One of the best features of this clock (I have one and have been messing with it a while now) is the up to 100 kHz refresh rate and an analog LCD driver circuit that makes the digits not flicker even a bit, even when dimmed, with a high speed camera.
That and the fact it's the highest precision clock display I've ever owned!
It is certainly not inexpensive, but it's more of an art piece than a practical instrument, unless your eyes can see in the thousands-of-hertz range!
Note that it's LED, with each segment driven continuously with a variable voltage for brightness control. Even most LED drivers are multiplexed, with each segment flashing for a short time then going dark. This uses a separate dedicated buffer for every segment. This design is important for working well with high-speed cameras, an obvious use.
"If we had a cellular modem, we could get the time from the cell towers, which is broadcast through a protocol called NITZ. This is how phones auto-update when you enter a different timezone, but it's somewhat unreliable depending on the carrier, and overall worse than using GPS. And I really don't want my clock to have a sim card."
Just out of curiosity, what can you do with a cellular modem but no sim card? Can you get the time?
CDMA networks broadcasted the time (from GPS) in the clear as part of the base station advertisements, so you used to see CDMA used as a precision time source without any kind of subscription required. CDMA equipment required accurate time for TDM coordination. Unfortunately GSM uses a different architecture for synchronization and does not require accurate time at all, so you have to be a subscriber to request time information and even then it is not all that reliable.
Timing is all about periodicity; if something beeps every second, you can measure intervals between two beeps but have no other information. It's often the case that timing is also synchronized to, say, second boundaries too, and most time sources would do this. Time would then be giving some indices to those beeps; the time source would beep and say that it was the N-th beep so that you can work the actual time out from N.
I assume they are referring to the channel timing? Cellular frequencies are segmented into time segments where each channel is allowed to be used by only some devices when it is their "turn" to transmit (this allows multiple phones to share the same frequency at the same time).
How can you access this timing I run a little mobile proxy service for myself and a few others, could add some value there. Mobile networks are quite interesting.
However, it seems that extra SIBs aren't necessarily broadcast but may be available on demand... and it wasn't clear to me whether making a on-demand SIB request can be done without SIM card.
I know this wasn't meant literally, but it does make for an interesting thought experiment - under what circumstances might it be valid to dial 911 to ask for the time?
My first thought was something like a nuclear operator - "we need to shut down the core at exactly 19:00, but our clocks are down!" so they call and wait for the operator to advise when the time is reached. Obviously contrived and not realistic, but interesting to think about.
> under what circumstances might it be valid to dial 911 to ask for the time?
Maybe, if all of your clocks don't work and you went to somewhere else and their clock isn't working either, and it is raining and you cannot use a sundial, and you tried to call everyone else already and they also cannot give you the time for whatever reason, then you might try to call 911 and ask them, because you tried everything else and it didn't work. (I once heard a (fictional) story where this happened. This is an unlikely scenario, but some of the things mentioned here (and other things) might happen, e.g. bad weather so you cannot go out, the television and computers are not working (and maybe the power is out but the telephone uses a separate power), and there are some problems with the telephone too (I have had problems before where some telephone numbers worked and some didn't), etc.)
Well given that virtually everyone's "telephones" are cell phones which would be incapable of making calls unless they connected to the network, and the network would be incapable of handling their calls without also telling the phone what time it is, this situation becomes far less tenable in 2025 vs the 20th century when land lines were king. :)
At least in many European countries, the landline phone companies offered a short number for obtaining the time from an answering robot.
Many, many years ago, I have designed a piece of equipment that was integrated in a phone exchange and it provided vocal messages to be sent to a caller, for errors like non-existent phone number, but also for replying to the dedicated number for the time service. The messages were something like "The time at the next beep will be ... hours ... minutes ... seconds".
I have not heard about a similar service for mobile phones, because here the phone gets the time automatically and it displays it.
It's not a good idea to call 911 with non-emergencies.
But until a few decades ago, the primary way most of us to set our clocks was to call a number the phone company provided, which in our case was TI4-1212. "At the tone, the time will be ..."
If a response packet contains a good timestamp, you could initiate a 911 call, get a reply packet, and cancel the call before it reaches any actual operators.
> I designed this clock years ago, with the intention to incorporate every feature request I ever received for the previous precision clock.
My first thought after reading this statement was to add one Ethernet port to run a NTP server and have PoE capability. Completely overkill for the intended purpose, but I can't help but giggle at the thought of hanging this on a wall in a datacenter and have the clock also provide time-keeping for it.
Using 0.56” segment displays would probably let you reduce the horizontal size to below what would fit in a 1U rack mount. It would look sick at the top of the rack in my living room.
NTP will typically have about 10ms error which would be visible on this display. So for more authentic giggles PTP is recommended. With compatible switches to compensate for queuing delays and cable length. And if your provider doesn't give you a good time, put your own GPS-backed network clock on the roof. For science.
No GPS in SCIFs. Agree ntp would be great. Also I’d prefer a way to manually switch between a fixed set of time zone, or even just showing UTC always as an option.
If you have a special project need, you might be able to get one-way IRIG time signal on fiber with guard boxes at both ends. There's really no technical reason why you couldn't do analog GPS baseband over fiber, but you do need approved equipment at both ends for policy reasons. (certified for no backflow)
This is insanely cool. It's a beautifully written article that covers all the major design decisions and issues that affected the final outcome. I'm considering purchasing one for myself and to support the developer. It would be a very cool addition to a homelab or tech/hacker space.
I think my own minor issue is the use of a micro-USB port instead of USB-C - I wonder why, I don't see it mentioned in the article. Maybe adding a PD controller would have complicated the already complex dance of components on the board. Standard USB can supply 5V at 1A (5.0W) no problem, so maybe that was just a simpler way to go.
