One thing we concluded is that we don't want to use transmission-line transformers. What we really need is wideband, isolation, and low leakage inductance.
It's usually the low leakage inductance that coaxial or twisted-pair or bifalar-wound transformers provide, and the txline coupling can be an undesirable parasitic.
The only coupling that we want between windings is magnetic.
On 8/25/25 17:07, john larkin wrote: [...]
Transmission lines transformers in Spice can be surprising. For one thing, the spice transmission line model has zero common mode admittance, so you have to add explicit elements if that matters to you.
Jeroen Belleman
Right. A short txline acts like an ideal 1:1 transformer.
And simulating, say, a hunk of coax or a twsited pair properly needs more than one txline element.
I used to hear 22 KHz. Now I'm about 8. There's not a lot to hear above 8 KHz.
I haven't used an analog scope in decades. My 11802 sampler has a CRT display (raster scan magnetic deflection with touch screen!) but the rest is all solid-state.
War story: In the mid 1980s, I ran into EEs who designed only digital logic, the new new thing then, using Spice (various ancient flavors including the original) to model shielded twisted pair transmission lines. They modeled this as two closely-spaced wires that happened to be within a common shield, but with no interaction between those two wires, so they could carry totally independent signals without interference.
Umm, No. It's a transformer as well, so what comes out will be the combination of what was sent on the individual wires, and no this cannot be "fixed". The Spice of that day did have a transformer model that kinda works, which they used, but no real transmission-line model.
Analog engineers know better, but not the digits, to this day.
Joe
"Leakage inductance" doesn't turn out to be a useful concept. Concentrate on the magnetic coupling between the windings.
Transmission line transformers are poor isolators. The capacitative coupling between the winding is high. If you need speed and isolation it's got to be opto-isolators.
"Bifilar windings" are twisted pair windings, and there's obviously a lot of interwinding capacitance
Pity about the laws of physics.
It's how we quantify the coupling between windings. You know, engineering.
For coupling a kilovolt pulse into a 50 ohm load?
Sometimes physics needs some help from design.
Have you looked into constructing your own 50-ohm load in the form of a shorted coaxial or twin transmission line made from some resistive material such as nichrome?
If it were coaxial and made from expanded metal sheet rolled into a cylinder, the resistance would be higher than the sheet metal alone and the cooling would be much better. If it were twin, ordinary nichrome wire could be used with the conductors firmly held at the correct spacing by heat-resisting insulators. There is a considerable force generated between parallel conductors carrying heavy current, so they need to be restrained.
To get 50 ohms you might need a fair length but I presume this is for your own testing purposes and you aren't intending to sell them.
The Caddock dpak 50 ohm resistors look very good. We banged one with a few hundred million 600 volt pulses and its value didn't change at all.
That's sure easy. I plan to dremel some dummy loads and see what their high-speed behavior is like.
I have some 2512 thinfilm resistors on order, ditto.
Building some mechanical thing would be a last resort. It's easy to pick-and-place parts on pc boards.
You shouldn't. The right way to quantify coupling is to measure it directly. What you are measuring is the proportion of the flux lines generated by one winding that couple into the next winding.
If you want lots of isolation you can't couple it directly.
Design can't help if you don't understand what you are trying to design, and the physical limits on what your components can do.
Some decades ago the Review of Scientific Instruments published a paper on a dummy load for big fast high-voltage pulses which used lots of more or less conventional resistors. Sadly none of them were surface mount back then.
If I remember rightly the idea was to organise the resistors into rings.
The individual resistor value in each ring went up as the square of the diameter of the ring - each ring could contain more resistors all working ion parallel, and each ring had more surface area to dump the power into.
A pick and place machine would need to be able to place the resistors in random oerientations in order to do that. A machine that could only do
0-degree and 90-degree orientations could only populate an array of nested rectangles. Adding two 45-dgree orientations would allow nested octagons, which would probably be good enough - a fat circular ring could contain an octagon.
How?
The usual approach is to short one winding and measure the inductance of the other. Since the shorted winding has a finite impedance, it isn't entirely accurate.
If you know the resistance of each winding - which you can measure perfectly accurately at low frequencies - and the parallel capacitance of each winding - which is lot harder to measure since the two windings interact - most people can do better with a series of measurements at different frequencies. You need to measure the in-phase and quadrature components of the output signals, which is a pest.
