Ways to design for low IM distortion

[DF96]

Having thought some more about it, there is a way to make IMD worse than THD: poor decoupling either of cathodes or screen grids. All devices create second-order distortion. If this is allowed to re-enter the same device from an electrode which can act as both an input and output (usually cathode or g2, but it could be the anode) then it reacts with the signal to create third-order. This is an example of the dangers of badly applied (or accidental) local feedback. Like global feedback, local feedback needs to be none or enough.

If there is some decoupling, then only low frequencies get fed back. So 400Hz and 420Hz generate some 20Hz and 820Hz due to second-order. The 820 gets decoupled to ground. The 20 goes back in, because the decoupler is not big enough. This mixes with the orginal signal to give some 380 and 440 too. Voila! 3rd order IMD made from 2nd order IMD, even though there may be no 3rd order harmonics.

So to answer the original question: you need to properly decouple things.

 

[Miles Prower]

I'm not completely clear on this, but I suspect that the source of distortion (like nonlineaity) will produce more THD and IM, but feedback loops can only reduce THD. But the fed back IM just introduces additional crud that isn't subtracted out or reduced. As it looks like part of the input signal when the feedback signal is merged with the input. That it produces yet more IM products, but at lower levels.

Not quite right. NFB increases linearity, and reduces both harmonic distortion and IMD. You can see this demonstrated on p. 185 of Oscilloscope Techniques (Alfred Haas; Gernsback Library No. 72) (It's available at Pete Millet's site -- highly recommended). It's true that NFB can't cure all harmonic distortion, and Crowhurst proved that over fifty years ago. It will greatly reduce near HD, but can't completely get rid of low level HD near the noise floor. Perhaps that's why too much NFB makes for a "solid statey" sound? Digitization likewise fills the noise floor with irreducible quantization noise that also has a similar harmonic profile. May be that's why some prefer vinyl to CDs?

If the above is even vaguely correct (!), it would seem that the best approach to amp design is to create the best amp you can with no feedback loops, and only after that, use a little feedback to make it more perfect. IOW, don't expect to clean up a crappy design with feedback.

That's how I design: make for the best open loop performance, then add just enough local (if needed) and gNFB to make it sound even better. This is something not taught in EE school, it would seem the graduates don't really know what they're doing, and solid state offers plenty of opportunity to just pour in enough gNFB to sweep those mistakes under the carpet. It also sweeps away the vitality of the music.

That is not the way to do it at all, solid state or hollow state, but it's all too common, and the result is obvious: sound-alike, Big Box amps that are mediocre at best. It's true now (and solid state does not have to sound as hideous as it does) and it was true sixty years ago.