A mechanical phase locked loop!

"This feedback loop functioned as an electromechanical version of a phase-locked loop..."

Rubbish, the function of a phase locked loop is to keep the phase of the

2 signals the same, within the constraints of the loop filter.

The clock *never* achieves this, it is open loop and applies a 'kick' to one pendulum the amplitude of which is NOT related to the difference in phase of the 2 pendulums.

A fixed kick is given without any knowledge that it will be of the correct amplitude to achieve an in phase or near in phase condition. There is NO feedback of an error signal that relates to the phase difference between the 2 pendulums.

The only time phase comes into the picture is the timing of when the 'kick' is given, so as not to disrupt the normal swing of the pendulum, and whether or not to give a kick at all.

It is and ingenious system, but not a phase locked loop.

I guess it could be closer to a PLL if the kick had its amplitude varied by the phase difference between the 2 pendulums, but you still have the problem that if you were in the state where no kick was required there is no way of slowing the second pendulum without waiting for it to drift back, so it is still open loop.

Jeff

..and of course everything on Wikki is correct!!!

Jeff

Exactly. The control is single path, master to slave, with no feedback to the reference, making it an open loop design. The master has no knowledge of the state of the slave at any time.

In a pll, there is continuous feedback from the vco to the phase detector, closing the loop and keeping the phase offset constant, The phase is continuously updated every cycle, whereas the Shortt clock can have significant accumulated error in the time between corrections...

Chris

Untrue.

The matter starts off when the slave signals to the master and drops the gravity link in the master, then, when the master pendulum is in a position to accept the impulse from that dropped gravity link, it signals back to the slave

But ... I'm still trying to google for the exact mechanisms because most URLs only hint at what is happening. (I'm also awaiting delivery of a couple of hope-jones' books about electric clocks)

I hope you have more success getting a copy of those books than getting the PA tuning instructions for a mass produced amateur TX.

Jeff wrote on 8/5/2017 5:45 AM:

You are making pointless distinctions. A phase locked loop is not defined by its mechanics but by the nature of its control. The Shortt clock maintains the relative *phase* of the two clocks by brief adjustments to the frequency via a spring. This is controlled by measuring the relative

*phase* of the two clocks.

It's that simple. You are just making things more complicated by talking about the details of how the adjustment works and the time function of the frequency. NO PLL can keep the two clocks perfectly in sync.

Calling it open loop is just absurd. The loop is closed because it

*measures* the phase of the clocks and adjusts the phase according to the measurement. It may be binary, but the adjustment is controlled by the measurement.

You aren't making sense. The reference is never adjusted in a PLL. That's why it's the *reference*.

There is no requirement in a PLL for continuous action or even frequent action.

The amplitude is not, but the frequency is - why do you think the amplitude should be related to the difference in phase?

Ah, yes there is, see below.

Are you referring to the kick given to the master pendulum? That is not part of the PLL system. The kicks given to the master pendulum are specifically designed not to affect the phase of the master pendulum at all.

If not, if you are referring to the kick given to the slave pendulum (these are quite different kicks) that is not how the clock works.

The slave pendulum is kicked from time to time, ad kicked a little more often when the phases get too far apart - the difference in phases is the error signal mentioned above - and these kicks do affect the phase of the slave pendulum.

That is exactly what a PLL is - and it is almost (though not quite) what this clock does. It is certainly what the slave does.

Not necessarily continuous - a bang-bang action is allowable, and does not prevent a system from being a PLL.

A PLL does not necessarily keep the phase offset constant, just within the interval =/- 2pi.

Not necessarily continuously updated, or updated every cycle - as long as the offset is continuously within the range -2pi to 2pi, the phases are locked.

Yes - but that doesn't mean it is not a PLL, as long as the error is less than +/- 2pi.

A phase-locked loop is a system which produces a (slave) vibration the integral of whose phase in comparison to the phase of another (master) vibration is continuously between -2pi and 2pi over long periods.

A last requirement is that the phase-locked loop system should have no effect whatsoever on the master vibration. That's it.

If it does that, the phases are locked - they may not be tightly locked, but the vibrations do not skip or add beats.

More advanced PLLs might keep the difference between phases much smaller, as in this clock - but that is not a requirement of a PLL. There is no such thing as absolutely tightly locked, there is only unlocked or locked.

Neither is continuous updating necessary, though the integral should be continuously in that interval.

In this clock the hit-and-miss synchroniser action undoubtedly does act as a PLL.

However it might be argued that the slave does subsequently have some (very small) input to the master, when it operates the gravity drive (whuzzat? I am not a clockmaker).

That certainly has an effect on the amplitude of the master; although as the idea an intention and practical effect is that it has no effect whatsoever on the phase of the master, thus the slave clock action overall most definitely should be considered a PLL.

-- Peter Fairbrother

ps; the +/- 2pi bit is not really a requirement either, as long as the system can keep count of the missing/extra beats - but as most systems don't do that we shall just gracefully ignore that for now ..

