Re: [gpsd-dev] PPS over USB

[Ed W]

On 07/05/2012 14:29, Eric S. Raymond wrote:
> Ed W<address@hidden>:
> > On 07/05/2012 04:20, Eric S. Raymond wrote:
> > > Hal Murray<address@hidden>:
> > > > Is there a list of serial-USB chips that are known to work and/or not work?
> > > You can generate one by grepping for TIOCMIWAIT in the driver tree.  Gary
> > > did this as one point.  The PL2303 is defintely good, the CP210x bad.
> > Apart from PPS signals, does TIOCMIWAIT improve gpsd accuracy in any
> > other way?  Phrased another way are CP210x any worse than a generic
> > PL2303 device?
> You haven't been following the thumbgps list, I see.  SiRF-III chips
> have a 100-170ms long-period wobble in their message timing, observed
> by Hal Murray and Gary Miller and myself. That's what PPS fixes,
> giving top-of-second to 1ms.
>
> > I couldn't get a definite opinion, but CP210x devices appear to be
> > USB2? I currently see around 1ms jitter on my device using just the
> > NMEA clock - perhaps I ought to be expecting lower than this?
> That's news to us.  What GPS chip is in your devices?  If it's
> Venus 6 that's interesting - suggests they don't have the timing
> wobble that SiRF chips do.

Yes it's a venus 6.  Resend of the link:
    
http://www.bluenext.co.uk/bluetooth-products/gps/92-bn-903s-gps-dongle.html

They cost around $15 or so.  Perhaps they can be had for less in 
quantity ($10 might be a reasonable target)

A little further research suggests that the CP210x is a "full speed 
12mbit USB2" device.  Google suggests that "full speed" devices are 
still polled at 1ms resolution and that you need a "High speed" device 
to achieve 125us resolution.  As such the expected jitter is at least 
500us, ie the normal 1ms jitter.

I see chrony state a 1ms jitter from the NMEA time, which is a bit 
higher than the expected 0.5ms jitter, but I haven't looked into how 
chrony computes it's calculation.  Also it's possible that there is some 
gremlin due to the way GPSD computes the NMEA time?  (eg I thought the 
algorithm might kick in once a whole sentence was read and reversed out 
the length of the sentence to get the start timestamp - if the sentence 
was read over two 1ms interrupts then we might acquire an additional 1ms 
jitter that could be eliminated if we used the sentence start timestamp, 
not the end timestamp? - ready to be told this is wrong...


> > Big picture question - Am I going to get better accuracy using your
> > new device than I am from these CP210x Venux6 devices?
> Maybe.  Gary is actually measuring 0.5 ms jitter.

I guess I need to go find those results - however, might Gary mind 
resending his results and how to compute them here?  For reference I 
have been feeding my results to chrony and examining the statistics from 
that.  Probably ntp computes it's statistics differently?

Thanks

Ed W