I've been thinking about releasing this idea into the wild to see what happens for quite some time now. I tried to get away from the ring oscillator method of spacing the phases, but haven't come up with anything that works comparatively well let alone elegantly.
There are some known problems.
Current sharing in the real world. In spice everything is intrinsically matched, so this problem goes missing there, although purposefully drifting various values has some interesting effects. For one thing there usually has to be at least the minutest obvious difference between phase legs or all of them tend to stay marching along together as if it were one big parallel inverting leg and complete waste of redundant circuitry.
The funny diode clamp at the comparator input is responsible for the "carrier-less" frequency locking. However, there is an irritating limitations with it that is kind of hard to explain because I don't even fully understand how it works when it does. But if you play with this thing long enough you soon notice that the opposed pair of coupled trapezoids coming off the capacitively loaded clamp end up getting compared against the long ends instead of the short ends. The result it that instead of slopes that run linear all the way up to 0/100 of duty cycle, they come apart and the phases let go of each other as the whole oscillator plan blows up, until the input signal drops down to the point where comparator inputs cross again at regular intervals. Whoever comes up with a way to get something like this to run all the way up to 100% duty cycle while keeping all the legs locked together and balanced, gets the prize.
Oh yeah, and in reality you need a handful of isolated gate drive supplies. 🙂
I have built a single phase version of this with a nifty little resonant gate supply and regular silicon FET output. If you take the clamp stuff away it basically turns into a Putzeys amplifier. Even one phase gets to be a lot of tiny transformer windings. For the pulse transformer I ripped the base drive core out of a bunch of old CFLs. For the comparator I used a AD790. I tried a real build with LT1711 on a ground plane board and it oscillated to beat the band. I mean that thing would not stop talking to itself. I might have been able to tame it but went another way. I doubted it would be better than the 790 in this application anyway.
If someone did this already, especially if simpler or better, let me know. EPC models are big and clunky. To run this you'll need them EPC Device Models
There are some known problems.
Current sharing in the real world. In spice everything is intrinsically matched, so this problem goes missing there, although purposefully drifting various values has some interesting effects. For one thing there usually has to be at least the minutest obvious difference between phase legs or all of them tend to stay marching along together as if it were one big parallel inverting leg and complete waste of redundant circuitry.
The funny diode clamp at the comparator input is responsible for the "carrier-less" frequency locking. However, there is an irritating limitations with it that is kind of hard to explain because I don't even fully understand how it works when it does. But if you play with this thing long enough you soon notice that the opposed pair of coupled trapezoids coming off the capacitively loaded clamp end up getting compared against the long ends instead of the short ends. The result it that instead of slopes that run linear all the way up to 0/100 of duty cycle, they come apart and the phases let go of each other as the whole oscillator plan blows up, until the input signal drops down to the point where comparator inputs cross again at regular intervals. Whoever comes up with a way to get something like this to run all the way up to 100% duty cycle while keeping all the legs locked together and balanced, gets the prize.
Oh yeah, and in reality you need a handful of isolated gate drive supplies. 🙂
I have built a single phase version of this with a nifty little resonant gate supply and regular silicon FET output. If you take the clamp stuff away it basically turns into a Putzeys amplifier. Even one phase gets to be a lot of tiny transformer windings. For the pulse transformer I ripped the base drive core out of a bunch of old CFLs. For the comparator I used a AD790. I tried a real build with LT1711 on a ground plane board and it oscillated to beat the band. I mean that thing would not stop talking to itself. I might have been able to tame it but went another way. I doubted it would be better than the 790 in this application anyway.
If someone did this already, especially if simpler or better, let me know. EPC models are big and clunky. To run this you'll need them EPC Device Models
Of course wherever a forescent score motion is required, or for use in uniliteral phase detractors, you can always use a drawn reciprocation dingle arm.