Hacker News Reader

Comment by gwbas1c | Hacker News Reader

Dumb question: What is the rollover issue?

Not a dumb question at all.

GPS used ten bits for datetime. With a week counter and seconds into that week since 6th of January, 1980. After 1024 weeks it resets back to 0. This has happened once already on 21st of August, 1999. This is the second rollover/reset of the week counter on 6th of April, 2019.

Newer versions of GPS messages use thirteen digits instead of ten digits allowing for a larger week counter (granting 8,192 weeks), causing the first rollover not to occur until the year 2137.

https://en.wikipedia.org/wiki/Global_Positioning_System#Time...

Instead of adding 3 bits, maybe they should have dropped 3 bits. A rollover every ~20 years is just asking for latent buts to blow up. A rollover every ~2.5 years is just business as usual. Go big or go small. Don't design things to break just after you retire but before your coworkers do.

OTOH, with the current setup a receiver will know the correct date as long as it is turned on at least once every 1024 weeks, since it can detect the rollover. More frequent rollovers might cause issues with timing on devices that are infrequently turned on. It's sort of a moot point for most of us once there's greater adoption of the L2 and L5 CNAV signals— those use a week counter that won't roll over until most if not all of us are dead. Then it can be someone else's problem :)

> OTOH, with the current setup a receiver will know the correct date as long as it is turned on at least once every 1024 weeks, since it can detect the rollover

Actually, probably every 512 weeks, as it needs to know (or assume) with side of the +/- it is on:

* https://en.wikipedia.org/wiki/Serial_number_arithmetic

* https://en.wikipedia.org/wiki/Date_windowing

* https://docs.ntpsec.org/latest/rollover.html#ntp_pivots

I believe that's what GLONASS used as a strategy:

Roll over so often that firmware/software implementations pretty much have to take it into account.

> GPS used ten bits for datetime.

Nit: GPS uses 29 bits for "datetime".

10 bits for the week since the epoch, and 19 bits to count the number of 1.5s intervals that have passed since the start of the week. Yes, 1.5s intervals.

See Section 2.3.5:

* https://www.gps.gov/technical/ps/1995-SPS-signal-specificati...

I had to look it up as well. I found this link: https://www.orolia.com/resources/blog/lisa-perdue/2018/gps-2... From the text comes: "However, the field that contains the week number is a 10-bit binary number. This limits the range of the week number to 0 – 1023, or 1024 total weeks."

https://news.ycombinator.com/item?id=19593227

Reply on news.ycombinator.com