In the 1950's, when Stereo was getting started, a very novel form of amplifier was introduced. The idea was that since a push pull or differential amplifier takes two input signals, one out of phase with the other, and stereo consists of two similar channels, if you invert one of the channels, feed each to the differential inputs, then matrix the output, you obtain a stereo amplifier "for the price" of one mono push pull amp. I ran across this in an article sent to me, IIRC, by Russ Sadd that was in Practical Wireless magazine in 1959, by A.G. Bourne. After I started investigating this, Gary Longrie sniffed out the entire article and sent it to me.
The general conclusion at the time was: low quality; suitable for cheap phonographs. However, as we will see, this conclusion was very premature.
One of the advancements over time, not generally available in the '50s was reasonable accurate analog modeling (Spice). A little Spice modeling immediately showed the limitations of that topology. The crosstalk between channels was terrible.
As you notice, this is not very good performance, and one of the reasons the circuit was abandoned. However, by adding just a few corrections to the circuit, this terrible performance can be turned into something quite good. This would not have been evident, nor was it to me, until I did some circuit simulations. Here's the same circuit with corrections:
This circuit usually gives people a headache trying to understand it, but it is really simple. Common mode signals (Left - Right) are "cancelled" in the output stage, and thus appear across the choke and are coupled into the second output transformer. The main "push pull" transformer carries the Left + Right signal. Those two signals are then recombined in the secondaries into LEFT and RIGHT. The original right channel inversion is corrected by connecting the "hot" of the right speaker to ground; that's why the speaker connections are drawn in the schematic the way they are.
The original circuit did not have the parafeed arrangement, nor C2, nor R14/19. It is these elements that turn this amplifier into a very good performer. The Parafeed arrangement corrects for transformer phase anomalies at low frequencies, improving lower frequency crosstalk, and C2 corrects for transformer leakage inductance issues at high frequencies. I used some old Sansui output transformers, but any good output transformer would work. You may need to adjust the values of these components for your particular transformers.
R14/19 form a blend/anti/blend circuit. Since the channels are matrixed, this resistor adjusts crosstalk. Near the middle, cancellation occurs (which is surprisingly independent of speaker impedance). Interestingly, the anti-blend creates an out of phase "other channel" component that can be used to improve the imaging on some highly processed CDs.
Why the odd output tubes? Found them at a local surplus place (NOS/NIB for cheap!). Since both are double tetrode, you get a "two tube" stereo amp. Looked so cool, I just had to use 'em.
The amplifier, as built produces about 5 watts per channel with either channel driven, and about 8 watts per channel with both channels driven. (Yep, not a typo).
Frequency response is down 3dB at about 8Hz and 50kHz, and output impedance is slightly below 4 ohms for the 8 ohm speaker. (Good damping).
The amplifier behaves like a single ended amp. That is, the even order distortion products are not suppressed, but in fact, the odd order components are lower than expected, particularly with both driven. Here's the distortion plot of the amplifier.
Note that the amplifier "gets more powerful" when both channels are driven, which is quite unusual.
How does it sound?
Absolutely marvelous. Very dynamic, clean, powerful. Clip point is almost completely inaudible. As built, this thing sounds like a pair of 845s.