More equipment dies at the bench than in the arc, usually for one reason: a scope's ground is bonded to mains earth, and the thing you are probing is not.
The one rule
There are two ways to meet that rule, and builders use both.
- Isolate the circuit under test. Ward's own test setup lists an isolation transformer "for powering the bridge electronics", wired for a 1:1 ratio. On HVF 646 Teravolt argues it on price: an ordinary dual-winding mains transformer "will be cheaper than a diferencial probe".
- Or leave the coil where it is and probe differentially. That costs more, as Teravolt says, and it is the only way to look at two nodes at different potentials at the same time (see the next section).
Once the electronics are isolated the probes need not be exotic. Ward asks for "at least 20MHz bandwidth, and a good set of 10:1 voltage probes so that you can view 200-400V signals safely". That is one builder's bench. We have no bench recommendation of our own.
One more trap, unrelated to voltage: on AC coupling the entire DC component sits across the scope's input blocking capacitor. That part is rated for what the input is rated for, not for your bus.
Probing the driver
A gate and its own emitter make one measurement. Two of them in the same leg cannot be measured at the same time with ordinary probes.
Both driver articles refer back here for their probing warning. For what the board is doing while you probe it, see what a DRSSTC driver does. The same board with its part values is inside the driver.
The probe is part of the circuit
Ward found sag and ringing on a gate to emitter trace that were not in the gate loop at all.
There are very large, high-frequency currents flowing in the primary circuit, and thus any loop of wire suddenly becomes an antenna with a voltage induced by the dB/dt.
The frequency gives it away. The droop and the ring sit at the primary's own resonance, "the natural Fo of the primary system", which the gate drive knows nothing about. His fix costs nothing: "squeeze down the loop area of the probe tip and the ground clip, preferably by twisting the ground clip around the probe body itself". The trace he publishes afterwards has neither the droop nor the ring.
This matters for the gate diagnosis too. Reading a gate waveform files sag under too few turns on the transformer and low frequency ringing under too much leakage inductance, and the cure in both cases is a rewind. Neither entry mentions the probe. Twist the ground lead against the probe body and take the trace again before you take a transformer apart.
Ward also runs the probe lead and the scope's own mains cable through ferrite toroids against common mode noise, and he is frank about how much it helps: "You can operate without the filters completely, but you may get waveforms that look noisier than they really are."
A probe can also start a coil, not just give a false reading. The symptom "it starts only while a scope probe touches the comparator input" is covered in every bang starts open loop, where the probe acts as an antenna into a comparator with an undefined idle state.
The two burdens are not the same
A UD2.x board has two current transformers with two burden resistors, and only one of them will take an ordinary probe.
- The feedback burden has one side on ground, so an ordinary probe goes across it. It reads the current, because the phase lead coil is in series and the same current runs through both. That is not what the comparator reads. The driver's input sees the resistor and the lead coil together, so the waveform on your screen and the waveform the loop switches on differ in phase.
- The overcurrent burden has neither side on ground. Clip a probe ground to one end and you have joined that end to the scope's earth, a connection the board was not built with. The comparator then sees something different from before, and "new faults appeared after probing" is on the site's symptom list for this reason. It needs a differential probe.
Check that there really are two rings before you assume it. The values, and why the two are kept apart on the boards that do so, are in two rings, not one, which also covers the UD3 putting gate drive synchronisation and current limiting on one input, on a single burden, by design.
Looking at the bridge output
Use two probes, one on each output of a full bridge, with both ground leads on the same node, the negative bus, and subtract in the scope. You get the differential without a differential probe, and it obeys the warning above because both ground clips are on one point.
The loop margin is measured the same way, on a bus low enough to be harmless: the margin is in the loop.
Looking at a gate
Probe it through an isolated supply. How much bus voltage the picture needs before you can set a gate resistor from it is covered in reading a gate waveform.
Ward reports two things from his own small half bridge that are useful when you bring a coil up on a bench supply. Both belong to his coil, not to yours as thresholds. Under 35VDC its feedback was too weak to hold a strong oscillation, so a coil that will not start at the bottom of the range is not necessarily faulty. And under about 50VDC, in his words, "IGBT damage is very difficult to achieve".
The current transformer, while you are in there
Where the ring goes, and why never between the primary and the tank capacitor, is in what a DRSSTC current transformer is, and what it senses.
A ring is not always the right instrument
All of the above assumes you want a current transformer. For measurement, as opposed to feedback, a resistor is a real alternative. The question came up when somebody wanted to replace shunts with CTs for a topload current measurement, and Uspring raised two objections (HVF 117):
Do you have an idea about their bandwidth considering e.g. ground strikes? … An advantage of using resistors is, that you can catch possible polarity dependent effects of the arc. A CT might filter that out.
