 400 feet in 16mm is 11 minutes. 1,000 feet in 35mm is the same time.
Well, in 11 minutes a 1MHz oscillator will have oscillated 660,000,000 times. If the oscillator exhibits 100ppm instability it will be off by +/- 6.6uS at the end of 11 minutes. A frame of film is 1/24 of a second or 41,666uS, so this error is only .016% (of a frame). As has been discussed above, the average clock oscillator can be accurate to only a few ppm, so I think that the film timing is pretty easy…
Posted by beowulff:
You’ve hit the nail on the head here; one of my biggest problems over the years was educating customers as to the difference between accurate and precise; the two may or may not be mutually exclusive, but either can be very difficult to achieve due to the parameters you’ve mentioned, as well as several others.
Well, your final sentence is key; money rules the design process. If you are an oscillator design engineer with an unlimited budget and an in-house crystal manufacturing facility you might be able to come up with what I think you are driving at; I somehow doubt it, though. I think, if I understand you, that you are crediting the microprocessor with way too much internal precision. In the usual case, you will find the crystal located between two pins of the microprocessor, with a capacitor on both sides of the crystal. Without going into too much detail here, those capacitors play a vital role in the accuracy of the output frequency and IF those capacitors vary significantly from their nominal value, the output frequency will deviate accordingly. That is precisely the usual case and I can almost guarantee that the internal capacitance values of the microprocessor DO vary significantly. The necessity of canceling these variations is what gives rise to VCXOs (Voltage Control Crystal Oscillators). A change in frequency results in a change in applied voltage to bring the frequency back to the desired value. And, bear in mind that we haven’t touched on frequency changes resulting from temperature changes; that is another kettle of fish. Basically, if I understand you, you want to minimize component count in order to achieve something that can only be achieved by adding components: Microprocessors are just not that accurate and precise. Or they weren’t during my day, at any rate.
If we are talking about the parasitic capacitances of in the driving circuit. The variance is a much higher percentage of the total capacitance but the total capacitance is much lower. So you really need to know if the geometries of the crystals have been scaling down as much as the geometries of driving circuits and packaging.
Bolding mine. I’ve never heard the theory that angle selection influences Q; The Q of a quartz crystal is a function of the design of the resonator plate. Angle selection does influence the frequency vs temperature characteristics, most notably in resonator plates that vibrate in thickness-shear.
In the case of a well made crystal unit, the resonator is encased in such a manner that it is shielded from contaminants; it certainly shouldn’t be sealed with contaminants present. The introduction of external contaminants implies a failure of the case seal; larger problems than aging with result in that case. And, a reduction in frequency due to mass loading is largely limited to thickness-shear resonators. To be a little nitpicky, the electrodes on today’s resonator plates are applied by metal evaporation in a vacuum chamber; electroplating implies a chemical process. While electrode detachment might have been a problem at one point, advances in technology have virtually eliminated that failure mode. Dramatic shifts in crystal frequency most often result from excessive mechanical shock and, more commonly, excessive drive current applied to the crystal.
The tuning fork resonator at 32.768 kHz resulted from a very intensive research and development program that took place mainly in Japan. It is based approximately on what is known as the D/T cut, which is a low-frequency cut. Low frequency quartz resonators (<1.0 MHz) are typically bar shaped, with higher frequencies resulting from shorter bars; at some point, the bar becomes too short to handle. Frequencies greater than 1.0 MHz are nearly always thickness/shear plates, with the A/T cut being dominant. There was a lot of research and development done with A/T cut resonators at a frequency of 4.916(?) MHz; the tuning fork design won the contest hands down. The technology of processing thickness shear resonators at lower frequencies has improved considerably, but the resulting crystals are still far too large physically to be used in a wrist watch. In any event, the so-called watch crystal is now a mature technology; it works very well, and the chances of changing it are slim to none. In any event, I believe there are now watch movements that are nearly all electronic, with no crystal at all.
And though I didn’t bold it, I believe most, if not all, atomic clocks utilize 5.0 MHz, 5th overtone thickness shear resonators as their basic controlling element.
As far as physical sizes go, I am astonished at the degree of miniaturization that has been achieved with quartz crystals; they are now being manufactured in sizes I would have thought impossible just a few years ago.
