Balancing the Futterman
It is well known that the degenerative feedback caused by the variation in the cathode voltage of the upper triode in the output stage, reduces the grid-to-cathode signal for the upper triode. If left un-addressed, this degenerative feedback causes the output stage to be unbalanced for anything but a zero load.
Futterman's well known solution was to bring the cathode resistor of the phase splitter to the output instead of to ground, shown in the diagram.
Futterman's Balancing Act
To get proper balance it is also necessary to have floating bias supply for the grid bias on the upper triode along with a fixed bias supply for the lower triode. One of the bias supplies is variable so that the DC offset can be set to zero. A stereo amp will require four separate bias supplies using this scheme, two of them floating.
In a power amplifier, where the idle current of the output stage may be fairly large, the cathode current of the phase splitter is negligible. In a headphone amplifier where the idle current might be only 10mA, a 2mA return current from the phase splitter will significantly unbalance the operating points of the output stage triodes.
The design goal having only one B+ supply and no bias supplies means that we've got to eliminate the two bias supplies and one of the bipolar supplies in the output (either + or -). A secondary design goal became to create a topology where the cathode current of the phase splitter was not returned through the lower triode of the output stage.
The first step in the process was to accept the fact that it is too difficult and too dangerous direct couple a tube output stage to headphones that you care about. Any number of failure modes can produce destructive voltages/currents in the headphones. I came to this conclusion after trying a number of servoing schemes, none of which could protect against all possible destructive failures. Finally, I made the design decision to use an output coupling capacitor. Once this decision was made, it was no longer necessary to have a bipolar supply for the output stage because we're no longer trying to set the output DC offset to zero.
With a coupling capacitor, the bottom end of the output stage is grounded and the top end is at the full B+. This eliminates one supply immediately (the negative output stage supply).
We still need the various bias supplies for the output stage grids.
Without going into detail, I eventually developed a biasing scheme that needed just one B+ and one small negative supply. Here is the diagram.
The bias for the lower output triode comes from the negative supply. The bias for the upper triode comes from a bootstrapped voltage divider from the B+. You can also see that the amp has a single gain stage and a split load phase splitter. This design also has the variable negative feedback scheme that has been successfully used in the Cavalli-Jones amplifier. This scheme allows the listener to adjust the feedback to suit the headphone impedance.
In amps where the output is nominally at zero volts, the Futterman balance technique works. With a unipolar supply, however, the output is not at zero volts, but at ½ the B+. The phase splitter cathode cannot be returned to the output. To remedy this, because the phase splitter is now operating at full B+, I simply returned the phase splitter's cathode to ground through an additional resistor. The Futterman feedback to balance the output is applied to the junction of the phase splitter's cathode resistor and its ground return resistor through a capacitor. This technique also eliminates the problem of unbalanced output currents in the output stage and balances the signals in the output stage for all loads.
I continued to think about how to build this amp because I still needed a negative bias supply. I posted the 6922 design as an idea in one of the Headwize forums. PRR, a member of the forum, proposed a simplifying change to the amp. Since the bias voltage on the 6922s was only 2.3 volts, PRR suggested using diodes for cathode bias instead of the more complicated biasing scheme that I developed. I realized immediately that this was the last piece of the puzzle.
The schematic changes are this:
This simplification eliminates quite a few resistors and capacitors, but more importantly it eliminates the need for any negative supplies at all. The diodes simply act as bypassed cathode resistors, as their dynamic resistance is only about 4 ohms.
At this point I had achieved the design goal of a full Futterman headphone amplifier with only two supplies. But, there was one more step. I realized that the JJ ECC99 could be directly substituted for the 6922 in the output stage without any changes, which would double the idle current from 10mA to 20mA, thereby quadrupling the power output. At this point I knew that I had to build this amplifier. It was only a matter of designing the power supply.
The Power Supply
I decided to use a tube rectifier so that I didn't have to build a solid state slow start circuit. My first choice was to use a 6X4. Mark Lovell, with whom I have collaborated on the Broskie-Cavalli-Jones and the Cavalli-Lovell amplifiers, persuaded me to use the 6V4 (EZ80) because it has smaller switching transients. Mark was generous enough to buy some EZ80s and to ship them from the UK.
I also decided to use the simple series-pass mosfet regulator that has worked so well in our other designs. And Mark, after seeing me refer to this amplifier as a real Gem, suggested that we call it the Bijou (the French work for Gem). And so we did.