i guess i disagree, that common mode chokes are wound for maximum leakage inductance.
Perhaps it would be better to say: they certainly aren't wound in any manner which would achieve a reduction in leakage (interleaving, multi-filar, paralleled strands..). Maximum leakage, while retaining current compensation, for a given core geometry, would probably be something like, having a dense multilayer construction for each winding, so the length of core occupied by each winding is minimal (and the distance between windings is maximal), and on a toroid for example, they would be placed exactly opposite each other.
The common "single layer on opposite sides of a toroid" construction gives a pretty close version of this, so those types probably achieve somewhere near "maximum leakage".
The bank wound style (used for C, E and figure-eight type cores) is probably a little higher in coupling coefficient, since the windings are adjacent on a common leg.
here's what i think is going on.
most common mode chokes have a relatively non optimal core topology.
i'm talking, a space available for winding copper in.. on the order of 7mm by 12mm.
yet the core cross section is only 4mm square.*
so naturally, when two coils are wound on a bobbin that offers them 6mm of creepage distance between the two coils.. yeah, there's a lot of leakage inductance, but there isn't really much that can be done about it. some common mode bobbins can be unwound without breaking the core.. and you'll find there is no air gap, the bobbin is in two pieces and is wound up on a machine.. these typically have 4 slots for copper to be wound in, and i suspect if you configure them psps you can get a relatively low leakage inductance.
Yeah, the figure-eight cores are solid, and that's how they can achieve mu up to 15k, even 20k, in ferrite. Pretty impressive, but a pain to wind. The bobbins are usually two-piece affairs, with a peculiar ratchet tooth shape going all the way around one side... gear drive, wound in place.
The cores themselves do tend to have a small cross section given the size, which is mainly to accommodate the large amount of copper and insulating material; they hardly need any flux density handling at all, whereas flux is a key aspect of transformer design.
Both considerations, together, mean you need that much more winding length for your GDT, which means more leakage.
BTW, leakage for a "twisted pair" construction is essentially the stray inductance of the pair to begin with. (After all, leakage is the flux that's NOT in the core, so the presence or absence of the core makes almost zero effect.) Given that twisted pair is typically around 100 ohms (transmission line characteristic impedance), and mu_0 is 1.26 uH/m and Zo is 377 ohms, you can figure the inductivity is around sqrt(100/377) times mu_0, or in the 0.7 uH/m range. If you need 1m of twisted pair, you'll have around 0.7uH leakage.
Which, in turn, if you have a nice low impedance driver (gate driver IC with < 5 ohm resistance?), means you effectively get a resonant tank between driven source, 0.7 uH series inductance, and whatever the gate capacitance equivalent is (10nF is pretty typical for power transistors of broadly this size; that would be 100nC at 10V). To minimize overshoot, you need to dampen that resonance, which has a characteristic impedance of Z = sqrt(L/C) = 8.3 ohms. A little extra series gate resistance might be handy (~4.7 to 10 ohms?).
If you're driving a half bridge, you'll have the same thing from the other transistor, acting in parallel, and you'll end up with the driver chip working into a resonant impedance of 4.15 ohms. A pair of 2 ohm drivers (~3-6A rated capacity, note: use CMOS type, not BiCMOS "boosted" kinds), acting in series with the LC (leakage + gate capacitance), should be just about critically damped, and pretty reasonably fast.
The risetime is around the LC time constant t = pi*sqrt(LC)/2, or 130ns. Pretty slow as switching supplies go, but if your switching frequency is under 100kHz, it'll be fine.
but you'll get a much lower leakage inductance if you get rid of the bobbin and wind say, cat 5 cable directly on the ferrite, feeding it through the core like you wind a toroid. this should result in half the copper physical length, and probably one fourth the leakage inductance or less.. simply because the turns are closer to the core and each other.
Yeah, rewinding with a good winding design (multifilar isn't bad -- good enough for modest transistor drive applications, but don't expect too much out of it) gives you the full performance of the core, without the limitations of whatever the original winding did. And the cores are intentionally optimized for just this sort of thing: a good pulse transformer has minimal current flow to ground (magnetizing current), just as a CMC wants minimum current flow through the AC line!
Tim
Home
Help
Search
Login
Register