Capacitance surrounding a crystal has always been a problem; it is easy enough to demonstrate that the frequency of thickness-shear resonators especially can be greatly influenced by external capacitance. An external load of 20.0 pf, for instance, will change the frequency more than an external load of 30.0 pf; this is the reason that symmetrical frequency changes cannot be achieved by a symmetrical change in load capacitance from a center value. It isn’t widely understood by users that a “lower” value of external capacitance will have a greater effect on the output frequency, nor is it widely understood that a “lower” value of external capacitance will result in an oscillator circuit that is more difficult to start; a higher value of drive current will be required at a “lower” value of external load.
There are some good technical articles here, some written by me and some by others. There are also a few application notes, if you can get the links to work. One of them, on the 32.768 kHz watch crystal was written by me.
I am no longer employed by Fox Electronics, so I have no commercial end in mind by recommending their site.
All very interesting. Maybe I can only add one refinement:
>In the usual case, you will find the crystal located between two pins of the microprocessor, with a capacitor on both sides of the crystal. Without going into too much detail here, those capacitors play a vital role in the accuracy…
I think you’re describing a microprocessor that is clocked by its own built in oscillator circuit that needs a quartz crystal to stabilize it. What you say is correct in this case. But I’m referring to timekeeping applications, where oscillator accuracy is a priority. I assume the quartz crystal I am interested in is packaged internally to an oscillator product, like a little ovenized can, where they give appropriate attention to precision in the circuit components that influence oscillator stability. My daydream includes a microprocessor which might get its clock signal from the precision oscillator, or might get it from a more typical microprocessor clock external crystal or 100 ppm packaged oscillator. My microprocessor is calculating slopes and offsets and corrections and rescaling the frequency to some more useful standard value like 10 MHz.
>Microprocessors are just not that accurate and precise. Or they weren’t during my day, at any rate.
In the sense that I claim microprocessors are precise, they can calculate Pi to a million digits. I’m talking about digital calculations, not analog circuit parameters concerning the microprocessor pins like the crystal pins for microprocessors with built in oscillators needing external crystals.
I have seen applications where a stock quartz crystal is used for timing and then it is calibrated by measuring the frequency using a more precise external reference. Say you want to generate a pulse every 1 ms. using a 10MHz crystal you would make a pulse every 10,000 cycles. You can adjust the frequency by changing this from 10,000 so say something like 10,005 to account for variations in the frequency of the circuit. You can gain greater accuracy by doing say 10,005 for a few pulses then change to 10,004 for a few more. There is a certain amount of jitter from a 1 KHz pulse but over the long run you can get arbitrarily close to an exact frequency.
In this manor you can calibrate out initial variation. It does not help you much with variations due to aging and temperature changes.
Okay, I didn’t fully understand; why don’t you start with a ready made oven controlled crystal oscillator? They are certainly readily available although they are somewhat expensive. You might look for a quote on an ovenized oscillator that utilizes the recently developed SC cut crystal; you’re talking almost unbelievable accuracy and stability with those things.
>Okay, I didn’t fully understand; why don’t you start with a ready made oven controlled crystal oscillator?
Well, I would, if I were picking from products that were available now. But I’m proposing a new product. It’s a quartz crystal oscillator only available in one surprising frequency value, with a frequency tolerance on that value of 10%, but whose stability (with respect to temperature and aging and other things) was more predictable than other similarly priced oscillators. You could build a more accurate clock around this oscillator than you could around any other oscillator for that price, because I didn’t compromise oscillator predictability to hit an arbitrary frequency target or minimize the size of the different instabilities, and because you have access to a microprocessor that could use the predictability to manage all the instabilities.
Again, I’m proposing that oscillator manufacturers shouldn’t worsen timekeeping, which is what quartz is good at, by trying to make the oscillator match an arbitrary frequency or temperature response, which the microprocessor would be better at (digitally). As it is now, they’d lower the Q of the oscillator to steer it directly at 10 MHz (say), just because one of the memory locations in the processor has 10000000 stored in it (in effect). I’d raise the Q and write a different number into that location in the processor.
You need to develop a complete specification for the crystal (or the packaged oscillator) that you want and submit it to a crystal (or oscillator) manufacturer for review.
–blink–
Do that math again. I dare ya. 
Well done, and I sit corrected. Film timing IS pretty easy next to the accuracies being discussed in here.