Hmmm, could one build a switching-power-supply with relays...?

I'm sure that it was done 70+ years ago.
Poor poor relays.

I think there existed some mechanical rectifications... and somebody mentioned a similar subject lately.

Interesting stuff here: 

http://www.secondchancegarage.com/public/83.cfm

Current-regulation and Voltage-regulation using relays and resistors. It looks like they "pre-bias" the relays by having two separate windings, one always active, but apparently not quite strong enough to suck the relay in, but when both are active, then it does... Not at all unlike what you've done with the biasing via resistors.

Anyways, this is all thoughts per @Ted Yapo's earlier "crazy thought" about trying a current-source, and your wanting to keep this system within the tech available in the WW2-era.

Another thought... You seem to be biasing your relays such that they have a single control-input... have you given any thought to whether some bits of logic might be simplified by controlling *both* sides of the relay-winding?

My friend, where are you dragging me to ?? :-D

OK so at least I know it was done, and the 150W behemoth I imagine could be powered through this, a large inductor (from a big transformer? a phosphorescent tube starter/ballast ?) and a crazy stack of capacitor.

Wait, I have such a stack that is waiting to be used... OH MY!

multi-input logic : yes I did. Look at the logs, I think I wanted to try the majority function, but the hysteresis is much too  prominent so it falls apart.
So far I seem to even be able to avoid the use of diodes, only resistors and capacitors are absolutely necessary. I make exceptions for non-esssential parts such as LEDs, which I can replace with Glühbirnchen but I wanted to be kind to the power supplies: 50mA per bulb, ×16bits×8registers=6.4A... If I load 0xFFFF in all the registers.
After all, good silicon diodes appeared in the 60s. Ge point-contact diodes couldn't sustain much current. Selenium was problematic.

"My friend, where are you dragging me to ??"

Nothing, nothing at all :) Just an interesting mind-experiment you and Ted brought me through...  "could voltage-regulation (or a current-source) be done with relays?" I had a vague image in my mind, but not the brain-power to design it... so search-fu worked in my favor this time! Just another thing to throw in the ol' toolbox/bag-o-tricks for later!

Don't Do It! ;)

I won't implement electromechanical regulation, don't worry :-D

I'll consider that the new version has enough ripple tolerance to work with an "unregulated" diode&cap AC/DC "bulk" converter (like they love to do in the "audiophile" world, "the more farads, the better")

And I'll just drop 3.3V POL modules, as I first described in #SPDT16: 16-bits arithmetic unit with relays to save power.

However it won't be a complete solution because the sensitive parts remain the hysteresis latches, I'll have to build #ReTest-RPi to sort the relays and set apart at least 150 of them with very closely matched trigger characteristics.

They were used in the first switching power supplies!  They were known as "vibrators" in the day.  The term has other connotations today.

https://en.wikipedia.org/wiki/Vibrator_(electronic)

I use to make high-voltage generators like this with relays as a kid.  They made that nice ozone smell and didn't last very long...

Did the same... but that's a *boost* converter... (and a "shocking" one, at that).

I'm talking about a *buck* converter... for your current-source-needs.

Yeah, that would kill the relays very fast...

With my system, I try to limit the current with a series resistor and so far it seems to work. But I should run a more thorough test with more relays (a dozen ?) for a longer period. That test is keeping my lab PSU busy for nothing...

Anyway, I don't believe I would use the final computing system for extended periods :-P however going from some tens of relays to 1500 will make unexpected problems surface so I try to be safe. I don't want to leave a stone unturned.

Yann, if you can design your logic so that about the same number of relays are always pulled in (like in the ring oscillator where it's either 2 or 3), maybe you can power the whole circuit with a current source so that they seek their own appropriate voltage.  Just a crazy thought, not sure how well it would work.

It's not a crazy thought.

Actually, at the moment when you were writing this, I was playing with the lab PSU in "current limit" mode, and saw the voltage wiggling.

Now, the question is more about practicality.

I'd like to design somethign "that could have been built during WW2" so no semiconductor voltage regulators... Current is proportional to supply voltage (with the resistance/conductance ratio) so I'll stick to "somewhat regulated power supplies".

I'll see how the range drifts...

The best current source I can think if is an infinite resistance in series with an infinite voltage...no semiconductors required :-)

Sorry, I can't find this on eBay or in junkyards :-P

Kind of hard to figure out the current when ∞/∞ is indeterminate...  :P

http://valvewizard.co.uk/ccs.html
>Using a triode as a constant current sink

Muahahahaha !

And please, no triode :-D

I'll just assume a 3.3V PSU with less than 100mV ripple coming from a huge toroid transformer and a crazy bank of capacitors.

@K.C. Lee Yeah, I always have that problem when I work on DSP projects: sinc(0) = sin(0) / 0 = 1 ?  [ insert joke about L'Hospital which I can't think of... ]

You could try a magnetic circuit if you insists.  :P

https://www.google.com/patents/US3397361

"Magnetic amplifier utilized as a static constant current regulator"