18

The clock sources in modern electronics seem to come invariably from quartz and MEMS oscillators, both of which generate vibrations mechanically. The amplitude and frequency of the vibration are orders of magnitudes different from the everday mechanical vibrations I observe in, say, musical instruments. Nevertheless, it's surprising to me that we don't get clock sources in the electromagnetic domain directly, say using capacitive or inductive elements.

I know that inductors especially are hard to manufacture without parasitic losses. But I would expect mechanical oscillators to be non-ideal as well.

You could use the propagation delay of electricity, but then it would be hard to make a small oscillator that operates at slow frequencies.

Is it really true we can make microscopic vibrating devices more ideally than we can make electrical oscillating components?

Gus
  • 431
  • 3
  • 11
  • 4
    Just a note -- Quartz crystals were the new, better frequency control for radios back in the 1920's. I have amateur radio magazines from 1928 where they're already an established technology (albeit way bigger than today's). For a while they were the best frequency control standard to be had, only being overtaken by atomic clocks in (I think) the 1940's or 1950's. So the **practical** answer to your question is because they work better and cheaper, and no one has been able to do better without being a whole lot more expensive. – TimWescott Nov 27 '18 at 20:33
  • 1
    Thanks for that note. Practicality aside, does it strike you as surprising? If someone told me that the voltage reference in a circuit comes from a generator connected to a constant-velocity reference. (or even better, from the amplitude of the current or voltage generated by the quartz crystal), I would think that's a little funny. I've known that crystal oscillators were mechanical for a while, but today it struck me as odd that it's actually good in practice. The electrical domain seems to win for signal processing, energy transfer, communication, and so on. – Gus Nov 27 '18 at 20:40
  • 3
    If I were to remain that surprised by everything that does not make immediate sense, I would not be able to get out of bed in the morning in my astonishment that the sun is up and gravity still works. I suppose it's kind of surprising, but it would require very deep study to find a really good "why". I tend to be distrustful of anything glib; I'm not sure that there really is a good, 100% true, and short explanation for this. – TimWescott Nov 27 '18 at 20:45
  • 8
    Quartz is simply *amazing*. It's piezoelectric effect is very large (the link between its mechanical/electrical properties). Its inherent temperature coefficient is very small. Any remaining temperature effect can be reduced by rotating crystal planes. Grinding/lapping can be done with great precision. Sometimes, the universe just gives you such a gift. – glen_geek Nov 27 '18 at 20:51
  • As a novice amateur radio operator in the mid 1950's, the FCC REQUIRED me to use quartz crystals. Fortunately, I found a source of cheap crystals around 6.5 MHz, and was able to re-grind them to around 7.15 MHz. – richard1941 Nov 30 '18 at 00:09
  • Quartz is not just good for radio frequencies. It is also good for around 0.01 Hz, as in the torsion pendulum that is used to measure the universal grav constant. – richard1941 Nov 30 '18 at 00:11

2 Answers2

21

Because the mechanical devices are much more stable than their electric counterparts. Let's compare a crystal oscillator to an LC oscillator:

Crystal:

  • Has a very high Q. According to wikipedia, a crystal oscillator has a typical Q of 10,000-1,000,000.
  • Stable with temperature. Many crystals are specified at <50ppm over their temperature range, and temperature compensated or controlled crystals are also available, down to ~1ppm with temperature
  • Manufactured to a tight tolerance. Cheap crystals are usually specified to ~25ppm, but tighter tolerances are available

LC or RC:

  • Not available as an integrated device, so must be assembled from off the shelf components (unless integrated into a mcu or similar)
  • Low Q, it's difficult to make an inductor with a Q higher than a few hundred
  • Temperature sensitive - making temperature stable inductors is difficult

  • Voltage sensitive - the threshold voltage and charging voltage in the feedback circuit is usually voltage dependent.

