[sdiy] VCO Jitter, Slop ... (was: Re: Smith's Evolver Desktop's spikey VCO waveforms)

[rburnett at richieburnett.co.uk]

Hi all,

The VCO jitter/fatness/slop/wander thing is something that I did some 
investigation into a few years back.  I had a GPS-locked atomic-clock 
source and very high precision data aquisition system on my bench for 
some other work unrelated to audio.  This allowed me to do some 
measurements on the instantaneous phase of synth VCO waveforms relative 
to the GPS clock's rock-solid 10MHz reference.  I've already posted 
about the test setup and results in detail here before, however a 
summary is as follows.  I tested a Roland TB-303, Roland SH-09, Roland 
JX-3P, Roland Alpha Juno 2, and Casio VZ-1 for oscillator stability 
relative to the GPS frequency reference.  The main contributors to pitch 
error over time were as follows:

1. Residual LFO feedthrough modulating the pitch even though the LFO 
mod is set at zero.  For the SH-09 this had the waveform of a high-pass 
filtered squarewave, and was identified to be due to 
undershoot/overshoot and subsequent recovery of the PSU rails when the 
LFO LED lights suddenly and extinguishes suddenly!  (If you set the LFO 
fast enough and do a long FFT on the VCO waveform you can clearly see a 
bunch of sidebands characteristic of frequency-modulation.)

2. Very noticeable 50Hz modulation of the VCO pitch on the SH-09.  PSU 
supply rail ripple for a full-wave rectified DC supply would be at 
100Hz.  The 50Hz FM was found to be due to capacitive coupling from the 
mains power wiring around the transformer to the sensitive 
high-impedance nodes of the VCO and it's expo current source.  Grounded 
Aluminium shielding of the mains wiring dramatically reduced the 
modulation sidebands in the VCO output due to 50Hz mains feedthrough.  
You can get very audible 50Hz FM of a VCO just by touching the 
integrator's virtual earth point with your finger, it doesn't take much 
current at all here to cause a disturbance!

3. Modulation of the VCO pitch by the LED multiplex waveform in the 
TB-303.  I suspect this is due to supply rail dip and recovery as 
various columns/rows of the LED array are abruptly switched in and out.  
I didn't try to "fix" this, or alter the multiplex process in any way 
since it's all tied up with the CPU and I didn't want to risk damage to 
those delicate phenolic PCBs!

4. Modulation of the VCO pitch by the tempo clock in the TB-303.  This 
was again very easy to see as modulation sidebands in a long-time FFT of 
the TB-303's raw oscillator.  This time the spacing between the 
sidebands could be changed by altering the tempo setting!  Presumably 
due to power supply disturbances again.  The TB-303's boost converter is 
quite primitive, so the load regulation is likely quite poor.  (I ran 
the 303 off batteries during the test, so don't know what the line 
regulation and susceptibility to external supply noise was like!)

5. The JX-3P showed noticeable pulling of the oscillator frequencies 
when they were set close.  Presumably due to slight electrical coupling 
between it's two LC free-running RF oscillators that are divided down to 
generate the integrator reset pulses for the two VCOs.  Certain settings 
would lock the oscillators in phase despite the little metal plate 
soldered between the two oscillators intended to mitigate this effect!  
Entrainment is quite common in analogue oscillators and is harder to 
stop at radio frequencies without good screening and decoupling.

6. The alpha Juno and VZ-1 showed very little long term drift at all.  
The VZ-1 showed much more short-term phase-noise in it's oscillator 
outputs particularly at high pitches.  This is presumably due to the low 
sampling rate of the digital VZ-1 when compared to the high clock rate 
used to drive the Alpha Juno's NCO based waveform generator chip.  (The 
aliasing in the VZ-1's oscillators is plain to hear at the top end of 
the keyboard, whilst you have to pitchbend up the highest note on the 
keyboard of the alpha juno and then look carefully at the spectrum to 
see aliased frequency components with it's 12MHz sampling rate!)

