[time-nuts] Msg to N.Z. time nuts

[Hal Murray]

lists at philpem.me.uk said:
> All of which are running the SiRF 3.2 firmware, so if there is a
> firmware bug  in play, they're all going to be doing much the same
> thing... 

I'm pretty sure that all the SiRF units I was watching had essentially the 
same behavior, and that included one using RS-232 without USB.


> They all use the Prolific PL2303 serial-to-USB chip.
> That might be what's causing the timing jitter, especially if there
> are other  devices on the USB bus. USB-to-serial chips aren't known
> for accurate timing  -- a few of them buffer incoming data and then
> send it over the USB bus in one  burst. 

USB has a bad reputation, but I think it's way way overblown.  Yes, it's 
polled, but that polling is done in hardware and the time scale is 1 ms.  If 
you are satisfied with an accuracy of a few 10s of ms, USB works fine.  The 
problem is the GPS unit.

I did the experiment of running one GPS unit through a splitter into a serial 
port and also into a serial-USB converter and into a USB port.  Mostly, what 
it showed was that the low-latency serial port option on that system was 
broken.  USB was much better than the serial port.  :)

Here is a graph of the offset ntpd sees from a Garmin GPS-18-USB.
  http://www.megapathdsl.net/~hmurray/ntp/GPS18USB-off.gif
peak-peak is under 20 ms so it's possible to get that sort of timing using 
USB.


The real problem with the SiRF chips (and Garmin 18x) is that there is a 
large low frequency component in the noise/jitter.  I'd call it wander rathre 
than jitter to emphasize the low frequency nature.  It's hard to filter that 
out when the clock you are trying to correct has temperature changes which 
happen much faster than that.  Typical numbers are 100 ms wander over a 
period of 12-24 hours.
  http://www.megapathdsl.net/~hmurray/ntp/GPSSiRF-off.gif

It might be fun to see if you can get good results on a box with a stable 
clock using one of these chips.  I'm thinking of a very long time constant on 
the PLL filter.  But how long?  See the old discussions about hanging 
bridges.  So maybe the goal should be to build a hanging-bridge detector.  :)



-- 
These are my opinions, not necessarily my employer's.  I hate spam.