> I think my own minor issue is the use of a micro-USB port instead of USB-C - I wonder why, I don't see it mentioned in the article. Maybe adding a PD controller would have complicated the already complex dance of components on the board. Standard USB can supply 5V at 1A (5.0W) no problem, so maybe that was just a simpler way to go.
You don’t need a PD controller. You need two resistors of a set value to be able to have USB-C supply 5V at up to 3A. No ICs needed on the input side.
A cheap GPS module without a disciplined 1PPS output (like the ublox MAX-M8 that's used in this project) typically has a jitter of about a few hundreds nanoseconds, so the potential is there for the precision to be much better than 1ms if the microcontroller on this clock has the 1PPS wired to an interrupt routine.
The 1PPS output may have that much jitter (due to it being clocked synchronously with the module's system clock) but I believe there is usually a way to query the device for the error between the 1PPS signal and the actual idealised pulse time.
Sylvain Munaut made a project using a high resolution TDC/time of flight converter to discipline another clock against the GPS idealised pusle time. I can't find a link right now unfortunately.
Yes, this feature exists, but it's usually only present on GPS modules that are sold specifically for time keeping instead of location tracking, such as the uBlox LEA-M8T.
They are much more expensive, think $100 instead of $10.
In a previous life, with a specific set of hardware, we used a measured value of 70 ns two sigma as the typical skew between two GPS-disciplined clocks. Don’t use these numbers without testing your own system, but confirms order of magnitude estimate.
I'll concede I didn't make it all the way through. I always have respect for people who make a physical product, especially one to sell.
I'm not really tracking the MHz signals thing considering there's the 10 MHz oscillator. It would've been interesting to see an EMI/C report. Usually I can see SMPS frequencies as a spur (especially in PFM mode).
For the 32K, at least for the H7 they recommended a ring of ground around it for some shielding. Unsure if that would've helped here. There are also micros that can use their internal RC 32K on vbat. And ones that can do USB without a crystal. The best micro is the one you know, the second best is the one you have. Maybe that's in play here.
The LED driver variability was nifty. My first reaction to large IO with tight timing would be an FPGA, but I don't know if they'd like the variable bank supply. It would probably be fine bit the solution used is interesting and works.
> It would've been interesting to see an EMI/C report.
It's a two layer board with no attempts made to keep loop areas small, it's going to be bad. Most of what I read of this seems fine at a glance but that part about EMI sticks out as complete nonsense.
Edit: I have nothing against it being a two layer stackup as part of the art of it, I however don't like that they're also claiming that it's designed to have extremely low radiated emissions while doing so.
That is not correct, with a charitable reading of the text. Which is not to say it’ll have good EMI characteristics - you never really know until you get it tested.
> Artificially confining ourselves to two layers is exactly the kind of challenge I am wont to be engulfed by. The trick, if you want the board to work well, is to work on only one layer, and keep almost the entirety of the other layer as ground. It is almost always possible to do this, if you're willing to put in enough thought. I rather find that designing circuit boards is a lot like Tetris, and once I'm in the swing of it I can route things for hours on end. It can become a multi-day trance, with dreams of signal integrity and current loop area.
Though on this subject:
> In contrast, said this one PCB designer, a thin track has a higher inductance, so if you make your power supply lines thin, you'll get a free bit of extra filtering.
Inductance in your power supply lines is bad, but I don’t have time to get into it right now.
I didn't see that bit, but the layout image[0] shows very large cuts in the ground plane with traces running right over the middle of them, especially between U8 and U7. There's plenty of space here to add some ground traces between them, but that hasn't been done.
> you never really know until you get it tested.
I suspect they haven't given the lack of a CE marking on the board.
As an aside that bothers me a bit given that they say they sold 1000 of the last version, which would easily cover the cost of the (legally required) testing. Normally I'm of the opinion that it's fine to ignore it for some obscure hobby product where the cost of testing would be higher than any expected profits, but in this case they have the money from previous versions of the same product.
For just the EU EMCD and for a simple device like this there's labs out there that will charge as little as $1000. You can technically also do things yourself although I don't know the exact legal requirements off the top of my head.
You can 100% do it yourself, though I think the fact that it has GPS makes it a bit of a weird area because it starts to become a radio even if it's not an intentional transmitter (I've had people who work at EMC test labs tell me they don't understand WTF the standards are actually supposed to mean for GNSS kit). But for a regular electronic gadget you don't need a lab with the right certificates or anything, it's just a declaration on your own liability that you think it meets the required standards. If you're happy to do this yourself and that gives you enough confidence to sign it off, you can. (And in practice the regulators don't really come knocking for small stuff like this unless it's somehow doing something catastrophically bad)
Around here (US East Coast), a day of testing at an FCC-accredited lab for self-certification (so without the lab having to do the TCF or their own documentation) of a small device without unusual IO or power requirements costs between USD $1K-$4K.
>Inductance in your power supply lines is bad, but I don’t have time to get into it right now.
Not always. You want to have appropriate nearby bypass capacitance, but it's common to filter power lines by deliberately adding inductance into them (this also requires care to avoid setting up resonances, but a thin trace will also add some resistance which helps with that). Either way it's not likely to make a detectable difference on this kind of PCB anyway.
A case would probably have it back in spec, if its even out of spec on it's own. The energy is going to be so tiny in those high frequency waves because it's just signal and clock. Never mind that they are still pretty low frequency and those traces are going to be awful emitters in that range.
The highest energy part of the device is going to be the LEDs, which are switching well below the bottom of the standard EMI compliance range (10Mhz).
MCU outputs can be nasty with their high switching speeds. Every STM32 has some slew rate limiting as far as I know, but I looked at their code on GitHub and they're explicitly turning it off for most pins so I assume it's not an option for their driver ICs for some reason.
