[time-nuts] M12+T sawtooth correction and additional filtering

[Azelio Boriani]

Tom,
very interesting. The Kalman filtering, as long as I know, requires a
model: you apply the model, make the measurements, compute the delta
between the model and the measurements and then the new gain to take the
next step. Let me ask: is this model to be determined? From your message it
seems to me that this model is already available waiting for someone to run
it. Don't look at me for this, only curious about the availability of such
a mathematical model. Actually I'm more interested on how to model an OCXO
to apply the Kalman filtering on such a device.

On Wed, May 9, 2012 at 1:19 AM, Tom Van Baak <tvb at leapsecond.com> wrote:

> Jerry,
>
> It depends on what your goal is. Hardware sawtooth correction is
> good at improving the short-term jitter of the 1PPS output.
>
> But if you're using the M12+T with a TIC simply to measure the
> performance of some time/frequency standard, then software
> correction is easier and gives slightly better results. This assumes
> you have a PC or equivalent, and that you like coding more than
> soldering.
>
> You'll note that both Trimble's Tboltmon and Rick's Tac32 programs
> allow you to combine sawtooth corrections and TIC readings into
> a single value by using two serial ports.
>
> Note there is a point where averaging lots of raw 1PPS data and
> sawtooth corrected 1PPS data is pretty much the same. Before
> you jump into this -- try a minute, an hour, and a day of data and
> see what you get.
>
> But there's more...
>
> As long as you're playing with @@Hn messages, Kalman filters,
> and software sawtooth correction, you might want to consider
> looking at the per-SV "fractional GPS local time estimates" in the
> Hn message. It's the mean of these values that determines the
> virtual 1PPS (= the physical 1PPS + sawtooth correction).
>
> It's my guess that taking a number of factors into account -- such
> as lat/lon, az/el, svn, snr, and fractional stddev -- that one could
> have a history-driven "intelligent weighted mean" instead of the
> simple realtime simple mean that is currently used in the M12.
>
> The M12 uses a fixed global elevation mask to decide if a SV is
> in the solution; so it's all or nothing. With your Kalman filter, over
> a few days or weeks, your software would have a more general
> equation with which to solve for the 1PPS each second. Each
> SV clock would have its own weight, each Az/El square degree
> of each SV would have its own weight. You can also weight by
> SNR or DOP. Or even temperature. Not only should this give a
> more precise virtual 1PPS but you also get wonderful 2D or 3D
> plots showing how weight varies with each factor.
>
> I have some data on this I can share; contact me off-line.
>
> /tvb
>
> ----- Original Message ----- From: "Jerry Mulchin" <jmulchin at cox.net>
> To: <time-nuts at febo.com>
> Sent: Tuesday, May 08, 2012 3:20 PM
> Subject: [time-nuts] M12+T sawtooth correction and additional filtering
>
>
>  OK, so before I commit to my Sawtooth correction design, I had a thought
>> about adding
>> additional filtering to the sawtooth corrections using a Kalman filter.
>> The Kalman filter would be driven using the @@Hn data coming from the
>> M12+T reciever.
>>
>> Is this something worth doing or am I expecting too much from the GPS
>> system timing or
>> the M12+T receiver?
>>
>> Advice and comments welcomed.
>>
>> Thanks
>> Jerry
>>
>
>
>
> _______________________________________________
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to
> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.
>