The Local Hero MK-III amplifier modification by Norman Koren is a well thought out design that seeks to eliminate some of the pitfalls encountered in most amplifiers - specifically those related to global negative feedback.
There's nothing wrong with negative feedback, especially when applied in moderation to an already well designed amplifier. There is however problems that creep up like inherent instability when improperly loaded (phase shifts), reactive loads and they do a butcher job on slew rate.
Mr. Koren did a wonderful treatise on converting one of the more powerful of Dynaco amplifiers, the Mark III to using local feedback only, here: http://www.normankoren.com/Audio/FeedbackFidelity2.html
Being an authorised manufactuerer and reseller of PCB's for his projects, I had a small run of PCB's made in order to more readily share this work of his.
However, I encountered a small problem when experimenting with it in my Dynaco test bed - it was not stable.
My test amplifier had a "heartbeat" every 30 seconds or so. Having used the amplifier with full stability with many other Dynaco and aftermarket designs, I ruled out the test bed as the cause.
Upon investigation, it was determined that in some cases that the cathode feedback going into the phase inverter was the cause.
This section has a rolloff increasing with frequency, due to C34, a 680pF capacitor joining the phase inverter cathodes. But the iron itself doesn't have infinate LF response. The result is a gain stage running wide open and uncontrolled below say, 4Hz.
The "ground the 4 ohm tap and use common and 16R as cathode feedback" is not a new trick. But Mr. Koren's use of feeding the phase inverter with the cathode feedback is quite unique.
To verify this as the cause, I simply pulled R3F and R4F. The "heartbeat" went away.
Some Googling led me to realize the situation was fairly unique, as only one other person complained of this too.
So I set out to stabilize the amplifier while still keeping most of Mr. Koren's circuit (and my boards) intact.
First, I had to lose the CFB to the phase inverter while still keeping feedback there to keep the 12AU7's in check (not the most linear tube).
Experienced tech's will recognize the method of cathode feedback used on a multitap secondary as one used by the audio manufacturer, Audio Research. Audio Research also used some anode-following feedback applied to the phase inverter cathode as seen here:
However, there is *very little* feedback from the anodes of the 6550's happening, as they are fed through two 1Meg resistors shunted to ground via low value cathode resistors. That's because they use a combination of three feedback systems: 1) traditional gNFB (green line), 2) Anode follower (blue lines) and cathode feedback (red lines).
We still want to keep gNFB out of the works, so originally I tried straight anode following and got this:
(12AT7 driver dealt with in a moment)
First thing I did was rip out the low value capacitors. They slowed the amplifier to the point a square wave was looking more like a pyramid.
I did my tests in triode, UL and pentode positions with my test bed (ST-70 with power supply on steroids and MK-III output iron). The results were as follows:
Triode power output: 21W
Full power F3: 16Hz - 26KHz
Local NFB: 5.8dB
UL power output: 45W
Full power F3: 24Hz - 36KHz
Local NFB: 8.5dB
Pentode power output: 50W
Full power F3: 25Hz - 35KHz
Local NFB: 10.1dB
On the squarewave test in pentode, there was a bit of a leading edge spike on the top and bottom waveforms due to the transformer flyback.
Sonics were bright, mids extremely detailed and bass neutral, but the only good full power 20Hz sinewave was on the triode position. Need better damping factor. Let's add cathode feedback again.
The cathode feedback only adds another 1.5dB to the NFB figures, but what happen to the F3 and square responses here?
Triode power output: 21W
Full power F3: 10Hz - 35KHz
Local NFB: 7.3dB
UL power output: 40W
Full power F3: 10Hz - 43KHz
Local NFB: 10dB
Pentode power output: 40W
Full power F3: 11Hz - 46KHz
Local NFB: 11.6dB
And there was no spikes on any leading edges in pentode on the square test.
Sonics test is forthcoming and I'll have a report on this, ASAP :-)
The two resistors in series (in my case, 56K @ 2W) are required because you have high AC peaks riding on an already high DC voltage. Unless you use a "barber pole" resistor, you will require the two two watters in series so they don't break down.
Now about the 12AT7 where the 12AX7 should be.....
In all tests, I found the gain way too high. Even with the stock board, 200mV RMS brought me to full tilt.
I modified the values for a 12AT7 and am running it at 3mA first stage. Now it requires a more sane 500mV RMS in triode, 550mV RMS in pentode and 600mV RMS in UL for full output.
Oh and have I mentioned... I do NOT like the sonics of 12AX7's, outside of heavy NFB RIAA stages? (the Koren PAS linestage was the FIRST 12AX7 linestage I liked. It has lotsa NFB)
Cheers!