    However, that doesn't mean that electric oscillators are never used, just that they're not used where great precision is needed. They do however have some advantages over crystal oscillators:

  • They can be easily integrated into another IC. Many microcontrollers now come with an integrated oscillator

  • They (sometimes) use less power. Often times a microcontroller will include a low power oscillator to run the watchdog timer, which uses less power than a high speed (MHz) crystal, and sometimes less power than a low speed (32.768kHz) crystal.
  • Since they can be integrated onto an IC, they can be used in places where a crystal would be far too large
  • They can be tuned fairly easily. A crystal can only really be shifted a few kHz off its calibrated frequency, but by adjusting the capacitance of the LC circuit (like with a varactor diode), the frequency can be adjusted over a fairly wide range. This means that LC oscillators can be used in circuits like PLLs or VCOs, possibly even locked to a crystal reference.

Non-mechanical oscillators are used in many devices, just not in those where accurate timing is required.

C_Elegans
  • 2,871
  • 14
  • 27
  • I asked in another comment, but I think I assumed we can build, for a similar cost, a more accurate voltage reference than we could a mechanical oscillator (that's a bit of an apples-to-oranges statement, but all other things being equal, measured both in frequency ppm of the final oscillator). – Gus Nov 27 '18 at 20:18
  • And I know that inductors are notoriously non-ideal. What about just using capacitors? – Gus Nov 27 '18 at 20:19
  • You can't use just capacitors for a time base. You'd need capacitors and resistors -- and that would be worse than capacitors and inductors. – TimWescott Nov 27 '18 at 20:22
  • 1
    One last question for now, I think of the Q factor as a measure of how well an system "rings" after you "strike" it. In a powered system, it would correspond to how much energy is needed to keep the oscillator running (not considering that the oscillation signal has to be sensed by some other circuit). Is this the big deal with a higher Q? Or does higher Q help ensure frequency stability as well? – Gus Nov 27 '18 at 20:22
  • @TimWescott, worse in what way? temperature sensitivity? Surely not worse in terms of parasitic effects, right? – Gus Nov 27 '18 at 20:23
  • 2
    The sensitivity of an oscillator to noise is inversely proportional to Q. That's part of the reason why an RC circuit would be worse than an LC circuit -- an LC circuit may have a Q of 100 or more, an RC circuit has a Q less than one, always. – TimWescott Nov 27 '18 at 20:24
  • 2
    High Q also relates to how stable the system is. A high Q oscillator has less phase noise than a low Q one, which is important for radio circuits and timing sensitive stuff (like controlling an ADC clock or DAC) – C_Elegans Nov 27 '18 at 20:25
  • 1
    @TimWescott. Thanks. Just to be sure, an RC circuit on its own does not oscillate, so I figured it does not even have a defined Q value. (In other words it's a first-order system.) – Gus Nov 27 '18 at 20:25
  • 1
    @C_Elegans. okay thanks for bringing that point in. I don't think I fully understand how Q relates to phase noise or sensitivity to noise. If you have any pointers, please let me know. I'm picturing two spring-mass-damper systems, at the same frequency, but one with much a worse Q factor, for whatever reason. If the one with the worse Q also has a much higher mass, then it seems that it would be less sensitive to external disturbances. It would take more energy to disturb it. It would also take more energy to keep it oscillating, but that seems like a separate issue. – Gus Nov 27 '18 at 20:31
  • No passive circuit will oscillate on its own. A single-R single-C circuit will oscillate if it's hooked up to a Schmitt trigger, but that's because the amplifier contributes a state. A 2-R, 2-C circuit will oscillate when connected to a suitable "straight" amplifier -- which is what you need for an LC or crystal circuit. – TimWescott Nov 27 '18 at 20:36
  • 1
    The system with the lower Q is easier to get to vibrate at a different frequency from its resonant frequency. Any noise in the circuit will tend to make the oscillator want to shift frequency slightly, and a higher Q circuit will resist that shift better. – C_Elegans Nov 27 '18 at 20:37
  • Your hypothetical heavier, lower-Q mass-spring-damper system would need a proportionally stouter damper. That stout damper would conduct disturbances to the mass with more authority than the higher-Q system. The lower-Q damper would also generate more thermal noise, if the system were linear down to the thermal limit. – TimWescott Nov 27 '18 at 20:38
  • @TimWescott, just for the sake of clarity, an LC circuit does "oscillate on its own", given the proper initial conditions. You might argue non-zero initial conditions means that the circuit isn't passive. But the differential equation governing an LC circuit and the one governing an RC circuit are structurally different in a way that allows one to oscillate and the other not to. That's why I don't think you can assign a Q value to an RC circuit. – Gus Nov 27 '18 at 20:43
  • OK, yes, an LC circuit exhibits damped oscillations. But nothing physical will exhibit infinitely sustained oscillations that can be measured without an active element involved. Even a \$Q=\infty\$ resonator couldn't be *measured* in its oscillation without extracting energy from it; that energy would have to be put back in with some active element. – TimWescott Nov 27 '18 at 20:47
  • @TimWescott Thanks for running with my hypothetical mass-spring-damper system. I will have to think more carefully, but I had thought that for any value of the damping and any value of the mass, you can choose a spring in order for the system to oscillate at some desired frequency. – Gus Nov 27 '18 at 20:49
  • 2
    " I think I assumed we can build, for a similar cost, a more accurate voltage reference than we could a mechanical oscillator". Only if you have an atomic clock handy. And some liquid nitrogen. See [this link](https://www.nist.gov/news-events/news/2013/04/primary-voltage-standard-whole-world). – TimWescott Nov 27 '18 at 20:50
  • 1
    "I had thought that for any value of the damping and any value of the mass, you can choose a spring"... Yes, but increasing the spring rate increases the Q, unless you increase the damping to match. – TimWescott Nov 27 '18 at 20:52
  • All piezo-electrics and MEMs are mechanical toodue to the motional capacitance SC cuts are >100x higher Q than std AT cuts. The nanufacturing process and structures determine the best technology to use. LC are the worst for Q and stability in 2nd order types – Tony Stewart EE75 Nov 27 '18 at 23:01
  • 3
    I can buy easily buy a crystal oscillator TCXO that is stable to within +/-50ppb over 0° to +70°C for less than $30 one-off. A 0.6ppm/°C temperature compensated voltage reference costs more than $150. Initial tolerance is +/-1ppm vs. 0.01%. So orders of magnitude worse for 5x the cost. That's not atypical. You can easily measure frequency better than ~\$10^{-10}\$ accuracy (1 year), but voltage is difficult to measure better than single digit ppm accuracy (I'll include Tim's Josephson Junction laboratory standard which lives in a dewar at 4.3 Kelvin as more than difficult..) – Spehro Pefhany Nov 27 '18 at 23:47
  • The application of mechanical devices goes beyond oscillators. They are important components of filters. As in quartz crystal filters and mechanical filters. That is where their high Q can be put to good use. – richard1941 Nov 30 '18 at 00:14
4

