Any thoughts on my latest brain aneurysm (Reply #26 - now with picture in the text )?
First thing, not related to your question, is that I always forgot to write you can just use a depletion mosfet and a resistor to get a current source withstanding 100s of Vs, rather than an LM317.
Now returning to your circuit... I've not simulated it but got the feeling that, with enough load current, it would be a great noise generator, a sort of low frequency inductor-less buck converter, but with unpredictable behaviour due to the current leakage through the zener diode which might be enough to switch on Q4 depending on the temperature... a resistor between BE would solve that problem, but not the other ones
You might reduce some of the spikes and get some more hysteresis through a capacitive divider, i.e. connecting for a short while a 100-470µF capacitor from the transformer to the 12.000µF one, it would need a series resistor to limit the peak current, a Pmosfet to open/close the circuit, a reverse polarized schottky diode to dissipate the capacitor charge when you switch off the Pmosfet and some sort of timer or another higher threshold Vpre-Vpost comparator to keep it switched on just a while... so it would just became much more complicated with little benefit.
If a vintage solution is your main goal, if I were you, I would explore an SCR solution...
The clean way is to charge a capacitor through a resistor discharging it on each zero-cross to get a 100Hz sawtooth sync-ed with the transformer secondary.
Then you compare the difference of Vout - Vprereg with the sawtooth value and as soon as the sawtooth goes above the delta you switch the SCRs on.
A several mH inductor (like those used in passive PFC PSU) would avoid a sudden rise of current through the transformer and when the voltage gets down so that there is no current through the SCRs and they switch-off, a schottky diode would let to discharge the rest of the energy stored in the inductor to the levelling capacitor.
As far as I can remember (I ran the simulations about one year ago), at least in the magic world of LTspice
, that circuit can provide high efficiency and low noise if you use mosfets rather than SCRs and update constantly the Vout-Vprereg threshold according to the set output voltage, the output current and the transformer voltage variations (not instantaneous, but the RMS ones, i.e. those due the AC line) and put a large non polarized capacitor in parallel to the transformer secondary.
But you might just keep it simple, still saving some power, by using a high enough Vout-Vprereg threshold and skipping all those sophistication reducing so the required components.
That is:
- skip the sawtooth comparison circuit which charge the levelling capacitor as suitable each 10 milliseconds
- just use a larger threshold and switch on the SCRs as soon as the voltage gets below that
Sometimes it would get 2 cycles to return quite above the threshold or you might skip many cycles when you have low load and so would have some very low frequency noise depending on the deltaVout/deltaVin capability of the linear regulator, but it should work.