Few bother. If you want the national standards lab take on the problem you can buy Rayner and Kibble's "Coaxial AC Bridges" ISBN 0-85274-3989-0 which cover's "mutual inductors" on pages 79-87. Bryan Kibble's name is attached to the modern standard of mass - the "Kibble Balance".
The real problem of talking in terms of leakage inductance is that it gives you the wrong point of view about what's going on in the transformer, which can be decidedly complicated.
We had some tricky problems with scan coils at Cambridge Instruments, and getting a more or less correct point of view turned out to be helpful.
Someone said that " doesn't turn out to be a useful concept"
It's easy to measure and maps directly into a Spice model.
And works.
Once the transformer is built, I generally TDR it. Yes, we actually build transformers.
I found the concept of leakage inductance very useful, but I did not measure it like that. Shorting a winding isn't practical in RF transformers where leakage inductance is in the single-digit nanohenry range.
I would measure the high frequency cut-off using a VNA and calculate the leakage inductance from that, then insert that into a spice model to confirm that this indeed resulted in the measured cut-off frequency.
There are other parasitic elements that affect upper cut-off.
I had one transformer where the leakage inductance worked out to about 500pH. It had a bandwidth of 10kHz-6GHz. There is no such thing as a dead short at 6GHz. You can only get so close.
Jeroen Belleman
You're an RF guy, and I live in time domain. So I use TDR to measure tiny inductances. They are mathematically about the same, but TDR is DC coupled.
I recently TDR'd some Coilcraft hex-a-path transformers (terrible) and a giant 300 watt planar power transfomer (surprisingly good)
We're using GaN fets to drive our transformers, with 700v edges in the
1-2 ns range. Slow!
It's an easy measurement that doesn't give quite the right answer.
It's certainly an easy measurement. The Spice model for a mutual inductor includes the coupling coefficient, but that scheme of shorting one winding and measuring the residual inductance of the other doesn't give you the coupling coefficient directly. You have to plug in more data to get there.
Well enough, most of the time. Life gets awkward when it doesn't.
Doing time domain reflectometry on a transformer is kind of odd. Do you load the secondary with some kind of terminating resistance?
But you much prefer to buy them off the shelf. And when you model them in Spice you mostly leave out the series resistance and parallel capacitance of the windings. The parallel resistance of the inductive loop in the core doesn't come up either (not that I've ever bothered to model that, but I do know enough to be aware that nickel-zinc ferrite cores have a higher resistance around that loop than manganese-zinc ferrites).
Nothing gives the exact answer. One key part of engineering is knowing what is important and what's not. If you factor in everything, you'll get nothing done.
Are you interested in getting things done?
I don't bother to compute K. I just set K=1 and add an inductor to the circuit.
Sometimes Spice seems to run faster if K=0.9999, so I might do that too.
Granted. Play it safe and never do anything odd.
When I can get a nice surface-mount part for 95 cents, I'd rather not design and manufacture it.
I leave them out when they don't matter. Sometimes very subtle things matter in magnetics, like magnetic domain noise, magnestriction, low-level stickiness, whatever. It's a matter of estimating magnitudes and ignoring the tiny stuff.
If you showed the netlist for this errant passive circuit it might be possible to take an educated guess why the solver is going haywire.
My money is that the unrealistic fast rise time of an "ideal" current source is doing something nasty to the transient solvers algorithm.
But you do have to know the extent of the corners you are cutting, and when to move up to the next level of approximation.
You do look more like a slap=dash chancer.
I've done lots of cleaning up behind people who were in too much of a hurry to get things done. Knowing when what you've got is good enough is a judgement call, and there tends to be a lot of pressure to make the call a little too soon.
That's foolish.
That almost as foolish.
Never apologise and never explain. Great image building and rotten engineering.
So you rely on the 50R termination in the oscilliscope, but can't be bothered to mention it.
But you rarely can. A transformer has lots of parameters you can chose if you design your own - if you design anything.
But how do you know they don't matter if you haven't modelled them.
But you seem to ignore quite significant stuff.
Have something to add? Share your thoughts — no account required.
Ask the community — no account required