Fw: [ntp:hackers] u-blox reference clock driver
Stromeko at nexgo.de
Sun Aug 12 07:33:36 UTC 2018
Gary E. Miller via devel writes:
> He added a u-blox driver to ntpd, and modifed the kernel to reflect the
> PPS pulse back to the u-blox. The u-blox measures the round trip time
> and provides a correction back to the driver.
As far as the analysis of the error sources goes, this is a bit on the
light side for a master's thesis.
> The results are an average PPS jitter of 222 ns!
Well, yes -- if you measure and eliminate the interrupt latency, which
by far dominates the jitter in the normal PPS processing, then the
residual jitter goes down.
The missed opportunity with this implementation is that you could
actually measure both the raw system clock frequency and the (NTP
corrected) phase independently and thus improve the convergence of the
FLL/PLL algorithm significantly by using both the rising and falling
It's also unfortunate that apparently no attempt was made to
independently measure the PPS echo latency via a TICC in order to get
some idea of how much residual offset there is.
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+
Factory and User Sound Singles for Waldorf Q+, Q and microQ:
More information about the devel