Ilya said:
Ok. Turned out that the cap did stop the oscillation, but the output stage still doesn't want to drive anything lower than ~2.5K. It behaves like this: the heavier the load, the more the output level increases, and THD increases as well.
That is definitely a sign that the negative impedance circuitry is not properly balanced. The ideal situation would be that the negative impedance compensates exactly the primary resistance. In practice, in order to avoid issues, the compensation is set a little lower, to make sure that there is no risk of unstability.
Wet finger says R33 should be about 1/10th of the DCR of the primary. Is that the case?
The only way to stop this is to ground the transformer primary. But in this case the overall output level is 3 dB lower which is undesirable.
This is not normal. Gain should change by a fraction of dB only. Again that's because too much positive feedback. I think it's because the xfmr you use has a much lower DCR than the original.
Is there anything I can look at to make this work with 600R loads?
What do you mean " the output stage still doesn't want to drive anything lower than ~2.5K"? What happens? Does it clip or cease completely to function?
Simulation indicates it should be capable of delivering about 3.5Vrms into 2 k (+13dBu) and about 1Vrms (+2 dBu) into 600 ohms.
It seems the only solution is to decrease the value of the resistor in the emitter of teh bottom output transistor R37.
27 ohms there suggests 3Vrms into 600 ohms (+12 dBu), and 15 ohms suggests 5Vrms.
"But in this case the overall output level is 3 dB lower which is undesirable."
The gain of this circuit is normally governed by the ratio of the NFB resistor R28 to the input resistor R7. The calculation gives a ratio of 8.5, or 18.5 dB. The additional 3 dB you get at the moment are due to excess of positive feedback, which should not be considered normal operation.