[sdiy] Analysis of frequency variation in analogue synths

[Magnus Danielson]

From: Nicholas Gregorich <nicksdsu at mac.com>
Subject: Re: [sdiy] Analysis of frequency variation in analogue synths
Date: Wed, 2 May 2007 22:31:44 -0700
Message-ID: <CF159199-CD5D-4128-BCBF-43D4FEE7469C at mac.com>

> 
> On Apr 30, 2007, at 11:31 PM, Magnus Danielson wrote:
> 
> >
> > The topic of weither the jitter & wander aspects is part of the  
> > "analog" or
> > "fattness" has been debated here and elsewhere to death before. I  
> > still
> > consider that issue unsettled. There are parties that argue for the  
> > case that
> > it is or it isn't jitter & wander related. Personally I advocate  
> > the point that
> > it *might* be phase modulations or for that matter amlitude  
> > modulations (I
> > wonder if the ear makes the distinction at lower modulations  
> > rates). I have
> > not seen any conclusive evidence that it is or it isn't. This have  
> > been beaten
> > to death here before, I'm just trying to refer to the issues.
> >
> > There is much more to be said about this, but I gotta run.
> >
> > Cheers,
> > Magnus
> 
> Lots of interesting information there, thanks.

You are most welcome. I wanted you to be aware of the debate and what kind of
dispersed opinions there are (which is good in a sense).

> I thought I read all  
> of the Moogy jitter posts but I saw some unread I will go check out.

OK.
 
> I realize its been debated a lot, and I don't think Tom or I was  
> trying to heat the debate. What we both seem to be after is finding  
> the conclusive evidence you mention.

Conclusive evidence is hard. Not only do you need to measure various sources
for a number of suspected effects (AM, PM, Moon phase or whatever), but you
need to measure them well enought from a large number of sources and then have
them blind-tested on a large enought scale of people that know whatever
psycho-acoustical sensation(s) you are after and is able to rate that sensation
for each of the test objects. This part of the test also needs about the same
signal or else the test subjects have too easy to spot a Moog from an ARP or
whatever. Thus the blindness needs to be fairly complete. Only after analyzing
this and seeing correlations between the test-persons judgement and some of
the measures you made on the source, you are approaching conclusive evidence.
You can naturally convince yourself by reducing the test-person to yourself,
but you must recall that you don't have conclusive evidence that will count
with everyone. The experiment I presented above would also have to be repeated
by independent groups and come to similar enought conclusions for it to attempt
to be conclusive. It also needs to be published in a good enought journal with
sufficient review to start to count as evidence. This is how the scientific
community works. Everything else we do is interesting exercises. Those can be
made more or less skillfull.

> My concern is measuring jitter of a signal and not knowing how how  
> much is being injected/caused by the measurement device. It seems  
> like the methods mentioned today by Ian and Paul are not robust  
> enough to conclude what is going on with PM/FM on a VCO.

You certainly need to investigate the measurement floor of your measurement
device. There can also be a number of signal-dependent properties which shifts
the measurement floor in significant ways. There can also be ways to circumvent
such shifts. It all depends on how you measure and what you measure. It is a
science in its own right. I can't expect my fancy counters to analyze the
"moogines" with full resolution directly, since their high-bandwidth input
expect high slewrate signals while there is a certain amount of noise due to
the high bandwidth. The low slew-rate of my MiniMoog (especially as it passes
through the full chain of filter, vca etc) would convert all that lovely noise
into time-shift. The obvious way to circumvent that is to make a through-zero
detector which significantly increases the slewrate. This can only be done with
a set of linear gain-stages (where the higher-bandwidth onces should have
output swing limitations such that they don't go into deep saturation). Such a
circuit is also used for high-resolution frequency comparision work when you
analyze the output of a mixer from two ultra-stable sources (think hydrogen
masers, ceasium fontains etc, but certainly also for normal caesium beams).

So, you can have somewhat of a state-of-the-art instrument with resolution you
could die for (I can reach downto 200 fs of singel-shot resolution) but that is
all wasted without propper signal treatement. You can expect the same to be
true for other types of measures.

Personally I have the fancy stuff for time/phase/frequency analysis, I just
need to have the time to build the necessary slewrate amplifier.

It is also highly recommended to make Allan deviation plots and Modified Allan
deviation. Also for these types of measures you need to be careful, and I have
several times proved how the improper handing of input spoils the measurement.
The reason is that these post-processings of the time-data is sensitive to the
drift-rate (derivate of the frequency). Even for very stable sources as
Rubidium cells, Ceasium beams and GPS receivers, inpropper handing of data will
spoil the result.

Cheers,
Magnus