Another potential quip but perhaps was intentional, there was a seemingly intentional use the term "MOST PRECISE" in the video. In the clock world I'd imagine the term "MOST ACCURATE" , used in the title, would have a substantively different meaning than the phrase "MOST PRECISE" To my non-nerdy audience here: An accurate baseball pitcher consistently throws strikes within the zone, while a precise pitcher can consistently pinpoint specific locations within or around the strike zone.
I've gone so far down the rabbit hole of Pi-related acronyms that, for the first few minutes of this video, I forgot what GPS stood for. This is technology at its finest. Love the work you do, Jeff.
I put one of these in my microwave's control board and now I can finally test my reflexes stopping the cook time down to the nanosecond like a true action hero.
This is awesome! Looking forward to hyper time-accurate interviews! And a huge thank you for helping raise money for ITDRC. I'm the Florida State Coordinator for the organization and can personally attest to the impact the donations make to help fund our mission. Disasters are never fun, so being able to provide some ounce of hope to the survivors is a true blessing.
Another good example for GPS synched clocks are DAB+ radio stations. For DAB+ you don't use a big transmitter for a large area, like for AM, but a lot of smaller transmitters serving smaller areas. GPS is used to keep all the transmitters synced up.
Hey Jeff, try picking up CHU and WWV and sending them a signal report. They'll send you back a QSL card in the mail. The neat thing about them is not only are they accurate time sources, even their radio transmitter's oscillation is disciplined by their atomic clocks. 73.
Isn't wwv specifically designated and intended as both a time and frequency reference? (Which is why the frequency is linked to the atomic clock, otherwise it wouldn't be a good frequency reference)
It's nice to have a super accurate time source but distribution using NTP introduces latency making the accuracy of the source largely irrelevant so what's the point? 😮
@@glennmcgurrin8397 Absolutely. WWV-DO and GPS-DO are both a thing. I wouldn't be able to say how much more accurate one over the other is - but GPS-DO is certainly enough to calibrate SHF ham signals in practice.
It isn't the high temperature that makes the crystal more precise. It's the fixed temperature. It's hard to cool components. So this means the oven is used to raise the temperature well above the inside temperature of the device the crystal will be used in. So assume +50°C ambient temperature, and maybe +75°C inside the device. Then the crystal needs to be heated well above +75°C so no unplanned extra warm day makes the inside of the device warmer than the oven. After they have decided the oven temperature, they will then cut the crystal to be at the nominal frequency at this temperature. That's like how wrist watch crystals can be cut to deliver 32768 Hz at close to 37°C from the arm. Don't use the watch and the crystal gets too cold. Leave it in the window and the watch gets too hot. And this makes the watch drift in one or other direction. A 32.768 kHz crystal for a home appliance is better optimized for maybe +23°C. Somewhere slightly above room temperature.
I could've been more clear there :) But yes, it's easier to maintain a fixed temperature when you heat things up beyond what the typical environmental temperature would be; cooling it would be much more difficult, especially inside enclosures that could vary wildly in temperatures. One design consideration for anyone using this card is maybe try to keep it out of a stream of heavy airflow, no use making that little SiTime chip work harder than it needs to to maintain its temperature.
@@thebamplayer It's very unusual now to use the mains power AC periods for time. It was quite common 30-40 years ago because crystals was expensive then. And older power supply designs made it easier. And in that case, at least for my country, you really would hardly see any drift because the frequency might be just below 50Hz during day but then run at just over 50Hz during night to keep a correct total number of cycles over the full 24 hour period. Just that the removal of old-school mains transformers and instead using switched power supplies means that in modern electronics you now need extra components all the way from the mains-power side. This adds cost - both he components and because of any certification that the mains power is properly isolated.
I have carried lots of interesting electronics and cables through the customs. The important thing - it doesn't smell like explosives. And it's good to be able to have batteries separated.
Yes; I'll be taking the batteries in a carry-on, while the rest will be packed neatly in my checked bag. I've only gotten taken aside for closer inspection one time carrying all sorts of strange contraptions through security :)
For your display… there might be some some signaling and compute delay in your python. Please check your offset calibration, values, and timing. I.e. set a log flag at the start of your script and log flag at the end (or whenever is appropriate) then use that for your screen off-set, and signaling / ptp / ntp offset. Love the project/product.
It will take a while even with an oven to age the crystal to a reasonable point of stability, we used to leave ours on soak for 2 months to get them stable enough for broadcast use.
This is so amazingly/hilariously over engineered….and I love it. While you’ll have the most accurate time there, Linus will still be late. Looking forward to seeing this at LTX!
