A mechanical phase locked loop!

Continuing my googling following last night's BHI lecture, and following up on the Shortt and Hope-jones clocks, here is a mechanical
phase locked loop, and in Meccano! ...
http://www.meccanotec.com/shortt.html
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On 01/08/17 12:00, Gareth's Downstairs Computer wrote:

While the article refers to a 'phase lock loop', it isn't really.. There doesn't seem to be any measurement of error in the slave which is then use use used to 'pull it' to reduce the error- which is how a true phase lock loop works.
The system seems to operate more as follows, the slave is designed to run 'very nearly right'. It receives precise pulses from the master which it will naturally sync to. The same will happen if you have two oscillators on nearly the same frequency if you 'feed' the output of one to the tuned circuit of the other. (Including harmonics.) This is used, for example, by some amateurs to lock radio oscillators to GPS locked references.
Still, it is an clever system and of interest.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Brian Reay wrote on 8/1/2017 7:58 AM:

The Shortt clock *does* make a measurement of the phase. It checks to see if the phase is fast or slow. In one case it invokes a spring that tweeks the phase of the slave. In the other case it does not invoke the spring allowing the clock to continue running unadjusted. The default behavior of the slave clock is to run a bit slow and the adjustments speed it up (or the other way round, I can't recall exactly).
The measurement may be binary and the adjustment is the same, but that does not make it anything other than a phase locked loop.
--

Rick C

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On 01/08/17 13:49, rickman wrote:

Hmm, I half see your point but I'm not entirely convinced.
I'm just not convinced that the description truly 'maps' to that of a true PLL.
I don't doubt that it works nor do I suggest it isn't a very clever bit of design. I'm just not sure about the terms used.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Brian Reay wrote on 8/1/2017 1:19 PM:

Ok, but I don't see what you can be confused about. I believe in electronics this phase detector is referred to as "bang-bang" where it outputs a 1 or a 0. So on every measurement the VCO frequency control signal receives an impulse of one polarity or the other.
The only difference between that and the Shortt clock is the Short clock only has one polarity of impulse and is adjusted to run a bit off so the required intermittent impulses will keep it in phase with the master.
If you are interested in mechanical clocks (the Shortt clock uses electricity to isolate the master and slave even though the master is purely mechanical) you can read about the Fedchenko AChF-3 time piece. It came well after the Shortt clock and not long before quartz and atomic clocks, but was amazingly accurate without any fancy footwork with master slave complexity.
Fedchenko used a compound spring for want of a better name. I've read that it corrects for the parabolic distortion introduced in the timing of a circular pendulum swing. This is a second order effect in that the coefficient in the term is rather small. But in these clocks it makes a difference. The way most clocks correct for it is to keep the amplitude of the pendulum swing as constant as possible minimizing the second order deviation. The Fedchenko clock uses a pendulum spring with two distinct lengths. This causes a different rate of spring over the range of angle. Some descriptions seem to say it actually causes the pendulum to swing in a parabolic arc. Either way it corrects for the second order term in the time equation of the pendulum making it less sensitive to variations in the amplitude of oscillation.
--

Rick C

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
rickman wrote on 8/1/2017 11:59 PM:

Thought I'd mention John Harrison's 'Clock B' too. It was designed 250 years ago, but never built that I am aware of until recently. It has proved to be nearly as accurate as the Shortt and Fedchenko clocks even though it was a much, much earlier design. I don't know any details of why it is so good other than that Harrison took into account every source of error and included a compensating factor to balance it out. I haven't see any further detail. Pretty impressive. Clearly the man was a genius.
--

Rick C

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On 02/08/2017 05:08, rickman wrote:

Oh yes, I recall the B clock- I have an interest in clocks (actually more watches) - and read up on Harrison's history, partly due to his work on clocks / watches directly but also as much of my engineering work was navigation related.
I recall reading of the building of the modern version of the B clock - it must have been in the 70s or early 80s.
As you say, Harrison was a genius- albeit an largely unrecognised / unappreciated one in his own time- at least by the Gov. of the day. I've seen the examples of his work in the National Maritime Museum- the quality is unbelievable, especially when you consider the technology of the time.
--

Suspect someone is claiming a benefit under false pretences? Incapacity
Benefit or Personal Independence Payment when they don't need it? They
  Click to see the full signature.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On 08/02/17 07:19, Brian Reay wrote:

I've had an interest in clocks as well. Working in computing, was interested in the IBM master clocks, which have a Graham deadbeat escapement and either an electrically wound spring, or weight driven mechanism, + an Invar pendulum. Found a mid 1930's example some time ago, which has been running now for about a year. Stripped down completely and rebuilt. IBM claim around 15 seconds a month error, but after rating for a few weeks, it shows an error of less than a second a month. There's noise on the stability, drifting +/- half a second or so from day to day, but was quite amazed at the accuracy of such an old clock...
Chris
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

I think the confusion occurs because at no time, are the phases of the 2 clocks locked together, even at the point of the impulse. By the very nature of the design the phase of the 2 pendulums (or should that be pendula to please Gareth) shift in relation to each other.
In an electronic pll, even one using a bang-bang phase detector, the phases of the 2 signals are locked together, within the constraints of the loop filter.
Jeff
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Jeff wrote on 8/2/2017 5:09 AM:

