Experimental Pendulum Clock

[John HaineParticipant @johnhaine32865]

Which one have you ordered please Dave?

[Joseph Noci 1Participant @josephnoci1]

Just tested on my old test box – It has a Neo-7M in it – an old UBLOX device. I used U-Center vers 8.13.

Connect via serial port to Ublox. On U-center display, lower of top left, select and set port number to the PC serial port to use. Right of that selection is the baud rate select – set to 9600.

The port symbol to the left should be green indicating serial comms. The displays on the right should fill with SAT info, if the GPS can see Sky. Oh, by the way, it must see sky to get SATS and a lock, to discipline the NCO…

The select VIEW, then Configuration View.

Then down the left, scroll till you see TP and TP5 – TP=Time Pulse.

TP = 1pps output, normally fixed, cannot modify.

TP5 shows a frequency setting – in Hz – enter what you want there, and click on 'SEND' lower left. Observe the TP output pin on your chosen GPS on a 'scope – it should show a clock.

My test Neo-7 gave stable clock @ 1,2,3,4,6,8 and 12MHz. Anything else had small to huge jitter. Greater than 12MHz has huge Jitter. 1,2,3,4,6,8,12 all go into 48 nicely, but nothing to do with powers of 2 – go figure….

EDIT – 24MHZ also works fine…

 

Edited By Joseph Noci 1 on 15/02/2023 14:53:48

[SillyOldDufferModerator @sillyoldduffer]

Posted by John Haine on 15/02/2023 13:29:51:

Which one have you ordered please Dave?

It's a Pi-Hut uBLOX MAX-M8Q Breakout for Active Antennas – 5V with Serial/I2C Level Converted version. The 3volt version is a bit cheaper. Note board needs an external active antenna if you haven't already got one.

I hope no-one uses their Google-foo to find a cheaper alternative! I spent a frustrating 30 minutes on the web looking for the best deal and failed. The last straw was finding the best buy was in Australia and they were out of stock.

Other annoyances, the right chip sealed inside it's own antenna with no pulse output on the connector and breakout &/or drone boards with the same problem. I didn't want to buy one of those in hope the pin was accessible, only to find it's not.

Dave

[Joseph Noci 1Participant @josephnoci1]

Dave, cancel your order…..

Look at the data sheet of the device before you choose – There should be 2 Timepulse pins, one is labeled Timepulse-1PPS, the other Timepulse only. – the latter is programmable to 'any' frequency

The NEO 6, 7, 8, M8 series all seem to have the two. Also the PAM-7Q

The MAX-7 does, the MAX-7Q does NOT.

NOT the SAM-M8Q, MAX-M8Q

[SillyOldDufferModerator @sillyoldduffer]

Posted by Joseph Noci 1 on 15/02/2023 15:13:36:

Dave, cancel your order…..

Aargh! I'll have another go…

Ta,

Dave

[John HaineParticipant @johnhaine32865]

I looked at the Neo/Lea M8T datasheet, it seems to say that the max frequency programmable is 10MHz.

1.8.2 Timepulse and frequency outputs The NEO-M8T and LEA-M8T modules provide two time pulse outputs that can be configured in rate from 0.25 Hz up to 10 MHz by message CFG-TP5. Time pulse alignment can be configured to UTC or GNSS time according to the standard used in signals being received or to an alternate standard where inter-standard calibration data is available (from the signals themselves or by aiding). The time pulses are generated on edges of an asynchronous clock; for pulse rates below 2 Hz, the exact phase of the TIMEPULSE output is reported before each pulse in the TIM-TP message. ☞ Times reported in navigation messages such as NAV-PVT report the time of the preceding pulse.

However, it also has this intriguing paragraph:

1.8.3 Time mark The NEO-M8T and LEA-M8T modules can be used for precise time measurements with submicrosecond resolution using the external interrupt pins (EXTINT0 and EXTINT1). Rising and falling edges of these signals are time-stamped to GNSS or UTC time, counted and the results reported in message TIM-TM2. The reference time is the same as set for TIMEPULSE with CFG-TP5.

