[sdiy] oscillator jitter / phase noise

[Magnus Danielson]

From: "Czech Martin" <Martin.Czech at Micronas.com>
Subject: RE: [sdiy] oscillator jitter / phase noise
Date: Thu, 3 Jun 2004 08:09:54 +0200
Message-ID: <D9D56E8FA1A73542BE9A5EC7E35D37FFF390A3 at EXCHANGE2.Micronas.com>

> > > For the phase noise I could imagine current source noise,
> > > integrator noise and perhaps most important comparator
> > > noise. Also noise on the power supply comming from other
> > > modules could be considered.
> > > 
> > > Anyway, has this ever been investigated?
> 
> Where? Most work I see concentrates on narrow
> band rf oscillator phase noise,
> because this is a problem for nearly all on chip plls.
> So far I have seen nothing interesting for audio frequency
> oscillators of the kind we use in synth gear.

Let's write it then!

Actually, we have been on and off on this topic. For instance did I discover
the PLL action mechanism of normal oscillators and you find that in the
archives I think.

> > It's the bandwidth resolution which is your limit. You want 
> > to zoom in like
> > hell on a single overtone to view the sideband spectra.
> 
> Well, I thought that the interesting noise must be low frequency,
> in order to create those phasing patterns. i.e. 1Hz and below.
> Noise in higher frequency (1kHz-20kHz) would make the signal sound
> noisy (hissing), but avering out the instantenious frequency.
> 
> I also thought that it might suffice to look at the fundamental
> and the side bands that go down to DC. Of course a very long DFT
> must be made to get good resolution in the low end.

But you need to rescale the result and have alot of dynamics and... no, you
don't want to directly analyse the waveform in it's sampled form. It gives you
a bad resolution.

> > Another way to do it is to give your frequency counter a 
> > heavy dose of steriods
> > and make time-interval samples back-to-back and analyse the 
> > data according to
> > the methods used in the time and frequency world. When you toss Allen
> > deviation and Modified Allen deviation analysis on a signal, 
> > you can start to
> > break up the different phase noise noise power spectrum 
> > curves and individual
> > amplitudes. Also, induces signals like overtones of line 
> > voltage also show up.
> 
> Unfortunately my frequency counter can not dump data.
> But couldn't a time measurement be made via audio sampling?

Not really usefull. The time resolution is way too coarse.

> I found this:
> http://www.analog-synth.de/avr/index.htm
> 
> Using the square output of the oscillator (given that the comparator
> is good enough, else use the original saw or tri)
> it should be possible to make such statistics
> for every cycle. This time domain analysis has also the advantage
> that no computer memory constraints apply, only hard disk space.

Right.

> With 48kHz sampling rate the time resolution apears to be rather poor:
> 20.8 us. But is this really the case? Could the real resolution
> not be a little bit better due to interpolation techniques?

No. You end up in the totally wrong place. For this kind of stuff you don't
use sampling at all. You can get a "feel" when using a digital sampling scope
with persistance. You can even do quite elaborate measures with that if used
properly, but it isn't as good a tool as when using a high resolution time
interval counter capable of back-to-back measures. Those tools are really not
replaceable by other means.

> Since most people do not have access to old gear, audio sampling
> would be also interesting. It is easier to obtain a DAT tape
> of a pure VCO recording then to get hold on the machine.

The information is essentially lost, so for this purpose it is not usefull.

> > Maybe I should put a synth under my arm as I go to work 
> > tomorrow and do a little evening-session?
> 
> This sounds like a very good idea!

Hmm... I could bring a MS-20 over without too much of a hazzel. I could also
try out a SEM, but not at the same time.

Cheers,
Magnus