Thanks a LOT for everyone offered help and advices!
Call me whatever, but after long search I found source for this not so obvious problem with distortions on low frequency/high signal level. The most difficult part was the fact that these distortions appear only below 50Hz starting from 10V of output RMS voltage.
When assembling amplifier, I occasionally swapped 33K and 47K resistors in phase splitter. Unbelievable, how so simple mistake lead to so long solving & difficult problem.
Additionally, amplifier seem to be very sensitive to twin triode mismatch in phase splitter, as well as mismatch of output tubes. As suggested Japanese tweaker, I have decreased NFB (down from 22dB), so amplification factor went from 10 to 15, THD increased a little bit but are not audible anyway.
After all, I've got very nice unit. Frequency response 20Hz-50KHz, +0.1/-0.36dB in range 20Hz-20KHz, noise level -89dB. Depending upon output power (and 6L6GC idle current 40mA), THD is in range of 0.1% - 0.96% (with pair of Shuguang 6L6GC which were overheated countless number of times, and draft assembly with very long interconnect wires all over the table).
Additionally, this amp proves push-pull output transformer design concept (method of equivalent generators with check at several critical points with and without DC BIAS) which allows to construct device which is very immune to idle current imbalance (without micro-air gap). Most units tolerate max 5mA, main has only -1.24dB drop at 20 Hz/40W and enormous imbalance, I1 = 50mA, I2 = 33mA (of course distortions are visible on oscilloscope, but sine-waves have symmetrical peaks and output voltage drop is very small). Transformer rated at 60W, I'm currently do not have amplifier to test at this power, but even at 40W this numbers are impressive.
A little more about square-wave test. I don't think it have much of usefulness, especially taking into account that interpretation of the result is quite difficult. Square wave composed of leading/falling edge (rise time of several MHz), and DC between them. This is not something large audio transformer supposed to work with. Take for example overshoot. Formula for calculating overshoot is quite complex, but combination of lower leakage inductance and lower distributed capacitance lead to higher overshoot. So, almost perfectly looking 2KHz square-wave on output may be a sign of not so good transformer after all.
Call me whatever, but after long search I found source for this not so obvious problem with distortions on low frequency/high signal level. The most difficult part was the fact that these distortions appear only below 50Hz starting from 10V of output RMS voltage.
When assembling amplifier, I occasionally swapped 33K and 47K resistors in phase splitter. Unbelievable, how so simple mistake lead to so long solving & difficult problem.
Additionally, amplifier seem to be very sensitive to twin triode mismatch in phase splitter, as well as mismatch of output tubes. As suggested Japanese tweaker, I have decreased NFB (down from 22dB), so amplification factor went from 10 to 15, THD increased a little bit but are not audible anyway.
After all, I've got very nice unit. Frequency response 20Hz-50KHz, +0.1/-0.36dB in range 20Hz-20KHz, noise level -89dB. Depending upon output power (and 6L6GC idle current 40mA), THD is in range of 0.1% - 0.96% (with pair of Shuguang 6L6GC which were overheated countless number of times, and draft assembly with very long interconnect wires all over the table).
Additionally, this amp proves push-pull output transformer design concept (method of equivalent generators with check at several critical points with and without DC BIAS) which allows to construct device which is very immune to idle current imbalance (without micro-air gap). Most units tolerate max 5mA, main has only -1.24dB drop at 20 Hz/40W and enormous imbalance, I1 = 50mA, I2 = 33mA (of course distortions are visible on oscilloscope, but sine-waves have symmetrical peaks and output voltage drop is very small). Transformer rated at 60W, I'm currently do not have amplifier to test at this power, but even at 40W this numbers are impressive.
A little more about square-wave test. I don't think it have much of usefulness, especially taking into account that interpretation of the result is quite difficult. Square wave composed of leading/falling edge (rise time of several MHz), and DC between them. This is not something large audio transformer supposed to work with. Take for example overshoot. Formula for calculating overshoot is quite complex, but combination of lower leakage inductance and lower distributed capacitance lead to higher overshoot. So, almost perfectly looking 2KHz square-wave on output may be a sign of not so good transformer after all.