Which means that the device will do the whole job I think!

[Joseph Noci 1Participant @josephnoci1]

The problem with that, John, is the 'sub-microsecond resolution' – the time stamping appears to be done by the GPS processor, not in hardware, and so the resolution has been shown to vary from 80ns to over 600ns – that's still 'sub-microsecond'..

Re the 10MHz max – as I indicated, all is not fact in the data sheets – 24MHz works well. The 'T' GPS's TP5 also jitter badly at the non 'divisor' rates..The 'T' devices however have better jitter performance on the 1PPS than none T devices ( which suffer the usual plus/minus 20-25ns jitter and associated walking bridges..)

You don't need a T device for this exercise though.

Maybe we should start a seperate thread on all this as the pendulum thread should not be contaminated with pseudo-times nuts…

 

 

Edited By Joseph Noci 1 on 15/02/2023 18:59:37

[SillyOldDufferModerator @sillyoldduffer]

The replacement 7490 chip arrived so I plugged it, thanking the deity I'd used a socket, and it worked. This circuit:

So have a box that outputs accurate 10MHz and 5MHz signals. (1MHz also available if I take the lid off and move a jumper.)

Next problem is to tune it as close to 10MHz as I can get. It's not far off, but I'd like it to be good to a couple of parts per 10,000,000.

No luck so far finding an affordable and available Ublox GPS breakout board with access to both TP and TP2 pins.

And my clock is still misbehaving badly, drifting nearly a minute in 12hours. Crystal clear I need to get on and make a decent suspension.

Dave

[Joseph Noci 1Participant @josephnoci1]

Posted by SillyOldDuffer on 15/02/2023 19:09:51:

 

So have a box that outputs …….accurate …….10MHz and 5MHz signals. (1MHz also available if I take the lid off and move a jumper.)

Next problem is to tune it as close to 10MHz as I can get. ……..It's not far off,………… but I'd like it to be good to a couple of parts per 10,000,000.

Dave

You will awaken the Pedant in me with such statements Dave! Accurate , and not far off, and good to a couple of parts….all in the one breath!

By the way, the 5MHz will have horrid , regular jitter on it…

 

I found a NEO-6M at a few shops in Oz – tween 13 and AU$30

Edited By Joseph Noci 1 on 15/02/2023 19:41:36

[Martin KyteParticipant @martinkyte99762]

Posted by SillyOldDuffer on 15/02/2023 19:09:51:

The replacement 7490 chip arrived so I plugged it, thanking the deity I'd used a socket, and it worked. This circuit:

So have a box that outputs accurate 10MHz and 5MHz signals. (1MHz also available if I take the lid off and move a jumper.)

Next problem is to tune it as close to 10MHz as I can get. It's not far off, but I'd like it to be good to a couple of parts per 10,000,000.

No luck so far finding an affordable and available Ublox GPS breakout board with access to both TP and TP2 pins.

And my clock is still misbehaving badly, drifting nearly a minute in 12hours. Crystal clear I need to get on and make a decent suspension.

Dave

Do you not like power supply decoupling capacitor Dave.
Only asking but you will potentially get some noise from the NAND gates switching on the 5V you use to derive your Voltage Reference.
regards Martin

[SillyOldDufferModerator @sillyoldduffer]

Posted by Martin Kyte on 15/02/2023 20:23:47:

Posted by SillyOldDuffer on 15/02/2023 19:09:51:

The replacement 7490 chip arrived so I plugged it, thanking the deity I'd used a socket, and it worked. This circuit:

So have a box that outputs accurate 10MHz and 5MHz signals. (1MHz also available if I take the lid off and move a jumper.)

Next problem is to tune it as close to 10MHz as I can get. It's not far off, but I'd like it to be good to a couple of parts per 10,000,000.

No luck so far finding an affordable and available Ublox GPS breakout board with access to both TP and TP2 pins.

