General > General Technical Chat
Clock circuit design idea
SiliconWizard:
--- Quote from: Connecteur on July 25, 2020, 06:34:44 pm ---
--- Quote from: m98 on July 25, 2020, 05:33:02 pm ---That's just not how this works. Clocks don't only drift from a constant frequency offset.
--- End quote ---
Seems to me whatever the cause, it could be mitigated by this simple time-correcting solution.
--- End quote ---
If it's kept in relatively constant conditions (temperature, RH...), that would sort of be OK I guess. But in real life conditions, that's rarely the case.
I don't really see a point these days though - even a good modern, but still not expensive TCXO, which is not boutique stuff at all, can get you less than 0.5 ppm drift over a large range of conditions, which would equate to approx. 1.3 s drift over a month, without any adjustement needed. I kind of doubt the scheme you suggest could get anything close to that, and it would require more "maintenance".
Of course you can get better than this yet, but that's quickly much more expensive (OCXOs, atomic clocks, etc.) and draws a lot more power.
Overall, the idea of a "learning" process is right only 1/ if the conditions that can influence the clock freq. are stable, or at least vary in a predictable, more or less constant way, or 2/ if your learning "algorithm" can actually take into account A LOT of parameters and compensate accordingly: that would mean using temperature, probably RH (depending on the clock generator), maybe pressure, vibrations... then also take into account how the clock generator reacts to these parameters... then feed it to some kind of neural network maybe. Eek. And even with that amount of sophistication, I still doubt you could get better than even a cheap TCXO. I'd be curious to see something like this implemented though, as an exercise.
Connecteur:
I don't know if I'm the only one, but my expectations for accuracy for clocks have been raised, especially when things like phones and smart TVs provide accuracy to the second all the time. When I set my AV equipment to turn on a program at a certain time, it's disappointing to miss the first minutes sometimes because the clock has lost time. It's not only that, but it's a bit annoying to be late for an appointment because your car clock is 10 minutes off because you never paid attention to it until that time you forgot your phone at home. No one is asking for the moon, but I think it would be great if some of the cheap clocks that seem to be everywhere nowadays could be more accurate, or at least self-adjusting with a simple program algorithm.
SiliconWizard:
--- Quote from: Connecteur on July 26, 2020, 06:44:29 pm ---I don't know if I'm the only one, but my expectations for accuracy for clocks have been raised, especially when things like phones and smart TVs provide accuracy to the second all the time. When I set my AV equipment to turn on a program at a certain time, it's disappointing to miss the first minutes sometimes because the clock has lost time. It's not only that, but it's a bit annoying to be late for an appointment because your car clock is 10 minutes off because you never paid attention to it until that time you forgot your phone at home. No one is asking for the moon, but I think it would be great if some of the cheap clocks that seem to be everywhere nowadays could be more accurate, or at least self-adjusting with a simple program algorithm.
--- End quote ---
That's because most cheap clocks use crappy oscillators. You wouldn't need anything really fancy. I gave some figures - you can expect a few seconds drift a year with a decent TCXO, but many cheap clocks have more like 100 ppm oscillators, and it can get worse than this in cars due to large temperature variations. And why is it so? Cost mainly. Even though a TCXO is not expensive these days, it'd still add a few bucks in the BOM, in a world where we shave off every cent. Secondly, because manufacturers likely don't even care - probably because in turn customers have been used to get crap clocks in common devices including cars, so they are not expecting any better.
A related thought regarding the market - what manufacturers sell, what consumers expect and buy: the clock ("watch") industry has followed this trend for a long time. Very accurate clocks with low drift have been considered luxury items for a long time, and often come at a premium. So in a car for instance, you're more likely to find an accurate clock in a luxury car.
Of course, as with mobile phones, as soon as some given device has GPS or an internet connection, it can get very accurate time. More and more recent cars come with some kind of internet connection, and it will likely increase in the future. Likewise, AV gear increasingly gets internet access. So I guess everyone will definitely consider this a solved issue, which may worsen the situation further as to getting low-drift, but fully autonomous clocks in the future.
duckduck:
--- Quote from: David Hess on July 26, 2020, 03:13:16 am ---
--- Quote from: Connecteur on July 26, 2020, 03:09:20 am ---I'm sure "close enough" is the reason we don't see it incorporated into most clock circuits today, but why couldn't it just be part of normal electronic clock design?
--- End quote ---
The first thing which occurs to me is that it requires state to be saved unless it is acceptable that the calibration be lost if power is removed. Saving state will mean some floating gate memory and nobody is going to build a cheap clock IC on a process which provides that, or add a second IC.
--- End quote ---
Amazon will sell me an "Atomic wall clock" delivered to my front door for US$20. When the shortwave time-keeping stations are shut down, your idea may find a place in non-internet-connected locations if dedicated GPS-set clocks are still fairly expensive, as they are now.
There's at least one utility that does time trimming via software for the built-in clocks in personal computers. The software doesn't actually modify the hardware, but uses software to modify the time based off of comparisons with NTP. I remember using this around the year 2005 or so. Nowadays I can't be bothered and everything is internet-connected anyways, right?
Speaking of, the head of IT Security at my company has a friend with one of those stupid internet-connected refrigerators. Every time security guy visits his friend's house, he makes sure to browse to a certain popular NSFW video website and leave the browser displaying it when he leaves.
Connecteur:
From my point of view, in the year 2020 there should be no excuse for clocks that don't keep time. It's fine if the device is already connected to the internet or shortwave; those may be the simplest solution, but for independent devices, self-correcting circuitry should be standard. Who can remember how to set the time on a device you got 6 months ago?
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version