That reminds me of production studio master clocks such as the bronics WCD-530W or the evertz equivalent. I'm more of a fan of the analog style such as the 1275T - https://evertz.com/products/12x5T.
What a beautiful execution. He seems to have included all requests and written about every detail.
Going to this kind of resolution, the only practical use case is with cameras. You record some frames and later want to check the exact timestamp. This is even useful for a single image, not just high speed cameras.
But taking images is not just a single point in time. There is an exposure time and it is hard to measure absolute timings on start and end of exposure. With a segment clock display, the numbers will blur.
For this a row of LEDs is fine - just light up one at a time. In the image you will see multiple leds on at once, giving you verification of start and stop.
But again, I needed just one clock and built it already.
When you're identical twin gets back from their extended space travels with their clock you're going to be in for some hard truths about the nature of precision and time.
They just need to go to IIS. IIRC the astronauts are already milliseconds younger than us. Too bad hard living up there prematurely ages them far faster than milliseconds per day.
To work at sea it would rather need a way to manually enter an offset, since customarily the local time on board is entirely at the discretion of the master.
When I encourage/cheerlead my extremely technically talented friends to “monetize their hobbies”, I am picturing exactly this.
A hyper technical, perfectly executed, fun/whimsical/smart product. You can share all the source code and have build instructions! And sell assembled versions and make a buck! And hopefully make money / thrive/survive doing what… you love?
((The reality, ofc, is that in the hardware biz, getting more orders can just mean a ton of additional problems re logistics of making, assembling and shipping the thing. But I still naively think that’s a good problem..?))
Extraordinary Clock, truly. Cheers, congrats on launching and wishing you all the sales!
Really interesting to see the deep dive into PCB design and EMI considerations here. It's a great reminder how much thought goes into balancing cost, manufacturability, and compliance, even for hobbyist products. The point about using one layer as a near-continuous ground plane is especially practical, and it's fascinating how even seemingly minor layout choices can have big implications for signal integrity. Thanks to everyone for sharing their expertise—it's one of the things that makes this community so valuable.
Cool project, but with this sort of power and area budget, can't we use chip scale atomic clocks ? Also, the quartz accuracy is listed as 1000 seconds to drift 1 millisecond. So that's 31 secs a year ? That's less than HAQ quartz watches (+- 5 secs a year) and definitely less than Citizen 0100 (+-1 sec a year).
Yes, I get that, but that is less interesting. Any clock can sync to GPS or NTP. It would be cooler to have an autonomous accurate clock if you are making a giant device with an ARM chip, flash storage etc.
I wouldn’t know how to obtain a clock that synchronizes to GPS. I’ve had situations in the past where it would have been useful to have such a clock for checking at a glance if the NTP server is screwing me up, or when you’re offline.
If you want a commercial GPS wall clock, Seiko makes several models. I have one - GP502W. They make others too. They are JDM models though, so you'll have to order one from Japan.
In the continental US it can be more convenient to get a clock that syncs to WWV, since it works without an external antenna. WWV is accurate to within 100ns, though you'll have propagation delays depending on your position in the US (I'm 50 miles from the station).
No, it doesn't. You need to set the offset to UTC manually. You also need to set the dates when daylight savings start and end if you live in a place that that uses daylight savings. But it runs on batteries that last for 3 years, so it's very low maintenance.
There is another project out there using a BeagleBone Black plus a small hardware RTC and a GPS module hooked up over GPIO/serial. The advantage there is that the BBB supports hardware PTP timestamping on its built-in NIC. I threw one together over a weekend, just need to design up a small box to hold it all in (though the cable mess of jumpers also made me want to design an interposer board to clean things up)
Those chips are also useful for getting better positions from navigation signals. One could imagine using them to help filter out spoofed signals, for example.
I've been testing various GPS/GNSS antennas indoors for the past year, and unless you're in a multistory building or have a full metal roof, there's a good chance you'll pick up 4+ satellites, good enough for a decent time.
But it can come and go, especially as there are other sources of interference in the 1-1.5 GHz range indoors, and reflections off various surfaces and objects make the signal way harder to distinguish.
But for by far the best chance of success, you should have the antenna outside, ideally with a clear view of most of the sky.
If inside, you could put it near a window or exterior wall, but a lot of modern construction materials can block things pretty badly.
I have an older garmin gps18x I use for ntp and it gets fantastic reception from a window shaded by trees. I wanted to test a monitor script so I stuffed it in a metal can, capped with a thick plate, and used excessive tinfoil to seal...If i recall it only dropped 2 satellites near the horizon, leaving 4+ with good signal. I suspect the signal was leaking in through the cable.
During past year running i don't believe its lost 3d fix once (not in the can)
I thought about doing precisely something like this, with a tcxo and an fpga! I’m happy to see someone else has already taken it to its logical conclusion.
The only other thing I’d add is NTP support, and perhaps I2C for driving external displays (such as a few other time zones).
Wouldn't be- a oscillator display, showing the quartz crystal oscillating, be the final display of precision? Like a pendulum, but for the digital age?
It's an artifact of the camera. The camera shutter is long enough that it averages the images over 33ms.
At some point in the video you can see that a high speed camera can see the correct display.
At the 7 minute mark the industrial 14k FPS camera shows essentially zero rollover. The earlier rollover does appear to be an artifact of the cheap consumer grade high speed camera used.
> Possibly the most requested feature, automatically setting the timezone based on the GPS coordinates was essential. However, most people don't realise how complex this process is, at least within the restrictions I'd given myself: no internet or cellular modem, nothing beyond the GPS data, and a USB port for occasional updates.
