Assume that the device does not have any reliable long term non-volatile memory that you can update.
In the absence of any clues your only reliable piece of knowledge is that the cold start date is somewhere after the date of manufacturing or, most often, firmware compilation date.
This is the simplest strategy - one that provides 20 years of device lifespan.  "Device" in this case can be anything, including a single chip module that only has mask ROM and 128bytes of OTP memory like most Ublox modules.

If you have NVRAM, you can store last seen full week number periodically.  Unless you don't power up the device for 20 years since last update you should be fine too.

Of course, if you have the ability to decode CNAV messages you can kick the can much further down the road.
Leo

Leo Bodnar leo@leobodnar.com wrote:

There's another relatively simple clue in the old GPS signal: the leap
second count! A device manufacturer could teach it what the leap second
count was at manufacturing time, and how to predict a lower bound on the
leap second count in the future (with a suitable safety margin / fudge
factor) which should allow it to live a bit more than 20 years.

Tony.

f.anthony.n.finch  dot@dotat.at  http://dotat.at/
Gibraltar Point to North Foreland: Northeasterly 5 or 6, occasionally 7 in
south. Moderate. Showers at first in south. Good, occasionally moderate at
first.

The idea was proposed 20+ years ago, Trimble even has a patent on it. Details here:

http://leapsecond.com/notes/gpswnro.htm

But it turns out not to work. Earth rotation is too difficult to predict 20, 40, or 60 years into the future. There was talk that the GPS receiver failures in 2015 were related to this algorithm. Look for any threads with subjects like: TS2100, TymServe 2100, 1995 rollover, Trimble ACE, Heol Design in:

http://lists.febo.com/pipermail/time-nuts_lists.febo.com/2015-May/
http://lists.febo.com/pipermail/time-nuts_lists.febo.com/2015-June/

/tvb

----- Original Message -----
From: "Tony Finch" dot@dotat.at
To: "Discussion of precise time and frequency measurement" time-nuts@lists.febo.com
Sent: Tuesday, April 09, 2019 4:08 AM
Subject: Re: [time-nuts] Garmin GPS12XL V3.51

I realize that this doesn't help old receivers, but I'm sort of
surprised that this wasn't addressed as the GPS system has had various
additions made to it such as WAAS.  Even 3-4 bits allocated to this
purpose in one of the datastreams would move us out beyond any
reasonable expectation of the life of the current GPS protocol.

Does anyone know if any of the global GNSS alternatives (aka GLONASS,
etc.) have similar limitations?

On Tue, Apr 9, 2019 at 7:01 AM Tom Van Baak tvb@leapsecond.com wrote:

--

• Forrest

Tom Van Baak tvb@LeapSecond.com wrote:

Oh, wonderful :-)

I was thinking in terms of getting a bit more lifetime than 20 years from
the firmware compilation date, but as you say it isn't reliable enough to
get much more than 20 years.

Tony.

f.anthony.n.finch  dot@dotat.at  http://dotat.at/
Biscay: Northerly 3 or 4, occasionally 5 in east. Moderate. Showers, thundery
at first in south. Good, occasionally poor until later.

Hi

The “fix” for this (adding 3 more bits) was worked out and announced back in 2004. The new
generation of sats will be transmitting a longer week number. The gotcha is that the old sats
are still up there doing their thing. They are built in a way that you can’t just shoot a firmware
patch into space and add more bits to the transmitted message.

The original plan was that the new message would be on the air well before the rollover event
that just happened. Even then, there was some debate about how many devices would be able
to use the “new” message. The whole implement / debug / deploy / patch cycle is not something
that happens overnight.

On all of our older gear, there pretty much are no patches for any of this. Doing a patch that
resets the magic “built on” date is pretty trivial compared to doing more bits in the message.
We don’t even have simple patches for this gear …..

Bob

The references I found in those archived postings explain that the problem
was that the GPS receiver had been 'fixed' in 2011 but that the fix only
worked until 2015. The solution was to replace the receiver.

I wasn't able to find any discussion of why the fix lasted such a short
time, but it seems far too short to be due to variation in the rate of leap
second insertions.

Assuming that the reason for that algorithm's failure is correct though, it
implies that leap second insertion rate might vary by at least 13
insertions in 20 years in order that an estimate of week counter cycles
produced from leap second count would be wrong.

This carries a further implication that the 7 bit count of leap seconds
will last not for 175 years but half that or perhaps worse. Or, if the
problem is with a lower than assumed rate of insertions, better.

How much variation in the rate of leap second insertion is there?

On Tue, 9 Apr 2019, 14:01 Tom Van Baak, tvb@leapsecond.com wrote:

Adrian Godwin artgodwin@gmail.com wrote:

1972-01-01 - 1992-01-01 : 17
1973-01-01 - 1993-01-01 : 16
1974-01-01 - 1994-01-01 : 16
1975-01-01 - 1995-01-01 : 16
1976-01-01 - 1996-01-01 : 15
1977-01-01 - 1997-01-01 : 15
1978-01-01 - 1998-01-01 : 15
1979-01-01 - 1999-01-01 : 14
1980-01-01 - 2000-01-01 : 14
1981-01-01 - 2001-01-01 : 13
1982-01-01 - 2002-01-01 : 12
1983-01-01 - 2003-01-01 : 11
1984-01-01 - 2004-01-01 : 10
1985-01-01 - 2005-01-01 : 10
1986-01-01 - 2006-01-01 :  9
1987-01-01 - 2007-01-01 : 10
1988-01-01 - 2008-01-01 : 10
1989-01-01 - 2009-01-01 :  9
1990-01-01 - 2010-01-01 : 10
1991-01-01 - 2011-01-01 :  9
1992-01-01 - 2012-01-01 :  8
1993-01-01 - 2013-01-01 :  8
1994-01-01 - 2014-01-01 :  7
1995-01-01 - 2015-01-01 :  6
1996-01-01 - 2016-01-01 :  7
1997-01-01 - 2017-01-01 :  6
1998-01-01 - 2018-01-01 :  6
1999-01-01 - 2019-01-01 :  6
2000-01-01 - 2020-01-01 :  6

Tony.

f.anthony.n.finch  dot@dotat.at  http://dotat.at/
Trafalgar: In east, northerly 5 or 6. In west, variable 3 or 4. In east,
moderate, occasionally rough. in west, moderate, occasionally rough in north.
In east, fair. In west, mainly fair. In east, good. In west, good.

Thanks. So the current estimate of 13 is towards the high end and variation
is large enough that 40 years with might be sufficiently low that only 1
rollover is assumed. Not likely to overflow the leap second count early
either.

On Thu, Apr 11, 2019 at 3:01 PM Tony Finch dot@dotat.at wrote:

Beyond the predictability of physical variation, there is also the unknown
probability of a human decision to abolish leap seconds, while also
continuing to satisfy the various legal requirements around the world that
civil time be tied to the Sun.
https://en.wikipedia.org/wiki/Leap_second#Proposal_to_abolish_leap_seconds

Tim N3QE

On Thu, Apr 11, 2019 at 3:01 PM Adrian Godwin artgodwin@gmail.com wrote: