[sdiy] Analogue Drift (was Re: HF VCOs and tracking problems)

[rburnett at richieburnett.co.uk]

Hi Andy,

> Great work Richie, thanks for sharing the results in textual form.

No worries.  I did this work a long time ago now, but its a subject 
that always seems to crop up now and again.  If Tom had asked about it 6 
months ago I could have shared the raw measurement data with you, but 
the PC's HDD died about 3 months back!

> I have noted that at lower pitches VCOs have more fuzz too them than
> at high pitches, which I am guessing comes from FM by hiss at both 
> the
> control signal input to the expo converter, and noise at the output
> which is integrated by the cap...

Sounds reasonable to me as any analogue system is subject to thermal 
noise, not least for a high-gain device like a comparator.

Unfortunately my measurement setup was not geared to measure really 
rapid fluctuations in the frequency.  It determined the frequency drift 
by continually measuring VCO waveform phase relative to a GPS locked 
"reference oscillator" over several seconds or minutes.  This makes it 
incredibly sensitive to tiny static or low-frequency tuning errors, 
(which I had initially thought would be where it was all happening!)  
The downside is that the measurment bandwidth is limited to around half 
the frequency of the VCOs note.  That's why I used high pitches like a 
top C for my tests in order to get as wide a measurement bandwidth for 
the drift or wobble as possible.  I started using higher and higher 
pitched notes once I realised that mains frequency (50Hz) and it's 
harmonics were significant players.

Although it excelled at characterising "slow drift" I might have been 
missing something interesting happening at the other end of the pitch 
range though.  And from what you've said, it looks like I have.  I think 
that an "upsampling and period measurement" technique might be better 
suited to characterising oscillator drift and instability with a PC in 
those low bass note regions.  Another option would be to use a low-cost 
PIC micro with a capture/compare module to directly measure each period 
of VCO oscillation against a known accurate clock in the 10's of MHz.  
The nice thing about low frequency notes is that you don't need 
particularly fine timing resolution to resolve small percentage changes 
in the period, and analysing low-pitched notes only produces a hundred 
or so period measurements every second so you're not dealing with masses 
of measurement data!

As engineers we have lots of tools available to use, it's just a case 
of picking the best one for the job.  Sometimes you don't know which is 
the best tool until you start the job.  Sometimes the job itself morphs 
into something else.

> In poly synths, in addition to all this FM, LFO, and LED modulation
> going on I think the main "fat" part of the sound is again not
> actually from drift but the basic fact that the scale and offset 
> trims
> of the expo converters all different for each voice.

Defo agreez,

Best regards,

-Richie,