Rds(on)*Coss (or Ciss or etc.) is a figure of merit that varies from generation to generation (so, look for newer parts), and proportional to the stats of the given material in use. So, among materials in present use for power switching, Si > SiC > GaN. Lower FoM being better, of course.
But again, it sounds like you're trying to solve a problem that no one else in the world has. Maybe there's some merit to that, that it's a new and unique problem; but it's more likely that you're approaching it from the wrong angle, and standard solutions exist, or if not solutions like existing products, then at least standard analyses that could be used. Suggesting, perhaps, further gaps -- in signals analysis, or network theory, for example. (Which should hardly be an embarrassment: most graduate EEs barely skate by signals theory as it is, and network theory isn't even part of undergrad, at least in most places, and I mean actual network theory, beyond the basics of AC steady state analysis.)
And no, of course not really "no one else", there's pulse type metal detectors for example which sound very similar. But why not just whole-ass pulse the coil, why use capacitors at all? (Other than switch capacitances, which will be easier to dampen against coil and stray inductances, than an extended network will.) Otherwise, you're just tuning for a narrower frequency band, at the antenna itself (the antenna is part of the signal filter chain), and starting and stopping any resonant current in the antenna, faster than its own time constant (~Q/Fo) would decay, simply isn't gaining you much of any signal energy (which only manifests over the same time constant!). More importantly, you're setting yourself up for all manner of nonlinearity (MOSFET Coss varies extraordinarily widely with Vds), making dynamic range a huge challenge. And maybe all this can be compensated out, or trimmed, or clever solutions found; but do you really want to spend so much time on that, only to find you've missed your targets on SNR or field strength or etc. and need to redo the whole damn thing -- and the whole thing will be keenly sensitive to device parameters so you have to go through the whole tuning process again?
A key insight, by way of anecdote:
Also, not to project someone else on you; not that you know anything about them other than I'm writing right here and now, obviously, but just to be clear, the guy was a bit of a jerk.
Anyway, bossman from my first job. Induction heating power supply design. He had an idea. Concern was, their existing designs were all so stupidly slow. And, they generally were: most used dominant-pole compensation with a time constant of fractional seconds; the audio-frequency ones, you could hear, feel even, as they ramped up. There's not really any reason to choose that, other than it's slow enough not to care about any particular load, and being that induction heating applications are usually pulsed over the seconds time scales, and the startup time is a constant adder, it's not really a big deal for anything like process consistency in production. But anyway, his idea was, not just to speed them up a little bit, but to create a
new, revolutionary design with
sub-cycle control that could meter power into the tank on a 1/4-cycle sample rate basis! It would be a digital control and FPGA and ARM soft core (??) and all that and it would be amazing and blow the socks off everyone else and...
Well, you can probably see a couple of odd points standing out from that already, and yes, they were odd, and have odd explanations (ranging from "we got the license might as well use it" to "we must succeed where others have failed"); but those aren't important here... Anyway, the relevant part here was the obsession with 1/4-cycle control.
His thought process was, if voltage and current are swapping around every 1/4 cycle, ...can't we
do something about that?
Alas, no amount of argument would convince him that 1/4 is an utterly arbitrary measure, or that no, we simply
can't, it's a physical impossibility; granted, I knew hardly anything about network theory at the time, so it's not like I could offer any sort of comprehensive proof; but also, now that I do know more about it, it's not like a comprehensive proof would've been at all convincing anyway (and, to be fair, I still don't know how I would go about constructing such a proof from first principles). That is: network theory is on a high enough level that, one cannot understand a proof written within it, without also understanding much of it; and any simplification or reduction thereof, must be taken for granted, and at that point it's just a confidence game like anything else.
Bossman had a degree in EE at least, maybe not at a practicing level for some years (decades?), but enough to be dangerous anyway; but also a degree in business, and I think the latter was the more primary interest. And, what is business but a confidence game? You roll the dice on new technology and you get what you get, right? Well, not to get political here, but you can see where that kind of mindset comes from, and what effect it has on, well, all of us, to some extent or another...
