[ntp:questions] ESR looking for good GPS clocks

unruh unruh at invalid.ca
Tue Mar 6 18:54:47 UTC 2012

On 2012-03-06, Rob <nomail at example.com> wrote:
> Ron Frazier (NTP) <timekeepingntplist at c3energy.com> wrote:
>> I haven't been following this thread extremely closely, but I did read 
>> ESR's blog post and exchanged a few emails with him.  He says SIRF GPS's 
>> exhibit a "wobble" of variance of outputting the NMEA time of 170 ms or 
>> so.  I've personally observed with my USB BU-353 this effect, where the 
>> Meinberg Server Monitor shows my computer locked into the GPS time 
>> within a few ms and a number of internet servers that I have programmed 
>> will show offsets of say 50 ms.  However, when I originally set the 
>> fudge factor, I had almost all the internet servers showing single digit 
>> offsets when the PC is locked to the GPS time.  Sometimes, it shows a 
>> positive offset for the internet servers and sometimes it shows 
>> negative.  Anyway, ESR says SIRF GPS's aren't suitable for timekeeping.  
> This is not correct.  The SiRF itself does not show this wobble.  If
> it would, it would also show a position inaccuracy of 170ms * c meters,
> and it would not be usable as a GPS.

The nmea sentences are part of the SiRF module. I agree that it
certainly determines the time internally to far better than 170ms (ie,
the circumpherence  of the earth) but it, not something else, reports in
its nmea sentences with this wobble. That is part of the sirf module. 

> What is showing this wobble is the output of NMEA sentences.
> NMEA is unsuitable for timekeeping, because the time in the sentence
> is not the current time, but the time at which the GPS last calculated
> a position fix.  The time delay between position calculations and NMEA

If the report came a set time after the determination, that would not
matter. Ie, if the nmea sentence started exactly 573.124ms after the
determination, it would not matter that it reported late. ntpd could
fudge that offset away. If that time varies by 170ms however, tht is a

> output is varying, because these two functions are carried out by separate
> tasks in the multitasking OS that is running on the processor of the
> GPS device, and there is no tight synchronization between them.
> All of this is not determined by the GPS device, but by the method it
> is being (ab)used in the setup, i.e. by using NMEA to convey timing.

Sure it is determined by the GPS device. It is, as you admit, the gps
device that allows there to be "no tight synchronization between them".
Had the designers decided that such a tight synchronization was
something to be aimed for, I am sure they could have done a whole lot
better. Probably to better than 1ms.

> SiRF chips should be used in SiRF binary mode.  ESR should know that.

??? Are you claiming that the reports are better? You have tested it?

More information about the questions mailing list