[time-nuts] GPS for Spirent Smartbits

Hal Murray hmurray at megapathdsl.net
Tue Feb 13 21:10:52 EST 2007


>  If you have an Internet access you can get a time reference with NTP
> and even do timed captures.

> We  would like to get 2 GPS timing systems in order to test influence
> of handovers in Wimax on moving vehicles.

What sort of timing accuracy do you need?

Assuming a few ms rather than a few microseconds is good enough...

I didn't see much about timing in that blurb.  It said it provides NTP 
service, but I didn't see anything fancy about where it got the time to 
"serve".  The ntp code is normally both a server and client, so I assume it 
can get the time from the net.  You can test that by pointing it at your 
local NTP server or one out on the net.  That should get you off the ground, 
but the accuracy may not be very good.

The favorite low cost GPS unit in the NTP community is the Garmin GPS-18-LVC. 
 (There are two other models of the GPS-18, but they don't have the PPS 
signal.)  You can get them for under $100.  It requires some "assembly"[1].  
The usual trick is to steal power from a USB port.  For more info, start here:
  http://ntp.isc.org/bin/view/Support/ConfiguringGarminRefclocks
or feed >garmin GPS-18 ntp< to google.

GPS may not work in a lab full of computers.  (You can test that with a 
normal hiking GPS unit.)  The clean solution is to put the antenna up on the 
roof.  Mine works most of the time inside my house, but only if it's up high 
near the ceiling.  You may have to locate the server near a window and/or 
kludge up some longer cables or ...

I'm assuming you can find two old PCs to dedicate for an NTP server on each 
end.  It would be simpler if you could get identical hardware and software at 
both ends.  It might work if you run it on a personal machine but that will 
probably add more jitter if the user is doing anything.  You might have to 
experiment.  NTP doesn't need much CPU or memory.

What I would use as a sanity check would be to slowly send short packets from 
A to B and measure the delay.  There will be two types of delay.  One is 
speed of light (and silicon processing).  The other is queuing.  The reason I 
said slowly is to make sure the test is not generating enough traffic to 
contribute to the queuing delays.

If you make a histogram of the delays, the left edge should be the no-queuing 
case.  If the histogram in one direction matches the histogram in the other 
direction your clocks are probably good.


1] "Some assembly required" may be an American joke.  It refers to Christmas 
where the father has to assemble the new toys for the kid.  Sometimes it's 
not easy.  They usually print a warning on the box but it doesn't tell you 
how much work it will be.



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






More information about the time-nuts mailing list