High Voltage Variable Supply on a Shoestring

[Pafi]

Below is a simulation of the worst case, at the sinewave peaks of 325V chopped by the 4KHz switching frequency.
If the mains has an impedance ~=∞, the fundamental leakage is ~10Vpp.
This looks large, but of course the mains impedance is nowhere near infinity

Why infinity? It's closer to 0. Then you can see that basically all current (divided by turns ratio) goes to the mains. At much higher freq 1 uF can help, but not at 4 kHz.

The 2 states of switching cycle seen by primary in your model should be not 325V and 0V, but 325V and a negative voltage depending on DC output voltage and turns ratio.

 

[Elvee]

Yes, the high impedance fundamental voltage will be about twice as large.

The actual voltage seen on the mains will be lower, but rather unpredictable since the impedance of the mains at these frequencies could be anywhere between a few tenths of an ohm to a few ohm.

 

[kenpeter]

One of thing of note about single stage PFC supplies is that
the current ripple still has to go 'somewhere'. One project
had me driving a constant voltage sting of LED's, and I saw
all the ripple current go to the LED string. No matter how
much bulk capacitance I put across the output, the string
proved the favored route for the ripple, and that current
peak was higher than the LED's rating...

Now, there was a fancier two stage supply on hand. It
isolated and switched down from the bulk capacitance
with an LLC converter. And that was fine to resolve the
discrepancy between input ripple current and rippleless
output. But I realized if my first stage was isolated and
approximately the right voltage, then the second stage
is only responsible to divert ripple current to the bulk.

The LLC stage, even with zero voltage switching, was
burning up several watts. In copper and ferrite, but
also the output diodes, which were Schottky but not
synchronus. I had a better dumb idea, same as yours.

I added an LDO second stage to my isolated flyback.
And didn't LDO to a specific voltage, but low passed.
Target set by watching low volt of the bulk capacitor.
That provided many K ohms of dynamic impedance to
the LED string, and allowed the bulk capacitor size to
be significantly reduced while still soaking up enough
ripple to allow for PFC. Wasted half as many watts as
the over-complicated LLC...

My second stage tracked the PFC, but the other way
round could have worked too. Just so linear dropout
is never more than needed.

To clarify: The two stage supply was non-isolated PFC boost + isolated LLC.
The single stage supply (before I added the LDO) was isolated PFC flyback.

Though saved about a Watt and a half, reduced size and cost, my solution
was rejected, "Because that's not what anyone wants."

No good deed shall go unpunished...

 

[Elvee]

One of thing of note about single stage PFC supplies is that
the current ripple still has to go 'somewhere'. One project
had me driving a constant voltage sting of LED's, and I saw
all the ripple current go to the LED string. No matter how
much bulk capacitance I put across the output, the string
proved the favored route for the ripple, and that current
peak was higher than the LED's rating...

Yes, LED strings are pretty demanding loads: they demand a pure and good current control, which requires a specific architecture unless you use insanely high switching frequencies, or "tricks"

I added an LDO second stage to my isolated flyback.
And didn't LDO to a specific voltage, but low passed.
Target set by watching low volt of the bulk capacitor.
That provided many K ohms of dynamic impedance to
the LED string, and allowed the bulk capacitor size to
be significantly reduced while still soaking up enough
ripple to allow for PFC. Wasted half as many watts as
the over-complicated LLC...

A gyrator in essence....
Another option is to go for the real thing, ie an inductor. It also has drawbacks, of course, but at least it doesn't burn any watt (or so little).
It provides the "elasticity" needed between the converter and the LEDs