[ §1 · primary ]

QCW primary design: the same part held fifty times longer

The same part, sized by the same arithmetic, holding the current fifty times longer while the frequency walks out from under it. Three published machines, and only one of them wound like a DRSSTC's.

CommonGnd23 August 202615 min read

A QCW's primary is the same part as a DRSSTC's, sized by the same arithmetic, and asked to hold the current for fifty times longer while the frequency walks out from under it. That is most of the difference. The rest is what builders do about it: how much tank, what shape to wind it, which way to detune it and whether to put ferrite in it, and on none of the four is the published record of one mind.

What it is and why

The impedance rule does not change here. Primary design is where it is worked, and none of it is superseded by anything below.

What changes is the ontime. A DRSSTC's primary carries its current for hundreds of microseconds; a QCW's carries it for milliseconds, ramp after ramp, so the copper and the joints are a thermal problem before they are an electrical one. And the frequency the tank is tuned against is not one frequency: it slides across the bang, which is the ramp's own page.

What you decide

How much tank, and why the band is so much smaller

Single digit nanofarads to about fifteen, and the comparison usually drawn against a DRSSTC understates the gap rather than overstating it. Barnkob's guide sizes a DRSSTC tank by the coil it feeds and gives 0.1 µF at the smallest of five sizes and 2 µF at the largest (the MMC chapter), which is hundreds of nanofarads and not tens. Both ends of that are in this corpus already: Shane Colton's DRSSTC measures 100 nF against a 10.6 µH primary, and Hynes's MMC tops out at 2.55 µF. Tens of nanofarads describes small high frequency machines, this site's own among them, rather than the class. This site's own coil is at 12.2 nF against a 16.96 µH primary, which is 37.3 Ω of tank impedance, and Gao's QCW 1 and 1.5 are both at a nominal 12.8.

The single digit end is not a rounding down to make the sentence read well. Gao's QCW 2 runs 5.875 nF into a "17 turn 1.2" dia 18AWG" primary at "60Apk Max Primary Current", on a secondary he puts at about 308 kHz unloaded, and it throws sparks over two feet out of a machine built to fit a postal box. So the band has both ends on the record and they are the same builder's neighbours rather than extrapolations: 5.875 nF at 60 A here, and 64.8 nF at 345 A on the machine in the note below, which is eleven times the capacitance and near six times the current.

It is not a preference. A low impedance tank accelerates the current too fast at the start of the bang, so the overcurrent detector fires before the ramp has finished and instead of a sword you get a pop. On a DRSSTC the same tank is simply a harder-working one, because the bang is over before the detector's threshold means anything about the shape of the arc.

The band is a property of an ordinary bridge, not of the topology, and one machine on the record is four times outside it. David Knierim's larger QCW runs 64.8 nF in a 3s72p of 2.7 nF parts, split into two 32.4 nF halves feeding two primary windings, against a 5.6 µH primary (HVF 2397). That is 9.3 Ω of tank impedance taking the tank whole, or 18.6 Ω counting one branch at a time, each winding being 11.2 µH against its own 32.4 nF half [derived, sqrt(L/C) on his stated figures]. Those are the same tank counted two ways rather than two readings of an unclear spec, which is why the factor between them is exactly two, and either way it lands two to four times below this site's 37.3 Ω. He gets away with it because he built the bridge for it: eight IGBTs, in pairs per switch. That the ceiling is a current and a detector threshold rather than a count of nanofarads is worked on the MMC page, on this same machine. But read what the current actually is before taking this as licence. On his own scope, a bus ramping to 240 V puts the average primary current at 220 A, "so 345A peak of sine wave", and 53 kW into the bridge at the end of the ramp. Gao's two machines run 100 to 126 A and 145 to 160 A, and those are peaks, so the figure to hold his against is the 345 A and not the 220. Peak against peak this is two to three times the current of an ordinary QCW [derived, 345/160 and 345/126], and low tank impedance is exactly how it got there: at a fixed frequency the primary current goes as the capacitance. What the tank does not explain is where the power goes. He first puts the loss at about 40 per cent of it, "mostly in primary foil winding", and revises that upward later in the same thread once he has a better model: "just over half of the power is dissipated in primary coil losses, even more than my earlier version that showed 40%." Take the half, not the 40. He does not put that down to the tank at all: "Foil provided high coupling but also high losses", because blocking the field is what makes the eddy currents as well as the coupling. So this machine costs twice over, once in current from the capacitance and once in loss from the foil, and the two are separate choices that happen to arrive together. Keeping them apart is the difference between reading it and copying it. Nobody should read 64.8 nF as a target. It is what a purpose-built bridge will carry, and the arithmetic that gets you there runs the other way for everybody else.

