Hacker News Reader

Comment by rrr_oh_man | Hacker News Reader

> The only big problems I can think of are devices that use the frequency as a reference, such as motors and clocks.

This was my March mindfuck:

That there are periods every couple weeks when the WHOLE ELECTRICITY GRID of a country is run with higher/lower frequency for a period of time to get your Chinese oven clocks back into sync.

View on HN

Long-term grid frequency stability is something you take for granted, until it suddenly fails, like in most of Europe back in 2018, due to a political disagreement between Serbia and Kosovo: https://www.reuters.com/article/idUSL5N1QP2FF/

I also find it quite funny that our cheap oven clock is probably among the most precise (non-internet connected) ones in our apartment :)

WWVB de-rating has got us all a bit down in North America. Lots of consumer level “atomic clock accuracy” devices just don’t update any more.

https://www.watchuseek.com/threads/wwvb-operating-with-reduc...

It’s a shame there aren’t more time tower sites broadcasting in a distributed manner for redundancy.

Noted my car got time from GPS the other day and it got me thinking, why aren't there unencrypted packets on all wifi that distribute NTP time so my oven/microwave/alarm/etc can grab it from any wifi it can see (even if it's not paired to them) - bonus points if APs also hand out local TZ data

Phone networks can already do this. And its true broadcast, you can decode it passively without ever connecting to the network, so no SIM/transmission required.

I have no idea this is actually enabled in networks in practice, but the specification exists.

And in my experience, it's often wrong on some towers. I can't enable the setting on my phone to automatically set the time from the network, because if I do, it gets several minutes out of sync with the real time, which breaks TOTP generation.

Yeah, a large German cell provider had their time off by a couple of minutes some years back, which made the news as it resulted in people's alarms ringing late and them being late to work/school etc.

Hopefully modern phones prioritize GPS or NTP derived time and timezone information over NITZ (which is how networks provide it). At least Android has the capability to do so, but it's ultimately up to device manufacturers, as far as I understand [1].

[1] https://source.android.com/docs/core/connect/time

Could the detection and reporting of this inaccuracy be crowdsourced, either with mobile devices or perhaps SDRs? Fetch tower time, fetch real time over GNSS or NTP, report delta with tower ID and drift to a central server? Perhaps even include a map, like the map of GPS error rates derived from ADS-B data feeds.

I'm almost certain that at least Apple already has such a database on the operator level, i.e. a list of MNCs and the (non-)trustworthiness of NITZ. I vaguely remember seeing some carrier settings keys to that extent somewhere.

Why bother? If you can download data from a central server you have an internet connection, so you can also retrieve time via NTP instead.

Distributed monitoring and centralized reporting would be used to correct anomalous towers providing bad time and can provide independent verification of precision. It is not enough to provide a service; you must also have a system to validate it, and correct when needed.

Then you'd need Internet access to get the data, and if you had that, you could just use NTP.

Unencrypted broadcast NTP timestamps sound like a good idea until your neighbor starts broadcasting false timestamps to prevent your loud smart leaf blower from charging overnight :)

Best practice would be to listen to all the local APs and choose what most of them say

Making one AP look like thousands is trivial.

Like the 6 independent 24-satellite (plus spare) constellations that constantly orbit Earth and output synchronized beacons accurate to nanoseconds?

How effective is receiving precision time from a GNSS constellation within structures, urban areas, and other challenging SNR envs? Another comment mentions cell towers broadcast this unencrypted and without the need for service, so perhaps the solution is to instead require strong SLAs around that service, with the gov potentially paying for that service level?

https://news.ycombinator.com/item?id=40388421

Yep, reception is a real problem with straight GNSS. WWVB suffered from it too, and cellular (which is locked to GPS) is a big upgrade. I suppose that now the thing to do is wait for the $0.10 cellular clock chip that makes it easy.

My interest is as a bit of a precision time enthusiast and an advocate of public goods. We’re awash in “good time”, GNSS, RF, NTP over IP, cellular, but it still feels like there is a gap to fill to make precision time super cheap and ubiquitous. Getting time over mains power or other previously convenient but potentially low precision sources should be over, but getting precision time should be as easy and reliable as mains power. Thanks for riffing with me!

I feel like once you're concerned about SLAs for precision timekeeping, you need to start worrying about spoofing as well, which is tricky to combat in an inherently unidirectional protocol.

Galileo has an interesting signal authentication service [1], but even that is subject to replay attacks by definition (although you can't skip ahead a receiver's time).

In comparison, networked bidirectional protocols like NTP are almost trivially secure-able, by including a challenge in the request and a cryptographic signature over the response.

