[sdiy] Opinions on Switchmode PSUs for analog synths?

[rburnett at richieburnett.co.uk]

A few points about SMPSU's in audio applications:

1. No decent SMPSU should generate significant audible output noise.  At 
least not when loaded with a fixed load in the range 10% to 100%.  If 
it's loaded less than 10% of it's maximum capacity it might start to 
cycle skip, (and this is almost guaranteed at some point if you reduce 
the load down to zero,) and this can obviously cause some ripple in the 
audio band.  (This is the reason why a lot of things like phone chargers 
will make an annoying whining noise if you disconnect the phone from the 
end of the cable, but are otherwise silent when charging.)  It's also 
possible you might see some 100Hz or 120Hz ripple if you overload the 
supply's output.  But generally SMPSUs do a pretty good job of filtering 
out mains related ripple due to some big 400VDC electrolytics at the 
input end, and a fast feedback loop to regulate the voltage at the 
output end.

2. SMPSUs generate RF noise!!!  It probably comes as no surprise that 
most of the noise you get out of typical SMPSUs is related to the 
switching action. For larger "forward" type power supplies (>70W) the 
output current is smoothed (usually by a big toroidal inductor) so you 
don't tend to get that much ripple at the switching frequency appearing 
at the output.  What you *do* tend to get is a relatively large amount 
of ripple at higher harmonics of the switching frequency, that looks 
like a train of tiny little impulses when viewed on a scope.  It is this 
HF switching hash (often with ringing) that is usually more noticeable 
on the output of bigger SMPSUs.

3. Unfortunately, smaller (<70W) SMPSUs usually use the flyback topology 
because it is a bit cheaper and lighter.  The flyback topology is evil 
because it has discontinuous input and output current (therefore 
generates a lot of noise) and it has no "buck choke" at the output to 
smooth the current!  That's why it's cheaper.  The output capacitor is 
left with the entire burden of holding up the output voltage during the 
intervals when there is no power coming from the transformer.  So in 
this arrangement you typically see a lot more ripple at the switching 
frequency present at the output side and the mains input side, and it is 
typically harder to filter out because it comes right down to the 
switching frequency.  Most modern wall-warts are flyback power supplies 
and are notoriously noisy.  Most people don't complain because they only 
use them to charge their phones, shavers, toothbrushes, laptops, etc.

4. As for reliability, MTBF etc, it depends on the choice of components, 
layout and *cooling*.  Low quality, standard temperature (85'C) 
electrolytics placed near to the transformer or heatsinks in a sealed 
enclosure with no forced-air cooling are not going to last very long at 
all in a SMPSU.  Conversely spreading the load current between a bunch 
of paralleled 105'C low-EST electrolytics from a reputable source like 
Rubicon placed away from other heat sources with forced-air cooling can 
surely last 40+ years.  Lifetime comes down to picking decent components 
and thermal management.

I'm not really much of an analogue synth design person, although I've 
tinkered and have a basic idea of how things like VCOs, filters, etc. 
work.  But I am an experienced EMC compliance engineer, so have a decent 
ability to foresee where problems might occur...  Although most SMPSU 
output noise is well above the audio band where it can't be heard 
directly, it has the potential to makes it's presence audible by some 
strange effects.  For instance, consider a standard saw-core analogue 
VCO, based around an expo-converter feeding an integrator that is reset 
periodically by a comparator.  This circuit normally performs very well 
in an RF-free environment.  Now think what might happen if you introduce 
noise in the form of a 200kHz pulse train of tiny RF pulses.  You won't 
hear the 200kHz directly.  That much we know.  But those little pulses 
could easily find their way onto the output of the integrator due to 
poor PSRR of the integrator op-amp, or they could find their way onto 
the reference voltage for the comparator.  If you get tiny pulses 
present at the inputs of the comparator, then the comparator is likely 
to trip early and terminate the integrator's charging cycle early.  The 
effect of this is a shift upwards in pitch!!!  Likewise if RF hash finds 
it's way into the highly non-linear expo-converter circuit, it's likely 
to also cause a shift in pitch.  So in summary, I wouldn't be at all 
suprised if RF hash from a poorly filtered SMPSU made an analogue VCO 
play out of tune, even though the average supply voltage is fine, and 
there's no audible ripple on the supply itself.  This is how I think 
problems would manifest themselves.  Has anyone experienced anything 
like this in analogue synths, modulars...!?!?

-Richie,