Member postings for Joseph Noci 1

John, Is this 'log' plot format what you mean?

I am also reaching the limits of the pendulum's environment now..

In my setup, looking at the bob period index pulse, referenced to the GPSDO 0.5Hz reference, on a 'scope is the quickest way of seeing what the pendulum period is doing.

After many hours of running and adjusting the weights I have it reasonably stable @ 0.5Hz with a 100us or so jitter. However, simply opening a 800mmx300mm cupboard door, 3 meters from the pendulum, causes a 3-400us jump within 1/2 a second of opening the door. The pendulum is in the study ( 2nd storey of the house..), on a thick concrete floor, but walking slowly on the floor 1 meter from the pendulum causes a similar jump - maybe motion, maybe air movement, or both. Also, I have the Atlantic ocean crashing 50 meters from the house...Perhaps Fishing would be a more successful pastime?

6sec in 44 days is rather impressive - 1.6us/sec slow....Not sure my approach will be as good.

Dave, that ADEV graph bothers me - it is shown reversed to the norm, ie, starts hi and drops, stabilises and then turns up again. Also, way down below 10 minus 13...! That's good GPSDO territory and I am afraid, not possible. Even the Shortt is around 10 minus 8 @ 10 ^5 sec.

Unless I misunderstand this curve completely, is looks like the curve you get when you do ADEV measurement on an oscillator using a time reference that is derived from the same oscillator - somewhat incestuous...doing so gives a curve that just keeps going down, almost forever - looks good, but is a lie.

EDIT : is your pendulum running in a vacuum on this test? 

Edited By Joseph Noci 1 on 30/08/2023 22:08:31

Thanks Michael - I read through briefly and will try study in depth, as I still do not follow - I understand the exponential decay - that applies to all oscillators, so in the realm of electronic oscillators, with transistors and coils and capacitors, I am comfortable in predicting and calculating the curve for a set of parts - But I am lost with the pendulum. I don't see how to compute the exponential decay without either knowing Q ( and then I have the decay anyway) or a plethora of other data - pivot friction, air resistance to the bob/rod, etc - all damping factors that would change the curve.

Data analysis is an Art itself. The size of the file and how much space it needs in RAM is no longer an issue these days, but I have yet to find Non-Rockefeller SW tools able to process such large files without needing a shave between runs!

The core sample rate of the ADC data for the angle sensor is 200Hz ( 5ms) - because I eventually want to use that to generate the phase offset for the coil drive, and to be able to adjust the phase in small angles - maybe 0.1deg or better. The Temps/pressure, etc I sample at 2sec intervals, and will probably push that out to 30sec.

You do seem reluctant to stem the Q tide against you - is it not practical on your setup to do a 'normal' run-down data log?

Dave:

I have made best effort to ensure the pendulum is in a stable system. Floor is 16 inches of concrete. The pendulum frame is very stiff and the A frames were welded at the base under sprung stress, so the frame is quite rigid - 20mm x1.5mm wall square tube, with a heavy base.

No draughts during the tests, all leveled up as best I can - Not sure how the rod could bend - only if there were resistance to pivot?

Temperatures, pressure, etc were very constant.

Ambient temp - 26-27deg, Rod : 23-24deg, Bob:22-23deg, Pressure 1014.4mb, Humidity - 44%

The Q's reported in the previous post have an error in the '31.8%' Q computation ( the Q of 14514)

The logged data is a large file, and during the latter portions of that log, 'someone' must have bumped the pendulum and buried in the data I found this:

That jump up in Q was to suspicious and when digging minute by minute in the file I found what had happened, so that Q computation is invalid.

I ran the test again today - twice, with consistent results both times , using all the variations of Q computations already discussed :

This is the plot of the last run-

It ran for 3hours 33minutes, with sample rate @ 20ms.

Q's computed :

@ 61% - 1940 beats * 2 * pi - Q = 12200

@ 50% - 2850 beats *4.5324 - Q = 12900

@ 36% - 4100 beats * pi Q = 12880

@ 21% - 6313 beats * 2.013 Q = 12714

There is a small improvement in Q from the 50% amplitude reduction onwards, and not what was indicated in the erroneous post. This is encouraging as it does indicate that the aerodynamic effects are not hugely impacting the bob, since there is little improvement at the lower amplitudes ( less moving air mass, but not a great Q improvement).

As more data becomes available for analysis, more improvements are made, and the cycle repeated...I have now discovered an oscillation in the stable system condition( pendulum running with amplitude loop locked) - logging the angle sensor data shows a variation of the peaks of the 0.5Hz period. It is very small - the peak to peak voltage is 1944mv, and the oscillation is 4mv. The peak angle is 1deg, so around 1000mv/deg. 4mV = 4milli degrees!

