I have a hankering to make a high-voltage SMPS, but I want to avoid having to wind my own transformer.

I have been looking at some old off-line ATX power supplies, and they all seem to use a half-bridge to step down half the supply voltage (about 300Vdc where I live) to +/-12V etc. I think the supplies I have are similar to this circuit: http://www.eejournal.org/wp-content/uploads/2010/10/PC-SMPS-2003-schematic1.gif

Therefore, can I simply use the same transformer, but backwards, driving the 12-0-12V winding with a push-pull switcher to get about 150V on the 'secondary'? Bascially, can I use something like this, but substituting an ATX transformer instead of a hand-wound one?
http://2.bp.blogspot.com/-IGrcj4RjyYs/UOqa9ISSgXI/AAAAAAAAAYY/PpmE6sGtLck/s640/circuit+50kHz.png

 

What you wanna do with that high voltage?

 

Drive a tube amp. I was actually planning to voltage double the output to get closer to 300V. It would be nice to get 100mA output or even more, but I would settle for 50mA.

 

a much simpler way would be to use two filament transformers back to back. since you need a filament supply of 6 or 12 volts anyway, just get two transformers and connect the second ones low voltage winding to the low voltage winding of the first. then you have a nice isolated high voltage source.
cliff

 

a much simpler way would be to use two filament transformers back to back. since you need a filament supply of 6 or 12 volts anyway, just get two transformers and connect the second ones low voltage winding to the low voltage winding of the first. then you have a nice isolated high voltage source.
cliff

My sentiments exactly. Why spend days designing a high speed switching supply when I can rummage around in a junk box and fix this in 10 minutes.

I expect your junk box is not as good as mine, but still...the hours of designing compared to the price of a filament transformer? You must be kidding.

That's my opinion.

 

Thanks, but I don't actually have a tube amp problem that needs solving, that's just a vague long-term use for the project. I'm already well experienced in linear supplies.
I'm more interested in whether my idea is sound in principle, as it would be a great motivation for me to explore SMPS, and I like the idea of recycling dead ATX supplies for my evil schemes! (I actually enjoy spending days designing things)

 

That's reasonable.
Yes, I think you can do it.
Don't get too ambitious. Small tube amps, 10W output and less, run on less than 100 ma.

 

Be careful which switcher you use. Old ATX power supplies use fairly low frequencies and the harmonics make lots of noise into your audio circuits. Thee newer ATX supplies are running over 1MHz (I know they are faster but 1MHz is inferred based on the size and material in the inductor - I haven't measured frequency) and can be filtered more easily (I would assume - I have not tried since my first failure with an older supply.

 

OK, here's what I have at the moment. I got the transformer and KA7500 from the same PSU, and I have set it up on breadboard as an open-loop flyback running off a bench PSU, but I'm getting a strange effect.

The KA7500 is producing a nice PWM signal which I can vary from about 20kHz to 50kHz per MOSFET. But the the output voltage from the transformer seems to be fixed at close to 70Vdc, no matter what frequency or duty cycle I use. The signal at the drains of the MOSFETs is peculiar- there seems to be a huge oscillation during the dead time (see scope image). Anyone know what's going on?

Yellow trace is gate of T1, blue trace is the drain. The bench PSU is registering an average supply current of around 70mA.

 

you need to use ultra fast rectifier diodes, not ordinary silicon ones for the bridge rectifier, also the atx in normal mode pulses 350v to 30v, thats an 11:1 ratio, so if your pulsing 12v that should give you 132v.

 

you need to use ultra fast rectifier diodes, not ordinary silicon ones for the bridge rectifier

How fast should I be looking for? The ones I have now have a reverse recovery of 75ns, which I thought was rather better than general purpose rectifiers...

also the atx in normal mode pulses 350v to 30v, thats an 11:1 ratio,

I think in half bridge mode it effectively only sees half the supply voltage, no?

 

when i modify atx psu's to variable outputs, i can get them to go to 30v dc at most, if you have an old atx to hand use the dual diode package which are bolted on the heatsinks, or Shottkley diodes, some atx transformers are tuned to one frequency, does it give a different waveform output at different frequencies?

 

use the dual diode package which are bolted on the heatsinks

Unfortunately they're S20C40C which are only rated for 40V.

some atx transformers are tuned to one frequency, does it give a different waveform output at different frequencies?

The supply I got it from appears to have been set up to run at 37kHz per driver transistor. I have tried setting mine up for the same frequency, but it seems to make no difference. The spurious oscillation seems to have a fixed frequency (150kHz?), so all I can do is increase the switching frequency until the dead time is too short for the oscillation to happen, but the output voltage remains pretty much constant (varies by a volt or two) whatever I do with the frequency or duty cycle

You can see it here more clearly when I drop the switching frequency down to 20kHz. (yellow = gate of T1, blue = drain.

 

Another question: Does a push-pull switcher like this require a minimum load current to work?

 

OK, I identified the original primary snubber (1n + 100R) and re-used it on my high-voltage 'secondary' in an attempt to kill the ringing. The drain waveform has certainly cleaned up, but the range of output voltage seems extremely limited, ranging from about 65V at minimum duty cycle, to 70V at 49% duty cycle. Is that normal??
Yellow = gate, Blue = drain.

 

have you built this, or is it in virtual world?

 

Given the screenshots are from a rigol oscilloscope, i'd say hes built it irl.

 

Yes, it is built on real live breadboard!