Brief post of a very long experiment. Unfortunately I’d have to dig out a long list of notes, tests and simulations to get the full thing out there. Unfortunately due to constraints, this will be brief but I hope it’d be entertaining.
The design is straightforward. JC Morrison has also worked with similar screen feedback arrangements but not with a hybrid mu-follower stage. The feedback loop is taken from the mu-output in my case. I’m using the EF37a with slight positive G3 bias (+12V) to minimise distortion at largest voltage swings (200Vpp). The gNFB is anode of output stage into cathode of the driver via R4 and R8. There are 2 tricky LF poles playing with the screen bias (C5+R9+divider pot) and C9+R14 in addition of the HF pass made up of C6 and R4.
I had some LF oscillation in my breadboard which I couldn’t fix eventually despite the help from JC and a few iterations.
The 6P36S is stressed to its limits. This valve is famous for being able to deliver much more than the specs. Yet, it was (not sure if still is) cheap as chips and I have plenty of them. I needed +5W, and here is the test with 7W:
Not bad at all 0.3% and decent harmonic profile.
I’d probably go with a different feedback arrangement if I were to do this again in future. Cathode drive is more interesting in my view and there are PMOS around to do this:
You will need a proper sized heatsink for the PMOS as it will dissipate 11W and R3 will help out a bit with 2.5W at least. 160Vpp are needed to drive the cathode so many of the good drivers are out there.