The bottom of a Tesla coil's secondary has to go somewhere, and it is not the wall socket. Outside the coil the ground is one side of the circuit. Inside the driver it is three separate things that must not be joined, and that half is what kills more QCW drivers than anything else.
Why the secondary needs a ground at all
Every streamer that leaves the topload is current, and that current has to come back. It travels through the air, into whatever it strikes or simply into the surroundings by displacement, and then back to the bottom of the winding. In the usual arrangement, one end of the winding at high voltage and the other at the base, the resonator is not a closed loop of wire; it is a loop that goes out through space and returns through the ground connection.
So the ground is not a safety afterthought. It is one side of the circuit.
It is more specific than that: the ground is one plate of a capacitor. The toroid's capacitance is capacitance to ground: 10.88 pF on one coil, which is about 34 kΩ of reactance at 426 kHz [derived, X = 1/(2πfC)]. The resonance runs on the whole of Ces, that 10.88 pF plus the winding's own 3.15, so 14.03 pF and 26.6 kΩ, and that is the resonator's own impedance, the number that, with the winding's inductance, fixes the frequency. That page reaches the same 26.6 kΩ by a different road, √(L/C) rather than 1/(2πfC), and the two agree because at resonance ωL and 1/ωC are one quantity [derived]. It owns the figure; this page is quoting it.
The clearest evidence sits in the calculator everyone already uses. Before JavaTC will return a number it asks for ground radius, wall radius and ceiling height. It is asking about the room, because without a counterpoise the room is the other plate: the floor, the walls, the bench, you. That still resonates, it is simply not a capacitance you can repeat, and it moves when you walk towards the coil.
Not mains earth
A counterpoise instead
A large sheet of metal under the coil, which the secondary's base connects to. Roofing aluminium beats foil or mesh, because it does not crumple and does not have a thousand joints. Round the corners, or they will corona.
The lead should be short, and the reason is not the impedance it adds. Ten metres of wire is about 27 Ω at these frequencies [derived, ten metres at the usual rule of thumb of a microhenry per metre is 10 µH, and 2πfL at 426 kHz is 26.8 Ω], against a toroid reactance of 34 kΩ, eight parts in ten thousand, and it changes nothing about the resonance. What matters is what the lead radiates. At a hundred kilovolts on the topload, that wire is carrying about three amps at 426 kHz [derived, I = 2πfCV on 10.88 pF]. Run it beside a signal cable and the signal cable now carries it too. Short, and alone.
What the lead is made of is less settled, and worth being honest about. There is no published gauge for it. What is worth knowing is that at 426 kHz the current runs on the surface, so a strap or a length of braid carries it better than a round wire of the same cross-section: size it for surface, not for area.
The base of the secondary connects to both the counterpoise and the strike ring.
The strike ring
A grounded ring around the coil, there to take the arc hits that would otherwise reach the ceiling, the operator, or the electronics. Run it to the RF ground on its own wire, not the counterpoise's and not through the frame: the two carry different currents and should not share a lead.
Give it a generous tube diameter and keep it clear of the winding, and leave it no sharp ends: cap them with balls or bend them back. A point on the ring coronas, and then the ring is a source instead of a sink.
How big it has to be
Half the topload's field leaves downward. A sheet of radius R sitting h below it intercepts ½(1 − h/√(h²+R²)) of the whole sphere. Of the half that goes its way, it catches:
R / h | of the downward field |
|---|---|
| 1× | 29 % |
| 2× | 55 % |
| 3× | 68 % |
| 10× | 90 % |
A radius of twice the height is where it stops being cheap; three times is where it stops mattering. Going from three to ten buys twenty two points for eleven times the aluminium.
There is a second rule for the same decision, and it is on a different axis. Barnkob sizes the ground by capacitance rather than by field: "To avoid spark formation at the bottom of the secondary coil, we need to have 10 times the capacitance in our grounding system than the topload has" (grounding guide). The two rules are not in competition, because they are answering different questions: the table above asks how much of the returning field you catch, and his ratio asks whether the bottom of the winding stays quiet. Ten to one is a target you can compute from the same geometry that produced the table, so run both and take the larger sheet.
And the word counterpoise is borrowed from a place where it cannot be honoured. In radio it means radials at least half a wavelength long. At 30 to 300 kHz that is "a half wave length of 5000 to 500 meter", so nothing a coil builder puts on a floor is a counterpoise in the original sense, and the sizing above exists precisely because the radio rule is unreachable. What Barnkob recommends instead is borrowed from wind turbines, a ring buried about 0.8 m down, "as this has the best potential distribution".
Which is the one place a number does exist for the shape. Comparing the voltage distribution of grounding geometries, lower being less chance of flashover from the grounding system before the energy reaches ground:
| geometry | ratio, lower is better |
|---|---|
| Right-angle turn | 8.41 |
| Three-point star | 6.45 |
| Four-point star | 5.50 |
| Six-point star | 4.61 |
| Eight-point star | 4.19 |
| Ring of wire | 3.49 |
A ring beats every star, and the gain from four points to eight is smaller than the gain from eight points to a ring. Read it against the impedance trade below, which has no numbers at all: shape is the part of grounding that has been quantified, and impedance is the part that has not.
