[time-nuts] decimation versus decimation

[Dana Whitlow]

Please note that I have little real interest in measuring ADEV.  My
interest lies almost
solely in measuring phase variations of a received signal via some
propagation path,
not in any statistical sense but looking at actual waveforms plotted as
phase versus
time.  I am working towards measurement runs spanning weeks (maybe even
months)
over some specified bandwidth (for example, 0.1 to 10 Hz), so that long
term phase
accumulation due to static frequency error and/or slow frequency drift of
my local
reference versus the incoming signal need not be fatal.  There is concern
if the phase
accumulation gets large enough to threaten loss of numerical precision,
however.
I'm currently working on a program which includes a rather slow AFC loop so
that the
phase accumulation is somewhat bounded.

I am currently doing a 2-stage decimation.  The first stage is by a ratio
of about 4 or 5 to 1,
simply so that I can get the array size down  to the point where an FFT
does not take
forever.  The FFT does the 2nd stage of decimation according to an upper
bandwidth I
specify at run time, and a much smaller IFFT gets me back to the time
domain where I
can do my stuff.  I'm currently stuck with a cheap DSO whose minimum sample
rate is
2 kHz.  I'd prefer to run at a sample rate of ~50 to 100 Hz so I can get
some reasonable
durations.  To this end I'm looking at buying an LGR-5325 from Measurement
Computing,
which should yield all the flexibility I can ever use.  But the price- Ouch!

Dana


On Tue, Feb 25, 2020 at 11:33 AM Anders Wallin via time-nuts <
time-nuts at lists.febo.com> wrote:

> Some of the confusion might come from different measurement systems dealing
> with different data, either phase (in seconds), or frequency (fractional).
>
> Sub-sampling works for phase-data, just throw away the in-between samples
> and look at phase samples that are tau-distance apart.
>
> Decimation works for frequency-data, if you collect with 1s gate-time and
> want to look at tau=10s, then bin together 10 samples and create a new
> time-series that is 10-fold shorter, and compute adev from that.
>
> FWIW IIRC all the allantools examples/tests that (successfully) compare
> results against Stable32, using frequency input data, just use simple
> non-overlapping 'boxcar' decimation - nothing more fancy or elaborate.
>
> AW
>
>
> On Tue, Feb 25, 2020 at 7:10 PM Dana Whitlow via time-nuts <
> time-nuts at lists.febo.com> wrote:
>
> > Let's see if I have this correct, then:
> >
> > "Decimation" refers to the two-step procedure (filtering followed by
> > picking every nth sample),   and
> > "Sub-sampling" (or "downsampling") properly refers to taking every nth
> > sample and discarding the rest.
> >
> > Is this correct?
> >
> > And thanks for all the responses.
> >
> > Dana
> >
> >
> > On Tue, Feb 25, 2020 at 10:09 AM George Watson via time-nuts <
> > time-nuts at lists.febo.com> wrote:
> >
> > >
> > > _______________________________________________
> > > time-nuts mailing list -- time-nuts at lists.febo.com
> > > To unsubscribe, go to
> > > http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> > > and follow the instructions there.
> > >
> > _______________________________________________
> > time-nuts mailing list -- time-nuts at lists.febo.com
> > To unsubscribe, go to
> > http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> > and follow the instructions there.
> >
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe, go to
> http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.
>