And my clock is still misbehaving badly, drifting nearly a minute in 12hours. Crystal clear I need to get on and make a decent suspension.

Dave

Do you not like power supply decoupling capacitor Dave.
Only asking but you will potentially get some noise from the NAND gates switching on the 5V you use to derive your Voltage Reference.
regards Martin

Oops. Missing from the diagram, but fitted to the real one. There's 100uF across the supply plus a 0.1uF decoupling pin 14 of the 7400 to ground, and another from pin 5 of the 7490 to ground, both short leads.

Well spotted!

Dave

[John HaineParticipant @johnhaine32865]

Joe, why should the divide by 5 jitter? Easy to make a synchronous /5 circuit where the output changes state every 5th edge, there's no reason why it should jitter surely? Existence proof: have a 3-bit register clocking the output of a 3-bit adder, force the adder output to be all 0s when the register state is binary 4, then the next state will be binary 0. So the register value cycles through 0, 1, 2, 3, 4, 0…. on successive clock edges.

[Martin KyteParticipant @martinkyte99762]

Posted by SillyOldDuffer on 15/02/2023 20:49:18:

Posted by Martin Kyte on 15/02/2023 20:23:47:

Posted by SillyOldDuffer on 15/02/2023 19:09:51:

The replacement 7490 chip arrived so I plugged it, thanking the deity I'd used a socket, and it worked. This circuit:

So have a box that outputs accurate 10MHz and 5MHz signals. (1MHz also available if I take the lid off and move a jumper.)

Next problem is to tune it as close to 10MHz as I can get. It's not far off, but I'd like it to be good to a couple of parts per 10,000,000.

No luck so far finding an affordable and available Ublox GPS breakout board with access to both TP and TP2 pins.

And my clock is still misbehaving badly, drifting nearly a minute in 12hours. Crystal clear I need to get on and make a decent suspension.

Dave

Do you not like power supply decoupling capacitor Dave.
Only asking but you will potentially get some noise from the NAND gates switching on the 5V you use to derive your Voltage Reference.
regards Martin

Oops. Missing from the diagram, but fitted to the real one. There's 100uF across the supply plus a 0.1uF decoupling pin 14 of the 7400 to ground, and another from pin 5 of the 7490 to ground, both short leads.

Well spotted!

Dave

Ah, thought that might be it. I really only commented ‘pour l’education des autres’ or whatever it is in French.

regards Martin

[SillyOldDufferModerator @sillyoldduffer]

Posted by Joseph Noci 1 on 15/02/2023 19:28:17:

Posted by SillyOldDuffer on 15/02/2023 19:09:51:

So have a box that outputs …….accurate …….10MHz and 5MHz signals. (1MHz also available if I take the lid off and move a jumper.)

Next problem is to tune it as close to 10MHz as I can get. ……..It's not far off,………… but I'd like it to be good to a couple of parts per 10,000,000.

Dave

You will awaken the Pedant in me with such statements Dave! Accurate , and not far off, and good to a couple of parts….all in the one breath!

By the way, the 5MHz will have horrid , regular jitter on it…

I found a NEO-6M at a few shops in Oz – tween 13 and AU$30

It's a fair cop, guv'nor. Slap on the bracelets.

Disaster if 5MHz has jitter, but will it? 5MHz is 10MHz ÷2. It's the 1MHz signal I'd expect to jitter, due to ÷5.

There are various cheap NEO boards in the UK, but I'm not sure they're fit for purpose. This ebay example has no PPS breakout, and this one has PPS, but no TP2

The datasheet suggests no TP2 on the chip either:

Anyway, I've been reading the clear as mud ublox M8 Receiver Description page 270 and wonder if the UBX-CFG-TP5 message, with payload, allows TP2 to be soft swapped with TP1 when only one timepulse pin is available? So it might allow pps or a user frequency. Wishful thinking I expect.

Dave

[duncan webster 1Participant @duncanwebster1]

This is all getting very heavy for poor souls like me. Can we go back several stages. I understand the need for an accurate 10+Mhz crystal (or whatever) and a processor to count the ticks. What does the RPi do? Just data logging? If so why can't it just send data straight to a laptop?