It's not really whether inductors and capacitors can be made more precisely than a mechanical oscillator. It's whether those components can operate in a stable manner over voltage/temperature ranges. Unless you want to design all of your circuits to have a band-gap voltage reference, a thermometer, and a heating circuit to keep voltage/temperature constant, you can't get inductors and capacitors to operate anywhere nearly as stable as a crystal does.

To tune a crystal to the correct frequency during manufacturing, I'm assuming they could just polish it until it's at the right size. You can also manufacture caps and inductors as accurate as you need. The problem is that it just won't stay there.

horta
  • 12,738
  • 22
  • 45
  • 1
    Is it important that the clock source be stable over voltage ranges? I had figured that modern electronics, like your cellphone, does have an accurate voltage reference (due to a band-gap). Stability over temperature makes more sense. There are oven-controlled crystal oscillators, so they must be sensitive to temperature as well, but to a lesser degree? – Gus Nov 27 '18 at 20:14
  • @Gus voltage range won't be nearly as important as temperature. For really accurate stuff, it makes sense to temp-control a crystal. – horta Nov 27 '18 at 20:19
  • GSM cellphones are trimmed in frequency, so the packets do not drift in timing; this ensures there always is the predicted rampup and rampdown time between packets and there never are missing or conflicting simultaneous packets. – analogsystemsrf Nov 28 '18 at 02:54