> If we had an internet connection, sure, we could just query an online service. But I don't want my clock to require wifi.
> If we had a cellular modem, we could get the time from the cell towers, which is broadcast through a protocol called NITZ. This is how phones auto-update when you enter a different timezone, but it's somewhat unreliable depending on the carrier, and overall worse than using GPS. And I really don't want my clock to have a sim card.
> I'd also like, just for the sake of it, for the clock to work anywhere in the world, even at sea.
> This means we need to keep a copy of the map data for the country outlines for the whole world. It also means we need a full copy of the timezone database, and have the processing power onboard to make use of all this.
In fact, this is one of the central design points he mentions within the first two lines of TFA: no point in a clock with a milliseconds place if it doesn’t display each millisecond legibly.
> Summary
The specifications for the clock were as follows:
* Millisecond precision, with no perceptible jitter
* Display clearly, without flicker, when filmed at very high framerates (20,000fps or more). The brightness should still automatically adjust, of course, and without the use of PWM
This reminded me how in Japan, 700 people were rushed to the hospital after watching pokemon on TV [1] due to 12Hz red/blue flashing frames, but now I understand that’s not the case here.
I’m guessing you’ve already realized this, but in case it helps clarify things for someone else:
The signals rates mentioned in the article are with respect to how frequently the segments are updated, not the brightness. (With the one exception being the colons being PWM controlled, IIRC).
PWM means Pulse Width Modulation. Controlling a (perceived) light intensity of an LED via PWM means (if we pretend for the sake of simplification that the voltage rise and fall is instantaneous, etc) quickly turning the LED fully on and fully off, varying the duty cycle to achieve the desired perceived brightness. This project avoids the flickering inherent with PWM by not using PWM: the voltage itself is set to some fixed value for a given target brightness.
(If one then wonders why this wouldn’t be the default approach for such things: microcontrollers often provide a built in means for PWM output, but it’s less common to have built in true, non-PWM analog outputs, requiring additional parts (thus more cost, more complexity) to implement variable voltage control).
Ah. I believe the “correct” solution is you record at high speed and down sample for the big screen in order to avoid the motion blur. There are a couple scenes before he gets to the real high speed where he’s showing the clock slowed to 1/10 speed at which point the 100ths place is only a little slow to update. I’m still seeing some alpha nerd room for making the LEDs tick over faster. Maybe the Mark V.
What part does this comment refer to? It looks like they're multiplexing each section of the display at 100 kHz rate (500 kHz overall), well outside that range.
I was just raising a general concern about flicker affecting people with photosensitive epilepsy, but if the multiplexing is indeed at 100 kHz per section, that’s definitely outside the problematic range.
One of the best features of this clock (I have one and have been messing with it a while now) is the up to 100 kHz refresh rate and an analog LCD driver circuit that makes the digits not flicker even a bit, even when dimmed, with a high speed camera.
That and the fact it's the highest precision clock display I've ever owned!
It is certainly not inexpensive, but it's more of an art piece than a practical instrument, unless your eyes can see in the thousands-of-hertz range!
Note that it's LED, with each segment driven continuously with a variable voltage for brightness control. Even most LED drivers are multiplexed, with each segment flashing for a short time then going dark. This uses a separate dedicated buffer for every segment. This design is important for working well with high-speed cameras, an obvious use.
Super random but I seem to remember that you were working on/testing a different time device/prototype recently in a GitHub comment.
Can you perhaps refresh my memory? Been trying to find the reference and it's driving me nuts this am. Thanks.
Check my time-pi repository on GitHub: https://github.com/geerlingguy/time-pi
"If we had a cellular modem, we could get the time from the cell towers, which is broadcast through a protocol called NITZ. This is how phones auto-update when you enter a different timezone, but it's somewhat unreliable depending on the carrier, and overall worse than using GPS. And I really don't want my clock to have a sim card."
Just out of curiosity, what can you do with a cellular modem but no sim card? Can you get the time?
CDMA networks broadcasted the time (from GPS) in the clear as part of the base station advertisements, so you used to see CDMA used as a precision time source without any kind of subscription required. CDMA equipment required accurate time for TDM coordination. Unfortunately GSM uses a different architecture for synchronization and does not require accurate time at all, so you have to be a subscriber to request time information and even then it is not all that reliable.
No. You have to have access to the network to get the time. You get timing from the network “for free” but it won’t give you time.
I'm not exactly clear on the difference between "time" and "timing"...? Like an accurate incrementing time offset from some unknown start point?
Timing is all about periodicity; if something beeps every second, you can measure intervals between two beeps but have no other information. It's often the case that timing is also synchronized to, say, second boundaries too, and most time sources would do this. Time would then be giving some indices to those beeps; the time source would beep and say that it was the N-th beep so that you can work the actual time out from N.
I assume they are referring to the channel timing? Cellular frequencies are segmented into time segments where each channel is allowed to be used by only some devices when it is their "turn" to transmit (this allows multiple phones to share the same frequency at the same time).
How can you access this timing I run a little mobile proxy service for myself and a few others, could add some value there. Mobile networks are quite interesting.
Is there any difference between a phone with no sim, and a phone with an old sim that's not linked to an active account?
Yes, on newer networks, the 5G NR System Information Block 9 (SIB9) provides UTC time.
However, it seems that extra SIBs aren't necessarily broadcast but may be available on demand... and it wasn't clear to me whether making a on-demand SIB request can be done without SIM card.
One possibility is to call 911 and ask what time is it.
I know this wasn't meant literally, but it does make for an interesting thought experiment - under what circumstances might it be valid to dial 911 to ask for the time?
My first thought was something like a nuclear operator - "we need to shut down the core at exactly 19:00, but our clocks are down!" so they call and wait for the operator to advise when the time is reached. Obviously contrived and not realistic, but interesting to think about.
