Ralph, thanks for the info. I recall that many years ago you explained
that global NF creates temporal intermodular (TIM) distortion because of
the infinitesimally small amount of time that it takes for the signal
(after phase inversion) to loop back from the output stage to the input
stage. That small time delay causes TIM distortion, ... if I recall
your explanation correctly.
@bifwynne Not TIM, but HD. You can look at it in terms of propagation delay (easily known and seen in any class D amp) or you can look at it in terms of phase shift. Either way it means that as frequency goes up, the feedback is increasingly erroneous. Eventually you arrive at a point where feedback causes oscillation since it is no longer negative. Put another way, you have two qualities in any amplifier: its gain bandwidth product which describes how much gain is available at a given frequency to support your feedback, and the phase margin of the amp, which is the frequency above which the amp will oscillate if feedback exists above that frequency. (This is why the Futterman amp could go into oscillation if presented with certain loads, since that load would affect the feedback and allow the amp to exceed its phase margin.)
So one conclusion you can draw from this is that at low frequencies you can have a lot of feedback. Its relatively easy to have 60dB at 10Hz; the real question is how much do you have at 10KHz or 20KHz?? Because most amps simply lack the gain bandwidth product, their feedback value falls off with frequency (meaning that distortion also increases with frequency, resulting in brightness and harshness). To support 60dB at 10KHz you're going to need open loop bandwidth of nearly 100MHz (and yes, that's an 'M')... right away you can see that no amp made can support a claim like that. This is why THD is usually measured at only 100Hz, where things are 'safe'. But that practice sweeps the dirt under the carpet because its between 3KHz and 7KHz where the ear is most sensitive (Fletcher Munson) and at these frequencies the amp will simply have less feedback.
Something also going on is that due to non-linearities existing at the feedback node there will be bifurcation of the audio signal created by the feedback. This results in a noise floor composed of harmonic (which are much higher ordered) and inharmonic (intermodulation) noise rather than just hiss as seen in a zero feedback circuit. Norman Crowhurst wrote about this in the 1950s; this is pretty well-known. IME this kind of noise floor is harder for the ear to penetrate while the ear can typically hear about 10dB or so into a 'hiss' noise floor. I suspect this has a lot to do with how well the amp can portray low level detail.
At any rate, if you really want the amp to sound musical, you need two things; the first is that you'll want to see is that the distortion is the same at all frequencies in the audio band. With traditional solid state this is very hard to achieve using feedback. Its also hard to do with tubes. One way around this is to build a wide bandwidth circuit that has good linearity and no feedback. That's been our solution for the last 46 years (as it is for Ayre and a few others). Of course you pay a price for this (as you do for anything); in our case it means that matching the amp to the speaker requires more attention.
Another solution is class D since you can set up a class D amp with so much feedback that it oscillates (because its phase margin is exceeded) and then use that oscillation as the switching frequency. Now you can have enough feedback that even though the bandwidth of the amp may not be that wide, it can have consistent feedback (and thus low distortion) at *all* frequencies with enough of it that phase shift can be as well controlled as a wide bandwidth amp of no feedback at all. This much feedback can also be enough to clean up the mess that feedback normally creates when there isn't enough of it.
The other thing you need to have is a proper distortion signature (something most people call the amp's 'sonic signature'; quite literally this is the difference we hear in all amps) that prevents the amp from sounding harsh due to unmasked higher ordered harmonic content.
