I think you are mixing the fact that I want a linear power supply with what happens inside the audiophoolery circles. These are two separate goals. One involves (me) designing a power supply with acceptable noise levels, even if it is not 100% efficient, the other revolves around expensive capacitors, unidirectional cables and snake oil.
FYI, there is a more technical, design-centered facet of audiophoolery, that likes to play with novel circuits -- diyAudio and Audiokarma are probably good examples. Audiophoolery is a huge sliding scale, from good engineers with a bit of a kink (an analogous and somewhat topical example might be Dave Berning's DC-transformer (chopper based) designs, worth a look), to the casual rich, looking to improve the chakra of their music with crystal pyramids supporting hand-woven silver cables.
Don't take it as an insult -- it's a sign of creativity. Just don't get disgusting with it, that's all.
Making it the primary regulation element, and then adding a pretty heavy filter, may not solve the noise completely.
Let's talk about that --
How much attenuation is "complete"? 40dB (100x)? 80dB? 200dB? Below the noise floor (whatever that might happen to be)?
Because it of course can never be complete-as-in-actually-zero, not in the real world, not at finite temperature and over finite time!
How much attenuation do you expect to obtain from the linear regulator stage?
It's not as easy a problem as you may envision, and you're open to common mode noise regardless of what circuitry is on
either side of the converter. The converter needs to be laid out properly to avoid this, regardless, and some additional filtering will inevitably be required (for some values of "complete"
).
So anywhere you're using a nasty rattling converter, you have to be careful to filter it, on all sides. The only true solution is going full linear, which as you already understand is quite inefficient, and is still not the greatest because you still have to deal with mains noise, diode recovery noise, and supply ripple. The noise floor is set by the output transistor, and/or overall output impedance*, which defines how much filtering you need, to reach that floor from some given noise source, whatever that might be.
*Note that capacitors have noise, just as everything else does: take the impedance at a given frequency, and convert it into noise with the Johnson noise relation:
https://en.wikipedia.org/wiki/Johnson%E2%80%93Nyquist_noise#Thermal_noise_on_capacitorsSince the impedance is usually low (~ohms), this is a small contribution (below 1nV/rtHz), but still nonzero.
The easiest low-noise source is a battery: the effective capacitance is extremely large, and as long as the current draw is small, the chemical overpotential is small as well. Noise is usually due to thermal and ionic motion, having a 1/f spectrum; keeping wind currents away, and current draw low, helps greatly.
Anything else is more complicated, whether a linear, switching or hybrid supply.
I recommend a well-filtered switching supply, because you may need filtering regardless of what topology you choose, and filters are easier to make than low-noise linear regulators: just throw more inductors and capacitors at it!
Tim