That's what the US Naval Observatory Master Clock phone service is for: 719-567-6742
> under what circumstances might it be valid to dial 911 to ask for the time?
Maybe, if all of your clocks don't work and you went to somewhere else and their clock isn't working either, and it is raining and you cannot use a sundial, and you tried to call everyone else already and they also cannot give you the time for whatever reason, then you might try to call 911 and ask them, because you tried everything else and it didn't work. (I once heard a (fictional) story where this happened. This is an unlikely scenario, but some of the things mentioned here (and other things) might happen, e.g. bad weather so you cannot go out, the television and computers are not working (and maybe the power is out but the telephone uses a separate power), and there are some problems with the telephone too (I have had problems before where some telephone numbers worked and some didn't), etc.)
Well given that virtually everyone's "telephones" are cell phones which would be incapable of making calls unless they connected to the network, and the network would be incapable of handling their calls without also telling the phone what time it is, this situation becomes far less tenable in 2025 vs the 20th century when land lines were king. :)
In the UK they are fine with you calling 999 to test your newly-configured office phone routing. They don't want a chat though.
That's a bit different than asking for the time. That's testing a newly configured emergency system to make sure it works.
At least in many European countries, the landline phone companies offered a short number for obtaining the time from an answering robot.
Many, many years ago, I have designed a piece of equipment that was integrated in a phone exchange and it provided vocal messages to be sent to a caller, for errors like non-existent phone number, but also for replying to the dedicated number for the time service. The messages were something like "The time at the next beep will be ... hours ... minutes ... seconds".
I have not heard about a similar service for mobile phones, because here the phone gets the time automatically and it displays it.
It's not a good idea to call 911 with non-emergencies.
But until a few decades ago, the primary way most of us to set our clocks was to call a number the phone company provided, which in our case was TI4-1212. "At the tone, the time will be ..."
You can still call NIST and get the time via "at the tone...": https://www.nist.gov/pml/time-and-frequency-division/time-di...
If a response packet contains a good timestamp, you could initiate a 911 call, get a reply packet, and cancel the call before it reaches any actual operators.
> I designed this clock years ago, with the intention to incorporate every feature request I ever received for the previous precision clock.
My first thought after reading this statement was to add one Ethernet port to run a NTP server and have PoE capability. Completely overkill for the intended purpose, but I can't help but giggle at the thought of hanging this on a wall in a datacenter and have the clock also provide time-keeping for it.
Using 0.56” segment displays would probably let you reduce the horizontal size to below what would fit in a 1U rack mount. It would look sick at the top of the rack in my living room.
You've got a rack in your living room? Sick! (in both meanings)
NTP will typically have about 10ms error which would be visible on this display. So for more authentic giggles PTP is recommended. With compatible switches to compensate for queuing delays and cable length. And if your provider doesn't give you a good time, put your own GPS-backed network clock on the roof. For science.
That would actually be really useful though: being able to see at a glance what the network time should be would be incredibly useful.
There's other types of environments where you isolated users like SCIFs or really anything airgapped.
Hell: that would convince me to buy one that for my house.
No GPS in SCIFs. Agree ntp would be great. Also I’d prefer a way to manually switch between a fixed set of time zone, or even just showing UTC always as an option.
How do SCIFs do time? Do they use optoisolated network connections and then do time sync over that? (SIPRnet, is it?)
The US Navy maintains NTP servers on a variety of networks, the public internet and SIPR included.
https://www.cnmoc.usff.navy.mil/Our-Commands/United-States-N...
If you have a special project need, you might be able to get one-way IRIG time signal on fiber with guard boxes at both ends. There's really no technical reason why you couldn't do analog GPS baseband over fiber, but you do need approved equipment at both ends for policy reasons. (certified for no backflow)
This is insanely cool. It's a beautifully written article that covers all the major design decisions and issues that affected the final outcome. I'm considering purchasing one for myself and to support the developer. It would be a very cool addition to a homelab or tech/hacker space.
I think my own minor issue is the use of a micro-USB port instead of USB-C - I wonder why, I don't see it mentioned in the article. Maybe adding a PD controller would have complicated the already complex dance of components on the board. Standard USB can supply 5V at 1A (5.0W) no problem, so maybe that was just a simpler way to go.
> I think my own minor issue is the use of a micro-USB port instead of USB-C - I wonder why, I don't see it mentioned in the article. Maybe adding a PD controller would have complicated the already complex dance of components on the board. Standard USB can supply 5V at 1A (5.0W) no problem, so maybe that was just a simpler way to go.
You don’t need a PD controller. You need two resistors of a set value to be able to have USB-C supply 5V at up to 3A. No ICs needed on the input side.
Great point! I suppose then it would be trivial to add ones own USB-C port if desired.
Most have very small spaces between the pins, they might not have been able to do them reliably on their home setup shown at the end of the post.
Power-only USB-C connectors don't have the same pitch issues as they only need to pin out the power and CC1/CC2 pins.
However, the 0.65/0.40 pads required for USB-C are no longer an issue for any of the board houses I have used in the past couple of years.
Yeah you should be able to find a USB c connector that works. Plus you only need to route the power, ground, and d+/d- signals
I use two of these clocks daily at work, to synchronize a high-speed process on video. I'll be buying a couple more.
How well synchronized are the updates of the two displays?
Assuming any decent GPS signal at all, they should be well within 1ms of each other, likely a bit better.
A cheap GPS module without a disciplined 1PPS output (like the ublox MAX-M8 that's used in this project) typically has a jitter of about a few hundreds nanoseconds, so the potential is there for the precision to be much better than 1ms if the microcontroller on this clock has the 1PPS wired to an interrupt routine.