What shape it is, which is not the DRSSTC answer

The three shapes an ordinary DRSSTC chooses between, flat spiral, helix and cone, are mostly not what the published QCWs are wound as. They are solenoids, or primaries interleaved with the secondary, or in one case the "half-donut" of Gao's QCW 2. One machine on the record is the exception, and it is described next. So a primary that sits inside its secondary rather than beside it is the norm here rather than a curiosity, and it is the fourth place where QCW practice runs the other way from DRSSTC practice.

Knierim's larger coil is the one written up in enough detail to copy, and it is the exception just named: four turns of 50 µm copper foil tape, as two four-turn windings interleaved with each other, each taking its own half of the tank, wrinkled into a conical shape. Interleaved with each other, note, not with the secondary. So it is a cone in the DRSSTC sense and a foil winding in nobody's, and the foil is the whole point of it: "The goal is to block all magnetic field penetration through primary winding, as it is solid copper from a side perspective. That is what makes coupling high."

That is the mechanism behind the whole divergence, and it is worth having plainly. A tube or a tapped wire lets field through the gaps between turns. A foil does not, because from the side it is a continuous sheet of copper, and a primary that blocks the field is a primary that shares it.

And he names two things he would change, which are worth more than a verdict on the whole machine. The smaller one is the foil's thickness: "100um foil would be better, but I didn't have much and it is much harder to wrinkle into proper conical shape." He had already treated the worst of it, since "the bottom turn is double-layer 50um tape" because FEMM put the maximum loss there. The larger one is the foil itself. "In hindsight I would have made a somewhat more standard primary. May decide to make such a new primary someday." He gives the reason plainly, that the gain in coupling is small against a large drop in Q and a ferrite core holds the coupling up anyway with ordinary wire or tubing. So the thickness is a fix he did not take, and the foil is a turning he would not take again. Read the machine as a scale nobody else on this page is working at, not as a warning: he stopped at 53 kW because "arcs are already problematic in my living room", with more available.

The coupling that buys is the other half of the divergence. A DRSSTC sits at 0.12 to 0.2 and a QCW at 0.30 to 0.50, a factor of two and a half at both ends, and the ferrite machines below are past even that.

Which way to detune, on which the record takes all three positions

The ferrite QCWs on the record hold the upper pole, and no two of them get there the same way.

David Knierim tunes the primary far below the secondary, with many times the resonant capacitance, on the first of his two. On the second he does not. That one runs matched, and he says why in the same breath: "But it makes sense for upper pole with high coupling and matched frequencies." His own figures agree, because 5.6 uH with 64.8 nF rings at 264.2 kHz and the secondary's 11.8 mH with its stated 32 pF rings at 259.0 kHz, two per cent apart rather than far below [derived, both]. So the same builder sits on two different sides of this question, and the pole check further down depends on it: f0/sqrt(1 - k) is the upper pole for a matched pair, which is why it reproduces his measured 485 kHz. Jan (dr. kilovolt) tunes his above it: 279 kHz against the secondary's 251, running at 400 kHz at the start of the ramp and 350 at the end.

And he says himself that it is the wrong way round. In his own words the primary "is not designed optimally, it would be better to use a lower inductance and higher capacitance and tune the primary lower". What makes that worth reading rather than filing is the next sentence of it, which is the warning below.

Whether to put ferrite in it

The two sections above assume copper in air. Put ferrite inside the winding and the primary stops being a coil you tap and becomes a transformer core, which is a different machine and not a better version of the one the impedance page describes.

David Knierim's replaceable ferrite-core primary is the worked case, and it is his first of two, itself following Jan's potted ferrite QCW on HVF 1073. The idea he states is "to pot the inside of the secondary with a cavity for the primary", so the core assembly slides out and can be changed. The stack is 44 E-55 halves in 3C92 with four flat E-64-50-10 cores for spacing and 3F4 top plates at 64 by 51 by 5 mm, and he prices it: "$158 for all the E-cores". The cavity is lined with "a Faraday cage with no closed loops. All the wires connect at only the top", which is the only way to put metal that close to a secondary without shorting a turn.

