[ntp:questions] Frequency adjustments in a local network

Maximilian Brehm maximilian.brehm at tu-ilmenau.de
Mon Apr 7 14:36:10 UTC 2014



David Woolley schrieb:
> On 07/04/14 10:43, Maximilian Brehm wrote:
>>
>>
>> David Woolley schrieb:
>>> On 04/04/14 12:44, Maximilian Brehm wrote:
>>>> Are there ways to achieve this using NTP that are implemented
>>>>> or may be implemented? If yes, can you recommend me starting points for
>>>>> either way?
>>>
>>> I guess you could treat the reference system as a PPS source, but
>>> there is still the problem that PPS assumes that the pulses are on the
>>> absolute time seconds not at a variable and drifting offset from them.
>>
>> Is a PPS signal not supplied via a hardware link? How would I simulate
>> the signal using the received timestamps?
> 
> No.  It is supplied by a software driver that gets it from a hardware link.
How would you do that? In an external program that reads the
timestamps, generates the PPS signal, and passes it to ntpd via shared
memory?
> 
>>
>> I believe I read somewhere that there are PPS systems that do not
>> supply a timestamp and an external NTP server has to be used to
>> synchronize time. Couldn't the implemention just ignore the timestamps?
>>
> 
> As far as I know, the timestamps are processed independent of the PPS feed.
> 
> Where you will have difficulty is that there is an assumption that
> there is some contant relationship between the timing of the
> timestamps and then PPS.  This often has quite large jitter, but the
> two don't drift relative to each other.  If you have a free running
> PPS source, that will not be true and ntpd will get confused once the
> drift exceeds a certain fraction of a second.  Also, the PPS signals
> will define exact seconds, so your driver will have to calibrate
> itself each time it is started, and you will have problems once the
> drift makes that calibration unreliable.
Couldn't I recalibrate the offset during runtime, using a state machine
for instance (for instance, with states recalibrate and reference)?


More information about the questions mailing list