> jitter of about a few hundreds nanoseconds
The 1PPS output may have that much jitter (due to it being clocked synchronously with the module's system clock) but I believe there is usually a way to query the device for the error between the 1PPS signal and the actual idealised pulse time.
Sylvain Munaut made a project using a high resolution TDC/time of flight converter to discipline another clock against the GPS idealised pusle time. I can't find a link right now unfortunately.
Yes, this feature exists, but it's usually only present on GPS modules that are sold specifically for time keeping instead of location tracking, such as the uBlox LEA-M8T.
They are much more expensive, think $100 instead of $10.
That it does.
Should be within tens of ns if you have good reception and decent antenna placement.
In a previous life, with a specific set of hardware, we used a measured value of 70 ns two sigma as the typical skew between two GPS-disciplined clocks. Don’t use these numbers without testing your own system, but confirms order of magnitude estimate.
I'll concede I didn't make it all the way through. I always have respect for people who make a physical product, especially one to sell.
I'm not really tracking the MHz signals thing considering there's the 10 MHz oscillator. It would've been interesting to see an EMI/C report. Usually I can see SMPS frequencies as a spur (especially in PFM mode).
For the 32K, at least for the H7 they recommended a ring of ground around it for some shielding. Unsure if that would've helped here. There are also micros that can use their internal RC 32K on vbat. And ones that can do USB without a crystal. The best micro is the one you know, the second best is the one you have. Maybe that's in play here.
The LED driver variability was nifty. My first reaction to large IO with tight timing would be an FPGA, but I don't know if they'd like the variable bank supply. It would probably be fine bit the solution used is interesting and works.
> It would've been interesting to see an EMI/C report.
It's a two layer board with no attempts made to keep loop areas small, it's going to be bad. Most of what I read of this seems fine at a glance but that part about EMI sticks out as complete nonsense.
Edit: I have nothing against it being a two layer stackup as part of the art of it, I however don't like that they're also claiming that it's designed to have extremely low radiated emissions while doing so.
> no attempts made to keep loop areas small
That is not correct, with a charitable reading of the text. Which is not to say it’ll have good EMI characteristics - you never really know until you get it tested.
> Artificially confining ourselves to two layers is exactly the kind of challenge I am wont to be engulfed by. The trick, if you want the board to work well, is to work on only one layer, and keep almost the entirety of the other layer as ground. It is almost always possible to do this, if you're willing to put in enough thought. I rather find that designing circuit boards is a lot like Tetris, and once I'm in the swing of it I can route things for hours on end. It can become a multi-day trance, with dreams of signal integrity and current loop area.
Though on this subject:
> In contrast, said this one PCB designer, a thin track has a higher inductance, so if you make your power supply lines thin, you'll get a free bit of extra filtering.
Inductance in your power supply lines is bad, but I don’t have time to get into it right now.
I didn't see that bit, but the layout image[0] shows very large cuts in the ground plane with traces running right over the middle of them, especially between U8 and U7. There's plenty of space here to add some ground traces between them, but that hasn't been done.
> you never really know until you get it tested.
I suspect they haven't given the lack of a CE marking on the board.
As an aside that bothers me a bit given that they say they sold 1000 of the last version, which would easily cover the cost of the (legally required) testing. Normally I'm of the opinion that it's fine to ignore it for some obscure hobby product where the cost of testing would be higher than any expected profits, but in this case they have the money from previous versions of the same product.
[0]: https://mitxela.com/img/uploads/clock/mk4/kicad-screenshot2....
Out of curiosity, what would getting it tested cost?
For just the EU EMCD and for a simple device like this there's labs out there that will charge as little as $1000. You can technically also do things yourself although I don't know the exact legal requirements off the top of my head.
You can 100% do it yourself, though I think the fact that it has GPS makes it a bit of a weird area because it starts to become a radio even if it's not an intentional transmitter (I've had people who work at EMC test labs tell me they don't understand WTF the standards are actually supposed to mean for GNSS kit). But for a regular electronic gadget you don't need a lab with the right certificates or anything, it's just a declaration on your own liability that you think it meets the required standards. If you're happy to do this yourself and that gives you enough confidence to sign it off, you can. (And in practice the regulators don't really come knocking for small stuff like this unless it's somehow doing something catastrophically bad)
Around here (US East Coast), a day of testing at an FCC-accredited lab for self-certification (so without the lab having to do the TCF or their own documentation) of a small device without unusual IO or power requirements costs between USD $1K-$4K.
>Inductance in your power supply lines is bad, but I don’t have time to get into it right now.
Not always. You want to have appropriate nearby bypass capacitance, but it's common to filter power lines by deliberately adding inductance into them (this also requires care to avoid setting up resonances, but a thin trace will also add some resistance which helps with that). Either way it's not likely to make a detectable difference on this kind of PCB anyway.
A case would probably have it back in spec, if its even out of spec on it's own. The energy is going to be so tiny in those high frequency waves because it's just signal and clock. Never mind that they are still pretty low frequency and those traces are going to be awful emitters in that range.
The highest energy part of the device is going to be the LEDs, which are switching well below the bottom of the standard EMI compliance range (10Mhz).
MCU outputs can be nasty with their high switching speeds. Every STM32 has some slew rate limiting as far as I know, but I looked at their code on GitHub and they're explicitly turning it off for most pins so I assume it's not an option for their driver ICs for some reason.
That reminds me of production studio master clocks such as the bronics WCD-530W or the evertz equivalent. I'm more of a fan of the analog style such as the 1275T - https://evertz.com/products/12x5T.
What a beautiful execution. He seems to have included all requests and written about every detail.
Going to this kind of resolution, the only practical use case is with cameras. You record some frames and later want to check the exact timestamp. This is even useful for a single image, not just high speed cameras.
