**broken link removed**
Completely new design
I've gone to a completely new design. It's more complex, but it works.
This circuit uses an LT3750 capacitor charger controller. I'd been avoiding that, because I didn't want to deal with soldering 0.5mm pitch components. But with more practice under the microscope, I can now do it. It's not fun.
The switcher here is the path from Vcc to the top of T1's primary, through MOSFET Q1, through current sense resistor R3 (0.061 ohm), to ground. The charger controller has a flip-flop which controls the MOSFET's gate. There's a comparator which monitors the current through R3 and turns off the flip-flop and gate power when the current reaches a setpoint. (Without that, this thing would short Vcc to ground in a few microseconds.) With the gate turned off, the voltage across the transformer primary drops, and when another comparator detects it's almost zero, the flip-flop and gate are turned on, restarting the cycle. This is a high-efficiency solution.
Over on the secondary side, there's the usual diode D1 and some capacitors, which get charged up to 120V. So that's the capacitor charger.
It takes tight layout and bypass caps to get this to work, but the LT3750 data sheet has good advice on that. This thing generates big spikes, some as narrow as 25ns, while oscillating around a megahertz. Efficiency is good; this can charge caps about 3x as fast as the previous design, while drawing only about 270mA from Vcc (+5). I've successfully run two different Teletype machines with a previous version of this board.
The latest rev eliminates a jumper and resistor used to set the output current, and adds a DMOS, Q2, set up as a current source at the output to make it self-adjust to different loads.
Entire project in KiCAD on Github