PPS over USB

[Dan Drown]

Quoting Hal Murray <hmurray at megapathdsl.net>:
> dan-ntp at drown.org said:
>> Here's 7 days worth of data:
>> https://dan.drown.org/rpi/usb-vs-gpio-pps.html
>
> Interesting graphs.  Thanks.
>
> What sort of setup were you using?  I assume the HAT graph was using data
> from a system using the HAT to set the time.  Was the USB graph using the
> USB-PPS for timing or also running on the HAT system when it was using the
> HAT for timing?

Both systems are using their local device for timing.

> If you are using a device for timing, it's easy to get a histogram of offsets
> that will show the shape of the curve, but if you are using a device to set
> the local clock, you can't tell if your clock has an offset.

You can tell how stable your upstream clock is - if it jumps around or  
not.  That in turn will tell you how stable the frequencies (the  
frequency of the local clock vs upstream clock) are in relation to  
each other.  And your limit in measuring that frequency difference.

> For something like a PPS over USB there are two ways to see the offset.  One
> is to run on a system that has a better way to get the time, for example a
> HAT.  The other is to monitor using NTP from another system that you trust.

Phase/offset requires an external trusted system or timesource, I agree.

> If you are using a device to set the local clock, that throws away any
> systematic offset.  You need something else to see that.  That's why I keep
> bugging you to setup a PC with a good PPS.  That will give you a place to
> stand and look at other systems.
>
> In this case, plugging the USB-PPS device into a Pi will let you see the
> offset.  In that case, you are "standing" on the Pi.  The USB Ethernet won't
> get in the way.

I actually have a GPIO-PPS on the same system as the USB-PPS device:

https://dan.drown.org/rpi/usb-vs-others.html

I just set it up last night, so I don't have as much data on it.  ntpd  
(classic, v4.2.8p7) rejects it as a falseticker for some reason.  If I  
force this machine to use the GPIO-PPS, I'd expect the same local  
offsets as the other machine.  It looks a lot more noisy because the  
local clock is following the USB-PPS's frequency around.