But taking images is not just a single point in time. There is an exposure time and it is hard to measure absolute timings on start and end of exposure. With a segment clock display, the numbers will blur. For this a row of LEDs is fine - just light up one at a time. In the image you will see multiple leds on at once, giving you verification of start and stop.
But again, I needed just one clock and built it already.
When you're identical twin gets back from their extended space travels with their clock you're going to be in for some hard truths about the nature of precision and time.
They just need to go to IIS. IIRC the astronauts are already milliseconds younger than us. Too bad hard living up there prematurely ages them far faster than milliseconds per day.
To work at sea it would rather need a way to manually enter an offset, since customarily the local time on board is entirely at the discretion of the master.
Maybe a feature request for mk V?
When I encourage/cheerlead my extremely technically talented friends to “monetize their hobbies”, I am picturing exactly this.
A hyper technical, perfectly executed, fun/whimsical/smart product. You can share all the source code and have build instructions! And sell assembled versions and make a buck! And hopefully make money / thrive/survive doing what… you love?
((The reality, ofc, is that in the hardware biz, getting more orders can just mean a ton of additional problems re logistics of making, assembling and shipping the thing. But I still naively think that’s a good problem..?))
Extraordinary Clock, truly. Cheers, congrats on launching and wishing you all the sales!
I've assembled two Mk III kits and they've been a delight each time. Instant impulse buy from me.
You just have to give the upvote for projects targeting extreme performance hardware by the DIY folks.
If I'm not mistaken, this is why HN was made.
I’m still waiting for several guys on YouTube to get their steel ball clock to display seconds.
Though their neighbors will kill them. Their prototypes are so goddamned loud.
The bike mechanic in the back of my head wants to know if you eventually filed down those bolts so they aren’t as proud of the nuts.
Really interesting to see the deep dive into PCB design and EMI considerations here. It's a great reminder how much thought goes into balancing cost, manufacturability, and compliance, even for hobbyist products. The point about using one layer as a near-continuous ground plane is especially practical, and it's fascinating how even seemingly minor layout choices can have big implications for signal integrity. Thanks to everyone for sharing their expertise—it's one of the things that makes this community so valuable.
yes
A philosophical, ontological problem, a clock or a display?
(Really nice project)
Cool project, but with this sort of power and area budget, can't we use chip scale atomic clocks ? Also, the quartz accuracy is listed as 1000 seconds to drift 1 millisecond. So that's 31 secs a year ? That's less than HAQ quartz watches (+- 5 secs a year) and definitely less than Citizen 0100 (+-1 sec a year).
The clock is synchronized to the GPS time signal. So the quartz accuracy is only for when you have no GPS signal, or between synchronizations.
Yes, I get that, but that is less interesting. Any clock can sync to GPS or NTP. It would be cooler to have an autonomous accurate clock if you are making a giant device with an ARM chip, flash storage etc.
I wouldn’t know how to obtain a clock that synchronizes to GPS. I’ve had situations in the past where it would have been useful to have such a clock for checking at a glance if the NTP server is screwing me up, or when you’re offline.
If you want a commercial GPS wall clock, Seiko makes several models. I have one - GP502W. They make others too. They are JDM models though, so you'll have to order one from Japan.
In the continental US it can be more convenient to get a clock that syncs to WWV, since it works without an external antenna. WWV is accurate to within 100ns, though you'll have propagation delays depending on your position in the US (I'm 50 miles from the station).
The reception of radio signals tends to be worse and more sensitive to your location. There was also talk of decommissioning them.
Thanks. Do you know if it automatically adjusts to the local time zone?
No, it doesn't. You need to set the offset to UTC manually. You also need to set the dates when daylight savings start and end if you live in a place that that uses daylight savings. But it runs on batteries that last for 3 years, so it's very low maintenance.
This is a good place to start: https://news.ycombinator.com/item?id=44145770
There is another project out there using a BeagleBone Black plus a small hardware RTC and a GPS module hooked up over GPIO/serial. The advantage there is that the BBB supports hardware PTP timestamping on its built-in NIC. I threw one together over a weekend, just need to design up a small box to hold it all in (though the cable mess of jumpers also made me want to design an interposer board to clean things up)
Aren't they significantly more expensive, $5k?
Feature request for Mk V: An onboard committee of three atomic clocks to ensure proper uncertainty estimation when GPS is lost.
They exist[0], but aren't cheap (~$1400 US/chip for the base model).
[0] https://www.microchip.com/en-us/products/clock-and-timing/co...
Rubidium oscillators are atomic clocks the way PhDs are doctors!
(Don't worry, I know they are considered atomic clocks, I have 2 Rb modules.)
Finally a good use case for the chip-scale atomic clock. https://www.microchip.com/en-us/product/csac-sa65
Heh, my thoughts exactly!
Those chips are also useful for getting better positions from navigation signals. One could imagine using them to help filter out spoofed signals, for example.
Beautiful write-up of taking a project from requirements to a shipping product. I wish I had any use for one at all!
Absolutely fascinating read and design process for something which I thought didn’t seem that hard to do ;)
i love mitxela and i love this project. don't know if i can stomach >250 GBP for a clock though.
I was about to buy two, but my daughter lost my debit card while running an errand for me and I had to have the card replaced.
That will take a few days, but it doesn’t matter, they’re sold out.
I’m amazed by the ambition, technical brilliance, and relentless dedication behind some personal projects on display here.
All of this for a clock! I don’t get it, but I’m in awe.
Make the clock ETERNAL with solar power to charge a small battery ;)
The attention to detail in the assembly instructions was enjoyable to read.