A second rule that does hold up sizes the ground by capacitance rather than by geometry. Mads Barnkob's grounding guide asks for ten times as much capacitance in the ground system as the topload has, and gives the pieces to count with: a thin rod is worth about 10 pF per metre, and a square of one metre by one metre 40.8 pF. Read that second figure the way he writes it. A plate's capacitance goes with the length of its sides and not with its area, and he says so in the same paragraph and again in his own table, where a 42 pF topload asks for a 10 by 10 metre plate rather than for ten square metres of any shape. The coil above carries 10.88 pF on its topload, which is JavaTC's own split of its Ces and not a figure worked back from the reactance above, so it wants about 110 pF underneath: eleven metres of rod, or a square about 2.7 metres a side, which is seven square metres of sheet [derived, 108.8 / 40.8 gives the side in metres, then squared]. This page printed "under three square metres" until now, which is what comes out if 40.8 is read as a figure per square metre, and it is the one reading Barnkob's own paragraph rules out. And with the topload a metre up, R = 2h is a two metre radius, so 12.6 square metres [derived, πR²] and 12.6 metres round the edge, which on his other rule of thumb of 10 pF per metre of circumference is about 126 pF against the 109 wanted [derived]. So a counterpoise sized by the geometry still clears the capacitance rule, but by a sixth rather than by the wide margin the old arithmetic suggested. The rules do not fight. Size for the geometry.
Joining strips is fine, and the arithmetic is not close. A 5 cm overlap along a metre of seam is 0.05 square metres of plate against plate; hold the two a whole millimetre apart with nothing but air between them, which is far worse than any real lap joint, and it is still 0.44 nF and under an ohm at coil frequencies [derived, C = ε₀A/d then X = 1/(2πfC) at 426 kHz, giving 0.84 Ω]. Against 34 kΩ of toroid reactance, at three amps, that is a couple of volts under a hundred kilovolts. This page previously printed "ten to fifteen ohms" for the same seam with nothing under it, and no plausible gap gets there. The check is to compare a seam against the toroid's reactance rather than against zero, and it is the same check that says taped toroid seams are acceptable.
| Buys | Costs | |
|---|---|---|
| Low impedance | less radiated interference | whiplash risk |
| High impedance | protects the winding | more interference |
That trade is Eric Goodchild's proposal, and Barnkob's grounding guide is where it is written down: coils on counterpoise grounds seemed to burn their secondaries more often, and a poor, high-impedance ground would absorb the whiplash instead of reflecting it. The guide prints it without settling it, and says plainly that there is no clear answer without measurements that are hard to make and need expensive equipment. Neither column of that table has a number against it, so it is a direction and not a setting.
One number does exist for the event itself, from the man who proposed the trade. Steve Ward, on the Tesla Coil Mailing List in 2012: "During a ground discharge, the voltage at the topload has been measured to collapse within 100nS (worst case) and 250nS (more typical)", and during that event "a wave front propagates down from the top of the coil". So the front is real and it is fast, a quarter of a microsecond on a machine whose bang lasts hundreds.
What is not established is what the front does when it arrives. Ward's own guess in the same exchange was that end-to-end capacitance, topload to winding included, would disperse it and take the steepness out. Paul Nicholson's answer concedes the dispersion and does not concede the conclusion: "Yes dispersion spreads the down-going transient although it could still build up quite a high volts/turn near the base. This has been discussed before as one possible cause of racing arcs" (TCML, 23 September 2012). One possible cause, discussed before, still open in 2012.
already most of the benefit; chasing zero is not a goal.
The frame under the primary is a shorted turn
The cure is to break the loop: nylon bushes, washers, spacers and Kapton, so that each upper rail is grounded at one end only. What that break has to stand is not the few volts a bolted joint suggests but whatever the primary induces around the frame at kiloamps, and this corpus has no measured figure for it: the 150 V per side printed here until now had no source anywhere. Nor would it have carried the rule it was attached to, that two millimetres of gap is not enough, because two millimetres of air stands off about 6 kV [derived, 3 kV per millimetre] and 150 V is forty times inside that. Break the loop and ground each rail once. Sizing a gap is the wrong question.
This is the piece that belongs to both halves of the subject. The frame is outside the electronics and carries the coil's own current, and it is also the thing most likely to put that current somewhere the driver can feel it.
Three grounds inside the driver
Separately from the coil's ground, the driver has three of its own and they are not interchangeable.
- GND2 is the high side device's emitter. It floats with the device. Do not join it to PGND.
- PGND is the low side's emitter, which is fixed.
- GND1 is logic ground. Join it to PGND at one star point, Kelvin.
The point of the star is that power currents should have no physical path through the logic ground.
The gate supply crosses the same boundary
And when the arc hits ground, a QCW does the opposite
On an ordinary DRSSTC a strike to ground is a low resistance short across the secondary, and the primary current slams into its limit.
Either way the topload voltage collapses and the arc takes the load off the secondary. How fast is not pinned, and this page had it an order of magnitude out. It carried 100 to 250 nanoseconds with no source underneath, while the one thread with a strike actually measured on it, HVF 117, has Uspring reading the duration as "around 1-2 us" and Hydron's capture of the same event is where the 25 kΩ below comes from. Take the microseconds; they are the measured ones. What the overcurrent detector does with that is on the same page, and the short version is that whether it fires is the threshold rather than the strike: half again over the normal ramp needs a channel of 22.1 kΩ and a fifth over needs 37.3, while the one strike anybody has published bottoms out at 25, between the two.
One thing that will surprise you
Metal closer than half the secondary's diameter kills the Q. A chassis, a case, a rack, a radiator: they load the resonator and steal energy into induction heating, and the coil simply performs worse for no reason you can see on a scope.
The earth symbol under the secondary on every department diagram is this, and what it is not connected to is the more important half.