The total control loop time constant is around 60minutes, so it seems impossible that the pendulum is being forced to oscillate like this.

I do suspect it has to do with aliasing during the sampling of the voltage by the DAC.

When I sample at 200Hz (5ms), the oscillation period is 12 seconds ; at 20ms sample rate ( 50Hz) the period is 15 seconds - I suspect a software sampling cycle that is being interrupted by serial comms at a rate that causes a repeating delta in the sampling rate, and that manifests as an aliase...What fun.

Thank You, Michael.

However, I am at the limit of my bounds on this - I don't know what to do next - so putting the Pendulum to sleep for a while!

Incorrect use of words...

This plot is 2-1/2 hours of data at 50Hz. The data was analysed at the points where the amplitude had fallen [by]:

61% - Q = 12000

50% - Q = 12400

31.8% Q = 14514

Meant to say :

This plot is 2-1/2 hours of data at 50Hz. The data was analysed at the points where the amplitude had fallen [to]:

Did a few runs today and logged data to calculate Q - Not impressive, but I think definitive.

The Processor reads data from - Angle sensor (sinusoid), Angle Peak detector, and all the temperature sensors, humidity and pressure. The angle and peak detectors are sampled at 200Hz and can be logged at 200Hz, 100Hz, 50Hz, 20Hz and 10Hz. 200Hz fills large files fast...

With the pendulum running and the control loop stable, the drive voltage to the drive coil was then disconnected and the pendulum angle data logged at 50Hz.

This plot is 2-1/2 hours of data at 50Hz. The data was analysed at the points where the amplitude had fallen by:

61% - Q = 12000

50% - Q = 12400

31.8% Q = 14514

50Hz data log of pendulum sinusoid amplitude over time: 

It is interesting that the Q increases with extended run-time. The amplitude of the swing at the start was 2deg full swing - Q seems to increase usefully with reduced swing angle. I will try another run with a 1.5deg full swing angle from the start to see if this is so.

The sine wave from the angle sensor looks good, sampled at 200Hz. The sample artifacts at 50 Hz are quite clear in the posted examples below. The 200Hz sampled waveform will be used to generate the digital 90deg phase shift to the coil drive, in due course.

200Hz sampling and logging

50Hz Sampling

With the 'knife-edge' pivot and efforts I put in on the Bob format, etc, I would have thought I should achieve a better Q than it appears, although my expectations are/were biased by what others have been saying they have achieved - I will placate myself somewhat in the belief that the Q figures bandied about have been very much over-estimated in light of the recent discussions on ways to calculate Q and methods to do so.. I think the method I used is definitive and is very repeatable and should enable quantitative evaluating of the effects of modifications and alterations to the system.

Edited By Joseph Noci 1 on 29/08/2023 22:53:20

John,

The Sync detector filter is a low pass with cutoff @ 1KHz. The angle sensor runs at 5KHz, so 1KHz cutoff is OK and the phase shift on that signal is very low in the 0.5Hz domain.

I read in detail the Bigelow texts re his oscillator, also the details in Matthuys - studied and modelled his circuits in Simulink, etc, but I believe the concept and approach is flawed - clipped sine wave will always cause perturbations - The 'thermal' analogue multiplier has interesting potential - rids signals of some noise issues, but not sure it is worth the effort. In addition, that concept provides no means of adjusting phase if in error/needed.

You said -

The op-amp GBW was 1 MHz, so there was a small excess phase shift at 0.5 Hz which meant that the impulse wasn't quite in phase with the pendulum.

Did you mean impulse? I guess so, since the clipped sine is more 'impulsive' - What I have tried to implement is really a 'pure' sine, at least as pure as the pendulum swing, with the coil current always in tune with the pendulum position and velocity - it can never differ, only the lead/lag can differ by changing the 90deg phase shift, and that is the ideal control point input.

The op-amps I used are 2MHz GBW, except for the differential instrument amp after the Sync detector - that is 500KHz. However, any build up of phase shift become unimportant as the 90deg shifter can be adjusted to shift by the amount needed to lock the loop. The peak detector voltage is monitored and that will stabilize while the final PID loop ( too be implemented still!) controls either to VGA gain, or the phase shifters setpoint.

OP amp choices are always a trade-off - I want really low noise and really good input offset performance, but that tends to oppose higher GBW, so overall system compensation is the way to go..

My 90Deg shifter is purely analogue right now - an all pass filter with 90deg point set at 0.5Hz. I will most likely implement this digitally - either a FIR all-pass or Hilbert transform - for concept testing right now the analogue all pass works just fine! It will have shift drift due to the tempco of the C's in the filter, but I believe all tempco's can be compensated for at system level, ie, an overall temp compensation, rather than compensating each circuit element.

