Just a supplement: the interesting part is IMO discharging the caps through the flyback chokes. It makes the pulses at no additional energy cost (beyond what has already been transferred through the 100M resistors), so they are both signal and power at the same time. Unfortunately, the amount of energy stored in that 470p capacitor charged to 1V is so tiny that the autotransformer is not able to boost the voltage (merely ~150mV at the low V_IN end) to open Q3. This is a constant energy regime: if you add more turns to L3, the voltage reaches some saturation point and the reset pulses are getting longer instead, as L and C both go up. This saturation point is ~400mV, too low for an NPN to notice (assuming zero power detection; you can't afford a proper PNP long tailed pair). A 400mV EPAD would cut the mustard, but another solution is to double the amount of energy deposited to the core. Hence the dual-pumping idea utilising L1 and L4. With that it works even down to 1.1V V_IN. Balancing these available picojoules is sort of obscene. :)
Best regards, Piotr