[Michael GilliganParticipant @michaelgilligan61133]

Posted by duncan webster on 16/02/2023 00:54:14:

.
This is all getting very heavy for poor souls like me. Can we go back several stages. […]

.

I’m with you there, Duncan … or more likely several chapters behind you.

I am writing at this ungodly hour, because my sleep has been disturbed; so this is probably drivel, but here goes

Back in Harrison’s day it was ‘easy but laborious’ to check the performance of a clock … because the baseline reference was the motion of the Heavens : A simple ‘sight’ on one’s window-frame was sufficient to detect the transit of a star, and calibration was therefore done in terms of the Sidereal Day [wobbly, but real and comprehensible].

However … Mankind has asserted a ‘definition’ of the Second which turns all this upside-down and instead of a Sidereal Day, our baseline reference is 1/9192631770 of a Second … This is incomprehensible in human terms, but it is what has been agreed.

So now, we are measuring the wavering performance of a notionally simple mechanical device with a digital resolution that is orders of magnitude finer. … Does this lead to better understanding of the pendulum? … I am really not sure.

Short of a massive win on the Lottery; it seems that the best any of us can do is hitch a ride on the GPS system and use it to ‘Discipline’ a local oscillator … typically having a frequency of 10MHz, but with ‘whatever’ instantaneous variation between cycles.

I am reminded of the old joke about the broken pocket-watch, which is exactly right twice a day.

Not having the expertise to even estimate the ‘Uncertainty of Measurement’ in any useful sense, for a real-world hobbyist system, I will just sit back and follow Dave’s exploits with admiration.

MichaelG.

.

Edit: __ removed underline from the word estimate 

[to avoid any implication that it might be hyperlinked]

Edited By Michael Gilligan on 16/02/2023 05:52:18

[Joseph Noci 1Participant @josephnoci1]

Posted by John Haine on 15/02/2023 20:51:50:

Joe, why should the divide by 5 jitter? Easy to make a synchronous /5 circuit where the output changes state every 5th edge, there's no reason why it should jitter surely? Existence proof: have a 3-bit register clocking the output of a 3-bit adder, force the adder output to be all 0s when the register state is binary 4, then the next state will be binary 0. So the register value cycles through 0, 1, 2, 3, 4, 0…. on successive clock edges.

Like I suggested, perhaps better to move this stuff to another topic? John, the 7490 divide by 5 section is a BCD MOD-10 counter – no fancy adders, etc – it counts 4 and then resets on the 5th clock, so the output is not symmetrical but has a 4:1 ratio…

Dave – I beg to borrow your Forum Handle for the day please – I have been sprouting so much BS I am ashamed…

There is no TP2 pin on any of the GPS's – I have misread all data sheets and spoken C..p!

There is only 1 pin, called timepulse, and it can be set to 1PPS ( mostly the default on all GPS's) or to anything else by means of the UBX-CFG-TP5 message as you describe – ON SOME, not ALL GPS's – so you have to dig deeper in the data sheet to see if its possible..If is indicates that the default is 1pps, but to refer to the RX description for more info, it can do it.

All -7 and -8 series seems to be able to do it. Only the 6T seems able, so my neo-6 suggestion is useless.

The easy way to set the TP5 message is to use the U-center SW tool. If the chosen GPS has FLASH, you can program it in hard and from next power on it is always there…

Your wishfull thinking is correct, valid and true!

AND LASTLY:

Mixing terms with brain in neutral…

Disaster if 5MHz has jitter, but will it? 5MHz is 10MHz ÷2. It's the 1MHz signal I'd expect to jitter, due to ÷5.