[1] https://berthub.eu/articles/posts/galileos-authentication-al...

It's probably about time for an update that can work with more compact receivers.

Governments(Or an industry group, or really anyone) could publish digitally signed timestamps every second with a a key that doesn't change, and you could have a BLE profile for sharing them.

There used to be a local hydro generation that fed the local city, they had 2 clocks on the wall, one on the local grid and one on the national one - some time late at night they'd speed up/slow down the local grid to keep the clocks in sync

It's surprisingly hard to keep a clock accurate over time. Typical crystal oscillators can gain or lose seconds per day. If you want to do better than that, you either need internet access (for NTP), a radio receiver of some kind (for atomic clock broadcasts, GPS, etc.), or a significantly more expensive oscillator. Having a frequency reference with perfect long-term accuracy that any electrically-powered clock can use for free is pretty convenient!

I've often wondered how hard it would be to make portable battery-powered clocks use 60 Hz noise as a timekeeping source. Maybe I should make a project out of it.

> Typical crystal oscillators can gain or lose seconds per day.

Are you sure?

I only owned a quartz watch when I was a kid, but all sources I find on the web say even the lowest quality quartz watch movements are accurate to +- 20-30 seconds a month.

For comparison: my several decades old manual Poljot loses some seconds per day.

The Casio G-Shock I got in the 90s was ~66 ppm fast, which worked out to ~5 seconds/day. There are cheap crystals on Digi-Key with ~10 ppm accuracy, which is around 0.86 seconds/day. But crystals drift with age (and a tiny bit with temperature). Perhaps watches have improved since I last looked, but I would want to measure the actual real-world accuracy before believing anything better than 1 second/day.

Yeah, that sounds about right. Crystal oscillators typically have an accuracy of about 20ppm, which is 1.73 seconds per day.

Keep in mind that those number are a guaranteed worst-case performance - real-world performance is often much better.

how more expensive is a "more expensive" oscillator?

A few cents, as far as I understand, but that's not the only problem:

You can get significantly more precision using an oven-controlled oscillator (frequency is temperature-dependent, and it's easier and more precise to control temperature to a known high setpoint than to compensate for the temperature-frequency curve using a thermometer), but these are, as the name implies, small ovens that consume additional electricity (on the order of 1-2 W, as far as I know).

An additional base load of 1-2 W per oven and microwave would be pretty bad for the energy budget on a global scale, so just providing a time source from the power line they're already connected to is more efficient :)

Other than oven-based oscillators, you can also use temperature-corrected oscillators; the Apple Watch does that, for example, and is supposedly about four times more accurate than a non-temperature-corrected MEMS, which in turn is already better than a quartz oscillator [1].

Not that it matters much, given that all models have at least a GPS receiver these days, as well as a battery life of about a day without charging, but I suppose it's a real benefit to all Apple Watch users working in faraday cage isolated bunkers without Wi-Fi :)

[1] https://www.watchuseek.com/threads/apple-watch-oscillator-an...

> A few cents, as far as I understand, but that's not the only problem

Nitpick: looking on Digi-Key, in quantity, a sub-ppm oscillator is ~5x more expensive than a 10 ppm crystal -- ~85 cents for the cheapest TCXO I can find vs. ~18 cents for a crystal. The frequencies are also much higher (tens of megahertz), which won't help with the power consumption.

You're right about the power consumption, though.

It depends. On the lower end of the spectrum, there are high volume products that offer somewhat better performance for just a tiny increase in cost, as mentioned in other comments. On the higher end of the spectrum there are laboratory-grade oscillators, for example the legendary Hewlett-Packard 10811 [1] and similar, which cost hundreds to thousands of dollars. These overlap in price with the lower end atomic clocks.

[1] http://etoysbox.jp/Memo/3_Test_Equipments/HP_10811_OCXO/HP_1...

The problem is that grid frequency has terrible mid-term performance. The frequency itself will be reasonably close to 50/60Hz, but it's guaranteed to be off by a little bit due to the way it operates - and that small offset accumulates over time.

For example, the European supergrid only makes time corrections if it has drifted by more than 20 seconds. That's good enough for the clock on your microwave, but it's pretty bad compared to literally any other time source.

Wow, that's pretty bad! In Texas it's +/- 3 seconds[1]. Which I wouldn't want to use for precise event logging, but for human use is fine.

[1] http://www.naesb.org/pdf2/weq_bklet_011505_tec_mc.pdf

I wouldn't be surprised if you could do it in software alone, using a wire on an ADC as the receiver

Reply on news.ycombinator.com