7. Loooooong-term thermal drift.  Both the TB-303 and the SH-09 showed 
long term thermal drift.  I tried to perform the tests in a temperature 
and humidity controlled environment where I worked, but it was 
innevitible that people opened doors to enter and leave the room 
occasionally.  The long-term thermal drift was best described as a 
gradual descending or ascending pitch over several minutes, with 
occasional more rapid slews where someone opened a door or wafted air 
past the synth.  (All of this was after 30mins time to aclimatise to 
ambient room temperature and reach steady-state operating temperature.)  
The Alpha Juno showed a very slight pitch change after initial switch on 
which I attributed to the crystal temperature changing.  It was only a 
few ppm but I can't remember the direction now.

I didn't see any evidence of 1/f noise or any exotic random walking of 
frequency etc.  However, maybe if I could have removed the LFO, LED 
multiplex, Tempo clock, and 50Hz supply contributions from the phase 
noise there would have been some "filtered noise" contribution lurking 
in there just above the wandering DC level caused by ambient temperature 
changes!  Different measurement setups are better at revealing 
deviations over different time-scales so in a way you almost need some 
prior knowledge of where to go looking for drift.

Also, I appologise for this analysis still sounding slightly hand wavy! 
It's down to two things.  Firstly it's a long email so I haven't gone 
into the technical details of the measurement, and secondly the PC that 
had the results on it at work has long since gone the way of the hard 
disk failure so I don't have a full set of measurement results any 
longer!

Hopefully some nuggets of useful info in there though as i'm not aware 
of any similar analysis, and appologies for the long post if this is not 
your particular area of interest,

The only thing I would add to Tom Wilthsire's discussion below is that 
very small frequency shifts in a single oscillator *ARE* audible in some 
circumstances.  Natural reverberations and comb filtering in the 
listening enviroment can make our ears much more sensitive to small 
frequency variations than we might expect.  When judging these things 
you have to keep your head rock steady whilst listening otherwise the 
change in sound simply due to multiple varying paths reaching your ear 
is equally as dramatic as any VCO instability.

-Richie,


-------- Original Message --------
Subject: Re: [sdiy] VCO Jitter, Slop ... (was: Re: Smith's Evolver 
Desktop's spikey VCO waveforms)
Date: 2013-06-10 22:07
 From: Tom Wiltshire <tom at electricdruid.net>
To: BRUCE DUNCAN <modcan at sympatico.ca>
Copy: "synth-diy at dropmix.xs4all.nl" <synth-diy at dropmix.xs4all.nl>

I totally agree with Bruce. "Fatness" is a concept related to multiple 
oscillators. I don't think the human ear  (most people's anyway) have 
anything like the accuracy required to hear a little drift in a single 
oscillator.

I also totally agree with Brian. We *really* need to test some of these 
hypotheses to get this nailed down once and for all. There is a 
widespread perception that analog oscillators (and some more than 
others) have some level of drift or noise inherent in their design that 
improves their sound, particularly with other oscillators. Digital 
technology has reached a point where we could/can model this property 
*if we can identify it*. Plenty of effort has already been made in this 
direction.

My own experiments in this area have shown me that a little linear 
phase noise in a waveform isn't a bad thing, and that a little linear 
frequency wobble (if that's a different thing, which it only just is) 
helps too. Whether these effects are what's going on in "proper" 
analogue oscillators I've never been able to show. I only have CEM3340 
oscillators around, and they don't show these effects, as far as I can 
determine.

One important thing to remember is that even quite minor shifts in 
phase have a significant audible effect on the higher harmonics if you 
mix a pair of oscillators. Even if the main pitches remain very close, 
the highest harmonics will drift in and out of phase. We can easily hear 
this, especially on the incredibly harmonically rich synth waveforms 
we're typically talking about. Thus you can put in a little bit of 
"motion" whilst barely affecting the pitch.

T.