[time-nuts] TymServe 2100 was: The GPS 1995 problem and the Heol Design solution.

[Hal Murray]

kb8tq at n1k.org said:
> The complexity is not in the data translation, it’s in the timing of the
> whole thing. The firmware in the TS2100 was designed and tested with a
> particular order of sentences and timing between them and the pps output of
> the Trimble ACE. Upset that timing (by delaying the data) and you may up
> upset the firmware’s expectations about when the data ...

It would probably take some experimentation to figure out what fields the 
TS2100 actually uses.

Assuming you know what it needs, my expectation is that the serial data 
stream would be delayed by one character time.  Mostly, it's just read a 
character from the input UART and copy it to the output UART.  Then you have 
to watch the data stream and find the sentences you want to modify.  If it's 
just the week number, that's as simple as add 0x?? to byte ?? of sentence 
type ??.  If the date used by the TS2100 is in year/month/day format, then 
it's replace several bytes with the precomputed correct data.  There is most 
of a second to do that computation.

If there is a checksum, that will have to be corrected on the fly.  That 
shouldn't be hard.

-----

Things get interesting if the local clock used by the output UART is slightly 
slower than the clock used to send to your input UART.

The output is probably double buffered.  That extra character will support 
some clock skew.  The critical factor is how long the data stream is between 
pauses.  If the worst case clock difference is 200 ppm (100 ppm each), it 
takes a 5000 character burst to overrun a 1 (extra) character buffer.  9600 
baud is 1000 characters per second so that's a 5 second burst.

(Ethernet hubs/repeaters have the same problem.  They have to buffer up 
enough data before starting to transmit so that the buffer doesn't run dry if 
the clocks are off in one direction and the buffer has to be big enough to 
hold the extra if the clocks are off in the other direction.  Both scale with 
the max packet length.)

Another approach would be to hack the transmit baud rate to be slightly fast, 
say 9602 so it won't be slower at the worst case clock speed difference.

-- 
These are my opinions.  I hate spam.