Oh, I went super hardcore into audiophoolery before I became a serious musician...and worked for a couple decades improving Dan Meyer's Tiger .01 design...
To the point it became utterly pointless as well as exposing some puff in the audio biz.
The result was a class...hmm, what to call this - AB mosfet output design with DC to 4Mhz flat response, and .00027% IMD measured at 100:101 khz 1::1 at any power I could see any IMD at at all.(*)
Not a typo. Three zeros.
With two months of building test gear and frontends to existing stuff to measure anything wrong at all (well, it sometimes had DC offset in the 10's of millivolts and drifted a little).
I can publish the schiz if anyone cares, and I'm still using the prototype because I have it. No golden ears can hear any difference between this and any other very good amp, on any source or speakers we've come up with - we hit the limits of what humans can perceive in this setting.
And what made me quit the audio biz...
I sold them for $400 a channel (which was ~ 10x parts) and sold diddly. No one believed it could be that good, though I sold some tens to people who used them as servo amps.
So, I jacked the price up to $3k and put together a fancy anodized finned spaceship case and sold the whole run right off. "Negative elasticity" as an economist would say,.
I felt so dirty I couldn't keep doing it. I went into more honest work....
Anyone familiar with the design I started from could assimilate what I did with a few word of description of the mods I did....it's not that crazy, just matching, better current sources, burn more current in the front end quad (see national's tutorial on what makes for fast slew in closed loop amps), some changes in the bias and values in the gainy output stage to accommodate the FETs tempco and also the gate capacity. Nothing that special, actually. Since you couldn't get truly complimentary N and P fets then (maybe still) the N fet had a little extra source resistance added to match the trans-conductance. Everything in there was hand matched.
Yes, they make a pretty good LF linear (to 80 meters), and I had to add a lowpass on the input to keep them from being transmitters if people ran crappy cables or had bad ground loops.
- In its accustomed place - my main squeeze
- innards, works despite the dust
The sharp eyed may note the accidental dead bug technique here.
And yes, class D used to stink universally. Either quantization noise, or slew limiting (like happens still in many sigma delta a/d's which help promote the myth that super fast sampling and more bits/sample are the answer when it's just that a 16bit/44.1k a/d is really only 12 bits good and slew rate limits at anything over 5khz full scale amplitude due to *design*).
But this one doesn't suck, and only cost ~ $12! Hence the amazement. I'd have to hook it up to my $20k monitors to be sure, but with some decent bookshelf speakers and a good d/a, nothing is obviously wrong, which wouldn't have been the case a couple years ago (and still in most class D stuff).
* In this design, neither side of the push pull ever really turns off, but does get down to a quiescent value fairly early in the off part of a cycle. But it stays on some, to facilitate instant recovery and keep the charge carriers alive...But it's not class A either - the "off" side is nearly off...
Edit: Based on this:
http://baselaudiolab.com/TIGER-01-207B.pdf Just burn a hell of a lot more current in the emitter followers in the output stages, change a few loop time constants, put fets as output devices, balance it all up.
Posting as just me, not as the forum owner. Everything I say is "in my opinion" and YMMV -- which should go for everyone without saying.