Tim, I had trouble understanding some of your post ![Confused :-//](https://www.eevblog.com/forum/Smileys/default/confused0024.gif.pagespeed.ce.5xOqKkq0Co.gif)
You are suggesting using two core materials in a stack? That's a great idea it sounds like the only way to get big bandwidth.
Yes. The permeability of a core drops with increasing frequency, just as the gain of a low-pass filter drops with increasing frequency (well, "just" is subject to material variations, with some falling faster than others).
So you can very easily have a modest-mu, but high frequency, material surpassing the permeability of a high-mu, low frequency material. Two combined means the best of both worlds!
I did not know anything about nanocrystalline material. Tapewound only cores, in strips and I haven't seen it before. Seems really good for high mu.
So I'm trying to visualize the "sandwich" - "wideband ferrite" and a tape wound core, two different (for example) toroids stacked?
Yup. Here's a listing of easily found parts:
http://www.mouser.com/Vacuumschmelze/Passive-Components/EMI-Filters-EMI-Suppression/EMI-RFI-Suppressors-Ferrites/Ferrite-Toroids-Ferrite-Rings/_/N-bw7t9?P=1yyvu4f(Magnetec is also German, and may well be a rebrand of VAC material.)
The permeability starts to drop at a mere 10kHz, which is where they measure the inductivity. So you will not have an inductor (i.e., impedance proportional to frequency) above that frequency -- still, figures like 40uH/t^2 are hard to beat, and despite the drop in inductance, the impedance (which is really what we're after, here) remains quite high into the 10s of MHz.
But I got lost thinking wavelength or transmission line for transformer design. I mostly use the core's AL value and inductance/Z target at the lowest frequency as a starting point.
The fundamental problem with the textbook introduction of a transformer is, you
never put the windings on opposite arms of the core. Because that's not where the waves are travelling. Waves don't magically jump from one winding to the other!
Flux is fine at DC, but not AC. It should be a huge clue that waves matter, because the winding length (that is, the actual wire used) is comparable to the wavelength, in the high frequency limit!
I would much rather see the transmission line model taught: a transmission line is a pair of pins at one end, and a pair of pins at the other. Each end exhibits a characteristic impedance, and each end communicates with the other through a propagation delay.
Transmission lines are easier to analyze with matched impedances in the time domain, whereas capacitors and inductors yield to frequency-domain analysis. It may well be easier to work with transmission lines, intuitively speaking -- the waveforms are what you see on the oscilloscope, in the time domain. You don't even have to understand Fourier analysis.
And that's all you're doing here -- you're using the impedance of a transmission line to couple one side to the other. Since you need isolation, you're doing it "sideways", so to speak: each end of the transmission line (plus and minus) connects to input (plus and plus) and output (minus and minus). Consider a coax cable: connect the shield to input (say), and core to output.
For very short time scales, each TL pair looks like a resistor of the characteristic impedance. Apply a voltage step, on the primary side, and that voltage is dropped across a loop of four resistances: the source resistance, the first TL pair, the load, and the second TL pair. (At low frequencies, the TL pairs communicate with little phase shift, and their voltage drops cancel out, thus giving low insertion loss. The cutoff frequency is where the phase shift between the two ports changes that significantly.)
I would recommend twisted pair for an isolation transformer -- there's no reason for the primary or secondary to be asymmetrical (it's a transformer, it works equally well both ways), so the winding might as well be, too.
Real cables have common mode impedance as well, but we can make reality approximate theory by wrapping the cable around a magnetic core. This gives a very high impedance between ends of the transmission line. That approximation breaks down at low frequency, giving the usual constraint (inductance and number of turns).
If you don't need any low frequency range (say, a bandwidth of around an octave), you don't need a core at all! The transmission line does the job on its own. These are common enough in RF applications, where the bandwidth doesn't need to be much. (Also common are 1/4 wave transformers, which are resonant, so even less bandwidth. That's a different thing though -- harnessing the power of waves directly, not just transmitting them with good matching.)
It looked like I would start 1/4 wavelength with say 1m coax/twisted pair, but is that physically large? Most core apertures are only a few mm-cm.
I do need isolation between primary/secondary.
Well, 1m is the maximum, but you'll probably need it. Also, the transmission line's cross section is arbitrary -- you could use high power hardline, but that would be... nevermind.
![Laughing :-DD](https://www.eevblog.com/forum/Smileys/default/smiley_laughing.gif.pagespeed.ce._hfWAz_QHO.gif)
Thin coax, or twisted pair, is going to be the most common choice.
You can't go so thin that the insertion loss (which will vary with frequency) blows out your flatness spec either. So, DCR << 1 ohm would seem to be a good idea. Still, that gives you anything from 36 gauge and down.
No, it probably won't fit on a binoc core, but those aren't the best choice for wideband transformers. They're best for small signal applications of modest to high bandwidth (or, often, of narrow bandwidth, but just so you don't have to care about what it is).
Tim