In mitigation, John missed this as well… I meant the divide by 5 output, not the 5MHz output…The div by 5 will always have non-regular mark-space ratio, see above…

Going to sit in the corner looking at my navel with a mirror for the rest of the day

[Joseph Noci 1Participant @josephnoci1]

Dave, do you have a frequency counter with variable timebase/gate – say to 10 sec maybe AND with an external ref clock input ( normally 10MHz)

Edited By Joseph Noci 1 on 16/02/2023 06:21:16

[HowiParticipant @howi]

have a look at this "https://www.clearskyinstitute.com/ham/gps/"

Relatively cheap to make and works well.

[SillyOldDufferModerator @sillyoldduffer]

Posted by Joseph Noci 1 on 16/02/2023 06:20:36:

Dave, do you have a frequency counter with variable timebase/gate – say to 10 sec maybe AND with an external ref clock input ( normally 10MHz)

…

Phew, a relief to read your corrections thanks! What the ublox chips do is wonderful but I find their documentation confusing; not clear which variant they mean. Just as well because Pi Hut confirmed the board I ordered was in the post before I was able to cancel it.

I have such a frequency counter, an elderly Racal, and see where you're going. I thought of something similar last night, which is to divide the GPS PPS output frequency by, say 4, with a chip, and count how many pulses the OCXO delivers in 4 seconds. Then adjust the OCXO to get a count of 40,000,000 exactly. A picPET will do this easily, and my ardPET could be finagled too. Or the right kind of frequency counter. I'm trying, as far as possible, to do the project with affordable off-the-shelf modules so that anyone else can do the same without a houseful of lab equipment.

Apologies to Duncan and Michael who are having trouble following the thread. Not their fault! This project has developed into a war and there are three different major battles raging at the moment, plus trouble brewing elsewhere,including the current diversion into electronics, and my need to explain what the hell is going on. As the battles overlap and are reported in parallel, the fog of war descends!

The three main battles are:

1. Building an accurate clock and adjusting it to keep good time. This one is about improving the bob, frame, suspension, sensors and the software that impulses the magnet and converts the pendulum's tick/tock events into meaningful human time, specifically UTC.
2. Measuring the running clock to identify problems and prove it really does keep good time, on two fronts:
   • Long term timekeeping is fairly easy to do by comparing the output of the clock to an another accurate time source. Could be the pips, or a decent wristwatch! As it's convenient for computer logging, I'm using bog-standard NTP, which is always within 100mS of atomic time, and usually better. NTP can be improved by disciplining it locally with a GPS receiver (or radio frequency standard like MSF), in which case the time will be within 'a few microseconds' of atomic time. Or I could use GPS directly. Choices, change and confusion are likely.
   • Ultra-short term timekeeping is also important. I want to measure tiny differences between each beat of the pendulum to determine how the bob is effected by Impulse power, and changing air pressure, temperature and humidity. Also, more subtle changes such as the tide (due to how the sun, moon, and earth align in space), micro-gravitational changes (could be groundwater movements), and earth tremors. (John H was able to show his pendulum reacted to the recent Turkish/Syrian earthquake.) This aspect of the project needs a different accurate time standard, one that measures very short time periods, microseconds, nanoseconds, and ideally picoseconds. This is where the OCXO and Ublox technologies come in.
3. Computing! There's a lot to explore and get wrong. Statistical tools are needed to extract information from a mass of data, and maths and statistics aren't in my comfort zone. The clock's microcontroller has to be programmed to drive the pendulum, read the sensors, compute human time, and send data for statistical analysis. The data is collected on a RaspberryPi, programmed in a different language to collect and timestamp it, and periodically copied to my far more powerful workstation which is programmed to crunch the numbers and draw graphs etc, Getting the computing right is important because it's where success and failure in the other endeavours come to light, often in confusing form.

So quite complicated to follow, for which I apologise. Be much clearer if I wrote a logically presented history, because most of the noise and diversions would be edited out.

Although I've won many battles on the way, it's not clear I'm going to win the war! Disappointing news from the front-line and some concern my overall strategy is wrong…

Dave

[SillyOldDufferModerator @sillyoldduffer]

Posted by Howi on 16/02/2023 09:50:52:

have a look at this "https://www.clearskyinstitute.com/ham/gps/"

Relatively cheap to make and works well.