This is another false dichotomy. The aspect of the Shortt clock you are referring to is that it is *discrete* rather than continuous. So you can clearly see the fact that the slave oscillator is not in perfect lock step with the master (reference). The same is true in *all* PLL circuits. The phase of the oscillator is adjusted by the error signal. There can be no adjustments without error, so the oscillator will not be in perfect lockstep with the reference. It will be within some tolerance... same as the Shortt clock. A PLL can be discrete and the phase will move in patterns with small offsets in frequency at all times. With a continuous phase comparison the frequency will vary continuously but still will not be "locked" to the reference with no error. In fact, PLLs are used to remove short term jitter from clocks by the use of a slow filter on the control signal.
--

Rick C

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Not correct the phases of the 2 pendulums are *never* in phase. Even when a kick is given, as of course if they were in phase there would be no need for a kick.

When a electronic phase lock loop is locked there is no error as the 2 signals are perfectly in phase. There will only be a change in locked control voltage if the phase drifts.

No, a phase locked loop has the same accuracy, or tolerance if you wish, as the reference.

No it will only vary in sympathy with the reference signal, or with signals that are not damped by the loop filter due to being faster than the loop filer can deal with.
Jeff
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Jeff wrote on 8/3/2017 5:32 AM:

You don't understand the meaning of "phase". If you said the two frequencies were never the same I would agree. The slave pendulum runs slower than the master with the intermittent impulse to adjust the phase. The relative phase varies with time as a sawtooth function and so at some point the phase *must* be aligned as the slave passes from being ahead to being behind. On the next adjustment the phase is adjusted or not. When properly adjusted the phase of the slave will only be "bumped" every other adjustment time. On the adjustment times when the slave phase is *not* adjusted the phase will be in alignment ideally.

You need to go back to PLL 101 class. When the PLL is "locked" it simply means the error in phase is small enough that the loop can compensate by varying the VCO frequency. If you understand the math you will see that this means it will *always* hunt for the perfect alignment. If there is no integral term in the feedback loop, there will always be a phase error dependent on the dF/dV slope of the VCO. If there *is* an integral term in the feedback loop the loop will have small fluctuations as the frequency adjusts to correct the phase, but when the phase error reaches zero the frequency error will *not* be zero and the phase error will immediately become non-zero.

There is always jitter in the output of the PLL that is independent of the reference clock.

Please review your PLL materials. There is no such thing as a PLL that aligns perfectly with the reference.
--

Rick C

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Phase is fundamentally linked to frequency.

That is a ridiculous statement, if it were true you could say that any 2 random signals were 'in phase' just because at some point in time they both had the same phase angle.
Jeff
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Jeff wrote on 8/3/2017 1:17 PM:

Not sure if you are referring to the Shortt clock or the PLL. But the statement applies equally to both. There is no magical stability in the PLL. It is a control loop and as such the thing being controlled will *never* remain in phase or at the same frequency as the reference.
--

Rick C

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On 08/03/17 17:27, rickman wrote:

I think the difference is that while a pll always has a phase offset the reference and vco are in phase lockstep once the loop has aquired lock. It's a closed loop system whereas the Shortt clock is an open loop system, only getting a kick back into sync from time to time.
Like a hit and miss governor ?...
Chris
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Chris wrote on 8/3/2017 3:05 PM:

I don't know what you guys are seeing. The two pendulums of the Shortt clock are in lock step. The fact that they are only compared every 30 seconds does not change the nature of the design.
The phase comparison signal from a PLL is typically "grainy" in the same way and has to be filtered to become a control signal. The only reason you say they are in "lock step" is because the grain is very fine. The Shortt clock grain is very fine as well typically adjusting only every other 30 second period.
I guess the difference is the Shortt clock is adjusting the instantaneous phase and the average frequency while a typical PLL adjusts the instantaneous frequency to try to keep the phase aligned. Both will see variations in phase over time.
--

Rick C

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On 08/03/17 21:31, rickman wrote:

I would see the Shortt clock as a frequency locked loop, not the same thing as a pll. Different level of instantaneous precision.
Semantics, semantics :-)...
Chris
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Chris wrote on 8/3/2017 6:33 PM:

Not sure why you say that. What is measured and adjusted is the phase. Either the slave is a bit ahead or a bit behind and it is either spurred on a bit or it is not. The frequency of the pendulum is not impacted other than at the moment of phase adjustment.
--

Rick C

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

What we are seeing is that even after the 30 second 'kick' the 2 pendulums are NOT in phase.
They may well be 'a bit closer' in phase, but the kick just moves the difference a fixed small amount in one direction, which may be sufficient to bring the phases closer, or it may be too much and go through the in phase point. With the design there is no time where the 2 pendulums are *held* in phase.
The design in fact relies on the fact that the phase of the 2 pendulums is constantly changing.
Jeff
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Jeff wrote on 8/4/2017 4:58 AM:

As is true for any PLL.
--

Rick C

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Polytechforum.com is a website by engineers for engineers. It is not affiliated with any of manufacturers or vendors discussed here. All logos and trade names are the property of their respective owners.