I stumbled across this thread today. I feel good and bad about it. I have a schematic I've been working on for about a month that is uncomfortably similar to roender's.
I have a background as a designer of IC amplifiers, including class D power amps and class AB headphone amps. So, my goal was to attempt to translate the very successful folded cascode topology from some of my IC designs to a discrete power amp.
I feel bad because mine no longer seems very original. I feel good because the RMI-FC100 amplifier has been universally acclaimed with countless diy builds over the past 13 years. Two main differences in my design are TL431 instead of LEDs for bias setting, and SSM2212 matched NPNs instead of JFETs for the input pair.
I am still working on optimum bias currents for everything, but I'm in the ballpark of the FC100. My final stage currents are lower, but my other currents are very close. I'm not using the finals with integrated diodes. I have another idea for temperature independent biasing.
I have been thinking about something like Q10 to help with DC offset, but have not yet. A lot of the discussion about compensation, bypass of R5 with C1 has been very informative to me.
I investigated CFP versus triple EF, and decided to do triple EF for bandwidth reasons. Now I am motivated to re-open my investigation of the CFP output stage. All in all, I am very impressed with roender's design & build, and the many builds that people have posted in this thread.
The only comment I have about the design is about the relative bias of the input pair relative to the folded cascode. Roender set the input pair at 6.25 mA each, and the folded cascodes at 2 mA each. This is an effective technique for maximizing gain of the amplifier. The transconductance of the input pair is higher, and the output impedance of the Q15 and Q16 is also higher. (the gain of the amplifier is gm*ro).
However, if the amp ever clips, the folded cascode transistors will run dry, resulting in some unsavory behavior. Basically, Q14 & Q16 and everything below them and to their right will have no current. Roender says he can never clip his amplifier in his system, so he has no worries. Others, however, will clip theirs. The solution is to increase the bias of Q13 and Q11 to a little above 13.5mA each. This will lower the OL gain by a factor of about 3, and I do not know what affect it will have on the sound. Probably several transistors will need heat sinks attached due to 3x dissipation. I learned this one on an IC design I was involved with once upon a time.