Both objections come from the same property. A ring will not pass DC and its low end rolls off, so it removes anything the arc does slowly or asymmetrically, and a ground strike is the event most likely to contain both. A resistor has neither limit but every other drawback, chiefly that it sits in the circuit at whatever potential the circuit puts it at.
In practice the ring is right for what the driver needs and the resistor is right for finding out what the arc did. If you are measuring to resolve an argument about arc behaviour, do not use the instrument that filters out the evidence.
The original question in that thread got no reply, but the answer is fairly clear. He was asking how to spend a second scope: on the primary and secondary base currents, or on one channel for the bridge output voltage. The bridge output is a square wave that sits at plus bus or minus bus and nothing else. Its only information is where the edges fall, and an external trigger input records that. He says as much himself, in a parenthesis added to the question rather than offered as the answer: "Might be able to have a 1-bit value recorded of that anyway via the ext trigger input." The currents are the analog signals, where the shape is what you want and one bit per sample tells you nothing.
Spend analog channels on signals whose shape matters and take the square wave as a trigger. That conclusion is ours, since no one in the thread replied to the question.
Making your own measurement transformer
This is not the driver's ring but a separate one whose only job is to put primary current on a scope. On HVF 1277 Daniel Uhrenholt asked for one covering 100 to 200 kHz after selling his Pearson, and got two useful answers.
Hydron's is the cheap one. "A couple of ferrite toroids (specification un-critical as long as they're not too low-Al) some wire and a resistor or two", cascaded so the ratios multiply, and in that band it "will work extremely well". He is just as plain about the downsides: no Pearson-grade electric field screen, "so it will pick up noise from fast switching transitions", and poor low frequency response. On HVF 914 he adds that his own DIY cascade "picks up a LOT of noise when not band limited to 20MHz, whereas the pearson picks up very little". Uhrenholt had already built one, two cores at roughly 1000:1, and reports the failure that matters most to a coiler: the waveform "looked awful in the start of a burst, compared to a Pearson monitor", which is probably the part of the bang you wanted to see. Hydron suggests nanocrystalline cores over ordinary ferrite as something to try, not a proven answer.
davekni's is the careful one, and he gives the full chain instead of a part number. An E80 ferrite core carries a single-layer secondary of 40 turns of 24AWG. Over that goes "just over one turn of copper foil, insulated at the overlap (no shorted turn), and connected to ground as a shield", and over that a one-turn primary of 0.2mm copper foil, made as two layers of 0.1mm. Both windings run almost the full width of the centre leg, which keeps the leakage inductance down. That stage feeds two ordinary toroidal transformers, each with a 2-turn primary and a 25-turn secondary, into a 5 ohm burden, and the sensitivity he quotes for the whole thing is 1V/100A.
That makes it a 500:1 chain, not 6250:1, and the arithmetic shows it. Taking the two toroids as two outputs off the one E80 secondary gives 40 times 25 over 2, so 500:1, and 100 A through it is 0.2 A into 5 ohm, which is the 1V/100A he states [derived, from his own turn counts]. Reading them as a cascade gives 6250:1 and 80 mV per 100 A [derived], which does not match his figure, so the parallel reading is right. The 1V/100A belongs to that chain and its 5 ohm burden, not to an E80 in general.
He reports it "works well to 2600A" and designed it for 3500A eventually. He also says he never explicitly measured its frequency response, so take the sensitivity as measured and the bandwidth as untested.
A Pearson is still the instrument to have if one turns up, and getting one takes patience more than money. Hydron's 2788s are specified "for use over 6 orders of magnitude (300Hz to 300MHz)", and he collected them slowly: "I got lucky on eBay a few times and collected a number of Pearson CTs (5 in total, averaging about 60£ each)". Uhrenholt, pricing a new one in the same thread in 2020, got "£300+" with shipping, import tax and handling into Denmark.
The main use for such a transformer is setting the current limit. Ward does it by measurement rather than calculation: "I typically set the limiter too low and watch the primary current on an oscilloscope (using a home made CT). I then operate the coil with free air streamers only if possible. I then set the limiter so that it just barely does not trigger with free air streamers. Then I check that the limiter does engage with ground arcs as it should" (general DRSSTC design guide). Why that order is right, and what the threshold is not, is covered in overcurrent is not a setting.
What can hurt you, as opposed to the equipment, is on its own page. Read it first.