In the end, I convinced him that, no, we're not going to do 1/4 cycle control, we don't have the
time and resources to develop that, but you can have 1-cycle control (which was done at a sample rate of 1 cycle with a pipeline delay of 1, so the actual maximum control bandwidth was 2 cycles, but shhh..).
Anyway, these days, I would present, at least the outline of a proof, something like this: for a given Q factor, and given excess inverter capacity (typically the inverter VAs are, say, double the required power, to allow some mismatch / adjustable range, without having to change the tuning settings -- transformer or inductor taps, resonant capacitors), the difference necessarily
must resonate in the tank by itself. That is, say we have 100kVA in the tank, a Q of 10, and inverter capacity of 20kVA. We're delivering 10kW and have 10 to spare*, so at most we could change the resonant tank energy by 20kVA per cycle, or five cycles to fully start/stop it.
*Which, if it's reactive, could then be 17.32kVAR, they sum vectorially; but if we're talking increasing or decreasing tank energy in as much of a stepwise manner as possible, we'll want that to be mostly real power, since we're delivering power to, or drawing power from, the resonant tank, exchanging with the DC link (which then also needs at least enough capacitance to absorb the tank energy at once, which will usually be fine). So we might constrain the reactive power output, and adjust timings to optimize for real power to dominate those 20kVA.
Which you'll notice is a more basic sort-of-time-domain and energy-based argument, which is also more convincing. It gets harder when the network is more complex; for a voltage-fed series resonant, or current-fed parallel resonant network, the above is straightforward, but it's less obvious if it's a voltage-fed inductor-matched parallel resonant circuit for example, which is actually what we were tasked with at that job.
And, control can be faster when Q is low, or indeed when Q <= (inverter capacity factor), it can be direct drive entirely (who needs capacitors!), but those are also less interesting as there aren't as many applications where that is true, and, well, it's fast to control regardless, so it doesn't really matter if you have a "1/4 cycle update rate" or anything.
Anyway -- that was quite a long story, but I wanted to make it clear I know where you're coming from (at least, to an extent), and provide some background about what I'm suggesting and why.
So, to say that I've thought about these things before, things like hard shutoffs, or fractional-cycle control; and I will likely think about them again from time to time. As have others; and indeed it seems at least one person had made it a small obsession -- always stuck sitting in one place, the destination in sight but the path otherwise unknown, lacking the knowledge to be able to circumscribe the canyon separating them from their destination. It's not that these are strictly impossible problems (you can scale or bridge a canyon!), but their solutions generally aren't practical; even if you resolve all the transients, and keep noise down, signal quality up, what do you have to gain, a dB or two of SNR? Such extensive effort needs justification; on what theoretical grounds can you prove such SNR improvement is possible, right? Stuff like that.
Also to be clear, I'm probably coming on too strong with things here, maybe a gentle nudge would work better than "you're wrong and this is why", well not quite that strong, but that's probably still within the realm of what one could read it as...
I'm not coming at this from a position of -- well, okay, to be even clearer, ego is inseperable, that's the human condition unfortunately, but I
at least like to think that I'm more interested in the science of this stuff, than to merely be right. I can't say it's not pleasing to discover things that are right, or point out things when they are wrong. But science is cool and everyone should know some, and I at least like to think that's the stronger priority.
And yeah, knowledge gaps happen, learning is the process of being wrong, less and less often; and humility is hard, especially as you learn more things and get (over?)confident in some areas, but that doesn't always translate to confidence in others. (You can read into this, whatever amount of projection you like, lol.) My hope is, not to point at gaps and laugh, but to highlight what misconceptions might've formed around them, and to show pieces that fit better. The real challenge is, the pieces are extended, with lots of wiggly tendrils that connect to myriad other topics; it simply takes years to develop that network of knowledge, no length of forum post can do that all at once.
Cheers
Tim
Home
Help
Search
Login
Register
