RasPi + NTP + PPS.

[Remco]

I followed both the cookbook recipes from here and here.

For whatever reason ntpd here does not acknowledge my
PPS timestamps when using the ATOM reflock (22.0).

All postings I have seen deal with PPS in conjunction with
(mostly) a NMEA GPS.

Setup here: another machine (preferred peer) delivers a
UTC locked PPS (Motorola Oncore).

With ppstest /dev/pps0 I get time stamps, from which I conclude
that pps_gpio in the kernel works well.

When starting up ntpd, it generally takes a while before the ATOM
refclock is taken into consideration, and this appears in my ntpd.log:

4 Oct 11:37:39 ntpd[24325]: refclock_params: time_pps_setparams: Invalid argument

Any hints/tips?

[Remco]

While the moderation of this topic was pending, I found out
that it was the fact that I use the falling edge of
the UTC locked PPS (flag 2 = 1 instead of 0).

I modified my ntp.conf (thus flag2 = 0) and yes.. after a while:

Code: Select all

ntpq -p
     remote           refid      st t when poll reach   delay   offset  jitter
==============================================================================
oPPS(0)          .PPS.            0 l    4   16   17    0.000  -198.80   0.049
*Time.remco.org  .GPS.            1 u   61   64   17    0.789   -0.562   0.282
+ntp.nmi.nl      .PPS.            1 u   62   64   17   20.043   -3.769   0.137

I get a lock, but it's the wrong one (i.e. the pulse width) around 198.8 ms here.

Being not a C-programmer at all, I looked into the GPIO kernel patch and see the line:

Code: Select all

+	.assert_falling_edge = false,

Might this be the reason that the falling edge of the PPS pulse cannot generate an interrupt?

With ppstest /dev/pps0 I overlooked the fact that I do not have a 'clear' time stamp.

Code: Select all

sudo ppstest /dev/pps0
trying PPS source "/dev/pps0"
found PPS source "/dev/pps0"
ok, found 1 source(s), now start fetching data...
source 0 - assert 1380886732.198747532, sequence: 99622 - clear  0.000000000, sequence: 0

A quick fix would be an inverter (transistor) or perhaps someone can integrate usage of the falling edge
of the PPS pulse in the GPIO bcm2708.c patch?

[Remco]

Couldn't wait, so I made a hardware mod. Picture is here.

The inverter has two purposes:
- of course invert the signal
- level the signal to 3.3V

Result after the first lock:

Code: Select all

ntpq -p
     remote           refid      st t when poll reach   delay   offset  jitter
==============================================================================
oPPS(0)          .PPS.            0 l    6   16    7    0.000    0.002   0.003
*Time.remco.org  .GPS.            1 u   50   64   17    0.775    0.013   0.073
+ntp.nmi.nl      .PPS.            1 u   47   64   17   20.284   -3.411   0.291

Looks good!

[Remco]

PLL offset of my Raspi* as ntp pool server
(serving several >1000 clients) is depicted below:



I run ntp servers for a long time but I never saw such (good)
performance of a 'clean' server, i.e. no thermostat on the CPU clock.

*Note: picture shown is actual and may vary on
my experiments with (several/port forwarded) ntp servers.
When it's 'Linux/3.6.11', it's the Raspi ; -)