[time-nuts] BC637PCI 1024 week rollover

[GandalfG8 at aol.com]

In a message dated 10/02/2014 21:56:25 GMT Standard Time,  
magnus at rubidium.dyndns.org writes:

On  10/02/14 11:15, GandalfG8 at aol.com wrote:
> Ah, I took 1999 as I thought  that was the only relevant date for another
> 1024 weeks, I'm not  familiar with the shifted 1024 week period so will 
take a
>    look at that.
>
> Does "shifted" imply a shift at the whim of the  manufacturer, ie  could 
it
> explain why these boards might have  been ok a few years  ago but not now?

Yes. We have seen week 500  and week 512 occuring.

Considering this simple code:

if (gpsweek  < 500)
gpsweek += 1024;

This means that GPS week  500 to 1023 maps straight and truncated GPS 
week 0 to 499 is mapped to GPS  week 1024 to 1523.

However, when GPS week 1524 occurs, GPS week 500 is  transmitted, so 
receivers jump from GPS week 1523 to GPS week 500 and the  NMEA readout 
date jumps 19.3 years. Woops.

The interesting thing is  that the GPS otherwise operate properly, as it 
is only the read-out date  which goes wrong, not the internal gears of 
the GPS, so the leap second  applied will be the current and not the one 
from 19 years  ago.
-------------------------------------
Yes, that's what I was seeing, anything received by the GPS module was  
passed through correctly, week number, leap seconds, etc, it was what the BC637 
 did with it after that wasn't quite so helpful.
-------------------------------------



> Oh dear, I think a wee light bulb has just  exploded:-)

Good. :)

> I haven't checked this yet, but if  shifting means to  start a 1024 week
> period that's approximately  from or not too  far before the date of
> manufacture, either for  individual units or just as  a ballpark for a 
given production
>  run, that would buy them nearly twenty  years from then, which would  
mean
> these boards should still be ok.

It's arbitrary. It could  be from writing the code to just before a 
certain batch. Who knows.  Adjusting it is trivial.

> If shifting means to do this say at the  design stage or starting with the
> first production run then they might  buy twenty years from then but
> regardless of individual manufacturing  date.

It's arbitrary. Considering that GPS week 500 and GPS week 512  have been 
found in equipment, and these are not "random numbers", it seems  like a 
random pick early in the design.

> I'm not too sure that  even the earliest of these boards should be twenty
> years old yet, but  if plan Z was to stick with some previously picked
> arbitrary   date, such as company formation or granny's birthday, then 
that might
>  well be  the answer:-)
>
> Thank you, will definitely look  more closely at this, perhaps it's not 
time
>   yet to put the  boards back into hibernation after all:-)

Good, now you learned  something. :)
------------------------
Certainly seems that way, perhaps the old brain cell does  still fire up 
now and again after all:-)
 
I was quite surprised though just how little a Google search threw up on  
1024 week offsets, however I worded it I got plenty of hits regarding the 
1024  week rollover itself, plus its implications, but virtually nothing  
regarding the use of offsets and any consequences of that.
-----------------------
 
 



> I agree re the TMS29F010, and I'm sure I could read  it, but would
> definitely need an adapter for that.

Ah.  Yes.

I don't know what FW my boards have, if it has the GPS FW latent  or not.
----------------------------
I bought a set of PLCC adapters on Ebay this afternoon, probably about time 
 my programmers joined the 19th century, so with a bit of luck, a following 
 wind, and a good head of steam, I might even have a dump of the  firmware 
by the weekend:-)
 
Regards
 
Nigel
GM8PZR