I see too many current sources. That might be a problem too. A real amp using the 12BH7 where moderate cancellation and works just fine together with moderate amounts of output stage cathode feedback + GRND feedback is this:
http://www6.plala.or.jp/Michi/100TH SE amp.html
As you can see each 12BH7 triode works at about 3 mA and it's screened by the 5998 driver. However in this case the amp is true Class A2. In A1 no need to use it. The THD is NOT frequency dependent, for sure between 100Hz and 10KHz.
http://www6.plala.or.jp/Michi/100TH SE amp.html
As you can see each 12BH7 triode works at about 3 mA and it's screened by the 5998 driver. However in this case the amp is true Class A2. In A1 no need to use it. The THD is NOT frequency dependent, for sure between 100Hz and 10KHz.
A triode with a 2 LED bias gives a known and fixed bias voltage.
A triode with a CCS in the plate gives the plate current you want.
But that combination may, or may not, give you the plate voltage you want, and the two channels plate voltage may not match (usually does not need to match, but has to be a plate voltage that allows for linear output over the peak to peak swing that you need).
A self biased cathode, plus a CCS plate load tends to make the plate voltages of the two channels be closer; versus the LED bias plus CCS circuit.
But of course, the self bias resistor has to have a bypass capacitor, or the gain will be lower than the LED biased circuit has.
A triode with a CCS in the plate gives the plate current you want.
But that combination may, or may not, give you the plate voltage you want, and the two channels plate voltage may not match (usually does not need to match, but has to be a plate voltage that allows for linear output over the peak to peak swing that you need).
A self biased cathode, plus a CCS plate load tends to make the plate voltages of the two channels be closer; versus the LED bias plus CCS circuit.
But of course, the self bias resistor has to have a bypass capacitor, or the gain will be lower than the LED biased circuit has.
Nice amps!
The circuit I posted isn't intended as a recommended input stage. It's solely a demo circuit to interrogate interactions between stage distortions under open loop and feedback conditions. The only requirement is that stages generate 2H. Ideal CSS loads and constant currents in this application lock two potential variables and simplifies analysis. My preference would have been to simulate a pair of ideal three terminal device (infinite input Z, zero output Z, in+, in-, out) with programmable 2H distortion but I couldn't figure out how to accomplish this in LTSpice.
Referential ambiguity on my part. "Increasing effectiveness as frequency increases" could have been clearer. The finding was that feedback's effectiveness increases with distortion harmonic number, not input frequency. 7H is reduced much more than 2H as feedback is dialled up for example. All sims were run at 40 Hz in the interest of calculation speed.