[time-nuts] Code for phase noise and allan variance

[Magnus Danielson]

From: "Poul-Henning Kamp" <phk at phk.freebsd.dk>
Subject: Re: [time-nuts] Code for phase noise and allan variance 
Date: Tue, 28 Jun 2005 00:46:44 +0200
Message-ID: <12558.1119912404 at critter.freebsd.dk>

> In message <20050628.004037.48976141.cfmd at bredband.net>, Magnus Danielson write
> s:
> >> Does anyone have code which will take an amplitude vs. time input
> >> stream and output phase noise and/or allan variance?
> >
> >I guess you could do it that way, but really, it would be alot of data at a
> >high sampling rate to make any usefullness.
> 
> You're wrong Magnus :-)
> 
> You overlook that the A/D converter gives you much more information
> about the timing of a zero-crossing than just the sample number:
> You can interpolate the zero-crossing to subsample accuracy and
> that way get far *better* resolution than the HP5370

Actually, I did think about it, but all I saw was problems, such as dependence
of waveshape (we see that in normal counters too) which makes steeper slopes
have you loose accuracy. I agree that the A/D performs a form of analog
interpolator.

But, I also said that it will be alot of unnecessary data and unless you skip
data between the last sample to the next sample.

> which otherwise
> holds the single-shot crown.

Actually, cutting edge equipment is at picosecond or even sub-picosecond in
single-shot resolution. Also, A/D convert based interpolators is used there but
on a much more controled signal than the raw waveform.

> I have played a lot with this with a 12bit 20MHz ADC card I have, and
> the USRP does 64MHz sampling which is a lot better.
> 
> If your signal is relatively noise-free and of good amplitude I
> would not at all be surprised to see single-shot timings well into
> the pico-second regime, (compared to the 15nsec a digital use of
> 64MHz would give).

I'd like to see the error analys of such a system. The signal-dependent error
profile would be an interesting exercise.

By propper pre-processing or a close to perfect signal, the system performance
reaches the quick-and-dirty estimate.

So, maybe I am a bit picky, but it can be a bit harder, but I agree that you
can acheive sub-sample resolution, but much care must be taken.

However, I do like the idea of taking something and use it for something
completely different! ;O) Just be a bit careful in verifying the actual
performance.

> And as I suggested in an email a couple of days ago: It would be
> almost perfect for phase-noise measurements as well, thanks to
> the dual inputs.

Actually, for phase noise you really *can* make use of the sampled waveform if
you make it long enought. Toss the waveform into FFT, scrape of the overtones,
sum around the base frequency to get the single-sided spectrum and correct for
6 dB slope and you are done. This quick-and-dirty method will be pretty
accurate for lower modulations. 

Cheers,
Magnus