What you are describing is how the phase measurement of the master is made. The gravity lever is simply a remontoire providing a consistent push to overcome the force of friction. It is designed to be invariant of small changes in timing of its release. You can see that in the animation linked below. The gravity arm is released at the point when the wheel is directly under the end of the gravity lever. A small change in timing changes the force only a tiny amount. This is critical to maintaining the swing of the free pendulum without affecting its period.

The animation happens in real time so it is hard to see the details of what is going on. The gravity lever and accompanying control is the magic of the clock. The rest is pretty straight forward. You need Flash to view this page. There is a button to see the wires.

What they fail to see is that the amplitude of the kick *is* adjusted. It's just the adjustment is binary, on or off. But that is still *adjustment* and is in response to the measured phase.

Not only that, but if you examine the equations for a PLL you will find it is *impossible* to maintain a constant phase offset with any variations in the reference or noise in the system.

In a typical PLL isn't the requirement to be within +/- pi rather than 2 pi? If you exceed a range of +/- pi from the intended alignment the feedback will start to push the controlled oscillator further out of alignment potentially aligning with another cycle of the master.

The usual cry of those who have not bothered to do any research on a subject and are shown a Wiki article that contradicts their position is that Wiki can be edited by anybody.

Wiki is more correct than most of the babble on USENET and the Wiki article has 18 external references to back it up.

Where is your annotated list of references?

Here's another site that says the same thing:

"The slave is kept in synchrony with the master in a phase locked loop."

Yup.

Compare with pwm (pulse width modulation) or ppm (pulse position modulation) - I forget what the actual modulation in the clock is called, but it is just another modulation, despite being binary and fixed in amplitude.

Indeed.. in some ultimate sense, perhaps that is the final purpose of a PLL.

Yes, in a typical PLL - however I was considering a more theoretical one where eg the phase offset was known to be positive or negative.

On reflection, is a system where the phases are several full cycles out-of-phase, but where the system over time adjusts the slave to (close to) the actual phase of the master, still a PLL?

On further reflection, I think it must be - so perhaps a better definition might be that the integral of the phase difference remains close to zero over long periods time (while leaving how close and how long as an exercise for the reader) :) .

-- Peter F

rickman wrote on 8/5/2017 11:08 AM:

One other part of the Shortt clock that requires careful thought is the relay and spring that perform the phase detection and correction. The slave pendulum has a leaf spring parallel to the rod and the control relay has a pick which is activated under control of the master gravity lever. The pick can intercept the leaf spring or not, depending on the timing. There is an issue with this which is impossible to eliminate, only minimize and that is metastability. A decision is being made and it can not be done with infinite resolution. So the pick and leaf spring must be designed to minimize the problem, likely done by making the spring thin as possible and making the edge on the pick as sharp as possible.

We see the same problem in electronics when trying to make decisions on the state of an input that is changing.

Just where did I say that ?. Having worked with pll's since the

4046 and earlier, I should know the difference.

That's probably why the Shortt clock is described as a hit and miss system and correction is unipolar, whereas a classic pll continually updates the vco every cycle, not multiples thereof.

Ok, the Shortt clock is probably as close as you can get to a classic pll using mechanics :-)...

Chris

You snipped the part I was replying to but you talked about the master knowing the status of the slave which would only be useful if you were adjusting the master.

"Classic"??? There is no such definition of a PLL to "continuously" update anything.

Yes, because it *is* a PLL. In fact the problem most people have with it is that it doesn't adjust the phase by adjusting the frequency of the slave. It adjusts the *phase* so clearly it *is* a phase locked loop.

All pendulums have circular error where the frequency is determined by the amplitude of swing, so for the half cycle where the phase is adjusted by abridging the swing by the hit of the hit and miss stabiliser, the frequency of the slave is, indeed, changed.

The standard formula given for the cycle time of pendulums ..

2 * PI * root( L / G)

... is only valid for those small angles where sin( theta ) = theta, and such angles are so infinitesimal that no visible movement of a pendulum would be seen!

This just won't go away, will it :-). Here we are, arguing over the semantics of phase locked loops, but the term pll didn't come into wide use until the 1960's, decades after the Shortt clock. I'll continue to think of it as a hit and miss governor, as it was originally described...

Chris

You seem to be confusing two different things

The error you refer to is due to the pendulum not actually taking a direct line between the ends of its travel, the error is small for small amplitudes. There was a famous experiment by a Frenchman in, I think Paris, he hung a huge pendulum and let it trace its path in sand, rather than it going 'to and fro' it actually went in arcs as it went to and fro.

The effect is minimised by reducing the amplitude.

As you correctly say, the frequency of a pendulum is given by the formula you state. If you 'give it a nudge' you may shorted one swing but the overall frequency is still determined by the formula.

The 'nudge' will change the phase of the swing, not the frequency- ie it will shorten one cycle.