If you want details of the angle sensor, I am happy to give my drawings and circuits - Do you have a desktop CNC router? Easy then to make the capacitor elements with FR4 PCB.

Absolutely amazing how Nature builds structures that challenge and overwhelm all man's scientific methods! That wing is gossamer thin and as strong as steel, but weighs nothing...Nice photo Sam.

I have/had all sorts of ideas for the bob and rod - as you are suggesting, but modelling such concepts is difficult and at pendulum velocities, involves lots of guesswork and many assumptions. To obtain definitive results an approach would be to use a functional, stable pendulum, and then make such a rod, for example, and see its effect on the performance. The problem is always, digging out the delta from all the other delta's - temp, humidity, etc. A road to Hell...and it takes a long time...

Edited By Joseph Noci 1 on 29/08/2023 07:22:31

My initial attempt at controlling to a fixed amplitude has met with some success and promise.

I used the angle ( = amplitude) of the pendulum as given by my angle sensor as reference, phase shift that sinusoid by 90deg and feed that to the coil driver, and control the level of that voltage so that the level of the voltage from the angle sensor remains constant.

A block diagram of the electronics:

A Sine generator generates a sine wave at 5KHz and feeds a 600ohm-600ohm transformer which drives the differential angle capacitor sensor. The pickup sensor of said device is buffered and feeds a synchronous detector, driven by a reference Sine from the sine generator. The resulting DC output is amplified differentially and filtered. This resulting voltage is a representation of the pendulum angle since the senor is fitted to the pivot.
This 0.5Hz sine wave feeds an ADC read by the Nucleo processor, for later processing.

The 0.5Hz also feeds a 90deg phase shifter. The shifted phase is required since the pendulum drive is via a drive coil atop the pendulum, above the pivot, and is fed with a sine wave drive per swing, all the time. The 90deg shift ensures that the sinusoidal drive voltage is at its peak when the pendulum is at BDC, and minimum when the pendulum is at the extremes of swing. There are no impulses, just a good sine wave drive, and I hope this will result in almost no pendulum perturbation.

The driving voltage level is maintained within a closed gain control loop - the pendulum angle sinusoid feeds a Peak detector, which generates a DC voltage equal to the peak of the voltage of the pendulum swing extremes. This DC voltage controls a voltage Gain Controlled amplifier. The input to the amplifier is the 90deg phase shifted sinusoid, and the output is that sinusoid, level controlled , to ensure that the pendulum swing angle is maintained. An increase in angle( amplitude) increases the Peak detector output and the VGA gain is reduced, reducing the drive sinusoid voltage to the pendulum drive coil. Reduction of swing is a reduction of the DC output of the Peak detector, causing an increase of Gain in the VGA, and increased drive to the pendulum coil.

The RMS current in the drive coil is 0.14mA when the loop is stable. The angle sensor Peak to Peak voltage output for a 2 deg total pendulum swing is 2 volts PP. This voltage varies by less than 0.2mV over 3 hours with the loop stable.
The drive voltage from the coil driver is 1.2V PP, and feeds the 10ohm coil via 3000ohm series resistance.

The Peak detector output is summed with the output from a 16bit DAC. The DAC output will be used to provide a voltage offset to the VGA, causing an increase or reduction of the swing amplitude. The intention is to perform temp/pressure/humidity compensation is software, and to feed corrections via the DAC to the VGA.

Presently there is no microprocessor control at all. There is a Time to Digital convertor that measures the time in nanoseconds from the rising edge of a 0.5Hz reference pulse from my GPSDO, and the rising edge of the pulse from the pendulum BDC sensor ( an opto interrupter). This will be used to Phase lock the pendulum to the GPSDO 0.5Hz pulse initially, by control of the DAC, breaking the Peak detector loop. When phase locked to the GPSDO reference, the uP will
log the variation of DAC control voltage to maintain lock, as well as logging Ambient temp, pressure, Humidity, bob temp, and rod temp.

I hope to find a way to implement some sort of amplitude compensation using all that data as reference input, at which point the pendulum will be released again and be free running, with only the amplitude loop controlling it, and the uP performing environmental corrections only.

The phase shifter is also able to be voltage controlled. If the amplitude control of the coil drive voltage does not achieve the control required ( since period is largely independent of amplitude) then the VGA loop will control amplitude, and the DAC will be used to adjust the drive voltage phase, allowing a lag or lead drive voltage to slow or speed up the swing.

A lot of electronics and a lot to experiment with....quite enjoyable.

Starting some logging now to see how the phase lock to GPSDO performs, and now that the ADC is in place, I want to do a few run-down swings to try calculate Q...

Some photos of the the growing electronics.

