Sean Breheny wrote:

> I don't think the GNDs for PC supplies are isolated from mains ground,

  You're right!  I thought someone would twig to that.

> so unless you float all three supplies, I don't think you can just tie
> GND to 5V on adjacent supplies.

  It's fairly straightforward - you open it up, identify which screw or
screws mounting the PCB are on a output ground land, and having removed
all the solder from the track, carefully cut the track all around the
mounting point.

  If you can figure an *absolutely reliable* way of simply insulating
the track without warping the PCB, that might be good too.  Note that
you must not remove the green wire from the IEC input connector to the
chassis, and you must not insulate the PCB mount points which ground the
input interference filter capacitors.

  While you're at it, you would probably un-solder all the output wires
(those connectors are too valuable to cut leads) and re-solder similar
ones to make the interconnects, about six or eight wires per terminal,
like the original, either directly from one PCB to the next (effective,
but unwieldy) or using (insulated) brass terminal blocks for
interconnect.  We could have a side argument about whether the leads
should be tinned before being screwed down.

>  I'm not sure, but I would probably want to check that the GNDs are
> isolated from the neutral wire, too.

  Check by all means, but the UL/ CE listing on the power supply should
guarantee this for supplies marketed in first-world countries. :)

Alan B Pearce wrote:

> This sounds like a constant voltage transformer.

  It's not, the description was for a constant-current or current-
limited transformer as used for neon signs.  Also called a "ballast"
transformer, with separate windings for primary and secondary and marked
magnetic leakage between.

  In fact, oven transformers have very little current-limiting action,
they're just power transformers.

>  They are commonly used in voltage stabilisers, and have a winding on
> them with a capacitor across it.  The value of the capacitor and the
> inductance of the winding are critical at the mains frequency and
> voltage, and changing either of these parameters means changing the
> tap the capacitor is connected to. I used to service a pack type disc
> drive that had one of these in its power supply.

  That description is correct; they are *very* inefficient as they
regulate by saturating.  Kinda like a shunt regulator - in fact they are
a shunt regulator of sorts.

> I believe it is actually possible to short circuit the output of one
> of these transformers without burning things up, but I never actually
> tried it.

  Whew!  Lucky!  You would have learned the difference between voltage-
limiting and current limiting, the hard way!

Chris Eddy wrote:

> Controlling phase of the AC mains on the primary side of the
> transformer is done, but I do not know the pitfalls.  It will probably
> cause horrible power factor figures, and will generate a lot of RF.

  Do remember that this applies equally to switchmode supplies as in PCs
which use reservoir capacitors on the primary.  In fact, the design
brief for this application is quite different because secondary ripple
and absolute regulation are unimportant.

  High power SMPS should use no primary capacitors (well actually,
little ones) so that their input follows the full (or most of) the
(half-wave rectified) input waveform.  This is implemented on "low power
factor", "power saving" compact fluorescents and similar applications
now (and it's cheaper!).

>  Will the transformer have to be better or higher bandwidth than a
> simple fillament transformer?

  Not particularly.

>  The up side is that you can put SCR's on the primary side and just
> rectify the output with relative ease.

  Be warned - you *must* make sure that the SCRs switch positive and
negative cycles at the same point so that there is no imbalance and net
DC current.

>  And the frequencies are lower.

  Big iron.  That's what you are trying to avoid.

>  And if the phase control method has poor speed response (suspect that
> is why they are not common in PSU's) who cares?  The cars will not
> know the difference.

  Correct, correct (using batteries).

  Finally, car batteries rather than deep-cycle are pretty much what you
want for this application because you want to smooth peaks and propose
to keep them charged otherwise.  Starting a car is basically that.
--
  Cheers,
        Paul B.