What it buys is coupling, and the size of the change is the point. He measures k = 0.909 with the ferrite in. Without it the thread gives two figures rather than one, 0.51 on a mock-up and an estimate of "around 0.7 or just under", so read the air-cored number as a range and the ferrite one as the measurement. His upper pole lands at 124 kHz with no topload and 106 with a 620 mm one.

Read that machine as the experiment, because he does. He opens the thread on his second coil with "After experimenting with a low-frequency (100kHz) QCW coil earlier", and the sentence that follows it is "Decided to try a somewhat-more-normal QCW coil" (HVF 2397). The normal one lands at k = 0.71 on a ferrite floor and centre post, with the top of the post in NiZn for low conductivity, and its upper pole unloaded at 485 kHz against 124 on the first. His own three figures close on themselves: 5.6 µH with 64.8 nF rings at 264.2 kHz, and f0/sqrt(1 - k) at his stated 0.71 puts the upper pole at 490.6 kHz against the 485 he measured, 1.2 per cent apart [derived, all three]. A spec that reproduces its own pole is worth more than one that only states it.

What will get you

The numbers

Three machines, all on the record, quoted from their own threads. Two of them are the same builder's, and the second is the one he calls normal.

Jan (dr. kilovolt), SiC phase-shift QCW
  primary            279 kHz          secondary   251 kHz
  running            400 kHz start -> 350 at the end
  k                  0.55, "very high, but ... not utilized fully"
  core               ferrite rod, I-cores of 3C97, about half the
                     secondary's height
  potting            polyurethane, against flashover
  arc                "over 2 meters", "approximately 2-2.5m"

David Knierim, replaceable ferrite-core QCW      his first, the experiment
  k                  0.909 with ferrite; 0.51 on an air-cored mock-up
  upper pole         124 kHz bare, 106 kHz with a 620 mm topload
  core               44 E-55 halves in 3C92, $158 for the E-cores

David Knierim, the larger one                   his second, "somewhat-more-normal"
  primary            5.6 uH, 4 turns of 50 um copper foil tape, conical,
                     as two 4-turn windings interleaved with each other,
                     bottom turn doubled where FEMM put the worst loss
  current            220 A average, "345A peak of sine wave", on a
                     bus ramping to 240 V
  bridge power       53 kW peak at the end of the ramp, of which about
                     40 % is loss, "mostly in primary foil winding"
  tank               64.8 nF, 3s72p of 2.7 nF 1600 Vdc PP, split into
                     two 32.4 nF halves, one per winding
  impedance          9.3 or 18.6 ohm, on how the 5.6 uH is read [derived]
  k                  0.71,  ferrite floor and centre post, no potting,
                     NiZn at the top of the post for low conductivity
  upper pole         485 kHz unloaded, 440 kHz at the end of the ramp
  secondary          11.8 mH, 170 mm across, 210 mm of winding on a
                     300 mm core, 255 turns and 4 spiralled
  topload            545 mm ring of 19 mm tube, ~32 pF with the coil
  bridge             8 IGBTs, FGH75T65SHDTLN4 in pairs per switch
  arc                none stated: he writes that there are no arc length
                     to secondary height records here

What goes wrong

The detector ends the bang before the ramp does. Tank impedance too low for the start of the ramp. It reads as a pop rather than as a protection event, which is why it is usually blamed on the modulator.
The coil runs but will not tune. If it is potted, that is the end of the conversation, and it is the failure Jan documents on himself.
The copper is warm after a run and was not on the DRSSTC it came from. Fifty times the ontime buys most of that, but not fifty times, and the page's own argument says why: a QCW runs the higher tank impedance and so the lower current, and heating goes as the square of it. Work the multiple on your two machines rather than taking the ontime ratio for it. One machine on the record does put a figure on the winding, just over half of 53 kW mostly in the primary foil, but read what it is a figure for: eddy currents in a foil chosen to block the field, not the ontime and not the tank. No temperature anywhere, on any of them.

Where next

Primary design, for the impedance rule, the two regimes and the detuning arithmetic this page assumes.
What a DRSSTC primary is for, if none of the above sounded like an answer to what the part does.
How the frequency slides down the ramp, because the tank here is tuned against a moving target.

primaryferriteQCW