Thanks for that Howi! The way Mr Elwood measures the DDS error frequency is identical to how I measure the error of my Arduino crystal, and closely related to the way a picPET works. And there's more. He explains how he sets accuracy very accurately by beating the GPS disciplined DDS signal against WWV, to get audio tones that can compared with FFT in SpectrumLab. He also gives the necessary maths. Excellent, a recipe I can follow without doing much work. The QEX article is here.

I have all the bits needed to build one apart from the DDS, and they're widely available on the web for under £10 each.

I've suddenly gone from only a hazy idea about how to generate an accurate frequency to having 3 good options.

Cheers,

Dave

[David TocherParticipant @davidtocher94033]

SOD

As I understand it you are, amongst many tasks, correcting your clock errors by stats analysis of the various environmental effects, some of which are correlated. One issue might be that they are correlated over the short term but in a different way depending on the season.

I haven't done any investigation but, for example, high pressure in winter brings artic air over the UK but in the summer the air comes from the tropics.

A more philosophical point; by using stats and an external precision time reference then your stats model hopefully corrects for all of the variation of your clock. You are using a complicated way of locking your clock to the external clock

[Joseph Noci 1Participant @josephnoci1]

Sorry Chaps, I will not add any more to the confusion I have already created!

[SillyOldDufferModerator @sillyoldduffer]

Posted by David Tocher on 16/02/2023 10:40:03:

SOD

As I understand it you are, amongst many tasks, correcting your clock errors by stats analysis of the various environmental effects, some of which are correlated. One issue might be that they are correlated over the short term but in a different way depending on the season.

I haven't done any investigation but, for example, high pressure in winter brings artic air over the UK but in the summer the air comes from the tropics.

A more philosophical point; by using stats and an external precision time reference then your stats model hopefully corrects for all of the variation of your clock. You are using a complicated way of locking your clock to the external clock

Not guilty of the last point: the clock's characteristics are determined by measuring them over a long time, longer the better. I measure:

• The period of the pendulum relative to an accurate external clock, then
• How that period varies with temperature, impulse power, humidity, and air pressure

The results of the analysis are fed into the clock, which then runs independently, except the pendulum's period is compensated for temperature and pressure. It's not locked to an external clock during normal running. At the moment, although the GPS module is switched on, the clock is ignoring it, and the PPS input is disconnected.

However, you're right about 'might be that they are correlated over the short term but in a different way depending on the season'. The quality of the statistical analysis depends on it having many samples covering a wide range of temperature and pressures. Not too concerned about pressure, because the weather has provided several highs and lows, but – it being winter – temperature indoors has been below average, and ranged only between about 10 and 20C, mostly about 15C. The sample is limited compared with the range of temperature the clock will be subjected too over several years, and might be skewed!

A second problem is temperature and pressure are interdependent, so I can't correct for them separately. The two have to be corrected together and I'm not convinced I'm doing that properly yet.

I'm also assuming that changing temperature and pressure both effect period linearly, which may not be true! Either or both could be curves, or logarithmic, or something else. Plotting amplitude against temperature suggests they're related in a very non-linear way, that I hope doesn't matter!

A curious feature of my approach is the accuracy of the clock should improve as it gains experience, perhaps taking a year or more to reach the best it can do. Assuming I get the pendulum working reliably mechanically of course, and it isn't yet.

Dave

[David TocherParticipant @davidtocher94033]

If your reference clock period is altered then your regression coefficients will also change, these corrections are then used by your clock to adjust its displayed time. Thus a change in the reference clock changes your clock's behaviour. You said that your clock will 'gain experience' which can only come from repeated observations and then changing your clocks behaviour compared to the external clock. This will lock the two clocks together.

No matter what clock and reference are used this will happen. A pendulum clock compared to star transits will become locked to 'true' sidereal time if repeated corrections are made to the pendulum clock output in light of any differences that are observed.

[Author]

Posts