@@JeffGeerlingfor Wan show there is no need for such accurate timing, it just need to be accurate ±2 hours, which means it could be an hourglass with rocks in it 😅
I have a Pi Zero running a few little things, including Chrony as my stratum 1 internal NTP, relying on an Adafruit GPS board with PPS. I have a 4x20 screen in an acrylic mount on my desk that serves as my geeky desk clock. It was under $100 all in, and has a few purposes for me.
Awesome! When I was checking on their current deployments, I saw they were helping some tornado victims just south of STL. They're deployed all over the place!
Welcome to the team and thanks for volunteering! Make sure you join the ITDRC slack community so you can keep up with everything goin on here in Region 4 and around the country.
I'm currently syncing Chrony with a stratum 1 server hosted at the Quebec Internet Exchange a few km away, but I'd love to have something like this in my lab! By the way it's OCXO, the X being short for Xtal/Crystal.
Ayoo, Loved seeing the m8f! Used it in my Bachelor Thesis where I implemented a OCXO (stratum 2 compatible) PLL chip with the PPS Signal coming from the GPS module. They are (even with the internal TCXO) really accurate and low jitter. Iirc somewhat around single digit nanoseconds compared to our big ocxo gps time reference. And yeah, just around 100$ by their own lol
End application was a phasor measurement unit which sampled the grid power based on the PLL output. Gotta keep good time synchronization when you're making the measurements and phasor estimations on the pi. Really great learning experience on how to measure >time
I used to war drive with my RP4, external GPS and USB WiFi on monitor mode. I captured a WiFi hot spot that was inside a parked cop car, with SSID "Purple Pig". I didn't stop to ask. What's interesting is that your GPS is sandwiched between the CM4 and your PCI board.
Could be a fun project! (And there are certainly cheaper ways of getting GPS signals into your main time server if you want to try that to get started!).
Another important difference between the OCXO and the rubidium card is that "board with a chip that gets hot but isn't powered on" is probably a lot easier to clear through customs compared to "board that is radioactive."
Hehe, though any decay in the isotope used in the Time Card (non-mini) would be practically un-detectable by airport security. They would think it looks a little sus though.
There is no radioactivity in a rubidium clock (unless you count the 50 billion year half life of the Rb-87 used in the filter - but with a half life much longer than the age of the universe, it is debateable whether to call this radioactive). Moreover, the clock shown uses a laser light source, rather than a filtered discharge lamp - so it doesn't even have the Rb87.
With all that python between the time source and the user I'm not sure you can claim the most accurate time, the underlying clock might be the best there but that doesn't do you any good if its displaying late by a probably wildly variable margin (at least in comparison to the clock accuracy). Might be the most drift proof clock present however and so be the most accurate eventually. But it probably isn't delivering the time to the display as accurately as the NTP will be on every other computer there, and even cut off over the duration of the event I'd expect the drift of the riff-raff PC's to be less than the inaccuracy caused by using the wrong language for such a hard realtime task! The most real of real time tasks!
The computer itself will be holding time accurately though-honestly the blinkstick and miniPiTFT are delayed enough even with a C library I'd end up with variable ms-level delays for end-user display. The Pi will be offering PTP output over Ethernet, though, so I could still say it's the most accurate time source-just not necessarily via it's visual outputs.
@@JeffGeerling True, and even if you went to the extreme of bare metal programming the blinkstick and screen so it is precise as the CPU cycle allow the LED warm up/pixel change times of those HID devices has a meaningful latency when talking this level of timing precision too. Make sure you serve at least once client with the correct time over the Ethernet as I think you can claim to be the most accurate timesource, but if nothing ever uses the really accurate time it could provide I don't think it counts - the Human interfaces on it are going to be the only things used, and they are not impressively accurate.
As soon as I saw that Ublox chip I knew it had to be GPS related. I've been trying to research accurate (especially vertically accurate) GPS for a while now and it's pretty hard to understand as a layperson and there's not much info out there anyway, it seems to be a very niche interest. If you could make a video about accurate GPS that'd be so helpful and I think a lot of people would find it cool.
Very cool! At some point LMG or your self should do a video on AoIP. PTP is huge in that space. Livewire, Wheatnet, Revenna, or Dante would be cool to see a deep dive on!
Just making sure all understand these: precision: how "finely divided" some measurement is accuracy: how closely a measurement agrees with some accepted standard So you can have a clock which has microsecond precision but is off (inaccurate) by minutes. Or you can have a clock that is accurate within milliseconds but only one second precision.
Oooh... if you have a Dante setup it would be fun to see if I could be the PTP source! Otherwise, I'll just gawk at your mics for a bit. I was trying to find a good ENG mic to throw in my bag for the livestream.