Edited By Joseph Noci 1 on 29/08/2023 00:06:05

Not what I said at all -

If you inserted a brass tube 25% the length of the bob, it would bring the bob suspension point to its midpoint, and being brass, the bob would then expand evenly up and down again. This allows you to play with the expansions amounts up and down. The inserted tube can be of materials that grow or shrink with temp, giving more freedom to compensate for expansion in the total system.

Brass brings you back to net zero, using other materials, steel, quartz, carbon fibre, aluminium, etc allows further compensation which is difficult to apply if the bob is suspended at midpoint and it is much easier to change a compensation tube than moving the bored hole at midpoint...

But, many ways to skin a cat...I like to follow sound concepts proposed by those skilled in the art, rather than inventing new ways to make the pendulum behave obtusely...

The cylindrical bob, stood on a table would expand upwards from the table when heated. If the bob is suspended in its middle, the CG, assuming a solid cylinder, then it would expand evenly up and down from that point, ie, expansion will be above and below the suspension point.

So if you attached the brass bob to the invar rod, only at the middle of the bob, bob expansion would have 'no' effect of the pendulum rate. Since the invar rod also has a coefficient, moving the bob suspension point up or down from the midpoint of the bob can compensate for that.

Placing that suspension point around 75% upwards in the bob allows you to insert a tube between the bob midpoint, and the bob suspension point. If you inserted a brass tube 25% the length of the bob, it would bring the bob suspension point to its midpoint, and being brass, the bob would then expand evenly up and down again. This allows you to play with the expansions amounts up and down. The inserted tube can be of materials that grow or shrink with tem, giving more freedom to compensate for expansion in the total system.

Something like this: ( the adjusting nut is just for completeness - not a practical implementation)

I think the brass bracket is structurally sound - I was referring to the 8020 Al profile section - I presume its a seconds pendulum and if the Al profile is the length of the pendulum or longer, it will tend to flex or rather twist with each cycle - your bob would be around 6 lbs so not trivial. I don't think there is a practical way to brace that profile - it is flat without sufficient dimensions ( as in 2 or 3..) to cross brace to. You would need to bolt it to a resilient surface.

For temp compensation - as you have an invar rod, the bob is the main item of expansion in this setup - bore into the bob about 2/3 up from the bottom, and make that the suspension point on the invar rod - As you know the idea is to get the bob mass to expand evenly up and down from the suspension point , so that the CG remains at the suspension point. To fine tune you can add short brass or steel ( or quartz..) sleeve between the suspension point and the bob to increase the upward expansion by different amounts.

If I am preaching, shut me down!

Thank You.

Can I copy an entire album to my PC? Or can I only copy/download photos already embedded in posts?

Sorry if this has been covered - I cannot find it in the confusion...

Well made SK...

There are always a few haunting thoughts when playing around with this stuff. Pendulums are an easy subject till you start reading, and the more tomes read, the less easy it becomes...Of course it's only hard if you are chasing that deep trough on the ADEV plot, but why else would we be making such complex structures for a simple pendulum?

So, in that vein of the extreme..

A 3 inch Bob swinging 2 inches from the rear structure will suffer some aerodynamic effects - esp a round bob - a rugby ball bob a little less, but the displaced air has to go somewhere, and it has nowhere to go between the bob and support. This has been evaluated in the literature with bob's in a clock case, etc..

Not sure what your bob weighs, but I suspect there will be some flexing of that vertical support, or rather a twisting, and even if only a few 100'ths, its enough to wreck the ADEV...Symmetry of support and structure is your friend in this, with diagonal struts and/or bracing a good aid.

Your have not indicated how the 8020 profile will 'mount' - if bolted well to a good wall, then flexing should be eliminated.

John commented on my cross support with the V's - - there is also good coverage on the concept in Matthys -

The friction between the rod and V is an issue and recommendations are a V of about 120deg angle , with the rod or pins riding the V being 3 or 4 mm diameter, hard and polished - also that the V should be of steel, polished if possible, but certainly very smooth, no machining or sanding marks. All in an attempt the get the pendulum hanging truly vertically. I have a 4mm polished pin and and mild steel polished V's and to the best I can measure, about 0.04 to 0.1 mm hysteresis when the pendulum bob is gently swung front'back and comes to a halt. I think 0.1mm is not great and will probably affect my ADEV in the end.

I will most likely get rid of the V block, and replace with flats and a spring blade pressure clamp, increasing pressure once the pendulum has swung for a while and 'stabilised' .

A +-1deg swing front to back should also dampen to a halt within 10 min or so, so knife edge supports in the forward plane are a no-no.

I would propose that it is a straight oblate spheroid, but with a missle nose type tip so that the air is separated cleanly rather than compressed and then squashed aside - but how long is a point, and where around the oblate does it start...I think just a point will do better than none.

As to ratio - I have seen proposed ( I am digging to find the reference again..) that the ellipse should fit inside a Golden Rectangle = Golden Ellipse