Is it possible to be an engineer and not think this is cool?
any chance that this clock wwould dbe available by Christmas 2025 Thanks Susan Morris USA
feature request: wifi and ntp server
How well does the GPS work indoors?
I've been testing various GPS/GNSS antennas indoors for the past year, and unless you're in a multistory building or have a full metal roof, there's a good chance you'll pick up 4+ satellites, good enough for a decent time.
But it can come and go, especially as there are other sources of interference in the 1-1.5 GHz range indoors, and reflections off various surfaces and objects make the signal way harder to distinguish.
But for by far the best chance of success, you should have the antenna outside, ideally with a clear view of most of the sky.
If inside, you could put it near a window or exterior wall, but a lot of modern construction materials can block things pretty badly.
I have an older garmin gps18x I use for ntp and it gets fantastic reception from a window shaded by trees. I wanted to test a monitor script so I stuffed it in a metal can, capped with a thick plate, and used excessive tinfoil to seal...If i recall it only dropped 2 satellites near the horizon, leaving 4+ with good signal. I suspect the signal was leaking in through the cable.
During past year running i don't believe its lost 3d fix once (not in the can)
£350, I just realized I'm poor
It’s cheaper than an Apple Watch Series 10. ;)
Damn. Out of Stock.
Wish the 2 ports were on the rear.
Best of luck.
I thought about doing precisely something like this, with a tcxo and an fpga! I’m happy to see someone else has already taken it to its logical conclusion.
The only other thing I’d add is NTP support, and perhaps I2C for driving external displays (such as a few other time zones).
Wouldn't be- a oscillator display, showing the quartz crystal oscillating, be the final display of precision? Like a pendulum, but for the digital age?
> Timezone and offset should be determined automatically from the GPS coordinates, no customisation or user interaction needed
Are there no good reasons why I might want to display time for a timezone other than the one I am currently in?
Umm.. it’s not time to the millisecond of the 100s and thousandths place smear to “88”.
I wonder if there’s some electronics hackers you can do to increase the response time on the last three elements.
It's an artifact of the camera. The camera shutter is long enough that it averages the images over 33ms. At some point in the video you can see that a high speed camera can see the correct display.
In the high speed you can see that they are distinct but you can see the rollover. Maybe mark V could tighten that up.
At the 7 minute mark the industrial 14k FPS camera shows essentially zero rollover. The earlier rollover does appear to be an artifact of the cheap consumer grade high speed camera used.
Easily my favorite part
> Auto timezone
> Possibly the most requested feature, automatically setting the timezone based on the GPS coordinates was essential. However, most people don't realise how complex this process is, at least within the restrictions I'd given myself: no internet or cellular modem, nothing beyond the GPS data, and a USB port for occasional updates.
> If we had an internet connection, sure, we could just query an online service. But I don't want my clock to require wifi.
> If we had a cellular modem, we could get the time from the cell towers, which is broadcast through a protocol called NITZ. This is how phones auto-update when you enter a different timezone, but it's somewhat unreliable depending on the carrier, and overall worse than using GPS. And I really don't want my clock to have a sim card.
> I'd also like, just for the sake of it, for the clock to work anywhere in the world, even at sea.
> This means we need to keep a copy of the map data for the country outlines for the whole world. It also means we need a full copy of the timezone database, and have the processing power onboard to make use of all this.
Hope it's not the case, but...
People with photosensitive epilepsy are affected by lights that have different flash or flicker rates from as low as 3 to as high as 60 per second.
In fact, this is one of the central design points he mentions within the first two lines of TFA: no point in a clock with a milliseconds place if it doesn’t display each millisecond legibly.
> Summary
The specifications for the clock were as follows:
* Millisecond precision, with no perceptible jitter
* Display clearly, without flicker, when filmed at very high framerates (20,000fps or more). The brightness should still automatically adjust, of course, and without the use of PWM
This reminded me how in Japan, 700 people were rushed to the hospital after watching pokemon on TV [1] due to 12Hz red/blue flashing frames, but now I understand that’s not the case here.
[1] https://www.neurology-asia.org/articles/19991_001.pdf
I’m guessing you’ve already realized this, but in case it helps clarify things for someone else:
The signals rates mentioned in the article are with respect to how frequently the segments are updated, not the brightness. (With the one exception being the colons being PWM controlled, IIRC).
PWM means Pulse Width Modulation. Controlling a (perceived) light intensity of an LED via PWM means (if we pretend for the sake of simplification that the voltage rise and fall is instantaneous, etc) quickly turning the LED fully on and fully off, varying the duty cycle to achieve the desired perceived brightness. This project avoids the flickering inherent with PWM by not using PWM: the voltage itself is set to some fixed value for a given target brightness.
(If one then wonders why this wouldn’t be the default approach for such things: microcontrollers often provide a built in means for PWM output, but it’s less common to have built in true, non-PWM analog outputs, requiring additional parts (thus more cost, more complexity) to implement variable voltage control).
Thank you for your time to explaining this. Really appreciated.
I’m missing the part where he succeeded at that. In the demo I just see #88 the entire time.
Look at later in the video: https://youtu.be/XL2cZjO5IUY?t=370
Ah. I believe the “correct” solution is you record at high speed and down sample for the big screen in order to avoid the motion blur. There are a couple scenes before he gets to the real high speed where he’s showing the clock slowed to 1/10 speed at which point the 100ths place is only a little slow to update. I’m still seeing some alpha nerd room for making the LEDs tick over faster. Maybe the Mark V.
What part does this comment refer to? It looks like they're multiplexing each section of the display at 100 kHz rate (500 kHz overall), well outside that range.
Thanks for the clarification.
I was just raising a general concern about flicker affecting people with photosensitive epilepsy, but if the multiplexing is indeed at 100 kHz per section, that’s definitely outside the problematic range.