richie burnett wrote ...

Those yellow (type-26) iron powder cores in PC PSUs are the main "buck
choke" and do indeed set the ripple current for the supply. The reason for
the little ferrite chokes that follow is that the main iron-powder core is
usually wound with multiple layers and multiple windings. (Sharing one core
improves cross-regulation of outputs because feedback is only taken from the
highest power rated output.) Interwinding capacitance in this "monster
choke" couples the fast switching edges through to the output capacitor,
particularly at maximum line voltage. Therefore it is followed by a single
layer ferrite "stick" inductor to remove high frequency spikes that would
drop voltage across the ESR of the output electrolytic. They could have
made a T-filter by placing a capacitor to ground between the two inductors,
but it's effectiveness is limited because of the self capacitance of the
buck-choke and any ESR of the capacitor chosen for this position. It also
introduces another pole in the transfer function which makes closing the
voltage loop very difficult.

This "two inductors in series" technique is quite common in active PFC
applications too. Most App Notes and text books say it's essential to wind
boost PFC chokes with a single layer to minimise end-to-end capacitance -
otherwise you get a lot of HF hash fed straight back from the switching node
into the supply. This is true, but it is certainly more compact and
cost-effective to make more use of the winding window and reduce HF hash
with a ferrite choke in series. (Or use ripple steering at high powers.)

wrote ...

Hi James,

I just noticed this post from you on 4HV, and have some input if you are
interested...

> I've tested Titch on a bench dc power supply (30V, 2A linear) with no
issues but when using the modified PC supply, ramping the power up to more
than what I estimate to be a maybe 10 watts (coil running interrupted at
50Hz, about 20% duty) the power supply cuts out in a similar fashion to when
the output is shorted out unintentionally.

This is common behaviour when you try to use a voltage-mode controlled SMPSU
to charge a capacitor in pulsed-power applications. I get exactly the same
behaviour if I try to use that Xantrex 600V-1kW supply to charge a cap bank
for a pulsed SSTC. It trips out on over-current when the SSTC is fired up -
even at low power.

> I've measured the output voltage - With 4 x 820uF, low impedance
capacitors on the output, there is about 0.5V to 1V of ripple when the coil
turns on.

That is the cause of the problem. Once the SSTC is fired and it discharges
the capacitors by a volt or so, the output of the PC power supply *DOES*
essentially see a short circuit. The supply is connected to a bunch of big
capacitors that are pulling its output down to a potential lower than its
output set-point. This is what trips the over current protection and causes
the supply to shut down.

> I haven't measured the ripple current, although I suspect that the pulsed
currents are enough to trip whatever over current protection exists in the
supply.

I would bet that the current out of the PSU skyrockets and hits the
over-current trip point when the capacitors are discharged.

In order to smoothly charge capacitors in a pulsed application you really
need to use a power supply with a constant-current output. Or at least use
a power supply that limits to a constant-current line when overloaded
instead of folding back or shutting down. A constant voltage power supply
that uses current mode control for the inner-loop and an outer voltage loop
can be made to work in the way that you want. This is because peak current
is still bounded even if the output voltage goes out of spec and the voltage
error amplifier is saturated. Just watch out for the voltage overshooting
when it charges the capacitors back up to the voltage set-point.

Unfortunately most cheap shitty PC power supplies use archaic topologies,
ie voltage-mode control with peak-current limiting via hiccup mode. In a PC
power supply s/c protection is important to prevent destruction when someone
accidentally shorts the output. Constant current limiting is not important,
and is probably undesirable because it would burn up PCB traces on
motherboards with shorted peripherals etc. Unfortunately there isn't much
you can do to alter the way the current limiting works. If you delay the
onset of current limiting the supply will most likely blow-up under a
short-circuit test, so this isn't a wise move. The best solution is to
build a current-mode buck or forward converter with average current mode
control or even a small flyback converter if the power level is below 100W.

I hope this helps,

-Richie

Marko wrote ...

I dremeled off all the connections to the main transformer secondaries (including center tap ground) and put up a full-wave (made of schottky diodes) rectifier + smoothing cap on the 12V winding. The feedback is taken from the cap to the TL494 pin 1

You did not use a buck inductor? Huh?

Marko wrote ...

I dremeled off all the connections to the main transformer secondaries (including center tap ground) and put up a full-wave (made of schottky diodes) rectifier + smoothing cap on the 12V winding. The feedback is taken from the cap to the TL494 pin 1

You did not use a buck inductor? Huh?

Dago wrote ...

I'll have to try the grounding too but I'm hoping to series these overclocked PSUs so I'm trying to avoid it.