Having involved years ago in buying ntp appliance + Rubidium oscillator for very accurate time, got a kick out of it. To be clear for a majority of users, ntp/chronyc is enough for miliseconds accuracy as Jeff mentioned... unless your doing trading for accurate time.
Oh yes, I've built a GPSDO before, using an OCXO, a GPS module and part of a PLL chip (specifically the phase comparator). No RPi needed. (edit) and no fancy hundred-dollars GPS module needed either. A $5 ublox7-compatible module is enough, with some skillful programming of its internal registers. :)
It does depend on what level of precision you are looking for, as well as the quality of your oscillator. My experiments with some of the cheapy ublox devices, show that they can easily show time deviations of +/- 100 ns, meaning that frequency stability at 100 seconds is only 1ppb on average (there can be larger excursions if there is a change in observing conditions). However, in the event that you need better absolute time precision, or a better precision frequency reference, then improving the GPS receiver may be needed. One interesting thing about making low cost GPSDOs is that obtaining "well-used" obsolete OCXOs from aliexpress is not just cheap, but the crystals are "pre-aged" and often have extraordinary stability. I've got a bunch of the $2 OSC5A2B OCXOs, and when using these for a GPSDO, the limting factor (by a huge margin) is the GPS receiver - you need to use PLL bandwidths in the order of 200 uHz, otherwise the GPS noise gets through to the output. I've got a couple of somewhat more expensive, old SC-cut OCXOs from Ali (about $50 each), which when measured significantly outperform the specifications of the rubidium oscillator shown in this video ( < 1x10^-12 adev @ 1000 seconds) with some software temperature compensation. That said, these miniature rubidium oscillators are a lot worse than the older ones. The laser based CPT method used in the miniature versions has much worse stability than the old-school rubidium lamp and microwave excitation design.
Nice video. We used a NavSync timing gps and feed its 10 MHz output to an FPGA and try to derive a 100 MHz signal from gps. This gives a 10 nanosecond pulse period. A cosmic ray signal event in our application is time stamped to 10 nanosecond in theory.
Rubidium clocks use the oscillation of the atom between two different excitation states as their clock reference, NOT radioactivity. Rubidium-87, the isotope used in atomic clocks is NOT radioactive.
The Ublox modules are fairly cheap, They are available on boards for drones. The OCXO is used for the 10MHz output that is used to sync test equipment. Check out Timenuts' if the subject interests you. I have two Meinberg GPS170PCI systems which is a computer card, and a GPS outdoor antenna with a nice molded plastic mounting arm that clamps around a mast, or tower leg. I have a Vk-172 Ublox 7 USB GPS that plugs into a laptop for field work.
Yeah; U-blox makes a HUGE variety of GPS modules. This high-end one includes the capability for external oscillators and a much better internal TCXO too (100 ppb I think?).
I have to use this whole setup for my master thesis and my prof doesn’t even have a budget for it. I’m just settling with NTP implementation and hope it works😅 Not sure it would give a good result at the end for my distributed systems algorithm.
It's well into the third quarter of 2023 here in Canada and the availability of Raspberry Pi 4 has still not arrived. I still have hope that Upton's promise will come true.
I've seen PiShop.ca getting stock at least a couple times a week lately. Down in the US it's available somewhere at least a couple times a day now, and it'll be in full stock soon.
"Too soon" 🤣. You'd fit right in down under, to quote Doug Anthony All Stars: "we had jokes written about the Challenger disaster before the #### thing hit the water!" (yt vid: DAAS Dead & Alive, 7m20s)
NTP tested against the 1PPS goes more on the order of 3 .. 10ms and occasionally worse than 10ms. With the Pi 4, you can test against the Pi clock to about 18 ns resolution. If you take in the 1 PPS and set up an interrupt, then you'll have sub-microsecond accuracy for time - even Jeff should be happy with that? So - you only need the serial output from the uBlox receiver ($50 - $80 at Sparkfun) breakout board for GPS data and the 1 PPS signal. I implement this in RTL rather than Linux. Of course this is not within a PC.
If you're using interrupts to capture time then it is likely that you have 10s to hundreds of microseconds of latency and considerable jitter. Linux is not a real time operating system. Interrupt latency varies wildly and is considerably faster in kernel space than user space. Hardware capture timers are needed for real accuracy.
@@therealtimwarren I'm not using Linux (as stated above). It's a barebones/bare metal set up. Latency and jitter verified against the Pi-4's 54 MHz clock of the system - indeed, with the 1 PPS I can also accurately characterize the 54 Mhz clock''s accuracy as 9 .. 11 PPM (on my Pi) depending on temperature.
