Re: [gpsd-dev] Mysteriously vanishing bugs don't make me happy

[Greg Troxel]

Andy Walls <address@hidden> writes:

> On Mon, 2013-11-04 at 16:14 -0500, Greg Troxel wrote:
>> Have people been seeing jitter low enough that microsecond quantization
>> looks like a real issue?
>
> AFAICT gpsd only deals with PPS sources associated with serial port
> control lines.  I'm going to guess that those sources have jitter in the
> +/- 1 microsecond range at best.

The reality should be "gpsd will use any PPS source as long as it can be
obtained via the RFC2783 interface", and I think that's true on Linux,
and will be true on BSD after a few hours of detangling.

> If gpsd would allow "arbitrary" PPS sources to be associated with a GPS
> serial stream, then maybe the precision handled by gpsd would need to
> stretch into the nanoseconds.

It probably should, but the OS may need a way to express this.


> As an example of numbers:
>
> I have a counter-timer capturing a 32-bit count every PPS rising edge.
> It uses a 13 MHz clock, so that's a 77 nanosecond period and
> quantization in the time of the PPS edge.  My GPS unit, when stationary,
> has PPS jitter of +/- 75 ns at 3 sigma (99.7%).  I.e. I'm on the cusp of
> being able to detect the PPS jitter.
>
> In measurements, I observe +/- 77 nsecs of jitter in that PPS capture,
> which I will attribute to sample alignment and the (in)stability of the
> 26 MHz crystal.  I just don't believe I'm observing that much of the
> 0.3% of PPS jitter that I could possibly see.

Sure, but you are in the regime where 1 us of quantization would hurt
you in observing the source.  (Whether it would impair the accuracy of
your system clock in any way which could be detected by *using* the
system clock is another story.)

> This example of 77 nanoseconds resolution really doesn't matter to
> gpsd's implementation, unless gpsd lets users associate an "arbitrary"
> PPS device node with a GPS serial stream.

That's just a small extension away.


I was curious whether normal PPS on (real serial) DCD got down under 1
us.  I wasn't trying to argue that the code shouldn't handle ns; it's
always better to have the representation be fairly far away from the
edge.

pgp3lT7Dl_MRU.pgp
Description: PGP signature