[time-nuts] RasberryPi, timing and GPS receivers

[shaun at impsec.net]

Michael Tharp <gxti at partiallystapled.com> wrote


[RE: raspberry pi ntp server]
>
> It's not a terrible idea, but the RPI has a USB ethernet transceiver so 
> in addition to the latency/jitter of the ethernet it also has the 
> latency/jitter of the USB.
I've been playing with a pair of Raspberry Pi ntp servers, one with a 
$15 Sure Electronics GPS evaluation board, which was very easy to 
interface to (3.3v uart and pps outputs) and the other connected to a 
Lucent RFTG, which once I figured out that the Pi apparenty tolerates 
the PPS out from the RFTG w/o a level shifter just fine (mostly - I was 
getting false extra pulses until I shielded the line, which I hadn't 
needed to do on the Sure GPS board).

I'm a junior time-nut at best but it looks to me like jitter from the 
USB Ethernet is acceptably low, based on ntpq -p anyway:
 
    remote           refid      st t when poll reach   delay   offset  
jitter
==============================================================================
oGPS_NMEA(0)     .GPS.            0 l    2   16  377    0.000    0.003   
0.004
+pi2             .PPS.            1 s    7   16  377    0.945   -0.004   
0.044
+pool-test.ntp.o 216.218.192.202  2 u   43   64  377   73.855    1.382   
0.262


    remote           refid      st t when poll reach   delay   offset  
jitter
==============================================================================
oPPS(0)          .PPS.            0 l    1    8  377    0.000   -0.024   
0.061
+ntp             .GPS.            1 s   14    8  252    0.983    0.217   
0.615
*tick.nullmodem. 128.252.19.1     2 u   60   64  377   35.066    0.562   
4.556


Some of the offset between the two might be that I am waiting for a 2nd 
GPS antenna to arrive for the RFTG and it's been undisciplined for a few 
days, it might be from the ethernet jitter, the Sure GPS might be off, 
or it might be from the fact that I may have a configuration issue 
someplace - ntpq -c rl has them at different precisions (-19 and -18), 
even though I think they're the same kernel and ntpd compile. 



>  I've also heard of stability problems just 
> keeping it running for weeks to months so you should integrate some kind 
> of watchdog timer if you can. 
I haven't seen the issue yet but I haven't been running more than a 
couple of weeks.  Part of the problem may be quality of power supply, 
I've seen some people reporting issues if the 5v usb power in isn't 
capable of providing enough current and a stable voltage.

>
> You will want to house the RPI and GPS receiver in a box where it will 
> not be subject to wide temperature swings,

I haven't been running either of mine in an enclosure, both are sitting 
up in the unheated attic at the moment, typical offsets from loopstats 
seem to be under 2us, but I should graph them. 

> It would also be interesting to upgrade the main oscillator to a 
> temperature-compensated model so NTP doesn't have to work as hard to 
> keep the frequency locked.
I am curious if I could replace the 19.2mhz system clock crystal with a 
stable source generated from the 15mhz RFTG output via a clockblock, ala 
what I have seen reported about the Soekris.
> Personally I would recommend getting a more robust single-board 
> computer
I don't disagree that the Pi has its flaws but it has been fun to play 
with.

-shaun